Structural Seismic Design Optimization

Structural Seismic
Design Optimization and

Earthquake Engineering:
Formulations and Applications
Vagelis Plevris
School of Pedagogical and Technological Education,
Greece National Technical University of Athens, Greece
Chara Ch. Mitropoulou

Institute of Structural Analysis and Seismic Research,
National Technical University of Athens, Greece
Nikos D. Lagaros
National Technical University of Athens, Greece Hellenic Army, Greece
Managing Director: Lindsay Johnston
Senior Editorial Director: Heather A. Probst
Book Production Manager: Sean Woznicki
Development Manager: Joel Gamon
Development Editor: Myla Harty
Acquisitions Editor: Erika Gallagher
Typesetter: Nicole Sparano
Cover Design: Nick Newcomer, Lisandro Gonzalez
Published in the United States of America by

Engineering Science Reference (an imprint of IGI Global)
701 E. Chocolate Avenue
Hershey PA 17033
Tel: 717-533-8845
Fax: 717-533-8661
E-mail: cust@igi-global.com
Web site: http://www.igi-global.com
Copyright 2012 by IGI Global. All rights reserved. No part of this publication may be reproduced, stored or distributed in
any form or by any means, electronic or mechanical, including photocopying, without written permission from the publisher.
Product or company names used in this set are for identification purposes only. Inclusion of the names of the products or
companies does not indicate a claim of ownership by IGI Global of the trademark or registered trademark.
Library of Congress Cataloging-in-Publication Data

Structural seismic design optimization and earthquake engineering: formulations and applications / Vagelis Plevris ... [et al.],
editors.
p. cm.
Includes bibliographical references and index.
Summary: This book focuses on the research around earthquake engineering, in particular, the field of implementation of
optimization algorithms in earthquake engineering problems, including simulation issues for the accurate prediction of the
seismic response of structures, design optimization procedures, soft computing applications, and other important advance-
ments in seismic analysis and design-- Provided by publisher.
ISBN 978-1-4666-1640-0 (hardcover) -- ISBN 978-1-4666-1641-7 (ebook) -- ISBN 978-1-4666-1642-4 (print & perpetual
access) 1. Earthquake engineering. 2. Structural optimization. I. Plevris, Vagelis, 1976-
TA654.6.S8673 2012
624.1762--dc23
2012003307
British Cataloguing in Publication Data

A Cataloguing in Publication record for this book is available from the British Library.
All work contributed to this book is new, previously-unpublished material. The views expressed in this book are those of the
authors, but not necessarily of the publisher.
Editorial Advisory Board
Christoph Adam, University of Innsbruck, Austria
Christian Bucher, Vienna University of Technology, Austria
Tadeusz Burczynski, Silesian University of Technology, Poland
Luis Esteva Maraboto, National University of Mexico, Mexico
Dan M. Frangopol, Lehigh University, USA
Michalis Fragiadakis, National Technical University of Athens, Greece
Hector Jensen, Universidad Tcnica Federico Santa Mara, Chile
Matthew G. Karlaftis, National Technical University of Athens, Greece
Manolis Papadrakakis, National Technical University of Athens, Greece
Shahram Pezeshk, University of Memphis, USA
Table of Contents
Foreword..............................................................................................................................................xiii
Preface.................................................................................................................................................. xiv
Acknowledgment................................................................................................................................. xvi
Chapter 1
Life Cycle Cost Considerations in Seismic Design Optimization of Structures..................................... 1
Bora Gencturk, University of Illinois at Urbana-Champaign, USA
Amr S. Elnashai, University of Illinois at Urbana-Champaign, USA
Chapter 2
Performance-Based Seismic Design: A Search-Based Cost Optimization with
Minimum Reliability Constraints.......................................................................................................... 23
Oscar Mller, University of Rosario, Argentina
Marcelo Rubinstein, University of Rosario, Argentina
Fabin Savino, University of Rosario, Argentina
Ricardo O. Foschi, University of British Columbia, Canada
Chapter 3
Discrete Variable Structural Optimization of Systems under Stochastic Earthquake Excitation.......... 51
Hctor Jensen, Universidad Tecnica Federico Santa Maria, Chile
Marcos Valdebenito, Universidad Tecnica Federico Santa Maria, Chile
Juan Seplveda, Universidad Tecnica Federico Santa Maria, Chile
Luis Becerra, Universidad Tecnica Federico Santa Maria, Chile
Chapter 4
A Multi-Hazard Framework for Optimum Life-Cycle Cost Design of
Reinforced Concrete Bridges................................................................................................................. 76
Azadeh Alipour, University of Massachusetts Amherst, USA
Behrouz Shafei, University of Massachusetts Amherst, USA
Masanobu Shinozuka, University of California, USA
Chapter 5
Efficient Robust Optimization of Structures Subjected to Earthquake Load and Characterized by
Uncertain Bounded System Parameters............................................................................................... 105
Subrata Chakraborty, Bengal Engineering and Science University, India
Soumya Bhattacharjya, Bengal Engineering and Science University, India
Chapter 6
Damage Assessment of Inelastic Structures under Simulated Critical Earthquakes........................... 128
Abbas Moustafa, Minia University, Egypt
Chapter 7
Metaheuristic Optimization in Seismic Structural Design and
Inspection Scheduling of Buildings..................................................................................................... 152
Chara Ch. Mitropoulou, Institute of Structural Analysis and Seismic Research, National
Technical University Athens, Greece
Vagelis Plevris, School of Pedagogical and Technological Education (ASPETE), Greece
Nikos D. Lagaros, Institute of Structural Analysis and Seismic Research, National Technical
University Athens, Greece
Chapter 8
Optimal Performance-Based Seismic Design...................................................................................... 174
Hamid Moharrami, Tarbiat Modares University, Iran
Chapter 9
Optimal Seismic Performance-Based Design of Reinforced Concrete Buildings............................... 208
Xiao-Kang Zou, AECOM Asia Company Ltd., Hong Kong
Chapter 10
Applications of Topology Optimization Techniques in Seismic Design of Structure......................... 232
Kazem Ghabraie, University of Southern Queensland, Australia
Chapter 11
Overall Conceptual Seismic Design and Local Seismic Capacity Design for
Components of Bridges........................................................................................................................ 269
Wan-Cheng Yuan, Tongji University, China
Yu-Guo Zheng, Tongji University and Hunan University of Science and Technology, China
Pak-Chiu Cheung, Tongji University, China
Chapter 12
Optimum Design of Structures for Earthquake Loading by a Cellular
Evolutionary Algorithm and Neural Networks.................................................................................... 306
Saeed Gholizadeh, Urmia University, Iran
Chapter 13
Fuzzy Identification of Seismically Excited Smart Systems............................................................... 323
JinSeop Kim, Massachusetts Institute of Technology, USA
Yeesock Kim, Worcester Polytechnic Institute, USA
Tahar El-Korchi, Worcester Polytechnic Institute, USA
Chapter 14
Health Assessment of Engineering Structures Using Graphical Models............................................. 342
Sankaran Mahadevan, Vanderbilt University, USA
Chapter 15
Optimal Design of Nonlinear Viscous Dampers for Protection of Isolated Bridges........................... 370
Alexandros A. Taflanidis, University of Notre Dame, USA
Ioannis G. Gidaris, University of Notre Dame, USA
Compilation of References................................................................................................................ 399
About the Contributors..................................................................................................................... 430
Index.................................................................................................................................................... 437
Detailed Table of Contents
Foreword..............................................................................................................................................xiii
Preface.................................................................................................................................................. xiv
Acknowledgment................................................................................................................................. xvi
Chapter 1
Life Cycle Cost Considerations in Seismic Design Optimization of Structures..................................... 1
Bora Gencturk, University of Illinois at Urbana-Champaign, USA
Amr S. Elnashai, University of Illinois at Urbana-Champaign, USA
The life-cycle cost (LCC) of a structure in seismic regions, which includes the initial and the post-
earthquake repair cost, is a critical parameter for structural engineers and other stakeholders. The LCC
analysis has been gaining prominence in recent years since civil infrastructure sustainability has been
identified as one of the grand challenges for engineering in the 21st century. The objective of this chapter
is to first identify the components in LCC evaluation that directly affect the outcomes, and propose strate-
gies to improve the reliability of the analysis. The shortcomings of existing studies on LCC optimization
of structures are identified. These shortcomings include simplified analysis techniques to determine the
structural capacity and earthquake demand, use of generalized definitions for structural limit states,
and inadequacies in treating uncertainty. In the following, the problem formulation and a brief review
of existing literature on LCC optimization of structures are provided. A LCC model is presented, and
techniques are proposed to improve the above mentioned shortcomings. Finally, LCC analysis of an
example reinforced concrete (RC) structure is employed to illustrate the methodology.
Chapter 2
Performance-Based Seismic Design: A Search-Based Cost Optimization with
Minimum Reliability Constraints.......................................................................................................... 23
Oscar Mller, University of Rosario, Argentina
Marcelo Rubinstein, University of Rosario, Argentina
Fabin Savino, University of Rosario, Argentina
Ricardo O. Foschi, University of British Columbia, Canada
An approach is presented to structural optimization for performance-based design in earthquake engineer-
ing. The objective is the minimization of the total cost, including repairing damage produced by future
earthquakes, and satisfying minimum target reliabilities in three performance levels (operational, life
safety, and collapse). The different aspects of the method are considered: a nonlinear dynamic structural
analysis to obtain responses for a set of earthquake records, representing these responses with neural
networks, formulating limit-state functions in terms of deformations and damage, calculating achieved
reliabilities to verify constraint violations, and the development of a gradient-free optimization algorithm.
Two examples illustrate the methodology: 1) a reinforced concrete portal for which the design parameters
are member dimensions and steel reinforcement ratios, and 2) optimization of the mass at the cap of a
pile, to meet target reliabilities for two levels of cap displacement. The objective of this latter example
is to illustrate model effects on optimization, using two different hysteresis approaches.
Chapter 3
Discrete Variable Structural Optimization of Systems under
Stochastic Earthquake Excitation.......................................................................................................... 51
Hctor Jensen, Universidad Tecnica Federico Santa Maria, Chile
Marcos Valdebenito, Universidad Tecnica Federico Santa Maria, Chile
Juan Seplveda, Universidad Tecnica Federico Santa Maria, Chile
Luis Becerra, Universidad Tecnica Federico Santa Maria, Chile
The reliability-based design optimization of structural systems under stochastic excitation involving discrete
sizing type of design variables is considered. The design problem is formulated as the minimization of an
objective function subject to multiple reliability constraints. The excitation is modeled as a non-stationary
stochastic process with uncertain model parameters. The problem is solved by a sequential approximate
optimization strategy cast into the framework of conservative convex and separable approximations. To
this end, the objective function and the reliability constraints are approximated by using a hybrid form
of linear, reciprocal, and quadratic approximations. The approximations are combined with an effective
stochastic sensitivity analysis in order to generate explicit expressions of the reliability constraints in
terms of the design variables. The explicit approximate sub-optimization problems are solved by an ap-
propriate discrete optimization technique. Two example problems that consider structures with passive
energy dissipation systems under earthquake excitation are presented to illustrate the effectiveness of
the approach reported herein.
Chapter 4
A Multi-Hazard Framework for Optimum Life-Cycle Cost Design of
Reinforced Concrete Bridges................................................................................................................. 76
Azadeh Alipour, University of Massachusetts Amherst, USA
Behrouz Shafei, University of Massachusetts Amherst, USA
Masanobu Shinozuka, University of California, USA
This chapter provides a comprehensive procedure for the time-dependant structural performance evalua-
tion and life-cycle cost analysis of reinforced concrete highway bridges located in extreme chloride-laden
environments. The penetration of chloride ions into the concrete is simulated through a finite difference
approach, which takes into account all the parameters that can affect the corrosion process. From simula-
tion results, the corrosion initiation time is predicted and the extent of structural degradation is calculated
over the entire life of bridge. A group of detailed bridge models with various structural attributes are
developed to evaluate the changes in the structural capacity and seismic response of corroded bridges.
For the purpose of the probabilistic seismic risk assessment of bridges, the seismic fragility curves are
generated and updated at regular time intervals. The time-dependent fragility parameters are employed to
investigate the life-cycle cost of bridges by introducing a performance index which combines the effects
of probable seismic events and chloride-induced corrosion. The proposed approach provides a multi-
hazard framework, which leads to more realistic performance and cost estimates. It also indicates the
inspection and maintenance intervals in a way that the inspection and maintenance costs are optimized,
while the safety of bridge is ensured.
Chapter 5
Efficient Robust Optimization of Structures Subjected to Earthquake Load and Characterized by
Uncertain Bounded System Parameters............................................................................................... 105
Subrata Chakraborty, Bengal Engineering and Science University, India
Soumya Bhattacharjya, Bengal Engineering and Science University, India
An efficient robust design optimization (RDO) procedure is proposed in the framework of an adap-
tive response surface method (RSM) for structures subjected to earthquake load and characterized by
uncertain but bounded system parameters. The basic idea of the proposed RDO approach is to improve
the robustness of a design by using a new dispersion index which utilizes the relative importance of the
gradients of the performance function. The same concept is also applied to the constraints. The repeated
computations of stochastic responses and their sensitivities for evaluating the stochastic constraint of
the associated optimization problem are efficiently obtained in the framework of an adaptive RSM. The
proposed RDO approach is elucidated through the optimization of a three-storied concrete frame struc-
ture. The numerical study depicts that the proposed RDO results are in conformity with the conventional
RDO results. However, definite improvements are achieved in terms of robustness and computational
time requirements indicating its efficiency over the conventional RDO approach.
Chapter 6
Damage Assessment of Inelastic Structures under Simulated Critical Earthquakes........................... 128
Damage of structures can be significantly reduced through robust prediction of possible future earth-
quakes that can occur during the life-time of the structure and through accurate modeling of the non-
linear behavior of the structure under seismic loads. Modern seismic codes specify natural records and
artificially generated ground accelerations as input to the nonlinear time-history analysis of the structure.
The advantage of using natural records is the inclusion of all important characteristics of the ground
motion (fault properties, path effects and local soil condition) in the design input. This option requires
selecting and scaling a set of proper accelerograms from the available records. However, the site under
consideration may have limited or scarce earthquake data. In such case, numerically simulated ground
motions can be employed as input to the dynamic analysis of the structure. This chapter deals with the
damage assessment of inelastic structures under numerically simulated critical earthquakes using non-
linear optimization, inelastic time-history analysis, and damage indices.
Chapter 7
Metaheuristic Optimization in Seismic Structural Design and
Inspection Scheduling of Buildings..................................................................................................... 152
Chara Ch. Mitropoulou, Institute of Structural Analysis and Seismic Research, National
Technical University Athens, Greece
Vagelis Plevris, School of Pedagogical and Technological Education (ASPETE), Greece
Nikos D. Lagaros, Institute of Structural Analysis and Seismic Research, National Technical
University Athens, Greece
Optimization is a field where extensive research has been conducted over the last decades. Many types
of problems have been addressed, and many types of algorithms have been developed, while their
range of applications is continuously growing. The chapter is divided into two parts; in the first part,
the life-cycle cost analysis is used as an assessment tool for designs obtained by means of prescriptive
and performance-based optimum design methodologies. The prescriptive designs are obtained through a
single-objective formulation, where the initial construction cost is the objective to be minimized, while
the performance-based designs are obtained through a two-objective formulation where the life-cycle
cost is considered as an additional objective also to be minimized. In the second part of the chapter, the
problem of inspection of structures and routing of the inspection crews following an earthquake in densely
populated metropolitan areas is studied. A model is proposed and a decision support system is developed
to aid local authorities in optimally assigning inspectors to critical infrastructures. A combined particle
swarm ant colony optimization based framework is implemented, which proves to be an instance of a
successful application of the philosophy of bounded rationality and decentralized decision-making for
solving global optimization problems.
Chapter 8
Optimal Performance-Based Seismic Design...................................................................................... 174
Hamid Moharrami, Tarbiat Modares University, Iran
In this chapter, the reader gets acquainted with the philosophy of performance-based design, its principles,
and an overview of the procedures for performance evaluation of structures. The essential prerequisites of
optimal performance-based design, including nonlinear analysis, optimization algorithms, and nonlinear
sensitivity analysis, are introduced. The methods of nonlinear analysis and optimization are briefly pre-
sented, and the formulation of optimal performance-based design with emphasis on deterministic type,
rather than probabilistic- (or reliability)-based formulation is discussed in detail. It is revealed how real
performance-based design is tied to optimization, and the reason is given for why, without optimization
algorithms, multilevel performance-based design is almost impossible.
Chapter 9
Optimal Seismic Performance-Based Design of Reinforced Concrete Buildings............................... 208
Xiao-Kang Zou, AECOM Asia Company Ltd., Hong Kong
In order to meet the emerging trend of the performance-based design approach and to improve the design
efficiency, this chapter presents a numerical optimization technique for both minimum material cost and
life-cycle cost design of building structures subject to multiple levels of linear elastic and nonlinear elastic
seismic drift performance design constraints. This chapter firstly introduces an elastic seismic drift design
of reinforced concrete (RC) building structures based on elastic response spectrum analysis method;
and then presents the inelastic design optimization based on the nonlinear pushover analysis method.
Finally, the optimal seismic performance-based design of RC buildings is posed as a multi-objective
optimization problem in which the life-cycle cost of a building is to be minimised subject to multiple
levels of seismic performance design criteria. The computer based optimization methodology developed
provides a powerful numerical design tool for performance-based design of RC building structures.
Chapter 10
Applications of Topology Optimization Techniques in Seismic Design of Structure......................... 232
Kazem Ghabraie, University of Southern Queensland, Australia
During the last two decades, topology optimization techniques have been successfully applied to a wide
range of problems including seismic design of structures. This chapter aims to provide an introduction
to the topology optimization methods and a review of the applications of these methods in earthquake
engineering. Two well-established topology optimization techniques are introduced. Several problems
including eigenfrequency control of structures, compliance minimization under periodic loading, and
maximizing energy absorption of passive dampers will be addressed. Numerical instabilities and ap-
proaches to overcome them will be discussed. The application of the presented approaches and methods
will be illustrated using numerical examples. It will be shown that in seismic design of structures, topol-
ogy optimization methods can be useful in providing conceptual design for structural systems as well
as detailed design of structural members.
Chapter 11
Overall Conceptual Seismic Design and Local Seismic Capacity Design for
Components of Bridges........................................................................................................................ 269
Wan-Cheng Yuan, Tongji University, China
Yu-Guo Zheng, Tongji University and Hunan University of Science and Technology, China
Pak-Chiu Cheung, Tongji University, China
From the perspective of overall conceptual seismic design, four design strategies are presented to de-
crease and balance the seismic force and displacement demands for some bridges working in a linear and
elastic state: the adjustment of the layout and detail of piers and expansion joints for a typical long span
continuous girder bridge, the adoption of a new-type spatial bridge tower for a long span cable-stayed
bridge, the study on the isolation mechanism of an elastic cable seismic isolation device for another
cable-stayed bridge, and the study on the seismic potential and performance for long span SCC (steel-
concrete composite) bridges. From the perspective of local seismic capacity design, three earthquake
resistant strategies are presented to achieve economical, applicable, and valid seismic design of local
components of bridges working in a nonlinear state: the adoption and the study on a new cable sliding
friction aseismic bearing, the study on the seismic capacities of single-column bridge piers wholly and
locally reinforced with steel fiber reinforced concrete (SFRC), the study on the seismic capacities, and
the hysteretic performance and energy dissipation capabilities of bridge pile group foundations strength-
ened with the steel protective pipes (SPPs). Research results show that these seismic design strategies
are effective to improve the seismic performance of bridges.
Chapter 12
Optimum Design of Structures for Earthquake Loading by a Cellular Evolutionary Algorithm and
Neural Networks.................................................................................................................................. 306
Saeed Gholizadeh, Urmia University, Iran
The present chapter deals with optimum design of structures for earthquake induced loads by taking
into account nonlinear time history structural response. As the structural seismic optimization is a time
consuming and computationally intensive task, in this chapter, a methodology is proposed to reduce the
computational burden. The proposed methodology consists of an efficient optimization algorithm and a
hybrid neural network system to effectively predict the nonlinear time history responses of structures.
The employed optimization algorithm is a modified cellular genetic algorithm which reduces the required
generation numbers compared with the standard genetic algorithm. Also, the hybrid neural network
system is a combination of probabilistic and generalized regression neural networks. Numerical results
demonstrate the computational merits of the proposed methodology for seismic design optimization of
structures.
Chapter 13
Fuzzy Identification of Seismically Excited Smart Systems............................................................... 323
JinSeop Kim, Massachusetts Institute of Technology, USA
Yeesock Kim, Worcester Polytechnic Institute, USA
Tahar El-Korchi, Worcester Polytechnic Institute, USA
In this chapter, a nonlinear modeling framework to identify nonlinear behavior of smart structural sys-
tems under seismic excitations is proposed. To this end, multi-input-multi-output (MIMO) autoregres-
sive exogenous (ARX) input models and Takagi-Sugeno (TS) fuzzy models are coalesced as the MIMO
ARX-TS fuzzy model. The premised part of the proposed MIMO ARX-TS fuzzy model is optimized
using the hierarchical clustering (HRC) algorithm, while its consequent parameters are optimized via the
weighted linear least squares estimation. The performance of the proposed model is investigated using
the dynamic response of a three-story shear planer frame structure equipped with a magnetorheological
(MR) damper subject to earthquake disturbances. Furthermore, the impact of the HRC algorithm on
the performance of the MIMO ARX-TS fuzzy model is compared with that of the subtractive and the
fuzzy C-means clustering algorithms. The equivalence of the original and identified data is numerically
shown to prove that the HRC MIMO ARX-TS fuzzy model introduced here is effective in estimating
nonlinear behavior of a seismically excited building-MR damper system.
Chapter 14
Health Assessment of Engineering Structures Using Graphical Models............................................. 342
Sankaran Mahadevan, Vanderbilt University, USA
A hybrid qualitative-quantitative health assessment of structures using the bond graph theory is presented
in this chapter. Bond graph (BG) is an energy-based graphical-modeling tool for physical dynamic
systems, actuators, and sensors. BG provides domain-independent framework for modeling dynamic
systems with interacting components from multiple domains. Discrete structures are modeled using one-
to-one bond graph elements, while continuous structures are modeled using finite-mode bond graphs.
BG facilitates the construction of temporal causal graph (TCG) that links the system response to the
damaged component or faulty sensor. TCG provides qualitative damage isolation, which is not possible
using most existing system identification techniques. This leads to rapid isolation of damage and sig-
nificant reduction in computations. Quantitative identification of damage size is performed by analyzing
the substructure containing the damaged component, using the nonlinear least-squares optimization
technique, thus reducing the computations. The health assessment algorithm developed in this chapter
combines the Generic Modeling Environment (GME), the Fault Adaptive Control Technology (FACT)
software, and Matlab Simulink. Numerical illustrations on BG modeling of a hydraulic actuator and
system identification of a fifteen-story shear building and a high-rise structure under earthquake loads
are provided.
Chapter 15
Optimal Design of Nonlinear Viscous Dampers for Protection of Isolated Bridges........................... 370
Alexandros A. Taflanidis, University of Notre Dame, USA
Ioannis G. Gidaris, University of Notre Dame, USA
A probabilistic framework based on stochastic simulation is presented in this chapter for optimal design
of supplemental dampers for multi - span bridge systems supported on abutments and intermediate
piers through isolation bearings. The bridge model explicitly addresses nonlinear characteristics of the
isolators and the dampers, the dynamic behavior of the abutments, and the effect of pounding between
the neighboring spans to each other as well as to the abutments. A probabilistic framework is used to
address the various sources of structural and excitation uncertainties and characterize the seismic risk
for the bridge. Stochastic simulation is utilized for evaluating this seismic risk and performing the as-
sociated optimization when selecting the most favorable damper characteristics. An illustrative example
is presented that considers the design of nonlinear viscous dampers for protection of a two-span bridge.
Compilation of References................................................................................................................ 399
About the Contributors..................................................................................................................... 430
Index.................................................................................................................................................... 437
xiii
Foreword
Earthquake engineering and aseismic structural design for many years was based on conventional meth-
ods of analysis, statistical description of earthquakes and very much engineering experience. Design
engineers smugly say that the medium quality of ordinary structures, the low or uncontrollable quality
of materials and the lack of guaranteed, maintenance in most parts of the world, did not justify the use
of more sophisticated methods. On the other hand modern methods for the optimal design of structures,
reliability analysis, and the most advanced methods of computational mechanics found their way into,
mainly, mechanical engineering applications. Integration of electronics and control, resulting in the so-
called smart structures, is another area with fruitful results. The existence of sensors allows us to think
on structural health monitoring and much more. All these techniques stayed away from the seismic
design of structures, although the safety of structures in which most of the population in earthquake-
affected countries lives and works is of obvious importance.
I am happy to see that this gap is gradually closing. Technology and knowledge becomes available to
broader parts of the population and is being applied to more classical fields. The present edited volume
contributes towards this direction. By studying selected topics of structural optimization, reliability-
based, performance-based, robust design and life cycle cost structural analysis the authors of the book
clearly demonstrate that sophisticated structural analysis and design tools slowly find their way in the
design of ordinary structures. Both users of buildings and infrastructure and professional engineers will
benefit from this development.
Georgios E. Stavroulakis
Technical University of Crete, Greece
xiv
Preface
Since the early seventies, optimization has been the subject of intensive research and several different
approaches have been advocated for the optimal design of structures in terms of optimization methods
or problem formulations. Most of the attention of the engineering community has been directed towards
the optimum design of structures with the assumption of linear elastic structural behavior under static
loading conditions. However, for a large number of real-life structural design problems, the assumption
of linear response may lead to vulnerable structural configurations, while seismic loading has also to be
taken into account in order to design earthquake resistant real-world structures. Parts of the book are
devoted to the formulation of design optimization frameworks in the field of earthquake engineering.
In recent years, probabilistic-based formulations of the optimization problem have been developed
in order to account for the uncertainty and randomness. The development of stochastic analysis methods
over the last two decades has stimulated the interest for the probabilistic optimum design of structures.
There are two distinguished design formulations that account for the probabilistic system response:
Robust Design Optimization (RDO) and Reliability-Based Design Optimization (RBDO). RDO meth-
ods primarily seek to minimize the influence of stochastic variations on the nominal dimensions and
values of the material properties of the design. On the other hand, the main goal of RBDO methods is
to design for minimum weight, while satisfying the limit on the allowable probability of failure. Parts
of the book are devoted to the formulation of optimization problems considering uncertainties under
earthquake loading conditions.
The research fields described above define hot topics of Earthquake Engineering involving the use
of advanced optimization tools. The basic idea of this book is to include all the aforementioned research
topics into a volume taking advantage of the connecting link between them, which is optimization. In
this direction, the book consists of 15 chapters in total. In the first chapter, the components in life-cycle
cost (LCC) evaluation are identified that directly affect the outcomes and propose strategies to improve
the reliability of the analysis, while the shortcomings of existing studies on LCC optimization of struc-
tures are identified.
In the second chapter an approach is presented to structural optimization for performance-based
design in earthquake engineering where the objective is the minimization of the total cost, including
repairing damage produced by future earthquakes, and satisfying minimum target reliabilities in three
performance levels. In the next chapter the reliability-based design optimization of structural systems
under stochastic excitation involving discrete sizing type of design variables is considered; while it is
formulated as the minimization of an objective function subject to multiple reliability constraints.
In the fourth chapter, a comprehensive procedure for the time-dependent structural performance
evaluation and life-cycle cost analysis of reinforced concrete highway bridges located in extreme
xv
chloride-laden environments is provided. The chapter also indicates the inspection and maintenance
intervals in a way that the inspection and maintenance costs are optimized while the safety of bridge is
ensured. In the next chapter, an efficient RDO procedure is proposed in the framework of an adaptive
response surface method for structures subjected to earthquake load and characterized by uncertain but
bounded system parameters.
The sixth chapter deals with the problem of damage assessment of inelastic structures under nu-
merically simulated critical earthquakes, implementing nonlinear optimization, inelastic time-history
analysis, and damage indices. In the next chapter the life-cycle cost analysis is used as an assessment
tool for designs obtained by means of prescriptive and performance-based optimum design methodolo-
gies; furthermore, the problem of inspection of structures and routing of the inspection crews following
an earthquake in densely populated metropolitan areas is studied.
In the eighth chapter, the reader gets acquainted with the philosophy of performance-based design,
its principles, and an overview of the procedures for performance evaluation of structures, while the
essential prerequisites of optimal performance-based design, including nonlinear analysis, optimization
algorithms, and nonlinear sensitivity analysis, are also introduced.
The ninth chapter presents a numerical optimization technique for both minimum material cost and
life-cycle cost design of building structures subject to multiple levels of linear elastic and nonlinear
elastic seismic drift performance design constraints. The next chapter aims at providing an introduction
engineering.
In the eleventh chapter, the overall conceptual seismic design and the local seismic capacity design
methods are proposed to give clear and correct directions for seismic design optimization of bridges
in order to obtain uniform and rational seismic demands and improved seismic capacities of structural
components. The next chapter deals with the optimum design of structures for earthquake induced loads
by taking into account the nonlinear time history structural response.
In the thirteenth chapter, a nonlinear modeling framework to identify nonlinear behavior of smart
structural systems under seismic excitations is presented; to this end, multi-input-multi-output autore-
gressive exogenous input models and Takagi-Sugeno fuzzy models are coalesced.
In the fourteenth chapter of the book, a hybrid qualitative-quantitative health assessment of struc-
tures using the bond graph theory is presented. In the last chapter, a probabilistic framework based on
stochastic simulation is presented for optimal design of supplemental dampers for multi-span bridge
systems supported on abutments and intermediate piers through isolation bearings.
Vagelis Plevris

Nikos D. Lagaros
xvi
Acknowledgment
The editors of the book would like to express their deep gratitude to all the contributors for their most
valuable support during the preparation of this volume, for their time and effort devoted to the comple-
tion of their contributions, and for their great expert help in the review process. In addition, we are most
appreciative to the members of the Editorial Advisory Board (EAB) of the book for their constructive
comments and suggestions offered during the preparation process and for their contribution in reviewing
some of the book chapters. We are also grateful to all the colleagues who, although they did not con-
tribute chapters to the book, were kind enough to offer their expert help during the review process. Fi-
nally, we would also like to thank all the personnel of IGI Global Publishers, especially Myla R. Merkel,
Lindsay Johnston, Emily E. Golesh, Erika L. Carter, and Kristin M. Klinger, for their most valuable
continuous support for the publication of this book.
Vagelis Plevris

Nikos D. Lagaros
1
Chapter 1
Life Cycle Cost Considerations
in Seismic Design
Optimization of Structures
Bora Gencturk
University of Illinois at Urbana-Champaign, USA
Amr S. Elnashai
University of Illinois at Urbana-Champaign, USA
ABSTRACT
The life-cycle cost (LCC) of a structure in seismic regions, which includes the initial and the post-
earthquake repair cost, is a critical parameter for structural engineers and other stakeholders. The
LCC analysis has been gaining prominence in recent years since civil infrastructure sustainability has
been identified as one of the grand challenges for engineering in the 21st century. The objective of this
chapter is to first identify the components in LCC evaluation that directly affect the outcomes, and pro-
pose strategies to improve the reliability of the analysis. The shortcomings of existing studies on LCC
optimization of structures are identified. These shortcomings include simplified analysis techniques to
determine the structural capacity and earthquake demand, use of generalized definitions for structural
limit states, and inadequacies in treating uncertainty. In the following, the problem formulation and
a brief review of existing literature on LCC optimization of structures are provided. A LCC model is
presented, and techniques are proposed to improve the above mentioned shortcomings. Finally, LCC
analysis of an example reinforced concrete (RC) structure is employed to illustrate the methodology.
INTRODUCTION zation of the properties of individual elements to

find the optimal solution. In shape optimization,
Structural optimization problems may be divided the contour of the boundary of a structural domain
into three classes: sizing, shape and topology opti- is optimized while keeping the connectivity of
mization. In sizing optimization, the locations and the structure the same, in other words, no new
number of the structural elements are fixed and boundaries are formed. Topology optimization
known. Usually, the problem is reduced to optimi- is the most general in the sense that both the size
DOI: 10.4018/978-1-4666-1640-0.ch001
Copyright 2012, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.
Life Cycle Cost Considerations in Seismic Design Optimization of Structures
and location of structural members are determined Sahab et al. (2005). As a consequence of the
and formation of new boundaries are allowed. The formulation of the problem, the majority of the
number of joints in the structure, the joint support single-objective optimization methods provide a
locations, and number of members connected to single optimal solution (which minimizes the
each joint are unknown. In other words, topology objective and satisfies the constraints). However,
and shape of the structure are optimized in addi- the decision maker does not have a broad view
tion to the shape of individual elements. Figure of to what extend the constraints are satisfied.
1 illustrates the sizing and topology optimization Thus, she has to either accept or reject the optimal
problems for a structural frame. The structure solution. On the other hand, since more than one
being optimized may vary significantly from a objective is considered in multi-objective optimi-
component of a mechanical device to a member zation problems, commonly a set of equivalently
of a full-scale structure depending on the problem. optimal solutions are obtained which provides the
Most problems in structural earthquake engineer- decision maker the flexibility to tradeoff between
ing fall into the category of sizing optimization the solutions and she may base her selection on
and deal with full-scale structures. rather transparent results. Some example multi-
The problems in structural optimization may objective studies include Li et al. (1999) and Liu
also be classified based on the number of objec- et al. (2006).
tives, i.e. single- and multi-objective. For single- It is natural to include multiple objectives (e.g.
objective seismic design optimization problems, structural performance, initial and life-cycle costs)
the objective is usually selected as the initial cost in the LCC optimization of structures; however,
for reinforced concrete (RC) and total weight for the number of objectives is not usually the only
steel structures; and the performance is determined difference between LCC and initial cost optimiza-
based on the conformance to the requirements of tion problems. The former necessitates the use of
a seismic design code. The code provisions are probabilistic formulations to evaluate the failure
usually introduced to the problem as constraints. probability at different limit states in addition to
Some example studies include Moharrami and the derivation of the probabilistic seismic hazard
Grierson (1993), Adamu et al. (1994), Memari at different intensities. Whereas, in the latter
and Madhkhan (1999), Zou and Chan (2005) and the evaluation of the structural performance at
Figure 1. Example illustration of (a) sizing, and (b) topology optimization problems
2
predefined hazard level(s) is sufficient. A brief The second item is critical to the evaluation of
review of existing literature on LCC optimization LCC in that the repair cost (after an earthquake
of structures is provided in the following section. event) is more directly related to the local behavior,
The objective of this chapter is to emphasize the rather than the global. There is existing literature
importance of the following in LCC optimization: on relating the global structural response to local
parameters (an example is predicting the damage
1. Use of advanced analysis, which provides state of a vertical member based on the interstory
the most rigorous assessment of structural drift); however, it is not accurate to generalize
capacity and earthquake demand, these relationships to structural configurations
2. Evaluation of the structural capacity (that other than that they are derived for. Therefore,
has direct impact on the LCC) by taking in order to accurately evaluate the LCC proper
into account not only the global behavior response metrics need to be chosen.
of the structure but also the local response, As for the third item, the capacity of a structure
such as reinforcement yielding and concrete depends on various factors, most importantly on
crushing, the force resisting system employed in design.
3. Use of system-specific limit states (rather The limit state values for a bearing wall system
than fixed value or generic limit states) to will be significantly different from those for a
define the structural capacity, frame system. It is simpler to use generic limit
4. Consideration of all major sources of states, which also reduces computational demand;
uncertainty, from seismogenic source however, accurate evaluation of the structural ca-
characteristics to material properties and pacity is key to seismic design and it warrants full
structural modeling in calculating the limit consideration. In other words, structural capacity
state exceedance probabilities. has to be evaluated specifically for the considered
structural configuration.
The majority of existing work on structural op- Finally, for the fourth item, the sources of
timization uses either elastic dynamic or nonlinear uncertainty have to be defined clearly. The uncer-
static analysis for seismic performance assessment tainty in exceedance probabilities of the structural
of structures. In cases where inelasticity is mod- damage states is mainly governed by the vari-
eled, lumped plasticity models are adopted. The ability in ground motion processes; nevertheless,
first item in the list above is significant because it is required to take into account the variability
the oversimplification of the structural assess- in material properties, and other capacity related
ment, even though the optimization framework is parameters.
robust and sound, might yield unrealistic results. In the following, first the background informa-
This chapter highlights the importance of utiliz- tion on structural optimization that will allow the
ing distributed inelasticity approach using which reader to follow the rest of the chapter is provided.
structural capacity and earthquake demand are Then a framework for LCC optimization of struc-
evaluated through nonlinear pushover analysis and tures that includes the definition of seismic hazard,
inelastic dynamic time history analysis, respec- evaluation of structural capacity and earthquake
tively. Widespread use of optimization tools for demand, LCC model and optimization algorithm,
seismic design of structures can only be achieved is proposed. Finally, the framework is applied to
by having practical and reliable approaches that an example RC structural frame.
can predict the structural performance with rea-
sonable accuracy.
3
BACKGROUND ON of a structure. Each hazard level is usually

STRUCTURAL OPTIMIZATION directly mapped to a single or multiple per-
formance levels.
In this section, first the terminology that is used Space of Design (decision) Variables
in the chapter is described. Then an overview of or Search Space: The boundaries of the
the most commonly used optimization algorithms, search space are defined by the ranges of
with a focus on the advantages and shortcomings, the design variables. The dimension of the
is provided. Finally, existing work on LCC opti- search space is equal to the number of de-
mization of structures is briefly reviewed. sign variables in the problem. The search
space can be continuous for continuous
Terminology design variables and discrete for discrete
design variables or certain dimensions can
Objective (merit) Function: A function be continuous and the rest can be discrete.
that measures the performance of design. Solution (objective function) Space:
For every possible design, the objective Usually the solution space is unbounded or
function takes a different value. Examples semi-bounded. The dimension of the solu-
include the maximum interstory drift and tion space is equal to the number of objec-
initial cost. tive functions in the optimization problem.
Design (decision) Variables: A vector that The optimal solution(s) is defined in the
specifies design. Each element in the vec- solution space. The set of optimal solu-
tor describes a different structural property tions in the solution space is referred to as
that is relevant to the optimization prob- Pareto-front or Pareto-optimal set, as de-
lem. The design variables take different scribed below.
values throughout the optimization pro- Pareto-Optimality: To define Pareto-
cess. Examples include section dimensions optimality, consider the function
and reinforcement ratios. f : which assigns each point, x
k l
Performance Levels: Predefined levels in the space of decision variables to a point,

that describe the performance of the struc- y = f ( x ) in the solution space. Here f
ture after an earthquake. Usually the fol- represents the objective functions, k is the
lowing terminology is used to define the number of decision variables and l is the
damage state (performance) of the struc- number of objective functions to assess the
ture: immediate occupancy, life safety and performance of each, x (or equal to the di-
collapse prevention. Occurrence of each mension of f). The Pareto-optimal set of
damage state is determined based on the solutions is constructed by comparing the
exceedance of a threshold value in terms points in the solution space based on the
of structural capacity. following definition: a point y in the solu-
Hazard Levels: Predefined levels used to tion space strictly dominates another point
describe the earthquake intensity that the y if each element of y is less than equal to
structure is subjected to. Hazard levels are the corresponding parameter of y (that is
usually described by earthquake return yi yi ) and at least one element, i*, is
periods (or annual frequencies of exceed-
strictly less (that is yi * < yi * ), assuming
ance) and represented by acceleration re-
that this is a minimization problem. Thus,
sponse spectrum. It is required to consider
the Pareto-front is the subset of points in
multiple levels of hazard to calculate LCC
4
the set of Y = f(X), that are not strictly computationally efficient, the problems that could
dominated by another point in Y. The be solved are limited due to fact that the objec-
Pareto-optimality is illustrated in Figure 2, tive functions, constraints and their sensitivities
the plot is in the solution space, and the fig- should be expressed explicitly in terms of the
ure axes are two objective functions, f1 and design variables. The latter condition imposes a
f2. Assuming that the objective is minimi- restraint on the analysis procedure that could be
zation of both f1 and f2 the Pareto-front lies selected for structural assessment. Furthermore,
at the boundary that minimizes both objec- the design variables should be continuous over
tives as shown in the figure. the search domain. In other words, discrete de-
sign variables such as the reinforcement ratio in
RC structures cannot directly be accounted for
Review of Most Commonly Used in gradient-based optimization algorithms. As a
Optimization Algorithms consequence, researchers used methods such as
the principle of virtual work in order to express
The algorithms used in structural optimization the objective function as well as the constraints
may be divided into two categories: gradient- analytically (e.g. Chan & Zou, 2004).
based and heuristic approaches. The earlier studies With the immense increase in available com-
on structural optimization utilized conventional putational power in the recent years, researchers
gradient-based algorithms to obtain the optimal started to incorporate more accurate analysis tools
solution (e.g. Bertero & Zagajeski, 1979; Cheng in structural optimization such as the static push-
& Truman, 1985; Pezeshk, 1998). The most over and dynamic time history analysis through
commonly used algorithms include optimality finite element modeling. However, due to the fact
criteria, linear and nonlinear programming, fea- that these analysis techniques required algorithms
sible directions, and state-space steepest descent. that do not entail the continuity of functions as
Although the gradient-based approaches are well as the existence of derivatives, researchers
Figure 2. Illustration of Pareto-optimality
5
had to abandon the conventional gradient-based provisions were followed to determine the validity
approaches. Heuristic optimization algorithms of design alternatives. Static pushover analysis
progressively became popular in structural opti- was used to derive an equivalent SDOF system
mization (Foley & Schinler, 2003; M. Fragiadakis, which was utilized in computing the maximum
Lagaros, & Papadrakakis, 2006b; N. D. Lagaros interstory drift ratios. Liu et al. (2004) studied
& Papadrakakis, 2007; Min Liu, et al., 2006). The the PBSD of steel moment-resisting frames us-
most commonly used approaches include genetic ing GA. Three merit functions were defined: the
algorithms (GA), simulated annealing (SA), taboo initial material and lifetime seismic damage costs,
search (TS), and shuffled complex evolution. A and the number of different steel section types.
further advantage of the heuristic algorithms is Maximum interstory drift was used for the per-
that they are very effective in terms of finding formance assessment of the frames through static
the global minimum of highly nonlinear and/or pushover analysis. Code provisions were taken
discontinuous problems where the gradient-based into account in design. Different sources of uncer-
algorithms are usually trapped at a local minimum. tainty in estimating seismic demand and capacity
were incorporated into analysis using the SAC/
Review of Studies on LCC FEMA guidelines (Cornell, Jalayer, Hamburger,
Optimization of Structures & Foutch, 2002). The results were presented as
Pareto-fronts for competing merit functions. Final
LCC analysis of structures has gained importance designs obtained from the optimization algorithm
in the recent years due to concerns related to were assessed using inelastic dynamic time his-
the civil infrastructure sustainability; therefore, tory analysis. Liu (2005) formulated an optimal
studies on LCC optimization of structures are design framework for steel structures based on the
relatively new compared to the rest of the lit- PBSD. The considered objectives were the mate-
erature on structural optimization. Below, a brief rial usage, initial construction expenses, degree
review of existing studies on LCC optimization of design complexity, seismic structural perfor-
is provided. The section is not intended to be mance and lifetime seismic damage cost. Design
comprehensive; the goal is to highlight some of variables were section types for members of the
the critical components of seismic LCC analysis frames. The validity of designs was performed
that are addressed in more detail in this chapter. based on the existing code provisions. A lumped
Wen and Kang (2001a) developed an analyti- plasticity model was used for structural modeling.
cal formulation to evaluate the LCC of structures Both static pushover and inelastic dynamic (only
under multiple hazards. The methodology is then when structural response parameters were directly
applied to a 9-story steel building to find the taken as objective functions) analysis were used.
minimum LCC under earthquake, wind and both Fragiadakis et al. (2006b) used evolutionary strate-
hazards (Wen & Kang, 2001b). In this study, the gies for optimal design of steel structures. Initial
simplified method based on an equivalent single- construction and life cycle costs were considered
degree-of-freedom (SDOF) system developed by as the two merit functions. The constraints were
Collins et al. (1996) was used for structural assess- based on the provisions of the European design
ment. Liu et al. (2003) investigated the optimal codes. A fiber-based finite element model was used
performance-based seismic design (PBSD) of to conduct static pushover analysis to determine
steel moment frame structures. Merit functions the inelastic response of structures. Deterministic
were selected as the initial material and lifetime structural damage states based on the maximum
seismic damage costs. Reducing design complex- interstory drift was employed; however, probabi-
ity was also incorporated in the algorithm. Code listic formulations were adopted for calculating
6
the life cycle costs. Zou et al. (2007) used the code as a basis. Steel structures were considered
method of virtual work to achieve an explicit and evolutionary strategies were used to solve the
formulation for multi-objective optimization of optimization problem. A fiber-based finite element
RC frames. Then, optimality criteria method was modeling approach was adopted. Either 10 natural
used to minimize the initial material cost and the or 5 artificial records were used to represent the
expected damage loss in a Pareto optimal sense. hazard. Material weight was selected as the design
The method was formulated in two stages: first objective. It was observed that lighter structures
elastic response spectrum analysis was performed could be obtained when inelastic dynamic time
where the cross-sectional dimensions were consid- history analysis (instead of elastic dynamic time
ered as the only design variables, second section history or modal analysis) and natural records
sizes were kept constant and the reinforcement (instead of artificial records that were compatible
ratio was taken as the design variable for the with a certain design spectrum) were used. Fragia-
static pushover analysis through which inelastic dakis and Papadrakakis (2008) studied the optimal
drift responses were calculated. design of RC structures. Both deterministic and
Although, LCC was not considered in the fol- reliability-based approaches were evaluated and
lowing studies, these studies are mentioned here the latter was found to provide more economical
because they are related to methods or derivations solutions as well as more flexibility to designer.
in this chapter by considering uncertainty, multiple The total cost of the structure was taken as the
hazard levels or advanced structural analysis to objective function and compliance with Euro-
obtain the seismic demand on structures. In Liu pean design codes was applied as a condition.
et al. (2005) conflicting objectives were defined Evolutionary strategies were used to solve the
as the initial material cost (including the cost due optimization problem. Three hazard levels were
to design complexity as a function of the number considered. To reduce the computational time,
of different structural shapes) and the seismic fiber-based beam-column elements were used only
performance. Two hazard levels were used and at the member ends and inelastic dynamic analysis
the performance criterion was selected as the was performed only if non-seismic checks per-
maximum interstory drift. Structural assessment formed through a linear elastic analysis were met.
was conducted using static pushover analysis LCC oriented seismic design optimization of
determined from seismic code provisions. GA structures is a relatively new subject, a review of
were employed to solve the optimization prob- existing literature, as provided above, indicates
lem. Fragiadakis et al. (2006a) used evolutionary that there is still need for further research due to
strategies for optimal PBSD of steel structures. following: (1) only a limited number of studies
Minimization of cost subjected to constraints on utilized advanced computational tools, structural
interstory drift was targeted. Both inelastic static performance assessment was usually performed
and inelastic dynamic analysis were employed. using code-based formulations or elastic analysis,
Discrete beam and column sections were selected simplified modeling techniques were adopted
as design variables. 10 earthquake records were whenever inelastic analysis was conducted; (2)
used to represent each hazard level and mean existing studies overwhelmingly focused on the
structural response was taken as the performance optimization of steel structures due to well defined
measure. Uncertainty associated with structural design variables (i.e. section types) and availabil-
modeling was also taken into account. Lagaros et ity of structural modeling tools; (3) most of the
al. (2006) evaluated modal, elastic and inelastic research effort was devoted to the development
dynamic time history analysis in an optimization of optimization methods; the real engineering
framework taking the European seismic design problem to be solved remained faint.
7
In the following section a LCC formulation used intensity measures in seismic design are the
is developed for optimal seismic design of RC peak ground acceleration (PGA) and spectral ac-
structures. Fiber-based finite element modeling celeration (Sa) at the fundamental period of the
is used to construct the structural models. In- structure. Example hazard curves for PGA and Sa
elastic dynamic time history analysis and static are shown in Figure 3(a). These curves are site
pushover analysis are used to accurately obtain specific and they can be obtained from PSHA for
the earthquake demand and structural capacity, re- seismic design purposes.
spectively. In addition, randomness due to ground For certain types of structural analysis such
motion variability (aleatory uncertainty) and errors as modal response spectrum and time history
in structural modeling and inherent variability in analysis, it is required to obtain the design spec-
material properties (epistemic uncertainty) are trum. The design spectrum could be obtained from
taken into account. the hazard curves depicted in Figure 3(a) at dif-
ferent return periods. The return period of an
earthquake, TR is simply the inverse of the mean
LIFE-CYCLE COST FORMULATION annual frequency of exceedance, v, i.e. TR = 1/v.
As examples, design spectra at different return
Definition of the Seismic Hazard periods obtained from the hazard curves shown
in Figure 3(a) are given in Figure 3(b). These are
In order to evaluate the LCC of a structure due to also referred to as uniform hazard spectra (UHS)
repair in future earthquakes, one has to evaluate because each spectral ordinate has the same prob-
the probability of demand exceeding capacity for ability of exceedance. As an alternative, the design
the whole-life time of the structure. There is more spectra could be obtained from seismic design
than one way of defining the seismic hazard that a codes. Although, most codes provide the necessary
structure could experience throughout its life time. information to draw design spectra for the maxi-
The most commonly used methods to evaluate mum considered earthquake only (e.g. ICC, 2006)
the earthquake hazard for a given location with and it is not possible obtain the hazard at different
known historical seismicity are the deterministic return periods which is required for LCC analysis,
and probabilistic seismic hazard analysis (DSHA more detailed seismic zonation maps have been
and PSHA). The main difference between the two developed in the recent years and design spectra
approaches is that PSHA incorporates the ele- for earthquake hazards with different probabilities
ment of time in hazard assessment. It is beyond of occurrence are now available (USGS, 2008).
the scope of this chapter to assess and validate
the two approaches for seismic hazard analysis; Evaluation of the Structural
however, PSHA has well made its way into the Capacity and Earthquake Demand
seismic design codes. As examples, UBC (ICBO,
1997) and FEMA 450 (FEMA, 2003) represented As described in the previous section, the continu-
design response spectrum based on probabilistic ous probabilistic hazard at a selected site, Figure
zonation maps. Therefore, here PSHA is used to 3(a), is discretized at certain probabilities, repre-
characterize the seismic hazard for a selected site, sented with return period in Figure 3(b). The objec-
while DSHA is also a valid option. tive behind this manipulation is to have a means
In PSHA mean annual frequency of exceedance of evaluating the probability of failure; that is the
(or probability of exceedance) is calculated for a probability of earthquake demand exceeding the
range of a selected intensity measure that repre- structural capacity. By having the design spectra
sents the earthquake hazard. The most commonly at different return periods, the hazard levels are
8
Figure 3. (a) Example hazard curves for different intensity measures obtained from PSHA, (b) design
spectra at different return periods obtained from PSHA (YRP: years return period)
very well defined. The more challenging part is levels were determined by the condition of both
the definition of structural limit states and their the structural and nonstructural components.
correspondence to hazard levels. Studies on PBSD Four performance levels were defined: fully
try to answer a similar question. operational, operational, life-safety, and near
As opposed to traditional seismic codes that aim collapse. Damage states of structural components
to provide structures with adequate strength and were mapped to performance levels. Finally, a
ductility for life safety, and stiffness for service- design performance objective was described as
ability limits states, PBSD is a broader approach the desired performance level for the building for
where the objective is to achieve stated perfor- each earthquake design level. Design performance
mance objectives (levels) when the structures are objectives were dependent on the buildings oc-
subjected to stated seismic hazard levels. Using the cupancy, the importance of functions occurring
multilevel (in terms of hazard) and multi-criteria within the building, economic considerations
(in terms of performance) approach offered by the related to repair due to building damage and busi-
PBSD it is aimed that the structural design will be ness interruption, and importance of the building
under direct and explicit control and the expecta- as a historical or cultural asset. Recommended
tions of stakeholders for more explicit codes for design performance objectives for buildings were
defining design objectives will be fulfilled. This mapped onto earthquake design levels as shown
is the very same reason for PBSD lending itself in Figure 4.
for use in LCC optimization of structures. In FEMA 273 (FEMA, 1997) a similar frame-
Vision 2000 (SEAOC, 1995) has been one of work to that of Vision 2000 (SEAOC, 1995) was
the key documents in the development of PBSD presented. However, different design performance
concepts. In Vision 2000, the hazard levels were objectives and earthquake design levels were
expressed in terms of return periods as shown in adopted (Figure 5). Threshold values for struc-
Table 1. A performance level was defined as the tural and nonstructural components at various
maximum allowable damage to a building for a performance levels were tabulated for various
given earthquake design level. The performance building types including steel, RC, masonry and
9
Table 1. Earthquake design levels in Vision 2000

ity (a system level indicator) and to the amount
(SEAOC, 1995)
of local deformation imposed on the vertical ele-
Earthquake Return Period Probability of ments and beam-column connections (component
Design Level Exceedance level indicators).
Frequent 43 years 50% in 30 years In this study the structural performance is de-
Occasional 72 years 50% in 50 years fined in three levels: immediate occupancy (IO),
Rare 475 years 10% in 50 years life safety (LS) and collapse prevention (CP); and
Very Rare 970 years 10% in 100 years these performance levels are mapped onto the
three hazard levels with return periods 75, 475
and 2475 years, respectively. The attainment of
wood. System performance levels were described each performance level is described by reaching
in terms of local (individual element) performance or exceeding a threshold value that defines the
levels. respective performance level (or structural limit
It is concluded from the review of documents state). It is noteworthy that the mapping between
Vision 2000 and FEMA 273 that conceptually the performance and hazard levels does not indi-
PBSD framework is very similar. In other words, cate that the respective hazard level is considered
it is agreed that seismic design should be based only in evaluating the probability of attaining
on multiple performance objectives for stated a given performance level. As described in the
earthquake hazard levels, however, the definition next section on the LCC model, first the fragility
for earthquake design and performance levels curve is derived and then it is integrated over the
show considerable variation. Furthermore, there entire range of the earthquake intensity measure
is consensus neither on metrics to be selected as to obtain the probability of reaching or exceeding
the indicators of performance levels, nor their each structural limit (or damage) state.
relation to different damage states. However, The structural capacity and earthquake demand
interstory drift is mostly preferred since it is are coupled. In other words, the capacity of a
closely related to the development of P- instabil- structure is not independent from the earthquake
Figure 4. Recommended performance objectives for buildings in Vision 2000 (Adapted from SEAOC, 1995)
10
Figure 5. Recommended rehabilitation objectives for buildings in FEMA 273 (Adapted from FEMA, 1997)
demand imposed on the structure. The capacity Finite element model (based on distributed
varies during the strong ground shaking which inelasticity) of the structure under consideration
also influences the seismic forces acting on the is built and pushover analysis is conducted. In
structure. The most elegant way of evaluating order to establish the limit state threshold values
the failure probabilities is through the joint prob- that define the structural capacity, the response
ability density function of capacity and demand metrics obtained from the analysis such as stresses,
which can be derived by Monte Carlo simulation strains, and interstory drifts are correlated to
(MCS). However, there are various sources of previously mentioned performance levels. Here
uncertainty in evaluation of the failure probabili- only structural damage is considered; however,
ties including the inherent variability of ground nonstructural damage could also be incorporated
motions and randomness in material properties. into the performance level definitions as suggested
Accounting for all the variability through MCS by Vision 2000 (SEAOC, 1995). Local (e.g.
requires a large number of structural analyses. At stresses and strains) and global (e.g. interstory
the same time, accurate prediction of structural drift) response measures could be combined to
capacity and earthquake demand is critical for define different performance levels. Here, only the
seismic design, and it is required to use analysis strains in longitudinal reinforcement and the core
methods that yield reliable estimates. Performing concrete are used as the criterion and the threshold
MCS becomes infeasible when computationally values are represented in terms of interstory drift
demanding methods such as inelastic dynamic by mapping the strains onto the pushover curves.
time history analysis is used. Therefore, here it A typical pushover curve is shown in Figure 6(a)
is assumed that structural capacity is independent alongside the limit state points based on strains
from earthquake demand and pushover analysis is in the longitudinal reinforcement and the core
used to evaluate the former while the latter is ob- concrete.
tained by inelastic dynamic time history analysis.
11
The LCC Model (IM) in Eqn. (3) defines the hazard curve, where
IM is PGA for this study. The conditional prob-
Once the hazard and performance levels are ability of demand being greater than the capacity
defined, the next step is the calculation of the (or fragility) is
damage state probabilities, i.e. the probability of
the structure attaining the pre-defined damage

P (D > C ,i | IM = im ) = P (
D > | IM = im )fC ,i ( ) d
states throughout its lifetime. Once the damage 0
state probabilities are calculated the cost of repair (4)

for each damage state is evaluated and the LCC
of the structure is easily found. In the following where is the variable of integration and fC,i is the
the full derivation of a LCC model is provided. probability density function for structural capacity
The expected LCC of a structure is calculated as for the ith damage state. This formulation assumes
that the demand and the capacity are independent
N of each other (as discussed in the previous section).
E C LC (t ) = C 0 + t C i Pi (1) Structural capacity is assumed to be lognormally
i =1
distributed with C,i and C that are, respectively,
the mean and the standard deviation of the cor-
where C0 is the initial construction cost, t is the responding normal distribution. As an example,
service life of the structure, N is the total number the lognormal probability density functions for
of limit-states considered, Pi is the total probability structural limit states are shown in Figure 6(b). In
that the structure will be in the ith damage state Figure 6(b), three limit states IO, LS, and CP with
throughout its lifetime, and Ci is the corresponding threshold values of 1%, 2.5% and 7% interstory
cost as a fraction of the initial cost of the structure. drift, respectively, are assumed and C is taken as
Pi is given by 0.3. The uncertainty in capacity (here represented
with C) accounts for factors such as modelling
Pi = P (D > C ,i ) P (D > C ,i +1 ) error and variation in material properties. A more
(2) detailed investigation of capacity uncertainty is
available in Wen et al. (2004) and Kwon and
where D is the earthquake demand and C,i is the Elnashai (2006).
structural capacity, usually in terms of drift ratio, The structural demand is also assumed to fol-
defining the ith damage state, as described in the low a lognormal distribution and the probability of
previous section. The probability of demand being demand exceeding a certain value, , is given by
greater than capacity is evaluated as
ln ( ) D|IM =im
dv (IM ) P (D > | IM = im ) = 1

P (D > C ,i ) = P (
D > C ,i | IM = im ) dIM D
dIM
0
(5)
(3)
where [] is the standard normal cumulative
where the first term inside the integral is the con-
distribution, D is the natural logarithm of the
ditional probability of demand being greater than
mean of the earthquake demand as a function of
the capacity given the ground motion intensity, IM.
the ground motion intensity, and D is the standard
This term is also known as the fragility function.
deviation of the corresponding normal distribu-
The second term is the slope of the mean annual
tion of the earthquake demand. Although, C and
rate of exceedance of the ground motion intensity.
D are dependent on ground motion intensity, in
12
Figure 6. (a) A typical pushover curve and the limit state points that delineate the performance levels,
(b) illustration of lognormal probability distributions for the three structural limit states (IO: immediate
occupancy, LS: life safety, CP: collapse prevention)
most studies they are taken as constants due to are shown in Figure 7. It is seen that although the
lack of information. mean of demand from three different methods
The dispersion in earthquake demand (here are similar, a higher dispersion is obtained when
represented with D) due to variability in ground natural and scaled ground motions are used.
motions is established here using a simple struc- The dispersion also increases with increasing
tural system (2-story 1-bay RC frame) and three ground motion intensity (i.e. earthquake return
sets of earthquake ground motions each represent- period). The focus of this chapter is optimal
ing a different hazard level at return periods of seismic design of structures considering the LCC
75, 475 and 2475 years. Each set includes seven (not assessment); therefore, the use of spectrum
ground motions which are selected from the PEER compatible records is suggested. For assessment
database (PEER, 2005) to represent the hazard at purposes, the use of unmodified (natural) records
a selected site in San Francisco, CA (more details is recommended.
are given in the example application below). The The mean, D, and standard deviation, D, of
correspondence between the hazard levels and the earthquake demand, as continuous functions of
ground motions is achieved using three different the ground motion intensity could be described
methods. In the first method, the natural records using (Aslani & Miranda, 2005)
are used without any modification. In the second
method, the records are scaled based on PGA to D (IM ) = c1 IM c2 or c1c2IM IM c3 (6)
match the PGA of the respective hazard level (in
this case 75, 475 and 2475 years return period
D (IM ) = c4 + c5 IM + c6 IM 2 (7)
earthquakes are represented with a PGA of 0.24 g,
0.51 g and 0.78 g, respectively). And in the third
method, spectrum matching is used to make the where the constants c1 through c3 and c4 through
acceleration response spectrum of each record c6 are determined by curve fitting to the data
compatible with the UHS corresponding to each points that match the PGA of the ground motions
return period shown in Figure 3(b). The results records with the mean and standard deviation,
13
Figure 7. Dispersion in earthquake demand for different representations of the earthquake hazard using:
(a) natural, (b) scaled, and (c) spectrum compatible records
respectively, of earthquake demand evaluated The Optimization Algorithm

using inelastic dynamic analysis. The mean and
standard deviation are the parameters of the cor- A brief review of most commonly used optimiza-
responding normal distribution that describes tion algorithms is provided above. The objectives
the earthquake demand. The curve fitting to the of the optimization problem considered here are
earthquake demand obtained from the spectrum highly nonlinear (due to the inelastic dynamic
compatible records that are mentioned above for analysis that is used to predict earthquake de-
the mean and standard deviation using the first mand) and the derivatives with respect to the
option in Eqn. (6) and Eqn. (7), respectively, is design variables are discontinuous. Furthermore,
shown in Figure 8. the design variables (i.e. section sizes and rein-
The hazard curve can also be described in forcement ratios) are discrete. Therefore, the use
mathematical form of gradient-based optimization algorithms is not
well suited. The evolutionary algorithms have
v (IM ) = c7 ec8 IM + c9 ec10 IM (8) shown to be very efficient in solving combinatorial
optimization problems as reviewed above. Here,
the taboo search (TS) algorithm is selected and
where c7 through c10 are constant to be determined
discussed in more detail. The same algorithm is
from curve fitting to the hazard curve.
also used to obtain the optimal solutions for the
With the above described formulation each
example application provided in the next section.
term in Eqn. (3) is represented as an analytical
TS algorithm, first developed by Glover (1989,
function of the ground motion intensity, IM. Thus,
1990), then it is adapted to multi-objective opti-
using numerical integration the desired probabili-
mization problems by Baykasoglu et al. (1999a,
ties of Eqn. (2) can easily be calculated. The cost
1999b). An advantage of TS algorithm is that a set
of repair for the IO, LS, and CP limit states, Ci,
of optimal solutions (Pareto-front or Pareto-set)
are usually taken as a fraction of the initial cost
could be obtained rather than a single optimal
of the structure. Finally, the LCC is evaluated
point in the objective function space. The meth-
through Eqn. (1).
odology presented in Baykasoglu et al. (1999a,
1999b) is used here with further modifications as
described below.
14
Figure 8. Curve fitting to obtain the (a) mean and (b) standard deviation of earthquake demand in
continuous form
TS algorithm has also been applied to structural to avoid multiple runs with the same combination
optimization problems. Bland (1998) applied TS of design variables. That is, a point in the taboo
algorithm to weight minimization of a space truss list is not evaluated again. The Pareto list includes
structure with various local minima and showed the points that are not dominated by other points
that TS algorithm was very effective in finding within the set for which the evaluation of objective
the global minimum when both reliability and functions is performed (i.e. the taboo list). The
displacement constraints were applied. Mano- seed list includes the points around which optimal
haran and Shanmuganathan (1999) investigated solutions are looked for. The latter are called as
the efficiency of TS, SA, GA and branch-and- the neighboring points and they are basically the
bound in solving the cost minimization steel truss adjacent elements of the multidimensional array,
structures. It was concluded that TS produced that defines the decision (or design) variables,
solutions better than or as good as both SA and around the given seed point. The modified TS
GA and it arrives at the optimal solution quicker algorithm works as follows:
than both methods. In a more recent study, Ohsaki
et al. (2007) explored the applicability of SA and A. Start with the minimum cost combination
TS algorithms for optimal seismic design of steel (or initial design), evaluate the objective
frames with standard sections. It was concluded function and add this point into taboo, seed
that TS was more advantageous over SA in terms and Pareto lists. Use this point as the initial
of the diversity of the Pareto solutions and the seed point.
ability of the algorithm to search the solutions B. Find the neighboring points around the cur-
near the Pareto front. rent seed. Here the number of neighboring
To describe the modified TS algorithm used points is chosen equal to the number of
here, first, it is required to make the following design variables and selected randomly
definitions. The taboo list includes points in the amongst all the adjacent elements of the
design space for which the objective functions are multidimensional array that defines the
evaluated for. Since inelastic dynamic time history design variables.
analysis is computationally costly, this list is used
15
C. Evaluate the objective function for all the for the application. Seven design variables are
neighboring points and add these into taboo defined for the optimization problem as given in
list. Table 2, alongside the minimum and maximum
D. Find the Pareto-front using the set of points values and increments. The combination of these
for which the objective function is evaluated design variables results in 30000 cases which
and update the Pareto list as the current constitute the search space. The objectives of
Pareto-front. the optimization problem are selected as initial
E. Amongst the neighboring points for the cur- and life-cycle cost, and structural performance in
rent iteration choose the one that is on the terms of maximum interstory drift. No constraints
Pareto-front and minimizes the cost function are defined because of the fact that code-based
as the next seed point and add this point into seismic design is not performed. The initial cost
the seed list. If there is no point which satis- of the frames considers only the material costs.
fies these conditions choose randomly one The unit prices for concrete and steel are assumed
of the points from the Pareto list amongst to $0.13/liters and $0.66/kg, respectively.
the ones that are not already in the seed list. A site at the intersection of 2nd and Market
F. Check if the predetermined maximum Streets in San Francisco, CA (with coordinates
number of objective function evaluations 37 47 21.58 N, 122 24 04.77 W) is se-
is exceeded; if yes stop, if not go to Step b. lected and the site-specific seismic hazard is
consistently derived. The soil conditions might
The steps required to perform LCC oriented significantly alter the characteristics of the ground
seismic design optimization according to the motions at a site, therefore, the soil conditions are
methodology presented in the preceding sections also taken into account in the development of the
is outlined in Figure 9. hazard curves. The soil at the selected site is
determined as D on the NEHRP (FEMA, 2003)
scale with a shear wave velocity in the range from
EXAMPLE APPLICATION 180 m/sec to 360 m/sec. Site specific hazard curve
for PGA is shown in Figure 11(a). UHS for three
In this section, the formulations presented in the different return periods (i.e. 75, 475, and 2475
preceding sections are applied to a building as an years) are obtained, see Figure 11(b), for record
illustration of the framework on LCC oriented selection and scaling purposes. These return pe-
seismic design optimization. The 2-story 2-bay riods are mapped onto three structural limit states
RC frame structure show in Figure 10 is selected IO, LS and CP, respectively. One earthquake
Table 2. Design variables and ranges for the considered structural frame
Minimum Maximum Increment

Column Reinforcement Ratio 1.0% 2.5% 0.50%
Beam Reinforcement Ratio 0.5% 2.0% 0.50%
Width of Exterior Columns (mm) 660.4 863.9 50.8
Width of Interior Columns (mm) 711.2 914.4 50.8
Depth of Columns (mm) 457.2 660.4 50.8
Width of Beams (mm) 508 711.2 50.8
Depth of Beams (mm) 406.4 508 50.8
16
Figure 9. Steps for performing LCC oriented seismic design optimization
ground motion is selected for each return period resented analytically by curve fitting to mathe-
(see the section The LCC Model for more in- matical form in Eqn. (6). The dispersion in
formation) and spectrum matching (Abrahamson, earthquake demand, D, is quantified as 0.25, 0.35
1993) is used to achieve direct correspondence and 0.45 for the three hazard levels with 75, 475
between the records and the hazard levels. The and 2475 years return period, respectively, by
acceleration response spectra of the records after running additional analysis. The details of this
spectrum matching are shown in Figure 11(b). derivation are omitted here for the sake of brev-
The life-time of the structure, t, is considered ity. As mentioned earlier, the structural limit states
as 50 years. The cost of repair for the IO, LS, and are also established by carefully investigating
CP limit states, Ci, are taken as 30, 70 and 100 different design alternatives in the search space
percent of the initial cost of the structure based by conducting pushover analysis and considering
on the correspondence of the limit states defined local response measures, i.e. strains in the longi-
previously and the information provided by Fra- tudinal reinforcement and concrete core. The
giadakis et al. (2006b). The hazard curve as a structural capacity is evaluated for a range of
function of PGA is shown in Figure 11(a). The design variables and the mean values for the three
functional form in Eqn. (8) is fitted to the hazard limit states IO, LS and CP are obtained as 0.4%,
curve to allow for the numerical integration of 2% and 3.5% interstory drift. The uncertainty in
Eqn. (3). The demand side of fragility relationship capacity is assumed to be equal to 0.35 taking
given in Eqn. (4), i.e. the first term, as a function previous research as a reference (Wen, et al.,
of PGA, is obtained by finding the maximum 2004). First the conditional probability (fragility)
interstory drift under the three hazard levels in Eqn. (4), then the total probability in Eqn. (3)
through inelastic dynamic analysis and it is rep- is calculated. The damage state probabilities are
17
Figure 10. RC frame structure used for the example application
found from Eqn. (2) and the expected LCC of the to the Pareto-front. The initial and the total costs
structure is obtained from Eqn. (1). of the building vs. the maximum interstory drift
During the optimization process, TS algorithm under the 2475 years return period earthquake is
is allowed to search 10% of the search space. In shown in Figure 12(b) [only for optimal solutions].
other words, the maximum number of objective It is seen that there is a compromise between
function evaluations is set 3000 for each hazard the initial cost, LCC and structural performance
level. Figure 12(a) shows the combinations of the (represented with maximum interstory drift). As
design variables (in the solution space) that are the initial cost increases it approaches the LCC
searched by the TS algorithm to obtain the Pareto- and the maximum interstory drift is reduced (bet-
front which is also shown in the same figure with ter structural performance) while the LCC and
a solid line with circle marks. It is seen that the the maximum interstory drift are significantly
algorithm is very effective in terms of narrowing higher for lower initial cost solutions. This type
down the search space to the points that are close of a representation is very important for the use
Figure 11. (a) Site specific hazard curve, (b) UHS and spectrum compatible earthquake records
18
Figure 12. (a) Example Pareto-front obtained using TS algorithm, (b) initial and LCC of the building
of LCC oriented seismic design optimization by Most importantly, for the formulation provided
the decision makers such as owners and engineers in this paper, the constants Ci in Eqn. (1) which
as well as for its implementation in future seismic define the cost of repair for the ith damage state
design codes, as the presentation of results in as a fraction of the initial cost have strong influ-
the Pareto-optimal form gives flexibility in the ence on the LCC. As discussed earlier, it is not
decision making process and can accommodate straightforward to determine damage states for
changes to the performance requirements through- structures and the associated cost of repair. The
out the course of design. effect of other parameters such as the dispersion
in capacity and demand on the LCC should also be
investigated. Eqn. (4) assumes that the structural
FUTURE RESEARCH DIRECTIONS capacity is independent of earthquake demand
which is another simplification that is usually
There are two major issues in LCC oriented adopted. The evaluation of workmanship cost is
seismic design of structures that requires further difficult as it is dependent on various parameters
research. The first one is the application of the (especially the local cost of labor). Finally, the
concept developed here to structural models with cost of non-structural damage after an earthquake
significant number design variables. The size of might exceed several times the cost of structural
the search space increases exponentially with damage and this also needs to be investigated.
increasing number of design variables and without
a compromise from the accuracy of structural as-
sessment it becomes very difficult to cope with the CONCLUSION
computational demand. Although simplified struc-
tural models and analysis techniques are always Optimization in structural design is a growing sub-
an option, these undermine the main objective of ject due to increased awareness of the community
the seismic design optimization by introducing regarding the sustainability of civil infrastructure
additional uncertainty into the procedure. and the use Earths nonrenewable resources. And
The second issue is sensitivity of the LCC the recent incidents have shown us once again that
results to the assumptions made in the derivations. the initial cost of a structure is not a satisfactory
19
criterion for the economy objective of seismic Baykasoglu, A., Owen, S., & Gindy, N. (1999a).
design, as the repair and indirect costs due to Solution of goal programming models using a
damage in future earthquakes might be signifi- basic taboo search algorithm. The Journal of the
cantly higher than the initial cost, therefore the Operational Research Society, 50(9), 960973.
life-cycle cost (LCC) should also be considered.
Baykasoglu, A., Owen, S., & Gindy, N. (1999b).
In this chapter the LCC considerations in seismic
A taboo search based approach to find the Pareto
design optimization of structures is discussed
optimal set in multiple objective optimization.
emphasizing the importance of the following: (1)
Engineering Optimization, 31(6), 731748.
use of advanced analysis, which provides the most
doi:10.1080/03052159908941394
rigorous assessment of structural capacity and
earthquake demand, (2) evaluation of the struc- Bertero, V. V., & Zagajeski, S. W. (1979). Optimal
tural capacity by taking into account not only the inelastic design of seismic-resistant reinforced
global behavior of the structure but also the local concrete framed structures. Paper presented at
response, (3) use of system-specific limit states the Nonlinear Design of Concrete Structures,
to define the structural capacity, and (4) consid- CSCE-ASCE-ACI-CEB International Sympo-
eration of all major sources of uncertainty, from sium, Ontario, Canada.
seismogenic source characteristics to material
Bland, J. (1998). Structural design optimization
properties and structural modeling in calculating
with reliability constraints using tabu search.
the limit state exceedance probabilities. A LCC
Engineering Optimization, 30(1), 5574.
formulation is provided where each component
doi:10.1080/03052159808941238
from seismic hazard, to structural assessment, LCC
model and optimization algorithm are rigorously Chan, C. M., & Zou, X. K. (2004). Elastic and
addressed. An example application is included to inelastic drift performance optimization for rein-
allow for easy implementation of the framework forced concrete buildings under earthquake loads.
presented here. Earthquake Engineering & Structural Dynamics,
33(8), 929950. doi:10.1002/eqe.385
Cheng, F. Y., & Truman, K. Z. (1985). Optimal
REFERENCES
design of 3-D reinforced concrete and steel build-
Abrahamson, N. A. (1993). Non-stationary spec- ings subjected to static and seismic loads including
tral matching program RSPMatch: users manual. code provisions (No. Final Report Series 85-20).
Washington, DC, USA: prepared by University
Adamu, A., Karihaloo, B. L., & Rozvany, G. I. of Missouri-Rolla, National Science Foundation,
N. (1994). Minimum cost design of reinforced US Department of Commerce.
concrete beams using continuum-type optimality
criteria. Structural and Multidisciplinary Opti- Collins, K. R., Wen, Y. K., & Foutch, D. A. (1996).
mization, 7(1), 91102. Dual-level seismic design: A reliability-based
methodology. Earthquake Engineering & Struc-
Aslani, H., & Miranda, E. (2005). Probability- tural Dynamics, 25(12), 14331467. doi:10.1002/
based seismic response analysis. Engineering (SICI)1096-9845(199612)25:12<1433::AID-
Structures, 27(8), 11511163. doi:10.1016/j. EQE629>3.0.CO;2-M
engstruct.2005.02.015
20
Cornell, C. A., Jalayer, F., Hamburger, R. O., & Glover, F. (1990). Tabu search - Part II. ORSA
Foutch, D. A. (2002). Probabilistic basis for 2000 Journal on computing, 2(1), 4-32.
SAC federal emergency management agency
International Building Code. (2006).
steel moment frame guidelines. Journal of Struc-
tural Engineering, 128(4), 526533. doi:10.1061/ Kwon, O.-S., & Elnashai, A. (2006). The effect
(ASCE)0733-9445(2002)128:4(526) of material and ground motion uncertainty on the
seismic vulnerability curves of RC structure. Engi-
FEMA. (1997). NEHRP guidelines for the seismic
neering Structures, 28(2), 289303. doi:10.1016/j.
rehabilitation of buildings, FEMA 273. Wash-
ington, DC: Federal Emergency Management
Agency. Lagaros, N. D., Fragiadakis, M., Papadrakakis,
M., & Tsompanakis, Y. (2006). Structural opti-
FEMA. (2003). NEHRP recommended provisions
mization: A tool for evaluating seismic design
for seismic regulations for new buildings and
procedures. Engineering Structures, 28(12),
other structures, FEMA 450, Part 1: Provisions.
16231633. doi:10.1016/j.engstruct.2006.02.014
Washington, DC: Federal Emergency Manage-
ment Agency. Lagaros, N. D., & Papadrakakis, M. (2007). Seis-
mic design of RC structures: A critical assessment
Foley, C. M., & Schinler, D. (2003). Automated
in the framework of multi-objective optimization.
design of steel frames using advanced analysis and
Earthquake Engineering & Structural Dynamics,
object-oriented evolutionary computation. Jour-
36(12), 16231639. doi:10.1002/eqe.707
nal of Structural Engineering, 129(5), 648660.
doi:10.1061/(ASCE)0733-9445(2003)129:5(648) Li, G., Zhou, R.-G., Duan, L., & Chen, W.-F.
(1999). Multiobjective and multilevel optimization
Fragiadakis, M., Lagaros, N. D., & Papadraka-
for steel frames. Engineering Structures, 21(6),
kis, M. (2006a). Performance-based earthquake
519529. doi:10.1016/S0141-0296(97)00226-5
engineering using structural optimisation tools.
International Journal of Reliability and Safety, Liu, M. (2005). Seismic design of steel moment-
1(1/2), 5976. doi:10.1504/IJRS.2006.010690 resisting frame structures using multiobjective op-
timization. Earthquake Spectra, 21(2), 389414.
Fragiadakis, M., Lagaros, N. D., & Papadrakakis,
doi:10.1193/1.1902952
M. (2006b). Performance-based multiobjective
optimum design of steel structures considering Liu, M., Burns, S. A., & Wen, Y. K. (2003).
life-cycle cost. Structural and Multidisciplinary Optimal seismic design of steel frame buildings
Optimization, 32(1), 111. doi:10.1007/s00158- based on life cycle cost considerations. Earth-
006-0009-y quake Engineering & Structural Dynamics, 32(9),
13131332. doi:10.1002/eqe.273
Fragiadakis, M., & Papadrakakis, M. (2008).
Performance-based optimum seismic design of Liu, M., Burns, S. A., & Wen, Y. K. (2005). Mul-
reinforced concrete structures. Earthquake Engi- tiobjective optimization for performance-based
neering & Structural Dynamics, 37(6), 825844. seismic design of steel moment frame structures.
doi:10.1002/eqe.786 Earthquake Engineering & Structural Dynamics,
34(3), 289306. doi:10.1002/eqe.426
Glover, F. (1989). Tabu search - Part I. ORSA
Journal on Computing, 1(3), 190-206.
21
Liu, M., Burns, S. A., & Wen, Y. K. (2006). Sahab, M. G., Ashour, A. F., & Toropov, V. V.
Genetic algorithm based construction-conscious (2005). Cost optimisation of reinforced concrete
minimum weight design of seismic steel moment- flat slab buildings. Engineering Structures, 27(3),
resisting frames. Journal of Structural Engineer- 313322. doi:10.1016/j.engstruct.2004.10.002
ing, 132(1), 5058. doi:10.1061/(ASCE)0733-
SEAOC. (1995). Vision 2000, performance based
9445(2006)132:1(50)
seismic engineering of buildings (Vol. I and II).
Liu, M., Wen, Y. K., & Burns, S. A. (2004). Life Sacramento, CA: Structural Engineers Associa-
cycle cost oriented seismic design optimization tion of California.
of steel moment frame structures with risk-taking
Uniform Building Code (1997).
preference. Engineering Structures, 26(10),
14071421. doi:10.1016/j.engstruct.2004.05.015 USGS. (2008). Documentation for the 2008 update
of the United States national seismic hazard maps
Manoharan, S., & Shanmuganathan, S. (1999). A
(No. Open-File Report 2008-1128). Reston, VA:
comparison of search mechanisms for structural
United States Geological Survey.
optimization. Computers & Structures, 73(1-5),
363-372. Wen, Y. K., Ellingwood, B. R., & Bracci, J. M.
(2004). Vulnerability function framework for
Memari, A. M., & Madhkhan, M. (1999). Optimal
consequence-based engineering (No. Project DS-4
design of steel frames subject to gravity and seis-
Report). Urbana, IL: Mid-America Earthquake
mic codes prescribed lateral forces. Structural
(MAE) Center.
and Multidisciplinary Optimization, 18(1), 5666.
Wen, Y. K., & Kang, Y. J. (2001a). Minimum
Moharrami, H., & Grierson, D. E. (1993). Com-
building life-cycle cost design criteria, I: Meth-
puter-automated design of reinforced concrete
odology. Journal of Structural Engineering,
frameworks. Journal of Structural Engineering,
127(3), 330337. doi:10.1061/(ASCE)0733-
119(7), 20362058. doi:10.1061/(ASCE)0733-
9445(2001)127:3(330)
9445(1993)119:7(2036)
Wen, Y. K., & Kang, Y. J. (2001b). Minimum
Ohsaki, M., Kinoshita, T., & Pan, P. (2007).
building life-cycle cost design criteria, II: Ap-
Multiobjective heuristic approaches to seismic
plications. Journal of Structural Engineering,
design of steel frames with standard sections.
127(3), 338346. doi:10.1061/(ASCE)0733-
9445(2001)127:3(338)
36(11), 14811495. doi:10.1002/eqe.690
Zou, X. K., & Chan, C. M. (2005). An optimal
PEER. (2005). Pacific Earthquake Engineering
resizing technique for seismic drift design of
Research (PEER) center: NGA database. Re-
concrete buildings subjected to response spec-
trieved January 1, 2009, from http://peer.berkeley.
trum and time history loadings. Computers &
edu/nga/
Structures, 83(19-20), 16891704. doi:10.1016/j.
Pezeshk, S. (1998). Design of framed struc- compstruc.2004.10.002
tures: An integrated non-linear analysis and
Zou, X. K., Chan, C. M., Li, G., & Wang, Q.
optimal minimum weight design. Interna-
(2007). Multiobjective optimization for per-
tional Journal for Numerical Methods in
formance-based design of reinforced concrete
Engineering, 41(3), 459471. doi:10.1002/
frames. Journal of Structural Engineering,
(SICI)1097-0207(19980215)41:3<459::AID-
133(10), 14621474. doi:10.1061/(ASCE)0733-
NME293>3.0.CO;2-D
9445(2007)133:10(1462)
22
23
Chapter 2
Performance-Based
Seismic Design:
A Search-Based Cost Optimization
with Minimum Reliability Constraints
Oscar Mller Fabin Savino

University of Rosario, Argentina University of Rosario, Argentina
Marcelo Rubinstein Ricardo O. Foschi

University of Rosario, Argentina University of British Columbia, Canada
ABSTRACT
An approach is presented to structural optimization for performance-based design in earthquake engineer-
ing. The objective is the minimization of the total cost, including repairing damage produced by future
earthquakes, and satisfying minimum target reliabilities in three performance levels (operational, life
safety, and collapse). The different aspects of the method are considered: a nonlinear dynamic structural
analysis to obtain responses for a set of earthquake records, representing these responses with neural
networks, formulating limit-state functions in terms of deformations and damage, calculating achieved
reliabilities to verify constraint violations, and the development of a gradient-free optimization algo-
rithm. Two examples illustrate the methodology: 1) a reinforced concrete portal for which the design
parameters are member dimensions and steel reinforcement ratios, and 2) optimization of the mass at
the cap of a pile, to meet target reliabilities for two levels of cap displacement. The objective of this
latter example is to illustrate model effects on optimization, using two different hysteresis approaches.
INTRODUCTION objective (e.g.,the total cost), with constraints

specifying minimum reliability levels for each
Structural optimization for performance-based performance requirement. Three performance
design in earthquake engineering aims at finding levels are normally considered in earthquake en-
optimum design parameters (e.g.,dimensions of gineering: serviceability, life safety and collapse.
structural members, steel reinforcement ratios The objective of the present Chapter is to pres-
for concrete structures, characteristics of energy ent a procedure for performance-based optimiza-
dissipation devices) corresponding to a minimum tion of a preliminary design, taking into account the
DOI: 10.4018/978-1-4666-1640-0.ch002
Performance-Based Seismic Design
uncertainties in the ground motion and structural for optimization and reliability estimation. Each
properties, and at the same time minimizing the of the responses of interest can be approximately
total cost. This includes, in addition to the initial represented by using response surfaces which,
cost of construction, that associated with repairs when properly adjusted, can be used as substi-
following each of a series of earthquakes within a tutes for the dynamic analysis (Hurtado, 2004).
given time interval. The preliminary design may be Different forms for approximating surfaces have
the result of applying capacity design approaches, been studied (Mller et al., 2009b), and neural
according to existing Codes, in order to satisfy networks have been shown to offer advantages
the performance levels. of flexibility and adaptability. Regardless of the
The subsequent sections of this Chapter discuss type of response surface used, a major advantage
the different aspects of the proposed design opti- of the substitute is the computational efficiency
mization process, including: 1) the quantification achieved with Monte Carlo-type simulations when
of the seismic hazard or possible ground motions estimating probabilities of non-performance.
at the site; 2) the use of a structural dynamics The approach chosen in the optimization prob-
analysis model that represents, as best as possible, lem must consider the presence of constraints,
the nonlinear response and the hysteretic energy the dimensionality and the form and number of
dissipation mechanisms; 3) the calculation of a objective functions. Optimization methods can use
response database for the estimated ranges of different approaches (Prez Lpez, 2005; Swisher
the different intervening input random variables et al., 2000), some requiring the calculation of
and design parameters; 4) a functional represen- gradients (steepest descent, conjugate gradients,
tation for the discrete databases, using neural Newton or quasi-Newton schemes) and others, not
networks response surfaces; 5) implementation using gradients, implementing genetic or search
of the response surfaces in the calculation of the algorithms. This Chapter proposes an efficient
reliability levels achieved, at each performance search-based algorithm which also accounts for
level, for specific values of the design parameters; constraints given by specified minimum reliability
6) representation of these reliability levels by levels at each performance level. The optimiza-
neural networks in terms of the design param- tion approach presented in this Chapter is based
eters, in order to achieve an efficient calculation on previous work by the authors (Mller et al.,
of reliability constraint violations and, finally, 7) 2007, 2008, 2009a,c), and introduces additional
an optimization algorithm that would search for contributions as follows:
a minimum total cost objective under minimum
reliability constraints for each performance level. The optimization process is started from a
Earthquakes show large ground motion un- preliminary, deterministic seismic design;
certainties which must be coupled with those in The previous search-based algorithm by
the structural capacities, and the demands will the authors has been modified to mini-
likely trigger a nonlinear structural response. mize the possibility of encountering local
For each performance level, the formulation of minima;
the limit-state functions requires the estimation The damage repair costs introduced into
of maximum responses (e.g., maximum inter- the objective function take into account the
story drift) during the duration of the earthquake. occurrence of multiple seismic events dur-
Since only discrete values can be obtained by ing the life of the structure.
numerical analyses for specific combinations of
the variables and parameters, the responses need The optimization approach presented here is
to be given a continuous functional representation illustrated with two performance-based design
24
examples. The first is for a simple, reinforced con- in which j(xd) are the reliability indices achieved
crete portal frame, for which the design parameters with the design parameters xd for each of the three
are the column and beam dimensions, as well as performance levels j: operational, life safety
the steel reinforcement ratios. The optimality of or controlled damage and collapse, with jT
the preliminary design is discussed, as well as the being the corresponding prescribed minimum
efficiency of the proposed optimization process. targets over the design life TD.
This example also considers the sensitivity of the The design parameters xd could be, for example,
optimal solution to the assumed relationship be- either dimensions, steel reinforcement ratios, or
tween calculated damage and repair costs, as well statistical parameters for some of the random
as to the reliability levels prescribed as minimum variables in the problem (e.g., the mean value of
constraints for each of the performance levels. the required steel yield strength).
Reliability estimates and optimization results Given a procedure to verify reliability constrain
are conditional on the analysis models used. The compliance, as described in Section 2.2, an opti-
second example is the optimization for the permis- mization technique is then applied to minimize the
sible pile-cap mass carried by a pile foundation. total cost while satisfying the minimum reliability
This main objective in this example is to discuss indices for each performance level. The optimiza-
the importance of the analysis model formulation tion algorithm proposed here is a gradient-free,
in the final optimization results. Two different search-based approach, as described in Section 5.
models are used, based on two different formula-
tions for hysteretic energy dissipation. 1.2 The Methodology for
Reliability Evaluation
1. THE OPTIMIZATION The optimization process requires the calcula-

FORMULATION tion of reliability levels achieved with the design
parameters xd. This is implemented through the
1.1 General Objective following steps, all prior to the optimization proper.
These steps, to be addressed in the following sec-
The aim of the optimization considered here is tions, include:
to find values for the design parameters, grouped
in a vector xd, to minimize the objective function A. Definition of the random variables X in the
C(xd). This function is the total cost, and it is given problem, and their corresponding upper and
by the sum of the initial construction cost C0(xd) lower bounds;
and the damage repair cost Cd(xd), required after B. Selection of combinations for the variables
the occurrence of earthquakes during the service X within their corresponding bounds, using
life of the structure. That is, design of experiments;
C. Construction of a structural response data-
C (xd ) = C 0 (xd ) + C d (xd ) (1) base: for each variable combination in b),
and for each record in a set of earthquakes
The minimum total cost must correspond to a considered likely to occur at the site, determi-
structure that also satisfies minimum reliability nation of the maximum structural responses
requirements for different performance levels: Ri(X) that enter in the formulation of the
performance functions, using a nonlinear
j (xd ) j T , j = 1, 2, 3 (2) dynamic time-step analysis;
25
D. For each variable combination, determina- that the development of a response database is the
tion of the mean maximum structural re- task involving the greatest computational effort.
sponse and corresponding standard deviation A nonlinear model for reinforced concrete, based
over the set of earthquake records, followed on bar elements, has been discussed elsewhere
by an approximation of these means and (Mller, 2001; Mller and Foschi, 2003) and
standard deviations by explicit, continuous has been used in the first application example
neural network response surface functions presented here in Section 7.
Fi(X). The reliability and, ultimately, the optimization
E. Using the mean and standard deviation results, are conditional on the adequacy of the
for each maximum response to represent analysis model, and a proper model error vari-
its corresponding cumulative probability able must be included in the random set X. Some
distribution by a lognormal. In turn, with of the issues to be addressed in the formulation of
these distributions, implementing the neural the dynamic analysis relate, for example, to how
networks from d), to calculate the probabili- the hysteretic energy dissipation is represented.
ties of non-performance or the associated This issue is considered in the second application
reliability indices j. for any given set of example in Section 8, a soil/structure interaction
design parameters. problem which highlights the dependence of the
F. For a set of parameters xd, chosen within their optimization results on the assumptions made in
bounds, approximation of the discrete results the analysis model.
from e) by means of neural network response
surface functions for the reliabilities j(xd).
Using these approximations directly, during 3. RESPONSE REPRESENTATION
the optimization, to calculate the reliability BY NEURAL NETWORKS
achieved at each performance level for a
particular vector xd, in order to efficiently 3.1 Neural Networks
evaluate constrain violations.
Neural networks are algorithms for the transmis-
sion of information between input and output.
2. NONLINEAR DYNAMIC This technique has been used here to represent the
ANALYSIS MODEL structural response databases. Because the neural
networks literature is quite extensive, only a brief
The optimization requires the development of a description of the type of network used here is
response database in terms of the random variables included for completeness.
X, for each earthquake record considered. This Several input parameters are assumed to oc-
calculation implies the application of a nonlinear cupy individual neurons in an input layer and,
dynamic analysis and a step by step integration of similarly, several outputs can form the output
the equations of motion over the duration of the layer containing the output neurons. The in-
earthquake. The focus of this Chapter is not on formation between input and output is assumed to
the details of the particular structural nonlinear flow through intermediate or hidden neurons, the
dynamic analysis used. Many modeling techniques strength of the information between two neurons
have been proposed, aiming at achieving an ad- j and i being given by weight parameters Wji. The
equate balance between accuracy and simplicity. architecture of the network is shown in Figure
It is important to emphasize, however, that each 1, for N inputs and K outputs. The information
approach has an associated modeling error, and received by a hidden or output neuron is modi-
26
fied by a transfer function h(t) before being sent either gradient- or search-based. For example, the
forward along the network. The mathematical back-propagation or other Newtonian algorithms
expression for this algorithm is implement steepest descent with either a controlled
or a calculated step. Although back-propagation is
the commonly used training method, this Chapter
J N
R(X) F (X) = h ( Wkj h( Wji X i +Wj 0 ) +Wk 0 )
j =1 i =1 presents in Section 4.2 a gradient-free search
(3) algorithm as a straight-forward approach to the
training optimization.
in which, in our case, R(X) is the true value for
the structural response obtained with the dy- 3.2 Search-Based Optimization
namic analysis for the input vector X, with com- as a Training Algorithm
ponents Xi, F(X) is the neural network approxima-
tion, Wkj and Wji are the weight parameters, and The following describes a gradient-free, search-
h(t) is the transfer function applied at the hidden based optimization algorithm for the neural
and output neurons. This function could take dif- network weights W. The optimization strategy is
ferent forms, and in this Chapter we use a sigmoid: called OPT. Let N be the number of input variables
and NP the number of input data combinations.
1.0 The input values are then X0(i,k), with i = 1, N and
h(t ) = (4)
(1 + exp(t )) k = 1, NP. Before proceeding, the data are scaled
to values X(i,k), between the limits 0.01 and 0.99,
The weights W must be obtained in such a in order to eliminate potential problems with dif-
way that the differences between R(X) and F(X) ferent units and magnitudes. Similarly, the output
be minimized. This optimization is defined results from the response analysis, T0(k), are also
as the training of the network, and different scaled to values T(k) between 0.01 and 0.99, tak-
minimization algorithms can be implemented,
Figure 1. Neural network architecture
27
ing into account that the sigmoid transfer function for the new 80% of the data corresponding to the
h(t) de Eq.(4) produces values between 0 and 1. greatest errors in absolute value. This process is
To start the optimization for the weights W, repeated NREP times (100 in the work presented
initial values are chosen randomly from an uni- here), looking for the possibility of local minima.
form distribution between -5 and 5. These limits The final solution corresponds to that W0 with the
are arbitrarily chosen, but are considered to be minimum overall error E0.
sufficiently wide to lead to a wide range in the
predicted outputs Y(k). For each weight selection 3.3 Subsequent Optimization
(e.g., 5000 combinations in this Chapter) the Using Gradients
corresponding prediction error E is calculated
from Eq.(5), The search-based algorithm described in 4.2 can be
complemented with a gradient-based optimization
NP
starting from the final weights W0. The following
E = (Y (k ) T (k ))
2
(5)
k =1
alternatives may be pursued:
A. Search for weights W trying to zero out the

keeping the combination of weights W0 which
error E
correspond to the minimum among the calculated
errors E.
If the error E is approximated as a linear func-
The absolute value of the NP errors (Y(k)
tion around E0,
T(k)), corresponding to the weights W0, are ranked
from smallest to largest. The set corresponding
E = E 0 + G0 (W W0 ) (6)
to 80% of the data with the largest errors is used
for the network training, with the remaining 20%
available for validation. in which Go is the gradient vector at Wo,
With W0 representing an anchor point with
corresponding error E0 over the training set, new E
G0 i = (7)
values for the weights W are randomly chosen Wi
W0
within a neighborhood of W0, and the correspond-
ing errors E are calculated with Eq.(5). Whenever
a set W is found for which E < E0, this set becomes the objective is to find W to make E = 0. Thus, the
the new W0 and the search process is repeated vector W W0 is taken along the gradient direction
around this new anchor. The maximum number (but in the negative direction):
of repetitions is limited to a prescribed number
(500 in the work presented here). (W W0 ) = G0 (8)
The number of samples of weights W around
a given anchor W0 is limited, e.g., to 1000. If this The step magnitude is then calculated from
limit is reached and all errors satisfy E > E0, then
convergence has been achieved and the mnimum E0
solution is estimated to be in correspondence with E = E 0 + G0 ( G0 ) = 0 =

G G0
0
the anchor W0. (9)
If either convergence has been achieved, or
the number of repetitions has exceded the stipu- giving a new W as
lated maximum, then the NP individual errors
are calculated and ranked, re-initiating the search
28
E0 G0 G0
W = W0
G0 (10) W = W0 G0 (15)
G0 G0 G0 H0 G0
Eq.(10) can be used to implement an iterative Eq.(15) can also be used to implement an
process which, however, may not converge. The iterative process, and convergence is improved
process can be stopped when the new calculated if only a fraction of the step is applied. As a
error is greater tan the previous one. To improve simplification, the full Hessian matrix can be
convergence, it is sometimes useful to advance replaced by using only its diagonal terms.
only a fraction of .
C. Combination of optimization approaches
B. Search for weights W trying to zero out the
gradient G The gradient-free search OPT described in
4.2 can be utilized to obtain a first solution for
A zero gradient is associated with a minimum the optimum weights W. This approach can be
value of the error E. Since, in general, the network subsequently combined with either the strategy
predictions will not all agree exactly with the input described in a) or b), in an attempt to improve the
data, searching for a minimum error is better than optimization results. This can be done after each
searching for E = 0. In this case, the gradient G is of the NREP repetitions of OPT.
expressed as a linear function around the gradient The components of the gradient or of the Hes-
G0 evaluated for the weights W0, sian matrix are calculated numerically.
G = G0 + H0 (W W0 ) (11)
4. SEARCH-BASED OPTIMIZATION
in which H0 is now the Hessian matrix FOR THE DESIGN PARAMETERS
2E The optimization process OPT described in 4.2

H0i j = (12) for the training of neural networks can also be
Wi Wj W0 adapted to the problem of determining optimum
design parameters, minimizing the total cost and
The search is now for a new vector W W0, satisfying the required minimum reliability con-
also in the gradient direction, with the objective straints. Thus, the numerical procedure follows
of achieving G = 0. Thus, the steps represented in Figure 2.
The starting point or anchor, xd0, is the result
(W W0 ) = G0 (13) of a preliminary design for the structure. This first
step normally utilizes deterministic, simpler
methods and may follow capacity design guide-
The step magnitude is now obtained from lines specified in the Codes. For this preliminary
choice of the design parameters, the neural net-
G0 G0 works for the reliability indices are used to estimate
= (14)
G0 H0 G0 the achieved levels j(xd0) and, using Eq.(1), the
total cost C(xd0), Figure 2(a). This Figure shows,
giving a new W as as a schematic illustration, the case of just two
design parameters, xd(1) and xd(2), which could
29
Figure 2. Schematic for the design parameters optimization process
either satisfy the reliability constraints or not. If still no combinations are found leading to
Should the preliminary design parameters xd0 not a cost lower than that for the anchor, the search
satisfy the reliability constraints, a search is made factor radius r1 is amplified to r2 = r1 + r, with
according to the following steps, Figure 2(b), r = 0.3 used in this work. A number m of com-
until one combination xd is found satisfying the binations xd is chosen within the amplified search
constraints. This combination is then taken as the volume, satisfying the reliability constraints, and
initial anchor. maintaining the same density as in the previous
Next, n combinations of xd are chosen randomly search domain corresponding to the radio r1. Of
in the neighborhood of xd0, Figure 2(c), choosing the m combinations (maximum of 2000), the
each of the design parameters, xdi, within the inter- first that is associated with a cost lower than that
val xd0i r1 xd0i (initially r1 = 0.1). The reliability for the anchor is retained and used as the new
constraints are verified for each combination, anchor. This sequence is repeated three times in
choosing a new combination if the constraints are order to reduce the possibility of encountering a
not satisfied. The total cost is calculated for the local minimum.
n combinations, keeping the one corresponding The search process is stopped when no com-
to the lowest cost. If this cost is lower than the bination is found, satisfying the constraints, with
one corresponding to the anchor, then the corre- a cost lower than for the anchor.
sponding combination becomes the new anchor
and the process is re-initiated. The work for this
Chapter used n = 100. 5. OBJECTIVE FUNCTION:
If none of the n combinations leads to a cost THE TOTAL COST
lower than the one corresponding to the anchor, the
search is densified by choosing more combinations Costs associated with the occurrence of earth-
within the interval, until one is found leading to a quakes and associated damage have been discussed
lower cost. The work for this chapter restricted the by Lagaros et al. (2010). Here, the total cost C(xd)
number of combinations to a maximum of n = 500. in Eq.(1) includes the initial construction cost
30
C0(xd) and the damage repair cost Cd(xd), assum- this damage occurs at a time t, the present value
ing that the structure will be completely repaired Cf0(DIES) is given by
after each earthquake that could occur during the
service life. C f 0 (DIES ) = C f (DIES ) exp(r t ) (17)
5.1 Initial Cost

in which r is the interest rate. The expected value
of Cf0(DIES) can be calculated once the probability
For a reinforced concrete structure, the initial
density function (PDF) for the time t is known.
cost is determined as a function of the concrete
For this, a Poisson arrival process is assumed,
volumes V and steel weights W in beams and
with mean arrival rate . Thus, the PDF for the
columns, and the unit costs of these items. Thus,
time t to the first earthquake event, t1, is
if CUC is the unit cost for concrete, and CUS is
the corresponding cost for steel, ft 1 (t ) = exp( t ) (18)
C 0 (xd ) = Vc,beam (xd ) + Vc,col (xd ) CUC + Ws , beam (xd ) +Ws , col (xd ) CUS
(16) which allows the calculation of the mean for the
time t1 as
The concrete volumen and steel weight are

functions of the design parameters xd. The unit 1
costs should include not only those for materials
t1 = t ft 1 (t ) dt =

(19)
0
but also for forming and labor. For the first appli-
cation example in this Chapter, unit costs of 1500
The probability P(t2 > t) that the time to the
$/m3 and 500 $/KN have been used respectively,
second event, t2, be greater than a time t must
for concrete and steel.
consider two posibilities: either that there are no
events before t or that there is only one. Thus,
5.2 Repair Costs
P (t2 > t ) = exp( t ) + t exp( t )
The repair costs, converted to present values,
(20)
depend on the level of damage inflicted during
the earthquakes, the uncertainty as to when these
from which the cumulative distribution Ft2(t) for
will occur and the number of earthquakes occur-
t2, and its PDF ft2(t) become
ring during the life time TD, and the interest rate
to be applied to repair funds accumulated from

the time of construction. P (t2 t ) = 1 (1 + t ) exp( t ) = Ft 2 (t ) ft 2 (t ) = 2 t exp( t )
The nonlinear dynamic analysis should include (21)

a parallel calculation of the damage accumulated
during the shaking. The analysis software used for The expected time to the second event is
the first application in this Chapter (Mller, 2001;
Mller and Foschi, 2003) quantifies the global 2
structural damage by means of an index DIES. If
t2 = t ft 2 (t ) dt =

(22)
0
Cf (DIES) is the function used to estimate the cost
of repairing a damage DIES, and assuming that
31
n
A similar reasoning allows the determination ( TD )
of the PDF for the arrival time tn of the nth event, P (XTD = n ) = exp( TD ) (27)
n!
and its mean value:
Finally,
n t n 1 n
ft n (t ) =
(n 1)!
exp( t ) tn = t ft n (t ) dt =

0
(23) n 1 i

TD ( T )n
Cd = C f (DIES ) i! t i exp ((r + ) t ) dt D
exp( TD )
n !
DIES
n =1 i =0 0
Under the assumption that the structure is fully (28)

repaired after each event, the total expected cost
(at present values and conditional on the damage In general, the costs Cn increase with n, but this
index DIES) becomes: is compensated by the diminishing probability of
occurrence of an increasing number of n events in
Cd = C 1 DIES + C 2 DIES + + C n TD. Thus, the outcome from Eq.(28) converges to a
DIES DIES
finite number as n is increased. In the application
(24)
results in this Chapter, the summation in Eq.(28)
was truncated when the relative contribution of
in which
the last term was less than 0.001.
TD The function relating the cost Cf to the damage
index DIES (which ranges between 0 and 1) was
C 1 DIES = C f0 (DIES ) ft1 (t ) dt
assumed to be of the form
0
TD
C 2 DIES = C (DIES ) ft2 (t ) dt b

DIES
f0
(25) C f (DIES ) = C 0 for DIES 0.60
0 0.60
C f (DIES ) = C 0 wheen DIES > 0.60
TD (29)
Cn DIES
= C f 0 (DIES ) ftn (t ) dt
0 in which C0 is the total replacement cost and the
coefficient = 1.20 is assumed to contemplate
Introducing Eq.(17) and (23) into Eq.(24) the extra costs for demolition and cleanup. Eq.(29)
final expected cost Cd|DIES, conditional on a dam- further assumes that complete replacement would
age level DIES, becomes be required should the damage index exceed a
value DIESLIM = 0.60. The exponent b needs to
i TD be set by calibration to practical experience, and
t i exp ((r + ) t ) dt
n 1

Cd DIES
= C f (DIES ) i! in the applications in this Chapter has been as-
i =0 0
(26) sumed to be b = 1.
Finally, the total expected repair cost must
Eq.(26) is still conditional on the number n of consider the PDF for the damage index DIES,
earthquakes during the service life TD.This number

is itself a random quantity and, assuming that it
obeys a Poisson distribution, the probability of n
C d (xd ) = C d DIES fDIES (DIES ) d(DIES )
0
events in TD is (30)
32
The PDF fDIES(DIES) in Eq.(30) is calculated The nonlinear dynamic analysis was then
by obtaining first, and then differentiating, the carried out for each of the combinations and for
cumulative distribution of DIES, using simulation each sub-combination, obtaining a database for
and the neural networks for the response DIES. the following response parameters:
This is represented by a Beta distribution, given
that the damage index is bounded by 0 and 1. UMAX: The maximum displacement at
In general, an earthquake does result in addi- the top of the portal
tional costs beyond those limited to the structure: DIST: The maximum inter-story drift
losses associated with contents or non-structural DILO: A maximum local damage index
components, cost for interruption of service, DIES: A maximum global damage index
insurance and losses for injury or casualties.
These additional costs can be taken into account Thus, to each of the response results Ri (i =
within the format proposed here, but have not 1, 450) obtained for the 450 combinations, cor-
been considered in the applications shown next responded NS = 10 results Rki (k = 1, 10) for the
in this Chapter. set of sub-combinations. These results can be used
to obtain the mean and the standard deviation of
each Ri over the set of secondary variables (differ-
6. APPLICATION EXAMPLE1: ent earthquake records and hysteretic properties):
OPTIMIZATION OF A SIMPLE
PORTAL FRAME 1 NS
1 NS
Ri = R ki Ri = (R ki Ri )2
NS k =1 NS 1 k =1
6. 1 Random Variables, (31)

Response Databases, and Neural
Network Representations The data for these statistics form two databases,
each with NP = 450 entries. Both databases are
The structure chosen for this application example represented by neural networks with the main
is a simple reinforced concrete portal frame, shown variables as input.
in Figure 3. The training of the networks used the OPT
The main input variables for the determination algorithm, with a subsequent attempt to improve
of the response databases are shown in Table 1 the solution by using a gradient-based approach
with their corresponding bounds. as described in Secion 4.3. This additional step
Within the bounds, 450 variable combinations only produced a small improvement in the results.
were generated by experimental design (Zhang, The agreement between the data T and the neural
2003). For each combination, 10 sub-combinations network outputs Y, are shown in Figure 4 for the
were generated for the following secondary vari- response DIST as a typical case. To quantify the
ables: (a) a set of random phase angles to gener- goodness of the regression, the linear correlation
ate an artificial earthquake record (Mller, 2001; coefficient YT is calculated,
Shinozuka, 1967), with the resulting record then
Y ) (Tk T )
Y T
scaled to the peak acceleration aG included in the
NP
1
Y T =
Y T
where Y T =
NP 1
(Y k
particular combination; and (b) the concrete and k =1
(32)
steel strength as they affect the variability in the
parameters for the hysteretic relationship moment-
The results in Figure 4 show a very good
curvature for beam and columns cross-sections
agreement and a correlation coefficient very close
(Mller et al., 2006).
33
Figure 3. Portal frame
to unity. Nevertheless, the lack of fit between the 6.2 Performance Functions
approximation F(X) and the actual data R(X) can and Neural Networks for
be quantified by calculating the standard deviation Reliability Estimates
of the relative error,
The performance function Gi for the ith-limit state
1 NP
Yk Tk 2 can be written, in general, as
r =
NP 1
(
k =1 Yk
) (33)
Gi (X) = RLIM i Ri (X) RLIM i Fi (X)
(35)
in which Yk is the value calculated by the network, Eqs.(36) through (43) show the performance
Tk is the target data from the nonlinear dynamic functions for the maximum relevant response
analysis and NP = 450, the size of the database. parameter and for three performance levels: op-
The neural network predictions can then be erational, life safety and collapse. Each of the
corrected to account for the lack of fit, as follows: capacity terms in these functions show, in paren-
theses, the mean and the coefficient of variation
F i = Y i ( 1. + m X N 1 ) Fi = Yi ( 1. + X N 2 ) (RLIM ,COVRL) . In G11, uy indicates the mean
(34)
yield displacement for the frame, below which
the structure remains elastic.
in which Y i , Yi are, respectively, the mean
value and the standard deviation calculated with Operational:
the corresponding neural network, and m , Elastic displacement limit
are the corresponding standard deviations of the G11 (X) = (uy , 0.10) UMAX (X) (36)
relative errors. XN 1, XN 2 are two additional random
variables associated with the fit error and assumed
to be Standard Normals.
34
Table 1. Application 1, main random variables, and definitions
Lower Upper
Variable Definition
bound bound
X(1) = m (KN s2 / cm2) 2.15x10-4 7.14x10-4 Applied mass per unit length
X(2) = bb (cm) 15 30 Width of the beam cross-section
X(3) = hb (cm) 40 70 Height of the beam cross-section
X(4) = bc (cm) 20 40 Width of the column cross-section
X(5) = hc (cm) 40 100 Depth of the column cross-section
X(6) = span 0.00298 0.01389 Steel reinforcement ratio, beam span
X(7) = end 0.00298 0.01389 Steel reinforcement ratio, over beam support
X(8) = col 0.008 0.04286 Steel reinforcement ratio, columns
X(9) = fr / fc0 0 0.15 Normalized concrete confinement pressure
X(10) = aG (cm / s2) 25 1200 Ground peak acceleration

X(11) = fg (Hz) 2 3 Soil central frequency
Interstory drift distribution over the sub-combinations. Thus, in

G12 (X) = (0.005, 0.10) DIST (X) (37) general,
Life safety: F (X) F (X)

F (X) = exp RN 1 ln(1. + ( )2 )
Interstory drift F (X) F (X)
1. + ( )2
G21 (X) = (0.015, 0.10) DIST (X) (38) F (X)
Maximum local damage index (44)
G22 (X) = (0.60, 0.10) DILO(X) (39)
Maximum global damage index in which RN1 is a Standard Normal variable that
G23 (X) = (0.40, 0.10) DIES (X) (40) introduces the variability over the sub-combina-
tions. F(X) and F (X) are, respectively, the mean
Collapse: and the standard deviation of the response calcu-
Interstory drift lated from the corresponding neural networks,
G 31 (X) = (0.025, 0.10) DIST (X) (41) using Eq.(34) to account for the regression error.
Maximum local damage index Eq. (44) is applied to each response, for example,
G 32 (X) = (1.00, 0.10) DILO(X) (42) whether F(X) is the maximum displacement
Maximum global damage index UMAX(X) or the maximum interstory drift
G 33 (X) = (0.80, 0.10) DIES (X) (43) DIST(X).
Since the damage indices DILO(X) and
In these performance functions, each response DIES(X) are limited between 0 and 1, the F(X)
demand F(X) is a function of the main random in these cases are represented by Beta distributions
variables X, and is represented with a lognormal with the same limits, with the mean F(X) and
standard deviation F (X) . These, in turn, are
35
Figure 4. Neural networks agreement for the response DIST
obtained from the corresponding neural networks. RLIM = RLIM ( 1. + COVRL X N 3 ) (45)
The Beta distribution for each of the damage
indices requires the introduction of an additional
A summary of the random variables, with
random variable, U1, uniformly distributed be-
their distribution types and parameters, is shown
tween 0 and 1.
in Table 2. The statistics for the peak ground ac-
The uncertainty in each of the capacities RLIM
celerations aG correspond to the seismicity for
is modeled with a normal distribution, introduc-
the city of Mendoza, Argentina, as provided by
ing an additional Standard Normal variable XN 3:
INPRES (1995).
Table 2. Random variables
Variable X X Type Variable X X Type
X(1) = m 5.1x10-4 5.1x10-5 Normal X(10) = aG 94 cm/s2 130cm/s2 Lognormal
X(2) = bb 20 cm 1 cm Normal X(11) = fg 2.50 Hz 0.375 Hz Normal
X(3) = hb ? cm 0.05 X Normal X(12) = aG 0 0.25 Normal
X(4) = bc 30 cm 1.5 cm Normal X(13) = aG X(13) = X(10) [1.0+X(12)]
X(5) = hc ? cm 0.05 X Normal X(14) = RN 1 0 1 Normal
X(6) = span ? 0.10 X Lognormal X(15) = U1 0 1 Uniform
X(7) = end ? 0.10 X Lognormal X(16) = XN 1 0 1 Normal
X(8) = col ? 0.10 X Lognormal X(17) = XN 2 0 1 Normal
X(9) = fr / fc0 0.10 0.01 Normal X(18) = XN 3 0 1 Normal
36
Each of the question marks (?) in this Table 6.3 Preliminary Design: Initial
indicates an optimization design parameter (the Values for the Design Parameters
mean values for the depth of the beams and col-
umns, and for the longitudinal steel reinforcement The preliminary design for the portal frame was
ratios). achieved according to the procedure described by
The reliability assigned to each performance Rubinstein et al. (2006), followed by application
level corresponds to a series system reliability for of the capacity design guidelines specified in the
all the performance functions G included in that Argentine norms INPRES-CIRSOC 103, Parte II
performance. Thus, the reliability for collapse cor- (2005). The results of this preliminary calculation
responds to that of a series system consisting of the are shown in Table 3, and provide the first anchor
three limit states given by Eqs.(41) through (43). combination xd0 for the optimization.
Databases for achieved reliabilities are devel- The design parameters satisfy additional con-
oped, as described, as a function of the design straints to provide adequate resistance to the
parameters, utilizing the performance functions gravitational loads on the top beam. Thus, fol-
and the response neural networks, These databases lowing Argentinas norm requirements,
are themselves represented by neural networks,
permitting the efficient estimation of reliabilities 1.4 q lb2
span + end (46)
achieved for any choice of input design parameters 6.48 bb hb2 fy
(within their bounds). The databases included
data from 180 design parameter combinations.
Neural network agreement with the input data is Furthermore, to ensure a minimum ductility in
shown in Figure 5 for the reliability ndices 1, 2, the beam cross-sections, the steel reinforcement
3 corresponding to the three performance levels. ratios satisfy
The dispersion or lack of agreement error is
represented by a normal distribution following 0.5 end span 2 end (47)
the same approach as described for Eq.(34). Fig-
ure 5 shows the very good approximation of the
reliability data by means of the trained neural
networks.
Figure 5. Neural network agreement with data for the reliability indices 1, 2, 3
37
Table 3. Initial values for the design parameters

Table 6. It is seen that the reliability levels for this
Design parameter Initial value design already satisfy the minimum values shown
in Tables 4 and 5. Starting from these preliminary
xd (1) = X(3) = hb [ cm ] 50 parameters, the optimization process was applied,
as described, using the target minimum reliabili-
xd (2) = X(5) = hc [ cm ] 45
ties from Table 4 (Paulay y Priestley, 1992). The
optimization took into account the bounds for
xd (3) = X (6) = span 0.00804
the design parameters (minimum and maximum
cross-sectional dimensions and steel reinforce-
xd (4) = X (7) = end 0.01143
ment ratios) specified in Table 1 and used in the
development of the response databases.
xd (5) = X (8) = col 0.03148
The final design parameters obtained are shown
in Table 7, with the final total cost and the reli-
ability levels achieved for each of the performance
6.4 Optimization Constraints: levels.
Minimum Reliability Levels It is interesting to compare the optimum solu-
tions obtained when the optimization is started
Minimum target reliabilities for each performance from conditions other than the preliminary design.
level are specified in terms of tolerable annual With this purpose, Table 8 shows five different
probabilities of non-performance, Pfannual. Using initial combinations of design parameters, two of
a Poisson arrival process, with a mean arrival them over-dimensioned in comparison to the
rate , the exceedence probability Pfannual can be preliminary design and two under-dimen-
converted to a probability of non-performance Pf sioned.
for the earthquake event, and finally expressed The evolution of the total cost during the op-
as a target event reliability index : timization, in terms of the number of steps, is
shown in Figure 6. Each step corresponds to one
Pfanual = 1. exp Pf Pf 1 (Pf ) evaluation of total cost with all reliability con-
(48) straints being satisfied. This figure shows that the
optimization converges to approximately the same
A mean arrival rate = 0.20, for earthquakes final total cost, regardless of the choice for the
with magnitudes M 5, was used for the city of initial design parameters. Those corresponding
Mendoza, Argentina. Table 4 shows annual and to the preliminary design provide a cost solution
event target probabilities as recommended by which is already quite close to the optimum,
Paulay and Priestley (1992). In this application, showing the adequacy of the methodology em-
more stringent targets were also considered, as ployed for the preliminary estimation.
shown in Table 5, to evaluate the sensitivity of When the optimization starts from an under-
the optimization outcome to the levels used for designed combination, the first cost that is calcu-
the targets. lated is controlled by the requirement to satisfy
the reliability constraints. It can be observed in
6.5 Optimization Results Figure 6 that the number of steps needed to achieve
the optimal solution varies, implying a varying
The total cost and the reliability levels correspond- number of anchor combinations and process
ing to the initial preliminary design are shown in repetitions. In each case, no lower total cost was
found when the optimization search radius r1 was
38
Table 4. Target minimum reliability levels (Paulay and Priestley, 1992)
Performance level Pf annual annual Pf

Operational 2 x 10-2
2.054 0.10101 x 10 0
1.276
Life safety 2 x 10-3
2.878 0.10010 x 10 -1
2.326
Collapse 2 x 10-4 3.540 0.10001 x 10-2 3.090
Table 5. Target minimum reliability levels proposed
Performance level Pf annual annual Pf

Operational 1 x 10-2
2.326 0.50252 x 10 -1
1.642
Life safety 1 x 10-3
3.090 0.50025 x 10 -2
2.576
Collapse 1 x 10-4 3.719 0.50003 x 10-3 3.291
expanded, indicating that at least a local minimum 6.6 Optimization Sensitivity to

had been found. Specified Minimum Target
The different initial conditions lead eventually Reliabilities
to essentially the same optimum total cost, but
the solutions contain different optimal combina- In this section we discuss the influence that the
tions of the individual parameters. This implies definition of the minimum reliability constraints
that the cost function may have different local may have in the optimum solution. Figure 7 shows
minima, but that in this case these minima are the evolution of the total cost either for the target
approximately of the same magnitude, with the reliabilities () in Table 4 (Paulay and Priestley,
largest difference in total cost being only of the 1992) or for those proposed in Table 5. In both
order of 2.1%. However, the difference in the cases the starting configuration is the preliminary
column steel reinforcement ratio, for example, design. Figure 8 shows higher total costs as a result
between using initially either the preliminary of imposing higher target reliability constraints,
design or the Under-design 1, is closer to 81%. as expected.
The final optimum results are shown in Table 9. It At the optimum solution, the constraints for
should be emphasized that all these different solu- the different performance levels are met but not
tions satisfy the minimum reliability constraints, to the same degree. For example, for the minimum
and that Table 9 implies a trade-off between the targets in Table 4, 1 = 1.276, 2 = 2.326 and 3 =
optimal depth of the beam, for example, and the 3.090, the final (optimum) reliability levels in
optimal amount of steel used. If the constraints Table 7 show 1 = 1.697, 2 = 2.422 and 3 = 3.102.
had been formulated in terms of specified narrow Thus, the constraint is just met for the collapse
reliability intervals, rather than just only minima, performance level, and satisfied with a greater
the number of optimum design possibilities would margin for performance levels 1 and 2.
have been correspondingly narrowed. Similar results are obtained using the proposed
target reliabilities from Table 5 (1 = 1.642, 2 =
2.576 and 3 = 3.291). In this case, the reliability
levels achieved at the optimum solution are 1 =
1.691, 2 = 2.749 and 3 = 3.296.
39
Table 6. Initial costs and reliabilities, preliminary design
Costs [$] Performance level

Initial C0(xd) 2839 Operational 1.837
Repair Cd(xd) 1113 Life safety 2.846
Total 3952 Collapse 3.489
We analyze now the effect of changing the 6.6 Optimization Sensitivity to the
more significant constraints, 2 and 3, using as a Assumed Damage-Cost Relationship
base the minimum targets specified in Table 4.
Thus, 2 and 3 are each individually changed by Finally we consider the effect on the optimization
10%, with the corresponding results shown in of the form used for the cost-damage relationship
Figure 8. (Eq.(29)). While maintaining b = 1, the relationship
When 2 is decreased by 10%, the results is modified by changing the level of damage index
are identical to those shown in Figure 7, since DIESLIMIT at which the structure is considered to be
constraint 2 is more easily met and constraint 3 destroyed and needs to be replaced. Thus, from the
remains the determinant one. On the other hand, value assumed for the base results, DIESLIMIT = 0.6,
if 2 is increased by 10%, constraints 2 and 3 now two different possibilities were considered: 0.4 or
become jointly important, leading to different 0.8. In the first, the cost increases more rapidly as
combinations of the design parameters but still a function of damage, while the reverse occurs in
with similar minimum total cost. the second case. The results are shown in Figure
If the target 3 is increased by 10%, the total 9, with corresponding differences of 15% and -7%
cost for the optimum solution also increases as with respect to the minimum total cost for the base
constraint 3 was already the most important. On the relationship. However, the optimum values for the
other hand, if 3 is decreased by 10%, constraints design parameters remain unaffected.
2 and 3 now have a similar influence, resulting
again in a similar minimum total cost but with a
different combination of design parameters.
Table 7. Optimum results starting from the preliminary design
Optimal Performance
Design parameter Costs [$]
value level
xd (1) = X(3) = hb [ cm ] 57.0 Initial C0(xd, ) 2601 Operational 1.697
xd (2) = X(5) = hc [ cm ] 40.3 Repair Cd(xd) 1020 Life safety 2.422
xd (3) = X (6) = span 0.00617 Total 3621 Collapse 3.102
xd (4) = X (7) = end 0.01028
xd (5) = X (8) = col 0.02130
40
Table 8. Different combinations of initial design parameters
Preliminary Over Over Under Under

Design parameter
design design 1 design 2 design 1 design 2
xd (1) = X(3) = hb [ cm ] 50 60 65 45 45
xd (2) = X(5) = hc [ cm ] 45 80 60 45 45
xd (3) = X (6) = spam 0.00804 0.01100 0.01300 0.00600 0.00390
xd (4) = X (7) = end 0.01143 0.01222 0.01011 0.00600 0.00390
xd (5) = X (8) = col 0.03148 0.03950 0.03950 0.00950 0.00890
7. APPLICATION EXAMPLE 2: amount . The problem is the assessment of the

OPTIMIZATION OF THE PILE-CAP probability that the displacement (the response
MASS FOR A PILE FOUNDATION, of interest) will exceed given levels, expressed
AND SENSITIVITY TO THE ANALYSIS as fractions of the pile diameter D. Furthermore,
MODEL USED the mass M is to be optimized so that the prob-
ability of the displacement exceeding a target
The second application example illustrates the be as specified.
influence that the modeling approach may have The performance function G is defined as
in the optimization results. With this objective,
Figure 10 shows a pile steel tube, of diameter D, G (X , d ) = D - (aG , S ,T , M , DR , r )
wall thickness t and length L, supporting a pile-cap (49)
mass M. The pile is embedded into a sandy soil
layer with relative density DR. Under earthquake in which
excitation, the mass will displace a maximum
Figure 6. Evolution of the total cost for different initial design parameters
41
Table 9. Optimum results starting from different initial design parameters as per Table 8
Preliminary Over Over Under Under

Design parameter
design design 1 design 2 design 1 design 2
xd (1) = X(3) = hb [ cm ] 57.0 58.4 56.5 53.6 58.0
xd (2) = X(5) = hc [ cm ] 40.3 41.1 40.3 45.6 40.4
xd (3) = X (6) = span 0.00617 0.00586 0.00813 0.00711 0.00644
xd (4) = X (7) = end 0.01028 0.01122 0.01300 0.01303 0.01051
xd (5) = X (8) = col 0.02130 0.01672 0.01553 0.01152 0.01814
Initial cost C0(xd, ) [$] 2601 2569 2559 2546 2568

Repair cost Cd(xd) [$] 1020 1008 1004 999 1007
Total cost [$] 3621 3577 3563 3545 3575
= capacity factor, fraction of D defining the The resulting model, developed for the similar
limit displacement; problem of a metal fastener in wood (Foschi, 2000)
aG = peak ground acceleration; depends solely on mechanical properties of the
S = soil frequency in the Kanai-Tajimi Power pile and the soil, and produces the hysteretic loop
Spectral Density function; for any input excitation, automatically developing
T = duration of the strong motion part of the ac- the pinching and degradation characteristics. The
celerogram record; p(w) relationship used here was taken from Yan
M = applied mass; and Byrne (1992),
DR = soil relative density;
r = nominal variable representing the different E w if w 2D
max
ground motion records. p (w ) = 0. 5

E D w if w > 2D
max d
The pile is considered as an elasto-plastic
beam on a nonlinear foundation. The pressure on (50)
the soil, p(w), is a function of the displacement w
which varies along the length L. in which = 0.5 (DR)-0.8 and DR is the soil relative
The structural analysis for the pile-cap dis- density. The modulus Emax depends on the specific
placement can be done by considering the weight of the soil and the depth of the soil layer
dynamic equilibrium of the mass M as a single and it is detailed by Yan and Byrne (1992). In this
degree of freedom system. Using a beam finite work only the relative density was considered to
element model of the pile, the restoring hysteretic be a random variable.
force F() can be calculated by integration of The calculation of can be integrated directly
the reactions p(w), after the deflected shape w is into the reliability calculation using the perfor-
determined. The nonlinear, compression-only, soil mance function in Eq.(49) or into the optimization
reactions p(w) take into account the development for the mass M which can be sustained to match a
of gaps between the pile and the surrounding soil. prescribed reliability level. As a modeling alter-
42
Figure 7. Evolution of the total cost for different minimum target betas
native, a different formulation could be used in good fit can be obtained for the cyclic response
describing the hysteresis behavior. Thus, it could used for the matching, there is no guarantee that
be matched to the response of the pile to a given a similar good representation would be achieved
cyclic displacement history for the pile-cap. This when using the same fitted rules for any other
history, obtained by either experiments or by cal- cyclic history or earthquake record. Perhaps the
culation, can then be fitted with a set of rules that most sophisticated of these calibrated approaches
permit the reproduction of the hysteresis observed to hysteresis representation is to use cyclic data to
for the specific cyclic displacements. determine parameters of a first-order differential
Many such set of rules are contained in dynamic equation, the solution of which, for a given history,
analysis software packages. However, although a can represent loops with pinching and degradation
Figure 8. Influence of 2 and 3 in the total cost
43
Figure 9. Influence of the cost-damage relationship used
characteristics. This model is commonly known previously discussed. Finally, the response in
as BWBN (Bouc, 1967; Baber et al., 1981, 1985). Eq.(49) was represented by using the lognormal
The objective in this second application is a format and the error representation as shown in
comparison of the influence that the two differ- Eqs.(44) and (34).
ent hysteretic modeling techniques have in either The pile was subjected to the displacement
reliability assessment or structural optimization. history (t) in Figure 11. The finite element ap-
The pile has a diameter D = 0.356m, with proach was used to calculate the hysteresis loop in
wall thickness t = 0.10m, and a length L = 30m. Figure 12. This response was used to calibrate the
Yield strength and elastic modulus of the steel parameters of a BWBN model, with the resulting
were assumed deterministic and to have nominal loop shown in Figure 13.
values (respectively, 250 Mpa and 200000 Mpa). Dynamic analyses for the different earthquake
Twenty earthquake records were simulated as sta- records were carried out with either the BWBN
tionary processes using a spectral representation representation of the hysteretic restoring force or,
based on the Clough-Penzien power spectrum alternatively, calculating each time the hysteresis
density function (Clough and Penzien, 1975), an via the finite-element model. In both cases, the
envelope modulation function (Amin and Ang, neural network methodology previously described
1968), and twenty different sequences of random was applied.
phase angles. Table 10 shows the statistical data assumed
For different combinations of the intervening for the intervening variables, and Table 11 the
variables, databases were constructed for the reliability results obtained for different values of
mean response and its standard deviation over the capacity factor .
the twenty records. These databases were then The parameters S and T define, respectively,
used to train corresponding neural networks, as the Clough-Penzien power spectrum density func-
44
Figure 10. Pile foundation under earthquake

parameter is the mean value of the mass M, al-
excitation a(t)
lowing for a coefficient of variation of 0.10. The
results are shown in Table 12, comparing again
results corresponding to the two approaches for
analysis. It is seen that the model used has now a
rather substantial effect on the optimal solution.
8. CONCLUSION
This Chapter has considered different is-

sues involved in performance-based design
and optimization in earthquake engineer-
ing, using performance criteria specified in
terms of displacements and damage. The
study has led to a proposal for a systematic
method of analysis, integrating the ground
motion, a nonlinear dynamic analysis of
the problem, neural network representation
for the calculated responses and for the
achieved reliabilities and, finally, a simple,
search-based, optimization algorithm to
minimize the total cost under minimum re-
liability constraints.
tion and the duration T of the strong motion in Earthquake engineering design normally
the modulation function. The statistics for the follows procedures and prescriptions given
peak ground acceleration aG correspond to the in Code guidelines. A design earthquake is
event occurrence, and are consistent with a design used, which has been chosen with a rela-
acceleration (475 years return period) of 0.31g, tively low probability of being exceeded
assuming that earthquakes have a Poisson arrival (e.g., 0.02 in 50 years). The calculated
rate of 0.2 (one every five years). structural responses for this earthquake
Although the agreement between Figures 12 are then modified by a set of coefficients
and 13 may be considered to be satisfactory, Table which should be calibrated to lead to a
11 shows that the calculated reliabilities, if not desired performance reliability. For rein-
drastically different, clearly depend on the model forced concrete structures, this preliminary
used for the analysis. design includes the determination of cross-
A similar conclusion may be reached if an sectional dimensions and steel reinforce-
optimization is carried out to determine the mass ments using a capacity design approach.
M that, optimally, will satisfy two performance The work presented in this Chapter con-
criteria: a displacement associated with moderate sidered the assessment of the reliability
damage level, with = 0.40, and another associ- achieved for different performance crite-
ated with more substantial damage, = 1.0. Target ria, and the possibility of optimizing the
reliability levels are prescribed: = 2.5 for the first design in terms of the total cost, includ-
criterion, and = 4.5 for the second. The design ing construction and the repair of damage
45
Figure 11. Cyclic displacement history
caused by an uncertain number of earth- by continuous, smooth functions, and this

quakes during the service life. Chapter has discussed the use of neural
The performance criteria must involve networks as an efficient approach to re-
maximum structural displacement repons- sponse representation.
es and the associated damage, the calcula- The random variables have been divided
tion of which requires a nonlinear dynamic into two groups: a first set of main vari-
analysis. This produces discrete values for ables and a second set of secondary ones
the responses for different combinations including those associated with the ground
of the intervening random variables. For motion. An approach has been presented
reliability assessment and optimization, by means of which the analysis data are
these discrete results must be represented the mean and the standard deviation of
Figure 12. Loop for cyclic displacement history, finite element approach
46
Figure 13. BWBN model regression for cyclic displacement history
the responses over the set of secondary mance criteria. The Chapter has described
variables. These means and standard de- a search-based algorithm for the optimiza-
viations are then represented by neural tion, not requiring the calculation of gra-
networks with the main variables as the dients. The optimization requires initial
input. This approach facilitates the use of values for the design parameters, and the
simulation in the estimation of the non- work has considered initial values to be
performance probabilities, at a great com- those from a preliminary design. Further,
putational saving. the Chapter considered the sensitivity of
The reliability indices associated with the optimal solution to the initial condi-
the performance criteria are thus calculated tions, with the conclusion that the final
in terms of the design parameters, and neu- result for total cost was essentially insensi-
ral networks for the reliabilities are trained tive, but that the corresponding individual
with those parameters as input. design parameters could be quite different.
The design optimization is carried out for
minimum total cost with minimum reli-
ability constraints for each of the perfor- Table 11. Reliability results and comparison be-
tween hysteretic representation approaches
Table 10. Statistical data for the intervening Limit

displacement, Reliability index
variables capacity factor
Hysteresis: Finite Hysteresis:
Standard
Variable Distribution Mean Element BWBN
Deviation
0.1 -0.143 0.716
aG (m/sec2) Lognormal 1.0 0.6
0.2 1.097 1.724
S (rad/sec) Normal 4
0.4 2.509 2.379
T (sec) Normal 12 2
0.6 3.197 2.675
M (kN.
Normal 150 15
sec2/m) 0.8 3.730 2.892
DR Normal 0.5 0.1 1.0 4.243 3.082
47
Table 12. Performance-based design
Hysteresis: Finite element

Optimal Mean Mass (kN.sec /m)
2
Performing criterion Target reliability Achieved reliability
0.4 2.5 2.59
139.83
1.0 4.5 4.51
Hysteresis: BWBN
Optimal Mean Mass (kN.sec2/m) Performing criterion Target reliability Achieved reliability
102.19 0.4 2.5 2.76
1.0 4.5 4.42
The optimization is then a problem which gree with which each minimum reliability
could admit multiple solutions. In practi- constraint is met may vary, the relationship
cal terms, the design engineer has then between the optimal solution and the pre-
the possibility of choosing from differ- scribed targets is not simple. Increasing or
ent alternatives, taking into account other decreasing a single target may affect the
practical requirements. It is also possible importance of another constraint, and the
to add to the optimization additional con- total cost for the optimal solution may or
straints among the design parameters, ori- may not be affected.
enting the optimum solution towards, for The assessment of reliability, and the op-
example, desired mechanisms for energy timization itself, are conditional on the
dissipation. quality of the response analysis model,
Data are required on the relationship be- particularly the quantification of the hys-
tween a calculated damage and the corre- teretic energy dissipation during the shak-
sponding repair cost. There is substantial ing. In order to illustrate this influence, a
uncertainty in this relationship, and the second application example has been dis-
Chapter has also explored the sensitivity cussed involving the earthquake response
of the optimal solution to the form of the of a steel pile foundation in a sandy soil.
damage-cost function. For the first applica- Two hysteretic representations are used:
tion example, it has been shown that there one implementing a calculation of the
is a significant influence in the optimum soil-structure interaction and the hyster-
minimum total cost, while the optimum etic force by means of a nonlinear finite
design parameters are not affected. Further element model; the other using a hysteresis
research is therefore needed to quantify the model fitted to results from a cyclic dis-
relationship, a task that must be guided by placement of the pile-cap. While the first
experiments and expert assessments. model can self-adapt to different excitation
The Chapter has also explored the sensitiv- histories or different earthquakes, the sec-
ity of the optimal solution to the minimum ond approach can only be said to provide
target reliability levels prescribed for each a good representation for the history to
performance criteria. In general, it can be which it was fitted. The example consid-
concluded that the results show higher total ered the question of how significant is the
costs as the minimum targets are increased. influence of these different analyses on the
This is expected. However, since the de- reliability and optimization results. It was
48
concluded that indeed there is an influence, REFERENCES

pointing to the use of an analysis as accu-
rate as possible, or a careful evaluation of a Amin, M., & Ang, A. H.-S. (1968). Non-stationary
model error to be used in conjunction with stochastic models of earthquake motions. Proceed-
the approximate approach. This is contrary ings ASCE, Journal of Engineering Mechanics
to the common assumption that structural Division, 94.
or hysteretic modeling is not so important Baber, T., & Noori, M. N. (1985). Ran-
because any uncertainty associated with it dom vibration of degrading, pinching sys-
is overwhelmed by the uncertainty in the tems. Journal of Engineering Mechanics,
ground motion. 111(8), 10101026. doi:10.1061/(ASCE)0733-
Sometimes it is proposed that reliabil- 9399(1985)111:8(1010)
ity assessment in earthquake engineering
should split the capacity from the hazard Baber, T., & Wen, Y. K. (1981). Random vibra-
demand. Thus, in this alternate approach, tion of hysteretic degrading systems. Journal
the total probability of non-performance of Engineering Mechanics, 107(6), 10691089.
would be obtained by integrating the vul-
Bouc, R. (1967). Forced vibration of mechanical
nerability, or probability of non-perfor-
systems with hysteresis. Proceedings of the 4th
mance conditional on a hazard level, with
Conference on Nonlinear Oscillations, Prague,
the probability density for the hazard. This
Czechoslovakia, Sept. 5-9.
approach has not been considered in this
Chapter because different vulnerabilities Clough, R. W., & Penzien, J. (1975). Dynamics
would have to be calculated each time that of structures. McGraw-Hill.
the design parameters are changed during
Foschi, R. O. (2000). Modeling the Hysteretic
the optimization process, with no savings
response of mechanical connectors for wood
in computational effort when compared to
structures. Proceedings, World Timber Engineer-
the approach proposed in this chapter.
ing Conference, Whistler, B.C.
Hurtado, J. (2004). Structural reliability Sta-
ACKNOWLEDGMENT tistical learning perspectives. Lecture Notes
in Applied and Computational Mechanics, 17.
This work was supported by grants from the Springer Verlag.
Research Proyect Diseo y optimizacin de
INPRES. (1995). Microzonificacin Ssmica del
estructuras bajo acciones dinmicas aleatorias,
Gran Mendoza. Instituto Nacional de Prevencin
1ING 300, Universidad Nacional de Rosario, Ar-
Ssmica, Publicacin Tcnica N 19.
gentina, and Neural networks for reliability and
performance-based design in earthquake engineer- INPRES-CIRSOC 103. (1995). Reglamento Ar-
ing, Natural Sciences and Engineering Research gentino para construcciones sismorresistentes,
Council of Canada: RGPIN 5882-04, University Parte II: Construcciones de Hormign Armado,
of British Columbia, Vancouver, Canada. INTI.
49
Lagaros, N. D., Naziris, I. A., & Papadrakakis, M. Mller, O., Foschi, R., Rubinstein, M., & Quiroz,
(2010). The influence of masonry infill walla in L. (2007). Optimizacin de prticos sismorresis-
the Framework of the performance-based design. tentes utilizando redes neuronales y algoritmo
Journal of Earthquake Engineering, 14, 5779. sin clculo de gradientes. [AMCA.]. Mecnica
doi:10.1080/13632460902988976 Computacional, 26, 18241839.
Mller, O. (2001). Metodologa para evaluacin Mller, O., Foschi, R., Rubinstein, M., & Quiroz,
de la probabilidad de falla de estructuras sismor- L. (2009b). Seismic structural reliability using
resistentes y calibracin de cdigos. Tesis de different nonlinear dynamic response surface ap-
Doctorado en Ingeniera, Universidad Nacional proximations. Structural Safety, 31(5), 432442.
de Rosario. doi:10.1016/j.strusafe.2008.12.001
Mller, O., & Foschi, R. (2003). Reliability evalu- Paulay, T., & Priestley, M. J. N. (1992).
ation in seismic design: A response surface meth- Seismic design of reinforced concrete and
odology. Earthquake Spectra, 19(3), 579603. masonry buildings. John Wiley & Sons, Inc.
doi:10.1193/1.1598200 doi:10.1002/9780470172841
Mller, O., Foschi, R., Quiroz, L., & Rubinstein, Prez Lpez, J. R. (2005). Contribucin a los
M. (2008). Optimizacin de prticos con accio- mtodos de optimizacin basados en procesos
nes ssmicas: diferentes estrategias numricas naturales y su aplicacin a la medida de antenas
utilizando redes neuronales. [AMCA.]. Mecnica en campo prximo. Retrieved from http://www.
Computacional, 28, 25832603. tesisenred.net/TDR-0305107-180847
Mller, O., Foschi, R., Quiroz, L., & Rubinstein, Rubinstein, M., Giuliano, A., & Mller, O. (2006).
M. (2009a). Performance-based seismic optimiza- Diseo preliminar de estructuras sismorresis-
tion implementing neural networks. In Frangopol, tentes: un tratamiento unificado de los efectos
D. M. (Ed.), Computational structural dynamics traslacionales y rotacionales. Memorias XIX Jor-
and earthquake engineering, structures and in- nadas Argentinas de Ingeniera Estructural, 49.
frastructures series (Vol. II, pp. 547564). Taylor
Shinozuka, M., & Sato, Y. (1967). Simulation
& Francis Group.
of nonstationary random processes. Journal of
Mller, O., Foschi, R., Quiroz, L., & Rubinstein, M. Engineering Mechanics, 93(1), 1140.
(2009c). Structural optimization for performance-
Swisher, J. R., Hyden, P. D., Jacobson, S. H., &
based design in earthquake engineering: Applica-
Schruben, L. W. (2000). A survey of simulation
tions of neural networks. Structural Safety, 31(6),
optimization techniques and procedures. In J. A.
490499. doi:10.1016/j.strusafe.2009.06.007
Joines, R. R. Barton, K. Kang, & P. A. Fishwick
Mller, O., Foschi, R., Rubinstein, M., & Quiroz, (Eds.), 2000 Winter Simulation Conference.
L. (2006). Momento-curvatura de secciones de
Yan, L., & Byrne, P. M. (1992). Lateral pile re-
hormign armado sismorresistentes utilizando
sponse to monotonic loads. Canadian Geotechni-
redes neuronales. [AMCA.]. Mecnica Computa-
cal Journal, 29, 955970. doi:10.1139/t92-106
cional, 25, 21452162.
Zhang, J. (2003). Performance-based seismic
design using designed experiments and neural
networks. PhD Thesis, Department of Civil Engi-
neering, University of British Columbia, Canada.
50
51
Chapter 3
Discrete Variable
Structural Optimization of
Systems under Stochastic
Earthquake Excitation
Hctor Jensen
Universidad Tecnica Federico Santa Maria, Chile
Marcos Valdebenito
Juan Seplveda
Luis Becerra
ABSTRACT
The reliability-based design optimization of structural systems under stochastic excitation involving discrete
sizing type of design variables is considered. The design problem is formulated as the minimization of an
objective function subject to multiple reliability constraints. The excitation is modeled as a non-stationary
stochastic process with uncertain model parameters. The problem is solved by a sequential approximate
optimization strategy cast into the framework of conservative convex and separable approximations. To
this end, the objective function and the reliability constraints are approximated by using a hybrid form
of linear, reciprocal, and quadratic approximations. The approximations are combined with an effective
stochastic sensitivity analysis in order to generate explicit expressions of the reliability constraints in
terms of the design variables. The explicit approximate sub-optimization problems are solved by an ap-
propriate discrete optimization technique. Two example problems that consider structures with passive
energy dissipation systems under earthquake excitation are presented to illustrate the effectiveness of
the approach reported herein.
DOI: 10.4018/978-1-4666-1640-0.ch003
Discrete Variable Structural Optimization of Systems
INTRODUCTION of physical systems characterized by uncertainty

models of small and medium sizes (less than 30
In a large number of practical design situations random variables).
the design variables must be selected from a list In this work attention is directed to discrete
of discrete values. Standard methods address reliability-based optimization of structural sys-
discrete variable design optimization problems tems under stochastic excitation. In particular,
by employing discrete or integer variable algo- excitation models defined in terms of hundreds
rithms to treat the problem directly in the primal or thousands random variables (high dimensional
variable space (branch and bound techniques, models) are considered here. One of the difficulties
combinatorial methods, evolution-based optimiza- in this type of problems is the high computational
tion techniques, etc.) (Kovcs, 1980; Goldberg, cost involved in the reliability analyses required
1989; Scharage, 1989).These methods are quite during the optimization process. This is due to the
general but are associated with a large number fact that the reliability estimation of stochastic
of function calls (evaluation of objective and dynamical systems involves the estimation of
constraint functions). On the other hand, in many failure probabilities in high-dimensional uncertain
practical applications the applied loads and system parameter spaces (Jensen et al., 2009; Valdebenito
parameters may be subjected to uncertainty or and Schuller, 2011). The objective of this work
variability. Under uncertain conditions the field is to present a general framework for solving this
of reliability-based optimization (RBO) provides class of reliability-based optimization problems.
a realistic and rational framework for structural The approach is based on the use of approximation
design and optimization (Enevoldsen and S- concepts, the application of an effective stochastic
rensen, 1994). Schemes for discrete structural sensitivity analysis, and the implementation of a
optimization considering uncertainties have not globally convergent optimization scheme. Special
been addressed as frequently as their deterministic attention is focused on the analysis and design of
counterpart. In most studies, ad hoc optimization structures protected by means of passive energy
algorithms have been integrated directly with dissipation systems. The basic function of these
a reliability method. For example, in (Hassan devices when incorporated into the structure is
and Crossley, 2008), the problem of optimiza- to absorb a portion of the input energy, thereby
tion under uncertainty has been approached by reducing energy dissipation demand on structural
means of genetic algorithms and direct Monte members and minimizing possible structural dam-
Carlo simulation. As both the optimization and age. In this regard, the consideration of discrete
the reliability algorithms require a large number optimal design processes for protected structural
of function calls, numerical costs can be very systems is one of the novel aspects of this work.
high. In order to reduce the computational efforts This type of problem is relevant from a practical
related to reliability analysis, the application of point of view since the potential advantages of
optimization methods in combination with the modern structural protective systems have lead
first-order reliability methods was investigated to the design and construction of an increasing
in, for example, (Tolson et al, 2004, Gunawan number of protected structures for the purpose
and Papalambros, 2007). Another approach for of mitigating seismic impact.
reducing computational efforts is the application The reliability-based optimization problem is
of meta-models, such as artificial neural networks formulated as the minimization of an objective
(see, e.g. (Papadrakakis and Lagaros, 2002; function subject to multiple reliability constraints.
Lagaros et al., 2008)). These approaches have All uncertainties involved in the problem (system
been applied to solve RBO problems of a series parameters and loading) are considered explicitly
52
during the design process. The basic mathematical Finally, two numerical examples that consider
programming statement of the structural optimi- structures with passive energy dissipation systems
zation problem is converted into a sequence of as structural protective systems are presented.
explicit approximate primal problems. For this
purpose, the objective function and the reliability
constraints are approximated by using a hybrid FORMULATION
form of linear, reciprocal and quadratic approxi-
mations. An approximation strategy based on an Consider a structural optimization problem de-
incomplete quadratic conservative approximation fined as the identification of a vector {y } of design
is considered in the present formulation (Groen- variables that minimizes an objective function,
wold et al., 2007; Prasad, 1983). An adaptive that is
Markov Chain Monte Carlo procedure, called
subset simulation (Au and Beck, 2001), is used Minimize f ({y }) (1)
for the purpose of estimating the failure prob-
abilities. The information generated by subset subject to design constraints
simulation is also used to estimate the sensitivity
of the reliability constraints with respect to the h j ({y }) 0 , j = 1, ..., nc (2)
design variables. The above information is com-
bined with an approximation strategy to generate
explicit expressions of the objective and reliability where h j represents a constraint function defined
constraints in terms of the design variables. The in terms of reliability measures, and nc is the
explicit approximate primal problems are solved number of constraints. As previously pointed out
either by standard methods that treat the problem a pure discrete variable treatment of the design
directly in the primal variable space (Goldberg, problem is considered here. Thus, the side con-
1989; Kovcs, 1980; Scharage, 1989; Tomlin, straints for the discrete design variables are writ-
1970) or by dual methods (Fleury and Braibant, ten as
1986; Haftka and Grdal, 1992; Jensen and Beer,
2010). The proposed optimization scheme exhibits yi Yi = {yil , l = 1, ..., ni }, i = 1, ..., nd (3)
monotonic convergence that is, starting from an
initial feasible design, the scheme generates a
where the set Yi represents the available discrete
sequence of steadily improved feasible designs.
values for the design variable yi , listed in ascend-
This ensures that the optimal solution of each ap-
proximate sub-optimization problem is a feasible ing order, and nd is the number of design variables.
solution of the original problem, with a lower It is assumed that the available values are distinct
objective value than the previous cycle. and they correspond to quantities such as cross
The structure of the chapter is as follows. First, sectional areas, moments of inertia, etc. The par-
the design problem considering discrete sizing ticular quantity to be used depends on the problem
type of design variables is presented. Next, the at hand. It is noted that alternative formulations
structural and excitation models are discussed in to the one proposed here should be considered
detail. The solution strategy of the problem in the for cases where the possible values of the discrete
framework of conservative convex and separable design variables are linked to a number of proper-
approximations is then discussed. This is followed ties simultaneously, for example, section groups
by the consideration of some implementation is- in steel structures.
sues such as reliability and sensitivity estimation.
53
The objective function is defined in terms of formance functions can also be considered in the
initial, construction, repair, or downtime costs. above formulation.
In the context of reliability-based optimization
of structural systems under stochastic excitation
a reliability constraint can be written as STRUCTURAL MODEL
h j ({y }) = PFj ({y }) PF*j 0 (4) A quite general class of dynamical systems under
ground acceleration can be cast into the follow-
ing form
where PFj ({y }) is the probability of occurrence
of the failure event Fj evaluated at the design {y } [M ]{x(t )} + [C ]{x(t )} + [K ]{x (t )} =
, and P is the corresponding target failure prob- [M ]{g }xg (t ) {f ({x (t )}, {z (t )})} (6)
*
Fj
ability. The failure probability function PFj ({y })

evaluated at the design {y } can be expressed in where {x (t )} denotes the displacement vector of
terms of the multidimensional probability integral dimension n , {x(t )} the velocity vector, {x(t )}
the acceleration vector, {f ({x (t )}, {z (t )})} the
PFj ({y }) =
Fj ({y })
q({})p({})d {}d {} vector of non-linear restoring forces, {z (t )} the
(5) vector of a set of variables which describes the
state of the nonlinear components, and xg (t ) the
where Fj ({y }) is the failure domain correspond- ground acceleration. The matrices [M ] , [C ] , and
ing to the failure event Fj evaluated at the design [K ] describe the mass, damping, stiffness, respec-
{y } . The vectors {}, i , i = 1, ..., nu , and tively. The vector {g } couples the base excitation
{}, i , i = 1, ..., nT represent the vector of uncer- to the degrees of freedom of the structure. All
these matrices are assumed to be constant with
tain structural parameters and random variables
respect to time. This characterization of the non-
that specify the stochastic excitation, respec-
linear model is particularly well suited for cases
tively. The uncertain structural parameters {}
where most of the components of the structural
are modeled using a prescribed probability den-
system remain linear, and only a small part behaves
sity function q({}) while the random variables
in a nonlinear manner. For a more general case,
{} are characterized by a probability density
the formulation can still be applied at the ex-
function p({}) . The failure probability function
penses of more computational efforts due to pos-
PFj ({y }) accounts for the uncertainty in the system
sibly necessary updating of the damping and
parameters as well as the uncertainties in the stiffness matrices of the system with respect to
excitation. It is noted that for structural systems time. In general, the matrices involved in the
under stochastic excitation the multidimensional equation of motion depend on the vector of design
integral (5) involves in general a large number of variables and uncertain system parameters and
uncertain parameters (in the order of thousands). therefore the solution is also a function of these
Therefore the reliability estimation for a given quantities. The evolution of the set of variables
design constitutes a high dimensional problem {z (t )} is described by a first-order non-linear
which is extremely demanding from a numerical differential equation
point of view. Finally, it is noted that constraint
functions defined in terms of deterministic per- {z(t )} = {h({x (t )}, {x(t )}, {z (t )})} (7)
54
Box 1.
1
{z ( j +1) (t + t )} = {z (t )} + [{k1( j +1) } + 2{k2( j +1) } + 2{k3( j +1) } + {k4( j +1) }] (10)
6
where {h } is a non-linear vector function that Equation (8). The iteration starting with solving
characterizes the behavior of the non-linear com- Equation. (8) needs to be repeated until the norm
ponents. From the last equation it is seen that the of the vector {z (t + t )} is in two consecutive
set of variables {z (t )} is a function of the displace- iterations sufficiently close. Numerical experience
ments {x (t )} and velocities {x(t )} , that is, shows that in general only few iterations are re-
{z ({x (t )}, {x(t )})}. quired within each time interval [t + t ].The
solution of the equation for the evolution of the
Numerical Integration set of variables {z } is obtained by a modified
Runge-Kutta method of fourth order. The solution
Since the differential equation that satisfy the at time t + t is written in Box 1, 2, and 3.
variables {z (t )} is non-linear in terms of the re- For actual implementation, the characterization
sponse {x (t )} , Equations. (6) and (7) must be of the vector of non-linear restoring forces is
solved in an iterative manner. Equation (6) is modeled in local component specific coordinates
solved first by any suitable step-by step integra- (local displacements and velocities) with a mini-
tion scheme, leading from the solution at time t mal number of variables. Therefore, the relation-
at the one at timet + t, that is ships given by Equations (11) and (12) are
evaluated in local displacements and velocities
[M ]{x(t + t )} + [C ]{x(t + t )} + [K ]{x (t + t )} =
increasing in this manner the efficiency of the
[M ]{g }xg (t + t ) {f ({x (t + t )}, {z (t + t )})} above numerical integration scheme.
(8)
It is seen that in order to compute the solution EARTHQUAKE EXCITATION MODEL

at timet + t , provided that the solution at time
t is known, the value of {z ()} is required at time The ground acceleration is modeled as a non-
t + t . To this end, at the beginning of the it- stationary stochastic process. In particular, a
eration within the time interval[t, t + t ], it is point-source model characterized by the moment
assumed that ( j = 0 ) magnitude M and epicentral distance r is con-
sidered here (Atkinson and Silva, 2000; Boore,
{z ( j ) ({x (t + t )}, {x(t + t )})} = {z ({x (t )}, {x(t )})} 2003). The model is a simple, yet powerful, means
(9) for simulating ground motions and it has been
successfully applied in the context of earthquake
The solution of Equation. (8), gives the re- engineering. The time-history of the ground ac-
sponses {x ( j +1) (t + t )} and {x ( j +1) (t + t )} . celeration for a given magnitude M and epicen-
Then, the nonlinear differential Equation (7) can tral distance r is obtained by modulating a white
be integrated to obtain new estimates for noise sequence by an envelope function and sub-
{z ({x (t + t )}, {x(t + t )})}, and the right sequently by a ground motion spectrum through
hand side {f ({x (t + t )}, {z (t + t )})} in the following steps:
55
Box 2.
{k1( j +1) } = t {h({x (t )}, {x(t )}, {z (t )})}

1
{k2( j +1) } = t {h({x ( j +1) (t + t / 2)}, {x ( j +1) (t + t / 2)}, {z (t )} + {k1( j +1) })}
2 (11)
1
{k3( j +1) } = t {h({x ( j +1) (t + t / 2)}, {x ( j +1) (t + t / 2)}, {z (t )} + {k2( j +1) })}
2
{k4( j +1) } = t {h({x ( j +1) (t + t )}, {x (k +1) (t + t )}, {z (t )} + {k3( j +1) })}
Box 3.
{x (t )} + {x ( j +1) (t + t )} {x(t )} + {x ( j +1) (t + t )}

{x ( j +1) (t + t / 2)} = , {x ( j +1) (t + t / 2)} = (12)
2 2
1. Generate a discrete-time Gaussian e(t, M , r ) and the ground acceleration spec-

white noise sequence trum S ( f , M , r ) are provided in the subse-
(t j ) = I / t j , j = 1, ..., nT , where quent sections.
j , j = 1, ..., nT , are independent, identi-
The probabilistic model for the seismic hazard
cally distributed standard Gaussian
at the emplacement is complemented by consid-
random variables, I is the white noise
ering that the moment magnitude M and epicen-
intensity, t is the sampling interval, and
tral distance r are also uncertain. The uncer-
nT is the number of time instants equal to
tainty in moment magnitude is modeled by the
the duration of the excitation T divided
Gutenberg-Richter relationship truncated on the
by the sampling interval
interval [M min , M max ], which leads to the prob-
2. The white noise sequence is modulated by
an envelope function e(t, M , r ) at the discrete ability density function (Kramer, 2003)
time instants
3. The modulated white noise sequence is b e bM
p(M ) = , M min M M max
transformed to the frequency domain e 6.0b e 8.0b
(13)
4. The resulting spectrum is normalized by the
square root of the average square amplitude
where b is a regional seismicity factor. For the
spectrum
uncertainty in the epicentral distance r, a lognor-
5. The normalized spectrum is multiplied by
mal distribution with mean value r (km) and
a ground motion spectrum (or radiation
standard deviation r (km) is used. The point
spectrum) S ( f , M , r ) at discrete frequencies
fl = l / T , l = 1, ..., nT / 2 source stochastic model previously described is
well suited for generating the high-frequency
6. The modified spectrum is transformed back
components of the ground motion (greater than
to the time domain to yield the desired ground
0.1Hz). Low-frequency components can also be
acceleration time history. Details of the
introduced in the analysis by combining the above
characterization of the envelope function
56
methodology with near-fault ground motion E ( f , M ) = C M 0 (M )Sa ( f , M ) (17)

models (Mavroeidis and Papageorgiou,2003).
where C is a constant, M 0 (M ) = 101.5M +10.7 is the
Envelope Function seismic moment, and the factor Sa is the displace-
ment source spectrum given by (Atkinson and
The envelope function for the ground accelera-
Silva, 2000)
tion is represented by (Saragoni and Hart, 1974;
Boore, 2003) 1
Sa ( f , M ) = + (18)
f f
2 2
1 + 1 +
a2 t
t a 3
e(t, M , r ) = a1 e tn
(14) f f
tn a b
where the corner frequencies fa and fb , and the

where
weighting parameter are defined, respectively,
as
e
a2
0.2ln(0.05) a2
a2 = , a3 = , a1 =
1 + 0.2(ln(0.2) 1) 0.2 0.2 log( fa ) = 2.181 0.496M ,
(15) log( fb ) = 2.41 0.408M , and
log() = 0.605 0.255M .
The envelope function has a peak equal to
unity when t = 0.2 tn , and e(t, M , r ) = 0.05 The constant C is given by
when t = tn , with tn = 2.0 Tgm , where Tgm is the C = URVF / 4s s R0 , where U is a unit
3
duration of ground motion, expressed as a sum dependent factor, R is the radiation pattern, V
of a path dependent and source dependent com- represents the partition of total shear-wave en-
ponent Tgm = 0.05 r 2 + h 2 + 0.5 / fa , where r ergy into horizontal components, F is the effect
is the epicentral distance, and the parameters h of the free surface amplification, s and s are
and fa (corner frequency) are moment dependent the density and shear-wave velocity in the vicin-
given by log(h ) = 0.15M 0.05 a n d ity of the source, and R0 is a reference distance.
log( fa ) = 2.181 0.496M (Atkinson and Silva, Next, the path effect P ( f , r ) which is another
2000). component of the process that affects the spectrum
of motion at a particular site is represented by
Ground Motion Spectrum functions that account for geometrical spreading
and attenuation
The total spectrum of the motion at a site
S ( f , M , r ) is expressed as the product of the P ( f , r ) = Z (R(r )) e fR(r )/Q ( f )s (19)
contribution from the earthquake source E ( f , M ),
path P ( f , r ), site G ( f ) and type of motion I ( f ), where R(r ) is the radial distance from the hypo-
i.e. center to the site given by R(r ) = r 2 + h 2 . The
attenuation quantity Q( f ) is taken as
S (f , M , r ) = E (f , M ) P (f , r ) G (f ) I (f ) (16) Q( f ) = 180 f 0.45 and the geometrical spreading
function is selected as Z (R(r )) = 1 / R(r ) if
The source component is given by R(r ) < 70.0 km and Z (R(r )) = 1 / 70.0 other-
57
Table 1. Parameters for the stochastic ground acceleration model
Parameter Numerical Value Parameter Numerical Value
r (km) 20.0 r (km) 9.0
b 1.8 U 1020
s (gm/cc) 2.8 s (km/s) 3.5
V 1/ 2 R 0.55
F 2.0 R0 (km) 1.0
T (s) 20.0 t (s) 0.01
M min 6.0 M max 8.0
wise (Atkinson and Silva, 2000). The modification SEQUENTIAL OPTIMIZATION

of seismic waves by local conditions, site effect
G ( f ), is expressed by the multiplication of a The solution of the reliability-based optimiza-
diminution function D( f ) and an amplification tion problem given by Equations (1, 2 and 3)
function A( f ). The diminution function accounts is obtained by transforming it into a sequence
for the path-independent loss of high frequency of sub-optimization problems having a simple
in the ground motions and can be accounted for explicit algebraic structure. Thus, the strategy
a simple filter of the form D( f ) = e 0.03 f (An- is to construct successive approximate analyti-
derson and Hough, 1984). The amplification cal sub-problems. To this end, the objective and
function A( f ) is based on empirical curves given the constraint functions are represented by using
in (Boore et al., 1997) for generic rock sites. An approximate functions dependent on the design
average constant value equal to 2.0 is considered. variables. In particular, a hybrid form of linear,
Finally, the filter that controls the type of ground reciprocal and quadratic approximations is con-
motion I ( f ) is chosen as I ( f ) = (2 f )2 for ground sidered in the present formulation (Haftka and
acceleration. The particular values of the different Grdal, 1992). For the purpose of constructing the
parameters of the stochastic ground acceleration approximations all design variables are assumed
model considered in this contribution are given to be continuous.
in Table 1. For illustration purposes Figure 1
shows the envelope function, the ground motion First and Second-Order
spectrum and a corresponding sample of ground Approximations
motion for a nominal distance r = 20 km, and
moment magnitude M= 7.0. For a detailed discus- Let p({y }) be a generic function involved in the
sion of this point-source model the reader is re- optimization problem, i.e. the objective or con-
ferred to (Atkinson and Silva, 2000; Boore, 2003). straint functions, and {y 0 } a point in the feasible
design space. The function p({y }) is first ap-
proximated about the point {y 0 } by using a hybrid
58
Figure 1. Envelope function, ground acceleration spectrum and sample ground motion for epicentral
distance r=20 km and moment magnitude M=7
form of linear and reciprocal approximations. In linearization is not guaranteed to be conservative

particular, the following approximation is con- in an absolute sense. That is, it is not known that
sidered the approximations are more conservative than
the original functions. One way to affect the con-
p({y }) pcl ({y }) = p({y 0 }) servatism of the approximation is by considering
p({y 0 }) second-order terms. For example, introducing
+ pl (yi , yi0 )(yi yi0 ) diagonal quadratic terms in the convex approxi-
yi
i
mation (20) yields
(20)
pcq ({y }) = pcl ({y }) + pq (yi , yi0 )(yi yi0 )2
where i
(22)
0
p({y })
pl (x i , yi0 ) = 1 if 0, and where pq (yi , yi0 ) are the coefficients of the second-
yi
y0 p({y 0 }) order diagonal terms. These coefficients are de-
pl (x i , yi0 ) = i if <0 (21) fined in terms of the second-order derivatives of
yi y i
the convex approximation pcl ({y }) at the point
An attractive property of this approximation, {y 0 } as
called convex linearization, is that it yields the
most conservative approximation among all pos- 2 pcl ({y 0 })
sible combinations of direct/reciprocal variables pq (yi , yi0 ) = i (23)
yi2
(Fleury and Braibant, 1986). In principle, convex
59
Box 4.
f ({y 0 }) f ({y 0 }) (y 0 )2 f ({y 0 }) y

fcq ({y }) = yi i
2if
yi ( i0 2) +

(i + ) yi
(i ) yi yi
(i ) yi yi
(25)
f ({ y 0
}) f ({y 0
})
f ({y 0 }) yi0 yi0 (2if 1)
+ y
(i ) i

(i ) y i

where Approximate Objective Function
2 pcl ({y 0 }) p({y 0 }) Applying the above approximation approach to

= 0 if 0, and
yi2 y i the objective function yields the following ap-
2 pcl ({y 0 }) 2 p({y 0 }) p({y 0 }) proximate function to be minimized as in Box 4.
= if <0 where the group (i + ) contains the variables for
yi2 yi0 y i y i
which the partial derivative of the objective func-
(24)
tion is positive and group (i ) includes the remain-
and i is a user-defined positive scalar. The ing variables.
purpose of this parameter is to enforce the con-
Approximate Constraint Functions
servatism of the approximation. For continu-
ously differentiable functions the manipulation
The construction of approximate constraint func-
of the higher-order curvatures will ensure the
tions is based on the approximation of transformed
conservatism of the approximations at least lo-
failure probability functions. Specifically, the
cally (Groenwold et al., 2009). The enforcement
following transformation is considered (Jensen
of the conservatism will be carried out during
and Beer, 2010)
inner loops throughout the optimization process
(see Solution Scheme section). The conservatism
h jt ({y }) = ln[PFj ({y })], j = 1, ..., nc (26)
of the approximations is important since affect
the global convergence of the optimization process
(Svanberg, 2002). It ensures that the optimal The transformed failure probability functions
solution of the sub-optimization problem is a are approximated by using the approximation ap-
feasible solution of the original problem, with a proach previously defined. That is,
lower objective value than the previous iteration.
In this manner, the process generates a sequence h t ({y 0 }) h t ({y 0 }) (y 0 )2
({y }) = yi
t j j i
hcqj
of steadily improved feasible designs if the process yi
(i j+ )
yi
(i j ) yi
starts from an initial feasible design. This prop- h t ({y 0 }) y
2i j
ht
yi ( i0 2) + h jt ({y 0 }) , j = 1, ..., nc
j
erty is significant from a practical view point since yi yi
(i j )
the optimization process can be stopped at any
(27)
stage still leading to better designs than the initial
feasible estimate.
with
60
Box 5.
f ({y 0 }) f ({y 0 }) (y 0 )2 f ({y 0 }) y

+ y yi y i
yi

2i
f
y y i ( i
0
yi
2)

(30)
(i ) i (i ) i (i ) i
Box 6.
h t ({y 0 }) h t ({y 0 }) (y 0 )2 h t ({y 0 }) x

j j
+ y yi y 2i
h tj
j i yi ( 0i 2)
yi yi yi (31)
(i j ) i (i j ) i (i j )
h jt ({y 0 }) + ln[PF*j ] , j = 1, ..., nc
h t ({y 0 })
Solution Scheme
h ({y }) = h ({y }) j
t 0 t 0
yi0
j j

yi
(i j+ )

The solution scheme of the optimization process
h ({y })
t 0
proceeds as follows:
yi0 (2i j 1)
ht
j
yi
(i )

j
(28) 1. Start from a feasible design. At the beginning
of the k th design cycle ( k = 0, 1, 2, ... ) the
where and mean summation over the objective function f ({y }) and constraint
(i j + ) (i j )
functions h j ({y }), j = 1, ..., nc are approxi-
variables belonging to group (i j+ ) and (i j ), re-
mated by using the approach introduced
spectively. Then, the approximate reliability
previously. The approximations require
constraints can be written as
f u n c t i o n e v a l u a t i o n s ( f ({y k }) ,
h j ({y k }), j = 1, ..., nc ) and sensitivity analy-
t
hcqj ({y }) = hcqj ({y }) ln[PF*j ] 0, j = 1, ..., nc
ses ( f ({y k }), h j ({y k }), j = 1, ..., nc ). In
(29)
this step, all design variables are assumed
Approximate Optimization Problem to be continuous.
2. Using this information an explicit sub-
Using the above approximations for the objective optimization problem is constructed. The
and constraint functions, the approximate sub- explicit sub-optimization problem can be
optimization problem defined at the point {y 0 } , solved either by standard methods that treat
P{y 0 } , takes the form in Box 5 and 6. the problem directly in the primal design
variable space (branch and bound techniques,
P{y 0 } : Minimize (see Box 5) subject to (see combinatorial methods, evolution-based
Box 6) j = 1, ..., nc , where all terms have been optimization techniques, etc.) or by dual
previously defined. methods (Fleury and Braibant, 1986). In the
present implementation, a genetic algorithm
is applied to solve the sub-optimization
problem.
61
3. The new point {y *k } is tested if it is ac- STOCHASTIC ANALYSIS

ceptable in terms of a conservative
criterion, i.e. if fcq ({y *k }) f ({y *k }) and The characterization of the sub-optimization
if hcqj ({y }) h j ({y }), j = 1, ...nc . I f
*k *k problems P{y k } , k = 0, 1, 2, ... requires the estima-
these conditions are satisfied (conservation of failure probabilities and their sensitivities.
tive step) the point {y *k } is used as the For that purpose an advanced simulation technique
current design for the next cycle, that is adopted and implemented in the present for-
is, {y k +1 } = {y *k }. If the design {y *k } mulation.
does not represent a conservative step
an inner loop is initiated to effect con- Reliability Estimation
servatism. For functions which are not
conservative at {y *k } the corresponding Subset simulation is adopted in this formulation
coefficients of the second-order diagonal for the purpose of estimating the corresponding
terms are increased by multiplying the failure probabilities during the design process (Au
scalars i by a constant greater than one. and Beck, 2001). In the approach, the failure
The modified approximations are used probabilities are expressed as a product of con-
to construct a new sub-optimization ditional probabilities of some chosen intermediate
problem to obtain a new point. failure events, the evaluation of which only re-
4. The overall design process is continued until quires simulation of more frequent events. There-
some convergence criterion is satisfied. fore, a rare event simulation problem is con-
verted into a sequence of more frequent event
It is noted that the requirement of a conserva- simulation problems. For example, the failure
tive step in the algorithm can be relaxed and probability PFj ({y }) can be expressed as the
demand that a feasible descent step is made instead, product
i.e. if f ({y *k }) < f ({y *(k 1) }) a n d i f
mFj 1
h j ({y *k }) 0, j = 1, ...nc . In this case, the con- PFj ({y }) = P (Fj ,1 ({y })) P (Fj ,k +1 ({y }) / Fj ,k ({y }))
servatism is only enforced when a feasible descent k =1
(32)
step could not be made. This approach which is
called, relaxed conservatism, inherits the global
convergence properties of the algorithm that
where Fj ,mF ({y }) = Fj ({y }) is the target failure
enforces conservatism at each design cycle. In j
general the above optimization scheme is very event and:

effective when the curvatures of the functions
involved in the optimization problem are not too Fj ,mF ({y }) Fj ,mF 1 ({y }) ... Fj ,1 ({y })
j j
large and relatively uniform throughout the design

space. For other cases, methods based on local is a nested sequence of failure events. Equation
response surfaces, trust regions and line search (32) expresses the failure probability PFj ({y }) as
methodologies may be more appropriate (Alex- a product of P (Fj ,1 ({y })) and the conditional
androv et al., 1998; Bucher and Bourgund, 1990; probabilities:
Jensen et al., 2009).
P (Fj ,k +1 ({y }) / Fj ,k ({y })), k = 1, ..., mFj 1 .
62
It is seen that, even if PFj ({y }) is small, by choos- where {y } = {y k } + {y } and {j } is a vector
ing mFj and Fj ,k ({y }), k = 1, ..., mFj 1 appro- of constant, real valued coefficients. For samples
priately, the conditional probabilities can still be ({i }, {i }), i = 1, ..., ms near the limit state sur-
made sufficiently large, and therefore they can be face, that is, j ({y }, {i }, {i }) 0 the perfor-
evaluated efficiently by simulation because the mance function is evaluated at ns = qs ms points
failure events are more frequent. The intermediate in the neighborhood of {y k } . That is, for each
failure events are chosen adaptively using infor- sample ({i }, {i }), qs designs are defined. These
mation from simulated samples so that they cor- points are generated as
respond to some specified values of conditional
failure probabilities (Au and Beck, 2001). {l }
{y lk } {y k } = {y } = R , l = 1, ..., ns
{l }
Sensitivity Estimation
(35)
As previously pointed out the approximation of
where the components of the vector {l } are in-
the reliability constraints requires the estimation
dependent, identically distributed standard Gauss-
of the gradient of the transformed failure probabil-
ian random variables, and R is a user-defined
ity functions h jt . For the purpose of estimating
small positive number. This number defines the
the gradients, it is assumed that the optimization radius of the hypersphere ({l } / {l } )R centered
variables can have non-discrete values during the
at the current design {y k } . The coefficients {j }
solution process. Therefore, all design variables
are treated as continuous during the first step of are computed by least squares. To this end the
the optimization process (see Solution Scheme following set of equations is generated
section). The sensitivity of the transformed failure
probability functions with respect to the design j ({y lk }, {i }, {i }) = j ({y k }, {i }, {i })
variables is estimated by an approach recently {l }
introduced in (Valdebenito and Schuller, 2011). +{i }T R
{l }
The approach is based on the approximate repre-
l = i + (q 1) ms , q = 1,..., qs , i = 1,..., ms
sentation of two different quantities. The first
approximation involves the performance function (36)
j while the second includes the failure probabil-
ity function. Recall that the failure domain Fj where the index l indicates a double loop in terms
of the indices i and q . Since the samples
for a given design {y } is defined as
({i }, {i }), i = 1, ..., ms are chosen near the
F ({ y}) = {{},{} | j ({ y},{},{}) 0}. limit state surface the approximate performance
j
function j is expected to be representative, on
(33)
the average, of the behavior of the failure domain
If {y k } is the current design, the performance Fj in the vicinity of the current design {y k }
function is approximated in the vicinity of the (Jensen et al., 2009). Next, the failure domain
current design as Fj for a given design {y } can also be defined in
terms of the normalized demand function as
j ({y }, {}, {}) = j ({y k }, {}, {}) + {j }T {y }
(34)
63
which is equal to one. Considering the definition

F ({ y}) = {{},{} | D j ({ y},{},{}) 1}.
j of failure probability in terms of the performance
(37)
function, the linear expansion of the performance
function in (34), and the approximation of the
where Dj ({y }, {}, {}) = 1 j ({y }, {}, {}).
failure probability function given in (38), the
The failure probability function, evaluated at the partial derivatives of the j -th transformed failure
current design {y k } , is approximated locally as probability function can be expressed as (Jensen
an explicit function of the normalized demand et al., 2009)
around Dj* = 1as
h jt ({y }) 1 e
0 + 1lj yl
e
0
lim
P Dj ({y k }, {}, {}) Dj* e 0 1 j ,
*
+ (D 1)
yl k
PF ({y }) y 0 yl
(38)
l
{y }={y k } j
*
D [1 , 1 + ] = 1jl , l = 1, , nd
j
(41)
where Dj* is a threshold of the normalized demand
(in the neighborhood of one) and represents a where jl is the l -th element of the vector {j },
small tolerance. The coefficients 0 and 1 can and all other terms have been previously defined.
be calculated by least squares with samples of the It is noted that the previous approach for estimat-
performance function j (or normalized demand ing the gradients of the failure probability functions
function Dj ) generated at the last stage of subset requires a single reliability analysis plus the
simulation (Valdebenito and Schuller, 2011). The evaluation of the performance functions in the
gradient of the j -th transformed failure probabil- vicinity of the current design. Validation calcula-
ity function at {y k } (see Equation (26)) is given tions have shown that this approach is quite ef-
by ficient for estimating the sensitivity of failure
probability functions with respect to design vari-
PFj ({y })
ables (Jensen et al., 2009; Valdebenito and
h jt ({y }) 1
= Schuller, 2011).
yl k
PFj ({y }) yl
{y }={y k } {y }={y k }
l = 1, , nd
(39) NUMERICAL EXAMPLES
where nd is the total number of design variables. Example No.1

On the other hand, the gradient of the j -th failure
probability function can be estimated by means Description
of the limit:
Consider the reliability-based optimization of the
PFj ({y }) k k
PFj ({y } + {(l )}yl ) PFj ({y }) two-story frame structure under earthquake load-
= lim ,
yl x l 0 yl ing shown in Figure 2. The floor masses are
{y }={y k }
l = 1, , nd mi = 107 kg, i = 1, 2, and the initial linear inter-

(40) story stiffnesses are K 1 = 9.0 109 N/m and
K 2 = 7.0 109 N/m. A 3 % of critical damping
where {(l )} is a vector of length nd with all is assumed in the model. The damping ratio is
entries equal to zero, except for the l -th entry, treated as uncertain and modeled as a log-normal
64
Figure 2. Structural system with a passive energy

gi ((t ), u(t )) = H (i (t ))
dissipation system under ground acceleration
iu(t ) uy
[H ( u(t ) u )
i y
u p uy
H (u p iu(t )) + H (iu(t ) u p )]

(44)
where i = 1(i 1) , i = 1, 2, H () denotes the

Heaviside step function, uy is a parameter speci-
fying the onset of yielding, and kd u p is the
maximum restoring force of the device. The val-
u e s u p = 6.0 103 m , uy = 0.7u p , a n d
kd = 6.0 108 N/m are used for each nonlinear
element. Because of the yielding, energy dissipa-
tion due to hysteresis is introduced in the struc-
tural response. For illustration purposes the be-
havior of the friction devices at the initial design
( K 1 = 9.0 109 N/m, and K 2 = 7.0 109 N/m)
is shown in Figure 3. In this figure typical dis-
random variable with a coefficient of variation of placement-restoring force curves of the friction
40 % . This high coefficient of variation accounts devices at the first and second floor are presented.
for the considerable uncertainty in estimating The non-linear incursion of the devices is clear
damping ratios in real structural systems. from the figure. In general the overall effect of
For an improved earthquake resistance, the the dissipation system is to decrease the response
model is reinforced with a passive energy dissipa- of the structural system. The system is subjected
tion system (shear panel) at each floor. The shear to a ground acceleration that is modeled as de-
panels follow the inter-story restoring force law scribed in a previous Section. Note that the dura-
tion of the excitation is 20 s and the sampling
r (t ) = kd ((t ) q 1 (t ) + q 2 (t )) (42) interval is equal to 0.01 s (see Table (1)). There-
fore the number of random variables involved in
where kd denotes the initial stiffness of the device, the characterization of the excitation is equal to
2000. This in turn implies that the estimation of
(t ) is the relative displacement between floors,
the failure probability for a given design represents
and q 1 (t )and q 2 (t ) denote the plastic elongations
a high dimensional reliability problem.
of the device. Using the auxiliary variable
u(t ) = (t ) q 1 (t ) + q 2 (t ), the plastic elongations
Discrete Optimization Problem
are specified by the differential equations
The objective function is defined in terms of an
qi (t ) = i (t )gi ((t ), u(t )), i = 1, 2 (43) initial cost which is assumed to be proportional
to the linear interstory stiffnesses. The reliability
where the nonlinear functions gi are specified by constraints are given in terms of the interstory
(Pradlwarter and Schuller, 1993) drift ratios. Failure is assumed to occur when the
65
Figure 3. Typical displacement-restoring force curves of the shear panels at the initial design. The first
and the second floor
interstory drift ratio reaches some critical level with side constraints
for the first time. A threshold level value equal to
0.02 m is considered in this case, and the failure K1
{3.0, 3.5, 4.0, 4.5, 5.0, 5.5, 6.0, 6.5, 7.0, 7.5}
events are defined as 109
{8.0, 8.5, 9.0, 9.5, 10.0, 10.5, 11.0, 11.5, 12.0}N/m
Fi = maxt [ 0,T ] | i (t ) |> 0.02 , i = 1, 2 (45) (48)
K2
where 1 (t ) and 2 (t ) are the relative displacements {3.0, 3.5, 4.0, 4.5, 5.0, 5.5, 6.0, 6.5, 7.0, 7.5}
109
of the first and second floor, respectively. The {8.0, 8.5, 9.0, 9.5, 10.0, 10.5, 11.0, 11.5, 12.0}N/m
target failure probability is taken equal to 103. (49)
The optimization variables are the linear inter-
story stiffnesses K 1 and K 2 , with initial design The problem is solved by using the sequential
K 01 = 9.0 109 N/m and K 02 = 7.0 109 N/m. approximate optimization approach previously
The optimization problem is written as described. Subset simulation is used to estimate
the failure probabilities and their sensitivities. To
Min f (K 1 , K 2 ) = K 1 + K 2 (46) smooth the variability of the estimates, the aver-
age of failure probability estimates over five in-
subject to dependent simulation runs is considered at each
design. The value of the parameter that controls
PFj (K 1 , K 2 ) PF*j , j = 1, 2 (47) the curvature of the second-order terms in the
approximations is taken as = 0.5. It is noted
66
Figure 4. Iteration history in terms of the objective function. Deterministic and uncertain models(Example 1)
that the optimal choice of this parameter from the tions. It is observed that the value of the objective
algorithm performance point of view depends on function at the final design of the uncertain model
the particular type of problem. The reader is re- is greater than the corresponding value of the
ferred to (Svanberg, 2002) for a general discussion deterministic model. This in turn implies that the
of the values of this parameter in the context of structural components (columns) of the uncertain
conservative, convex and separable approxima- model are bigger than the corresponding compo-
tions. nents of the deterministic model. This result can
be seen from Figure 6, where the final designs of
Results the deterministic and uncertain models are shown
in the design space corresponding to the uncertain
The iteration history of the optimization process in model. Note that the final design obtained from
terms of the objective function and the reliability the deterministic model is not feasible. Finally,
constraints is shown in Figure 4 and figure 5, re- it is seen from Figures 4 and 5 that the method
spectively. For comparison, the results obtained generates a series of steadily improved feasible
with the deterministic model are also shown in designs that move toward the optimum.
the figures. In this context, the deterministic
model considers the damping ratio equal to its Example No.2
most probable value. The objective function is
normalized by the cost of the initial design. It is Structural Model
seen that, starting from a feasible initial design,
the process converges in less than four iterations. A four-story reinforced concrete building under
Therefore, the entire optimization process takes earthquake motion is considered for analysis. A
few excursion probability and sensitivity estima- 3D view of the system is shown in Figure 7. Each
67
Figure 5. Iteration history in terms of the reliability constraints. Deterministic and uncertain models
(Example 1)
Figure 6. Final designs of the deterministic and uncertain models in the design space
68
Figure 7. Four-story building model under earthquake excitation
of the four floors is supported by 43 columns of acceleration is modeled as described in a previous

circular cross section. All floors have a constant Section.
height equal to 3.0 m, leading to a total height of
12.0 m. It is assumed that each floor may be Dissipation Model
represented sufficiently accurately as rigid
within the x y plane when compared with the For an improved earthquake performance the
flexibility of the columns. Hence, each floor can structural system is enforced with vibration con-
be represented by three degrees of freedom, i.e. trol devices.
two translatory displacements in the direction of Ten devices connected to the structure every
the x axis and y axis, and a rotational displace- two floors as indicated in Figure (7) are placed
ment. The associated active masses in the x and in the structural system. A typical configuration
y direction are taken constant for all floors and of the devices is shown in Figure (8). They con-
equal to 658 ton, while the mass moment of in- sist of brace and plate elements as indicated in
ertia is equal to 1.8 105 ton-m 2 for all floors. the figure. Between the plates there are a series
The Youngs modulus E and the modal damping of dissipators in the form of metallic U-shaped
ratios i of the structural model are treated as flexural plates (UFP).
uncertain system parameters. The Youngs The devices exhibit a one-dimensional hys-
modulus is modeled as a Gaussian random vari- teretic type of non-linearity modeled by the restor-
able with most probable value E = 2.5 1010 N/ ing force law
m 2 , and coefficient of variation of 20%, while
the damping ratios are modeled by independent r (t ) = nufp rufp (t ) (50)
Log-normal random variables with mean values
i = 0.05 and coefficients of variation of 40% .
The structural model is excited horizontally by a where nufp is the number of U-shaped flexural
ground acceleration applied at 45 with respect plates in the device and rufp (t ) is the dissipator
to the plan of the system. The induced ground force given by
69
Figure 8. Vibration control device and metallic U-shape flexural plates
rufp (t ) = ke (t ) + (1 ) ke U y z (t ) (51) ke = 4 108 N/m, U y = 5 103 m, = 0.01,

1 = 1.0, 2 = 2.0, 3 = 0.5 are used in this
where ke is the pre-yield stiffness, U y is the yield case. These model parameters generate hyster-
displacement, is the factor which defines the etic behaviors similar to the one observed from
extent to which the restoring force is linear, z (t ) experimental data (De La Llera et al., 2004)
is a dimensionless hysteretic variable, and (t )
is the relative displacement between the floors Design Problem
where the device is connected. The nonlinear
restoring force of the device acts between the The design variables denoted bydi , i = 1, ..., nd ,
floors where it is placed with the same orientation are the diameters of the reinforced concrete column
as the relative displacement (t ). The hysteretic elements. Each floor is associated to one design
behavior of each U-shaped flexural plate is defined variable and therefore nd = 4 in this example.
in terms of the auxiliary variable z (t ) which sat- The objective function f is proportional to the
isfies the first-order non-linear differential equa- total volume of the resistant elements (columns),
tion while the failure events are defined in terms of
the interstory drift ratios, and given by
U y z(t ) = (t ) 1 z (t )(2 sgn((t )) + 3 sgn(z (t )))

(52) Fi ({d }, {}, {}) =
maxt ,k =1,..,2001 | i (tk , {d }, {}, {}) |> * (53)
k
where 1 , 2 and 3 are dimensionless quanti-
ties that characterize the properties of the hys-
teretic behavior, sgn() is the sign function, and where i (tk , {d }, {}, {}) is the relative displace-
all other terms have been previously defined. ment between the (i 1, i ) -th floor evaluated at
The quantities 1 , 2 and 3 correspond to the design {d }, tk are the discrete time instants,
scale, loop fatness and loop pinching parameters, * is the critical threshold level and equal to 0.006
respectively. The above characterization of the m (0.2 % of the floor height), {} is the vector
hysteretic behavior corresponds to the Bouc-Wen that represents the uncertain structural parameters
type model (Baber and Wen, 1981). The values (Youngs modulus and damping ratios), and {}
70
is the vector that specifies the stochastic excita- indicated. The objective function is normalized
tion. The corresponding multidimensional prob- by its value at the initial design. From Table 2 and
ability integrals involved in the estimation of the figure 9 it is observed that the dimensions of the
probabilities of failure include more than two column elements at the final design of the system
thousand random variables in this case. Thus, as without the vibration control devices are greater
in the previous example the estimation of the than the corresponding elements of the model
failure probability for a given design represents with the devices, as expected. The total weight of
a high-dimensional reliability problem. The reli- the unprotected model increases almost 40 % with
ability-based optimization problem is written as respect to the weight of the enforced model. This
result highlights the beneficial effects of the vibra-
Min f ({d }) (54) tion control devices in protecting the structural
system. For illustration purposes Figure 10 shows
subject to a typical displacement-restoring force curve of
one of the U-shaped flexural plates at the final
PFi ({d }) PF* = 103 i = 1, 2, 3, 4 (55) design. The nonlinear incursion of the dissipator
is clear from the figure.
It is found that the process converges in less
with side constraints than four iterations in this case and therefore the
design process takes few excursion probability
di Di , i = 1, ..., 4 (56) and sensitivity estimations. This computational
cost is substantially different for the case of direct
where the set Di represents the available discrete optimization. In that case the number of excursion
values for the design variable di . The set Di is probability and sensitivity estimations increases
dramatically with respect to the proposed ap-
defined as Di = {0.40 , 0.41 , , 0.75} m. It is
proach. In direct optimization the excursion prob-
seen that each of the design variables can be
abilities and their sensitivities need to be esti-
chosen from a discrete set of 36 diameters. There-
mated for every change of the design variables
fore, the discrete set of design variables involves
during the optimization process. As in the previ-
more than 106 possible combinations. The opti-
ous example the method generates a series of
mization problem is solved by the sequential
steadily improved feasible designs that moves
approximate optimization strategy discussed in
toward the final design. That is, the design process
previous sections.
has monotonic convergence properties.
Final Designs
CONCLUSION
The initial and the final designs of the structural
system are presented in Table 2. The correspond-
A general framework for reliability-based design
ing iteration history of the optimization process
optimization of a class of stochastic systems in-
in terms of the objective function is shown in
volving discrete sizing type of design variables
Figure 9. Note that the initial design is an inte-
has been presented. The reliability-based design
rior point in the design space, as it is required by
problem is formulated as an optimization prob-
the algorithm (see first step of the Solution Scheme
lem with a single objective function subject to
section). For comparison the final design of the
multiple reliability constraints. The high compu-
system without vibration control devices is also
tational cost associated with the solution of the
71
Table 2. Final designs
Final design
Design variable Initial design ND D
d1 (m) 0.75 0.71 0.62
d2 (m) 0.70 0.68 0.59
d3 (m) 0.65 0.64 0.54
d4 (m) 0.60 0.56 0.44
Normalized objective function 1.00 0.92 0.66

ND= without devices, D= with devices
Figure 9. Iteration history in terms of the objective function (Example 2)
optimization problem is addressed by the use of takes into account the uncertainty in structural and
approximate reliability analyses during portions excitation model parameters explicitly during the
of the optimization process. This is achieved optimization process. The design scheme exhibits
by implementing a sequential optimization ap- monotonic convergence that is, starting from an
proach based on global conservative, convex initial feasible design the scheme generates a se-
and separable approximations. The combined quence of steadily improved feasible designs. This
use of approximation concepts, advanced simu- property is important from a practical view point
lation techniques, efficient sensitivity analyses since the optimization process can be stopped at
and discrete optimization techniques provides any stage still leading to better designs than the
an effective reliability-based design optimiza- initial feasible estimate. Recall that each design
tion strategy for sizing problems which involve cycle involves a considerable computational effort
discrete design variables. The proposed approach for complex structural systems. Numerical results
72
Figure 10. Typical displacement-restoring force

ACKNOWLEDGMENT
curves of one of the U-shaped flexural plates at
the final design
This research was partially supported by CONI-
CYT (National Commission for Scientific and
Technological Research) under grant number
1110061. This support is gratefully acknowledged
by the authors.
REFERENCES
Alexandrov, N., Dennis, J. Jr, Lewis, R., & Tor-

czon, V. (1998). A trust-region framework for
managing the use of approximation models in
optimization. Structural Optimization, 15(1),
1623. doi:10.1007/BF01197433
Anderson, J., & Hough, S. (1984). A model for
the shape of the Fourier amplitude spectrum of
acceleration at high frequencies. Bulletin of
the Seismological Society of America, 74(5),
19691993.
Atkinson, G., & Silva, W. (2000). Stochastic
obtained from the example problems and from ad-
modeling of California ground motions. Bulletin
ditional validation calculations have shown that the
of the Seismological Society of America, 90(2),
number of reliability estimations required during
255274. doi:10.1785/0119990064
the optimization process is in general very small.
The results have also indicated that the presence Au, S., & Beck, J. (2001). Estimation of small
of passive vibration control devices has a positive failure probabilities in high dimensions by
impact on the reliability and global performance subset simulation. Probabilistic Engineering
of systems under stochastic loading. In fact, the Mechanics, 16(4), 263277. doi:10.1016/S0266-
results of this study highlight the beneficial effect 8920(01)00019-4
of these devices in protecting structural systems
Baber, T., & Wen, Y. (1981). Random vibration
under environmental loads such as earthquake
hysteretic, degrading systems. Journal of the Engi-
excitations. Based on the previous observations
neering Mechanics Division, 107(6), 10691087.
it is concluded that the proposed methodology
can be very advantageous from a numerical point Boore, D. (2003). Simulation of ground motion
of view for solving discrete variable structural using the stochastic method. Pure and Applied
optimization problems of an important class of Geophysics, 160(3), 635676. doi:10.1007/
stochastic dynamical systems. PL00012553
73
Boore, D., Joyner, W., & Fumal, T. (1997). Gunawan, S., & Papalambros, P. Y. (2007).
Equations for estimating horizontal response Reliability optimization with mixed continuous-
spectra and peak acceleration from western North discrete random variables and parameters. Jour-
American earthquakes: A summary of recent work. nal of Mechanical Design, 129(2), 158165.
Seismological Research Letters, 68(1), 128153. doi:10.1115/1.2406085
doi:10.1785/gssrl.68.1.128
Haftka, R., & Grdal, Z. (1992). Elements of
Bucher, C., & Bourgund, U. (1990). A fast and structural optimization (3rd ed.). Kluwer.
efficient response surface approach for structural
Hassan, R., & Crossley, W. (2008). Reliability-
reliability problems. Structural Safety, 7(1),
based optimization under uncertainty including
5766. doi:10.1016/0167-4730(90)90012-E
discrete variables. Journal of Spacecraft and
De la Llera, J., Esguerra, C., & Almazn, J. Rockets, 45(2), 394405. doi:10.2514/1.28827
(2004). Earthquake behavior of structures with
Jensen, H., & Beer, M. (2010). Discrete-contin-
copper energy dissipators. Earthquake Engi-
uous variable structural optimization of systems
neering & Structural Dynamics, 33(3), 329358.
under stochastic loading. Structural Safety, 32(5),
doi:10.1002/eqe.354
293304. doi:10.1016/j.strusafe.2010.03.007
Enevoldsen, I., & Srensen, J. (1994). Reli-
Jensen, H., Valdebenito, M., Schuller, G., &
ability-based optimization in structural engi-
Kusanovic, D. (2009). Reliability-based optimi-
neering. Structural Safety, 15(3), 169196.
zation of stochastic systems using line search.
doi:10.1016/0167-4730(94)90039-6
Computer Methods in Applied Mechanics and En-
Fleury, C., & Braibant, V. (1986). Structural gineering, 198(49-52), 39153924. doi:10.1016/j.
optimization: A new dual method using mixed cma.2009.08.016
variables. International Journal for Numeri-
Kovcs, L. (1980). Combinatorial methods of
cal Methods in Engineering, 23(3), 409428.
discrete programming. Mathematical Methods
doi:10.1002/nme.1620230307
of Operation Research, 2.
Goldberg, D. (1989). Genetic algorithms in search,
Kramer, S. (2003). Geotechnical earthquake
optimization, and machine learning. Reading,
engineering. Prentice Hall.
MA: Addison Wesley.
Lagaros, N., Garavela, A. T., & Papadrakakis, M.
Groenwold, A., Etman, L., Snyman, J., & Rooda,
(2008). Innovative seismic design optimization
J. (2007). Incomplete series expansion for function
with reliability constraints. Computer Methods in
approximation. Structural and Multidisciplinary
Applied Mechanics and Engineering, 198, 2841.
Optimization, 34(1), 2140. doi:10.1007/s00158-
doi:10.1016/j.cma.2007.12.025
006-0070-6
Mavroeidis, G., & Papageorgiou, A. (2003).
Groenwold, A., Wood, D. W., Etman, L. F. P., &
A mathematical representation of near-fault
Tosserams, S. (2009). Globally convergent optimi-
ground motions. Bulletin of the Seismo-
zation algorithm using conservative convex sepa-
logical Society of America, 93(3), 10991131.
rable diagonal quadratic approximations. AIAA
doi:10.1785/0120020100
Journal, 47(11), 26492657. doi:10.2514/1.41975
74
Papadrakakis, M., & Lagaros, N. (2002). Svanberg, K. (2002). A class of globally conver-
Reliability-based structural optimization using gent optimization methods based on conservative
neural networks and Monte Carlo simulation. convex separable approximations. SIAM Journal
Computer Methods in Applied Mechanics and on Optimization, 12(2), 555573. doi:10.1137/
Engineering, 191, 34913507. doi:10.1016/ S1052623499362822
S0045-7825(02)00287-6
Tolson, B. A., Maier, H. R., Simpson, A. R.,
Pradlwarter, H. J., & Schuller, G. I. (1993). & Lence, B. J. (2004). Genetic algorithms for
Equivalent linearization - A suitable tool for ana- reliability-based optimization of water distribution
lyzing MDOF-systems. Probabilistic Engineering systems. Journal of Water Resources Planning
Mechanics, 8(2), 115126. doi:10.1016/0266- and Management, 130(1), 6372. doi:10.1061/
8920(93)90005-G (ASCE)0733-9496(2004)130:1(63)
Prasad, B. (1983). Approximation, adaptation Tomlin, J. (1970). Integer and non-linear program-
and automation concepts for large scale structural ming. In Branch and bound methods for integer
optimization. Engineering Optimization, 6(3), and non-convex programming (pp. 437450).
129140. doi:10.1080/03052158308902462 Amsterdam, The Netherlands: North Holland.
Saragoni, G., & Hart, G. (1974). Simulation of Valdebenito, M., & Schuller, G. (2011). Ef-
artificial earthquakes. Earthquake Engineering & ficient strategies for reliability-based optimiza-
Structural Dynamics, 2(3), 249267. doi:10.1002/ tion involving non linear, dynamical structures.
eqe.4290020305 Computers & Structures, 89(19-29), 1797-1811.
Scharage, L. (1989). Optimization modeling with
LINGO. Technical report, LINDO Systems inc.,
Chicago.
75
76
Chapter 4
A Multi-Hazard Framework
for Optimum Life-Cycle
Cost Design of Reinforced
Concrete Bridges
Azadeh Alipour
University of Massachusetts Amherst, USA
Behrouz Shafei
University of Massachusetts Amherst, USA
Masanobu Shinozuka
University of California Irvine, USA
ABSTRACT
This chapter provides a comprehensive procedure for the time-dependant structural performance evalua-
tion and life-cycle cost analysis of reinforced concrete highway bridges located in extreme chloride-laden
environments. The penetration of chloride ions into the concrete is simulated through a finite difference
approach, which takes into account all the parameters that can affect the corrosion process. From
simulation results, the corrosion initiation time is predicted and the extent of structural degradation is
calculated over the entire life of bridge. A group of detailed bridge models with various structural attri-
butes are developed to evaluate the changes in the structural capacity and seismic response of corroded
bridges. For the purpose of the probabilistic seismic risk assessment of bridges, the seismic fragility
curves are generated and updated at regular time intervals. The time-dependent fragility parameters
are employed to investigate the life-cycle cost of bridges by introducing a performance index which
combines the effects of probable seismic events and chloride-induced corrosion. The proposed approach
provides a multi-hazard framework, which leads to more realistic performance and cost estimates. It
also indicates the inspection and maintenance intervals in a way that the inspection and maintenance
costs are optimized, while the safety of bridge is ensured.
DOI: 10.4018/978-1-4666-1640-0.ch004
A Multi-Hazard Framework for Optimum Life-Cycle Cost Design
1. INTRODUCTION water-to-cement ratio, ambient temperature, rela-

tive humidity, concrete age, free chloride content,
From a long-term point of view, the durability and binding capacity, are considered to obtain a
of reinforced concrete (RC) highway bridges is precise prediction of the chloride content in dif-
significantly impacted by the deterioration of ferent depths of RC members with the progres-
their structural members. When investigating sion of time. By comparing the chloride content
the damaged bridges, the deterioration caused values with certain critical thresholds suggested
by the corrosion of reinforced concrete members in the literature, the corrosion initiation time is
is usually found to be one of the main sources estimated. After corrosion initiation, the time-
of structural degradation which may eventually dependent characteristics of corroded bridges are
result in the serviceability failure of bridges under identified through the extent of the cracking and
service or extreme loading conditions. An accurate spalling of the concrete cover, reduction of the
estimation of the extent of degradation during the steel bar cross-section area, and decrease in the
structures life-cycle provides both engineers and yield strength of reinforcing bars. Based on that,
decision-makers with valuable information which the probabilistic life time fragility parameters of
helps to ensure the safety of bridges while reduc- a group of RC bridges with different structural at-
ing the associated costs. Towards this goal, the tributes are evaluated over the time using fragility
current chapter focuses on the corrosion process analysis procedure.
caused by the chloride ions attack and evaluates Furthermore, the life-cycle cost of RC bridges
its effects on the life-cycle performance and cost under corrosion attack is studied in this chapter.
of RC bridges. The total life-cycle cost of the bridge is calculated
Chloride-induced corrosion is one of the from the present value of the construction cost,
deterioration mechanisms caused by the rapid inspection and maintenance costs, serviceability
intrusion of chloride ions into the concrete. This and earthquake-induced failure costs, and finally
mode of corrosion is expected when the bridge user costs associated with them. These costs are
is exposed to aggressive environments (e.g., reviewed in detail and the relevant assumptions
coastal environments or the application of deicing are discussed to provide a more realistic estimation
salts). The penetration profile of chloride ions in of the total cost. Among the mentioned costs, a
a reinforced concrete member demonstrates the special attention is paid to the serviceability and
highest chloride content near the surface with a earthquake-induced failure cost. The serviceability
decreasing trend towards the depth of the member. failure cost is incurred from necessary repair and
The chloride transport mechanism in concrete is replacement actions due to the concrete cover
a complex phenomenon that may occur in several spalling and steel rebar corrosion while the earth-
forms, such as ionic diffusion, capillary suction, quake-induced failure cost is due to the repair and
and permeation. When the concentration of chlo- rehabilitation actions after a specific seismic event
ride ions in the pore solution within the vicinity of and is dependent on the occurrence probability of
reinforcing bars becomes high enough to depas- different damage states. This cost is estimated here
sivate the protection film of the reinforcement, the from the results of probabilistic life-time fragil-
layers of rust start to form on the reinforcing bar ity analysis by introducing a performance index
surface and the corrosion of steel begins. which represents the expected performance of a
In this chapter, an integrated computational corroded bridge under a particular seismic hazard
methodology is developed to simulate the penetra- risk. This index is updated regularly over the time
tion of chloride ions into the reinforced concrete and takes into account the combined effects of
members. Through a comprehensive study, the seismic hazard and chloride-induced corrosion in
effects of various influential parameters, such as the calculation of life-cycle cost of RC bridges.
77
2. CORROSION INITIATION (2000) suggest an idealized isotherm which

AND PROPAGATION simulates the exposure conditions of marine
structures. This isotherm is called Freundlich
Chloride-induced corrosion is identified as one isotherm and its relationship can be expressed as:
of the major causes for the structural deteriora-
tion of reinforced concrete bridges. This type of C
C b = FC fF b = F FC fF 1 (3)
corrosion is initiated by the ingress of chloride C f
ions into structural concrete members during the
concentration and diffusion cycles. The sources
where F and F are the Freundlich binding con-
of chloride ions are mainly air-borne sea-salts
stants equal to 1.05 and 0.36, respectively. The
in coastal areas and deicing salts used in winter
chloride diffusion coefficient in Equation 2, DCl,
times. This chapter focuses on the corrosion
is calculated by taking into account the effects
resulting from sea-salt particles floating in the
of major influential parameters, such as water
air and assumes that the diffusion process is the
to cement ratio, ambient temperature, relative
dominant mode of chloride intrusion. In order to
humidity, age of the concrete, free chloride con-
study the diffusion process, it is essential to find
tent, and chloride binding capacity. The chloride
the changes in the chloride content at different
diffusion coefficient can be determined from a
depths of the concrete member. The total chloride
diffusion coefficient estimated for a reference
content refers to the total acid-soluble chloride in
temperature and humidity, DCl,ref, multiplied by
concrete, which is the summation of free chlorides
the modification factors as below:
and bound chlorides. The relationship among the
total, Ct, free, Cf, and bound, Cb, chloride content
DCl = DCl ,ref F1 (T )F2 (h )F3 (te )F4 (C f ) (4)
in unsaturated concrete is as follows:
C t = C b + weC f (1) where F1(T) accounts for the dependence of

chloride diffusion coefficient on the ambient
where we is the evaporable water content (m3 of temperature, F2(h) represents the influence of
evaporable water per m3 of concrete). According relative humidity, F3(te) denotes the influence
to the Ficks second law which is based on the of concrete age, and F4(Cf) considers the effects
mass conservation principle, the diffusion process of free chloride content. Table 1 summarizes the
is expressed as the change in the free chloride mathematical expressions for the modification
content over the time, t (Equation 2). factors used in Equation 4 (Baant and Najjar,
1972, Saetta et al., 1993, Bamforth and Price,
1996, Xi and Baant, 1999, Martin-Perez et al.,
C f DCl
= div (C f ) 2001, and Kong et al., 2002).
t 1 + (1 / we ) (C b / C f ) In Table 1, F1(T) is calculated using the tem-
(2) perature data, T, gathered for a specific region
where the structure is located. In the current study,
where DCl is the chloride diffusion coefficient and the temperature data of the Los Angeles area dur-
(C b / C f ) is the binding capacity. The chloride ing last 15 years (from 1995 to 2009) has been
binding capacity characterizes the relationship collected from the National Oceanic and Atmo-
between the free and bound chloride ions in con- spheric Administration (NOAA). It is evident
crete at a constant temperature and it is also referred from this database that the temperature has a
to as the binding isotherm. Martin-Perez et al. periodic trend over the year and a sinusoidal func-
78
Table 1. Summary of modification factors used for the estimation of chloride diffusion coefficient
Formulation Parameters Assigned values
log DCl ,ref = a + b log(w / c) a, b: empirical coefficients a = -10.6, b = 1.9
E: activation energy for diffusion process E = 44.6 4.3 (kJ/mol)

E 1 1 R: gas constant R = 8.314 (kJ/mol. K)
F2 (T ) = exp Tref: reference temperature Tref = 296 (K)
R Tref T
1 h
4

F3 (h ) = 1 / 1 +
1 hc hc: critical humidity level hc = 0.75

m tref: reference time tref = 28 (day)
tref m: empirical age factor m = 0.04
F4 (te ) =
t
F5 (C f ) = 1 (C f )
n
k, n: empirical parameters k = 8.366, n = 0.5
tion can be fit very well to that. The temperature methods. In the current study, a finite difference
at the j-th day of the year can be found from algorithm is developed to evaluate the chloride
Equation 5: diffusion process at different time steps by con-
stant updating of the diffusion coefficient. This
T ( K ) = 291 15 sin (2 j / 365) (5) algorithm considers the effects of all influential
parameters and provides a more accurate estima-
tion of chloride content in the concrete.
Similar to the ambient temperature, the local
For the numerical solution, it is assumed that
humidity information is needed for F2(h). The
the free chloride content in the concrete is zero
average monthly relative humidity data for the
at the initial condition. This value will gradually
Los Angeles area has been obtained from NOAA
increase by the intrusion of chloride ions over the
to find the annual trend of humidity. The relative
time. On the other hand, the free chloride content
humidity is periodic in nature and is repeated
is always constant at the concrete surface. The
throughout the years. Hence, it can be simulated
surface chloride content, Cs, depends on various
for the j-th day by a half-sinusoidal function as
parameters, such as the composition of the con-
below:
crete, location of the structure, orientation of its
surface, chloride concentration in the environment,
h (%) = 0.65 + 0.13 sin ( j / 365) (6)
and general conditions of exposure with regard to
rain and wind (Bertolini, 2008). A range of 2.95
A review of Table 1 shows that the chloride dif- kg.m-3 (McGee, 1999) to 7.00 kg.m-3 (Val, 2004)
fusion coefficient is a nonlinear parameter which has been suggested for the surface chloride content
varies over the time. Hence, the governing partial at bridges located near coastlines. In this study
differential equation given by Equation 2 cannot the average value of 5.00 kg.m-3 is assigned to Cs.
be solved without using appropriate numerical
79
Using the developed finite difference algo- chloride concentration causing depassivation of
rithm, a set of simultaneous equations are solved the concrete protection film and initiation of the
in one-day time steps to determine the chloride corrosion process. There have been many research
content in the concrete. Assuming that the water efforts during the past three decades to determine
to cement ratio is equal to 0.5, the calculated an appropriate threshold for the critical chloride
free chloride content at the depth of 50 mm has content. It had been first suggested that the critical
been shown in Figure 1. After obtaining the free value should be determined by investigating the
chloride content at each time step, the bound and free chloride concentration, but the study of Glass
total chloride contents can be calculated using and Buenfeld (2000) on the chemical aspects of
Equations 1 and 3. The changes in bound and chloride binding capacity showed that the bound
total chloride contents during a 30-year period chloride should also be taken into account. As a
can also be found in Figure 1. result, the threshold value is expected to represent
The calculated free chloride content at differ- the total chloride content. Figure 2 demonstrates
ent time steps can be used to estimate the corrosion a summary of data available in the literature
initiation time, ti. The corrosion initiation time is regarding the measured or suggested values for
determined as the time when the chloride concen- the critical chloride concentration. For this study,
tration near the reinforcing bars reaches the the critical chloride concentration is considered
threshold chloride concentration. This means: to be 1% of the cement weight which is assumed
to equal 350 kg.m-3.
C f (ti , dc ) = C critical (7) The total chloride content profile given in
Figure 1 can be used to evaluate the corrosion
initiation time. Assuming the threshold value as
where dc is the depth at which the reinforcing bars
3.5 kg.m-3 (1% of cement weight), the initiation
are placed (usually equal to the concrete cover
time can be estimated accordingly as 12.66 year
depth). In Equation 7, the Ccritical is the threshold
for. It is evident that different chloride binding
Figure 1. Change in the free, bound, and total chloride contents during a 30-year period
80
Figure 2. Summary of data available in literature for the critical chloride concentration required to
initiate the corrosion process
isotherms result in some change in the estimation relates the mass of steel consumed over the time
of the corrosion initiation time, which indicates to the amount of current that flows through the
the importance of a reasonable choice for the electrochemical corrosion cell. The rate of mass
chloride binding model. The initiation time for loss per unit length of bar subjected to the corro-
different cover depths of 40, 50, and 60 mm would sion, Mloss (gr.cm-1), for a time step of t (sec)
respectively be 7.23, 10.40, and 14.20 years can be described as:
(Figure 3). It can be seen that the values of cor-
rosion initiation time obtained from the developed M loss (t ) = k D(t )icorr t (8)
algorithm lie well within the range of 7 to 20
years, observed by Kong et al. (2002).
where D(t) is the reduced diameter of reinforc-
ing bar during the corrosion process, k, the mass
transport coefficient equal to 2.89310-9, and icorr,
3. STRUCTURAL DEGRADATION
the current per unit area of the reinforcing bar. For
DUE TO CORROSION
the icorr in Equation 8, a range of 10 to 25 A.cm-2
has been suggested by Rodriguez et al. (1994).
After the corrosion initiation time, the protection
This range corresponds to the high reinforcement
film of the reinforcing bar is depassivated and the
corrosion risk because it is larger than 1 A.cm-2
transport of iron ions starts. This results in the
(Andrade et al., 1993). In the current study, icorr is
formation of rust layers around the rebar during
assumed to equal 10 A.cm-2. By taking the steel
the corrosion process. This process continues until
mass density, s, equal to 7.8 (gr.cm-3), the change
the volume of rust reaches a level that causes the
in the volume of corroded steel, Vloss (cm3.cm-1),
concrete to crack due to the excessive expansion
can be simply calculated from Mloss. The reduced
of rust layers. In this study the crack initiation
rebar diameter after each time step of corrosion
time is calculated using the Faradays law which
is calculated as:
81
Figure 3. Estimation of corrosion initiation time for the cover depths of 40, 50, and 60 mm
D(t ) = D02 4Vloss (t ) / (9) m(t) is equal to the consumed mass of steel per
unit length divided by the original steel mass.
The reduced diameter and the remaining yield
where D0 is the initial diameter of the rebar.
strength of rebars are calculated at different time
The residual strength of corroded reinforcing
steps during the life-cycle of the bridge (Table 2).
bars was investigated experimentally by Du et
These values are used to update the characteristics
al. (2005a and b). They conducted both acceler-
of bridge models during the structural capacity
ated and simulated corrosion tests on the bars
estimation and seismic performance evaluation
embedded in the concrete and concluded that
of corroded bridges.
the strength of steel bars decreases significantly
Through a step-by-step analysis, the time in
with chloride penetration. Their test results are
which the concrete starts cracking is determined
in reasonable agreement with other studies, such
as the time when the percentage of steel mass
as Andrade et al. (1991), Lee et al. (1996), and
loss, m(t), becomes equal to a critical level, mcritical.
Morinaga (1996). Therefore, the below empirical
The mcritical which can be defined as a function of
equation proposed by Du et al. (2005a and b) is
rebar dimensions and concrete properties (El
used to estimate the time-dependent loss of yield
Maaddawy and Soudki, 2007) is calculated in the
strength in corroded reinforcing bars:
present study equal to 20%. Based on this critical
level, the crack initiation time is found to be 51
fy (t ) = (1 0.005m (t )) fy 0 (10)
days (0.14 year) for the structures under consid-
eration. To calculate the crack width after crack
where fy(t) is the yield strength of corroded rein- initiation, wcrack (mm), the analytical equation
forcement at each time step, fy0, the yield strength proposed by Vidal et al. (2004) can be used:
of non-corroded reinforcement, t, the time elapsed
since corrosion initiation (year), and m(t), the wcrack (t ) = K crack (As (t ) A0 ) (11)
percentage of steel mass loss over the time. The
82
Table 2. Reduction in the mass, diameter, and yield strength of reinforcing bars at 5-year time intervals
Time* (year) 0 5 10 15 20 25 30 35 40 45 50
(Mloss/M0)x100 0.00 6.38 12.24 18.10 23.96 29.82 35.68 41.54 47.40 53.26 59.12
D (mm) 35.80 34.75 33.61 32.46 31.32 30.17 29.03 27.88 26.74 25.59 24.45
fy/fy0 1.00 0.97 0.94 0.91 0.88 0.85 0.82 0.79 0.76 0.73 0.70
*after corrosion initiation
where As(t) is the steel loss of the rebar cross This group consists of two-span bridges with three
section during the corrosion process (mm2), A0, variations in the span lengths, representing the
the steel loss of the cross section needed for crack short-, medium-, and long-span bridges. All the
initiation (mm2), and Kcrack, an empirical coeffi- bridges have two columns at each bent and their
cient equal to 0.0577. Assuming the crack width height varies from 7.5 to 12.5 m. This provides a
of 0.3 mm as one of the first serviceability limits, range of span length-to-column height ratios from
the time in which this limit is exceeded has been 1.2 to 6.0. For the purpose of this study, the effect
calculated to equal 117 days (0.32 year). Compar- of skewness is not considered and as a result, the
ing the time to crack initiation (0.14 year) and the skew angle is assumed to equal zero degrees.
time to exceed the crack width of 0.3 mm (0.32 A schematic view of the bridges under study is
years) with the time to corrosion initiation (10.40 illustrated in Figure 4 and their dimensions are
years), it can be clearly seen that the two former summarized in Table 3.
times are negligible within the whole life-cycle OpenSees (2009) is used in this study to carry
of the bridge. Hence, considering the fact that the out a series of static and dynamic analyses which
crack initiation occurs shortly after the corrosion can provide a comprehensive performance assess-
initiation time, it is assumed that the time corre- ment of the bridges subjected to the time-depen-
sponding to the serviceability threshold is equal dent corrosion process. At different bridge ages,
to the corrosion initiation time. Furthermore, it is the remained structural capacity and expected
widely accepted that a crack width of more than seismic response are calculated by analyzing the
1 mm indicates the performance failure of the bridge mathematical models which consist of a
concrete cover. The time required for reaching variety of elements defined for the superstructure,
this crack width limit has also been calculated to pier, abutment, and foundation. The developed
equal 542 days (1.48 year) after corrosion initia- models are representative of the bridge geometric
tion. Since the capacity of structures under study characteristics, boundary conditions, material
will be evaluated every 5 years after the corrosion properties, mass distribution, and nonlinear be-
initiation time, it is assumed that the concrete havior of selected components. The detailed as-
cover is destroyed from the first analysis interval. sumptions made for each of the bridge components
are discussed below.
4. BRIDGE MODELING 1. Material Properties: The compressive

strength of concrete, fc' , is assumed to equal
To develop the probabilistic life time fragility pa- 35 MPa for the bridge columns and super-
rameters of RC bridges located in chloride-laden structure. The Poissons ratio is 0.2 and the
environments, a group of 9 box girder bridge concrete modulus of elasticity, Ec, is calcu-
models are developed and analyzed in this study. lated for the normal weight concrete using:
83
Ec = 4700 fc' (ACI-318, 2008). To capture is considered by two additional circular

the effects of confinement in columns, the layers of rebars.
properties of confined and unconfined con- 4. Abutments: The bridge abutment is modeled
crete are both taken into account following using a rigid element with a length equal to
the equations given by Mander et al. (1988). the superstructure width, connected to the
For the reinforcing bars, the yield strength superstructure centerline through a rigid
is assumed to equal 470 MPa before corro- joint. This element is supported at each end
sion begins. As discussed in the previous by three springs in longitudinal, transverse,
section, the steel yield strength decreases and vertical directions. While an elastic
over the time due to the corrosion progress spring is used for the vertical direction, the
(Equation 10). longitudinal and transverse springs are ex-
2. Superstructure Model: The bridge super- pected to present a nonlinear behavior. For
structure has been designed for four traffic this purpose, nonlinear zero-length elements
lanes, two in each direction (as it can be seen are placed in the perpendicular horizontal
in Figure 4). The roadway width is equal directions with the properties determined
to 23.0 m and the concrete cross-sectional from the Caltrans seismic design criteria
area is 12 m2. The bridge deck is modeled (SDC, 2006). The abutment stiffness and
by linear-elastic beam-column elements maximum resistance are dependent upon the
placed at the centroid of the deck cross sec- material properties of abutment backfill and
tion. These elements are subjected to linear- account for an expansion gap, here assumed
distributed loads which represent the bridge to equal 5 cm.
mass per unit length. Since the columns and
abutments are designed to experience the As discussed earlier, the corrosion process may
nonlinear behavior, no nonlinear properties cause a significant structural capacity loss which
are assigned to the superstructure elements directly affects the bridge performance under any
and they always remain in the elastic range. service and extreme loading conditions. From a
Furthermore, because the concrete super- multi-hazard point of view, the combined effects
structure always experiences some cracks of a natural event, here an earthquake, and an
due to loading conditions, the flexural stiff- environmental stressor, here chloride-induced
ness of deck section is modified by a factor corrosion, are studied over the time and the vulner-
of 0.75 according to the recommendation ability of bridge as one of the key infrastructure
of Caltrans seismic design criteria (SDC, components is evaluated. Towards this goal, the
2006). nonlinear time-history analysis is employed to
3. Pier Columns: A nonlinear three dimen- estimate the seismic response of various bridges at
sional beam-column element is used to different ages. Obtained results will be used later
model the bridge columns. This element for the probabilistic life-time fragility analysis.
is based on the iterative force formulation In order to perform nonlinear time-history
and considers the spread of plasticity along analysis, a suite of 60 earthquake ground motions
the column (OpenSees, 2009). The concrete is selected. These ground motions were originally
cross section is discretized into a number generated through the FEMA/SAC project (1997)
of fibers (total of 18 wedges and 20 rings) for the Los Angeles area and include records from
defined by the fiber module available in historic earthquakes as well as artificially-gener-
OpenSees (2009) and the steel reinforcement ated time histories. The selected suite consists of
84
Figure 4. A schematic view of the two -span bridges under study
three sets of records corresponding to earthquakes response time-histories are recorded as forces and
with 4.0410-4, 2.1010-3, and 1.3910-2 annual displacements at various bridge components. For
frequency of exceedance, which are equivalent the sake of brevity, the current study demonstrates
to 2%, 10%, and 50% probability of occurrence the seismic response only in terms of deck drift
in 50 years. Figure 5 shows the spectral accelera- ratio which is one of reliable response measures
tion values over a range of natural periods for the among diverse response parameters. The deck
three categories mentioned. For each category of drift ratio (DDR) is defined as the relative dis-
records, the median spectral acceleration has also placement of deck centroid divided by the column
been indicated which represents the expected level height. This ratio is calculated in both longitudi-
of seismic demand. nal and transverse directions, but since the ground
Using the selected suite of ground motions, motions are applied in the longitudinal direction,
the dynamic response of the bridge cases are the transverse DDR can be neglected compared
evaluated. A series of 60 nonlinear time-history to the longitudinal one. The distribution of lon-
analysis is performed for each bridge case and gitudinal DDR at different time steps is shown in
Table 3. Main dimensions of the reinforced concrete bridges under study
Bridge Case No. Number of Spans Span Length (m) Column Height (m) Column Diameter (m)
1 2 15-15 7.5 1.3
2 2 15-15 10 1.3
3 2 15-15 12.5 1.3
4 2 30-30 7.5 1.6
5 2 30-30 10 1.6
6 2 30-30 12.5 1.6
7 2 45-45 7.5 1.9
8 2 45-45 10 1.9
9 2 45-45 12.5 1.9
85
Figure 6 for the bridge with medium span length this study, the damage limit states are assumed to
and column height of 10.0 m. The review of the equal the ductility of 1.0, 2.0, 4.0, and 7.0 for the
drift data at each time step indicates that a log- slight, moderate, extensive, and complete damage
normal distribution can be fit very well to the states, respectively. The estimation of these limit
DDR response. It can be seen that both the me- states are beyond the scope of this chapter, but
dian and standard deviation of the obtained dis- the suggested values are in accordance with the
tributions increase as the corrosion progresses limit states available in the literature for similar
and the scatter plot tends to the higher DDRs. bridges (Hwang et al., 2000, Choi et al., 2004,
and Yang et al., 2009).
Under a ground motion excitation with the
5. PROBABILISTIC LIFE-TIME peak ground acceleration of PGAi (here i = 1,...,
FRAGILITY ANALYSIS 60), a bridge sustains failure in a specific dam-
age state if its ductility is larger than the ductility
Fragility analysis is considered as a powerful corresponding to that damage state. Depending
tool for the probabilistic seismic risk assessment on whether or not the bridge sustains the state
of highway bridges. Through this analysis, a set of damage under different ground motions, the
of fragility curves is developed to estimate the parameters of each fragility curve (i.e., median,
conditional probability statements of the bridge ck and log-standard deviation, k) are estimated
vulnerability as a function of ground motion in- using the maximum likelihood procedure given
tensity measure. The damageability of the bridge in Shinozuka et al. (2000b). For the k-th damage
can be assessed by expert opinions (ATC, 1985), state (k = 1, 2, 3, and 4), the fragility curve is
empirical data from past earthquakes (Basoz and developed following the formula below:
Kiremidjian, 1999 and Shinozuka et al., 2000a),
and analytical methods (Mander and Basoz, 1999 ln(PGA / c )
and Shinozuka et al., 2000b). The current study Fk (PGAi | k , ck ) = i k
(12)

k
uses the later approach and defines four limit states
of damage. The definitions of damage states are
derived from HAZUS-MH (2007) and can be sum- where Fk is the probability of exceeding the
marized as: (at least) slight, E1, (at least) moderate, damage state of k and [.] is the standard normal
E2, (at least) extensive, E3, and complete damage, distribution function. The fragility curves of the
E4. Based on these damage states, the analytical intact two-span bridges with the column height
fragility curves of the bridge cases are generated of 10.0 m, having a range of short, medium, and
at different ages after the corrosion initiation time. long span lengths, are illustrated in Figure 7.
To perform fragility analysis, the column Additionally, the estimated median values (ck) of
curvature ductility is taken here as the primary the fragility curves developed for all the bridges
damage measure. The curvature ductility is under study are summarized in Table 4 for the
defined as the ratio of maximum column cur- four damage states considering the intact bridge
vature recorded from a nonlinear time-history conditions before the corrosion initiation time. For
analysis to the column yield curvature obtained the log-standard deviation (k), it is seen that dif-
from moment-curvature analysis. Following the ferent deviation values may result in intersecting
procedure given by Priestley et al. (1996), the the fragility curves of different damage states. To
curvature ductility values of all the bridge cases avoid any intersection, Shinozuka et al. (2000b)
are calculated under the set of 60 ground motions suggest considering one common deviation value
and then compared with damage limit states. In for all the damage states. In this study, since the
86
Figure 5. Spectral acceleration plots developed

estimated log-standard deviations are very close to
for three sets of 20 ground motions with differ-
each other (ranging from 0.50 to 0.70), the median
ent seismic hazard levels: (a) 50% probability
value of log-standard deviations (equal to 0.60) is
of occurrence in 50 years; (b) 10% probability
selected as the identical log-standard deviation.
of occurrence in 50 years; (c) 2% probability of
To evaluate the effects of chloride-induced
occurrence in 50 years
corrosion on the seismic damageability of RC
bridges, the fragility curves are generated for the
case-study bridges at different time steps during
their life-cycle. Considering the extent of struc-
tural degradation (as discussed in Section 3), the
median and log-standard deviation of fragility
curves are estimated for the corroded bridges
following the procedure described for the intact
bridges. The change in the median values of fra-
gility curves is shown in Figure 8 for all the
bridges with the column height of 10.0 m. This
figure indicates that after 50 years, the overall
average of median values obtained for the four
damage states drops by 38%. For further illustra-
tion, the time-dependent fragility curves of the
bridges with medium span length and column
height of 10.0 m are depicted in Figure 9. It can
be understood from this figure that for a specific
PGA value, the probability of exceeding any
damage state increases over the time due to the
corrosion process. This increases the seismic
damageability of bridge and makes it more vul-
nerable to natural hazards. The time-dependent
fragility curves developed in this section will also
be used to predict the life-cycle cost of bridges
in a multi-hazard framework.
6. LIFE-CYCLE COST ANALYSIS OF

DEGRADED BRIDGES
The life-cycle cost (LCC) of a structure is defined

as the total cost of the structure from the begin-
ning of planning for construction to the end of
its service life time. The LCC analysis provides
a framework that helps to allocate appropriate
87
Figure 6. The distribution of DDR response over the time for the two-span bridge with the medium span
length and column height of 10 m
resources for design, construction, and operation of C sfu , the user cost associated with the probable
the structure. The focus of the current study is on service failure. The inflation is also taken into
the LCC analysis of RC bridges located in extreme account by dominating future maintenance ex-
chloride-laden environments. From the resources penditures in base year prices. These recurrent
point of view, it is important to optimize the inspec- maintenance costs are combined by weighing
tion and maintenance schedules in a way so that them according to a discount factor that takes into
the total cost of structure is minimized while the account the time value of the money. The discount
structure satisfies the performance requirements. factor, z(t), is defined as:
The LCC of a bridge consists of a one-time
initial cost associated with design and construction z (t ) = (1 + r )t (14)
of the bridge and regular inspection and mainte-
nance costs necessary at certain time intervals.
The general formula for LCC analysis can be where r is the discount rate indicating the expected
expressed as follows: market rate of return on an investment. The choice
of discount rate is often disputable in LCC analy-
LCC = C c + [C IN + C M + C Mu ] + [C sf + C sfu ] sis. In practice, the discount rate ranges from 2%
to 8%. Generally, choosing a high discount rate
(13)
favors short service life time while a low discount
rate encourages a longer service life time.
where Cc is the initial construction cost, CIN, the
The initial cost of construction is assumed to
inspection cost, CM, the maintenance cost, C Mu ,
be the summation of the costs of the bridge com-
the user cost associated with the maintenance
ponents. The cost of the deck is computed as the
procedure, Csf, the bridge service failure cost, and
88
Figure 7. Fragility curves developed for different damage states of the intact two-span bridges with the
medium span length and column height of 10 m: (a) at least slight damage state; (b) at least moderate
damage state; (c) at least extensive damage state; (d) complete damage state
total area of the deck multiplied by a cost per unit bridge piers, the construction cost includes the cost
deck area. Caltrans contract cost data (2008) sug- of steel work and concrete work. After calculating
gests $380 per m2 for deck construction cost. For the total volumes of the steel and concrete, the
Table 4. Median values of the fragility curves estimated for all the bridge cases under different damage
states
Two-Span short-span medium-span long-span

Damage State E1 E2 E3 E4 E1 E2 E3 E4 E1 E2 E3 E4
Hcol = 7.5 m 0.38 0.90 2.46 6.0* 0.32 0.68 1.36 1.90 0.42 0.78 1.36 1.90
Hcol = 10.0 m 0.42 1.28 2.46 6.0* 0.34 0.82 1.64 2.84 0.36 0.76 1.48 2.46
Hcol = 12.5 m 0.66 1.66 2.84 6.0* 0.38 0.82 2.46 6.0* 0.36 0.88 1.56 3.36
* indicates that no case of complete damage, E4, was observed.
89
Figure 8. Time-dependant median values of fragility curves obtained for the two -span bridges with the
column height of 10 m: (a) at least slight damage state; (b) at least moderate damage state; (c) at least
extensive damage state; (d) complete damage state
n
pier cost can be estimated by assuming $4 per kg C IN = Sz(i t ) (15)
of steel and $840 per m3 of concrete according to i =1
Caltrans contract cost data (2008). The average
total construction cost of short-, medium-, and where S is the cost of each inspection, n, the num-
long-span bridges under study can be seen in ber of maintenance intervals, and z, the discount
Table 5. Their inspection and maintenance costs factor from Equation 14. It can be seen that by
as well as their failure costs will be discussed in increasing the number of inspections, n, during
the upcoming sections. the life-cycle of the bridge, the inspection cost
rises accordingly. For the bridges under study,
6.1. Inspection and Maintenance the inspection cost is assumed to equal 0.5% of
Costs the construction cost. The maintenance cost over
a structures life span can be expressed as:
The cost of inspection and maintenance is expected
to be incurred at regular time intervals, t. The
inspection cost is calculate as:
90
Figure 9. Time-dependant fragility curves for the two-span bridge with the medium span length and
column height of 10 m; (a) at least slight damage state; (b) at least moderate damage state; (c) at least
extensive damage state; (d) complete damage state
Table 5. Life-cycle cost of short-, medium-, and long-span bridges under study, including initial con-
struction cost, inspection and maintenance costs, and service failure costs
Bridge Case short-span medium-span long-span

Cc $361,151 $500,117 $639,083
CIN $122,557 $169,716 $216,874
CM $154,744 $214,288 $273,832
CMu $95,171 $95,171 $95,171
Csf $90,085 $124,749 $159,412
Csf u
$45,471 $45,471 $45,471
Ctotal $869,179 $1,149,512 $1,429,843
91
n
C M = Mz (i t ) (16) fic is assumed to be 100,000 hour.vehicle.year-1,
i =1 which makes the user cost of the maintenance
program equal to $875,000. It is also assumed
where M is the cost of maintenance activity in here that the inspection procedure is so short that
base year prices. The maintenance cost is usually it leads to negligible disruptions on bridge traffic,
assumed between 0.5% to 1.0% of the construction and as a result, it causes no user cost.
cost, but since it is expected that the maintenance
cost increases as the bridge ages, it is assumed here 6.2. Service Failure Costs
that the maintenance cost has a linear increasing
trend from 0.5% to 1.0% during the structures The expected value for the service failure cost
service life time. of a bridge can be calculated using Equation 18:
In addition to the direct maintenance costs, the n
user cost associated with the temporary closure C sf = C f z (i t )p f (i ) (18)
of facilities should also be considered. The actual i =1
user cost during the regular maintenance work

depends on the extent and duration of service where Cf is the repair cost due to the service
disruption (Chang and Shinozuka, 1996). This failure, assumed to be equal to 20% of the con-
can be expressed as: struction cost. In order to obtain the expected
service failure cost, the repair cost should be
n
C Mu = t mbmuz (i t ) (17) multiplied by the relevant probability of failure
i =1 during each time interval of the bridge life-cycle.
Since the current chapter studies the effects of the
where tm is the duration of maintenance activities, corrosion process on LCC of bridges, the prob-
and bm, the index of usage disruption (0 bm ability of failure due to the corrosion process
1). For example, if maintenance entails closure between (i-1)-th and i-th time intervals, pf(i),
of one of the two lanes of a bridge, bm would be has been calculated using a recursive formula
0.5. In Equation 17, u is the unit user cost which suggested by Val and Stewart (2003) in Box 1:
depends on the volume and type of traffic crossing where p is the cumulative distribution function
the bridge as well as the availability of convenient for the time of service failure. Since the crack
alternative routes. The user cost also typically initiation and propagation time (calculated in
includes the increased costs associated with travel Section 4.2) is small comparing to the corrosion
delays and accidents. In the current study, it is initiation time, the service failure is assumed to
assumed that for the maintenance program, only occur after the corrosion initiation time. The
1/4 of bridge is closed at each period of time for values of pf(i) have been calculated for different
one week. Hence, the average hourly user cost is inspection intervals and are shown in Figure 10
calculated to be $8.75 per vehicle. The annual traf- for the entire life-cycle of the bridge, which is
Box 1.
i 1
p f (i ) = p (i t ) p (i 1) t + p f ( j ) {p (i j ) t p[(i j 1) t ]} (19)
j =1
92
assumed to be 50 years after the corrosion initiation and maintenance costs as well as the service
tion time. It is evident that as the number of in- failure costs of all the bridges are calculated and
spections decreases, the probability of failure due the average costs are summarized in Table 5. The
to the corrosion process increases. According to values of Table 5 are obtained with the assumption
Chang and Shinozuka (1996), the expected user of 4% annual inflation rate and scheduling an-
cost associated with the probable service failure nual inspection and maintenance program. This
can be calculated as below: table shows that the user associated costs, C Mu
andC sfu , do not change for different bridge types,
n
C sfu = tsf b sf u z (i t )p f (i ) (20) which indicates that they are independent of the
i =1 construction cost.
In order to evaluate the effects of inspection
where tsf is the duration of repair activities after and maintenance intervals on the LCC of bridges,
failure, and bsf, the usage disruption parameter different inspection and maintenance strategies
due to the service failure. In this study, it is as- have been examined. In this study, it is assumed
sumed that it takes one month for the bridge to that the inspection and maintenance intervals are
be repaired and during this time half of the bridge the same and can be scheduled every 1, 2, 3, 4, or
will be closed to traffic. 5 years. Table 6 indicates the average inspection,
maintenance, and service failure costs for the
6.3. Optimized Inspection and medium-span bridge inspected and maintained
Maintenance Intervals according to the proposed time schedules. The ratio
of different cost types to the initial construction
Based on the total construction cost of the short-, cost are depicted in Figure 11. Referring to Table
medium-, and long-span bridge cases, the inspec- 6 and Figure 11, when the inspection intervals
Figure 10. Probability of service failure due to the corrosion process for various inspection intervals
93
increase, the costs associated with the inspec- where Cef is the bridge failure cost due to earth-
tion and maintenance decrease but this causes quake event, and C efu , the user cost associated
a significant increase in the service failure cost. with the probable bridge failure.
The total LCC of the bridge has been shown in The failure cost is assumed to equal the cost
Figure 12, considering the decreasing trend of associated with the repair and replacement of the
inspection and maintenance costs and increas- damaged parts of a bridge during probable natural
ing trend of the service failure costs. From this hazards. Since the current chapter evaluates the
figure, it can be understood that the total LCC of seismic performance of corroded bridges, the state
the bridge can be minimized if the inspection and of damage after an earthquake event can be consid-
maintenance intervals are scheduled for every 2 ered as an assessment measure for the estimation
or 3 years. This schedule optimizes the inspection of failure cost. Through a probabilistic approach,
and maintenance costs while ensuring the safety the expected service failure cost is calculated for
of the bridge. each of the damage states and the results are then
combined with appropriate weighting factors.
6.4. Earthquake-Induced Failure Cost This procedure is repeated over the entire life-
cycle of the bridge (n time intervals) to calculate
This section focuses on the LCC analysis of RC the total failure cost by taking into account the
bridges considering the combined effects of natu- failure probabilities which are updated at each
ral hazards and environmental stressors on the time interval based on the corrosion process. The
estimation of the LCC. The extent of structural general formula for the estimation of failure cost
degradation and capacity loss due to the corrosion is as below:
process were discussed in the previous sections
and it is evident that an optimized plan for the n 4
inspection and maintenance of bridges is necessary C ef = PI (dk , i t )rk z (i t )C c (22)

i =1 k =1
to avoid any structural failure under the service
or seismic loads. To satisfy overall performance
requirements while minimizing the total resource where rk is the damage ratio (will be discussed
costs, the LCC of bridges is evaluated in this study later) and PI is the performance index for the
by taking into account the structural performance damage state of dk (k = 1, 2, 3, and 4) at i-th time
criteria in addition to the key cost parameters. interval. The performance index represents the
As mentioned earlier the total LCC of the overall performance of a particular corroded state
bridge includes a one-time initial cost required of the bridge under a specified seismic hazard risk
for the design and construction of the bridge, and it can be determined in terms of the annual
some regular inspection and maintenance costs probability of exceeding a given damage state
necessary at certain time intervals, and the costs considering the effects of deteriorating mecha-
associated with the serviceability failure of the nisms. This index is calculated from Equation
corroded bridge due to earthquake events. Based 23, as follows:
on the mentioned costs, Equation 13 is updated

as follows to account for the earthquake-induced dH (x )
PI (dk , i t ) = P (DS > d
k ,i k | x) dx
failure cost for the degraded bridges. 0
dx
(23)
LCC = C c + [C IN + C M + C Mu ] + [C ef + C efu ]
(21)
94
Table 6. Effects of inspection/maintenance intervals on the life cycle cost of the medium-span bridge
Interval annual 2-year 3-year 4-year 5-year

Cc $500,117 $500,117 $500,117 $500,117 $500,117
CIN $169,716 $86,887 $54,368 $39,966 $32,798
CM $214,288 $112,618 $69,391 $51,288 $43,249
CM u
$95,171 $48,723 $30,488 $22,412 $18,392
Csf $124,749 $237,982 $340,877 $431,840 $500,949
Csfu $45,471 $86,744 $124,249 $157,405 $182,595
Ctotal $1,149,512 $1,073,071 $1,119,490 $1,203,028 $1,278,100
where Pk,i is the probability of suffering the damage location of the bridge. For further clarification,
state of dk (i.e., DS > dk) under the ground motion the current chapter demonstrates an application
intensity (here PGA) of x. This probability can be of the explained approach in the calculation of
obtained at each time interval from the updated the failure cost of the case study bridges. It is
fragility curve developed for that damage state. assumed that these bridges are located at three
On the other hand, the probability of exceeding different parts of the Los Angeles area which are
the ground motion intensity of x during the service similar in terms of exposure to chloride ions but
life-time of the bridge can be calculated from a seis- are different in terms of the seismic hazard risk.
mic hazard curve, H(x), generated for the specific The seismic hazard curves corresponding to these
Figure 11. Ratio of different cost items to the initial construction cost
95
Figure 12. Total LCC of the medium-span bridge obtained for different inspection and maintenance
intervals
locations have been extracted from the USGS The required steps for this procedure have also
database and shown in Figure 13. For the purpose been schematically illustrated in Figure 14.
of this study, these hazard curves represent high, In addition to the failure cost which is di-
medium, and low hazard risks and indicate a PGA rectly related to the repair and replacement costs
of 0.48, 0.32, and 0.20g for 10% probability of of a bridge, there are user costs associated with
exceedance in 50 years, respectively. the closure of part or whole of a bridge to traffic
Using Equation 23, the performance indices after an earthquake. Similar to the user cost of the
for the three levels of seismic hazard risk are maintenance procedure, the total user cost of the
calculated in different states of damage at a range bridge failure is calculated based on a unit user
of PGA values. The obtained results are multiplied cost, u, which is assumed to be identical in both
by damage ratios, rk, which are employed as cases. By taking into account the time- and
weighting factors to adjust the economic loss due damage-dependent performance indices obtained
to the different damage states. HAZUS-MH (2007) from Equation 16, the total user cost of the bridge
provides some ranges for damage ratios and also failure, C efu , can be expressed as:
suggests the best estimates within each range for
the slight, moderate, extensive, and complete n 4
C efu = PI (dk , i t ) t f (dk )bf (dk )u z (i t )
damage states (Table 7). By repeating the explained i =1 k =1
procedure for all time steps, the earthquake in- (24)
duced failure cost is estimated from Equation 22.
96
Figure 13. Seismic hazard curves for three different locations in the Los Angeles area (USGS)
where tf is the duration of repair activity and bf is for the regular inspection and maintenance ac-
the usage disruption parameter due to the seismic tivities, which include 1-, 2-, and 5-year time
damage (0 bf 1). These two parameters depend intervals. Considering three levels of seismic
on the state of damage, dk, and the values assigned hazard risk, the ratio of total failure cost to initial
to them are summarized in Table 8. construction cost is shown in Figure 15 for all the
The summation of failure cost and its associ- bridge cases. As it can be seen from this figure,
ated user cost is calculated for each of the bridge the obtained ratios have an increasing trend when
cases and then the total failure cost is normalized the inspection and maintenance activities are less
to the relevant initial construction cost for com- frequent and as a result, the bridges have become
parison purposes. There are also three assumptions more vulnerable to seismic events. As a case in
Table 7. Damage ratios for RC highway bridges (HAZUS-MH,2007)
Damage State Range of Damage Ratios Best Estimate Damage Ratio

slight 0.01-0.03 0.03
moderate 0.02-0.15 0.08
extensive 0.10-0.40 0.25
complete 0.03-1.00 1.00*
*If the number of spans is greater than two, then the best estimate
damage ratio for complete damage is [2/(number of spans)]
97
Figure 14. Summary of required steps for the calculation of the failure cost of corroded bridges due to
probable earthquake events
98
Figure 15. Ratio of total failure cost to initial Table 8. Values assumed for bf and tf at different
construction cost for three levels of seismic hazard damage states
risk (Note: The geometric characteristics of each
bridge case number can be found in Table 3): Damage State bf tf
(a) low seismic hazard risk; (b) medium seismic slight 0 0
hazard risk; (c) high seismic hazard risk moderate 0.25 1 month
extensive 0.50 6 months
complete 1.00 24 months
point, the ratio of total failure cost increases be-

tween 5% to 19% for various bridge cases as the
inspection and maintenance interval changes from
every 1 year to every 5 years. On the other hand,
planning for more frequent inspection and main-
tenance intervals may cause an increase in the
total life-cycle cost of a bridge. Selecting the
two-span bridge with the medium span length and
column height of 10.0 m as the case-study, it is
found that the ratio of total life-cycle cost (LCC)
to initial construction cost is equal to 2.7, 2.4, and
3.3 for the inspection and maintenance intervals
of 1, 2, and 3 years, respectively. This indicates
that a 2-year interval would be probably an opti-
mized interval which saves more money while
limiting the probability of bridge failure.
As discussed earlier, the current chapter pro-
vides a multi-hazard framework for the estimation
of the total life-cycle cost of bridges. But one may
question the importance of the proposed frame-
work considering the efforts required to take into
account the deterioration process in addition to
the seismic hazard risk. To respond to this concern,
a comparison has been made between the ratios
of total failure cost to initial construction cost
calculated with and without the effects of the
corrosion process (Figure 16). It can be found that
this ratio is underestimated by a factor of 47% on
average in case that the corrosion process is ig-
nored. This indicates that the accuracy of life-
cycle cost estimations can be improved signifi-
cantly through the developed framework.
In order to simplify the evaluation of the cor-
rosion process and easily incorporate it into the
99
Figure 16. Comparison between the ratios of total failure cost to initial construction cost calculated with
and without the effects of corrosion (Note: The geometric characteristics of each bridge case umber
can be found in Table 3)
life-cycle cost analysis, adoption of some relation- the rate of change in the performance index value
ships which predict the extent of change in the over the time given a specific state of damage
performance of a bridge over the time is proposed. (dk). The expected values of A(dk) obtained from
For this purpose, the performance index is chosen the detailed analysis of the bridge cases for dif-
here as an appropriate measure since it considers ferent seismic hazard risks and damage states are
the combined effects of corrosion and earthquake summarized in Table 9. The performance indices
scenarios and can be directly used in the calcula- extracted from Equation 18 can be used to predict
tion of the life-cycle cost. For the bridges consid- the failure costs in bridge cases with similar dete-
ered in the current study, with the given struc- rioration scenarios and seismic hazard conditions.
tural degradation rate and seismic hazard risk, a
series of linear regression analysis is conducted
to provide the users with an equation which updates 7. CONCLUSION
the value of the performance index over the life-
cycle of the bridge. This Equation can be expressed Reinforced concrete highway bridges are continu-
as follows: ously exposed to different environmental stressors
during their service life time. Chloride-induced
PI (dk , t ) = PI 0 (dk ) + A (dk )t (25) corrosion is one of deterioration mechanisms that
causes serious structural degradation and may
result in the service failure of the entire bridge.
where PI0(dk) is the performance index calculated
This chapter develops a comprehensive framework
for a bridge in the intact condition (no deteriora-
to study the chloride intrusion mechanisms and
tion) and A(dk) is the slope of the line which shows
100
Table 9. Suggested values for A(dk) in Equation 25
Seismic Hazard Risk

Damage State Low Medium High
slight 3.26 10 -05
5.44 10-05
9.93 10-05
moderate 5.53 10-06 1.05 10-05 2.38 10-05

extensive 1.66 10 -06
3.74 10-06
9.12 10-06
complete 5.46 10-07 1.43 10-06 3.66 10-06
predict the progress of corrosion in reinforced total life-cycle cost of the bridge while maintains
concrete members. Through this study, the effects its expected structural performance.
of various influential parameters, such as water This chapter also provides a multi-hazard
to cement ratio, ambient temperature, relative framework that evaluates the life-cycle perfor-
humidity, concrete age, free chloride content, and mance and cost of reinforced concrete highway
binding capacity, are carefully considered for an bridges. The bridges under consideration are lo-
accurate estimation of the chloride penetration cated in seismic areas and they are continuously
profile in deteriorating structural members over exposed to the attack of chloride ions. As a result,
the time. it is necessary to study the combined effects of a
The corrosion initiation time is estimated by natural hazard and an environmental stressor over
comparing the chloride content values in the vi- the time. The time-dependent seismic fragility
cinity of reinforcing bars with critical thresholds. curves of the bridges are then generated using a
After the determination of the corrosion initiation set of damage states for the purpose of seismic
time, the rate of reduction in geometry and material risk assessment. In addition to the probabilistic
properties due to the corrosion process is calculated structural evaluation, the results of fragility analy-
at different time steps. This chapter specifically sis are employed to estimate the total life-cycle
considers the reduction rate of diameter and yield cost of the bridges. To consider both earthquake
strength of reinforcing bars in corroded members. and corrosion scenarios, a performance index is
The crack initiation and propagation in concrete introduced which represents the vulnerability of a
cover is also investigated in order to modify the corroded bridge under a specified seismic hazard
confinement assumptions required for capac- risk. Comparing the life-cycle costs calculated with
ity evaluation. Based on the updated structural and without the effects of corrosion, it is found
characteristics of corroded members, the capacity that the suggested performance index plays a key
loss of a group of RC bridges with short, medium, role in obtaining more realistic estimates of the
and long spans is evaluated. In addition to the total costs and it provides valuable information to
structural degradation, the effects of the corrosion optimize the inspection and maintenance intervals.
process on the life-cycle cost of bridges are also The time-dependent variation of this index is also
studied. From the detailed calculation of construc- formulated to be directly used in the life-cycle cost
tion, inspection, maintenance, and service failure analysis of similar cases without going through
costs of the bridges, this chapter examines various the detailed corrosion models.
inspection and maintenance strategies and sug-
gests the optimized inspection and maintenance
intervals. The proposed procedure minimizes the
101
REFERENCES California Department of Transportation. (2008).

Caltrans contract cost data. Sacramento, CA:
American Concrete Institute. (2008). ACI-318: Author.
Building code requirements for structural concrete
and commentary. ACI. Chang, S., & Shinozuka, M. (1996). Life cycle cost
analysis with natural hazard risk. Journal of In-
Andrade, C., Alonso, C., Garcia, D., & Rodriguez, frastructure Systems, 2(3), 118126. doi:10.1061/
J. (1991). Remaining life time of reinforced con- (ASCE)1076-0342(1996)2:3(118)
crete structures: effect of corrosion in mechanical
properties of the steel, life prediction of corrodible Choi, E., DesRoches, R., & Nielson, B. (2004).
structures. Proceedings of the International Sym- Seismic fragility of typical bridges in moder-
posium of the National Association of Corrosion ate seismic zones. Journal of Engineering
Engineers, 12/1-12/11, Cambridge, U. K. Structures, 26(2), 187199. doi:10.1016/j.eng-
struct.2003.09.006
Andrade, C., Alonso, C., & Molina, F. J. (1993).
Cover cracking as a function of rebar corrosion: Du, Y. G., Clark, L. A., & Chan, A. H. C. (2005a).
Part I: Experimental test. Materials and Structures, Residual capacity of corroded reinforcing bars.
26(3), 345353. Magazine of Concrete Research, 57(3), 135147.
doi:10.1680/macr.2005.57.3.135
Applied Technology Council. (1985). ATC-13:
Earthquake damage evaluation data for Califor- Du, Y. G., Clark, L. A., & Chan, A. H. C. (2005b).
nia. Redwood City, CA: ATC. Effect of corrosion on ductility of reinforcing bars.
Magazine of Concrete Research, 57(7), 407419.
Bamforth, P. B., & Price, W. F. (1996). An inter- doi:10.1680/macr.2005.57.7.407
national review of chloride ingress into structural
concrete. Rep. No. 1303/96/9092. Middlesex, UK: El Maaddawy, T., & Soudki, K. (2007). A model
Taywood Engineering Ltd., Technology Division. for prediction of time from corrosion initiation to
corrosion cracking. Journal of Cement and Con-
Basoz, N., & Kiremidjian, A. (1999). Development crete Composites, 29(3), 168175. doi:10.1016/j.
of empirical fragility curves for bridges. Proceed- cemconcomp.2006.11.004
ings of 5th US Conference on Lifeline Earthquake
Engineering, ASCE, New York, USA. Glass, G. K., & Buenfeld, N. R. (2000). The influ-
ence of chloride binding on the chloride induced
Bazant, Z. P., & Najjar, L. J. (1972). Nonlinear corrosion risk in reinforced concrete. Corrosion
water diffusion in unsaturated concrete. [Paris, Science, 42(2), 329344. doi:10.1016/S0010-
France.]. Materials and Structures, 5(25), 320. 938X(99)00083-9
Bertolini, L. (2008). Steel corrosion and service HAZUS-MH/MR3. (2007). FEMAs software
life of reinforced concrete structures. Structure program for estimating potential losses from
and Infrastructure Engineering, 4(2), 123137. disasters.
doi:10.1080/15732470601155490
Hwang, H., Liu, J. B., & Chiu, Y.-H. (2001).
California Department of Transportation. (2006). Seismic fragility analysis of highway bridges.
Caltrans seismic design criteria. Sacramento, Tennessee, USA: Center of Earthquake Research
CA: Author. and Information, The University of Memphis.
102
Kong, J. S., Ababneh, A. N., Frangopol, D. M., Priestly, M. J. N., Seible, F., & Calvi, G. M. (1996).
& Xi, Y. (2002). Reliability analysis of chloride Seismic design and retrofit of bridges. USA: John
penetration in saturated concrete. Journal of Wiley and Sons Inc. doi:10.1002/9780470172858
Probabilistic Engineering Mechanics, 17(3),
Rodriguez, J., & Ortega, L. M. (1994). Assessment
302315. doi:10.1016/S0266-8920(02)00014-0
of structural elements with corroded reinforce-
Lee, H. S., Tomosawa, F., & Noguchi, T. (1996). ment. Corrosion and Corrosion Protection in Con-
Effect of rebar corrosion on the structural perfor- crete. Sheffield, UK: Sheffield Academic press.
mance of single reinforced beams. In Lacasse, M.
Saetta, A. V., Schrefler, B. A., & Vitaliani, R. V.
A., & Vanier, D. J. (Eds.), Durability of building
(1993). The carbonation of the concrete and the
material and components (pp. 571580). London,
mechanism of moisture, heat and carbon dioxide
UK: E & FN Spon.
flow through porous material. Journal of Ce-
Mander, J. B., & Basoz, N. (1999). Seismic fragil- ment and Concrete Research, 23(4), 761772.
ity curve theory for highway bridges. Proceedings doi:10.1016/0008-8846(93)90030-D
of the 5th US Conference on Earthquake Engi-
Shinozuka, M., Feng, M. Q., Kim, H.-K., & Kim,
neering TCLEE No.16, (pp. 31-40).
S.-H. (2000b). Nonlinear static procedure for fra-
Mander, J. B., Priestly, M. J. N., & Park, R. (1988). gility curve development. Journal of Engineering
Theoretical stress-strain model for confined Mechanics, 126(12), 12871296. doi:10.1061/
concrete. Journal of the Structural Division, (ASCE)0733-9399(2000)126:12(1287)
114(8), 18041826. doi:10.1061/(ASCE)0733-
Shinozuka, M., Feng, M. Q., Lee, J., & Naanu-
9445(1988)114:8(1804)
ma, T. (2000a). Statistical analysis of fragil-
Martin-Perez, B., Pantzopoulou, S. J., & Thomas, ity curves. Journal of Engineering Mechanics,
M. D. A. (2001). Numerical solution of mass 126(12), 12241231. doi:10.1061/(ASCE)0733-
transport equations in concrete structures. Journal 9399(2000)126:12(1224)
of Computers and Structures, 79(13), 12511264.
Somerville, P., Smith, N., Punyamurthula, S., &
doi:10.1016/S0045-7949(01)00018-9
Sun, J. (1997). Development of ground motion
McGee, R. (1999). Modeling of durability per- time histories for Phase 2 of the FEMA/SAC steel
formance of Tasmanian bridges. Proceedings of project. SAC joint venture, California, USA.
the ICASP8 Conference, (pp. 297-306). Sydney, U.S. Geological Survey. (USGS). Retrieved from
Australia. www.usgs.gov
Morinaga, S. (1996). Remaining life of reinforced Val, D., & Stewart, M. G. (2003). Life-cycle cost
concrete structures after corrosion cracking. Du- analysis of reinforced concrete structures in marine
rability of Building Material and Components (pp. environments. Journal of Structural Safety, 25(4),
127136). London, UK: E & FN Spon. 343362. doi:10.1016/S0167-4730(03)00014-6
OpenSees Development Team. (2009). OpenSees: Val, D. V. (2004). Aspects of corrosion in rein-
Open system for earthquake engineering simula- forced concrete structures and its influence on
tion. Berkeley, CA, USA. structural safety. Report No. 2002950, National
Building Research Institute.
103
Vidal, T., Catel, A., & Francois, R. (2004). Analyz- Yang, C. S., DesRoches, R., & Padgett, J. E. (2009).
ing crack width to predict corrosion in reinforced Fragility curves for a typical California box girder
concrete. Cement and Concrete Research, 34(1), bridge. Proceedings of the 2009 ASCE Technical
165174. doi:10.1016/S0008-8846(03)00246-1 Council on Lifeline Earthquake Engineering Con-
ference, San Francisco, California, USA.
Xi, Y., & Bazant, Z. P. (1999). Modeling chlo-
ride penetration in saturated concrete. Journal
of Materials in Civil Engineering, 11(1), 5865.
doi:10.1061/(ASCE)0899-1561(1999)11:1(58)
104
105
Chapter 5
Efficient Robust Optimization
of Structures Subjected
to Earthquake Load and
Characterized by Uncertain
Bounded System Parameters
Subrata Chakraborty
Bengal Engineering and Science University, Shibpur, India
Soumya Bhattacharjya
Bengal Engineering and Science University, Shibpur, India
ABSTRACT
An efficient robust design optimization (RDO) procedure is proposed in the framework of an adaptive
response surface method (RSM) for structures subjected to earthquake load and characterized by un-
certain but bounded system parameters. The basic idea of the proposed RDO approach is to improve
the robustness of a design by using a new dispersion index which utilizes the relative importance of the
gradients of the performance function. The same concept is also applied to the constraints. The repeated
computations of stochastic responses and their sensitivities for evaluating the stochastic constraint of
the associated optimization problem are efficiently obtained in the framework of an adaptive RSM. The
proposed RDO approach is elucidated through the optimization of a three-storied concrete frame struc-
ture. The numerical study depicts that the proposed RDO results are in conformity with the conventional
RDO results. However, definite improvements are achieved in terms of robustness and computational
time requirements indicating its efficiency over the conventional RDO approach.
DOI: 10.4018/978-1-4666-1640-0.ch005
Efficient Robust Optimization of Structures Subjected to Earthquake Load
INTRODUCTION or nominal value of a performance function disre-

garding its variation due to uncertainty. The RBDO
The response of a structural system under en- ensures a target reliability of a design for a specific
vironmental loads such as wind, water wave, limit state. An efficient RBDO was presented for
earthquake, etc. is highly uncertain and can be best linear (Jensen, 2005) and nonlinear (Jensen, 2006)
modelled as a stochastic process. The optimization deterministic dynamic systems under earthquake
of structure under such loads is normally dealt in load. Jensen et al. (2008) have further extended
the literature in the form of standard nonlinear the approach to include the randomness in the
optimization problem. The dynamic responses system parameters. Lagaros et al. (2008) proposed
to define the stochastic constraints of the related an RBDO procedure for computationally intensive
optimization problem are obtained by random system under earthquake load and random system
vibration theory. Subsequently, a standard non- parameters. Mohsine et al. (2005) presented an
linear optimization problem is formulated where RBDO method where solution has been achieved in
the weight of the structure or a desired stochastic a hybrid design space (HDS) considering probabi-
response quantity is minimized. The procedure is listic variations of parameters under deterministic
termed as stochastic structural optimization (SSO). dynamic load. The HDS considers reliability level
The details of the relevant developments can be in the same design space of the objective func-
found in (Nigam, 1972), Kang et al. (2006). It may tions and constraints. An optimum seismic design
be underlined here that in a typical SSO procedure criterion was proposed by Marano et al. (2006)
the dynamic load is considered to be the only for elastic structures considering deterministic
source of randomness in many cases and all other system parameters. The optimum design of struc-
system parameters are assumed to be determinis- ture considering system parameter uncertainty as
tic. But, uncertainty in the system parameters is discussed above is mainly accomplished in the
inevitable to model a realistic structural system framework of RBDO to ensure a target reliability
and incorporation of such uncertainty creates an of structure with respect to desired performance
interaction between the stochastic descriptions of modes. It may be noted here that the studies on
the loads and the uncertain parameters (Jensen, the optimization of dynamic system considering
2002). Furthermore, the effect of system parameter system parameter uncertainty primarily apply
uncertainty is important as the safety of structure the total probability theory concept to obtain the
may be endangered due to this (Chaudhuri, & unconditional response or the failure probability
Chakraborty, 2006) and can affect the final optimal of the system which is subsequently used as the
design significantly (Schuller, & Jensen, 2008). performance measure. However, the optimization
Thus, there is a growing interest to consider the has been performed without any consideration to
effect of uncertainty in the optimization process the possible variation of the performance of the
for economic design of structure ensuring neces- structure due to system parameter uncertainty. It
sary safety requirements. may be realized that such a design approach not
The developments in the optimum design of necessarily corresponds to an optimum design in
structure under uncertainty can be divided into terms of minimum dispersion of the performance
three broad categories: (i) performance based objective of the design. Rather, the system may
design optimization (PBDO), (ii) reliability based be sensitive to the variations of the system pa-
design optimization (RBDO), and (iii) robust rameters due to uncertainty. In order to obtain a
design optimization (RDO). The limitation of the more viable optimum design, the RDO approach
PBDO is quite obvious as it optimizes the mean is more desirable which optimizes a performance
106
index expressed in terms of the mean value of the The RDO Method: Developments
performance function (obtained by the so called
RBDO approach) as well as its dispersion due to The concepts of RDO have been developed
uncertainty. Thereby, a design configuration of the independently in different scientific disciplines
structure can be achieved so that the performance and the developments in the recent past are note-
objective is less sensitive to the variations due to worthy as evident from the works of Park et al.
system parameter uncertainty. Furthermore, the (2006); Beyer, & Sendhoff (2007). The limited
RBDO of structure, based on a probabilistic de- information on uncertainty is usually integrated
scription of uncertain parameters attains its limita- with a nondeterministic optimization framework
tion when sufficient reliable data are not available to obtain an RDO (Park, Lee, & Hwang, 2006).
for describing the system parameter uncertainty This approach is often referred as sensitivity
of a real life system. In fact, often a probabilistic based approach. There are various such RDO
description of uncertainty arising from insufficient approaches adopted by different researchers e.g.
information is warned in order to incorporate our robust counterpart approach (Lewis, 2002), semi-
partial knowledge about the system. A preferable definite programming (Ben-Tal, & Nemirovski,
approach is to model the system parameters as 2002; Bertsimas, & Sim, 2004), worst case sensi-
uncertain but bounded (UBB) type. In such a case, tivity region concept (Gunawan, & Azarm, 2005),
the RDO approach becomes an attractive alterna- minimization of sensitivity matrix (Al-Widyan, &
tive to the RBDO approach. The RDO approach Angeles, 2005) etc. A new semi-analytical method
is fundamentally concerned with minimizing the to calculate the sensitivity of stability boundary
effect of uncertainty in the Design Variables (DVs) for a system of delay differential equations was
(the specific system parameters designer needs presented by Kurdi et al. (2008). Guo et al. (2009)
to optimize to achieve a desired performance) proposed a bi-level programming technique us-
and the Design Parameters (DPs) (which cannot ing a semi-definite programming to solve RDO
be controlled by the designer or are difficult and problem under non-probabilistic and non-convex
expensive to control). The subject of the present stiffness and load uncertainties. The study on
chapter is the RDO of structures under stochastic RDO procedure in the field of stochastic dynamic
load (earthquake to be specific) considering UBB systems is comparatively a less attempted area
type system parameters. compared to the deterministic design optimization
(DDO) and the RBDO procedures. Hwang et al.
(2001) minimized the mean and the variance of
BACKGROUND displacement at the first resonance frequency of
an automobile mirror considering system param-
To present the proposed RDO approach, it will be eter uncertainty. Zang et al. (2005) reviewed the
informative to discuss the background of RDO of applications of optimization of dynamic system
structures with emphasis on optimization of sto- and presented an RDO procedure for a vibration
chastic dynamic system. In doing so, the related absorber considering mass and stiffness uncertain-
developments in the field of RDO are presented ties under deterministic sinusoidal load. Son and
first to justify the relevance of the present study. Savage (2007) proposed a probabilistic design of
Subsequently, the concept of conventional RDO vibration absorber parameters to reduce the mean
approach, the SSO under earthquake load as- as well the variance of dynamic performance
suming deterministic system parameters and the measure. Marano et al. (2008) investigated RDO
metamodelling based approximation of stochastic solution for a tuned mass damper (TMD) system
dynamic responses are briefly presented. in seismic vibration mitigation. Taflanidis and
107
Beck (2008) proposed an RDO procedure for it is intuitively expected that the importance of
a base-isolation system under earthquake load the individual gradient should also provide use-
considering system parameter uncertainty. The ful information to measure the robustness of the
unconditional performance function was evaluated performance of a design and the concept can be
through stochastic simulation. Guedri et al. (2009) applied in the RDO procedure. In this regard,
proposed a stochastic metamodel based approach it is of worth mentioning that most of the engi-
integrated with an RDO procedure to reduce the neering design problems are strongly based on
cost of uncertainty analysis. In a recent study, computationally expensive complex computer
Marano et al. (2010) presented an RDO criterion code and numerical analysis. For a large-scale
for a TMD system in seismic vibration control of system design, a preferable strategy is to utilize
structures. Taflanidis (2010) presented an RDO the metamodelling technique to approximate the
procedure of a linear dynamic system under implicit performance functions and constraints
stochastic stationary excitation using simulation (Jurecka, Ganser, & Bletzinger, 2007). However,
based approach. the accuracy of the metamodel based optimization
The review of currently available literature approach relies on how accurate the response sur-
reveals that the developments in the field of face method (RSM) is in capturing the performance
RDO procedures have been taken place in three variations during the iteration cycles of a typical
distinct areas: (i) mathematical formulation of numerical optimization procedure (Jin, Chen, &
RDO procedure as a SSO problem or deterministic Simpson, 2001). Generally, the RSM is based on
programming (Du, & Chen, 2000; Lee, & Park, the least-squares method (LSM) which is primar-
2001; Gunawan, & Azarm, 2005), (ii) solution ily a global approximation of scatter position data
strategies (Du, Sudijianto, & Chen, 2004; Beyer, (Myers, & Montogmery, 1995). It is well-known
& Sendhoff, 2007), and (iii) assessment of robust- that the LSM is one of the major sources of error
ness (Huang, & Du, 2007). The literature on RDO in the response approximation by the RSM. The
procedures largely indicates that the applications moving least-squares method (MLSM), basically
of RDO procedures in dynamics is lesser compared a local approximation approach is found to be
to its applications in static. Moreover, in many more efficient in this regard (Kim, Wang, & Choi,
such applications, the dynamic load is consid- 2005). However, the studies addressing the RDO
ered to be deterministic in nature. It is generally of structure using the MLSM based metamodel-
observed that the existing RDO formulations put ling technique is observed to be scarce than the
equal importance to each individual gradient of applications of the MLSM addressing the RBDO
the performance function and constraints. But, it or the DDO procedures.
is well-known to the structural reliability com- The focus of the present chapter is on an
munity that all the gradients of a performance improved RDO strategy for structures subjected
function are not of equal importance (Gupta, & to stochastic earthquake load and characterized
Manohar, 2004; Haldar, & Mahadevan, 2000). by UBB type DVs and DPs. The formulation is
In fact, when a large numbers of DVs and DPs proposed in the framework of MLSM based adap-
are involved in a structural reliability analysis tive RSM. The basic idea of the proposed RDO
problem, the dominant parameters having rela- approach is to improve the robustness of the per-
tively stronger influence on the reliability are formance function using a new dispersion index,
identified by using the relative importance of the which utilizes the weight factors proportional to
gradients. The concept has been successfully used the importance of each gradient of the performance
in the reduction of number of random variables function. The same concept is also applied to the
in large scale reliability analysis problems. Thus, constraints. The repeated computations of the
108
stochastic dynamic responses and their sensitivi- for a performance function, f (u), its nominal
ties for evaluating the stochastic constraint of the value is obtained as: f = f (u) , where u is a vec-
related optimization problem have been avoided tor comprising of nominal values of u. Generally,
by applying the MLSM based adaptive RSM. The the RDO is performed by improving the robust-
proposed RDO approach is elucidated through the ness of the performance function by minimizing
optimization of a three-storied concrete frame. The a gradient index obtained through first-order
numerical results obtained by the conventional Taylor series expansion (Lee, & Perk, 2001) as
and the proposed RDO approaches are presented defined below:
to demonstrate the effectiveness of the proposed
RDO approach. N
f
f = ui (2)
i =1 u i
The Conventional RDO Approach
Most of the developments in the field of RDO as

discussed in the previous section primarily use where, N is the total number of DVs and DPs and
the weighted sum method (WSM) for examples as ui quantifies their uncertainty amplitudes. The
adopted by Du &Chen (2000), Lee & Perk (2001). objective of an RDO is to achieve optimum per-
This approach is termed in the present study as formance of the design as well as its less sensitiv-
the conventional RDO approach in which the ity with respect to the variations of DVs and DPs
robustness of a design performance is expressed due to uncertainty. This leads to a dual criteria
in terms of the dispersion of performance function performance function. This dual criteria perfor-
from its nominal value. mance function is transformed to an equivalent
Let, u = (x, z) is a vector composed of n-di- single objective function as following:
mensional DVs, x = [x 1 , x 2 , ..x n ] and l-dimen-
sional DPs, z = [z 1 , z 2 , .., z l ] . The lower and f f
minimize: (1 ) + , 0 1
upper bounds of the ith UBB type DV or DP, uiI f* f *
(3)
are denoted by uil and uiu , respectively. In interval
mathematics uiI is expressed in Box 1 (Moore,
1979). where, is a weighting factor in the above bi-
In the above, ui is the nominal value of ui , and objective optimization problem, f * and f * are
ui denotes the maximum variation of ui from the optimal solutions at two ideal situations ob-
its nominal value, termed as dispersion. If a tained for =1.0 and 0.0, respectively. The
practical estimate of the nominal value is avail- maximum robustness will be achieved for =0.0,
able, it can be directly assigned to ui . In absence and = 1.0 indicates optimization without any
of that, ui is usually taken as, (uil + uiu ) / 2 . Then, consideration for robustness.
Box 1.
uiI = [uil , uiu ] = [ui ui , ui + ui ] = [ui , ui ]+ [ ui , ui ] (1)
109
The constraint functions are exactly satisfied to uncertainty in the DVs and DPs can be esti-
in a DDO. However, these are expected to vary mated through first-order Taylor series expansion
due to the presence of uncertainty in the DVs and of the constraint function (Du, & Chen, 2000) as:
DPs. As a consequence, the final design obtained
by the DDO approach may become infeasible in N g j
presence of uncertainty and the constraint func- g j = ui (4)
i =1 ui
tions are required to be revised to include the
effect of their variations due to uncertainty. The
robustness of a constraint is the feasibility of the To further enhance the quality of robustness
constraint that needs to be guaranteed for the of the constraint, it is multiplied by a penalty
considered uncertainty ranges of the DVs and factor, k j and the constraint function can be ex-
DPs. Further details of this can be found in the pressed as (Lee, & Park, 2001):
works of Lee, & Park (2001); Wang, Peng, Hu,
& Cao (2009). When the DVs and DPs are char- g j (u) = g j + k j g j 0 (5)
acterized by random variables with known prob-
ability density function (pdf), the probabilistic It may be noted that a direct relationship to
feasibility of the constraint can be approximated ensure a target reliability level can not be achieved
ensuring a target reliability level for the considered in case of UBB type DVs and DPs and the value
constraint (Zang, Friswell, & Mottershead, 2005). of k j is introduced in equivalence to the probabi-
A general probabilistic feasibility formulation for
listic feasibility formulation, thereby indirectly
jth constraint can be expressed as:
ensuring safety requirement of a design. The
P [g j (u) 0] Poj , j = 1, ...., m , where, Poj is
selection of k j is somewhat ad hoc. Obviously,
the probability one desires to satisfy for the jth
the larger value of k j means one is more conser-
constraint feasibility. Assuming g j (u) to be nor-
vative to enhance the feasibility of the associated
mally distributed, this probabilistic feasibility of
constraint.
the constraint can be approximated as
Finally, the conventional RDO formulation
g j (u) + k j g j 0 , where, g j (u) and g j are the
is expressed by combining Eqs. (3) and (5) as:
mean and standard deviation of g j (u) . The pen-
alty factor, kj is used to enhance the feasibility of (1 )
f
+
f
0 1
minimize: ,
g j (u) and can be obtained from, kj= -1 (Poj ) , f* f *
such that: g j + k j g j 0 , j = 1, 2, ......, m.
where -1 (.) is the inverse of the cumulative
(6)
density function of a standard normal distribution.
For example, to ensure a reliability level of Poj
The conventional RDO approach as presented
=99.87%, one should take kj=3.0. above and also by the proposed RDO formulation
However, when the DVs and DPs are of UBB to be presented in the later part of the chapter are
type, above probabilistic feasibility formulation based on linear perturbation based approximation
for the constraints cannot be adopted. In order to of response functions about the mean values of the
enhance the feasibility of the jth constraint, an UBB parameters. The accuracy and efficiency of
additional quantity ( g j ) is introduced to con- such linear perturbation based approach are well
sider the effect of uncertainty in the DVs and DPs. documented in stochastic finite element literatures
The maximum variation ( g j ) of the jth constraint (Vanmarcke et al., 1986; Ghanem, & Spanos, 1990;
with respect to its nominal value, g j = g j (u) due Kleiber, & Hien, 1992). The study of accuracy
110
of perturbation based approach for evaluation of is the vector comprising of displacements of the
nominal response and its dispersion in this regard MDOF system due to ground motion ug (t ) at
may be found in Chen, Song, & Chen, (2007). base of the system. {L} is the influence coefficient
It has been numerically shown that the error in matrix. The displacement of the system subjected
estimation by linear perturbation approximation
to ground motion, ug (t ) = ug ( )e i t can be as-
goes up as the amplitude of uncertainty of the
interval variables increases. Thus, the approach s u m e d a s {Y (t )} = {H Y ( )e i t } , w h e r e
will be applicable for systems having small de- {H ( )} is the complex frequency response func-
Y
gree of uncertainty so that the linear perturbation tion (FRF) vector. Using this, the equation of
based analysis will be satisfactory. An extension motion can be expressed in the frequency domain
to the linear perturbation analysis is to use higher as following:
order perturbation method. Stochastic simulation
approach may be also applied for larger ampli- ([K] [M] + i [C]) {H ()} = [M]{L} u () i.e.
2 D ( ) {H Y ( )} = {F ( )} .
Y g
tude of uncertainty (Datta, 2010). However, the (8)
stochastic simulation will require assumptions
of pdf functions. One may choose conservative In the above, D ( ) is the dynamic stiffness
uniform distribution for this purpose. This aspect
matrix, {F ( )} is the forcing vector and ug ( )
needs further study and it is beyond the scope of
the present chapter. is the Fourier amplitude of ground motion. In Eq.
(8), {H Y ( )} is a function of u and can be ex-
Stochastic Structural plicitly re-written as,
Optimization (SSO)
{F (, u)}
1
D (, u) {H Y (, u)} = {F (, u)} or {H Y (, u)} = D (, u)

As already discussed, the SSO procedure under (9)
random earthquake load involves solution of a
nonlinear optimization problem involving sto- For a linear dynamic system, the power spec-
chastic performance measure. The constraint in tral density (PSD) function SYY (, u) of any
a generic form can be stated as, the probability response variable Y(t) can be readily obtained as
of not to exceed a given threshold of a stochastic (Lutes, & Sarkani, 1997; Datta, 2010),
response measure in a given time period, must be
greater than some prescribed minimum value. The SYY (, u) = {H Y (, u)} Su u ( ) {H*Y (, u)}
T
related formulation of stochastic dynamic analysis g g
to obtain the constraint of the related optimization (10)

problem is briefly presented here.
The dynamic equilibrium equation of a linear Where, {H*Y (, u)} is the complex conjugate
multi degree of freedom (MDOF) system under of {H Y (, u)} and Sug ug ( ) is the known PSD
seismic excitation can be written as,
function of stochastic ground motion.
The records of the ground motion at site are
[ M ] {Y (t )} + [C] {Y (t )} + [ K] {Y (t )} = [ M ]{L} ug (t ) necessary for realistic seismic reliability analy-
(7) sis. However, in scarcity of sufficient data for
statistical descriptions, various statistical models
Where, [ M ] , [C ] and [ K ] are the global mass, are developed to describe the stochastic ground
damping and stiffness matrix, respectively. Y (t ) motion process. The simplest such stochastic
111
model is the well-known stationary white noise

j
process, whose correlation function is Delta Dirac j (u) = SYY (, u)d ,

and associated PSD function is constant at each
frequency. However, it cannot sufficiently describe Y (u) = o (u) and Y (u) = 2 (u)
the spectral behaviour of many real stochastic (13)
dynamic loads. To improve the spectral charac-
terization of ground motion, a second order filter The reliability of the structure based on the
is introduced to colour the white noise, known as first passage failure criterion for double barrier
the Kanai-Tajimi model (Tajimi, 1960). This model problem can be obtained as,
has been widely applied in the random vibration
analysis of structures as it provides a simple way T
to describe the ground motion characterized by r (T ) = r0 exp h(t )dt (14)
0
a single dominant frequency. However, actual
earthquake recorded data show a non-stationary
nature both in the amplitude and frequency
contents and a more generalized non-stationary where, r0 is the survival probability at time, t=0
model obtained by enveloping the stationary input and h(t ) is the hazard function and T is the dura-
stochastic process should be used for more refined tion of the ground motion. Following the Poissons
analysis. The Kanai-Tajimi PSD function used in assumption of rare and independent threshold
the present study is represented by, crossings events, h(t) can be replaced with un-
conditional threshold crossing rate, +Y which can
(4g2 g2 2 + g4 ) be expressed as:
Sug ug ( ) = S 0 (11)
( 2 g2 )2 + 4g2 g2 2
1 Y 1 2
Y+ ( ) = ( ) (15)
2 Y 2 Y
where, g and g denote the natural frequency and
damping ratio of the soil layer, respectively. The In the above, is the first time bi-lateral dis-
PSD of the white noise process at bed rock, S 0 can placement crossing barrier. If the failure is due to
be related to the peak acceleration, ( xg ,max ) by, double symmetric threshold crossing (as consid-
ered herein), the threshold crossing rate can be
0.0707 g xg2,max given by,
S0 = (12)
g (1 + 4g2 )
( ) = 2Y+ ( ) (16)
The spectral moments ( j ) and the associated The applicability of the above is limited to
root mean square (RMS) values ( Y and Y ) of the assumptions that the structure can recover
immediately after suffering failure (threshold
the responses useful for reliability evaluation can
crossing) and such failures arrive independently,
be evaluated as,
that is, they constitute a Poisson process. More
112
Box 2.
log (1 / Rmin ) 1 Y 1 2 log (1/Rmin )

g (u ) = ( , u ) = exp ( / Y ) 0 (19)
T Y 2 T
details may be found in Lutes, & Sarkani, (1997); stochastic dynamic response of structure under
Datta, (2010). earthquake load modelled as stationary stochastic
The limit state corresponds to the first excur- process. However, application of the proposed
sion of a structural response Y(u,t ) can be ex- RDO presented in this chapter is not restricted to
pressed as, G [Y(u, t)] > 0, where G is a function. such stochastic random process only and extension
For a stationary stochastic process, the reliability to non-stationary earthquake model will be straight
of structure can be written as, forward. However, this will involve time depen-
dent response statistics evaluations and subse-
R(t, u) = P {G[Y(u, t )] 0 | t <T }. (17) quently to deal with time dependent performance
function in the optimization procedure.
The optimum DVs must satisfy one or more
probabilistic constraint(s) consisting of limiting
Stochastic Dynamic
the failure probability for a given value of reli-
Response Approximation
ability, Rmin . For a specified threshold barrier (
It can be noted that in a typical SSO procedure,
), the stochastic constraint becomes,
the safety measures are required to be evaluated
several times to obtain an optimal design. The
{
R(, T, u) = P G[Y(u, t ) 0] t <T } = exp ( (, u)T ) R
min
evaluations of safety measures for every change
(18)
of the DVs require several evaluations of the
dynamic responses of a structural system. The
Using Eqs. (15) and (16), the constraint can
dynamic responses as described by Eq. (9) are
be finally expressed in Box 2.
implicit function of u and normally obtained by
Finally, the SSO problem under stationary
numerical methods like the finite element proce-
earthquake model can be expressed in Box 3.
dure. Furthermore, the formulations described in
In the above, f (u) is the objective function
the previous section assume that all the DVs and
of the optimization problem. Usually the weight DPs are deterministic. If the effects of uncer-
of the structure or some important stochastic re-
tainty are considered in the analysis, D (, u)
sponse measure is considered as the objective
and thereby the FRF vector will also involve
function. The SSO presented here is based on the
uncertainty. Therefore, the analysis will require
Box 3.
find {x} to minimize: f (u)

log(1 / Rmin ) 1 Y 1 2 log(1 / Rmin ) (20)
such that : g(u) = (, u) = exp ( / Y ) 0.
T Y 2 T
113
complex sensitivity analysis of the stochastic the variable xij, which denotes the ith observation
dynamic system (Chaudhuri, & Chakraborty, of the jth variable xj obtained by a suitable design
2004) as the RDO problem requires the gradients of experiments (DOE), following matrix form can
of the objective function and associated con- express the relationship between the responses (
straints. For systems of practical interest, re- y ) and the variables ( x ),
peated evaluations of dynamic responses and their
sensitivities will be extremely time-consuming. y = xa + y (21)
Thus, the use of direct optimization procedure is
not suitable to perform RDO under stochastic
In the above, x , y , a and y are the design
excitation. In the present study, an alternative to
the direct optimization methods is proposed. An matrices containing the input data obtained by
RSM based approximation is adopted judicious- the DOE, the response vector, the unknown coef-
ly to approximate the dynamic response required ficient vector and the error vector, respectively.
to obtain the stochastic constraint of the related The least-squares function Ly (x ) can be defined
SSO problem. The MLSM, a local approximation as the sum of the weighted errors,
approach is observed to be elegant in this regard q
(Kim, Wang, & Choi, 2005) is adopted in the Ly (x )= wi i2 = T W(x ) = (y xa)T W(x )(y xa)
present study. The essential concept of the MLSM i =1
(22)
is briefly described here in order to outline the
procedure. The further details about this may be
found elsewhere (Kim, Wang, & Choi, 2005;
where, W(x ) is the diagonal matrix of the weight
Bhattacharjya, & Chakraborty, 2009; Kang, Koh,
function. It can be obtained by utilizing the weight-
& Choo, 2010).
ing function such as constant, linear, quadratic,
The MLSM based RSM is a weighted LSM
higher order polynomials, exponential functions,
that has varying weight functions with respect to
etc. (Kim, Wang, & Choi, 2005), as described in
the position of approximation. The weight associ-
Box 3.
ated with a particular sampling point xi decays
RI is the approximate radius of the sphere of
as the prediction point x moves away from xi.
The weight function is defined around the predic- influence, chosen as twice the distance between
tion point x and its magnitude changes with x. If the centre point and extreme most experimental
yi is the ith response (i=1,2,.,q) with respect to point. The value of RI is chosen to secure sufficient
number of neighbouring experimental points to
Box 3.
w(x -x i ) = w(d )
Constant: 1. 0

Linear: 1- d / RI

(if d / RI 1.0), Quadratic: 1- (d / RI )2
= th
4 order polynomial: 1- 6(d / RI )2 + 8(d / RI )3 -3(d / RI )4

Exponential: exp (-d / RI )

(if d / RI 1.0), 0.0
(23)
114
avoid singularity. An exponential form of weight Theoretical Formulation

function has been adopted in the present nu-
merical study. Eventually, a weight matrix W(x) It is well-known that the sensitivity information
can be constructed by using the weighting func- are useful to a designer as it provides a measure
tion in the diagonal terms as: of performance deviations in a design associated
with an increase or decrease of the respective vari-
w(x -x 1 ) 0 0 0 ables. To reduce the number of significant random

0 w(x -x 2 ) .. 0 variables before in reliability evaluation process

W(x )= .. .. .. .. the use of importance measure as given below is
.. quite common (Bjerager, & Krenk, 1989; Haldar,
.. .. ..
0 & Mahadevan, 2000; Gupta, & Manohar, 2004):
0 .. w(x -x n )
(24) G
2 N 2
I i = G

u (26)
By minimizing the least-squares estimators i k =1 uk
Ly (x ) , the coefficients a(x)can be obtained by
the matrix operation as below,
where, G is the failure surface defining the safe
1
a(x ) = xT W(x )x x W(x )y
T
(25) and unsafe regions and N is the total number of
random variables in a generic structural reliability
analysis problem. The applicability of the above
It is important to note here that the coefficients form of importance measure was studied by
a(x ) are the function of the location x , where the Gupta, & Manohar (2004) through Monte Carlo
approximation is sought. Thus, the procedure to Simulation (MCS) study. Based on the entries of
calculate a(x ) is a local approximation and mov- Iis, the uncertain variables can be grouped into
ing processes performs a global approximation important and unimportant variables. The un-
throughout the whole design domain. certain variables for which the failure surface is
more sensitive are identified as dominant variables.
The Proposed RDO APPROACH This is hinged on the fact that all the gradients
are not equally important in the expression of
As mentioned earlier, the present chapter deals the failure surface in a typical reliability analysis
with an efficient RDO procedure for structures problem. This intuitively indicates that all the
subjected to stochastic earthquake load and char- gradients do not have equal importance in the
acterized by UBB type DVs and DPs. A heuristic expression of the dispersion of the performance
algorithm is proposed here which allows the use function as defined by Eq. (2). Hence, it is ap-
of importance factor obtained using the respec- parent that the importance measure should also
tive sensitivity information of the performance play a role to indicate the measure of robustness
function and constraints. In the following sub- of the performance.
sections, the related theoretical developments, It can be readily realized from Eq. (2) that
implementation of the algorithm and numerical the dispersion f depends on two factors: (i)
study are presented. the amplitude of uncertainty in the DVs and DPs
represented by the corresponding dispersion,
ui ' s and (ii) the gradients of the performance
115
function, f / ui ' s . If the amplitudes of un- Integrating the information presented so far, the
certainty, i.e. the dispersions of DVs and DPs proposed RDO scheme can be finally represented
do not change, the dispersion of the performance as an equivalent DDO problem as following:
function f will solely depend on the gradients
f / ui ' s . It is obvious that a change in the f f
minimize: (1 ) + w*
dispersion fi (due to the variation of the ith DV f* fw
such that: g j + k j g jw 0, j = 1, 2, ......, m,
or DP) will be more than fj (due to the varia-
tion of the jth DV or DP) if the associated im- (29)
portance measure I i > I j or vice versa. Hence,
it seems to be more logical to use the importance In the above, k j is as defined in Eq. (5). The
measure of the associated gradients as well in above nonlinear optimization problem can be
defining the measure of robustness. A new mea- solved by available optimization techniques. Once
sure of robustness is thus proposed in the present the optimum design point is obtained by the pro-
RDO study by redefining the performance dis- posed importance factor based RDO approach,
persion as following: the dispersion of the performance function can
be evaluated at the optimum design point using
f
2 N 2
Eq. (2). It may be noted that the importance fac-
f f
N
fw = I fi ui ,where, I fi = N
i =1 u i ui k =1 uk tors are directly incorporated in the optimization
(27) formulation and such factors are evaluated at
updated design point during each iteration cycle
In the above, fw is the new dispersion index of the optimization process. Doing so, more sen-
and I fi is the importance factor for the ith DV or sitive DVs and DPs are automatically got re-
DP. It is to be noted here that the importance fac- amplified yielding a better robust solution in
tor as defined above is multiplied by the total lesser computational time. Basically, the incor-
number (N) of the DVs and DPs. However, the poration of the importance factors changes the
summation of all such factors will be always equal feasible domain of the original optimization
to N, whatever is the individual value of the im- problem and captures the more flat zone of the
portance factor. This will keep the consistency of performance function. In this regard, it is worth
the definition of this index to measure the robust- mentioning that many researchers adopt a separate
ness with the usual dispersion index used in the sub-problem to find a search direction for a quick
conventional RDO approach. convergence towards the robust optima (Lee, &
The basic idea to improve the robustness of the Park, 2001; Wang et al., 2009). The search direc-
performance function using the new index utilizing tion should emphasize on the more sensitive
the importance factors proportional to the impor- variables for an efficient and quick convergence
tance of the gradients of the performance function to the robust optima. As the importance factor
can be readily extended to the constraints as well. based dispersion index includes this aspect of
A new dispersion index to achieve the robustness assigning more importance to more sensitive DVs
of jth constraint feasibility is defined as: and DPs, the proposed RDO procedure con-
verges to robust optima efficiently without requir-
N 2 ing any such separate sub-problem for direction
g j
2
N g j g
g jw = I g ui , where, I g = N j finding.
i =1 ui ji ji ui k =1 uk
(28)
116
Figure 1. Implementation of the proposed RDO
Implementation of the responses as per the DOE in step (i), and iii) finally
RDO Approach checking the convergence to obtain the robust
optimum solution. It can be noted from the flow
The implementation procedure of the proposed chart that during each update of the DVs in the
RDO for linear dynamic system is demonstrated optimization process the MLSM based response
through a flow chart in Figure 1. The procedure approximation is re-called and a new approxima-
is basically a three-stage interlinked procedure, tion function is formed. However, this is not the
viz.: i) stochastic dynamic analysis of the finite case for the LSM based RSM, where a single ap-
element model for selected set of input variables proximation function gets operated throughout the
as per the DOE to obtain the required stochastic optimization process. It is worth mentioning here
responses and subsequently the constraint of the that the choice of initial solution is an important
optimization problem, ii) evaluation of the con- issue in metamodel based optimization procedure.
straint function at the iteration point following For the conventional LSM based RSM approach,
the MLSM based RSM during each iteration of a poor initial guess not only increases the number
the optimization process utilizing the computed of iterations, but also warns the convergence of
117
the optimization problem. As the MLSM based Unless mentioned specifically Rmin , T and
approximation technique can capture the actual are taken as 0.99, 10 sec and 3Y ( Y is the
response even beyond the sampling zone, the maximum RMS displacement), respectively. The
convergence is faster than the conventional LSM MLSM based adaptive RSM technique as dis-
based techniques (Kim, Wang, & Choi, 2005). cussed previously is used to approximate the
constraint function. In the present numerical study,
Numerical Study the second-order polynomial without cross terms
is considered. The DOE is constructed consider-
A three storied concrete building frame (as shown ing the centre and axial points following the
in Figure 2) subjected to earthquake motion is Saturated Design (SD) method (Bucher, & Macke,
taken up to elucidate the effectiveness of the 2005). However, the points chosen are at the
proposed RDO approach. The frame structure is nominal value ( x i ) of the input variable (xi) and
idealized by fifteen two-nodded plane frame ele- at axial points xi = x i hi x i , where hi is a
ments having three degrees of freedom at each
positive integer. For each input variable six axial
node. Both the DVs and DPs are considered to
points (hi=1, 2, 6) are considered on each axis
be of UBB type and are described by their respec-
taking x i as 5% of x i to cover the different
tive dispersions, ui , representing the maximum
amplitude levels of uncertainty. Thus, for the
possible ranges of variations expressed in terms
present study, the number of required training
of the percentage of the corresponding nominal
points with respect to dimension of the input vec-
values. The necessary information with associ-
tor (N) is: (6x2N+1) i.e. the total number of
ated notations are summarized in Table 1. The
sampling points is sixty one with five input vari-
deterministic dimension L and h are taken as 6.0
ables. As more axial points are considered than
m and 4.0 m, respectively. The natural frequency
that required by the SD method, the design be-
( g ) and the damping ratio ( g ) of the soil layer
comes a redundant design (Bucher, & Macke,
are considered to be 18.85 rad/sec and 0.65, re- 2005). To study the effectiveness of the MLSM
spectively. The peak ground acceleration ( xg,max ) based adaptive RSM, the computed responses are
is taken as 0.2g, where g is the acceleration due compared with that obtained by the conventional
to gravity. LSM based RSM. The maximum RMS displace-
The objective function f(u) considered in the ment of the frame is shown for a wide range of
present study is the weight of the frame. Apart column size in Figure 3. The results obtained by
from the stochastic constraint, as discussed ear- the direct random vibration analysis are also shown
lier, a size constraint (g2) limiting the ratio between in the same figure for ease in comparison. It can
the depth and width of the beam members is also be readily observed from the figure that the pre-
considered in the SSO formulation. The SSO dicted responses by the MLSM based RSM better
formulation under earthquake load with deter- matches with the direct random vibration analysis
ministic DVs and DPs can be defined as: results. The error using the conventional LSM
based RSM drastically increases beyond the range
find x = { bc db bb } to minimize: f = (9hbc2 + 6Lbbdb )c
of data points which are used in the DOE to con-
2
1 Y 1 log (1 / Rmin )
such that: g1 (x, z) :
Y
exp
2 Y T
0 struct the metamodel. During the iteration stage

g 2 (x, z) : db 3bb 0 0.1 bc , bb , db 1.2 of any gradient-based optimization algorithm, the
(30) DVs may take values outside the sampling range
118
Figure 2. The building frame

The coefficient of determination ( R 2 ) is de-
fined as:
p 2 p
R 2 = (yi yi ) (y y )2
(32)
i =1 i =1 i i

where, yi is the mean value of the actual response.

The value of this coefficient close to one represents
a good metamodel for response approximation.
The average prediction error ( m ) is defined as,
yi yi
m = 100 (33)
yi
Table 2 shows the results of the statistical test

for both the LSM and MLSM based metamodels.
It can be observed that the lesser RMSE and m
depending on the specific nature of the optimiza- values and the higher R 2 value are attained by the
tion problem. This is a potential problem in it- MLSM based RSM compared to the LSM based
erative optimization process and warns the ap- RSM. This clearly indicates the accuracy of the
plicability of the LSM based RSM approach for MLSM based RSM over the LSM based RSM.
RDO of complex dynamic system under stochas- The CPU time required for complete generation
tic load. Furthermore, to study the suitability of of all yi by the direct MCS is about 2.5 hrs,
the proposed adaptive RSM, the statistical metrics whereas the CPU time needed for computation
as described below have been also computed of yi is only 11 minutes by the LSM based RSM
(Bouazizi, Ghanmi, & Bouhaddi, 2009): and 12 minutes by the MLSM based RSM. The
The Root Mean Square Error (RMSE) is de- MLSM based RSM needs more computational
fined as: time compare to the LSM based RSM due to the
fact that the MLSM based approach needs re-
2 peated evaluation of the response surface as it is
p
(yi yi )
RMSE = p
(31) required to be generated afresh for each updated
i =1
DV set. Whereas, the LSM based approach may
need more rigorous DOE compare to the MLSM
In the above, yi is the predicted response ob- based RSM to obtain a comparable accuracy
tained by the considered LSM or MLSM based level. It may be noted here that one needs to
metamodel and yi is the actual response obtained evaluate the structural responses analysis at all
by the direct MCS for ith sample. The sample these training points involving the solution of a
size, p is taken as 1000 for the present numerical system having number of unknowns in the order
investigation. of few thousands to million. This obviously in-
119
Table 1. The details of the uncertainty in the DVs and DPs
DPs Nominal values Dispersion
Density of concrete, c 2400 kg/m3 10%
Modulus of Elasticity 20 GPa 10%

of concrete, Ec
DVs
Size of square column, bc To be optimized 10%
Depth of beam, db To be optimized 10%
Width of beam, bb To be optimized 10%
Table 2. Performance of the LSM and MLSM based RSM
RMSE R2 m
LSM based RSM 7.91% 0.97 2.5%
MLSM based RSM 4.72% 0.99 0.07%
Figure 3. The comparison of RMS displacement by the LSM and MLSM based RSM
120
creases the time requirement by the LSM based the penalty factor, kj and the bi-objective weight
approach. Thus the computational involvement factor, are considered to be 1.0 and 0.5, respec-
with regard to generation of training data point tively. It can be noted that the optimization results
will be substantially high whereas, there is a obtained by the proposed approach indicate an
higher computational demand with regard to re- improved robustness compared to the conven-
peated evaluation of response surface. Of course, tional RDO method.
it needs further study to comprehend which one The optimal weight and its dispersion are plot-
will be more efficient. ted in the Figures 6 and 7, respectively, with respect
Now, the optimization task is performed by to the upcrossing level, for different values of
the sequential quadratic programming (SQP) us- kj. The dispersion of the inputs and are set as
ing built-in MATLAB routine. The weight of the 10% and 0.5, respectively. It can be noted from
frame is optimized by the proposed importance these figures that the same nature of variations
factor based RDO approach as described by Eq. and improvement are observed for all the values
(29) and the conventional RDO approach as de- of kj considered in the study. In general, it has
scribed by Eq. (6) considering uncertainty in both been observed that the effect on the robustness is
the DVs and DPs. The results are studied through more prominent for smaller values of . This is
Figures 4, 5, 6, 7 and 8. The optimal weight of obvious as the smaller value of represents more
the frame versus the dispersion of the DVs and stringent failure criterion and makes the constraints
DPs are plotted in Figure 4. For comparison, the more critical. Thus, the effect of the importance
results obtained by the conventional RDO ap- of the sensitivity derivatives introduced by the
proach are also shown in the same figure. The proposed RDO approach becomes more.
variations of the optimal weight of the frame One of the important tasks in an RDO proce-
(indicates the measure of robustness) versus the dure like any other multi-objective optimization
dispersion of the input DVs and DPs, as obtained problem is to obtain the Pareto front (Deb, 2001).
by the proposed and conventional RDO ap- It is generally observed that there is a trade-off
proaches are depicted in Figure 5. The value of between the objective value of a design and its
Figure 4. The optimal weight of the frame with increasing range of uncertainty
121
Figure 5. The variation of optimal weight of the beam with increasing range of uncertainty
Figure 6. The optimal weight of the frame with varying upcrossing level
robustness. If one desires more robustness, the RDO approaches are plotted in Figure 8. The
design will be further away from its ideal optimal uncertainty ranges of the DVs and DPs are taken
value. The situation can be studied further in terms as per Table 1 and kj is considered to be 1.0 to
of Pareto-front. The Pareto-front is one where any develop this figure. The Pareto front is determined
improvement in one objective can only achieve by evaluating the optimal solutions (the objective
through worsening of at least one other objective. function and its associated dispersion) for differ-
If one chooses a design that is not Pareto-optimal, ent settings of . Maximum robust solution is
one essentially forfeits improvements that would obtained when is1.0. The designer puts maxi-
otherwise entail no compromise. The Pareto fronts mum emphasis on optimal objective, not on its
obtained by the proposed and the conventional robustness, if is zero. Thus, the designer can
122
Figure 7. The dispersion of optimal weight of the frame with varying upcrossing level
Figure 8. Comparison of Pareto front as obtained by the proposed and the conventional RDO approaches
tune the robustness of a solution through suitable RDO approach is also observed to be lesser than
choice of this parameter. It can be clearly observed that obtained by the conventional RDO approach.
from the figure that for a specific level of disper- To quantify the computational involvement of the
sion f, if the desired objective is to minimize the proposed RDO approach, the CPU time required
weight of the frame, the proposed RDO approach to generate the complete Pareto front by both the
always yields lesser weight compared to the con- approaches are computed. It is observed that the
ventional RDO approach. Furthermore, for a proposed RDO approach needs 25 minutes
prescribed weight of the frame, the dispersion of whereas the conventional RDO approach needs
the optimal weight f obtained by the proposed 40 minutes for this purpose. This clearly indicates
123
the efficiency of the proposed RDO approach. In is possible to achieve compare to the conventional
this regard, it is worth mentioning here that one RDO approach is problem dependent. It will spe-
may fail to identify the complete Pareto front by cifically depend on the nature of the performance
the present WSM approach for non-convex Pa- function and the entire solution domain, i.e., the
reto front. This is a potential drawback of WSM nature of the variation of the objective function
and to overcome this limitation Evolutionary and constraints. This of course needs further
algorithms like Non-domination Sorting Genetic study. The results shown in the numerical example
Algorithm-II (Deb, 2001) can be applied. are for some specific values of the bi-objective
weight factor and penalty factor. However, the
general trend is observed to be similar for other
SUMMARY AND CONCLUSION values. The proposed RDO approach is generic
in nature and can be applied to optimum RDO of
An efficient RDO procedure for structural system structures under other stochastic dynamic load,
subjected to stochastic earthquake load and char- like wind, wave, etc.
acterized by UBB type DVs and DPs is presented.
The associated stochastic constraint is derived
by imposing a limit on the failure probability. To REFERENCES
exclude the complexity of repeated evaluations
of random dynamic responses and their sensi- Al-Widyan, K., & Angeles, J. (2005). Model-
tivities (required for RDO solution), the implicit based formulation of robust design. Journal of
constraint function is approximated through a Applied Mechanics Transaction, 127(3), 388396.
potentially superior MLSM based adaptive RSM doi:10.1115/1.1829728
method. The proposed RDO approach improves Ben-Tal, A., & Nemirovski, A. (2002). Robust
the robustness of the performance function using optimization-methodology and applications.
a new dispersion index which utilizes the weight Mathematical Programming, 92(3), 453480.
factors proportional to the importance of the gra- doi:10.1007/s101070100286
dients of the performance function with respect
to uncertain DVs and DPs. The improvement Bertsimas, D., & Sim, M. (2006). Tractable ap-
is achieved in terms of more robustness in the proximations to robust conic optimization prob-
optimization by allowing more importance to the lems. Mathematical Programming, 107(1-2),
uncertain variables which influence the variations 536. doi:10.1007/s10107-005-0677-1
of the performance function and the constraint
Beyer, H., & Sendhoff, B. (2007). Robust opti-
violations. Numerical results indicate that the
mization -A comprehensive survey. Computer
trends and variations of the optimization results
Methods in Applied Mechanics and Engineer-
are in conformity with the well-known two criteria
ing, 196(33-34), 31903218. doi:10.1016/j.
RDO results. It can be noted that the economy is
cma.2007.03.003
not affected in achieving better robustness, as the
proposed RDO approach yields more reduction of Bhattacharjya, S., & Chakraborty, S. (2009). Ro-
weight for a specified dispersion of the weight. bust optimization of linear dynamic system with
This implies that the importance factor based RDO random parameters under stochastic earthquake
approach captures the optimal design point on a excitation. International Journal of Reliability
comparatively flat region than the optimal point Quality and Safety Engineering, 16(3), 261279.
captured by the conventional RDO approach. doi:10.1142/S0218539309003393
However, how much improvement in robustness
124
Bjerager, P., & Krenk, S. (1989). Para- Du, X., & Chen, W. (2000). Towards a better un-
metric sensitivity in first order reliability derstanding of modelling feasibility robustness in
theory. Journal of Engineering Mechanics, engineering design. ASME Journal of Mechanical
115(7), 15771582. doi:10.1061/(ASCE)0733- Design, 122(4), 385394. doi:10.1115/1.1290247
9399(1989)115:7(1577)
Du, X., Sudijianto, A., & Chen, W. (2004). An
Bouazizi, M.-L., Ghanmi, S., & Bouhaddi, N. interval framework for optimization under un-
(2009). Multi-objective optimization in dynam- certainty using inverse reliability strategy. ASME
ics of the structures with nonlinear behavior: Journal of Mechanical Design, 126(4), 562570.
Contributions of the metamodels. Finite Ele- doi:10.1115/1.1759358
ments in Analysis and Design, 45(10), 612623.
Ghanem, R., & Spanos, P. D. (1990). Stochastic
doi:10.1016/j.finel.2009.03.007
finite element analysis: A spectral approach.
Bucher, C., & Macke, M. (2005). Response sur- Berlin, Germany: Springer.
faces for reliability assessment. In Nikolaidis, E.,
Guedri, M., Ghanmi, S., Majed, R., & Bouhaddi,
Ghiocel, D. M., & Singhal, S. (Eds.), Engineering
N. (2009). Robust tools for prediction of vari-
design reliability handbook. Boca Raton, FL:
ability and optimization in structural dynamics.
CRC Press.
Mechanical Systems and Signal Processing, 23(4),
Chaudhuri, A., & Chakraborty, S. (2004). Sen- 11231133. doi:10.1016/j.ymssp.2008.08.013
sitivity evaluation in seismic reliability analysis
Gunawan, S., & Azarm, S. (2005). Multi-objective
of structures. Computer Methods in Applied
robust optimization using a sensitivity region
Mechanics and Engineering, 193(1-2), 5968.
concept. Structural and Multidisciplenary Op-
doi:10.1016/j.cma.2003.09.007
timization, 29(1), 5060. doi:10.1007/s00158-
Chaudhuri, A., & Chakraborty, S. (2006). Re- 004-0450-8
liability of linear structures with parameter
Guo, X., Bai, W., Zhang, W., & Gao, X. (2009).
uncertainty under nonstationary earthquake.
Confidence structural robust design and opti-
Structural Safety, 28(3), 231246. doi:10.1016/j.
mization under stiffness and load uncertainties.
strusafe.2005.07.001
Chen, S. H., Song, M., & Chen, Y. D. (2007). gineering, 198(41-44), 33783399. doi:10.1016/j.
Robustness analysis of vibration control struc- cma.2009.06.018
tures with uncertain parameters using interval
Gupta, S., & Manohar, C. S. (2004). An improved
algorithm. Structural Safety, 29(2), 94111.
response surface method for the determination
doi:10.1016/j.strusafe.2006.03.001
of failure probability and importance measures.
Datta, T. K. (2010). Seismic analysis of Structural Safety, 26(2), 123139. doi:10.1016/
structures. Singapore: John Wiley & Sons. S0167-4730(03)00021-3
doi:10.1002/9780470824634
Haldar, A., & Mahadevan, S. (2000). Reliability
Deb, K. (2001). Multi-objective optimization assessment using stochastic finite element analy-
using evolutionary algorithms. New York, NY: sis. USA: John Wiley and Sons.
John Wiley & sons.
Huang, B., & Du, X. (2007). Analytical robust-
ness assessment for robust design. Structural and
Multidisciplinary Optimization, 34(2), 123127.
doi:10.1007/s00158-006-0068-0
125
Hwang, K.-H., Lee, K.-W., & Park, G.-J. (2001). Kang, S.-C., Koh, H.-M., & Choo, J. F. (2010).
Robust optimization of an automobile rearview An efficient response surface method using mov-
mirror for vibration reduction. Structural and ing least squares approximation for structural
Multidisciplinary Optimization, 21(4), 300308. reliability analysis. Probabilistic Engineering
doi:10.1007/s001580100107 Mechanics, 25(4), 365371. doi:10.1016/j.pro-
bengmech.2010.04.002
Jensen, H. (2002). Reliability-based optimization
of uncertain systems in structural dynamic. AIAA Kim, C., Wang, S., & Choi, K. K. (2005). Efficient
Journal, 40(4), 731738. doi:10.2514/2.1705 response surface modelling by using moving least-
squares method and sensitivity. AIAA Journal,
Jensen, H. (2005). Structural optimization of linear
43(1), 24042411. doi:10.2514/1.12366
dynamical systems under stochastic excitation: A
moving reliability database approach. Computer Kleiber, M., & Hien, T. D. (1992). The stochas-
Methods in Applied Mechanics and Engineering, tic finite element method. Chichester, UK: John
194(12-16), 1757-1778. Wiley and Sons.
Jensen, H. (2006). Structural optimization of non- Kurdi, M. H., Haftka, R. T., Schmitz, T. L., &
linear systems under stochastic excitation. Proba- Mann, B. P. (2008). A robust semi-analytical
bilistic Engineering Mechanics, 21(4), 397409. method for calculating the response sensitivity
doi:10.1016/j.probengmech.2006.02.002 of a time delay system. Journal of Vibration and
Acoustics, 130(6), 16. doi:10.1115/1.2981093
Jensen, H. A., Valdebenito, M. A., & Schuller, G.
I. (2008). An efficient reliability-based optimiza- Lagaros, N. D., Garavelas, A. T., & Papadrakakis,
tion schme for uncertain linear systems subject to M. (2008). Innovative seismic design optimization
general Gaussian excitation. Computer Methods with reliability constraints. Computer Methods
in Applied Mechanics and Engineering, 198(12- in Applied Mechanics and Engineering, 198(1),
16), 72-87. 2841. doi:10.1016/j.cma.2007.12.025
Jin, R., Chen, W., & Simpson, T. (2001). Com- Lee, K., & Park, G. (2001). Robust optimization
parative studies of metamodelling techniques considering tolerances of design variable. Com-
under multiple modeling criteria. Structural puters & Structures, 79(1), 7786. doi:10.1016/
and Multidisciplinary Optimization, 23(1), 113. S0045-7949(00)00117-6
doi:10.1007/s00158-001-0160-4
Lewis, A. S. (2002). Robust regularization. Techni-
Jurecka, F., Ganser, M., & Bletzinger, K. U. (2007). cal Report. Columbia: Simon Fraser University.
Update scheme for sequential spatial correlation
Lutes, L. D., & Sarkani, S. (1997). Stochastic
approximations in robust design optimization.
analysis of structural and mechanical vibrations.
Computers & Structures, 85(10), 606614.
Upper Saddle River, NJ: Prentice Hall.
doi:10.1016/j.compstruc.2006.08.075
Marano, G. C., Greco, R., & Sgobba, S. (2010).
Kang, B., Park, G. J., & Arora, J. S. (2006). A
A comparison between different robust optimal
review of optimization of structures subjected to
design approaches: Application to tuned mass
transient loads. Structural and Multidisciplinary
dampers. Probabilistic Engineering Mechan-
Optimization, 31(2), 8195. doi:10.1007/s00158-
ics, 25(1), 108118. doi:10.1016/j.probeng-
005-0575-4
mech.2009.08.004
126
Marano, G. C., Sgobba, S., Greco, R., & Mezzina, Son, Y. K., & Savage, G. J. (2007). Optimal proba-
M. (2008). Robust optimum design of tuned mass bilistic design of the dynamic performance of a vi-
dampers devices in random vibrations mitiga- bration absorber. Journal of Sound and Vibration,
tion. Journal of Sound and Vibration, 313(3-5), 307(1-2), 2037. doi:10.1016/j.jsv.2007.06.032
472492. doi:10.1016/j.jsv.2007.12.020
Taflanidis, A. A. (2010). Reliability-based optimal
Marano, G. C., Trentadue, F., & Greco, R. design of linear dynamical systems under stochas-
(2006). Optimal design criteria for elastic tic stationary excitation and model. Engineering
structures subject to random dynamic loads. Structures, 32(5), 14461458. doi:10.1016/j.
Engineering Optimization, 38(7), 853871. engstruct.2010.01.023
doi:10.1080/03052150600913028
Taflanidis, A. A., & Beck, B. L. (2008). An efficient
Moore, R. E. (1979). Methods and applications framework for optimal robust stochastic system
of interval analysis. Philadelphia, PA: SIAM design using stochastic simulation. Computer
Bookmart, PA. Methods in Applied Mechanics and Engineering,
198(1), 88101. doi:10.1016/j.cma.2008.03.029
Mohsine, A., Kharmanda, G., & El-Hami, A.
(2005). Reliability-based design optimization Tajimi, H. (1960). A statistical method of determin-
study using normal and lognormal distributions ing the maximum response of a building during
with applications to dynamic structures. In G. earthquake. In G. W. Housner (Ed.), Proceedings
Augusti, G. I. Schuller, & M. Ciampoli (Eds.), of the Second World Conference on Earthquake
Proceedings of the 9th International Conference Engineering, 11-18 July 1960 Tokyo and Kyoto,
on Structural Safety and Reliability, 19-23 June Japan, (pp. 781-797). Tehran, Iran: University
2005 Rome, Italy (pp. 3773-3779). Rotterdam, press.
The Netherlands: Millpress.
Vanmarcke, E. H., Shinozuka, M., Nakagiri, S.,
Myers, R. H., & Montogmery, D. C. (1995). Re- Schuller, G. I., & Grigoriu, M. (1986). Random
sponse surface methodology: Process and product fields and stochastic finite elements. Struc-
optimization using designed experiments. USA: tural Safety, 3(3), 143166. doi:10.1016/0167-
John Wiley and Sons. 4730(86)90002-0
Nigam, N. C. (1972). Structural optimization in Wang, W. M., Peng, Y. H., Hu, J., & Cao, Z. M.
random vibration environment. AIAA Journal, (2009). Collaborative robust optimization under
10(4), 551553. doi:10.2514/3.50151 uncertainty based on generalized dynamic con-
straints network. Structural and Multidisciplinary
Park, G. J., Lee, T. H., Lee, K., & Hwang, K.
Optimization, 38(2), 159170. doi:10.1007/
H. (2006). Robust design: An overview. AIAA
s00158-008-0271-2
Journal, 44(1), 181191. doi:10.2514/1.13639
Schuller, G., & Jensen, H. (2008). Computational
methods in optimization considering uncertain-
ties- An overview. Computer Methods in Ap-
plied Mechanics and Engineering, 198(1), 213.
doi:10.1016/j.cma.2008.05.004
127
128
Chapter 6
Damage Assessment of
Inelastic Structures under
Simulated Critical Earthquakes
Abbas Moustafa
Minia University, Egypt
ABSTRACT
Damage of structures can be significantly reduced through robust prediction of possible future earth-
quakes that can occur during the life-time of the structure and through accurate modeling of the nonlinear
behavior of the structure under seismic loads. Modern seismic codes specify natural records and arti-
ficially generated ground accelerations as input to the nonlinear time-history analysis of the structure.
The advantage of using natural records is the inclusion of all important characteristics of the ground
motion (fault properties, path effects and local soil condition) in the design input. This option requires
selecting and scaling a set of proper accelerograms from the available records. However, the site under
consideration may have limited or scarce earthquake data. In such case, numerically simulated ground
motions can be employed as input to the dynamic analysis of the structure. This chapter deals with
the damage assessment of inelastic structures under numerically simulated critical earthquakes using
nonlinear optimization, inelastic time-history analysis, and damage indices.
1. INTRODUCTION tion techniques. The critical excitation method

relies on the high uncertainty associated with
Damage indices describe the state of the structural the occurrence of the earthquake phenomenon
damage and correlate well with actual damage and on the safety requirements of important and
displayed during earthquakes. The critical ground lifeline structures. The earthquake input is taken to
motion for a give structure is estimated by solving maximize the damage index of the structure while
an inverse nonlinear dynamic problem in time satisfying predefined constraints that are quanti-
domain using constrained nonlinear optimiza- fied from the earthquake data available at the site.
Numerical illustrations for damage assessment of
DOI: 10.4018/978-1-4666-1640-0.ch006
Damage Assessment of Inelastic Structures
one-storey and two-storey plane frame structures the structure under possible future earthquakes
under possible future earthquakes are presented. while maintaining optimal use of the construc-
Earthquakes continue to claim thousands of tion material. The design objectives in current
lives and to damage structures every year (Comar- seismic building codes are to ensure life safety
tin et al, 2004). Each earthquake brings out new and to prevent damage of the structure in minor
surprises and lessons with it. In fact, the unexpected and moderate frequent earthquakes, and to control
loss of lives and the severe damage of infrastruc- local and global damage (prevent total collapse)
tures and buildings during past strong earthquakes and reduce life loss in a rare major earthquake.
(e.g., 1994 Northridge, 1995 Kobe, 2010 Haiti and This can be achieved through: (1) robust predic-
the most recent 2011 Tohoku earthquakes) have tion of expected future strong ground motions
raised significant concern and questions on life at the site, (2) accurate modeling of the material
safety and performance of engineering structures behavior under seismic loads, and (3) optimal
under possible future earthquakes. The occurrence distribution of the construction material.
of strong earthquakes in densely populated regions, Early works on seismic design have dealt
especially in developing countries with vulnerable with the specification of earthquake loads using
building stock and fragile infrastructure, could the response spectrum method, the time history
lead to catastrophic consequences. A notable of the ground acceleration or using the theory of
example is the 2010 Haiti earthquake that killed random vibrations. The nonlinear time-history
250,000 people and left a long-term suffering for analysis method is recognized as the most accurate
the residents of this developing country (USGS/ tool for dynamic analysis of structures (Pinho,
EERI 2010). On the other hand, the severe dam- 2007). Many researchers have also established
age caused by the 2011 Tohoku earthquake and deterministic and probabilistic hazard spectra
associated tsunami in Japan has raised significant for the site (Reiter, 1990, McGuire, 1995). The
challenges to one of the most developed countries development of mathematical models to describe
as well (Takewaki et al, 2011). Hence, the assess- the hysteretic nonlinear behavior of the structure
ment of seismic performance of structures under during earthquakes has also been pursued in
strong ground motions is an important problem in several studies (e.g., Takeda et al, 1970, Otani,
earthquake engineering. Structures need to resist 1981, Akiyama, 1985). New design concepts
unknown future earthquakes which adds more and methods, such as energy-, performance- and
complexity to the problem (Moustafa 2011, 2009, displacement-based design, base-isolation and
Moustafa & Takewaki 2010a, Abbas & Manohar structural control have been recently developed
2007, Takewaki 2002a, 2007). The consideration (Priestely et al, 2007, Takewaki, 2009, Fardis,
of the earthquake inherent uncertainty, the vari- 2010). Similarly, the optimal design of the struc-
ability in the structure parameters and modeling tures under earthquake loads has been investigated
the nonlinear behavior of the structure is essential in several studies (Fardis, 2010, Elishakoff &
for the accurate prediction of the actual response Ohsaki, 2010, Plevris, 2009, Haldar, 2006, Liang,
of the structure. Earthquake uncertainties include 2005). The evaluation of the current procedures
time, location, magnitude, duration, frequency and new practical procedures for ground motion
content and amplitude, referred to as aleatory selection and modification are provided in the
uncertainties. recent special issue on earthquake ground motion
The earthquake-resistant design of structures selection and modification for nonlinear dynamic
has been an active area of research for many analysis of structures (Kalkan & Luco, 2011). The
decades (e.g., Penelis & Kappos 1997). The struc- two edited books by Papadrakakis et al (2009)
tural engineer aims to ensure safe performance of and Tsompanakis et al. (2008) and the doctoral
129
thesis by Plevris (2009) present the state-of-the-art and other related subjects has been extensively
on advances and applications of optimal seismic investigated by several researchers, especially
design of structures considering uncertainties. at Stanford University (e.g., Baker & Cornell,
The definition of the worst (also known as 2006, Moustafa et al, 2010, Baker, 2011, Baker et
critical) ground motion represents a major chal- al, 2011, Haselton et al, 2011, Buratti et al 2011,
lenge in earthquake-resistance design of structures. Bommer & Acevedo 2004). If the number of the
This is because of the high uncertainty involved available records is small, records from other
in the occurrence of the earthquake phenomenon sites with similar soil condition or artificially
compared to the relatively low variability in the simulated ground motions could be employed.
structures properties. Strasser & Bommer (2009) The critical excitation method provides another
pose an important question on whether we have alternative in case of scarce, inhomogeneous or
seen the worst ground motion yet, and the answer limited earthquake data at the site. This method
is not. They identify the worst ground motions as has been used to assess the structures response
those having large amplitude. It may be empha- under mathematically simulated earthquake inputs
sized that the worst ground motion for a structure representing possible worst future earthquakes
may not be the worst input for a different structure. (e.g., Moustafa, 2011, 2009, Takewaki, 2002a,
For example, the 2002 earthquake of magnitude 2007). The method relies on the high uncertainty
7.9 occurred along Alaskas Denali fault killed associated with the occurrence of the earthquake
no one and did a little serious damage (Worth phenomenon, associated characteristics and also
2005), while the 1995 Kobe earthquake of 6.9 on the safety requirements of important and lifeline
magnitude killed 5,100 people and caused billion structures (nuclear plants, storage tanks, industrial
of dollars in structural damage. Early studies on installations, etc.). The critical earthquake input
defining severity of strong ground motion and for a given structure is computed by minimizing
earthquake capability to create large damage have the structures performance while satisfying pre-
focused on the earthquake intensity, peak values defined constraints observed in real earthquake
of ground acceleration, velocity and displacement records. The structural performance may be de-
(PGA, PGV, PGD), effective PGV, etc. (Housner, scribed in terms of the structures response or in
1970, Housner & Jennings, 1976). Near-field terms of reliability measures or damage indices
ground motion with pulse-like characteristic is a (Abbas & Manohar, 2007, Moustafa, 2009). The
phenomenon representing one scenario of sever- optimum design of the structure under varying
ity in the near-field region. Other scenarios of critical earthquake loads has also been studied
earthquake severity include repeated occurrence of (e.g., Fujita et al, 2010, Saikat & Manohar, 2005,
ground motion in sequences (Elnashai et al, 1998, Takewaki, 2002b). Several practical applications
Moustafa & Takewaki 2009). Deep soft soils can have evolved from the concept of critical excita-
also amplify earthquake amplitudes and modify tions. This includes design of structures to critical
frequency content. Secondary causes include also excitations, deriving critical response spectra for
the travel path effects. In reality, some of these the site, estimating critical cross power spectral
causes could exist together. density functions of multi-point and spatially-
To carry out nonlinear time-history analysis, a varying ground motions, and reliability analysis of
set of suitable accelerograms need to be selected structures to partially specified earthquake loads.
from available records (see, e.g., PEER, 2011). In The method of critical excitations has also been
this context, the criteria based on which records employed in identifying resonant accelerations and
are selected and scaled represents an interest- in selecting critical recorded accelerograms based
ing subject. A notable effort in this direction on the notion of the entropy principle (Moustafa,
130
2009, 2010). Comprehensive reviews on these 2.1 Energy Dissipated by Inelastic

aspects can be found in Moustafa (2011). Structures
This chapter deals with the damage assess-
ment for inelastic structures under worst future The energy dissipated by an N multi-degree-of-
earthquakes using the critical excitations method. freedom (MDOF) structure under the ground
The novelty of this research is in combining acceleration x(t )can be computed by integrating
damage indices, for the first time, with nonlinear the equations of motion as follows (Zahrah &
optimization and nonlinear time-history analysis Hall, 1984, Akiyama, 1985, Uang & Bertero,
for assessing the structural performance under 1990, Takewaki, 2004, Kalkan & Kunnath, 2008):
possible future ground motions. The use of dam-
age indices provides a quantitative measure for t
1 N
X ( )d =
mi xi2 (t )
T
damage and necessary repair for the structure. E K (t ) = ( )MX
0
2 i =1
Bilinear and elastic-plastic force-displacement (1a)
relationships are taken to model the material
nonlinearity, and thus the present work is lim- t N t
T ( )CX
( )d =
ited to non-deteriorating structures. Numerical E D (t ) = X xi ( )fDi ( )d
examples for one-storey and two-storey plane 0 i =1 0
(1b)
frames without irregularities are provided. Future
practical applications of the proposed methodol- t
N
ogy in seismic analysis and design of structures E H (t ) = xi ( )fsi ( )d Es (t ) (1c)
are also discussed. i =1 0
2. DAMAGE ASSESSMENT where, M, C, are the mass and damping matrices

OF INELASTIC STRUCTURES of the structure, respectively, fsi (t )is the ith hys-
UNDER EARTHQUAKE LOADS teretic restoring force, X(t) is the structure dis-
placement vector and dot indicates differentiation
Damage assessment for structures is generally
with respect to time. The quantities EK(t), ED(t),
based on the nonlinear response quantities under
Es(t) and EH(t) represent the kinetic, damping,
earthquake loads (see Table 1). The bilinear and
strain and hysteretic energies, respectively
the elastic-plastic models are shown in Figure 1.
(Moustafa, 2009). For viscous damping models,
The evaluation of the structural damage is usu-
the damping energy reduces to
ally carried out using damage indices which are N N t
quantified in terms of the structure response and c x ( )x ( )d .
ij i j
the associated absorbed energy. Therefore, the i =1 j =1 0
quantification of damage indices is carried out after Note that equations (1) provide the relative
performing nonlinear time-history analysis for the energy terms. Note also that, by the end of the
structure. The nonlinear time-history analysis for earthquake duration the kinetic and elastic strain
the structure is performed by solving the equations energies diminish. Thus, the earthquake input
of motions using numerical integration schemes energy to the structure is dissipated by hysteretic
(Moustafa, 2009, Hart & Wong 2000). and damping energies. The next section demon-
strates the use of the structures response and the
hysteretic energy in developing damage indices.
131
Table 1. Response descriptors for inelastic buildings under earthquake ground motion
Response parameter Definition

Maximum ductility
x (t )
max = max | |
0t t
f xy
Number of yield reversals Number of times velocity changes sign
x p,i
x p,i = max | |
Maximum normalized plastic deformation range 0t t f xy
N | x
p ,i |
ac . = +1
Normalized cumulative ductility i =1 xy
x (t f )
res =| |
Residual (permanent) ductility xy
tf
1
EI = E (t )dt
Normalized earthquake input energy fy x y 0 I
t
1 f
E = E (t )dt
Normalized total hysteretic energy dissipated H fy x y 0 H
EH
rE =
Ratio of total hysteretic energy to input energy EI
1 dE (t )
PI ,max = max I
Maximum rate of normalized input energy fy x y 0t t f dt
1 dE (t )
PD ,max = max D
Maximum rate of normalized damping energy
fy x y 0t t f dt
1 dE (t )
PH ,max = max H
fy x y 0t t f dt
Maximum rate of normalized hysteretic energy
2.2 Damage Measures for during the ground shaking. This measure does not
Inelastic Structures incorporate information on how the earthquake
input energy is imparted on the structure nor how
The literature on damage measures for structures this energy is dissipated. Earthquake damage oc-
under earthquake loads is vast (e.g., Cosenza et curs due to the maximum deformation or ductil-
al, 1993, Ghobarah et al, 1999). Damage indices ity and the hysteretic energy dissipated by the
are quantified in terms of a single or a combina- structure. Therefore, the definition of structural
tion of structural response parameters. Table 1 damage in terms of the ductility is inadequate.
summarizes several damage measures that are The last three measures indicate the rate of the
based on a single response parameter (Powell & earthquake input energy to the structure (i.e., how
Allahabadi 1988, Cosenza et al, 1993). The first fast the input energy E I is imparted by the earth-
measure represents the ultimate ductility produced
132
Figure 1. (a) Force-displacement relation for nonlinear materials
quake and how fast it gets dissipated). Damage E H / ( fy x y )

DI H = (4)
indices can be estimated by comparing the re- u 1
sponse parameters demanded by the earthquake
with the structural capacities. Powell & Alla-
A robust damage measure should include not
habadi (1988) proposed a damage index in terms
only the maximum response but also the effect
of the ultimate ductility (capacity) u and the
of repeated cyclic loading. Park and co-workers
maximum ductility attained during ground shak- developed a simple damage index, given as (Park
ing max : et al, 1985, Park & Ang 1985, Park et al, 1987):
x max x y max 1 x max E EH

DI = = (2) DI PA = + H = max +
xu xy u 1 xu fy x u u fy x y u
(5)
However DI does not include effects from
hysteretic energy dissipation. Additionally, this Here, x max , E H are the maximum absolute
damage index may not be zero for undamaged displacement and the dissipated hysteretic energy
structures. A damage index that overcomes this excluding elastic energy. x u is the ultimate defor-
problem has been proposed by Khashaee (2004): mation capacity under monotonic loading and
is a positive constant that weights the effect of
1
1 max cyclic loading on structural damage. Note that if
DI K = (3) = 0, the contribution to DIPA from cyclic load-
1 u1
ing is omitted.
The state of the structure damage is defined
Fajfar (1992) and Cosenza et al., (1993) quan- as: (a) repairable damage, when DI PA < 0.40 ,
tified damage based on the structure hysteretic
( b ) d a m a g e d b e y o n d r e p a i r, w h e n
energy E H :
0.40 DI PA < 1.0 , and (c) total or complete
133
collapse, when DI PA 1.0 . These criteria are tion which are selected to span satisfactory the
based on calibration of DIPA against experimen- frequency range of xg (t ) . In constructing critical
tal results and field observations in earthquakes seismic inputs, the envelope function is taken to
(Park et al., 1987). Note that Eq (5) reveals that be known. The information on energy E, peak
both maximum ductility and hysteretic energy ground acceleration (PGA) M1, peak ground
dissipation contribute to the structural damage velocity (PGV) M2, peak ground displacement
during earthquakes. Eq. (5) expresses damage as (PGD) M3, upper bound Fourier amplitude spec-
a linear combination of the damage caused by tra (UBFAS) M 4 (), and lower bound Fourier
excessive deformation and that contributed by amplitude spectra (LBFAS) M 5 () are also taken
repeated cyclic loading effect. Note also that the to be available which enables defining the fol-
quantities x max , E H depend on the loading history lowing nonlinear constraints (Abbas & Manohar,
while the quantities , x u , fy are independent of 2002, Abbas, 2006):
the loading history and are determined from ex-
perimental tests. It should also be emphasized
1
2
that Eqs (2-5) can be used to estimate damage for x2 (t ) dt E
g
a member in a structure which defines the local 0
damage. To estimate the global damage of the max | xg (t ) | M 1
0<t <
structure, a weighted sum of the local damage
max | xg (t ) | M 2 (7)
indices need to be estimated (Park et al, 1987). 0<t <
In this chapter, Eq (5) is adopted in quantifying max | x g (t ) | M 3

0<t <
the structural damage. The next section develops
M 5 () | X g () | M 4 ()
the mathematical modeling of critical future
earthquake loads.
Here, X g () is the Fourier transform of xg (t )
. Note that the constraint on the earthquake en-
3. DERIVATION OF WORST ergy is related to the Arias intensity (Arias 1970).
FUTURE EARTHQUAKE LOADS The spectra constraints aim to replicate the fre-
quency content and amplitude observed in past
The worst future ground acceleration is repre- recorded accelerograms on the future earthquake.
sented as a product of a Fourier series and an The ground velocity and displacement are obtained
envelope function in Box 1. from Eq. (6) and seen in Box 2:
Here, A0 is a scaling constant and the param- Making use of the conditions x g (0) = 0 and
eters 1 , 2 impart the transient trend to xg (t ) . lim xg (t ) 0 (Shinozuka & Henry, 1965), the
t
Ri and i are 2N f unknown amplitudes and phase
constants in the above equation can be shown to
angles, respectively and i , i = 1, 2, ..., N f are be given as (Abbas & Manohar, 2002, Abbas,
the frequencies presented in the ground accelera- 2006):
Box 1.
Nf Nf
xg (t ) = e(t ) Ri cos(it i ) = A0 [exp(1t ) exp(2t )] Ri cos(it i ) (6)

i =1 i =1
134
Box 2.
Nf t Nf t
xg (t ) = Rie( ) cos(i i )d + C 1 ; x g (t ) = Rie( )(t ) cos(i i )d + C 1t + C 2

i =1 0 i =1 0
(8)
Finally, the problem of deriving critical future

Nf
earthquake loads on inelastic structures can be
C 2 = 0;C 1 = Rie( ) cos(i i )d
i =1
posed as determining the optimization variables
0
(9) y = {R1 , R2 , ..., RN f , 1 , 2 , ..., N f }t such that the
damage index DIPA is maximized subjected to
The constraints of Eq (7) can be expressed in the constraints of Eq (10). The solution to this
terms of the variables Ri , i , i = 1, 2, ..., N f (see nonlinear constrained optimization problem is
Box 3) tackled by using the sequential quadratic program-
Here i = 1 . To quantify the constraints ming method (Arora, 2004). The following con-
quantities E, M1, M2, M3, M 4 (), and M 5 () it vergence criteria are adopted:
is assumed that a set of Nr earthquake records
| fj fj 1 | 1 ; | yi , j yi , j 1 | 2 (12)
denoted by vgi (t ), i = 1, 2, ..., N r are available for
the site under consideration or from other sites
with similar geological soil conditions. The values Herein, fj is the objective function at the jth
of energy, PGA, PGV and PGD are obtained for iteration, yi,j is the ith optimization variable at the
each of these records. The highest of these values jth iteration and 1 , 2 are small quantities to be
across all records define E, M1, M2 and M3. The specified. The structure inelastic deformation is
available records are further normalized such that estimated using the Newmark -method which is
the Arias intensity of each record is set to unity built as a subroutine inside the optimization pro-

gram. The details of the optimization procedures
(i.e.,[ vgi2 (t )dt ]1/2 = 1, Arias, 1970), and are involved in the computation of the critical earth-
0
quake and the associated damage index are shown
denoted by {vgi }iN=1r . The bounds M 4 () and
in Figure 2. Further details can also be found in
M 5 () are obtained as: Abbas (2006).
It may be emphasized that the quantities (t )
M 4 () = E max | Vgi () |; M 5 () = E min | Vgi () | and E H (t ) do not reach their respective maxima
1i N r 1i N r
(11) at the same time. Therefore, the optimization is

performed at discrete points of time and the op-
Here Vgi (), i = 1, 2, ..., N r denotes the Fou- timal solution
rier transform of the ith normalized accelerogram
vgi (t ) . The bound M 4 () has been considered y * = [R1* , R2* , ..., RN* f , 1* , 2* , ..., N* f ]t
earlier (Shinozuka, 1970, Takewaki, 2001, 2002).
The lower bound was considered by Moustafa is the one that produces the maximum DIPA across
(2002) and Abbas & Manohar (2002). all time points. The critical earthquake loads are
characterized in terms of the critical accelerations
135
Box 3.
1
Nf Nf 2
A2 E
0 Rm Rn [exp(1t ) exp(2t )]2
cos(m t m ) cos(n t n ) dt
m =1 n =1 0
Nf
max | A0 [exp(1t ) exp(2t )] Rn cos(nt n ) | M 1

0<t <
n =1
Nf t
max | A0 Rn [exp(1 ) exp(2 )] cos(n n )d

0<t <
n =1 0
Nf
(10)
A0 Rn [exp(1 ) exp(2 )] cos(n n )d | M 2
n =1 0
Nf t
max | A0 Rn [exp(1 ) exp(2 )](t ) cos(n n )d

0<t <
n =1 0
Nf
A0t Rn [exp(1 ) exp(2 )] cos(n n )d | M 3

n =1 0
Nf
M 5 () | A0 Rn {exp[1 ] exp[2 ]} cos(n n ) exp[i ]d | M 4 ()

n =1 0
and associated damage indices, inelastic deforma- also been developed in the existing literature,
tions and energy dissipated by the structure. The mainly by seismologists (see, e.g., Brune, 1970,
next section provides numerical illustrations for Hanks & McGuire, 1981, Boore, 1983, Queck et
the formulation developed in this section. al, 1990). In using these models, one needs to
In the numerical analysis, the constraints input values for a host of parameters and the suc-
quantities E, M1, M2, M3, M 4 (), and M 5 () cess of the model depends on how realistically
are estimated using past recorded earthquake data. this is done. It is possible to formulate the critical
This approach is considered to be consistent with earthquake models based on the latter class of
the aspirations of the ground motion models that models in which one can aim to optimize the
are commonly used by engineers, which, aim to parameters of the model so as to realize the least
replicate the gross features of recorded motions, favorable conditions. Note that the class of admis-
such as, amplitude, frequency content, nonsta- sible functions, in the determination of critical
tionarity trend, local soil amplification effects, excitations, in this case, becomes further con-
and duration. It is of interest to note in this context strained by the choice that one makes for the
that, predictive or physical models for ground physical model. The approach adopted in this
motions, which take into account several details, study, in this sense, is nonparametric in nature. A
such as, fault dimension, fault orientation, rupture comparison of results based on this approach with
velocity, magnitude of earthquake, attenuation, those from model-based approaches is of inter-
stress drop, density of the intervening medium, est; however, these questions are not considered
local soil condition and epicentral distance, have in the present study.
136
Figure 2. Flowchart for deriving critical earth-

parameters are changed later to study their influ-
quake loads
ence on the estimated worst earthquake loads and
the associated damage. The yield displacement is
taken as 0.10 m and the structure is taken to start
from rest. The objective function is adopted as
the Park and Ang damage index DIPA given by
Eq (5). The parameters of the Newmark -method
are taken as = 1 / 2; = 1 / 6 and t = 0.005
s.
4.1.2 Quantification of Constraints
A set of 20 earthquake records is used to quan-

tify the constraint bounds E, M1, M2, M3, M 4 ()
and M 5 () (COSMOS, 2005). Table 2 provides
information on these records. Based on numerical
analysis of these records, the constraints were
computed as E = 4.17 m/s1.5, M1 = 4.63 m/s2
(0.47 g), M2 = 0.60 m/s and M3 = 0.15 m and the
average dominant frequency was about 1.65 Hz.
The envelope parameters were taken as A0 = 2.17,
1 = 0.13, and 2 = 0.50. The convergence limits
1 , 2 were taken as 10-6 and the convergence
criterion on the secant stiffness is taken as 10-3
N/m. The frequency content for xg (t ) is taken as
(0.1-25) Hz. Additionally, in distributing the
frequencies i , i = 1, 2, ..., N f in the interval
(0.1, 25) , it was found advantageous to select
some of these i to coincide with the natural
frequency of the elastic structure and also to place
relatively more points within the modal half-
power bandwidth.
The constraint scenarios considered in deriv-
4. NUMERICAL EXAMPLES ing the worst earthquake inputs are listed in Table
3. The constrained nonlinear optimization problem
4.1 Bilinear Inelastic Frame Structure is solved using the sequential quadratic optimiza-
tion algorithm fmincon of the Matlab optimiza-
A SDOF frame structure with mass 9 103 kg, tion toolbox (Caleman et al, 1999). In the nu-
initial stiffness k1 = 1.49 105 N/m and viscous merical calculations, alternative initial starting
damping of 0.03 damping ratio is considered solutions, within the feasible region, were exam-
(initial natural frequency = 4.07 rad/s). The strain ined and were found to yield the same optimal
hardening ratio is taken equal to 0.05. These solution. To select the number of frequency terms
N f a parametric study was carried out and N f =
137
51 was found to give satisfactory results. Figure the total energy and PGA, the worst input is nar-
3 depicts the influence of Nf on the convergence row band (highly resonant) and the structure
of the objective function for constraints scenarios deformation is conservative (see Figure 4 and
1 and 4 (see Table 3). Table 4). Furthermore, most of the power of the
Fourier amplitude is concentrated at a frequency
4.1.3 Numerical Results and close to the natural frequency of the elastic struc-
Discussions ture. This amplitude gets shifted towards a
higher frequency when the strain hardening ratio
The numerical results obtained are presented in increases. The Fourier amplitudes at other frequen-
Figures 4, 5, 6, 7, 8 and 9 and Table 4. Figure 4 cies are low and uniformly distributed. This result
shows results for constraint scenario 1 and similar is substantially different from that for the elastic
results for case 4 are shown in Figure 5. Each of structure where all power of the acceleration
these figures shows the Fourier amplitude spec- amplitude is concentrated around 0 with no
trum of the worst ground acceleration, the inelastic amplitude at other frequencies (Abbas & Manohar,
deformation, the hysteretic force and the energy 2002). Additional constraints on the Fourier am-
dissipated by the structure. Figure 6 shows the plitude spectra (see Table 3) force the Fourier
time history of the ground acceleration. Based amplitude of the worst acceleration to get distrib-
on extensive analysis of the numerical results, uted across other frequencies. The critical accel-
the major observations are summarized below. eration possesses a dominant frequency that is
The frequency content and Fourier amplitude close to the average dominant frequency observed
of the worst earthquake are strongly dependent in past records (see Figure 5). The realism of the
on the constraints imposed (see Table 3). If avail- earthquake input is also evident from the maximum
able information on earthquake data is limited to damage index it produces. For instance, the dam-
Table 2. Information on past recorded ground motion records for a firm soil site
Earthquake date Magnitude Epic. Dist. Comp- PGA PGV PGD Energy* Site
(km) onent (m/s2) (m/s) (m) (m/s1.5)
Mamoth lakes 6.2 1.5 W 4.02 0.21 0.05 3.73 Convict Greek
05.25.1980 S 3.92 0.23 0.05 4.01
Loma prieta 7.0 9.7 W 3.91 0.31 0.07 3.82 Capitola
10.18.1989 S 4.63 0.36 0.11 2.61
Morgan hill 6.1 4.5 S60E 3.06 0.40 0.07 2.33 Halls valley
04.24.1984 S30W 1.53 0.30 0.02 1.64
San Fernando 6.6 27.6 N69W 3.09 0.17 0.04 2.07 Castaic old ridge
02.09.1971 N21E 2.66 0.28 0.10 2.47
Parkfield 5.0 9.1 W 2.88 0.44 0.01 1.33 Parkfield fault
12.20.1994 S 3.80 0.10 0.01 1.74
Caolinga 6.5 30.1 W 2.83 0.26 0.10 2.67 Cantua creek
05.02.1983 N 2.20 0.26 0.10 2.14
Northridge 6.7 5.9 S74E 3.81 0.60 0.12 4.17 Canoga park
01.17.1994 S16W 3.43 0.34 0.09 3.50
Cape Mendocino 7.0 5.4 W 3.25 0.45 0.15 2.44 Petrolia general
04.25.1992 S 2.89 0.24 0.08 2.31
Westmorland 5.0 6.6 E 4.35 0.33 0.11 3.26 Westmorland fire
04.26.1981 S 3.54 0.44 0.15 3.25
Imperial valley 6.4 17.4 S45W 2.68 0.22 0.10 2.30 Calexico fire
10.15.1979 N45W 1.98 0.19 0.15 2.14

* E = [ vg2 (t ) dt ]1/2 (Arias, 1970).

0
138
Table 3. Nomenclature of constraint scenarios

examined by comparing maximum response from
considered
these accelerations with those from past recorded
Case Constraints imposed ground motions. Thus, the maximum ductility
1 Energy and PGA
factor of the structure from the worst earthquake
2 Energy, PGA, PGV and PGD is about 3.9 (case 1) and 2.6 (case 4) times that
3 Energy, PGA and UBFAS
4 Energy, PGA, UBFAS and LBFAS
from the Kobe earthquake and is 2.7 (case 1) and
1.5 (case 4) times that from the San Fernando
earthquake.
To examine the effect of the strain harden-
age index for case 4 is 0.37 which is substan-
ing ratio on the design earthquake acceleration
tially smaller than 1.15 for case 1 (Table 4). The
computed, limited studies were carried out. The
constraints on PGV and PGD were not found to
value of was changed and the critical input
be significant in producing realistic critical inputs
was determined by solving a new optimization
compared to the constraints on UBFAS and LB-
problem. Namely, was taken as 0.20, 0.10, 0.05,
FAS. Also, the realism of the critical acceleration
and 0.01. The strain hardening ratio was not seen
for case 4 can be examined by comparing the
to significantly influence the frequency content
Fourier amplitude spectra and frequency content
of the critical earthquake input. It was observed,
of the worst acceleration (Figures 4, 5) with the
however, that the inelastic structure with lower
Fourier amplitude spectra of past recorded earth-
values of yields more frequently compared to the
quakes (Figure 7). Note that, while the constraint
same system with higher values. Accordingly,
scenario 1 leads to pulse-like ground motion, such
the cumulative hysteretic energy dissipated was
scenario is observable in past recorded earthquakes
observed to decrease for higher values of (Fig-
(e.g., 1971 San Fernando, 1985 Mexico, and 1995
ure 8(a)). This feature is particularly remarkable
Kobe earthquakes). Resonant or pulse-like earth-
at the end of the earthquake duration. It was also
quakes are also observable in near-field ground
observed that the results on critical earthquake
motion with directivity focusing, known as for-
accelerations for bilinear inelastic structure with
ward- and backward-directivity ground motion
= 0.01 are close to those for the elastic-plastic
(Housner & Hudson, 1958, Kalkan & Kunnath,
structure (Abbas, 2006).
2006, He & Agrawal, 2008, Moustafa 2008). The
realism of worst earthquake loads can be also
Figure 3. Convergence of objective function in terms of frequency terms Nf (a) Case 1 (b) Case 4
139
Figure 4. Optimal earthquake input and associated structural responses for case 1 (a) Fourier ampli-
tude of the ground acceleration (b) Normalized inelastic deformation (c) Hysteretic restoring force (d)
Dissipated energy
To investigate the influence of the damping compared with the same structure with lower
ratio on the computed worst earthquake load, damping ratio (see Figure 8(b)). The damage
limited studies were carried out. The damping index also reduces when the damping ratio in-
ratio was changed, namely, 0.01, 0.03 and 0.05, creases.
while all other parameters were kept unchanged. To assess the structure safety, Eq. (5) was used
The critical earthquake is computed by solving a to estimate the damage index of the structure
new optimization problem for each case. The ef- subjected to the critical earthquake load. The ef-
fect of the change in 0 was seen to be similar to fect of the parameter on the damage index is
that due to . In other words, the value of the examined first. Based on experimental tests, it
damping ratio was not seen to significantly influ- was reported that ranges between 0.05 and 0.20
ence the frequency content of the earthquake with an average value of 0.15 as suggested by
acceleration. It was observed, however, that the Park et al, (1987). Figure 9(a) shows the influence
ductility ratio and the maximum inelastic defor- of on the damage index. To study the effect of
mation of the structure decrease for higher damp- the initial natural frequency of the structure on
ing ratios. Thus, the ductility ratio decreases to the damage index, the structure stiffness was
2.43 when the damping ratio is taken as 0.05 while varied while keeping all other parameters un-
the ductility ratio increases to 2.89 when the changed and the critical earthquake was com-
damping ratio reduces to 0.01. It was also observed puted for each case. Subsequently, the value of
that the inelastic structure with higher damping DIPA was calculated for each case. In the nu-
ratio dissipates more energy through damping merical calculations was taken as 0.15 and
140
Figure 5. Optimal earthquake input and associated structural responses for case 4 (a) Fourier ampli-
tude of the ground acceleration (b) Normalized inelastic deformation (c) Hysteretic restoring force (d)
Dissipated energy
x max , max are taken as 0.10 m and 2.64, respec- of the structure is expected. The value of DIPA
tively. for the structure with 0 greater than about 1.70
The value of u was taken as 6 in Figures. 9(a) Hz is less than 0.40 and thus the structure does
and 8 in Figure 9(b). It was found that the damage not experience total damage but repairable dam-
index for the structure with initial natural fre- age. This observation is consistent since the site
quency smaller than 1.65 is higher than 0.40 and dominant frequency is around 1.65 Hz and since
thus either total collapse or damage beyond repair the Fourier amplitude of the ground acceleration
Figure 6. Optimal earthquake acceleration and velocity (a) Case 1 (b) Case 4
141
Figure 7. Fourier amplitude of recorded earthquakes (a) San Fernando 1971 (b) Hyogoken-Nanbu 1995
Figure 8. (a) Effect of strain hardening ratio on dissipated yield energy (b) Effect of damping on dis-
sipated damping energy
Figure 9. (a) Effect of the value of on the damage index (b) damage spectra for inelastic SDOF build-
ings
142
Table 4. Response parameters for alternative constraint scenarios ( = 0.05, = 0.03)
Case xmax (m) max xp (m) Nrv* DIPA Damage status

1 0.47 4.65 0.07 60 1.15 Total collapse
2 0.45 4.53 0.06 54 0.97 Damaged beyond repair
3 0.41 4.14 0.07 49 0.72 Damaged beyond repair
4 0.26 2.64 0.05 44 0.37 Repairable damage
* Nrv = number of yield reversals (see Table 1)
is seen to be located in the stiff side of the initial x y = Ly / cos = 0.0381 m.

frequency of the inelastic structure.
Thus, brace 1 yields when | x 1 | = 0.0381
4.2 Inelastic Two-Story m and brace 2 yields when | x 2 x 1 | = 0.0381
Frame Structure m. The objective function is taken as the weight-
ed damage indices in braces 1 and 2. In the nu-
A two-story braced building frame is considered merical analysis, the parameters of the Newmark
to demonstrate the formulation developed in this -method were taken as = 1 / 2; = 1 / 6 and
chapter for MDOF inelastic structures (Moustafa the time step t = 0.005 s.
2009). The material behavior of the braces is The results of this example are shown in Fig-
taken as bilinear ( k2 = k1 ) as shown in Figure ures 10 and 11. In general, the feature observed
1(a). The floor masses are taken as for the future earthquakes in the previous example
m1 = m 2 = 1.75 105 Ns2/m, the cross-section- was also observed in this example. However, the
al areas of the braces are A1 = A2 = 6.45 104 inelastic deformation and the associated damage
m2, the Youngs modulus = 2.59 1011 N/m2, and were seen to depend on the two vibration modes.
the strain hardening ratio = 0.10 (i.e., ratio of the Thus, the maximum ductility ratio for case 1 is
post-yield stiffness to the pre-yield stiffness). 4.34 while that produced from constraint case 4
When both braces are behaving elastically, the is 2.27. Similarly, the maximum response reduces
stiffness matrix Ks = Kel , if brace 1 yields from 0.15 m to 0.08 m when the constraints on
Ks = K1 , if brace 2 yields Ks = K2 and if both UBFAS and LBFAS are brought in. The earthquake
braces yield Ks = K12 . These matrices are given input energy to the inelastic system is mainly dis-
sipated by yielding and nonlinear damping of the
in Moustafa (2009). The structure is assumed to
structure. The hysteretic and damping energies are
start from rest. The first two natural frequencies
significantly higher than the recoverable strain and
of the elastic structure were computed as
kinetic energy. The kinetic and recoverable strain
1 = 6.18 rad/s and 2 = 16.18 rad/s. A Rayleigh
energies are small and diminish near the end of the
proportional damping C = aM + bKs with a =
ground shaking. The energy dissipated by yield-
0.2683, b = 0.0027 is adopted. These values are ing is significantly higher than that dissipated by
selected such that the damping ratio in the first damping. The weighted damage index for case 1
two modes is 0.03. This implies that the damping was about 0.96 implying total collapse while for
forces in braces are nonlinear hysteretic functions case 4 the damage index was about 0.35 implying
of the deformed shape of the structure. Let the repairable damage.
yield strain of the braces y = 0.002 for both It may be noted that the numerical illustrations
tension and compression. The braces will yield of the formulation developed in this paper were
at a relative displacement demonstrated for simple structures with bilinear
143
Figure 10. Critical acceleration xg (t ) for inelastic structure for case 1 (a) Time history (b) Fourier
amplitude spectrum
Figure 11. Critical acceleration xg (t ) for inelastic structure for case 2 (a) Time history (b) Fourier
amplitude spectrum
and elastic-plastic force-deformation laws. The This includes: (1) the weak cumulative component
application of the proposed method to more com- for practical cases given the typical dominance
plex structures and the use of more detailed of the peak displacement term over the accumu-
degradation models (e.g., trilinear degradation, lated energy term, (2) the use of a linear combina-
Takeda and Clough models) need to be investi- tion of deformation and energy in spite of the
gated. Additionally, in this paper Park and Ang obvious nonlinearity of the problem and the inter-
damage index has been used to assess the structure dependence of the two quantities, and (3) the lack
performance. It may be emphasized that this dam- of considering the loading sequence effect in the
age index has some limitations (Mehanny & cumulative energy term. Furthermore, when EH
Deierlein, 2000, Bozorgenia & Bertero 2003). = 0 (elastic behavior), the value of DIPA should
144
be zero. However, the value of DIPA computed is of substantial importance in deriving critical
from Eq (5) will be greater than zero. Similarly, earthquake loads for inelastic structures. This is
when the system reaches its maximum mono- because damage indices imply that the structure
tonic deformation, while DIPA should be 1.0, is damaged by a combination of repeated stress
however, Eq (5) leads to DIPA greater than 1.0. reversals and high stress excursions. This also
Chai et al (1995) proposed modification to DIPA facilitates assessing the safety of the structure by
to correct for the second drawback only. The study, providing a quantitative measure on the neces-
also, examined experimentally the implication of sary repair.
the energy-based linear damage model of DIPA. In this chapter, the worst earthquake load is
Despite the drawbacks of DIPA, it has been exten- derived based on available information using
sively used by many researchers, mainly due to inverse nonlinear dynamic analysis, optimization
its simplicity and the extensive calibration against techniques and damage indices. It was seen that if
experimentally observed seismic structural dam- available information is limited to the energy and
age during earthquakes (mainly for reinforced PGA, the resulting earthquake is highly resonant
concrete structures). Bozorgenia & Bertero (2003) and produces conservative damage. When extra
proposed two improved damage indices that information on the Fourier amplitude spectra is
overcome some of the drawbacks associated with available, more realistic earthquake loads (in
DIPA. terms of frequency content, amplitude, inelastic
In this chapter, worst earthquakes that maxi- deformations and damage indices produce) are
mize the structures damage were obtained using obtained. The influences of the strain hardening
deterministic methods. Critical earthquakes can and damping ratios on the estimated design loads
be formulated using stochastic processes, random were studied. Critical damage spectra for the site
vibration theory and reliability analysis which were also established. These spectra provide upper
provides a powerful alternative to the methodol- bounds on the structural damage and necessary
ogy developed here (see, e.g., Abbas & Manohar, repair under possible future earthquakes. The
2005, 2007). formulation developed in this chapter was dem-
onstrated for inelastic frame structures modeled
with bilinear and elastic-plastic force-deformation
5. CONCLUDING REMARKS laws. In other words, non-deteriorating structures
are only considered. Future extension of the
A methodology for assessing damage in structures present research requires the use of nonlinear
under critical future earthquake loads is developed degradation models that facilitate the development
in this chapter. The novelty of this research lies of plastic hinges in the structure. In this case the
in combining damage indices, nonlinear opti- computations will increase considerably due to the
mization and nonlinear time-history analysis in complexity in estimating the structural response.
assessing the structural performance under future Finally, it may be emphasized that in the present
earthquakes. Damage descriptors are introduced work, the structural properties have been kept
in deriving the worst earthquake ground motion. unchangeable. It is possible to apply the proposed
The structural damage is quantified in terms of methodology for optimal design of the structure
Park and Ang damage indices. As is well known, under future earthquakes. Herein, an initial guess
damage indices describe the damage state of the for the dimensions of the structures members
structure and correlate well with the actual damage needs to be assumed and an iterative procedure
displayed during earthquakes. The quantification has to be carried out leading to the optimal de-
of the structures damage using damage indices sign of the structure, the system-dependent worst
145
earthquake and the associated damage (Takewaki, Baker, J. W. (2011). Conditional mean spectrum:
2002b, 2007, Saikat & Manohar, 2005, Fujita et Tool for ground motion selection. Journal of Struc-
al, 2010). tural Engineering, 137(3), 322331. doi:10.1061/
(ASCE)ST.1943-541X.0000215
Baker, J. W., & Cornell, C. A. (2006). Spectral
ACKNOWLEDGMENT
shape, epsilon and record selection. Earthquake
Engineering & Structural Dynamics, 35, 1077
This research work is partly supported by research
1095. doi:10.1002/eqe.571
funds from the Japanese Society for the Promotion
of Science (JSPS) under Grant No. JSPS-P-08073. Baker, J. W., Lin, T., Shahi, S. K., & Jayaram, N.
The support is gratefully acknowledged. (2011). New ground motion selection procedures
and selected motions for the PEER transporta-
tion research program. PEER Technical Report
REFERENCES 2011/03. 106p.
Abbas, A. M. (2006). Critical seismic load inputs Bommer, J. J., & Acevedo, A. B. (2004). The use of
for simple inelastic structures. Journal of Sound real earthquake accelerograms as input to dynamic
and Vibration, 296, 949967. doi:10.1016/j. analysis. Journal of Earthquake Engineering, 1,
jsv.2006.03.021 4391. doi:10.1080/13632460409350521
Abbas, A. M., & Manohar, C. S. (2002). Inves- Boore, D. M. (1983). Stochastic simulation of
tigations into critical earthquake load models high-frequency ground motions based on seis-
within deterministic and probabilistic frameworks. mological models of the radiated spectra. Bul-
Earthquake Engineering & Structural Dynamics, letin of the Seismological Society of America, 73,
31, 813832. doi:10.1002/eqe.124 18651894.
Abbas, A. M., & Manohar, C. S. (2005). Reliabil- Bozorgenia, Y., & Bertero, V. V. (2003). Dam-
ity-based critical earthquake load models. Part 2: age spectra: Characteristics and applications to
Nonlinear structures. Journal of Sound and Vibra- seismic risk reduction. Journal of Structural
tion, 287, 883900. doi:10.1016/j.jsv.2004.12.003 Engineering, 129(4), 13301340. doi:10.1061/
(ASCE)0733-9445(2003)129:10(1330)
Abbas, A. M., & Manohar, C. S. (2007).
Reliability-based vector nonstationary random Bozorgnia, Y., & Bertero, V. V. (Eds.). (2004).
critical earthquake excitations for parametrically Earthquake engineering. New York, NY: CRC
excited systems. Structural Safety, 29, 3248. Press. doi:10.1201/9780203486245
doi:10.1016/j.strusafe.2005.11.003 Brune, J. N. (1970). Tectonic stress and the spectra
Akiyama, H. (1985). Earthquake-resistant limit- of seismic shear waves from earthquakes. Jour-
state design for buildings. Tokyo, Japan: Univer- nal of Geophysical Research, 75, 49975009.
sity of Tokyo Press. doi:10.1029/JB075i026p04997
Arias, A. (1970). A measure of earthquake inten- Buratti, N., Stafford, P. J., & Bommer, J. J. (2011).
sity: Seismic design of nuclear power plants (pp. Earthquake accelerogram selection and scaling
438468). Cambridge, MA: MIT Press. procedures for estimating the distribution of drift
response. Journal of Structural Engineering,
Arora, J. S. (2004). Introduction to optimum de- 137(3), 345357. doi:10.1061/(ASCE)ST.1943-
sign. San Diego, CA: Elsevier Academic Press. 541X.0000217
146
Caleman, T., Branch, M. A., & Grace, A. (1999). Elishakoff, I., & Ohsaki, M. (2010). Optimiza-
Optimization toolbox for the use with Matlab. tion and anti-optimization of structures under
Users guide. USA: The MATH WORKS Inc. uncertainty. Singapore: Imperial College Press.
doi:10.1142/9781848164789
Chai, Y. H., Romstad, K. M., & Bird, S. M.
(1995). Energy-based linear damage model for Fajfar, P. (1992). Equivalent ductility factors, tak-
high-intensity seismic loading. Journal of Struc- ing into account low-cyclic fatigue. Earthquake
tural Engineering, 121(5), 857864. doi:10.1061/ Engineering & Structural Dynamics, 21, 837848.
(ASCE)0733-9445(1995)121:5(857) doi:10.1002/eqe.4290211001
Choi, H., & Kim, J. (2006). Energy-based design Fajfar, P., & Krawinklere, H. (1997). Seismic
of buckling-restrained braced frames using hys- design methodologies for the next generation of
teretic energy spectrum. Engineering Structures, codes. Rotterdam, The Netherlands: Balkema.
28, 304311. doi:10.1016/j.engstruct.2005.08.008
Fardis, M. N. (Ed.). (2010). Advances in perfor-
Comartin, C., Brvez, S., Naeim, F., Greene, mance-based earthquake engineering. New York,
M., Blondet, M., & Cherry, S. (2004). A chal- NY: Springer.
lenge to the earthquake engineering profession-
Fujita, K., Moustafa, A., & Takewaki, I. (2010).
als. Earthquake Spectra, 20(4), 10491056.
Optimal placement of viscoelastic dampers and
doi:10.1193/1.1809130
supporting members under variable critical excita-
Conte, J. P., & Peng, B. F. (1997). Fully nonstation- tions. Earthquakes and Structures, 1(1), 4367.
ary analytical earthquake ground motion model.
Ghobara, A., Abou-Elfath, H., & Biddah, A.
Journal of Engineering Mechanics, 123, 1524.
(1999). Response-based damage assessment of
doi:10.1061/(ASCE)0733-9399(1997)123:1(15)
structures. Earthquake Engineering & Structural
Cosenza, C., Manfredi, G., & Ramasco, R. (1993). Dynamics, 28, 79104. doi:10.1002/(SICI)1096-
The use of damage functionals in earthquake engi- 9845(199901)28:1<79::AID-EQE805>3.0.CO;2-
neering: A comparison between different methods. J
Goel, R. K. (1997). Seismic response of asym-
22, 855868. doi:10.1002/eqe.4290221003
metric systems: Energy-based approach.
COSMOS. (2005). Consortium organizations for Journal of Structural Engineering, 123(11),
strong-motion observation systems. Retrieved 14441453. doi:10.1061/(ASCE)0733-
from http://db.cosmos-eq.org/scripts/default.plx 9445(1997)123:11(1444)
Decanini, L. D., & Mollaioli, F. (2001). An Haldar, A. (Ed.). (2006). Recent developments
energy-based methodology for the assessment of in reliability-based civil engineering. Singa-
seismic demand. Soil Dynamics and Earthquake pore: World Scientific Publishing Co. Pte. Ltd.
Engineering, 21, 113137. doi:10.1016/S0267- doi:10.1142/9789812707222
7261(00)00102-0
Hanks, T. G., & McGuire, R. K. (1981). The char-
Drenick, R. F. (1977). The critical excitation of acter of high frequency ground motions based on
nonlinear systems. Journal of Applied Mechanics, seismic shear waves. Bulletin of the Seismological
18, 333336. doi:10.1115/1.3424047 Society of America, 71, 20712095.
147
Hart, G. C., & Wong, K. (2000). Structural dy- Khashaee, P. (2004). Damage-based seismic
namics for structural engineers. New York, NY: design of structures. Earthquake Spectra, 21(2),
John Wiley & Sons. 371387. doi:10.1193/1.1896366
Haselton, C., Baker, J. W., Liel, A. B., & Deier- Kiureghian, A. D., & Crempien, J. (1989). An
lein, G. G. (2011). Accounting for ground mo- evolutionary model for earthquake ground motion.
tion spectral shape characteristics in structural Structural Safety, 6, 235246. doi:10.1016/0167-
collapse assessment through an adjustment for 4730(89)90024-6
epsilon. Journal of Structural Engineering,
Liang, Q. Q. (2005). Performance-based optimi-
137(3), 332344. doi:10.1061/(ASCE)ST.1943-
zation of structures. New York, NY: Spon Press.
541X.0000103
doi:10.4324/9780203334713
He, W. L., & Agrawal, A. K. (2008). Ana-
Mahin, S. A., & Bertero, V. V. (1981). An evalu-
lytical model of ground motion pulses for the
ation of inelastic seismic design spectra. Journal
design and assessment of seismic protective
of the Structural Division, 107(ST9), 17771795.
systems. Journal of Structural Engineering,
134(7), 11771188. doi:10.1061/(ASCE)0733- McGuire, R. K. (1995). Probabilistic seismic
9445(2008)134:7(1177) hazard analysis and design earthquake: Closing
the loop. Bulletin of the Seismological Society
Housner, G. W., & Hudson, D. E. (1958). The
of America, 85, 12751284.
Port Hueneme earthquake of March 18, 1957.
Bulletin of the Seismological Society of America, Mehanny, S. S., & Deierlein, G. G. (2000). Model-
48, 163168. ing of assessment of seismic performance of com-
posite frames with reinforced concrete columns
Iyengar, R. N. (1972). Worst inputs and a bound
and steel beams. Report No. 135, The John Blume
on the highest peak statistics of a class of non-
Earthquake Research Center, Stanford University.
linear systems. Journal of Sound and Vibration,
25, 2937. doi:10.1016/0022-460X(72)90593-7 Moustafa, A. (2002). Deterministic/reliability-
based critical earthquake load models for linear/
Kalkan, E., & Kunnath, S. K. (2006). Effects of
nonlinear engineering structures. Doctoral disser-
fling step and forward directivity on seismic re-
tation, Department of Civil Engineering, Indian
sponse of buildings. Earthquake Spectra, 22(2),
Institute of Science, Bangalore, India.
367390. doi:10.1193/1.2192560
Moustafa, A. (2009). Critical earthquake load
Kalkan, E., & Kunnath, S. K. (2008). Rel-
inputs for multi-degree-of-freedom inelastic
evance of absolute and relative energy con-
structures. Journal of Sound and Vibration, 325,
tent in seismic evaluation of structures. Ad-
532544. doi:10.1016/j.jsv.2009.03.022
vances in Structural Engineering, 11(1), 118.
doi:10.1260/136943308784069469 Moustafa, A. (2009). Discussion of the effect of
energy concentration of earthquake ground mo-
Kalkan, E., & Luco, N. (Eds.). (2011). Earthquake
tions on the nonlinear response of RC structures.
ground motion selection and modification for
Soil Dynamics and Earthquake Engineering, 29,
nonlinear dynamic analysis of structures. Jour-
11811183. doi:10.1016/j.soildyn.2009.02.004
nal of Structural Engineering, 137(3), 277467.
doi:10.1061/(ASCE)ST.1943-541X.0000355
148
Moustafa, A. (2009). Discussion of a new ap- Moustafa, A., & Takewaki, I. (2010). Charac-
proach of selecting real input ground motions terization and modeling of near-fault pulse-like
for seismic design: The most unfavourable real strong ground motion via damage-based critical
seismic design ground motions. Earthquake En- excitation method. Structural Engineering &
gineering & Structural Dynamics, 38, 11431149. Mechanics, 34(6), 755778.
doi:10.1002/eqe.885
Moustafa, A., & Takewaki, I. (2011). Response of
Moustafa, A. (2010). Identification of resonant nonlinear SDOF structures to random accelera-
earthquake ground motion. Indian Academy of tion sequences. Engineering Structures, 33(4),
Sciences, 35(3), 355371. 12511258. doi:10.1016/j.engstruct.2011.01.002
Moustafa, A. (2010). Discussion of analytical Moustafa, A., Takewaki, I., & Wijeyewickrema,
model of ground motion pulses for the design and A. K. (2010). Selection of earthquake records
assessment of seismic protective systems. Jour- for time-history analysis of structures. Joint 7th
nal of Structural Engineering, 137(1), 229230. International Conference on Urban Earthquake
doi:10.1061/(ASCE)ST.1943-541X.134 Engineering and 5th International Conference on
Earthquake Engineering, 3-5 March, Tokyo, Japan.
Moustafa, A. (2010). Closure to discussion of
critical earthquake load inputs for multi-degree- Moustafa, A., Ueno, K., & Takewaki, I. (2010).
of-freedom inelastic structures. Journal of Sound Critical earthquake loads for SDOF inelastic
and Vibration, 330, 356360. doi:10.1016/j. structures considering evolution of seismic waves.
jsv.2010.09.002 Earthquakes and Structures, 1(2), 147162.
Moustafa, A. (2011). Damage-based design earth- Nakashima, M., Saburi, K., & Tsuji, B. (1996).
quake loads for SDOF inelastic structures. Jour- Energy input and dissipation behavior of structures
nal of Structural Engineering, 137(3), 456467. with hysteretic dampers. Earthquake Engineering
doi:10.1061/(ASCE)ST.1943-541X.0000074 & Structural Dynamics, 25, 483496. doi:10.1002/
(SICI)1096-9845(199605)25:5<483::AID-
Moustafa, A., & Mahadevan, S. (2010). Probabilis-
EQE564>3.0.CO;2-K
tic critical earthquake excitations using earthquake
response spectra. [Building and Housing]. Asian Newmark, N. M., & Hall, W. J. (1982). Earthquake
Journal of Civil Engineering, 11(3), 295319. spectra and design. Monograph, Earthquake
Engineering Research Institute. EERI.
Moustafa, A., & Takewaki, I. (2010). Modeling
critical strong ground motion sequences on inelas- Otani, S. (1981). Hysteretic models of reinforced
tic structures. Advances in Structural Engineering, concrete for earthquake response analysis. Journal
13(4), 665679. doi:10.1260/1369-4332.13.4.665 of the Faculty of Engineering. University of Tokyo,
36(2), 407441.
Moustafa, A., & Takewaki, I. (2010). Determin-
istic and probabilistic representation of near- Pacific Earthquake Research (PEER) Center.
field pulse-like ground motion. Soil Dynamics (2011). Retrieved from http://peer.berkeley.edu/
and Earthquake Engineering, 30, 412422. peer_ground_motion_database
doi:10.1016/j.soildyn.2009.12.013
149
Papadrakakis, M., Charmpis, D. C., Lagaros, N. D., Priestely, M. J. N., Calvi, G. M., & Kowalsky, M.
& Tsompanakis, Y. (Eds.). (2009). Computational J. (2007). Displacement-based seismic design of
structural dynamics and earthquake engineering. structures. Pavia, Italy: IUSS Press.
London, UK: Taylor & Francis.
Quek, S. T., Teo, Y. P., & Balendra, T. (1990).
Park, Y. J., & Ang, A. H.-S. (1985). Mecha- Non-stationary structural response with evolu-
nistic seismic damage model for reinforced tionary spectra using seismological input model.
concrete. Journal of Structural Engineering, Earthquake Engineering & Structural Dynamics,
111(4), 722739. doi:10.1061/(ASCE)0733- 19, 275288. doi:10.1002/eqe.4290190210
9445(1985)111:4(722)
Reiter, L. (1990). Earthquake hazard analysis.
Park, Y. J., Ang, A. H.-S., & Wen, Y. K. (1985). New York, NY: Columbia University Press.
Seismic damage analysis of reinforced concrete
Riddell, F. (1995). Inelastic design spectra ac-
buildings. Journal of Structural Engineering,
counting for soil conditions. Earthquake Engi-
111(4), 740757. doi:10.1061/(ASCE)0733-
neering & Structural Dynamics, 24, 14911510.
9445(1985)111:4(740)
doi:10.1002/eqe.4290241106
Park, Y. J., Ang, A. H.-S., & Wen, Y. K.
Saikat, S., & Manohar, C. S. (2005). Inverse
(1987). Damage-limiting aseismic design of
reliability based structural design for system de-
buildings. Earthquake Spectra, 3(1), 126.
pendent critical earthquake loads. Probabilistic
doi:10.1193/1.1585416
Engineering Mechanics, 20, 1931. doi:10.1016/j.
Pecker, A. (Ed.). (2007). Advanced earthquake probengmech.2004.03.006
engineering analysis. Udine, Italy: Springer.
Sarkar, A. (2003). Linear stochastic dynamical
doi:10.1007/978-3-211-74214-3
system under uncertain load: Inverse reliability
Penelis, G. G., & Kappos, A. J. (1997). Earth- analysis. Journal of Engineering Mechanics,
quake-resistant concrete structures. London, UK: 129(6), 665671. doi:10.1061/(ASCE)0733-
E & FN SPON. 9399(2003)129:6(665)
Philippacopoulos, A. J., & Wang, P. C. (1984). SEAOC. Vision Committee. (2002). Performance
Seismic inputs for nonlinear structures. Jour- based seismic design engineering. Sacramento,
nal of Engineering Mechanics, 110, 828836. CA: Structural Engineers Association of California
doi:10.1061/(ASCE)0733-9399(1984)110:5(828) (SEAOC) Report.
Pinho, R. (2007). Non linear dynamic analysis of Shinozuka, M. (1970). Maximum structural
structures subjected to seismic action. In Pecker, A. response to seismic excitations. Journal of En-
(Ed.), Advanced earthquake engineering analysis. gineering Mechanics, 96, 729738.
New York, NY: Springer Wien. doi:10.1007/978-
Shinozuka, M., & Henry, L. (1965). Random vibra-
3-211-74214-3_5
tion of a beam column. Journal of Engineering
Plevris, V. (2009). Innovative computational Mechanics, 91, 123143.
techniques for the optimum structural design
Strasser, F. O., & Bommer, J. J. (2009). Review:
considering uncertainties. Ph. D. dissertation,
Strong ground motions Have we seen the worst?
Institute of Structural Analysis and Seismic Re-
Bulletin of the Seismological Society of America,
search, National Technical University of Athens,
99(5), 26132637. doi:10.1785/0120080300
June 2009.
150
Takeda, T., Sozen, M. A., & Nielsen, N. (1970). Tsompanakis, Y., Lagaros, N. D., & Papadrakakis,
Reinforced concrete response to simulated M. (Eds.). (2008). Structural design optimization
earthquakes. Journal of the Structural Division, considering uncertainties. London, UK: Taylor
96(ST12), 25572573. & Francis.
Takewaki, I. (2001). Probabilistic critical ex- Uang, C.-M., & Bertero, V. V. (1990). Evalua-
citation for MDOF elastic-plastic structures tion of seismic energy in structures. Earthquake
on compliant ground. Earthquake Engineer- Engineering & Structural Dynamics, 19, 7790.
ing & Structural Dynamics, 30, 13451360. doi:10.1002/eqe.4290190108
doi:10.1002/eqe.66
USGS/EERI Advance Reconnaissance Team.
Takewaki, I. (2002a). Seismic critical excita- (2010). The Mw 7.0 Haiti earthquake of January
tion method for robust design: A review. Jour- 12, 2010.
nal of Structural Engineering, 128, 665672.
Westermo, B. D. (1985). The critical excita-
doi:10.1061/(ASCE)0733-9445(2002)128:5(665)
tion and response of simple dynamic systems.
Takewaki, I. (2002b). Robust building stiff- Journal of Sound and Vibration, 100, 233242.
ness design for variable critical excitations. doi:10.1016/0022-460X(85)90417-1
Journal of Structural Engineering, 128(12),
Wong, K. K. F., & Yong, R. (2002). Earthquake
15651574. doi:10.1061/(ASCE)0733-
response and energy evaluation of inelastic
9445(2002)128:12(1565)
structures. Journal of Engineering Mechanics,
Takewaki, I. (2004). Bound of earthquake input 128(3), 308317. doi:10.1061/(ASCE)0733-
energy. Journal of Structural Engineering, 9399(2002)128:3(308)
130, 12891297. doi:10.1061/(ASCE)0733-
Yamaguchi, H., & El-Abd, A. (2003). Effect of
9445(2004)130:9(1289)
energy input characteristics on hysteretic damper
Takewaki, I. (2007). Critical excitation methods efficiency. Earthquake Engineering & Structural
in earthquake engineering. Elsevier Science. Dynamics, 32, 827843. doi:10.1002/eqe.250
Takewaki, I. (2009). Building control with pas- Zahrah, T. F., & Hall, W. J. (1984). Earthquake en-
sive dampers. Singapore: John Wiley & Sons. ergy absorption in sdof structures. Journal of Struc-
doi:10.1002/9780470824931 tural Engineering, 110, 17571772. doi:10.1061/
(ASCE)0733-9445(1984)110:8(1757)
Takewaki, I., Murakami, S., Fujita, K., Yoshi-
tomi, S., & Tsuji, M. (2011). The 2011 off the
Pacific coast of Tohoku earthquake and response
of high-rise buildings under long-period ground
motions. Soil Dynamics and Earthquake En-
gineering, 31(11), 15111528. doi:10.1016/j.
soildyn.2011.06.001
151
152
Chapter 7
Metaheuristic Optimization
in Seismic Structural
Design and Inspection
Scheduling of Buildings
Institute of Structural Analysis and Seismic Research, National Technical University Athens, Greece
Vagelis Plevris
School of Pedagogical and Technological Education (ASPETE), Greece
Nikos D. Lagaros
Institute of Structural Analysis and Seismic Research, National Technical University Athens, Greece
ABSTRACT
Optimization is a field where extensive research has been conducted over the last decades. Many types
of problems have been addressed, and many types of algorithms have been developed, while their range
of applications is continuously growing. The chapter is divided into two parts; in the first part, the
life-cycle cost analysis is used as an assessment tool for designs obtained by means of prescriptive and
performance-based optimum design methodologies. The prescriptive designs are obtained through a
single-objective formulation, where the initial construction cost is the objective to be minimized, while
the performance-based designs are obtained through a two-objective formulation where the life-cycle
cost is considered as an additional objective also to be minimized. In the second part of the chapter, the
problem of inspection of structures and routing of the inspection crews following an earthquake in densely
populated metropolitan areas is studied. A model is proposed and a decision support system is developed
to aid local authorities in optimally assigning inspectors to critical infrastructures. A combined particle
swarm ant colony optimization based framework is implemented, which proves to be an instance of a
successful application of the philosophy of bounded rationality and decentralized decision-making for
solving global optimization problems.
DOI: 10.4018/978-1-4666-1640-0.ch007
Metaheuristic Optimization in Seismic Structural Design and Inspection Scheduling of Buildings
INTRODUCTION maximum inter-story drift and floor acceleration,

the loss of rental cost, the income loss cost, the
Earthquake loading transfers large amounts of en- cost of injuries and the cost of human fatalities.
ergy in short periods of time, which might produce Furthermore, the influence of uncertainties on the
severe damages on the structural systems. During seismic response of structural systems and their
the last century, significant advances have been impact on Life Cycle Cost Analysis (LCCA) is
made towards the improvement of the seismic examined. In order to take into account the uncer-
design codes. The philosophy underlying modern tainty on the material properties, the cross-section
codes is that the building structures should remain dimensions and the record-incident angle, the
elastic for frequent earthquake events. Under Latin hypercube sampling method is integrated
rare earthquakes, however, damages are allowed into the incremental dynamic analysis procedure.
given that life safety is guaranteed. Hence, the In addition, the LCCA methodology is used as an
main task of the design procedures is to achieve assessment tool for the designs obtained by means
more predictable and reliable levels of safety of prescriptive and performance-based optimum
and operability against natural hazards. Through design methodologies. The prescriptive design
extensive research studies it was found that the procedure is formulated as a single-objective op-
Performance-Based Design (PBD) concept can timization problem where the initial construction
be integrated into a structural design procedure in cost is the objective to be minimized; while the
order to obtain designs that fulfill the provisions performance-based design procedure is defined
of a safety structure in a more predictable way as a two-objective optimization problem where
(ATC-40,1996, FEMA-350, 2000, ASCE/SEI the life-cycle cost is considered as an additional
Standard 41-06, 2006, FEMA-445, 2006, ATC- objective also to be minimized.
58, 2009). According to the PBD framework the Infrastructure networks are vital for the well-
structural behavior is assessed in multiple hazard being of modern societies; national and local
levels of increased intensity. Consequently, it is economies depend on efficient and reliable net-
very important to use robust and computation- works that provide added value and competitive
ally efficient methods for predicting the seismic advantage to an areas social and economic growth.
response of the structure in order to assess its The significance of infrastructure networks in-
capacity under different seismic hazard levels. creases when natural disasters occur since restora-
In the first part of the chapter, 3D reinforced tion of community functions is highly dependent
concrete (RC) buildings with regular and iron the affected regions receiving adequate relief
regular plan views were considered in order to resources. Infrastructure networks are frequently
examine the sensitivity of life-cycle cost value characterized as the most important lifelines
with reference to the analysis procedure (static or in cases of natural disasters; recent experience
dynamic), the number of seismic records imposed, from around the World (hurricanes Katrina and
the performance criterion used and the structural Wilma, Southeastern Asia Tsunami, Loma Prieta
type (regular or irregular). In particular, nonlin- and Northridge earthquakes and others) suggests
ear static analysis and multiple stripe analysis, that, following a natural disaster, infrastructure
which is a variation of IDA, were applied for the networks are expected to support relief opera-
calculation of the maximum inter-story drift and tions, population evacuation, supply chains and
the maximum floor acceleration. The life-cycle the restoration of community activities.
cost was calculated for both regular and irregular Infrastructure elements such as bridges, pave-
in plan test examples taking into consideration ments, tunnels, water and sewage systems, and
the damage repair cost, the cost of loss of con- highway slopes are highly prone to damages
tents due to structural damage, quantified by the caused by natural hazards, a result of possible
153
poor construction or maintenance, of design expected benefits of this work include improve-
inconsistencies or of the shear magnitude of the ments in infrastructure network restoration times
natural phenomena themselves. Rapid network and minimization of adverse impacts from natural
degradation following these disasters can severely hazards on infrastructure networks.
impact both short and long run operations resulting
in increased fatalities, difficulties in population
evacuation and the supply of clean water and food LIFE-CYCLE COST
to the affected areas. Much of the state of the art ASSESSMENT OF OPTIMALLY
in this research area indicates that attention must DESIGNED REINFORCED
be given to three important actions: (i) Fail-safe CONCRETE BUILDINGS
design and construction of infrastructure facili- UNDER SEISMIC ACTIONS
ties; (ii) Effective maintenance and management
of the available facilities; and, (iii) Planning and In the framework of the present study, two multi-
preparing actions to deal with rapid reparation story 3D RC buildings, shown in Figure 1 (a) and
of infrastructure following the disasters (Altay (b), have been optimally designed to meet the
et al. 2006, Dong et al., 1987, Peizhuangm et Eurocode (EC2 (2004) and EC8 (2004)) or the
al. 1986, Tamura et al., 2000, Mendonca et al., PBD requirements. Furthermore, the two buildings
2001, Mendonca et al., 2006, Karlaftis et al. 2007, (optimally designed according to EC2 and EC8)
Lagaros et al., 2011). have been considered in order to study the influ-
The second part of the chapter focuses on is- ence of various factors on LCCA procedure and
sues that are related to inspecting and repairing to perform critical assessment of seismic design
infrastructure elements damaged by earthquakes, a procedures. Therefore, the investigation presented
highly unpredictable natural disaster of consider- in this study is composed by three parts. In the
able importance to many areas around the world. first part the single and multi-objective optimiza-
An explicit effort is made to initiate the develop- tion problems are solved, in the second part the
ment of a process for handling post-earthquake influence of various parameters on the LCCA
emergency response in terms of optimal infrastruc- procedure is quantified while in the last part a
ture condition assessment, based on a combined critical assessment of the two design procedures
Particle Swarm Optimization (PSO) Ant Colony with reference to the life-cycle cost is presented.
Optimization (ACO) framework. Some of the
Figure 1. Test cases: (a) Eight-story 3D view, (b) Five-story 3D view
154
Single and Multi-Objective in real-life design problems that have to be dealt

Optimization Problems with simultaneously. This situation forces the
designer to look for a good compromise among
In the following paragraphs, the single and the the conflicting requirements. Problems of this kind
two-criteria design optimization problems and the constitute multi-objective optimization problems.
optimum designs obtained are presented. In general, a multi-objective optimization problem
can be stated as follows:
Problem Formulations
min sF [C IN (s),C LS (t,s)]T
The mathematical formulation of the optimization where C IN (s) = C b (s) + C sl (s) + C cl (s) + C ns (s)
problem for the single-objective formulation, as it subject to g Capacity
j (s) 0 j =1,....,k
was presented in (Lagaros et al., 2004), is defined g jPBD (s) 0 j =1,...,k
as follows: (2)
min sF C IN (s)
where s represents the design vector, F is the
where C IN (s) = C b (s) + C sl (s) + C cl (ss) + C ns (s)
feasible region where all the constraint functions
subject to g SERV
j (s) 0 j =1,...,m gCapacity and gPBD are satisfied for the PBD formula-
gULT
j (s) 0 j =1,...,k tion. The objective functions considered are the
(1) initial construction cost CIN and the life-cycle
cost CLS. Several methods have been proposed for
where s represents the design vector, F is the treating structural multi-objective optimization
feasible region where all the serviceability and problems (Coello, 2000, Marler & Arora, 2004).
ultimate constraint functions (gSERV and gULT) In this work, the Nondominated Sorting Evolu-
are satisfied. In this formulation the boundaries tion Strategies II (NSES-II) algorithm, proposed
of the feasible region are defined according to by Lagaros and Papadrakakis (2007), is used in
the recommendations of the EC8. The single order to handle the two-objective optimization
objective function considered is the initial con- problem at hand. This algorithm is denoted as
struction cost CIN, while Cb(s), Csl(s), Ccl(s) and NSES-II(+) or NSES-II(,), depending on the
Cns(s) correspond to the total initial construction selection operator.
cost of beams, slabs, columns and non-structural Various sources of uncertainty are considered:
elements, respectively. The term initial cost of on the ground motion excitation which influences
a new structure corresponds to the cost just after the level of seismic demand and on the model-
construction. The initial cost is related to material, ing and the material properties which affects the
which includes concrete, steel reinforcement, and structural capacity. The structural stiffness is
labor costs for the construction of the building. The directly connected to the modulus of elasticity
solution of the resulting optimization problem is Es and Ec of the longitudinal steel reinforcement
performed by means of Evolutionary Algorithms and concrete respectively, while the strength is
(EA) (Mitropoulou, 2010). influenced by the yield stress fy of the steel and
In practical applications of sizing optimiza- the cylindrical strength for the concrete fc and the
tion problems, the initial cost rarely gives a hardening b of the steel. In addition to the mate-
representative measure of the performance of rial properties, the cross-sectional dimensions are
the structure. In fact, several conflicting and considered as random variables. Thus, both for
usually incommensurable criteria usually exist beams and columns four random variables are
155
considered; the modulus of elasticity (Es and Ec), Optimum Design Results
the yield and cylindrical strength stresses (fy and
fc), the hardening parameter b of the stress-strain For both formulations the designs variables of
curve and the cross-sectional dimensions (B and the optimization problems are defined through
H). One random variable is considered for both the dimensions of the columns and beams cross-
confined and unconfined concrete. Furthermore, section. The columns are chosen to be rectangular
one random variable is considered for the ground and they are grouped into four categories (C1, C2,
motion excitation and one for the incident angle. C3 and C4) for the eight-story test example while
In order to account for the randomness of the in- they are grouped into three categories (C1, C2
cident angle, the ground motions are applied with and C3) for the five-story test example, while the
a random angle with respect to the structural axes beams for both test examples are grouped into two
uniformly distributed in the range of 0 to 180 de- categories (more details can be found in a study of
grees. The characteristics of the random variables Lagaros et al. (2004)). The two dimensions of the
are given in Table 1, i.e. probability density func- columns/beams along with the longitudinal, trans-
tion (PDF), mean value, coefficient of variation verse reinforcement and its spacing are the five
(CoV) and type. Therefore, the total number of design variables that are assigned to each group
random variables considered is: 54 (4+2 groups of the columns/beams. Therefore, the structural
of structural elements times 9 random variables) elements (beams and columns) are separated into
for the eight-story RC building (since 4 groups of 14 groups, 12 groups for the columns and 2 for
columns and 2 groups of beams are considered) the beams, resulting into 70 design variables for
and 45 (3+2 groups of structural elements times 9 the eight-story test example; while for the five-
random variables) for the five-story RC building story test example the structural elements (beams
(since 3 groups of columns and 2 groups of beams and columns) are separated into 10 groups, 8 for
are considered) plus one random variable for the the columns and 2 for the beams, resulting in 50
seismic record and one for the incident angle. design variables in total.
Table 1. Random variables (Ellingwood et al., 1980, Dolsek, 2009)
Random
Distribution (PDF) Mean CoV Type
variable
Earthquake Uniform - - aleatory
Incident angle* Uniform - - aleatory
mean fc Lognormal 20 MPa 4% epistemic
fc Lognormal mean fc 15% aleatory
Ec Lognormal 2.9107 kN/m 15% aleatory
Material mean fy (steel) Lognormal 500 MPa 4% epistemic
fy (steel) Lognormal mean fy 5% aleatory
Es (steel) Lognormal 2.110 kN/m
8
5% aleatory
b (steel) Lognormal 1% 5% aleatory
b Normal design value 5% aleatory
Design variables
h Normal design value 5% aleatory
* In the Range of 0 to 180 degrees.
156
Based on the prescriptive seismic design for- the initial cost is the dominant criterion and DPBD2
mulation, both buildings have been designed for obtained from the region where the life-cycle cost
minimum initial cost following an optimization is the dominant criterion. The steel and concrete
strategy proposed by Mitropoulou et al. (2010). quantities for the columns and the beams along
In particular, for the solution of the single objec- with the RC frame cost and total initial cost, for
tive optimization problem formulated as shown the three optimum designs, are presented in Tables
in Eq. (4) the EA( + ) optimization scheme is 2 and 3 corresponding to the designs of the eight-
employed (Lagaros et al. 2004) with ten parent story and five-story test example, respectively.
and offspring (==10) design vectors for both For the eight-story symmetric test example,
test examples. On the other hand, the second op- as shown in Table 2, it can be said that compared
timization problem is formulated as a two-criteria to Ddescr the DPBD1 requires 9% more concrete both
design optimization problem, as presented in Eq. for beams and columns while it requires 22% and
(5) where the initial construction cost CIN and the 31% more longitudinal steel reinforcement for
life-cycle cost CLS are the two objectives both to the beams and the columns, respectively. On the
be minimized, while for solving the problem the other hand, DPBD2 requires 37% and 30% more
NSES-II(100+100) optimization scheme was concrete for beams and columns, respectively;
employed. while it requires 70% and 56% more longitudinal
Solving the optimization problem of Eq. (4) steel reinforcement for the beams and the columns,
results to a single design denoted as Ddescr corre- respectively. Furthermore, with reference to the
sponding to the prescriptive design procedure. On RC frame initial cost, where the cost of the plates
the other hand, solving the optimization problem of is also included, it can be said that DPBD1 is by
Eq. (5) results to a group of designs that define the 10% more expensive compared to Ddescr; while
Pareto curve. In order to compare the behavior of DPBD2 is by 26% more expensive. On the other
the different designs of the Pareto front curve two hand though, with reference to the initial cost, the
characteristic designs were selected, correspond- three designs vary by 2% and 4% only.
ing to the PBD optimum designs, which they are The five-story non-symmetric test example has
denoted as DPBD1 obtained from the region where a similar trend. Based on the concrete and steel
Table 2. Eight-story test example: comparison of steel and concrete quantities in the three designs
Design Columns Beams CIN, RC Frame CIN

procedure Concrete (m3) Steel (kg.) Concrete (m3) Steel (kg.) (103MU) (103MU)
Ddescr 1.68E+02 1.84E+04 2.27E+02 1.06E+04 2.40E+02 1.44E+03

DPBD1 1.84E+02 2.41E+04 2.48E+02 1.29E+04 2.64E+02 1.46E+03
DPBD2 2.19E+02 2.87E+04 3.11E+02 1.80E+04 3.03E+02 1.50E+03
Table 3. Five-story test example: comparison of steel and concrete quantities in the three designs
Design Columns Beams CIN, RC Frame CIN

procedure Concrete (m3) Steel (kg.) Concrete (m3) Steel (kg.) (103MU) (103MU)
Ddescr 8.86E+01 5.20E+03 6.57E+01 1.45E+03 1.11E+02 7.36E+02

DPBD1 1.04E+02 6.87E+03 7.40E+01 1.75E+03 1.20E+02 7.45E+02
DPBD2 1.27E+02 8.24E+03 9.16E+01 2.50E+03 1.33E+02 7.58E+02
157
reinforcement quantities and initial costs given elastic response. Thus, the seismic design loads
in Table 3, it can be said that compared to Ddescr are reduced by the behavior factor q. According
the DPBD1 requires 13% and 18% more concrete to EC8, the nonlinear deformation of the structure
for beams and columns, respectively; while it caused by the seismic load is equal to q times the
requires 21% and 32% more longitudinal steel corresponding deformation of the linear analysis.
reinforcement for the beams and the columns, In accordance to the previous section, the three
respectively. On the other hand, DPBD2 requires 39% designs are also considered for the comparative
and 43% more concrete for beams and columns, study with reference to the life-cycle cost and the
respectively; while it requires 72% and 59% more impact of the various sources of randomness of
longitudinal steel reinforcement for the beams the LCCA procedure. The median values of the
and the columns, respectively. Furthermore, with life-cycle cost of the three designs are shown in
reference to the RC frame initial cost, where the Tables 4 to 7 corresponding to the deterministic and
cost of the plates is also included, it can be said probabilistic formulations, while the histograms of
that DPBD1 is by 8% more expensive compared to Figures 2 and 3 show the probabilistic distribution
Ddescr; while DPBD2 is by 19% more expensive. On of the life-cycle cost values for the deterministic
the other hand though, with reference to the initial and probabilistic formulations implemented into
cost, the three designs vary by 1% and 3% only. the Multi-Stripe Dynamic Analysis (MSDA) for
the three different designs along with the 90%
Prescriptive vs Performance- confidence bounds.
Based Design For the eight-story symmetric test example,
comparing the histograms of Figures 2(a) and
The difference between EC8 and PBD formula- 2(b) it can be noticed that the width of the 90%
tions is demonstrated in terms of the life-cycle confidence bounds of the life-cycle cost values
cost analysis of selected designs. The EC8 for- of design DPBD2, is much narrower compared to
mulation implements a linear analysis procedure the other two confidence bounds both for the
where the behavioral factor q is used to take into deterministic and probabilistic formulations.
account the inelastic behavior of the structure. Furthermore, it can be said that with reference to
Most of the contemporary seismic design codes the mean value of the life-cycle cost (as shown
rely on the ability of the structure to absorb energy in Table 4) DPBD1 is by 18% less expensive com-
through inelastic deformation using the reduction pared to Ddescr; while DPBD2 is by 52% less expen-
or behavior factor q. The capacity of a structure sive when the deterministic formulation is imple-
to resist seismic actions in the nonlinear range mented for 60 records. On the other hand, as
generally permits the design seismic loads to be shown in Table 5, it can be said that design DPBD1
smaller than the loads corresponding to a linear is by 5% less expensive compared to Ddescr; while
Table 4. Eight-story test example: median value of the life-cycle cost (103 MU) for the four cases and
the three designs for the deterministic formulation
Number of records
Design
10 20 40 60
Ddescr 3.13E+03 2.73E+03 2.61E+03 2.72E+03
DPBD1 2.87E+03 2.02E+03 1.83E+03 2.30E+03
DPBD2 1.91E+03 1.82E+03 1.79E+03 1.79E+03
158
DPBD2 is by 57% less expensive when the proba- by 52% less expensive when the deterministic
bilistic formulation is also implemented for 60 formulation is implemented for 60 records. On
records. For the five-story symmetric test ex- the other hand, as shown in Table 7, it can be said
ample, comparing the histograms of Figure 3(a) that design DPBD1 is by 40% less expensive com-
it can be noticed that the width of the 90% con- pared to Ddescr; while DPBD2 is by 53% less expen-
fidence bounds of the life-cycle cost values, of sive when the probabilistic formulation is also
design Ddescr, is much narrower compared to the implemented for 60 records. Comparing the de-
other two confidence bounds both for the deter- terministic with the probabilistic formulation with
ministic formulation while for the probabilistic reference to the median values, they appear to be
one it is DPBD2 design that shows the narrower increased by 12% to 30% for the eight-story test
confidence bounds. Furthermore, it can be said example and by 11% to 13% for the five-story
that with reference to the mean value of the life- building.
cycle cost (as shown in Table 6) DPBD1 is by 38%
less expensive compared to Ddescr; while DPBD2 is
Figure 2. Eight-story test case Prescriptive vs PBD (a) frequency of occurrence deterministic approach
and (b) frequency of occurrence probabilistic approach, all for the case of 60 records
Figure 3. Five-story test case Prescriptive vs PBD (a) frequency of occurrence deterministic approach
and (b) frequency of occurrence probabilistic approach, all for the case of 60 records
159
Table 5. Eight-story test example: median value of the life-cycle cost (103 MU) for the four cases and
the three designs for the probabilistic formulation
Number of records
Design
10 20 40 60
Ddescr 3.16E+03 3.04E+03 3.08E+03 3.14E+03
DPBD1 3.25E+03 2.97E+03 2.95E+03 2.98E+03
DPBD2 2.05E+03 2.00E+03 1.99E+03 2.00E+03
Table 6. Five-story test example: median value of the life-cycle cost (103 MU) for the four cases and the
three designs for the deterministic formulation
Number of records
Design
10 20 40 60
Ddescr 5.10E+03 3.54E+03 3.28E+03 4.18E+03
DPBD1 4.70E+03 3.17E+03 2.84E+03 3.04E+03
DPBD2 4.37E+03 2.58E+03 2.27E+03 2.75E+03
Table 7. Five-story test example: median value of the life-cycle cost (103 MU) for the four cases and the
three designs for the probabilistic formulation
Number of records
Design
10 20 40 60
Ddescr 4.79E+03 4.71E+03 4.47E+03 4.72E+03
DPBD1 3.90E+03 3.26E+03 3.17E+03 3.36E+03
DPBD2 3.66E+03 3.00E+03 2.90E+03 3.08E+03
METAHEURISTIC OPTIMIZATION and subsequently repair infrastructure elements

FOR THE INSPECTION SCHEDULING will delay search and rescue operations and relief
OF BUILDINGS efforts. The objective of this work is scheduling
structure and infrastructure inspection crews
Natural hazards such as earthquakes, floods and following an earthquake in densely populated
tornadoes can cause extensive failure of critical metropolitan areas. A model is proposed and a
infrastructures including bridges, water and sewer decision support system is designed to aid lo-
systems, gas and electricity supply systems, and cal authorities in optimally assigning inspectors
hospital and communication systems. Follow- to critical infrastructures. A combined Particle
ing a natural hazard, the condition of structures Swarm Ant Colony Optimization based frame-
and critical infrastructures must be assessed and work is developed which proves an instance of
damages have to be identified; inspections are a successful application of the philosophy of
therefore necessary since failure to rapidly inspect bounded rationality and decentralized decision-
160
(i )
making for solving global optimization problems N IG nSB
min [d(SBk ,C i ) D(k )]
(Plevris et al., 2010). i =1 k =1
(i )
nSB
1
Problem Formulation xC i = (i ) x
nSB k =1
k
(4)
A general formulation of a nonlinear optimization
(i )
nSB
1
problem can be stated as follows yC i = (i )
nSB
y k =1
k
D(k ) = A(k ) BP (k )
minn f ( x) x = [x 1 , , x n ]T
x R
Subject to (3) where nSB(i )

is the number of structural blocks al-
gk ( x) 0 k = 1, , m located to the ith inspection crews, d(SBk,Ci) is the
xL x xU distance between the SBk building block from the
centre of the ith group of structural blocks (with
where x is the design variables vector of length coordinates xC and yC), while D(k) is the demand
n, f(x): RnR is the objective function to be for the kth building block defined as the product
minimized, the vector of m inequality constraint of the building block total area times the built-up
functions g(x): RnRm and xL, xU are two vectors percentage (i.e. percentage of the area with a
of length n defining the lower and upper bounds structure). This is defined as a discrete optimiza-
of the design variables, respectively. tion problem since the design variables x are in-
The main objective of this work is to formulate teger numbers denoting the inspection crews to
the problem of inspecting the structural systems which each built-up block has been assigned and
of a city/area as an optimization problem. This thus the total number of the design variables is
objective is achieved in two steps: in the first step, equal to the number of structural blocks and the
the structural blocks to be inspected are optimally range of the design variables is [1, NIG].
assigned into a number of inspection crews (as-
signment problem), while in the second step the
problem of hierarchy is solved for each group of STEP 2: INSPECTION
blocks (inspection prioritization problem). In the PRIORITIZATION PROBLEM
formulation of the optimization problems consid-
ered in this work, the city/area under investigation The definition of this problem is a typical Travel-
is decomposed into NSB structural blocks while ling Salesman Problem (TSP) (Colorni et al., 1992)
NIG inspection crews are considered for inspect- which is a problem in combinatorial optimization
ing the structural condition of all structural and studied in operations research and theoretical
infrastructure systems of the city/area. computer science. In TSP a salesman spends his
time visiting N cities (or nodes) cyclically. Given
a list of cities and their - pair-wise - distances,
STEP 1: OPTIMUM the task is to find a Hamiltonian tour of minimal
ASSIGNMENT PROBLEM length, i.e. to find a closed tour of minimal length
that visits each city once and only once. For an N
The assignment problem is defined as a nonlinear city asymmetric TSP if all links are present then
programming optimization problem as follows there are (N-1)! different tours. TSP problems are
also defined as integer optimization problems,
161
similar to all problems that have been proven to and position according to its own experience
be NP-hard (Lawler, 1985). as well as the experience of other (neighbouring)
Consider a TSP with N cities (vertices or particles. Particles experience is built by tracking
nodes). The TSP can be represented by a complete and memorizing the best position encountered.
weighted graph G=(N,A), with N the set of nodes As every particle remembers the best position it
and A the set of arcs (edges or connections) that has visited during its flight, the PSO possesses
fully connects the components of N. A cost func- a memory. A PSO system combines local search
tion is assigned to every connection between two method (through self-experience) with global
nodes i and j, that is the distance between the two search method (through neighbouring experience),
nodes di,j (ij). In the symmetric TSP, it is di,j=dj,i. attempting to balance exploration and exploitation.
A solution to the TSP is a permutation p={p(1),
, p(N)} of the node indices {1, , N}, as every Mathematical Formulation of PSO
node must appear only once in a solution. The
optimum solution is the one that minimizes the Each particle maintains two basic characteristics,
total length L(p) given by velocity and position, in the multi-dimensional
search space that are updated as follows
N 1
L( p) = (dp(i ),p(i +1) ) + dp(N ),p(1) (5)
i =1 v j (t + 1) = wv j (t ) + c1 r1 ( x Pb,j x j (t )) + c2 r2 ( x Gb x j (t ))
(7)
Thus, the corresponding prioritization problem
is defined as follows x j (t + 1) = x j (t ) + v j (t + 1) (8)
nSB
(i )
1 where vj(t) denotes the velocity vector of particle
min d(SBk , SBk +1 ) + d(SBn(i ) , SB1 ) , i = 1, ..., N IG

k =1 SB
j at time t, xj(t) represents the position vector of
(6) particle j at time t, vector xPb,j is the personal best
ever position of the jth particle, and vector xGb is
where d(SBk, SBk+1) is the distance between the the global best location found by the entire swarm.
building block SBk and k+1th. The main objective The acceleration coefficients c1 and c2 indicate
is to define the shortest possible route between the degree of confidence in the best solution found
the structural blocks that have been assigned in by the individual particle (c1 - cognitive param-
Step 1 to each inspection group. eter) and by the whole swarm (c2 - social param-
eter), respectively, while r1 and r2 are two random
vectors uniformly distributed in the interval [0,
PARTICLE SWARM 1]. The symbol of Eq. (7) denotes the Had-
OPTIMIZATION ALGORITHM amard product, i.e. the element-wise vector or
matrix multiplication.
In a PSO formulation, multiple candidate solutions Figure 4 depicts a particles movement, in
coexist and collaborate simultaneously. Each solu- a two-dimensional design space, according to
tion is called a particle that has a position and Eqs. (7) and (8). The particles current position
a velocity in the multidimensional design space. xj(t) at time t is represented by the dotted circle
A particle flies in the problem search space at the lower left of the drawing, while the new
looking for the optimal position. As time passes position xj(t+1) at time t+1 is represented by the
through its quest, a particle adjusts its velocity dotted bold circle at the upper right hand of the
162
Figure 4. Visualization of the particles movement in a two-dimensional design space
drawing. It can be seen how the particles move- & Eberhart,1995), as it was introduced later by
ment is affected by: (i) its velocity vj(t); (ii) the Shi and Eberhart (1998) in a successful attempt
personal best ever position of the particle, xPb,j, to improve convergence. Large inertia weights
at the right of the figure; and (iii) the global best will force larger velocity updates allowing the
location found by the entire swarm, xGb, at the algorithm to explore the design space globally.
upper left of the figure. Similarly, small inertia values will force the veloc-
In the above formulation, the global best loca- ity updates to concentrate in the nearby regions
tion found by the entire swarm up to the current of the design space.
iteration (xGb) is used. This is called a fully con- The inertia weight can also be updated during
nected topology (fully informed PSO), as all iterations. A commonly used inertia update rule is
particles share information with each other about the linearly-decreasing, calculated by the formula:
the best performer of the swarm. Other topologies
have also been used in the past where instead of w max w min
wt +1 = w max t (9)
the global best location found by the entire swarm, t max
a local best location of each particles neighbour-
hood is used. Thus, information is shared only
where t is the iteration number, wmax and wmin are the
among members of the same neighbourhood.
maximum and minimum values, respectively, of
The term w of Eq. (7) is the inertia weight,
the inertia weight. In general, the linearly decreas-
essentially a scaling factor employed to control
ing inertia weight has shown better performance
the exploration abilities of the swarm, which
than the fixed one.
scales the current velocity value affecting the
Particles velocities in each dimension i (i = 1,
updated velocity vector. The inertia weight was
,n) are restricted to a maximum velocity vmaxi.
not part of the original PSO algorithm (Kennedy
The vector vmax of dimension n holds the maximum
163
absolute velocities for each dimension. It is more criterion presumes prior knowledge of the global
appropriate to use a vector rather than a scalar, as optimal value, which is feasible for testing or fine-
in the general case different velocity restrictions tuning the algorithm in mathematical problems
can be applied for different dimensions of the when the optimum is known a priori, but this
particle. If for a given particle j the sum of ac- is certainly not the case in practical structural
celerations of Eq. (7) causes the absolute velocity optimization problems where the optimum is not
for dimension i to exceed vmaxi, then the velocity known a priori.
on that dimension is limited to vmax,i. The vector In our study, together with the maximum
parameter vmax is employed to protect the cohesion number of iterations, we have implemented the
of the system, in the process of amplification of convergence criterion connected to the rate of
the positive feedback. The basic PSO has only few improvement of the value of the objective function
parameters to adjust. In Table 8 there is a list of for a given number of iterations. If the relative
the main parameters, their typical values as well improvement of the objective function over the
as other information (Perez & Behdinan, 2007). last kf iterations (including the current iteration) is
less or equal to a threshold value fm, convergence is
Convergence Criteria supposed to have been achieved. In mathematical
terms, denoting as Gbestt the best value for the
Due to the repeated process of the PSO search, objective function found by the PSO at iteration t,
convergence criteria have to be applied for the the relative improvement of the objective function
termination of the optimization procedure. Two can be written for the current iteration t as follows
widely adopted convergence criteria are the
maximum number of iterations of the PSO algo- Gbestt k f +1 Gbestt
rithm and the minimum error requirement on the fm (10)
Gbestt k f +1
calculation of the optimum value of the objective
function. The selection of the maximum number
of iterations depends, generally, on the complexity In Table 9 there is a list of the convergence
of the optimization problem at hand. The second parameters of the PSO used in this study with
description and details.
Table 8. Main PSO parameters
Symbol Description Details

NP Number of particles A typical range is 10 40. For most problems 10 particles is sufficient
enough to get acceptable results. For some difficult or special prob-
lems the number can be increased to 50-100.
n Dimension of particles It is determined by the problem to be optimized.
w Inertia weight Usually is set to a value less than 1, i.e. 0.95. It can also be updated
during iterations.
xL, xU Vectors containing the lower and upper They are determined by the problem to be optimized. Different ranges
bounds of the n design variables, respec- for different dimensions of particles can be applied in general.
tively
vmax Vector containing the maximum allowable Usually is set half the length of the allowable interval for the given
velocity for each dimension during one dimension: vmaxi = (xUi - xLi)/2. Different values for different dimen-
iteration sions of particles can be applied in general.
c1, c2 Cognitive and social parameters Usually c1=c2=2. Other values can also be used, provided that 0 <
c1+c2 < 4 (Perez & Behdinan, 2007)
164
Table 9. PSO convergence parameters
Symbol Description Details

tmax Maximum number of iterations for the Determined by the complexity of the problem to be optimized, in
termination criterion. conjunction with other PSO parameters (n, NP).
kf Number of iterations for which the relative If the relative improvement of the objective function over the last kf
improvement of the objective function satis- iterations (including the current iteration) is less or equal to fm, con-
fies the convergence check. vergence has been achieved.
fm Minimum relative improvement of the value
of the objective function.
PSO for Integer Optimization The algorithm was inspired by the behaviour of
real ants in nature. In many ant species, individu-
Since both problems defined in previous section als initially wander randomly and upon finding
are integer optimization problems, discrete opti- a food source return to their colony, depositing
mization algorithms are required. For the Step 1 a substance called pheromone on the ground.
optimization problem described in previous sec- Other ants smell this substance, and its presence
tion, a discrete version of the PSO algorithm is influences the choice of their path, i.e. they tend
employed. In the continuous version of the PSO to follow strong pheromone concentrations rather
method, both particle positions and velocity are than travelling completely randomly, returning
initialized randomly. In this work, the particle and reinforcing it if they eventually find food.
positions are generated randomly over the design The pheromone deposited on the ground forms
space using discrete Latin Hypercube Sampling, a pheromone trail, which allows the ants to find
thus guaranteeing that the initial particle positions good sources of food that have been previously
will be integers in the acceptable range. Further- identified by other ants.
more, in the case of discrete optimization and in As time passes, the pheromone trails start to
particular in integer programming, at every step evaporate, reducing their strength. The more time
of the optimization procedure, integer particle it takes for an ant to travel down a path and back
positions should also be generated. In order to again, the more time the pheromone trail has to
satisfy this, Eq. (7) is modified as follows evaporate. A short path gets marched over faster
than a long one, and thus the pheromone density
v j (t + 1) = round wv j (t ) + c1 r1 ( x Pb,j x j (t )) + c2 r2 ( x Gb x j (t )) remains high as it is laid on the path faster than

(11) it can evaporate. If there was no evaporation, the
paths chosen by the first ants would tend to be
where the vector function round(x) rounds each excessively attractive to the following ants and
element of the vector x into the nearest integer. as a result the exploration of the solution space
would be constrained. In that sense, pheromone
evaporation helps also to avoid convergence to
ANT COLONY OPTIMIZATION a locally optimal solution. Positive feedback
eventually leads to most of the ants following a
The Ant Colony Optimization (ACO) algorithm single optimum path.
is a population-based probabilistic technique for The idea of the ant colony algorithm is to mimic
solving optimization problems, mainly for finding this behaviour with simulated ants walking around
optimum paths through graphs (Dorigo, 1992). the graph representing the problem to solve. The
165
first algorithm was aiming to search for an optimal the higher the pheromone and the heuristic value
path in a graph. The original idea has since diversi- associated to an edge, the higher the probability
fied to solve a wider class of numerical problems the ant will choose that particular edge. Once all
and, as a result, several problems have emerged, the ants have completed their tour, the iteration is
drawing on various aspects of the behaviour of complete and pheromone values on the connec-
ants. The initial applications of ACO were in the tions are updated: each of the pheromone values
domain of NP-hard combinatorial optimization is initially decreased by a certain percentage and
problems, while it was soon also applied to rout- then it receives an amount of additional phero-
ing in telecommunication networks. mone proportional to the quality of the solutions
In ACO, a set of software agents called artificial to which it belongs.
ants search for good solutions to the optimization
problem of finding the best path on a weighted Ant Colony Optimization Algorithm
graph. The ants incrementally build solutions by
moving on the graph. The solution construction Consider a population of m ants where at each
process is stochastic and it is biased on a phero- iteration of the algorithm every ant constructs a
mone model, that is, a set of parameters associated route by visiting every node sequentially. Ini-
with graph components (either nodes or edges) tially, ants are put on randomly chosen nodes. At
whose values are modified at runtime by the ants. each construction step during an iteration, ant k
To apply ACO to the TSP, the construction applies a probabilistic action choice rule, called
graph is considered, defined by associating the random proportional rule, to decide which node to
set of cities with the set of vertices on the graph. visit next. While constructing the route, an ant k
The construction graph is fully connected and currently at node i, maintains a memory Mk which
the number of vertices is equal to the number of contains the nodes already visited, in the order
cities, since in the TSP it is possible to move from they were visited. This memory is used in order
any given city to any other city. The length of the to define the feasible neighbourhood Nki that is
edges (connections) between the vertices are set to the set of nodes that have not yet been visited by
be equal to the corresponding distances between ant k. In particular, the probability with which ant
the nodes (cities) and the pheromone values and k, currently at node i, chooses to go to node j is
heuristic values are set for the edges of the graph.
Pheromone values are modified during iterations at (i , j ) (i , j )
runtime and represent the cumulated experience of pk
= , if j Nki
((i , ) (i , ) )
i,j
the ant colony, while heuristic values are problem Nki

dependent values that, in the case of the TSP, are (12)
set to be the inverse of the lengths of the edges.
During an ACO iteration, each ant starts from where i,j is the amount of pheromone on con-
a randomly chosen vertex of the construction nection between i and j nodes, is a parameter
graph. Then, it moves along the edges of the to control the influence of i,j, is a parameter to
graph keeping a memory of its path. In order to control the influence of i,j and i,j is a heuristic
move from one node to another it probabilisti- information that is available a priori, denoting the
cally chooses the edge to follow among those desirability of connection i,j, given by
that lead to yet unvisited nodes. Once an ant has
visited all the nodes of the graph, a solution has 1
i , j = (13)
been constructed. The probabilistic rule is biased di , j
by pheromone values and heuristic information:
166
According to Eq. (13), the heuristic desirability al., 1992). In general, connections that are used
of going from node i to node j is inversely propor- by many ants and which are parts of short tours,
tional to the distance between i and j. By definition, receive more pheromone and are therefore more
the probability of choosing a city outside Nki is likely to be chosen by ants in future iterations of
zero. By this probabilistic rule, the probability of the algorithm.
choosing a particular connection i,j increases with
the value of the associated pheromone trail i,j and
of the heuristic information value i,j. CASE STUDY
The selection of the superscript parameters
and is very important: if =0, the closest cities The real world case study considered is the city of
are more likely to be selected which corresponds Patras in Greece, which is used in order to define
to a classic stochastic greedy algorithm (with both the problem of the inspection assignment and
multiple starting points since ants are initially the inspection prioritization. The city of Patras
randomly distributed over the nodes). If =0, only is decomposed into 112 structural blocks having
pheromone amplification is at work, that is, only different areas and built-up percentages, while
pheromone is used without any heuristic bias (this two different sets of inspection groups (crews of
generally leads to rather poor results (Dorigo & inspectors) are considered. A non-uniform distri-
Sttzle, 2004). bution of damages is examined with respect to the
damage level encountered on the structures due
Pheromone Update Rule to a strong earthquake. Four areas with different
structural damage levels are considered: (i) Level
After all the m ants have constructed their routes, 0 no damages, (ii) Level 1 slight damages,
the amount of pheromone for each connection (iii) Level 2 moderate damages and (iv) Level
between i and j nodes, is updated for the next 3 extensive damages. The subdivision of the city
iteration t+1 as follows of Patras into 112 structural blocks and the mean
damage level for each region are shown in Figure
m
i , j (t + 1) = (1 ) i , j (t ) + ik, j (t ), (i, j ) A 5. Damages are assumed to follow the Gaussian
k =1 distribution with mean value 0, 1, 2 and 3 for the
(14) four zones of Figure 5. The final distribution of
damages over the structural blocks can be seen in
where is the rate of pheromone evaporation, a Figure 6, where a big circle denotes severe damage.
constant parameter of the method, A is the set of In order to account for the influence of the
arcs (edges or connections) that fully connects distribution of the damages in the citys regions,
the set of nodes and ki,j(t) is the amount of the formulation of the optimal assignment problem
pheromone ant k deposits on the connections it given in Eq. (4) is modified as follows
has visited through its tour Tk, typically given by
(i )
N IG nSB
1
if connection (i,j ) belongs to Tk min [d(SBk ,C i ) D(k ) DF (k )] (16)

ik, j = L(Tk ) i =1 k =1

0
otherwise
(15) where DF(k) is the damage factor corresponding to
each damage level, as shown in Table 10. Figures
The coefficient must be set to a value <1 to 7(a) and 7(b) depict the solutions obtained for the
avoid unlimited accumulation of trail (Colorni et
167
Figure 5. City of Patras Subdivision into struc- Table 10. Damage Factor (DF) corresponding to
tural blocks and the mean damage level distributed each damage level
over the structural blocks
Damage level Damage Factor (DF)
0 1.0
1 1.2
2 1.5
3 2.0
CONCLUSION
In this study the application of metaheuristic op-

timization and in particular Evolution Strategies,
Particle Swarm Optimization and Ant Colony
Optimization is examined in two problems of
great significance, the structural seismic design
optimization problem and the inspection schedul-
ing problem after a seismic hazard attack.
In the first problem examined in this study it
was found that with reference to the factors in-
fluencing the life-cycle cost estimation it can be
concluded that 10 to 20 records are not enough to
obtain reliable life-cycle cost analysis prediction
results. The structural type of the building affects
its structural performance. It has been verified that
a symmetrical structure sustains less damage and
optimum allocation problem for the two different therefore less repair cost during its life compared
numbers of inspection crews. to a non-symmetric structure. In both test examples
In the second step, the inspection prioritization the effect of the other sources of uncertainty like
problem defined in Eq. (6) is solved by means of material properties, damping and mass proper-
the Ant Colony Optimization algorithm. Figures ties is very significant varying considerably the
8(a) and 8(b) depict the optimum routes achieved, mean, the standard deviation and the fractiles of
corresponding to the least time consuming route the seismic response. Neglecting the influence of
required for each inspection group imitating from modeling uncertainties (i.e. material properties and
their base. The base is the same for every inspec- design variables) in the prediction of the seismic
tion crew. The distances for the first and second response can significantly underestimate the val-
group are 17121 and 31540 respectively for the ues of the seismic damage indices considered. As
two inspection groups while for the four are 9633.7, a result the estimated value of the life cycle cost
10939, 11383 and 15740. varies considerably (up to 30%) compared to the
Figure 9 depicts the convergence histories of case where the cumulative impact of all sources
the ACO algorithm. The vertical axis is the of randomness is considered. Furthermore, it has
minimum distance path among the ants for every been shown that designs obtained in accordance
iteration. to the European seismic design code are more
168
Figure 6. City of Patras Distribution of the damage levels
Figure 7. City of Patras - Subdivision into structural blocks (a) two and (b) four inspection crews
169
Figure 8. City of Patras Best route (a) two and (b) four inspection crews
Figure 9. City of Patras Optimization history of the last group (a) for the case of two and (b) the case
of four inspection crews
170
vulnerable to future earthquakes compared to REFERENCES

similar designs, in terms of initial construction
cost, obtained with the performance-based design Altay, N., & Greene, W. G. (2006). OR/MS
procedure. This vulnerability increases for designs research in disaster operations management.
selected from the part of the Pareto front curves European Journal of Operational Research, 175,
where the initial construction cost is the dominant 475493. doi:10.1016/j.ejor.2005.05.016
criterion. Even though these conclusions cannot ASCE/SEI Standard 41-06. (2006). Seismic re-
be generalized, they provide an indication of the habilitation of existing buildings, prepublication
quality of the designs obtained according to a edition. Structural Engineering Institute, American
prescriptive design code and to a performance- Society of Civil Engineers, 2006. ATC-40. (1996).
based design procedure. Seismic evaluation and retrofit of concrete build-
On the other hand following a natural hazard, ings. Redwood City, CA: Applied Technology
the condition of the critical infrastructures must Council. ATC-58. (2009). Guidelines for seismic
be assessed and damages have to be identified. performance assessment of buildings. Redwood
Inspections are therefore necessary, immediately City, CA: Applied Technology Council.
after the catastrophic event, since failure to quickly
inspect, repair and/or rehabilitate the infrastructure Coello, C. A. (2000). An updated survey of GA-
system, particularly in densely populated metro- based multi-objective optimization techniques.
politan regions, might delay search and rescue ACM Computing Surveys, 32(2), 109143.
operations and relief efforts, which increases the doi:10.1145/358923.358929
suffering of the survivors. Specialized crews must
Colorni, A., Dorigo, M., & Maniezzo, V. (1992).
be dispatched and inspect critical infrastructures.
An investigation of some properties of an ant
The objective of the present work was to schedule
algorithm. In R. Manner, & B. Manderick (Eds.),
critical infrastructures inspection crews following
Proceedings of the Parallel Problem Solving from
an earthquake in densely populated metropolitan
Nature Conference (PPSN 92), (pp. 509-520).
regions. In this work two formulations have been
Brussels, Belgium: Elsevier Publishing.
successfully implemented: in the first, the struc-
tural blocks are assigned to different inspection Colorni, A., Dorigo, M., & Maniezzo, V. (1992).
groups with an effort to equally distribute the Distributed optimization by ant colonies. In F.
workload between the groups, while in the second Varela & P. Bourgine (Eds.), Proceedings of the
the optimal route for each group was determined First European Conference on Artificial Life,
with an effort to minimize the distance that each (pp. 134-142). Paris, France: Elsevier Publishing.
inspection group has to cover. A Particle Swarm
Dolsek, M. (2009). Incremental dynamic analysis
Optimization and an Ant Colony Optimization-
with consideration of modelling uncertainties.
based framework were implemented for dealing
with the problem at hand and they both resulted
38(6), 805825. doi:10.1002/eqe.869
in tractable and rapid response models.
Dong, W. M., Chiang, W. L., & Shah, H. C. (1987).
Fuzzy information processing in seismic hazard
analysis and decision making. Soil Dynamics
and Earthquake Engineering, 6(4), 202226.
doi:10.1016/0267-7261(87)90003-0
171
Dorigo, M. (1992). Optimization, learning and Lagaros, N. D., Fragiadakis, M., & Papadrakakis,
natural algorithms. Milano, Italy: Politecnico M. (2004). Optimum design of shell structures with
di Milano. stiffening beams. AIAA Journal, 42(1), 175184.
doi:10.2514/1.9041
Dorigo, M., & Sttzle, T. (2004). Ant colony
optimization. The MIT Press. Lagaros, N. D., & Karlaftis, M. G. (2011). A criti-
cal assessment of metaheuristics for scheduling
EC2. (2004). Eurocode 2. Design of concrete
emergency infrastructure inspections. Swarm
structuresPart 1: General rules and rules for
and Evolutionary Computation, 1(3), 147163.
buildings. (The European Standard EN 1992-1-
doi:10.1016/j.swevo.2011.06.002
1). Brussels, Belgium: European Committee for
Standardisation. Lagaros, N. D., & Papadrakakis, M. (2007). Seis-
mic design of RC structures: A critical assessment
EC8. (2004). Eurocode 8: Design of structures for
in the framework of multi-objective optimization.
earthquake resistance. (The European Standard
EN 1998-1). Brussels, Belgium: European Com-
36(12), 16231639. doi:10.1002/eqe.707
mittee for Standardisation.
Lagaros, N. D., & Papadrakakis, M. (2007). Robust
Ellingwood, B. R., Galambos, T. V., MacGregor,
seismic design optimization of steel structures.
J. G., & Cornell, C. A. (1980). Development of
Structural and Multidisciplinary Optimization,
a probability-based load criterion for American
33(6), 457469. doi:10.1007/s00158-006-0047-5
national standard A58. Washington, DC: Na-
tional Bureau of Standards. FEMA-350. (2000). Lawler, E. L., Lenstra, J. K., Rinnooy Kan, A.
Recommended seismic design criteria for new H. G., & Shmoys, D. B. (1985). The traveling
steel moment-frame buildings. Washington, DC: salesman problem: A guided tour of combinatorial
Federal Emergency Management Agency. FEMA- optimization. New York, NY: Wiley.
445. (2006). Next-generation performance-based
Marler, R. T., & Arora, J. S. (2004). Survey
seismic design guidelines, program plan for new
of multi-objective optimization methods for
and existing buildings. Washington, DC: Federal
engineering. Structural and Multidisciplinary
Emergency Management Agency.
Optimization, 26(6), 369395. doi:10.1007/
Karlaftis, M. G., Kepaptsoglou, K. L., & Lampro- s00158-003-0368-6
poulos, S. (2007). Fund allocation for transporta-
Mendonca, D., Beroggi, G. E. G., van Gent, D., &
tion network recovery following natural disasters.
Wallace, W. A. (2006). Designing gaming simula-
Journal of Urban Planning and Development,
tions for the assessment of Group decision support
133(1), 8289. doi:10.1061/(ASCE)0733-
Systems in emergency response. Safety Science,
9488(2007)133:1(82)
44, 523535. doi:10.1016/j.ssci.2005.12.006
Kennedy, J., & Eberhart, R. (1995), Particle
Mendonca, D., Beroggi, G. E. G., & Wallace, W.
swarm optimization. IEEE International Confer-
A. (2001). Decision support for improvisation dur-
ence on Neural Networks, Piscataway, NJ, (pp.
ing emergency response operations. International
19421948).
Journal of Emergency Management, 1(1), 3038.
doi:10.1504/IJEM.2001.000507
172
Mitropoulou, C., Lagaros, N. D., & Papadraka- Plevris, V., Karlaftis, M. G., & Lagaros, N. D.
kis, M. (2010). Building design based on energy (2010). A swarm intelligence approach for emer-
dissipation: A critical assessment. Bulletin of gency infrastructure inspection scheduling. In
Earthquake Engineering, 8(6), 13751396. Gopalakrishnan, K., & Peeta, S. (Eds.), Sustain-
doi:10.1007/s10518-010-9182-x able and resilient critical infrastructure systems:
Simulation, modeling, and intelligent engineering.
Peizhuangm, W., Xihui, L., & Sanchez, E. (1986).
Springer. doi:10.1007/978-3-642-11405-2_8
Set-valued statistics and its application to earth-
quake engineering. Fuzzy Sets and Systems, 18(3), Shi, Y., & Eberhart, R. (1998). A modified par-
347356. doi:10.1016/0165-0114(86)90011-4 ticle swarm optimizer. IEEE World Congress on
Computational Intelligence, Anchorage, AK,
Perez, R. E., & Behdinan, K. (2007). Particle
USA, (pp. 69-73).
swarm approach for structural design optimiza-
tion. Computers & Structures, 85, 15791588. Tamura, H., Yamamoto, K., Tomiyama, S., &
doi:10.1016/j.compstruc.2006.10.013 Hatono, I. (2000). Modelling and analysis of
decision making problem for mitigating natural
disaster risks. European Journal of Operational
Research, 122(2), 461468. doi:10.1016/S0377-
2217(99)00247-7
173
174
Chapter 8
Optimal Performance-
Based Seismic Design
Hamid Moharrami
Tarbiat Modares University, Iran
ABSTRACT
In this chapter, the reader gets acquainted with the philosophy of performance-based design, its principles,
and an overview of the procedures for performance evaluation of structures. The essential prerequisites
of optimal performance-based design, including nonlinear analysis, optimization algorithms, and nonlin-
ear sensitivity analysis, are introduced. The methods of nonlinear analysis and optimization are briefly
presented, and the formulation of optimal performance-based design with emphasis on deterministic
type, rather than probabilistic- (or reliability)-based formulation is discussed in detail. It is revealed
how real performance-based design is tied to optimization, and the reason is given for why, without
optimization algorithms, multilevel performance-based design is almost impossible.
INTRODUCTION tive meaning of optimization; i.e., the degree of

professionality of any engineering design can be
According to the archaeologists discoveries the measured based on its degree of optimality. This
ancient Egyptians believed that to keep a ceiling aspiration has been followed in two directions:
safely supported by columns, the columns have 1) A better understanding from the behaviour of
to have one third of area of the ceiling. Since structure for enhancing design knowledge and
ever, the effort of engineers has been devoted to 2) Achieving the best (optimum) design in the
minimizing the size of columns to retrieve more framework of structural design knowledge.
space. This indicates that engineering has an intui-
DOI: 10.4018/978-1-4666-1640-0.ch008
Optimal Performance-Based Seismic Design
Improving Design Practice wind, earthquake, etc. In this chapter, we will learn
how to find the performance level of a structure.
In the traditional design of structures, attention was To that end, since some methods of performance-
mostly paid to strength of structure and the defor- based designs require nonlinear analysis, a brief
mation control was often a secondary check. The introduction to nonlinear analysis procedures is
designer aimed to design a structure in such a way also provided to complete the discussion.
that it withstands the applied loads with sufficient
reserved resistance capacity. The structure was in Optimizing the Design
fact designed for amplified loads and checked for for Seismic Effects
deflection or side-sway. It is not too far that the phi-
losophy of design changed to load and resistance The desire of optimum design has a long history
factored design (LRFD) in which reduced ultimate and goes back to the time of Galileo and even be-
strengths of structural elements are compared to fore. However, the new era of structural optimiza-
the corresponding amplified internal forces. In this tion starts with the time of emergence of electronic
method, different loads are amplified differently computers. Considerable research works can be
based on the reliability of their evaluation; and found in the literature that have focused on the
the resistance is decreased differently for bending structural optimization under static loading. Many
moment, shear, etc. for similar reasoning. After of them have suggested efficient algorithms with
California earthquakes, including the 1989 Loma relatively good degrees of success.
Prieta and 1994 Northridge events, the need for a As the design practice evolved and the de-
better control on performance of structure became mand for structural design under earthquake
more serious. Seismic-related optimization was effects increased, the research on optimizing the
supposed to address this problem. Alternatively, structural design for seismic effects increased.
Performance-Based Design (PBD) of structures The primary works in this field were devoted
that aims to design a structure for required duc- to structural optimization under dynamic loads.
tility and targeted displacement in expected risk Most of research works in the field considered
levels was proposed to satisfy this need. This the elastic behaviour of structures. The degree
latter new design philosophy is so attractive that of success was not fully satisfactory because of
has the potential of being the next generation of the complexity of the problem, objectives to be
design philosophy. optimized, mathematical formulation, and the
In this new design philosophy, the structure solution schemes for nonlinear optimization
is expected to be such designed that it behaves problems. Among these research works some
nonlinearly under severe loadings while it behaves attempted to suggest a mathematical model for
linearly under service loads, small wind and minor a better solution of optimization problems and
earthquake effects. The design has to have enough speed up of the solution process; some tried to
ductility to tolerate specified drift for severe wind change the dynamic behaviour of the structure.
and earthquakes. Depending on behaviour of mate- Among the others, some recent papers by Masson
rial used in the structure, being ductile or brittle, et al. (2002), Besset and Jezequel(2007), Chen,
the importance of the structure, the risk level and et al. (2002), and Mills-Curran (1985), may be
the severity of loading, different design criteria consulted in this field. Park et al. (2003, 2005
may apply. In other words, different performances & 2010) suggested an equivalent static loading
may be expected from a structure with different procedure that generates the same response field
material and different load intensities including in linear static analysis that nonlinear dynamic
175
analysis does. Moharrami and Alavinasab (2006) Since the goal of this chapter is to promote
proposed a method for structural optimization the readers in particular young researchers and
under equivalent static and pseudo-dynamic practicing engineers to become acquainted with
earthquake loading. In their method, the effect of practical performance-based design optimization,
change of structural components on the amplitude and on the other hand, from the practical point of
of equivalent static loading (because of change in view the life-cycle design optimization requires
natural period of the structure) was accounted for. development of quantitative definitions for quali-
In this way, they could end up to a solution that tative subjects that are not yet well documented,
had maximum ductility leading to minimum lateral this second type of performance-based design,
earthquake loads and consequently lowest cost. despite its generality, is not covered in this chapter.
A different approach to seismic-related op- However, for those readers who are interested in
timization was also followed in seismic design, having more information in this field, among many
by considering the uncertainties that encounter others, the papers by Frangopol (2011), Fragia-
in demand and capacity assessment. This topic dakis and Lagaros (2011), Lagaros et al (2008),
attracted considerable research works in differ- Jalayer and Cornell (2009), Cornell(2008), and
ent branches. Some researchers considered the Esteva et al. (2010) are suggested to start with.
uncertainties involved in the response and per- With the promotion of deterministic perfor-
formance prediction of a structure during its life. mance-based design philosophy, some researchers
This type of problem formulation that is called in the field of structural optimization focused their
life-cycle design optimization is a more general attention to various aspects of this new subject
aspect of optimal design, and may be considered with different approaches to the problem. Since
as an overall performance-based design optimiza- optimal performance-based design is the main
tion. The optimization of the structural system on subject of this chapter, some selected research
life-cycle performance has to consider the initial works in the field will be reviewed in more detail
cost of construction of a building, vulnerability in the forthcoming sections.
of damage to structural components, its expected When we seek the optimum performance-based
costs of preventive maintenance, the expected seismic design (OPBSD), we need to know: how
costs of inspection, the estimated costs of repair or to formulate our optimization problem, what are
rehabilitation, and predicted costs of consequences the prerequisites of the problem formulation,
of failure. This in turn, requires some deep study how to choose the design variables and how to
on risk-related engineering applications including choose the optimization algorithm to solve it. In
studies about risk analysis on system survival and this chapter, all these items will be discussed.
possibilities of partial or total collapse of a structure The prerequisites of the OPBSD will be briefly
and corresponding damage assessment. It is also explained to enable the reader write his own com-
necessary to have a highly efficient and robust puter program for his especial type of structure and
algorithm to estimate the deformation capacity material properties. Special attention will be paid
of the structure and evaluate the damage to the to the sensitivity analysis that is a fundamental
structure for given deformation. To that end, some tool for classical optimization algorithms. Two
vulnerability and damage functions have to be types of formulations of sensitivity analysis for
defined in monetary terms that take into account 2D steel and reinforced concrete structures will be
not only the damage and rehabilitation-related discussed. The differences between the sensitivity
costs, but also the cost of social consequences of analyses will be presented by some samples from
structural collapse related to human life. the literature.
176
INTRODUCTION TO C. Life Safety (LS), in which the structure

PERFORMANCE-BASED DESIGN tolerates sever damage, but it remains safe
for the occupants to evacuate the building.
As was mentioned before, the performance-based D. Collapse Prevention (CP). This level of
design integrates the design for deformation performance is the final stage of life of a
with the design for strength. In this new design structure in which the structure has reached
philosophy, the ductility of structure and the its instability level, and an increase in load or
resistance for the internal forces have the same deflection results in collapse of the structure.
order of importance. It is expected from the design
engineer, to design the structure in such a way Depending on the degree of importance of a
that it exhibits different desired behaviours under structure, it may be desirable to have different
different intensities of loads. Although this design levels of performances for different levels of
strategy is general, it is mainly proposed and used earthquake intensities. For example, it may be
for design of structure under earthquake excita- desirable to design a hospital so that even for se-
tion. Accordingly in this chapter our discussion vere earthquake excitations, its performance does
is limited to this field of application. not go beyond life safety (LS) situation. This is
Although Japanese are likely the first who because hospitals are to serve to the people after
proposed and used the performance-based design, earthquake. For a residential building, it may be
the studies made in the United States of America uneconomic to be designed similar to a hospital;
are the best known and the most referred ones. therefore, it may be designed such that it experi-
Since this design philosophy is not yet mandatory ences CP situation for severe earthquakes. Table
for newly designed structures, but is essential in 1 shows an example of this kind of strategies.
rehabilitation of buildings, most of research works, FEMA356 (2000) suggests a more general
practical recommendations and applications are definition as in Table 2 for rehabilitation objec-
made in seismic rehabilitation of buildings. As an tives. Here, we call it performance objectives.
introduction to the research activities in this field, According to FEMA356, the basic safety
one may consult a paper by Ghobarah (2001) who objective (BSO) for an ordinary building is to
has reviewed the serious challenges up to 2001. satisfy the K+P criteria. i.e., every building has
ASCE41 (2007) that is the latest published to be safe enough to satisfy LS criteria under an
recommendations for rehabilitation, considers earthquake with the probability of 10% per fifty
four levels of performances for structures. These years, and withstand the earthquake with the
performance levels that are virtually adopted probability of 2% per fifty years with CP condi-
worldwide are: tion. Any design that satisfies either of M, N or
O alone, or satisfies any of A, E, I, B, F, and J in
A. Operational (OP), which is the level of per- addition to K+P, is considered to have enhanced
formance in which no damage is accepted for performance objective. Reversely, if a design does
the structure, and it behaves elastically under not meet K or P criteria is designated as limited
service loads, minor wind, and earthquake performance objective. A design that uses a
effects. lower seismic hazard or lower target Building
B. Immediate Occupancy (IO), that is the Performance Level than the BSO may be called
level of performance in which little damage a reduced performance design.
may occur in the structure, and it is safe to To design a structure for specified performance
be reoccupied immediately following the level, the design engineer needs to know the target
earthquake. displacement; i.e., the extent of deformations that
177
Table 1. A sample of desired performances for different buildings
Earthquake hazard level OP IO LS CP

The most severe in 72 years
Unimportant building such as a sunshade
Ordinary buildings such as residential buildings
Important buildings such as hospitals
Table 2. Performance objectives (FEMA356, 2000)
Earthquake hazard level OP IO LS CP

50% per 50 years A B C D
20% per 50 years E F G H
10% per 50 years I J K L
2% per 50 years M N O P
the structure has to tolerate for the specified per- and the demand that an earthquake imposes to it,
formance level. Some publications in the literature in respect to its strength and ductility character-
claim that once the target displacement is known, istics. This comparison is made in member level
it is possible to determine the design base shear as well as structure level.
and consequently to design the structure for the The capacity of a structure is a resultant of
base shear. A book by Priestley et al. (2007), that capacity of its components. If a force-deformation
instructs the direct displacement-based seismic relation is defined for elements of a structure, the
design methodology, and research works by Fajfar capacity of structure can be found from nonlinear
(2000) that propose the N2 method, and research analysis. Since computer programming for non-
works by Aschheim (1999) that suggest YPS linear analysis is difficult and is time consuming,
method may be consulted in this regard. Since engineers prefer to conduct a linear analysis. Four
these methods have not yet been adopted by design methods of analyses are accepted for seismic
codes, the iterative performance-based design is evaluation of a structure:
often used. The iterative process consists of assum-
ing a trial design, finding its performance level, 1. Linear static analysis
modifying the design and repeating the process 2. Linear dynamic analysis
again and again until the required specifications 3. Nonlinear static analysis
that meet the performance criteria are obtained. 4. Nonlinear dynamic analysis
How to Find the Performance If a static type of analysis is performed, a

Level of a Structure modification factor has to be applied to the results
to make them reflect the nonlinear behavior. This
The spirit of almost all seismic evaluation methods will be discussed later.
is a comparison between the capacity of structure
178
The Vertical Distribution of Lateral Load usually more accurate than those predicted by the
linear static analysis; accordingly compared to the
To perform a linear or nonlinear static analysis, first method, there are fewer limitations on use
the distribution of base shear or pseudo-lateral of linear dynamic analysis. Whenever applicable,
force over the storys in the vertical direction of a either of the response spectrum method or time
building, has to be such applied that it simulates history method may be used for linear dynamic
the earthquake effect accounting for dynamic analysis.
characteristics. According to ASCE41, the base When the linear type of analysis is used for
shear should be vertically distributed according demand evaluation, the structure is loaded to the
to the following equations. extent that reaches the target displacement. At
this stage, every structural component experi-
Fx = C vxV (1) ences a deformation and corresponding internal
force. All structural components have to be
checked against acceptance criteria pertaining
w x hxk
C vx = n
V (2) to the desired performance level. Discussion
w h k
i i
on how the acceptance criteria are obtained
i =1 is out of the scope of this chapter, but it worth
knowing that these criteria are specified using
where actual laboratory test results, supplemented by
the engineering judgment of various development
Cvx = vertical distribution factor; teams of FEMA1. Acceptance criteria are different
for deformation-controlled (ductile) and force-
k = 2.0 for T 2.5 sec; k= 1.0 for T 0.5 sec controlled (inductile) members. They also vary
and linear interpolation shall be used to calcu- depending on the type of structural component,
late values of k for intermediate values of T. (i.e., beam, column etc.), its material, physical
properties, and performance level. Acceptance
V = Base shear or pseudo-lateral force; criteria are given differently for the results of lin-
ear and nonlinear type of structural analyses. For
wi = portion of the effective seismic weight W forced controlled members, depending on the type
located on or assigned to floor level i; of structural element and the performance level
in question, if a linear type of analysis is used,
wx = portion of the effective seismic weight W
the internal force of member due to lateral load is
located on or assigned to floor level x;
multiplied by a reduction factor, to rationalize it
for the given performance level. Alternatively for
hi = height from the base to floor level i; and hx
deformation controlled components, the strength
= height from the base to floor level x.
of structural component is multiplied by an m-
factor and checked against design forces due
Linear Analyses
to the gravity and earthquake loads. FEMA356
(2000) recommends the following formula for
Among the above four methods, the first one is the
deformation-controlled components:
most accessible method for engineers. However,
there are some limitations on its use for irregular
mkQCE QUD (3)
structures. The distribution of demands predicted
by a linear dynamic analysis (the 2nd method), is
179
where, m is the component or element demand in question. Otherwise, a lower performance level
modifier (factor), to account for expected ductility is examined.
associated with this action at the selected Structural
Performance Level; i.e., the m-factors modify Nonlinear Analyses
the results so as to be similar to the results of a
nonlinear analysis. The m-factors are specified in Parallel and simultaneous improvements in com-
Chapters 4 through 8 of FEMA356. putational facilities in commercial software and
QCE is the expected strength of the component the theories supporting nonlinear static seismic
or element at the deformation level under consid- (pushover) analysis, is making the pushover
eration for deformation-controlled actions. method more accessible and reliable for engineers,
QUD is Deformation-controlled design action and from a practical point of view, it is foreseen
due to gravity loads and earthquake loads. to be the most popular analysis procedure in the
k is the Knowledge factor (0.75 0r 1.0) de- future for performance-based seismic design.
pending on the accuracy of collected data and The outcome of pushover analysis is the inelastic
performance objectives. capacity curve of the structure. This curve defines
For a force-controlled structural component, the capacity of the building independent of any
in the absence of either a nonlinear or limit state earthquake. To make it useful for evaluation of
analysis the following formula applies to the performance point for a given earthquake, this
design force, QUF. curve has to be converted to spectral ordinates.
This will be discussed later.
QE On the other hand because of its complexity,
QUF = QG (4) the nonlinear dynamic analysis (the 4th method)
C 1C 2C 3J
is hardly used by engineers. However, since it
is a reliable analysis method, it is more often
QUF is the Force-controlled design force due used for research purposes. When this method
to gravity plus earthquake loads. J is Force- is used the demand data can be directly used for
delivery reduction factor defined by FEMA356 evaluation of performance of the structure. Design
and is greater than 1. Coefficients C1, C2, and optimization using nonlinear dynamic analysis is
C3 are amplification factors and are defined in an extraordinarily difficult subject that is not yet
FEMA356. They are used to amplify the design used for practical design problems.
base shear for achieving displacement target at When nonlinear analysis is used for demand
the desired performance level. This pre-standard evaluation, similar to deformations, the internal
recommends the following formula for forced forces obtained for components, are used without
controlled members. substantial modification. One should realize that
unlike to nonlinear dynamic analysis, nonlinear
kQCL > QUF (5) static (pushover) analysis does not fully reflect
the nonlinear behaviour of a structure. It does not
k was previously defined. QCL is Lower-bound account for cycle strength and stiffness degrada-
strength of a component or element at the deformation. Also, the ductility demand concentration of
tion level under consideration for force-controlled individual stories is not tracked in this method.
actions. Two accepted methods i.e., the method of coef-
If structural components have capacities more ficients and method of capacity spectrum have
than the demand, it satisfies the performance level been proposed to convert the outcome of nonlinear
180
pushover analysis, the capacity curve, to usable Sa

R= Cm (8)
data as of nonlinear dynamic analysis. Vy W
The Coefficient Method Vy is yield strength calculated using nonlinear

static pushover analysis.
In the coefficient method that is recommended by
FEMA273, FEMA356 and ASCE41, the target W is effective seismic weight.
displacement, t at roof level, shall be calcu-
lated in accordance with an equation similar to Cm is effective mass factor and is given in a
Eq. (6). The internal forces and deformation of table in ASCE41.
every component will be accordingly obtained.
The coefficients Ci and Sa, have slightly differ- Sa is the response spectrum acceleration, at the
ent definitions in these references, but they are effective fundamental period and damping ratio of
essentially the same. According to ASCE41, that the building in the direction under consideration.
is believed to provide better results compared to It is shown graphically in Figure 1 in which Ts
FEMA356, the target displacement is obtained is characteristic period of response spectrum and
from the following equation. B1 is obtained from B1 = 4/[5.6 - ln(100)] and
is effective viscose damping ratio.
Te2 C2 is the adjustment factor to represent the
t = C 0C 1C 2Sa g (6)
4 2 effect of pinched hysteretic shape, cyclic stiffness
degradation and strength deterioration on maxi-
where: mum displacement response. For Te>0.7 sec,
Te is the effective fundamental period of build- C2=1.
ing in the direction under consideration. (clause
2
3.3.3.2.6) 1 R 1
C2 = 1 + (9)
C0 is the modification factor to relate spectral 800 Te
displacement of an equivalent SDOF system to the
roof displacement of the MDOF building system.
C1 is modification factor to relate expected
maximum inelastic displacements to displace-
ments calculated for linear elastic response. It is
calculated from the following equation. According Figure 1. General horizontal response spectrum
to ASCE41, some limitations apply.
(R 1)
C1 = 1 + (7)
Te2
is site class factor given by ASCE41.
R is the ratio of elastic strength demand to cal-

culated yield strength.
181
With this target displacement, the internal Having the capacity curve converted to capac-
forces and deformation of structural components ity spectrum curve it remains to draw the demand
are obtained and checked to the acceptance criteria. response spectrum in the same coordinate. Noting
From this comparison, the performance level of that response spectrum as specified by a design
structure is known. The checking process is simi- code is a curve specified in Sa-T coordinates, and
lar to linear analysis, but with different criteria; we need demand spectrum in Sa-Sd coordinates,
i.e., it is similar in the sense that it depends on the horizontal axis should be transformed to Sd
the properties of structural component whether simply noting that Sd=T2/42Sa g. The outcome of
it is deformation-controlled or force-controlled this conversion is an acceleration-displacement
component. response spectrum (ADRS) curve.
As building yields in response to seismic de-
The Capacity Spectrum Method mand, it dissipates energy with hysteretic damping;
that means its damping increases. With an increase
In the capacity spectrum method that is recom- in damping, the demand spectrum reduces. i.e.,
mended by ATC40, the performance point that is the ADRS curve is not unique, and it depends on
alternatively called the target point is obtained by the effective damping property of the structure.
intersecting the capacity spectrum and demand According to ATC 40, the effective equivalent
spectrum. The capacity spectrum is indeed the viscose damping, is obtained from eq=0+0.05
capacity curve converted from force-displacement in which 0.05 is the inherent viscose damping of
ordinates to spectral acceleration-spectral dis- the structure that is considered to be constant. 0
placement ordinates. This is done by converting is the hysteretic damping represented as viscose-
the base-shear to Sa and displacement to Sd. The equivalent hysteresis damping and is calculated
following formulae apply. from: 0=ED/(4ES0) in which ED is the damping
dissipated energy and ES0 is the maximum strain
VW energy. According to ATC 40 (section 8.2.2),
Sa = (10)
1 ED and ES0 are calculated based on performance
point that is the intersection of the ADRS curve
and the reduced demand spectrum. This implies
roof
Sd = (11) that the performance point is obtained through an
PF1 (roof ,1 ) iterative process.
There are several techniques for finding this
n
2
intersection. Here, we explain one of them. Since
m

1 i =1
i i 1 the intersection of demand and capacity spectra,
1 = (12) (the target/performance point) is not known a
W n
mi (i 1 ) 2
priori, an initial target point is assumed on the
i =1 capacity spectrum curve. Then with the infor-
mation of this point and the capacity spectrum
n
curve, the corresponding viscose-equivalent
m i i1
hysteresis damping 0 and eq are obtained and
PF1 = n
i =1
(13) the corresponding reduced demand spectrum is
m (
i =1
i i1
2
) constructed and intersected to capacity spectrum
curve to produce a new performance/target point.
If the obtained target point is close enough to the
assumed one, the performance point is found;
182
otherwise a new target point has to be assumed If for a certain degree of freedom at a node,
on capacity spectrum, and the process should be the internal forces of all members connected to
repeated. Figure 2 illustrates the procedure. that node are summed up and equated to the corre-
Having the nonlinear demand obtained via any sponding external load, an equation of equilibrium
of aforementioned analyses and subsequent tech- for that particular degree of freedom will be gen-
niques, and comparing it to the specification of erated. If this is done for all degrees of freedoms,
acceptance criteria at desired performance level(s) a set of nonlinear equations can be obtained. The
identifies the performance level of a structure. nonlinear analysis procedure is the art of finding
a displacement vector that satisfies the set of all
nonlinear equations of equilibrium in Eq.(14),
A BRIEF LOOK INTO simultaneously. Once the displacement vector
NONLINEAR ANALYSES in Eq.(14) is obtained, it remains to calculate the
internal forces in all members. At this stage, the
Any nonlinear analysis procedure requires the pre-defined force-deformation relation is used to
establishment of nonlinear set of equations of obtain internal forces.
equilibrium as follows: Since the performance-based design in its
general form requires nonlinear analysis, design
[K (x , )]{} = {P } (14) engineers in this field are recommended to study in
details some useful publications before employing
any commercial software. This will help to under-
This in turn, necessitates the definition of a
stand the merits of different nonlinear schemes
force-deformation relation in member level. If
with respect to each other. Crisfield (2000) and
the force-deformation relation is known in its ex-
Owen and Hinton (1980) have cited good sum-
plicit mathematical form, the assembly of explicit
maries of classical nonlinear analysis techniques.
nonlinear stiffness matrix is theoretically possible,
A brief description of some of these methods that
otherwise, as it often happens, the stiffness matrix
are suitable for monotonically increasing curves is
will be a numerically established matrix that is an
provided hereunder and summarized graphically
implicit function of deformation.
in Figure 3.
Figure 2. Capacity spectrum method
183
Incremental Scheme: As shown in Figure tial stiffness method in the sense that both
3a, in this method, displacement is found methods start with the initial stiffness and
by gradual loading and modification of the continue analysis with unbalanced loads,
stiffness matrix of the structure at the be- however, Newton Raphson method uses
ginning of each load increment. To get a tangential stiffness of the structure at the
satisfactory result from this method, load- beginning of each analysis instead of ini-
ing increments should be quite small; oth- tial stiffness. In this method too, the pro-
erwise, it results in incorrect displacement cess continues until the unbalanced load
values. becomes infinitesimally small.
Initial Stiffness method: In this method, Combined Methods: To achieve a more
which is shown in Figure 3b, the structure efficient method, the above techniques
is analyzed with its initial stiffness. After can be combined. For example, Figure 3d
each analysis, internal forces of members, demonstrates the combination of the initial
which can be measured by the previously stiffness method with the incremental so-
obtained displacement, are used to find lution. Figure 3e shows Newton Raphson
the unbalanced forces in nodes. The un- method in conjunction with the incremen-
balanced forces are then applied again to tal loading scheme. A combination of three
the structure with its initial stiffness. The methods can also be seen in Figure 3f. In
process of analysis of structure under un- this latter method, loading on structure is
balanced forces is repeated until external divided into several segments. At the be-
and internal forces and moments reach ginning of every loading increment, stiff-
equilibrium. ness matrix of structure is established/
Newton-Raphson method: As Figure 3c modified and displacement vector of struc-
shows, this method is similar to the ini-
Figure 3. A schematic presentation of nonlinear analysis
184
ture is obtained using iterative initial stiff- Despite its robustness, this method suffers from
ness technique. the considerable number of variables that enter
in the QP sub-problem. Recently Tin Loi has
There are also some other techniques that have improved this technique. Among others, a paper
been established for inelastic analysis of structures by Tangaramvong and Tin Loi (2011) may be
based on theorems of Structural Variation. Struc- consulted in this field.
tural variation theory studies the effect of change
of properties, or even removal, of a member on
the entire structure. It takes advantage of linear DESIGN OPTIMIZATION
analysis and sensitivity of structure to some self
equilibrating unit loads that are applied at the The traditional design as shown in Figure 4a
end nodes of changing members. This technique consists of a cycle of four components. They
has been applied to analysis of several types are: Design, Analysis, Feasibility-check and
of inelastic skeletal structures including space Design improvement. If the initial design is not
trusses, frames, and grids, etc. It has been also satisfactory or feasible, a revision on the initial
extended to nonlinear finite elements analysis. design is made, and the design cycle is repeated.
Although this method takes advantage of initial The process stops when the design is deemed
stiffness matrix and does not require a change satisfactory. A design that is obtained in this way
in the stiffness matrix of structure during the may be satisfactory but not optimum. If, as shown
analysis process, it is a hierarchical and step by in Figure 4b, in the design cycle an optimization
step method of analysis in which every step uses component is added to enhance the design intel-
information from the previous step and is not ligently, the outcome of the design cycle may be
a proper nonlinear analysis that is to be joined both satisfactory and optimum.
to a performance-based design program. As an Therefore, the design optimization may be
example in this field one may start with a paper defined as a mathematical means that is used to
by Saka (1997) evolve a structure from its initial form to a final
Nonlinear analysis of structures by the math- form with characteristics of being the optimum.
ematical programming is another field of research This process may be employed for a member by
in this ground. De Donato (1977) presented member optimization or whole-structure. To
fundamentals of this method for both holonomic optimize a design, it should be written in the form
(path independent) and nonholonomic material of the standard optimization problem as follows:
behaviors. In this method, it is assumed that dis-

placement of nodes of an elasto-plastic structure Minimize Z (x )
comprises two parts namely elastic and plastic
Subject to : g j (x ) 0 ; j = 1, 2, ..., m
parts. Then, the problem of finding total displace-
ment vector of a structure is formulated in the hk (x ) = 0. ; k = 1, 2,, ..., p
form of a quadratic programming (QP) problem (15)
with some complementary yield constraints. These
yield constraints state that individual members where x is the generalized vector of design vari-
either are stressed within elastic limits and do not ables including x1 to xn. The design variables
accept plastic deformations or, are stressed up to establish the design space. Z is the objective func-
yield limit, and as a result, undergo some plastic tion. g and h are inequality and equality design
deformations. The output of this sub-problem is constraints, respectively. Design constraints divide
linear and nonlinear deformation of the structure. the design space into feasible and infeasible sec-
185
Figure 4. Design cycles: a) traditional, b)optimal
tions. The solution of constrained optimization a large part of body of classical optimization
problem places within feasible space. The feasible and are most commonly used. The second order
space for problems with equality constraints is optimization algorithms that use the second or-
the boundary constructed by the intersection of der derivatives of constraints are fairly efficient
all equality constraints. In the engineering design algorithms, but because of calculation of second
problems, we rarely confront to problems with order derivatives, they are not proper algorithms
equality constraints. Considerable research effort for problems with numerous design variables.
has been paid to propose some solution algorithms Many attempts have been made to formulate
that efficiently solve the problem. Depending on the optimization problem in such a way that it
the order of derivatives that optimization algo- is easily solved. Optimization problems may be
rithms use for solution, Haftka (1992) divides categorized into linear and nonlinear. If the con-
them into three main groups. They are: zero order, straints and objective function of a problem are
first order and second order algorithms. linear, the problem is called a linear programming
Zero order algorithms do not use any derivative problem and may be solve by Simplex method.
of objective functions or constraints. There are a Most of structural design optimization problems
few zero order classical algorithms, but there are are naturally nonlinear problems because of the
many hierarchical and probabilistic algorithms nonlinear nature of indeterminate structures. For
such as Genetic Algorithms, Particle Swarm Op- determinate structures, the fully stressed design
timization, Ant Colony, etc. Most of zero order (FSD) optimization or similar algorithms (such as
algorithms are used for unconstrained optimization simultaneous failure mode (SFM), etc.) which are
problems. Since almost all engineering design a member to member optimization strategy may
optimization problems are constrained problems, be used. In most of engineering problems, the
to use a zero order algorithm for their solution, design constraints cannot be defined in explicit
the constrained optimization problems have to be form in terms of design variables. Accordingly,
converted to unconstrained optimization problems the Taylor series expansion is used to express the
via one of Penalty Function methods. A book by constraints in explicit form as follows:
Arora (2004) will assist in this regard. The first
order algorithms use the first order derivatives of
constraints. This group of algorithms comprise
186
QP sub-problems should be repeated. These strat-

1 egies are called Sequential Linear Programming
F (x ) = F (x 0 ) + F (x 0 )T x + x T 2F (x 0 )x + ...
2 (SLP) and Sequential Quadratic Programming
(16)
(SQP), respectively. Although SLP and SQP can
often capture the optimum solution, they are not
where, x is the vector of deign variables; F (x 0 )
so much efficient algorithms. Considerable effort
is the gradient of the function F at the current
has been paid to find efficient and robust optimiza-
point and 2F (x 0 ) is the matrix of second order
tion algorithms. Among successful optimization
derivatives, named Hessian matrix of the function.
algorithms are Dual method and Conlin algorithm
Although the design constraint may be implicit
by Fleury (1983 and 1989) and DOT algorithm
function of design variables, its derivatives can
by Vanderplatts (1997). The degree of success
be numerically obtained by sensitivity analysis.
for an optimization algorithm somehow depends
The calculation of Hessian of a function, particu-
on the nature of the optimization problem. An
larly if it is a function of numerous design variables
algorithm may work very well for a problem and
is relatively difficult and computationally time
not well for another. There is not a firm conclu-
consuming. Some algorithms such as DFP and
sion about the efficiency and robustness of any
BFGS use the first order derivatives to gradually
especial optimization algorithm. An optimization
construct the Hessian matrix or its inverse from
algorithm may solve one optimization problem
its first order derivatives.
efficiently, and not solve another. For example if
If a function is approximated with its first two
an optimization problem is expressed in terms of
terms of Taylor series, it is called linearized form
reciprocals of design variables, it may be solved
of the function. Keeping the first three terms of
in an excellent way, but if it is defined in terms
Taylor series approximates the function with a
of direct design variables it may not be solved
quadratic form. Except for problems with a few
properly. This may happen vice versa.
design variables, the quadratic term is rarely used.
As a conclusion, the choice of the optimization
Expressing the constraints and objective function
algorithm for a design optimization problem does
in their linear form, makes it possible to define
not have a certain rule. However, some algorithms
a Linear Programming (LP) sub-problem and
have shown that in many applications are often
solve it by Simplex algorithm. Objective func-
more efficient than the others. Optimality Criteria
tion is usually given explicitly in terms of design
(OC) methods are believed to be more efficient
variables. If the objective function is written in
than other classical optimization algorithms. These
quadratic form and the constraints expressed in
methods do not directly consider the objective
linear form, a Quadratic Programming (QP) prob-
function in the solution process. They seek the
lem is established. The solution of an optimization
optimality of the solution in satisfaction of some
problem can be sought by converting it to LP or
predefined criteria. The FSD and SFM methods
QP sub-problem and solving it. The QP and LP
are among this group of optimization algorithms.
problems have a straight-forward solution scheme
Some research works focused their attempt to
and most of mathematical programming software
combine the principals of OC methods with those
have these solution options. The outcome of LP
of classical optimization algorithms. The Kuhn-
or QP sub-problem will not necessarily insure the
Tucker-based OC method is an example of this
feasibility and optimality of the solution because
kind. In this type of OC algorithm, the necessary
the constraints of these problems are some ap-
conditions for optimality of a design point that
proximation of actual design problem. Therefore,
are called Kuhn-Tucker conditions, are chosen
the process of establishing and solution of LP or
as a basis for the establishment of recursive for-
187
mulae for updating design variables. Moharrami of FD method depends on the degree of nonlinear-
(2006) improved the convergence of this type of ity of the function and the value of di . Habibi
OC algorithm by establishing and solving a QP and Moharrami (2010) showed that FD method
sub-problem for finding the Lagrange Multipliers. not only does not give accurate sensitivity values
This improvement completes the efficiency and but also sometimes results in false evaluation.
robustness of the algorithm.
Analytical Sensitivity Analysis
SENSITIVITY ANALYSIS Another way of evaluating the sensitivity of a

structural behavior is to differentiate the basic
The change in a behaviour Bj of structure due to equation from which that behavior is evaluated.
change in a design variable di that is expressed as As an example, the sensitivity of the displace-
ment vector when a structure behaves linearly can
B j be derived as follows:
di The basic equation that the displacement vector
is calculated from is:
is called the sensitivity of Bj with respect to di.
The sensitivity calculation is performed in several K = P (18)
ways. Here, we point to some of them.
Differentiating this equation with respect to
Finite Difference Method any generic design variable, di results in:
As was mentioned before, the behaviour B is usu- K P

ally implicit function of design variable. Therefore, +K = (19)
di di di
it is not possible to find the sensitivity via a direct
differentiation of a function. As a remedy for this
problem, the Finite Difference (FD) method has Rearranging the above equation for di
been widely used. In this method, the difference results in the sensitivity equation as follows:
between the values of a function at two adjacent
points are used to find the approximate slope of the P K
= K 1 (20)
curve (sensitivity) at the desired point as follows: di di di
B j B j (d 1 ) B j (d 2 )
(17) Note that all terms in the right hand side of
di (di1 di2 ) the above equation is known and although is
not known explicitly in terms of design variables,
The numerator of the right side of Eq.(17) is its derivatives can be calculated accurately. To
the difference between the values of Bj at two calculate the sensitivity di from the above
adjacent design points d1 and d2 where all com- equation efficiently, the paper by Arora and
ponents of vector of design variables remain Haug,(1979) is recommended for study.
unchanged except di that can assume any value As another example, consider the evaluation
of di1 di i.e., the current point and a point before, of sensitivity of a number of vibrating frequencies
or current point and the point after, or the points of a structure. The basic equation for calculating
before and after the current point. The accuracy
188
the frequencies and modal shapes is the follow- [K (d, )]{} = {P } (14)
ing equation:
That is the stiffness matrix itself becomes a
K M 2 = 0 (21)
function of displacement. Provided that the non-
linear relation of the stiffness matrix with the
Differentiating the above equation with respect displacement vector is explicitly known, theo-
to the design variable, di, and rearranging the terms retically saying, the equation of equilibrium can
results in the following equation. be solved by mathematics and the specific dis-
placement vector can be obtained in terms of
design variables. However, if the relation is not
K M 2 = 2M + 2 M K known explicitly, the calculation of displacement
d di di di
i vector and its derivative di has to be fol-
(22)
lowed in a different manner. Differentiating the
above equation with respect to any design variable
If the above equation is pre-multiplied by T
di, results in the following complicated equation:
and noted that
K (d, ) ndf K (d, ) P

K M 2 = K M 2 = 0 ,
+ + K (d, )] =
r
{ } [
T
d
i r =1 r di
d
i i
d

(25)
then the following equation is obtained:
Obviously the above equation cannot be eas-
T M K ily solved for di ; therefore, a different
2 M + 2T T =0
di di di strategy has to be followed for the sensitivity of
(23) structural nonlinear behavior. There are a number
of papers that have paid attention to this problem.
If the vector f is normalized in such a way that For the sake of illustrations, the sensitivity cal-
T M = 1 , then the sensitivity of frequency can culations in moment resisting steel frames by
be obtained easily from the following expression: Gong (2003) is explained and compared to those
of reinforced concrete frames by Habibi and
1 T K M Moharrami (2010).
= 2 (24)
di 2 d di
i
A Sensitivity Analysis Procedure
for Nonlinear Steel Frame
Nonlinear Sensitivity Analysis
In his research for performance-based design,
In the sensitivity calculations in Eqs.(18-20), since Gong used a special nonlinear static (pushover)
the structure is in its linear behaviour state, the analysis scheme proposed by Hasan et al.(2002).
stiffnesses of structural members do not depend In this nonlinear analysis, the progressive degrada-
on the deformation, , and therefore, K di tion of stiffness of a frame structure is referenced
and consequently di are easily calculated. to its plastification. As the bending moment, M,
However, if the structure is in nonlinear behavior goes beyond the yielding moment My, a partial
state, the equation of equilibrium becomes as in plasticity (Plastification) develops in the section.
Eq.(14) that is repeated here. As the plastification increases, the rigidity of the
189
section decreases. Hasan et al. simulated the partial This is to say that sum of internal forces at a
plasticity in a section with a semi-rigid connection node is equal to external loads. Differentiating
in which the rigidity of connection is progres- Eq.(27) with respect to any design variable di
sively decreased as the moment in the section is results in the following relation.
increased. In particular, a potential plastic-hinge
section was simulated with a semi-rigid connec- F F P
+ = (28)
tion whose stiffness variation is measured by a di di di
plasticity-factor p that ranges from unity, for ideal
elastic, to zero, for fully plastic. If this degrada-
Noting that F is actually the global
tion is monitored, its influence on the nonlinear
tangential stiffness matrix, Eq.(28) can be written
behavior of the member and the overall behavior
as:
of a structure under increasing lateral load can be
traced. Hasan et al. assumed an elliptic moment-
P F
curvature relation for post elastic behavior of a KT = (29)
di di di
section. Then he replaced the plasticity factor
with rigidity factor and performed the nonlinear
analysis in an incremental scheme. As the lateral As per above explanations, the tangential stiff-
load is gradually increased in infinitesimal steps, ness matrix, KT, for a given displacement vector
the stiffness matrix of the structure is deteriorated. is easily calculated. Therefore, the sensitivity of
This deterioration is considered by modifying the displacement can be obtained from the following
Cs and Cg coefficient matrices in the following equation that is similar to Eq.(20) except that the
formula based on plasticity factors. calculation of P di and F di requires
further consideration.
K = S CS + G Cg (26)
P F
= KT 1 (30)
where, S is the standard elastic stiffness matrix; di di di
CS is a correlation matrix expressed in terms of
plasticity factors; G is the standard geometric In the above equation, the first term in pa-
stiffness matrix and Cg is corresponding correction renthesis is indeed the change in the external
matrix, formulated in terms of plasticity factors. earthquake forces. Recalling Eq.(1), this may be
In this way, they could obtain the magnitude of interpreted in two ways. First, the change in the
overall deformation of the whole-structure as well base shear V, that may occur due to change in the
as all members. stiffness of the structure, as a result of change in
The sensitivity analysis that Gong proposed design variables. The second, the change in the
for his problem is defined in this framework of vertical distribution factor Cvx. However, since
analysis. Since the sensitivity of displacement is in the nonlinear pushover analysis, the load level
the basis of evaluation of sensitivities of other is invariant of design variables, the sensitivity
structural behaviours, it is usually found first. To of lateral load depends on sensitivity of vertical
obtain sensitivity of displacement vector, Eq. (14) distribution of lateral loads that is a function of
was written in the following form. parameter k in Eq.(2) and k is given in terms of
natural period of the structure. See FEMA356 or
F (d, ) = P (d ) (27) ASCE41 in this regard. Therefore, utilizing chain
190
rule in differentiation, the sensitivity of lateral the value of F n di can be computed in an ac-
load distribution becomes as follows: cumulative process. It is now sufficient to substi-
tute for F di and P di in Eq.(30) to obtain
C vx C vx k T the sensitivity of displacement vector to change
= (31)
di k T di in a design variable.
If Eq.(24) is used for finding the sensitivity A Sensitivity Analysis procedure for
of natural period of the structure, and the change nonlinear reinforced Concrete frame
of the mass of the structure due to change in the
design variables is ignored, one can easily find: In this section the sensitivity analysis proposed by
Habibi and Moharrami (2010), will be described
T T3 K to show a different sensitivity analysis procedure
= 2 T (32) and emphasize on the fact that formulation of
di 8 di
any design sensitivity analysis has to be derived
based on assumptions made on the corresponding
The second term in the parenthesis of Eq.(30), nonlinear analysis.
however, is the variation of internal forces due to As per any nonlinear analysis of structures that
a change in design variable when there is no requires the stiffness-displacement relation for as-
change in external loads. This is in fact, the main sembling Eq.(14), and requires force-displacement
challenge in nonlinear sensitivity analysis. To relation for solution of nonlinear Eq.(14) in the
calculate this quantity, Gong pointed out that form of Eq.(27), it is necessary to define a non-
internal forces in members are the accumulation linear moment-curvature and a stiffness-curvature
of their incremental forces, F , induced in mem- relation for concrete structural elements. Habibi
bers during incremental analysis procedure i.e.: adopted a tri-linear moment-curvature relation,
proposed by Park and Ang (1985) as shown in
F n n
F n
KTl Figure 5. This assumption helps to achieve the
= = l (33)
d i l =1 di l =1 di amount of curvature for a given moment and
amount of the moment for a given curvature.
Therefore, the sensitivity of internal force can The first line represents the without-crack
be accumulatively obtained at every load step as situation. The second line stands for post-crack
follows: to yielding state and the third line corresponds to
yield to the ultimate state. With the help of these
definitions, the distribution of curvature of a beam
F n F n 1 KTn
= + n (34) under applied loads can be obtained and plotted
d i di di
as per Figure 6a. The distribution of curvature
makes it possible to create a nonlinear model at
Noting that KTn is the tangential stiffness ma- the element level. Park et al. (1987) proposed a
trix at iteration n and is an explicit function of nonlinear model for R/C members (used in IDARC
plasticity index that in turn is a function of inter- software) in which the flexural stiffness in any
nal forces in members, one can easily calculate section is related to its curvature. Since both
the value of KTn di . Having n calculated flexural deformation and flexibility have recipro-
from analysis at the iteration n, the second term cal relation with stiffness of element, there will
in the right hand side of Eq.(34) is known, and be an analogy between flexibility of the section
191
Figure 5. A trilinear moment curvature relation
and its flexural deformation. Accordingly Figure penetration coefficients, and L is the free length
6b can be considered for the flexibility distribution of the element. Having A and B together with
in the RC elements. In Figure 6b, EI A and EI B EI A and EI B , facilitates obtaining the flexibility
are the flexural stiffness of the section at end A curve of a member and its corresponding defor-
and end B, respectively, and EI 0 is the initial mation, etc.
stiffness of the element; A and B are the yield
Figure 6. (a) Curvature distribution along a RC element, (b) flexibility assumption along a RC element
192
Habibi and Moharrami (2010) adopted the negligible. The following equations are used
Parks nonlinear model and modified it to take to find the displacement vector in the nonlinear
the axial force into account and formulated the Modified Newton Raphson method.
nonlinear stiffness matrix of the nonlinear concrete
elements based on these assumptions. For a rein- KTl j = P l F j 1 j = KTl 1 (P l F j 1 )
forced concrete frame under gravity and lateral (37)
loads, it was assumed that prior to execution of
lateral loads, the structure behaves linearly. The l = j ; D = l (38)
nonlinear behavior starts when the lateral loads j l
are applied and increased to a certain level.

Similar to nonlinear analysis of steel frames, it is where, KTl is the tangential stiffness matrix at the
assumed that in the reinforced concrete frame, beginning of load step l ; j is the incremental
the unloading does not happen. Habibi (2008) displacement due to the unbalanced force
used the Modified Newton Raphson method for (P l F j 1 ) at the jth iteration of the lth load step;
nonlinear analysis. In this form of nonlinear l is the subtotal displacement for the load step
analysis, the internal forces in members are ob- l and D = l is the total displacement for all
tained using tangential stiffness matrix. At the l
beginning of every load step, as shown in Figure load steps, up to load level l.
3f the tangential stiffness matrix, Kl, is updated, The sensitivity analysis proposed by Habibi
and the displacement vector is obtained assuming and Moharrami (2010) is based on the foregoing
an elastic behavior for the structure. With the nonlinear analysis. Assuming linear behavior
displacement found, the internal forces in all during every unbalanced force analysis, the
members are obtained from the following equation Eq.(37) can be utilized to produce the sensitivity
and the equilibrium is checked in all nodes. of incremental displacement j with respect to
any design variable di as follows:
Fel , j = Kel ej 1 (35)
j 1 P
l
F j 1 KTl
= KTl j (39)
d i d i d i d i
where, ej 1 is the displacement sub-vector for Noting that during any load segment, the prop-
the element at (j-1)th iteration of load step l. Real- erties of tangential stiffness matrix and the load
izing that the internal forces in members at any level presumably do not change, their derivatives
analysis stage is the sum of increments, the fol- remain unchanged within any load step. Therefore,
lowing equation is used. Eq.(30) can be used during repeated analysis in
any particular load segment, and the sensitivity
Fel , j = Fel 1 + Fel , j (36) of displacement vector at any level of loading can
be obtained as follows:
If equilibrium is not satisfied in any node, the
j j 1 P l F j 1 KTl j
unbalanced force in the node is calculated and the = =KTl n l
di di di di di
structure is analyzed for unbalanced forces in all
l l

(40)
nodes. This gives the increment in the displace-
ment vector. This process is repeated several times
where, nl is the number of iterations of N-R pro-
until the unbalance forces in all nodes become
cess at the relative load step l. The sensitivity
193
P l x i can be calculated based on how it is to Figure 3a, followed a similar procedure for
obtained. For example, if Eq.(1) is used for its sensitivity analysis of displacement. On the other
calculation, this equation should be differentiated hand, Habibi used the modified Newton-Raphson
with respect to design variables and Eqs.(31 & method as shown in Figure 3f. In this method,
32 should be used. Since we know how to calcu- consideration of unbalanced forces is a must.
late P in terms of design variables, its derivatives Therefore, in the establishment of corresponding
may be easily obtained. sensitivity analysis, this matter was taken care
The second term in Eq.(40) is derivative of of. Another common point in the two sensitivity
the resultant of internal forces in nodes. It can methods was the fact that sensitivity assessment
be calculated by summing up the derivatives of was obtained through an accumulative calcula-
internal forces in elements. The derivatives of in- tion process. This is not to say that the sensitivity
ternal forces in elements, in turn, can be calculated calculation is an iterative process, but is to say
using Eq.(33) and (34). Note that in Eq.(40), the that during the analysis process, some derivatives
second term is the derivative of internal forces in have to be obtained and summed up for use in
the previous iteration, and from Eqs.(35 & 36 it sensitivity computation in upcoming stages.
is understood that the internal forces themselves In Gongs method of analysis, a lumped in-
are determined based of deformations of previ- elasticity was considered, and a plasticity factor
ous loadings. was defined to simulate the partial plasticity with
The third term in the right hand side of Eq.(40), rigidity degradation in a semi-rigid connection.
is easily calculated by assembling the derivatives This particular model with all its circumstances
of the stiffness matrix of all elements with respect was exactly used for sensitivity analysis. Obvi-
to any design variable. This is because for any ously the outcome of the sensitivity analysis is
specific amount of deformation, the tangential consistent with the basic assumptions made.
stiffness matrix can be expressed explicitly, and However, in Habibis nonlinear analysis, a spread
its derivatives can be obtained accordingly. plasticity model was adopted for consideration
Summing up the derivatives of displacements of nonlinear flexibility and nonlinear stiffness
in all load segments, results the derivative of total characteristics of a concrete beam column. There
displacement in a nonlinear force-displacement is no doubt that the corresponding sensitivity
environment. analysis procedure is fundamentally different from
Gongs method. Another difference between the
Discussion on Sensitivity Analyses two methods returns to the method of natural
period evaluation. Gong used a Rayleigh method
In previous sub-sections, two distinctive sensitiv- for determination of natural period of the structure
ity analyses were briefly described. Here in this while Habibi used the analytical method for
section some similarities and dissimilarities of the evaluation of natural period. Of course, the sen-
two methods are discussed to suggest any future sitivity calculations for T di are remarkably
research activities in this field. different.
A common point in these two methods was the As a conclusion, the nonlinear sensitivity
fact that the formulation of nonlinear sensitivity analysis formulation has to be built based on
analysis in both methods was established based adopted nonlinear analysis procedure; otherwise,
on the nonlinear procedures that were followed the results may not be consistent and may result
for nonlinear analyses. For example, Gong who in difficulty in convergence of the optimization
used an enhanced incremental method similar algorithm. The similarities and dissimilarities of
194
the sensitivity procedures may be interpreted in pattern. The load pattern can be different for
the context of nonlinear analysis procedure. various levels of risks. It also can be different in
a single risk level for different directions. The
lateral earthquake load pattern can be either uni-
OPTIMIZATION IN PERFORMANCE- form, triangular, exponential, or etc.; FEMA356
BASED DESIGN recommends the exponential form. To be in the
safe side one may use more than one or two load
To optimally design a structure, as was already patterns. n , n and ns are number of stories,
mentioned, the design problem should be ex- number of sections that the plastic rotation has to
pressed mathematically in terms of design vari- be controlled, and number of sections for strength
ables in the framework of a standard optimization check, respectively. and are inter-story drift
problem, Eq.(15). To express design constraints and its allowable upper limit. Similarly, and
in its explicit form, Eq.(16) is employed. The are plastic rotation and its allowable upper bound.
derivatives in this equation are obtained from F and S are internal force and strength respec-
sensitivity analysis. Some examples of sensitiv- tively. DI stands for damage index for any load
ity analysis were discussed. In this section, the pattern and performance level; damage index
general formulation of OPBSD is presented and p
should be less than the pre-specified value DI .
some, not all, published research works in the
The constraint group C1 and C2 are the main
field will be described.
constraints in PBSD; they control the drift and
plastic rotation. The constraint group C3 are
General OPBSD Problem
strength constraints and for higher performance
levels, may be considered as secondary perfor-
For optimal Performance-based seismic design,
mance constraints. The C4 set of constraints are
the general standard optimization problem can be
not used so much. They are optional and stand
written in the form seen in Box 1.
for limitations on damage index. One may adopt
In the above formulation, Z is the objective or
different upper bounds for different performance
cost function. C1 to C5 are groups of constraints.
levels on a certain damage index. The C5 group of
The superscript p relates to the performance
constraints are lower bound and upper bounds for
level and np is the number of performance levels.
design variables. Obviously some other constraints
If it is to design a structure for four performance
can be added to the optimization problem. For
levels, np=4. The superscript lp denotes the load
Box 1.
Minimize Z = C (X )
Subject to C1 : lp p
p o
s (s = 1, 2, ...n ) ; (p = 1,...,nnp) ;(lp = 1,...,nlp)
C2 : lp p
r rp o (r = 1, 2, n ) (41)
C3 : lp p
i
p
Fi S o (i = 1, 2, ...ns )
p
lp
C4 : DI p DI o
C5 : x lj x j o, x j x uj o (j = 1, 2, ...nx )
195
example if it is desired to provide some constraint earthquake. The second phase involved mini-
on protection of equipments and contents in a floor, mizing the cost of steel reinforcement subject to
some constraints such as limits on velocity or ac- constraints on inelastic displacements. Pushover
celeration of floors have to be added to the design analysis was performed based on the assumption
optimization problem. Evidently the formulation that the fundamental mode of vibration was the
of these particular constraints requires employing predominate response and did not change during
a nonlinear dynamic analysis. It is noted that the nonlinear behavior.
C1, C2, C3 and C4 group of constraints may be The details of the first phase that produces the
different for various levels of performances, and optimal dimensions of members for minor wind
they all can enter into the optimization problem. or earthquake loadings was previously addressed
The difference between research works on the by Moharrami (1993) and is not the concern of
OPBSD is a consequence of their assumption on this chapter. The detail of the second phase, as
the objective function, constraints and the solu- it is addressed by Zou and Chan, is presented
tion procedure. Some examples of recent research hereunder. To establish a nonlinear analysis pro-
works in this ground that utilize the classical cedure, the following assumptions were made for
optimization algorithms are discussed hereunder. the structural model of reinforced concrete frame.
Many other researches have been excluded not
because of their scientific value, but because of 1. All inelastic deformations occur at the plastic
their weak relation to the subject of this chapter. For hinges, which are located at the ends of each
example, Ganzerli et al. (2000), who were prob- frame member and, members are perfectly
ably the first that incorporated pushover analysis elastic between the plastic hinges.
and the performance-based design concept, used 2. Beam column connections are rigid zone,
the idea of OPBSD for a one story one bay R/C and the plastic hinges are assumed to be
frame with pushover analysis. They minimized the frictionless and have zero length.
cost of R/C frame including costs of concrete and
reinforcements under plastic rotation constraints As a requirement for the solution of nonlinear
at the ends of members. They considered perfor- equations of equilibrium, Eq.(14), the authors
mance constraints at immediate occupancy, and expressed the pushover displacement in two parts
checked for the satisfaction of design conditions i.e. elastic and plastic (inelastic) displacements.
in other levels. However, in their publication,
there was no sensitivity analysis and no explicit j = j ,e + j ,p (42)
design constraints. Therefore, their work is not
presented here in detail.
In Eq.(42), j stands for the story number. Us-
ing the principle of virtual load method, j,e and
Example 1
j,p can be obtained from Eq.(43) and Eq.(44)
respectively. It is to be noted that fs and ms are
The procedure of OPBSD of reinforced concrete
internal forces and moments due to application
frames suggested by Zou and Chan (2005) is a
of a unit load in the direction in question, and Fs
good example to be presented here. These au-
and Ms are internal forces due to applied loads.
thors decomposed the optimal design process
into two single-criterion phases. The first phase Nj
Li FX fX FY fYj FZ fZ j M X mX j MY myj M Z mzj
involved an elastic design optimization in which = j
+ + + + + dx
j ,e
i =1
0 EA
GA GA GI EI EI X
X Y Z X Y
i
the cost of concrete is minimized subject to elastic (43)
spectral displacement constraints due to a minor
196
Ni
2 0
= m pjh ph (44)
j ,p
i =1 h =1

i
M y = 0.5 fcBJd (Jd 3 d ') + fy Bd(d d ') ;

Zou and Chan (2005) also adopted the follow- d '
nsc2 + 2 + ' nsc ( + ') nsc
2
J = ( + ')
ing assumptions for relating the plastic rotation d
to the moment and reinforcements in the beam (46)
and column. 2
They assumed a bilinear M- relation similar to fy BD D Jd J 2 fy Bd (d d ')
My = + ;
2nsc 2 3 1J 2d 1J
the one shown in Figure (7); accordingly a linear
2
F F D F
relation, as in Eq.(45), can be found between p, the J = 2 + X nsc + 2 + X nsc2 + 2nscc + X

fy Bd
fy Bd d fy Bd
plastic rotation, and the moment in excess of My.
(47)
M M
p = PU
y
(45)
MU M y
in which nsc is the ratio of modules of elasticity of
steel, Es and concrete Ec. and are the ratios
Zou and Chan assumed MU = 1.1M y and of reinforcements in tensile and compression zone
could write p as a function of M of concrete beam. Zou and Chan (2005) assumed
Zou and Chan (2005) used an explicit rela- that and are linearly dependent to each other.
tionship between My and reinforcement ratios They were assumed equal in columns.
and for beams, Eq.(46), and beam-columns,
Eq.(47), as follows: They expressed the relationship between plastic
rotation and reinforcement ratios by second order
Taylor series expansion as follows:
p ( 0 ) 2 0
1 p ( )
p () = p ( 0 ) + ( 0 ) + ( 0 )2
2 2
(48)
Figure 7. A bilinear M- relation To find the derivatives of p with respect to

design variables it is noted that Eqs.(46 and 47
provide explicit relation between M and . There-
fore, using Eq.(45) p can be expressed in terms
of and the derivatives can be easily obtained.
Substituting p from Eq.(48) into Eq.(44) and
utilizing chain rule in differentiating, produces an
explicit relation between pushover displacement
and reinforcement ratios.
j (i ) = j (i0 ) +
Ni
j (i0 ) 2 0
1 N i j (i )
(i i0 ) + (i i0 )2

2
i =1 i 2 i =1 i
(49)
197
Figure 8. The ten story reinforced concrete frame,

Upon establishing the explicit formulation of
Example 1
the inelastic displacement, Eq. (49), Zou and Chan
could explicitly write the optimization problem
of minimizing the steel construction cost of a
multistory RC building with drift constraints in
terms of the design variables, i.
In Zou and Chans study (2005), since the
dimensions of frame elements were fixed, the
objective function of the OPBSD problem Ws was
the volume or weight of reinforcements. The drift
constraints were written in terms of design vari-
ables, , in the form of quadratic functions at the
target displacement. The design optimization
problem became in Box 2. Where dUj is the upper
limit on story drift and 1i and 2i are first and
second derivatives of differential displacement
j with respect to . They are expressed in terms
of two end moments and first and second deriva-
tives of plastic hinge rotations, , with respect to
. (See Zou and Chan 2005).
The above optimization problem was solved
utilizing Optimality Criteria method for a ten story
2D frame of Figure (8), in which, the concrete
cylinder strength and steel reinforcement yield
strength, fy, for all members were assumed to be
20 and 335 MPa, respectively.
The problem was solved for two different
gravity loads, to show the effect of axial force on
the pushover performance and capacity of frame-
work. The followings have been reported as the
outcome of design optimization process. loads on tensile reinforcement of columns. After
In the load Case A, the gravity is 30 kN/m while design optimization, there were 45 plastic hinges
that of Case B is 10 kN/m. The results show that, in Case A while there were 55 plastic hinges in
as expected, more reinforcements are required for Case B. The distribution of plastic hinges was
the beams in Case A relative to Case B. This is albeit different in the two Cases. Overall, the
because the gravity forces have more influence on reinforcement in Case B was more than Case A.
bending moment of beams compared to columns.
For columns, the situation is somehow differ- Example 2
ent. Most of columns in Case B required more
reinforcement in comparison to Case A and there As the second example consider the three story
were more plastic hinges in columns of Case B 2D frame in Figure (9) that was studied by Xu
compared to A. This might be because the gravity et al. (2006).
forces have a reverse effect compared to lateral
198
Box 2.
Ni
Minimize F (i ) = W si i
i =1
1 Ni Ni
2 (50)
Subject to : g j (i ) = j ( 0 ) + 1i (i i0 ) + 1 2i (i i0 ) dUj j = 1, 2, 3, .....N j

hj 2 i =1
i =1
iL i Ui i = 1, 2, 3, ....N i
To optimize the design of this frame under used one set of constraints, C1, in Eq.(41). The
applied earthquake load, Xu et al. (2006) consid- constraints of group C3 as well as C5, were
ered a combination of two objective functions for treated as side constraints; i.e. they were checked
this example. The first is F1, a measure of weight after analysis; if they were violated, the design
of the structure, and the second is F2, a measure was scaled via a proper way.
of uniformity of distribution of relative drift (see The pushover analysis of Hasan (2002) was
Box 3). used for determining the displacements under ap-
In Eq.(51), and H stand for lateral displace- plied earthquake load at IO, LS and CP levels. To
ment and height, respectively; r and s represent express the objective function explicitly in terms
the roof and story; AiU is the upper bound of area of design variables, as a requirement of Eq.(16),
Ai of member i. 1 and 2 are named combination the derivative of F2 had to be calculated using a
factors, and for this example they were taken proper sensitivity analysis technique. Xu et al.
1=0.95 and 2=0.05. The first objective function, (2006) realized that a better approximation can be
F1, is an explicit function of design variables, A, obtained using reciprocal variables. Accordingly,
while the second objective function, F2, is an they reformulated the problem it terms of recip-
implicit function of design variables. Xu et al. rocal variables and used the sensitivity analysis
technique of Gong (2003), i.e. Eq.(30) for the
Figure 9. A three story four bay moment frame, (Xu et al. 2006)
199
Box 3.
L A i i
1 ns 1 H
2
s s 1 (51)
F = 1F1 + 2F2 ; F1 (x ) = i =1
n
; F2 (x ) =
ns
H

L A
i =1
i
U
i
s =1 r r
sensitivity calculations. Having the optimization Table 4 shows how the drift ratio for different
problem fully defined, it was solved by the Dual performance levels varied during the optimization
method by Fleury (1983). Table 3 shows some history.
of the history of the change in design variables The most critical constraint was satisfied with
during the optimization process. a response ratio of 1. The plastic states of the
optimal design corresponding to the LS and CP
Table 3. Example 2: primary design variable results (Xu et al. 2006)
Cycle A1 A2 A3 A4 A5 A6 A7 A8 A9
index (mm )2
(mm ) 2
(mm ) 2
(mm )2
(mm ) 2
(mm )
2
(mm )
2
(mm )
2
(mm2)
1 152.900 152.900 152.900 152.900 152.900 152.900 67.100 90.320 92.900
4 67.440 49.605 42.595 74.970 71.835 59.624 35.095 37.093 26.600
7 55.460 38.210 32.100 64.115 63.030 49.960 30.985 30.465 19.880
10 50.010 34.460 29.860 57.820 61.250 49.850 29.825 27.680 17.870
Discrete 48.770 33.420 33.420 58.970 65.160 48.770 31.930 28.060 17.900
Design W14257 W14176 W14176 W14311 W14342 W14257 W33169 W30148 W2494
Table 4. Design history for critical drift, objective functions f1 and F2 (Xu et al. 2006)
Design /H Overall (roof) drift Structural Ductility

Cycles S/hS Critical story drift Weight/Wmax Demand
IO LS CP (f1) (f2)
1 { 0.0020 0.0027 0.0042 1.000 0.067
0.0024 [2]
0.0033 [2] 0.0051 [2]
4 { 0.0058 0.0078 0.0162 0.402 0.045
0.0067 [2] 0.0089 [2] 0.0188 [2]
7 { 0.0067 0.0087 0.0305 0.333 0.037
0.0075 [2] 0.0098 [2] 00349 [2]
10 { 0.0070 0.0091 0.0476 0.311 0.026
0.0077 [3] 0.0101 [3] 0.0510 [2]
Discrete design 0.0069 0.0090 0.0456 0.319 0.022
0.0078 [3] 0.0102 [3] 0.0481 [2]

The numbers in brackets show the critical story.
200
Figure 10. Structural plastification: A) At LS level, B) At CP level, (Xu et al. 2006)
performance levels _i.e., 2.5% and 5% roof drift computer program utilizes the CONLIN optimi-
can be compared in Figure 10A and figure 10B, zation algorithm with some modifications made.
respectively. Habibi optimized the example 1, with
CONOPT software, for several design assump-
Example 3 tions. In his design optimization formulation, the
dimensions of beams, columns as well as rein-
Habibi (2008) whose method of nonlinear analysis forcement ratios, were chosen as design variables;
and nonlinear sensitivity analysis were discussed i.e. he nearly doubled the design variables and
earlier, formulated the OPBSD of reinforced significantly increased the number of design
concrete frame of example 1, a ten story two bay constraints because of considering almost all
RC frame, in the form of the standard design groups of constraints in Eq.(41). He even entered
optimization problem of Eq.(41). He wrote his the voluntary constraints of damage index into
own optimization software named CONOPT. This the design optimization problem. He used the
stiffness degradation index (DI ) = 1 (k j / kOP )
201
suggested by Ghobarah (1999) as a damage index. dently the target point displaces from the initial
He suggested and applied different damage indi- point. Corresponding curves to Zou and Chan is
ces for different performance levels. To investigate shown in Figure 11-a while that of Habibi is in
the effectiveness of design optimization algorithm, Figure 11-b.
Habibi first assumed the same dimensions for This figure shows that in both solutions, the
frame elements that Zou and Chan had considered spectral accelerations and consequently the base
in their second phase optimization. The results shears have increased and corresponding target
were compared to the results of Zou-Chan and point or spectral displacements have been de-
Asadi (2006) who optimized the same RC frame. creased. In the automated performance design, it
It worth to mention that Asadi sought the uni- is possible to impose any constraint on behavior
form inter-story drift for his measure of optimality of the structure. Here in this example the drift has
of design, and similar to Zou and Chan assumed been limited to 1% in both solution strategies. In
that distribution of rebars in R/C frames has a both studies while the initial designs pertained to
substantial effect on the drift control. Therefore, more expensive cost, they included some vio-
considering fixed values for dimensions of frame lated constraint but the optimal designs not only
members and variable reinforcement, he indeed reduced the cost, but also satisfied all drift con-
chose the same design variables as Zou and Chans straint. In fact, this is the advantage of automated
but followed different design objective. optimum design that reduces the cost while satis-
As an alternative design optimization, Habibi fies user-defined conditions/constraints.
considered variable dimensions and considered
the cost of formwork in addition to the cost of
concrete and reinforcement. As a general rule, FUTURE RESEARCH DIRECTIONS
when the number of design variables increases, a
more optimal design can be achieved. On the other As was stated earlier, the trend of design phi-
hand, when the number of constraints increase, losophy is towards performance-based design.
the cost function increases. Obviously Habibi The second and third examples, especially the
could reduce the cost of material used in trade comparison of the results from the first and the
of more computational effort. Table 5 shows the third example, clearly showed that to design a
comparison of the results. structure for several performance levels simulta-
To see what happens to performance point, the neously, there is no way other than optimization.
demand and capacity spectra have to be drawn In fact, optimization puts the design problem in
after completion of design optimization. Evi- a mathematical framework and solves the design
Table 5. Cost of initial and optimal design of ten story frame
Cost basis (US$) Initial Optimal Design

Design
OC UD CONOPT CONOPT
(Fixed Dimensions) (Variable Dimensions)
Reinforcement 3386 3956 3204 2972 2333
Concrete +Formwork 3558 3558 3558 3558 3382
Total 6994 7515 6762 6530 5722
Optimality Criteria algorithm used by Zou and Chan (2005)
Uniform Ductility distribution criterion by Asadi (2006)
202
Figure 11. The initial and final (optimal) target point. a)Zou and Chan(2005), used with permission,
b)Habibi
problem mathematically. When we need simul- The OPBSD has a bright future. This method-
taneous solution of a number of equalities and ology can be used as a powerful robust tool for
inequalities one way is to satisfy equalities and scientist engineers, to revise their suggestions on
check for inequalities. This is the common practice acceptance criteria based on correlations that may
in design offices. The design engineers satisfy be discovered between several aspects of opti-
equilibrium equalities and check for acceptance mally designed structures. For example, a design
criteria inequalities for a structure. Another way engineer may have queries on the specification
to solve the set of equalities and inequalities is of a good design so as to have proper energy dis-
to get assisted by mathematics. The design opti- sipation, limited plastic rotations in connections,
mization is this second way. Design optimization controlled relative drift, comfort of occupants, etc.
helps the designer to design a structure in such These questions may be answered by a massive
a way that not only satisfies the equilibrium and research on different types of optimally designed
acceptance criteria but also reduces the cost. In structure including different objectives and con-
this methodology, the designer is allowed to put all straints. This tool may be also used to discover
self-interested conditions among the constraints. the controversy of different outcomes for different
Albeit, the constraint limits require experience to design criteria.
be determined realistically. It is the responsibility Preparing the ground for application of OPBSD
of all Engineering faculties to consider the design can be followed in several directions:
optimization as a highly vital subject and put it
among the obligatory courses. This will help more A. Enhancing the knowledge on the material
understanding about this subject and provides the nonlinearity models for better predicting
ground for more developments in the field. the response of a structure.
203
B. Improving the nonlinear analysis and cor- such that he/she could begin to write his own
responding sensitivity analysis procedures. computer program for OPBSD. The start points
C. Developing more efficient optimization for different parts of an OPBSD program includ-
algorithms. ing the knowledge for starting performance-based
D. Rationalizing acceptance criteria and corre- design, nonlinear analysis, formulation of design
lating them to structures overall behaviour optimization problem and sensitivity analysis were
with the help of OPBSD. clearly introduced. The algorithms of optimiza-
E. Developing optimal performance-based tion, although were particularly relevant to the
seismic design criteria applicable in com- subject of this chapter, were not explained much
bination with nonlinear dynamic analysis because of shortage of space; however, some use-
F. Establishing relations between expected ful references were introduced. If the interested
values of the response and performance reader establishes his own performance-based
indicators, estimated by means of refined design optimization problem, its solution may
models of the system with those estimated be facilitated by one of available optimization
with the aid of structural analysis and design software.
ordinarily applied in typical structural design
practice.
REFERENCES
CONCLUSION Arora, J. S. (2004). Introduction to optimum de-

sign. San Diego, CA: Elsevier Academic Press.
In this chapter, performance-based design and Arora, J. S., & Haug, E. J. (1979). Methods
its prerequisites were briefly introduced. Due to of design sensitivity analysis in structural op-
shortage of space, and wideness of the field, we timization. AIAA Journal, 17(9), 970974.
could only point to titles of the related subjects. doi:10.2514/3.61260
However, this chapter provides a ground for
more deep studies in the subject for the interested Asadi, P. (2006) Optimization of distribution of
reader. The performance-based design philosophy reinforcement in the seismic design of R/C frames.
and corresponding practical procedures were MSc. Thesis Civil Eng. Dept., Sharif University
explained on the whole, to provide an overview of Technology.
of what is going on in design offices. The goal of
ASCE. (2007). Seismic rehabilitation of exist-
design optimization and its application to some
ing buildings (pp. 4106). Reston, VA: ASCE
design problems was briefly presented. The
Publications.
sensitivity calculation that is an essential task in
practical design optimization was explained. It Aschheim, M. (1999). Yield point spectra: A sim-
was shown how nonlinear sensitivity calculation ple alternative to the capacity spectrum method.
has to be built on nonlinear analysis assumptions SEAOC 1999 Convention, (pp. 373-379).
and procedure. Some examples of the research
Besset, S., & Jezequel, L. (2007). Optimization
activities in the field of Optimum Performance-
of structural dynamic behaviour based on effec-
Based Seismic Design were presented, and various
tive modal parameters. International Journal
aspects of this progressing desire were discussed
for Numerical Methods in Engineering, 70(5),
within the description of examples. The aim of
523542. doi:10.1002/nme.1890
the author was to help the interested reader to
obtain an overall understanding of the subject,
204
Chen, J., Che, J., Sun, H., Ma, H., & Cui, M. (2002). Fragiadakis, M., & Lagaros, N. D. (2011). An
Structural dynamic optimization with probability overview to structural seismic design optimisation
constraints of frequency and mode. Structural frameworks. Computers & Structures, 89(11-12),
Engineering & Mechanics, 13(5), 479490. 11551165. doi:10.1016/j.compstruc.2010.10.021
Cornell, A. C. (2008). Uncertainty propagation Frangopol, D. M. (2011). Life-cycle perfor-
in probabilistic seismic loss estimation. Struc- mance, management, and optimisation of
tural Safety, 30(3), 236252. doi:10.1016/j. structural systems under uncertainty: Ac-
strusafe.2006.11.003 complishments and challenges. Structure and
Infrastructure Engineering, 7(6), 389413.
Crisfield, M. A. (2000). Non-linear finite element
doi:10.1080/15732471003594427
analysis of solids and structures, Vol. 1: Essen-
tials; Vol. 2: Advanced topics. New York, NY: Ganzerli, S., Pantelides, C. P., & Reaveley, L. D.
John Wiley & Sons. (2000). Performance-based design using structural
optimization. Earthquake Engineering & Structural
De Donato, O. (1977). Fundamentals of elastic-
Dynamics, 29(11), 16771690. doi:10.1002/1096-
plastic analysis. In Cohn, M. Z., & Maier, G.
9845(200011)29:11<1677::AID-
(Eds.), Engineering plasticity by mathematical
EQE986>3.0.CO;2-N
programming (pp. 325349). New York, NY:
Pergamon Press. Ghobarah, A. (2001). Performance-based design
in earthquake engineering: State of develop-
Esteva, L., Lpez, O. D., & Hernndez, E. I.
ment. Journal of Engineering Structures, 23(8),
(2010). Seismic vulnerability functions of multi-
878884. doi:10.1016/S0141-0296(01)00036-0
story buildings: estimation and applications.
Structure and Infrastructure Engineering Special Ghobarah, A., Abou-Elfath, H., & Biddah, A.
Issue: Vulnerability Assessment of Structures and (1999). Response-based damage assessment of
Infrastructures, 6(1-2), 316. structures. Earthquake Engineering & Struc-
tural Dynamics, 28(1), 79104. doi:10.1002/
Fajfar, P. (2000). A nonlinear analysis method for
(SICI)1096-9845(199901)28:1<79::AID-
performance-based seismic design. Earthquake
EQE805>3.0.CO;2-J
Spectra, 16(3), 573599. doi:10.1193/1.1586128
Gong, Y. (2003). Performance-based design of
FEMA. 356. (2000). Prestandard and Commen-
steel building frameworks under seismic loading.
tary for the seismic rehabilitation of buildings.
PhD Thesis, University of Waterloo, Ontario,
Reston, VA: ASCE Publications.
Canada, 2003.
Fleury, C. (1983). Dual methods for optimizing
Habibi, A. (2008). Optimal seismic performance-
finite element flexural systems. Computer Meth-
based design of 2D reinforced concrete frames.
ods in Applied Mechanics and Engineering, 37(3),
Doctoral dissertation, Tarbiat Modarres Univer-
249275. doi:10.1016/0045-7825(83)90078-6
sity, Tehran, Iran.
Fleury, C. (1989). CONLIN: An efficient dual op-
Habibi, A., & Moharrami, H. (2010). Nonlinear
timizer based on convex approximation concepts.
sensitivity analysis of reinforced concrete frames.
Structural Optimization, 1(2), 8189. doi:10.1007/
Finite Elements in Analysis and Design, 46(7),
BF01637664
571584. doi:10.1016/j.finel.2010.02.005
205
Haftka, R. T., & Gurdal, Z. (1992). Elements of Moharrami, H. (2006). Optimality criteria: A
structural optimization. Norwell, MA: Kluwer robust nonlinear optimization algorithm. The
Academic Publishers. 7th International Congress on Civil Engineering,
Tarbiat Modarres University, Tehran, Iran.
Hasan, R., Xu, L., & Grierson, D. E. (2002). Push-
over analysis for performance-based seismic de- Moharrami, H., & Alavinasab, A. (2006). An
sign. Computers & Structures, 80(31), 24832493. optimization procedure for automated design
doi:10.1016/S0045-7949(02)00212-2 of seismic resistant steel frames. International
Journal of Civil Engineering, 4(2), 86105.
Jalayer, F., & Cornell, C. A. (2009). Alterna-
tive non-linear demand estimation methods for Owen, D. R. J., & Hinton, E. (1980). Finite ele-
probability-based seismic assessments. Earth- ments in plasticity- Theory and practice. Swansea,
quake Engineering & Structural Dynamics, 38(8), UK: Pineridge Press.
951972. doi:10.1002/eqe.876
Park, G. J., & Kang, B. S. (2003). Validation of
Kim, Y.-I., & Park, G. J. (2010). Nonlinear dy- a structural optimization algorithm transforming
namic response structural optimization using dynamic loads into equivalent static loads. Journal
equivalent static loads. Computer Methods in of Optimization Theory and Applications, 118(1),
Applied Mechanics and Engineering, 199(9-12), 191200. doi:10.1023/A:1024799727258
660676. doi:10.1016/j.cma.2009.10.014
Park, K. J., Lee, J. N., & Park, G. J. (2005).
Lagaros, N. D., Garavelas, A. T., & Papadrakakis, Structural shape optimization using equivalent
M. (2008). Innovative seismic design optimization static loads transformed from dynamic loads.
with reliability constraints. Computer Methods International Journal for Numerical Methods
in Applied Mechanics and Engineering, 198(1), in Engineering, 63(4), 589602. doi:10.1002/
2841. doi:10.1016/j.cma.2007.12.025 nme.1295
Masson, G., Cogan, S., Bouhaddi, N., Lombard, Park, Y. J., & Ang, A. H. S. (1985). Mecha-
J. P., & Bonini, J. (2002). Parameterized reduced nistic seismic damage model for reinforced
models for efficient optimization of structural concrete. Journal of Structural Engineering,
dynamic behavior. Collection of Technical Pa- 111(4), 722739. doi:10.1061/(ASCE)0733-
pers - AIAA/ASME/ASCE/AHS/ASC Structures, 9445(1985)111:4(722)
Structural Dynamics and Materials Conference,
Park, Y. J., Reinhorn, A. M., & Kunnath, S. K.
(pp. 1434-1440).
(1987). IDARC: Inelastic damage analysis of
Mills-Curran, W. C., & Schmit, L. A. (1985). frame sheare-wall structures. Technical Report
Structural optimization with dynamic behaviour NCEER-87-0008, National Center for Earthquake
constraints. AIAA Journal, 23(1), 132138. Engineering Research, State University of New
doi:10.2514/3.8881 York at Buffalo.
Moharrami, H. (1993). Design optimization of Priestley, M. J. N., Calvi, G. M., & Kowalsky, M.
reinforced concrete frameworks. Ph.D. Thesis, The J. (2007). Displacement-based seismic design of
University of Waterloo, Waterloo, Ontario, Canada structures. Pavia, Italy: IUSS Press.
206
Saka, M. P. (1997). The theorems of structural Xu, L., Gong, Y., & Grierson, D. E. (2006).
variation for grillage systems. In Topping, B. H. Seismic design optimization of steel building
V., & Leeming, M. B. (Eds.), Innovation in com- frameworks. Journal of Structural Engineering,
puter methods for civil and structural engineering 132(2), 277286. doi:10.1061/(ASCE)0733-
(pp. 101111). Edinburgh, UK: Civil Comp Press. 9445(2006)132:2(277)
doi:10.4203/ccp.50.8.2
Zou, X. K., & Chan, C. M. (2005). Optimal seismic
Tangaramvong, S., & Tin Loi, F. (2011). Math- performance-based design of reinforced concrete
ematical programming approaches for the safety buildings using nonlinear pushover analysis.
assessment of semirigid elastoplastic frames. In- Engineering Structures, 27(8), 12891302.
ternational Journal of Solids and Structures, 48(6), doi:10.1016/j.engstruct.2005.04.001
10111023. doi:10.1016/j.ijsolstr.2010.12.003
Vanderplaats, G. M. (1997). DOT users manual.
Colorado Springs, CO: VMA. ENDNOTE
1
Federal Emergency Management Agency
(FEMA) of USA
207
208
Chapter 9
Optimal Seismic Performance-
Based Design of Reinforced
Concrete Buildings
Xiao-Kang Zou
AECOM Asia Company Ltd., Hong Kong
ABSTRACT
In order to meet the emerging trend of the performance-based design approach and to improve the design
efficiency, this chapter presents a numerical optimization technique for both minimum material cost and
life-cycle cost design of building structures subject to multiple levels of linear elastic and nonlinear elastic
seismic drift performance design constraints. This chapter firstly introduces an elastic seismic drift design
of reinforced concrete (RC) building structures based on elastic response spectrum analysis method;
and then presents the inelastic design optimization based on the nonlinear pushover analysis method.
Finally, the optimal seismic performance-based design of RC buildings is posed as a multi-objective
optimization problem in which the life-cycle cost of a building is to be minimised subject to multiple
levels of seismic performance design criteria. The computer based optimization methodology developed
provides a powerful numerical design tool for performance-based design of RC building structures.
1. INTRODUCTION satisfactory. The acceptability checking proce-

dures may employ various linear or nonlinear
The concept of performance-based design ap- analysis methods to assess the seismic responses
pears to be the future direction of seismic design of structures in relation to the acceptable design
codes (SEAOC 1995; ATC-40 1996; FEMA criteria. Response spectrum analysis, which is
356 2000; FEMA 440 2005; FEMA 445 2006; one of the most common linear elastic methods,
ASCE41 2007). According to the newly developed provides designers with a simple but rational
performance-based seismic design approach, an basis for determining the responses of structures
acceptability analysis needs to be conducted at under minor or moderate earthquake loading.
various design load levels in order to ensure that In the advent of advanced computational tech-
the corresponding performance objectives are niques, nonlinear analysis procedures become
DOI: 10.4018/978-1-4666-1640-0.ch009
Optimal Seismic Performance-Based Design of Reinforced Concrete Buildings
significantly necessary to identify the pattern and (Charney 2000; Foley 2002). Zou (2002), Chan
level of damage and to understand the modes of and Zou (2004), Zou and Chan (2001, 2005a,
failure of structures during severe seismic events. b), Zou et al (2007a, b), Zou (2008), Wang et al
In assessing the nonlinear seismic behaviour of (2010) and Zou et al (2010) had been working at
framework structures, pushover analysis has the performance-based seismic design optimi-
provided an effective means for distinguishing zation of RC building structure and developed
between good and bad seismic performance of an effective method for design optimization of
structures (Krawinker 1994). buildings subject to seismic elastic and inelastic
It has been recognized that the displacement or seismic drift performance criteria.
lateral drift performance of a multi-story building The traditional prescriptive design optimiza-
can be a good measure of structural and non- tion based on linear elastic techniques has been
structural damage of the building under various researched for many years. One major drawback of
levels of earthquake motions (Moehle and Mahin these elastic techniques is that it does not directly
1991). The performance-based seismic design address structural inelastic seismic responses
provisions for multi-story buildings can be based and thus cannot effectively deal with damage
upon controlling story drifts to prescribed limit loss due to structural and non-structural failure
states under different design levels of earthquakes. during earthquakes. As a result, the long-term
Lateral drift design requires the consideration of risk and benefit implications cannot be assessed
a proper distribution of the stiffness of all struc- using the traditional linear elastic design method.
tural elements and, in a severe seismic event, also In seismic performance-based design, the design
the occurrence and redistribution of plasticity in objective function may consist of two main parts:
structure elements. initial construction cost and expected future failure
Numerous studies on structural optimization loss caused by earthquakes (SEAOC 1995; Foley
in the seismic design of structures have been pub- 2002). The final design should be established
lished in the past two decades, including Cheng considering a good balance between the initial
and Botkin (1976), Feng et al. (1977), Bhatti and structural cost and its loss expectation in the design
Pister (1981), Balling et al. (1983), Cheng and life period. Due to the fact that the life-cycle cost
Truman (1982), Arora (1999). However, most of involves the consideration of construction cost and
these previous research efforts were concerned damage loss which are inherently conflicting, the
with optimization through prescriptive-based minimum life-cycle cost design can be viewed as a
design concepts. Recently, Beck and his associates multi-objective optimization problem. Given with
(1998) developed an optimization methodology the initial conditions of a building project, it is,
for performance-based design of structural sys- in general, a relatively easy task to estimate the
tems operating in an uncertain dynamic environ- construction cost of the building. However, the
ment. Foley (2002) provided a comprehensive estimation of the seismic damage to a building is
literature review of current state-of-the-art such a complex problem that it involves not only
seismic performance-based design procedures engineering cost but also costs associated with
and presented a vision for the development of social, economic, political, cultural and ethical
performance-based design optimization. It has aspects. A number of different damage loss models,
been recognized that there is a pressing need for established by numerous researchers (Park and
developing optimized performance-based design Ang 1985; Gao and Bao 1985; Park et al. 1987;
procedures for seismic engineering of structures Ang et al. 1997; Wen and Kang 1997, 2001), in-
209
volve realistic modelling of loading and resistance 2. SINGLE OBJECTIVE DESIGN

uncertainty, initial construction cost, damage cost, OPTIMIZATION PROBLEM
failure consequence cost, maintenance cost and
discount cost of distant future failure. Cheng and In seismic design, it is commonly assumed that
Chang (1988), Cheng and Li (1997), Li (1998) a building behaves linearly elastic under minor
and Zou et al (2007b) studied the application of earthquakes and may work nonlinearly inelastic
minimum expected life-cycle cost. Recently, Liu when subjected to moderate and severe earth-
et al. (2003) presented a multi-objective genetic quakes. Under such an assumption, the entire
algorithm for optimal seismic design of steel design optimization process can therefore be de-
frames based on life cycle cost considerations. composed into two phases (Zou and Chan 2001;
All these research efforts indicate the importance Zou 2002). The first phase is an elastic design
of consideration of a life-cycle cost on making optimization in which the structural concrete cost
rational design decisions. can be minimized subject to elastic spectral drift
This chapter presents an effective single- or responses under minor earthquake loading; and
multi-objective optimization technique for the concrete member sizes are taken as design vari-
elastic and inelastic seismic drift performance ables since concrete material plays a more domi-
design of reinforced concrete buildings under nant role in improving elastic drift performance
response spectrum loading and pushover loading. of a building. In this phase, all concrete sections
Using the principle of virtual work, the modal drift are assumed to be uncracked and to behave linear-
response can be explicitly formulated in terms of elastically. Once the optimal structural member
element sizing variables and the peak drift values sizes are determined at the end of the first phase
can be estimated by modal combination methods. of the optimization, the steel reinforcement quan-
With careful tracking of the location and extent of tities can then be considered as design variables
plastic hinge occurrence, the inelastic pushover in the second phase. In controlling the inelastic
drift can also be explicitly expressed in terms of drift responses, steel reinforcement is the most
the sizing variables using the same principle of effective element that provides the ductility of
virtual work and the Taylor series approximation. RC building structure beyond first yielding (Zou
The total life-cycle cost optimization is formulated 2002). In this second design phase or the so-called
as a multi-objective optimization problem subject inelastic design optimization, the member sizes
to seismic inelastic inter-story drift responses are kept unchanged and the cost of steel reinforce-
under pushover loading. Once the multi-objective ment is minimized subject to inelastic inter-story
function (including both the initial material cost drift performance constraints based on nonlinear
and the predicted damage loss) and the design pushover analysis.
performance constraints are explicitly formulated,
the constraint method is then applied to produce 2.1 First Phase: Elastic Design
a Pareto optimal set, from which the decision of the Optimization Problem
best compromise solution for the multi-objective
design problem can be achieved. The optimization Consider a multi-story concrete framework hav-
methodology for each Pareto optimal solution is ing i= 1, 2,, Ni member (or member fabrication
basically established based on a rigorously derived groups). Assuming that the concrete elements are
Optimality Criteria (OC) approach. Finally, one uncracked and have rectangular cross sections
RC building frame example is presented to il- such that the width ( Bi ) and depth ( Di ) are
lustrate the effectiveness and applicability of the taken as design variables, the design objective of
proposed automated optimization approaches.
210
the first phase elastic optimization is to minimize Based on the modal response of a building,
the concrete material cost of the structure as ex- which is computed by commonly available engi-
pressed explicitly in terms of design variables as neering software, the nth modal elastic displace-
ment at the jth floor level, u (j n ) , can be expressed
Ni
explicitly by the principle of the virtual work in
Minimize: concrete cost f1c = wci Bi Di
i =1 Box 1.where Li is the length of member i; E and
(1) G are the axial and shear elastic material moduli;
AX, AY, and AZ are the axial and shear areas for the
where wci is the unit cost coefficient of concrete cross-section; IX, IY, and IZ are the torsional and
for member i. flexural moments of inertia for the cross-section;
The intent of the elastic drift design is to ensure FX(n ) , FY(n ) , FZ(n ) , M X(n ) , MY(n ) , and M Z(n ) are the
that a building remains operational or serviceable nth modal element internal forces and moments;
under the action of minor earthquakes. In checking fXj, fYj, fZj, mXj, mYj, and mZj are the virtual
the seismic drift response of a building, an elastic element forces and moments due to a unit of
analysis procedure can be employed and the j = virtual load applied to the building at the location
1, 2,, Nj inter-story drift should comply with corresponding to the story displacement, uj.
the following requirement: Considering rectangular concrete elements
with the width ( Bi ) and depth ( Di ) taken as
u j design variables and expressing the cross section
j ( j = 1, 2, ..., N j ) (2)
hj properties in terms of Bi and Di , the modal dis-
placement Eq. (3) can be simplified as
where u j is the elastic inter-story drift of the jth
Li F
(n ) (n ) (n )
1 X fXj + FY fYj + FZ fZj dx +
story; h j is the jth story height and j is the
E

Bi Di 0 5G / 6 i
specified inter-story drift ratio limit for the jth Ni

1 Li M (n )
Z mZj
story. u j (Bi , Di ) =
(n )
3 0
dx +
i =1 B D
E / 12

i i

Elastic linear response spectrum analysis
i

1 Li M (n )
m M (n )
m

method, widely used in modern building codes B 3D 0 G + E / 12 dx
X Xj Y Yj

i i i
such as UBC (1997) and GB5011-2001 (2001),
(4)
is adopted in this study. This method eliminates
the time variable and provides a relatively simple
denotes the torsional coefficient that depends
method for determining the maximum structural
on the ratio value of depth to width of the element
responses in which individual modal responses
i .For typical rectangular sections, it can be ap-
are first calculated and the maximum responses
proximately set to 0.2.
are then obtained by combination rules.
Once the modal story displacement is formu-
lated explicitly, the maximum value of the inter-
Box 1.
N i Li (n )
FX fXj FY(n ) fYj FZ(n ) fZj M X(n )m Xj MY(n )mYj M Z(n )mZj
u (j n ) = + + + + + dx (3)
i =1 o EAX GAY GAZ GI X EIY EI Z
i
211
story drifts can be expressed using the general where BiL and BiU , DiL and DiU correspond to
and accurate complete quadratic combination the lower and upper size bounds specified for the
(CQC) method as section width, Bi , and depth, Di , respectively.
2.2 Second Phase Inelastic

Nn Nn
u j (Bi , Di ) (u (j n ) u (j n)1 )(u (jm ) u (jm1) )
CQC = nm
n =1 m =1 Design Optimization Problem
(5)
2.2.1 Nonlinear Analysis Procedure
where N n denotes the total number of modes
considered in the response spectrum analysis ; In the newly developed performance-based seismic
u (j n1) , u (jm1) , u (jm ) are the respective nth and mth design approach, nonlinear analysis procedures
modal elastic displacements at the (j-1)th and jth become important in identifying the patterns and
floor levels; nm is a modal correlation coefficient levels of damage to assess a structures inelastic
for the nth and mth modes, which can be obtained behaviour and to understand the modes of failure
from the following equation with the constant of the structure during severe seismic events.
damping ratio, , as Pushover analysis is a simplified, static, nonlinear
procedure in which a predefined pattern of earth-
8 2 (1 + nm )nm2
3
quake loads is applied incrementally to framework
nm = 2
(6) structures until a plastic collapse mechanism is
(1 nm )2 + 4 2 nm (1 + nm )2
reached. This analysis method generally adopts a
lumped-plasticity approach that tracks the spread
where of inelasticity through the formation of nonlinear
plastic hinges at the frame elements ends during
n the incremental loading process.
nm = 1 (7)
m
As graphically presented in Figure 1, the
nonlinear static analysis procedure requires de-
Upon establishing the explicit formulation of termination of three primary elements: capacity,
the elastic inter-story drift Eqs. (4) and (5), based demand and performance. The capacity spectrum
on the response spectrum analysis method, elas- can be obtained through the pushover analysis,
tic spectral inter-story drift constraints using the which is generally produced based on the first
CQC method can be written in terms of the design mode response of the structure assuming that the
variables, Bi and Di , as fundamental mode of vibration is the predominant
Subject to: response of the structure. This pushover capacity
curve approximates how a structure behaves be-
g j (Bi , Di ) =
1
u j (Bi , Di ) 1 ( j = 1, 2, ..., N j )
yond the elastic limit under seismic loadings. The
e
hj
j demand spectrum curve is normally estimated by
(8) reducing the standard elastic 5% damped design
spectrum by the spectral reduction method. The
Besides, the member sizing constraints can intersection of the pushover capacity and demand
be defined as, spectrum curves defines the performance point
as shown in Figure 1. At the performance point,
BiL Bi BiU ; DiL Di DiU the resulting responses of the building should then
(i = 1, 2, ..., N i ) (9) be checked using certain acceptability criteria. The
212
responses can be checked against acceptability element to be used to control the inelastic perfor-
limits on both global system levels (such as the mance of the structure. Based on these consider-
lateral load stability and the inter-story drift) and ations, in the second phase, the tension steel re-
local element levels (such as the element strength inforcement ratio, i , and the compression steel
and the sectional plastic rotation) (ATC-40 1996). reinforcement ratio, i , of rectangular cross sec-
When the responses of a structure do not meet the tions are taken as design variables, for a flexural
targeted performance level, the structure needs to concrete building having i=1, 2,?, Ni members
be resized and the design process repeated until and 2Ni plastic hinges (assuming one hinge at
a solution for the desired performance level is each end of a member). The width Bi and depth
reached. In general, the determination of the satis- Di of the cross section are fixed in this phase.
factory performance response that fulfils both the If the topology of a building structural system
system level response and element level response is predefined, the design objective of the reinforced
requires a highly iterative trial-and-error design concrete framework in the second inelastic design
procedure even with the aid of todays finite ele- phase is to minimize the steel reinforcement cost
ment analysis software. f1s , which can be expressed in terms of steel re-
inforcement design variables:
2.2.2 Design Optimization Problem and
Explicit Drift Formulation Minimize: steel cost
Ni
While the concrete material plays an important f1s = wsi (Lsi i + Lsi i) (10)
i =1
role in controlling the elastic displacement re-
sponse of a RC building, the steel reinforcement
can have a significant effect on the inelastic dis- where wsi is the cost coefficient for steel reinforce-
placement and ductility of the RC building beyond ments; and Lsi and Lsi are the lengths of the
the linear elastic limit. Moreover, when an RC lower and upper steel reinforcements for member
structure works in the inelastic stage, steel rein- i. Herein, only the longitudinal reinforcement of
forcement is generally the more cost-effective member sections is considered as design variables
and the transverse reinforcement is assumed to
Figure 1. Nonlinear analysis procedure
213
be invariant in the optimization under the presup- In order to facilitate a numerical solution of
position that adequate shear strength is provided the drift design problem, it is necessary that the
for all members. implicit story drift constraint Eq. (12) be expressed
The total construction cost, f1 , of a reinforced explicitly in terms of the design variables, i and
concrete framework, which consists of the concrete i . Before the drift formulation can be discussed,
cost f1c given in Eq. (1) and the steel reinforcement three assumptions must be made. The first is that
cost f1s shown in Eq. (10), can be expressed as all the inelastic deformation is assumed to occur
at the plastic hinges, which are located at the ends
f1 (Bi , Di , i , i) = f1c + f1s = wci Bi Di Li + wsi (Lsi i + Lsi i) of each frame member and members are fully
Ni Ni
i =1 i =1
elastic between the plastic hinges. Secondly, the
(11)
plastic hinges are assumed to be frictionless and
have zero length. The third assumption is that
It has been previously shown that, in the elas-
beam-column joints are much stronger than any
tic design optimization, the concrete material cost
adjacent framing components so that the joint
of the structure with respect to the width, Bi , and
region may be modelled as a stiff or rigid zone.
depth, Di , is minimized and is kept to the mini-
Based on the internal element forces and mo-
mum value in the inelastic design optimization. ments of the structure obtained from the pushover
Under the condition that the concrete element analysis at the performance point, the principle
dimensions are fixed to their minimum values, of virtual work can be employed to express the
the steel material cost is to be minimized in the pushover displacement. The pushover story dis-
inelastic design stage. placement, u j , at the performance point includes
In seismic performance-based design, it is
the virtual work, u j ,memb , produced by the struc-
necessary to check the capacity of a structure
against the demand of an earthquake. At the tural members and the virtual work, u j ,hinge ,
performance point, where capacity equals demand generated by the plastic hinges. That is,
as shown in Figure 1, the resulting responses
of the building should then be checked using u j = u j ,memb + u j ,hinge (13)
certain acceptability criteria. In this chapter, the
inelastic drift responses at the performance point in which
of a building, generated by a severe earthquake
demand, are to be checked against appropriate N i Li F f
u j ,memb = X Xj + FY fYj + FZ fZj + M X m Xj + MY mYj + M Z mZj dx
limits corresponding to a given performance ob- i =1 o

EA
X GA Y GA Z GI X EI Y E I Z

i
jective. The inelastic interstory drift constraint at
(14)
the performance point is defined as below.
Ni
2 0
u j u j u j 1 u j ,hinge = m pjh ph (15)
= j ( j = 1, 2, ..., N j )
i =1 h =1

hj hj i
(12)
Considering rectangular concrete elements
where u j and u j 1 are the respective displacement with width ( Bi ) and depth ( Di ) and expressing
of two adjacent j and j-1 floor levels; j is the the cross sectional properties in terms of Bi and
specified inelastic inter-story drift ratio limit for Di , the displacement, u j ,memb , in Eq. (14) can be
the jth story in the inelastic design phase. simplified in terms of Bi and Di , as shown in
214
Figure 2. Moment-rotation curve

Eq.(4). During the inelastic drift design optimiza-
tion process, u j ,memb is kept unchanged since Bi
and Di of each member section are fixed. The
emphasis here is on the displacement, u j ,hinge ,
caused by the formation of the plastic hinges. In
Eq. (15), m pjh
0
is the virtual end moment at the
location of the hth hinge of a member; ph is the
actual plastic rotation experienced by the hth
plastic hinge, which is equal to zero when no
plastic hinge is found. As shown in Figure 2, the
behaviour of a plastic hinge is modelled as a bi-
linear curve: the elastic segment, AB, and the
hardening segment, BC. Based on the line seg-
ments A-B-C, the plastic rotation, p , can be
the inelastic optimization process requires a cor-
given as follows
responding update on the values of M and M y .
M My In pushover analysis, pure moment hinges as
p = Up Up (16) well as axial moment hinges are widely used and
Mu My
are generally assigned to the two ends of each
beam or column. In fact, the inelastic displace-
where Up is the ultimate plastic rotation which ment, u j ,hinge , in Eq. (13) includes the displacement
can be established based on experimental tests or generated by moment hinges (usually in the beams)
can be obtained directly from design guidelines and that generated by axial moment hinges (usu-
such as the ATC-40 (1996); M is the applied ally in the columns). By the force equilibrium
moment at the location of the plastic hinge; M y shown in Figure 3, where fc is the stress at the
is the bending moment at the first yielding of the extreme compression concrete fibre, fs is the
tensile steel; and M u is the ultimate moment of stress in the compression steel, fy is the yield
resistance. Given the quantity of the steel rein- strength of the tension steel, and d is the effective
forcement used in a concrete section, the values depth, which is equal to the distance from the
of M y and M u can then be determined. For sim- extreme compression fibre to the centroid of the
plicity, M u can be approximately related to M y tension steel, M y for a moment hinge (where there
as M u = 1.1M y (ATC-40 1996). For the explicit is no co-existent axial force) can be expressed in
problem formulation, it is necessary that the terms of design variables, and , as
plastic rotation, p , be accurately expressed in
kd
terms of the design variables (i.e., and ' ). M y = 0.5 fcBkd( d ') + fy Bd(d d ')
3
Furthermore, a good formulation should reflect (17)
accurately the change in the plastic rotation, p ,
due to a change in the design variables during the where k is the neutral axis depth factor at the first
optimization resizing process. In other words, any yield and it is given as
change in the design variables, and , during
215
Figure 3. Double reinforced member section at first yield
d' lor series approximation for evaluating the value

k = ( + )2 nsc2 + 2( + )nsc ( + )nsc
d of p is given as
(18)

2
p 1 p
p () = p = 0
+ = 0
( 0 ) + = 0
( 0 )2
in which nsc = Es Ec where Ec and Es are the 2 2
(21)
moduli of elasticity of the concrete and the steel,
respectively.
where the tension steel ratio, , is considered as
Similarly, M y for an axial-moment hinge
the major design variable; for simplicity, the
(wherein there is co-existent axial force) can be
compression steel ratio, , is assumed to be
expressed in terms of design variables, and
linearly related to for beams and to be the same
as
as for columns. Given the explicit expression
fy Bd D k d k 2 fy Bd(d d ')2
of M y as a function of from Eqs. (17) and (19),
My = +
2nsc 2 3 1k 2d 1k p
the gradient, , and the second-order term,
(19)
2p
where k is given as , can be analytically calculated from Eq.
2
2 (16).
F F D F
k = 2 + X nsc + 2 + X nsc2 + 2nsc + X By substituting the explicit plastic rotation,
fy Bd fy Bd d fy Bd
p (), given in Eq. (21) into Eq. (15), the pushover
(20) displacement, u j , in Eq. (13) can also be explic-
itly expressed in terms of the design variable, i
To take into account the change in p due to a in Box 2.
change in and while maintaining an instan-
taneously fixed value of M, a second-order Tay-
216
Box 2.
2
Ni
u j 1 Ni u j
u j (i ) = u j i =i0
+ = 0
(i i0 ) + i =i0
(i i0 )2 (22)
i =1 i i i
2 i =1 i2
2.2.3 Plastic Rotation Constraint and My M Mu (24)

Sizing Constraint
On the one hand, by setting M y = M where
Besides checking the inter-story drift responses
M y can be found from Eq. (17) for a moment
discussed above, local response quantities (i.e.,
sectional plastic rotation and strength of all mem- hinge and Eq. (19) for an axial-moment hinge,
bers) at the performance point also must not exceed the corresponding value of i can be solved and
appropriate response limits. Therefore, the plastic this value can then be taken as the instantaneous
rotation, ph , at the hth end of member i (where upper bound value of i such that Ui = i . On
the subscript h represents one end of a member the other hand, by setting M u = M and assuming
and h=1,2) should be constrained in the optimiza- M u = 1.1M y , the instantaneous lower bound
tion by L
value of i can then be found such that i = i .
As a result, based on Eq. (24), the lower and up-
ph Up (23) per bounds of i for each plastic hinge can be
instantaneously established during the OC itera-
where Up is the rotation limit of member i for a tive resizing process.
specific performance level. Once the designer
determines the performance levels of the structure 2.2.4 Design Optimization
(e.g., Immediate Occupancy, Life Safety, Collapse Problem Formulation
Prevention), the corresponding limiting value of
Up is then determined. In addition, in order to Upon establishing the explicit inelastic drift for-
mulation, Eq. (22), the optimization problem of
reduce the practical building design problem to
minimizing the steel construction cost Eq.(10)
a manageable size, the strength design of each
can be explicitly written in terms of the design
member is not considered explicitly as a design
variable, i , as
constraint; rather, the strength-based steel rein-
forcement ratios in accordance with code speci- Ni
fications are first calculated and these values are Minimize: f1s (i ) = wsi i (25)
then taken as the lower size bound for each mem- i =1
ber in the inelastic seismic drift design optimiza-

tion. Subject to:
It is found from Figure 2 that, in order to
maintain the relationship of 0 p Up , the 1 Ni
+ 1i (i i0 ) +
1 Ni
g j (i ) = p
hj
u j
i =i0 2i (i i0 )2 1
2 i =1
internal moment, M , leading to the occurrence j i =1
of a plastic hinge must satisfy the following con-

dition: ( j = 1, 2, 3, ..., N j ) (26)
217
iL i Ui (i = 1, 2, 3, ..., N i ) (27) analysis using commercially available soft-

ware such as SAP2000 (2000).
where 3. Read all the necessary input and output re-
sults of the analysis and establish the explicit
u j 2 ph elastic design optimization formulation of
1i = = m jh0 Eqs. (1) and (9).
i i =i0 h =1 i =i0
i 4. Apply the recursive OC optimization algo-
(28) rithm to resize the concrete element sizes.
5. Repeat Steps 2 and 4 for statically indeter-
2u j 2 2ph minate structures until the concrete cost of
2i = = m jh0 i =i

i =i0
the structure between two successive design
0
i2 h =1 2 i
(29) cycles converges to be within certain accept-
able criteria (say, 0.5%).
In Eq. (25), wsi is the cost coefficient for the 6. After the elastic design optimization, fix the
optimal member sizes, Bi and Di , in the
steel reinforcement, i ; jp is the allowable in-
inelastic design optimization. Based on the
elastic inter-story drift ratio. Eq. (26) defines the
member size derived from the elastic opti-
set of seismic inelastic inter-story drift perfor-
mization, determine the minimum and
mance constraints under specified earthquake
maximum size bounds of the steel reinforce-
ground motions. Eq. (27) defines the sizing con-
ment ratios, i and i , in accordance with
straints for the steel reinforcement, where iL and
the strength-based code requirements.
Ui correspond to the lower and upper size bounds
7. Carry out the nonlinear pushover analysis
specified for the tensile steel reinforcement vari-
to determine the inelastic responses of the
able, i , and they should be updated after each
structure at the performance point.
nonlinear pushover analysis. For the sake of 8. Read all the necessary input and output
simplicity, the compressive steel reinforcement, results of the pushover analysis, establish
i , has been assumed to be simply related to i the lower and upper bounds of i for the
and therefore, it is not included in the explicit members with plastic hinges using Eq. (24)
optimization problem Eqs. (25)-(29). and determine the values of the first-order
and second-order derivatives of inelastic
2.3 Design Procedure drift responses using Eqs (28) and (29).
9. Establish the explicit inelastic inter-story
The overall design optimization procedure for drift constraints using a second-order
limiting lateral elastic and inelastic drifts of an Taylor series approximation and formulate
RC building structure is listed as follows: the explicit inelastic design problem, Eqs
(25)-(29).
1. Assume the initial member sizes and deter- 10. Apply the recursive OC optimization algo-
mine the design spectra corresponding to rithm to resize all steel reinforcement design
minor and severe earthquakes. variables.
2. In the first phase, i.e., the elastic design op- 11. Repeat Steps 7 and 10 until the convergence
timization, carry out the response spectrum of the values of the steel cost objective func-
analysis of the structure subject to a minor tion and the inelastic drift design constraints
earthquake and conduct a static virtual load is achieved.
218
3. MULTIOBJECTIVE DESIGN the steel reinforcement cost f1s , as given in Eq.

OPTIMIZATION PROBLEM (11).
3.1 Objective Function 3.2 Expected Future Structural Loss f2
In seismic performance based design, the total The structural failure losses due to an earthquake
life-cycle cost of structures should be considered attack consist of direct loss and indirect loss. The
and the final design should be established based direct loss is the cost of repair or replacement of
on a good balance between the initial structural structural members, contents, non-structural
cost and its loss expectation in the design life components and equipment, and so on. The indi-
period. Due to the fact that the life-cycle cost rect loss may include the losses associated with
involves the consideration of construction cost structural malfunction, injuries and fatalities,
and damage loss which are inherently conflicting, psychological and political influence, etc. The
i.e., the decrease of one increases the other, it is accurate estimation of both direct and indirect
necessary to formulate the design optimization losses is generally a very complex task involving
as a multi-objective optimization problem. In not only engineering analysis but also many
this research study, the initial construction cost other issues. The failure of each structural perfor-
of an RC building frame is estimated simply in mance may lead to a different failure loss.
terms of the material costs of the concrete mem- Herein, Lis work (1998) is introduced as an
ber sizes and steel reinforcements. In contrast, example of derivation of the expected future
the establishment of an approximate but rational structural loss f2 for convenient discussion in the
cost function to explicitly represent the expected following. According to the work by Li (1998),
damage loss of building structures in terms of the total loss expectation can be defined by the
sizing design variables remains one of the chal- summation of the product of the occurrence prob-
lenging difficulties. ability of earthquake and the system failure loss.
If the topology of a building structural system That is, for N r performance levels, the loss ex-
is predefined, the design objective is to minimize pectation function f2 can be stated as
the multi-objective life-cycle cost of the structure.
Using F to denote the life-cycle cost function Nr
including the initial structure cost f1 and ex- f2 = Pr Lr (31)

r =1
pected future damage loss f2 , the multi-objective
function can be expressed as
where r denotes the seismic design level and
Minimize: F = { f1 , f2 } r = 1, 2, ..., N r ; Pr is the occurrence probabil-
(30)
ity of an earthquake for the rth seismic design
level, which can be determined from specified
In order to facilitate the numerical solution of code requirements; Lr is the structural failure loss
the optimization problem, the implicit objective including direct and indirect losses under the rth
function in Eq. (11) needs to be first expressed seismic design level. In Lis study (1998), the
explicitly in terms of the design variables. As damage loss of a structure corresponding to dif-
presented previously, the initial structure cost f1 ferent discrete levels of performance criteria is
of an RC framework is simply assumed to be the expressed in terms of the maximum inter-story
summation of the concrete material cost f1c and drift index under minor, moderate and severe
219
earthquakes. Five classes of structural damage 3.3 Explicit Multi-Objective Problem

(namely negligible, slight, moderate, severe and Formulation
complete damage) of a building are given based
on the maximum inter-story drift value in accor- Upon obtaining the explicit nonlinear objective
dance with the Chinese code for seismic design function (i.e., the initial structure cost f1 in Eq.
of buildings (GB50011-2001). Due to the fact that (11) and the damage loss f2 in Eq. (32)), and the
the classification of the five different degrees of explicit inelastic drift formulation in Eq. (22), the
damage is discrete in nature and the set of inter- multi-objective design optimization problem can
story drift responses of a building is continuous, be written in terms of the design variables i as
the fuzzy-decision theory is employed to best Minimize:
estimate the damage loss with specified proba-
bilities of occurrence of different levels of earth- F (i ) = { f1 , f2 } (33)
quakes. For simplicity, is used to represent the
inter-story drift index such that = u R h ,
where u is the inter-story drift; h is the story Subject to:
height; R is equal to 500 for concrete frames and

1000 for concrete shear-wall structures and frame-
Ni
1 1 Ni
g j (i ) = u j
i =i0
+ 1i (i i0 ) + 2i (i i0 )2 1
h j jp 2 i =1
wall structures. Finally, the total structural damage i =1
loss f2 presented in Eq. (31) can be explicitly

expressed in terms of the variable (u ) as ( j = 1, 2, ..., N j ) (34)
Nr Nj iL i Ui (i = 1, 2, 3, ..., N i ) (35)
f2 = f1 Pr (a1(u j ) + a2 ) (32)
r =1
j =1

Eqs. (33) defines the life-cycle cost function
F , which consists of the construction cost f1 and
where a1 and a2 are the coefficients of an ex-
the damage loss f2 ; Eq. (34) defines the inelastic
pected failure loss, which depend on not only
damage levels but also building classes and details inter-story drift constraint at the structural perfor-
are presented in Li (1998). For example, assum- mance point for a specified ground motion; Eq.
ing that the importance of a frame building belongs (35) defines the lower and upper size bounds
to Class B, the values of a1 and a2 can be derived specified for the design variables, i ; 1 and 2
as shown in Table 1. Details of derivation can be are given in Eqs.(28)-(29).
found in Zou (2002) and Zou et al (2007).
Table 1. The coefficients of expected failure loss for a Class B building
< 0.5 0.5 < 1.5 1.5 < 3 3 <7 7 < 10 10
a1 0.000 0.080 0.800 1.350 4.770 0.000
a2 0.02 -0.020 -1.100 -2.750 -26.667 21.000
220
3.4 Multi-Objective iL i Ui ( i = 1, 2, ..., N i ) (39)

Optimization Algorithm
By varying the upper bound value, 2 , for the
The multi-objective optimization problem given in objective f2 (i ) and minimizing the objective
Eqs. (33)-(35) can be solved by a Pareto optimal
f1 (i ) , all Pareto optimal points are, in principle,
set formed by a large number (infinite number)
attainable. Theoretically, there is no limitation on
of Pareto optimal solutions. The Pareto optimal
the range of 2 ( 2 + ) regardless of
set can provide an overview of all the tradeoffs
to the designer. To establish a Pareto optimum the convexity conditions. However, this may
point, multi-objective optimization algorithms result in extensive computational time to find a
should be constructed, one of which transforms Pareto set. In fact, the main difficulty of the
the multi-objective problem into a single-objective constraint method lies in finding the range of
optimization. The key to the transformation is reasonable values for the upper bound, 2 . Details
that the solution of the single-objective problem can be found in Zou (2002) and Zou et al (2007).
should be a point in the Pareto set with respect to
a feasible region and a set of objective functions. 3.5 Multi-Objective Design
Among transformation methods, the constraint Optimization Procedure
method is one of the commonly used approaches in
practical problems (Kaisa 1999; Marler and Arora The multi-objective inelastic optimal design
2004). The constraint method is a technique procedure is outlined as follows:
that transforms a multi-criteria objective function
into a single criterion by retaining one selected 1. The optimal member sizes Bi and Di are
objective function as the primary criterion to be first found based on the elastic design opti-
optimized and treating the remaining criteria as mization, which are fixed in the inelastic
constraints. design optimization. The initial reinforce-
Herein, the construction cost, f1 (i ) , is taken ment ratios for each member are taken as
as a primary objective, while the damage loss, the minimum values of the design variable,
f2 (i ) , is transformed into a design constraint. i , obtained from the elastic optimization
The reason for this consideration is that both the results.
damage loss f2 (i ) in Eq. (32) and the inter-story 2. Establish the explicit optimization problem
drift constraints in Eq. (34) are expressed explic- given in Eqs. (33)-(35).
itly in terms of the inter-story drifts. The similar- 3. Transfer multi-objective design problem Eqs.
ity may bring advantages in numerical calcula- (33)-(35) into a single objective optimization
tions. Thus, the multi-objective optimization Eqs.(36)-(39).
problem given in Eqs. (33)-(35) can be transformed 4. Change 2 in Eq.(40) and generate one Pareto
as optimum solution by employing the OC
method. The global convergence is checked
Minimize: f1 (i ) (36) based on the change in the objective function
and the violation of the constraints.
5. Repeat Steps 4 and 5 until the upper bounds
Subject to: f2 (i ) 2 (37)
of the active design variables are achieved.
g j (i ) 1 ( j = 1, 2, ..., N j ) (38)
221
4. ILLUSTRATIVE EXAMPLE jective optimization where other conditions are

the same as Case B.
A ten-story, two-bay planar frame is used to il-
lustrate the proposed optimal design method. The 4.1 A Single Objective Optimization
geometry of the example is given in Figure 4(a). (i.e., Cases A and B)
Concrete with the cylinder strength of 20MPa and
steel reinforcement with the yield strength, fy , of Two levels of earthquake loads are considered in
335MPa are used for all members. To illustrate this example. One represents a minor earthquake
the effectiveness of the optimal design technique, load with a peak acceleration of 0.32g according
three cases are conducted in this example. Cases to the acceleration response spectrum in GB5011-
A and B are elastic and inelastic design optimiza- 2001(2001), as shown in Figure 4(b). Another
tion for a single objective, respectively, while load level represents a severe earthquake with an
Case C considers life-cycle cost, i.e., multi-ob- initial peak acceleration of 1.4g. In the elastic
phase of the optimization, the concrete cost of the
Figure 4. A ten-story, two-bay frame
222
structure is to be minimized subject to the elastic structure. During the pushover analysis, the lateral
spectral drift constraints under the minor earth- loads are applied incrementally in proportion to the
quake loading condition. The unit construction initial loads but the gravity loads, shown in Figure
cost of concrete is assumed to be US$90/m3, 4(a), are assumed to be fixed. The design process
including the concrete material cost and labour is deemed to converge when the difference in the
cost. Elastic inter-story drift constraints are taken structure costs for two successive design cycles
into account with an allowable inter-story drift is within 0.5% and when the difference between
ratio limit of 1/450. The initial sizes are arbi- the active inter-story drift value and its allowable
trarily chosen to be 350 350mm for the columns limit at the performance point is within 0.5%.
and 200 350mm for the beams. Size bounds are Figure 5 presents the optimal design history
defined as 350~1000mm for the depths and the for both the elastic and inelastic drift optimiza-
widths of the columns, 200~350mm for the widths tion processes. In the elastic optimal design,
of the beams and 350~450mm for the depths of rapid and steady convergence of the concrete cost
the beams. from the initial US$1528 to the finial US$2205
In the inelastic phase of the optimization, the after 6 design cycles has been found. The rapid
design objective is to minimize the steel reinforce- convergence can be explained by the fact that the
ment cost subject to the performance-based in- member force distribution for such structures is
elastic drift constraints under the severe earthquake somewhat insensitive to changes in member size.
loading condition. The unit construction cost of In contrast, the inelastic optimal design process
steel reinforcement is assumed to be US$960/ converges quite slowly but steadily in 11 design
tonne including the steel material cost and the cycles. Relatively slow, but steady, convergence is
labour cost. Inelastic inter-story drift constraints inevitable due to the need of maintaining a small
are considered with an allowable inter-story drift change in the steel reinforced ratios during the
ratio limit of 1/100 and the P-Delta effect is con- nonlinear design optimization process. However,
sidered in the example. Initial reinforcement ratios the OC design method is able to achieve a smooth
are calculated based on the strength requirements and steady convergence to the optimal design as
of members after the elastic phase design process. evidently shown in Figure 5.
Such strength-based reinforcement ratios are In the elastic design optimization subject to
taken as the lower bounds for the inelastic design elastic inter-story drift constraints, the total mate-
process. Their upper bounds are assumed to be rial cost only includes the concrete cost, which is
6.0% for columns and 4.0% for beams. For sim- minimized from the arbitrary, initial US$1528 to
plicity, symmetrical arrangement of steel rein- the final US$2205. In the inelastic design opti-
forcement is assumed such that i = i . Flex- mization subject to inelastic inter-story drift
ural moment hinges and axial-moment hinges are constraints, the initial total cost US$3898 consists
assigned to the end locations of the beams and of the steel reinforcement cost of US$1693 cal-
columns, respectively. The ultimate plastic hinge culated based on code-specified strength require-
rotation, Up , is assumed to be 0.02 radian for the ments and the concrete cost of US$2205, which
moment hinges on the beams and 0.015 radian is to be fixed in this inelastic phase of the opti-
for the axial-moment hinges on the columns. mization. Since initial violations in inelastic inter-
Initial applied lateral loads applied in push- story drift constraints are found in the strength-
over analysis are shown in Figure 4(a), which is based design, an increase in the steel reinforcement
proportional to the product of the story mass and is necessary, resulting in a final total construction
the first mode shape of the elastic model of the cost of US$4184. In addition, it is found from
Case B that there is a relatively large increase of
223
Figure 5. Design history of structural cost
16% in the steel cost from the initial US$1693 to The initial and final inter-story drift ratios are
US$1978. shown in Figure 7(a) for Case A and Figure 7(b)
Figure 6 presents the optimal steel reinforce- for Case B. The initial inter-story drift constraints
ment ratios for Cases A and B. In the elastic phase, at the second through the eighth floors are found
the member sizes are increased after the elastic to violate substantially the allowable inter-story
design optimization since the initial design is drift ratio limit of 1/100 for both cases, resulting
found to be infeasible. The starting strength-based in the occurrence of the weak stories on the these
design, in the second inelastic design phase, is also floor levels of the building. However, these push-
found to be infeasible in terms of the assumed over inter-story drift constraints are found to be
allowable inter-story drift limit. After the inelas- close to and within the allowable values after the
tic design optimization, the steel reinforcement optimization, indicating that a rather uniform
ratios of the beams greatly increase particularly inter-story drift distribution over the height of the
in the lower levels of the structure, while those building has been achieved and the occurrence of
of columns are found with little changes for Case weak story has been prevented at the final per-
B. Reasons lie on that the columns of Case B are formance point.
found to be mainly controlled by the minimum Figure 8(a) includes a table showing the num-
strength requirements with the final steel ratios ber of plastic hinges at three different performance
being set to the initial strength based limits, states, i.e., Immediate Occupancy (IO), Life
whilst the beams are governed by the inelastic Safety (LS) and Collapse Prevention (CP) for
inter-story drift performance requirements with Case B. Figures 8(b)-(c) show the initial and final
the final steel ratios being larger than the initial plastic hinge distributions under the pushover
strength based limits. loading at the performance point of the structure.
224
Figure 6. Optimal member sizes and reinforcement ratios for three cases
Figure 7. Interstory drift responses of three cases
225
Figure 8. Initial and final plastic hinge distribution of Case B
No plastic hinge rotation is found to exceed the rather linear deflected profile of the inelastic
specified threshold of plastic rotation, Up . As design. Such a result further indicates that the
shown in Figure 8(b), the rotations of twenty optimization method developed can automati-
plastic hinges of the initial design are found to be cally resize the steel reinforcements of all mem-
located between the LS-CP state. However, after bers to attain a uniform ductility demand along
the optimization, most of the plastic hinges are the height of the multi-story building.
found to be in the B-IO and IO-LS states and only
one hinge is in the LS-CP state, as can be observed 4.2 Multi-Objective Optimization
from the optimized framework in Figure 8(c). (Case C)
Furthermore, the inter-story drifts along the height
of the building are also found to be almost all The occurrence probability Pr for the severe
fully constrained at the optimum, resulting in a earthquake is assumed to be 3.9% according to
226
the design return period of 50 years. This building US$7522 at Point C0 is the maximum value of
frame is assumed to be a Class B building in this the damage loss, f2 .
example and the coefficients of expected failure Firstly, it is observed from Figure 9 that an
loss are shown in Table 1. Other conditions stay improvement in the damage loss leads to an in-
the same as Case B. crease in the initial material cost of the structure.
Figure 9 presents a Pareto optimal set, which From the Pareto optimal set, a designer can have
provides the best compromise between the initial an overview of the tradeoffs. Secondly, any point
structure cost, f1 , and the damage loss, f2 . The in the Pareto optimal set can be selected as an
minimum structure cost, f1 , for the single objec- optimum design, which represents a designers
tive optimal design problem given in Eqs. (36)- preference. A designer has many choices rather
(39), i.e., corresponds to Point C0 in Figure 8 than only one and he/she may finally choose a
where f1 = US$4184 and the corresponding dam- particular solution, which he/she thinks, should
age loss, f2 , is found to be US$7522 . The optimal be realized. Thirdly, the life-cycle cost can be
steel reinforcement ratios corresponding to Point further written as F = 1 f1 + 2 f2 , where 1 and
C0 are taken as initial steel ratios of members in 2 are the weighting factors and their magnitudes
the subsequent multi-objective inelastic design are dependent on a designers decision. A preferred
optimization (see Eqs. (33)-(35)) and the optimal balance of the initial and damage losses can
Pareto set curve can then be generated. Besides, therefore be obtained. It is worth noting that dif-
US$4184 at Point C0 is the minimum value of ferent values of 1 and 2 will lead to different
the initial structure cost, f1 , satisfying design minimum values of the life-cycle cost.
constraints at this Pareto curve; correspondingly, Figure 6 lists the optimal steel reinforcement
ratios at Points C0 and C*, respectively. The steel
Figure 9. A Pareto optimal set of Case C
227
reinforcement ratios of beams are largely increased found necessary to ensure a smooth and steady
from the first to the eighth floors from Point C0 to convergence of the inelastic drift design process.
Point C* (i.e., an increase of 17%~47%), while It has been demonstrated that steel reinforce-
there is no change of steel reinforcement ratios ment plays a significant role in controlling the
in most columns due to no occurrence of plastic lateral drift beyond first yielding and in providing
hinges. Such an increase indicates that steel rein- ductility to an RC building framework. It is also
forcement has a significant effect on improving demonstrated that the OC design method is able
inelastic drift performance and further reducing to attain automatically and gradually the optimal
damage loss. performance-based inter-story drift design. At op-
Figure 7(c) presents the results of the inter- timum, a uniform lateral drift or ductility demand
story drift ratios at Points C0 and C*, respectively. over all stories of the building with the minimum
It is found that all the inelastic inter-story drift cost is achieved, thus preventing the occurrence
constraints are satisfied corresponding to either of soft story mechanisms in such structures.
Point C0 or Point C*. In the optimal solution (i.e., The optimization technique developed for the
corresponding to Point C0) of the single objective nonlinear multi-objective design of inelastic drift
design optimization, the inter-story inelastic drift performance of RC frameworks under pushover
constraints from the second to the seventh floors loadings provides a very effective way to simul-
are close to the limiting ratio of 1%. However, in taneously optimize the structural life-cycle cost
the multi-objective design optimization (i.e., Point while satisfying all drift performance design cri-
C*), all final drift constraints are not necessarily teria. A multi-objective optimization algorithm,
found to be close to the limiting ratio 1%. This the constraint method, has been effectively
is due to the fact that steel ratios of the structure applied to handle the conflicts between the initial
members are further enhanced in order to meet the structure cost and damage loss by producing a
best balance between the initial structure cost and Pareto optimal set, from which a decision maker
damage loss. Such a result seems to indicate that can directly select the best compromise solution.
for the multi-objective design optimization, the It is believed that the single- and multi-
lateral load resisting system can be automatically objective optimization methodology provides a
improved by the OC procedure to seek the best powerful computer-based technique for seismic
balance so that lateral drift constraints are satis- performance-based design of multi-story RC
fied simultaneously with the least life-cycle cost. building structures. The proposed optimization
methodology provides a good basis for more
comprehensive performance-based optimization
5. CONCLUSION of structures as multiple levels of performance
criteria and design objectives are to be considered.
Using the principle of virtual work, both the elastic
spectral drift constraints and the nonlinear push-
over inelastic drift constraints have been explicitly ACKNOWLEDGMENT
formulated in terms of the design variables. Rapid
and steady convergence for elastic drift optimi- The author would acknowledge the thoughtful
zation has been found. In contrast, the inelastic suggestions and support from Prof. C.M. Chan of
optimal design process converges quite slowly the Department of Civil Engineering, the Hong
but steadily. The restrictive move limit imposed Kong University of Science and Technology.
on the steel reinforcement design variables is
228
REFERENCES Charney, F. A. (2000). Needs in the development

of a comprehensive performance based design op-
Ang, A. H.-S., Lee, J.-C., & Pires, J. A. (1997). timization process. In M. Elgaaly (Ed.), Advanced
Cost-effectiveness evaluation of design criteria. Technology in Structural Engineering - Structural
Proceedings of the International Workshop on Congress 2000, ASCE, May 8-10, (CD-ROM).
Optimal Performance of Civil Infrastructure
Systems (pp.1-16). New York, NY: ASCE. Cheng, F. Y., & Botkin, M. E. (1976). Nonlinear
optimum design of dynamic damped frames.
Arora, J. S. (1999). Optimization of structures Journal of the Structural Division, 102, 609628.
subjected to dynamic loads. In Leondes, C. T.
(Ed.), Structural dynamic systems computational Cheng, F. Y., & Chang, C. C. (1988). Safety-based
techniques and optimization (pp. 173). Gordon optimum design of nondeterministic structures
and Breach Science Publishers. subjected to various types of seismic loads. NSF
Report, U.S. Department of Commerce, VA, NTIS
ASCE. 41. (2007). Seismic rehabilitation of exist- No. PB90-133489/AS.
ing buildings ASCE/SEI 41/06. ASCE Standard
No. ASCE/SEI 41-06. ATC-40. (1996). Seismic Cheng, F. Y., & Li, D. (1997). Multiobjec-
evaluation and retrofit of concrete buildings, Vol. tive optimization design with Pareto genetic
1. ATC-40 Report. Redwood City, CA: Applied algorithm. Journal of Structural Engineering,
Technology Council. 123(9), 12521261. doi:10.1061/(ASCE)0733-
9445(1997)123:9(1252)
Balling, R. J., Pister, K. S., & Ciampi, V. (1983).
Optimal seismic-resistant design of a planar Cheng, F. Y., & Truman, K. Z. (1982). Optimization
steel frame. Earthquake Engineering & Struc- algorithm of 3-D building systems for static and
tural Dynamics, 11, 541556. doi:10.1002/ seismic loading. In Ames, W. F. (Ed.), Modeling
eqe.4290110407 and simulation in engineering (pp. 315326).
North-Holland Pub. Co.
Beck, J. L., Papadimitriou, C., Chan, E., & Irf-
anoglu, A. (1998). A performance-based optimal Computer and Structures, Inc. (CSI). (2000).
structural design methodology. Report No. EERL SAP2000/NL-PUSH software, Version 7.40.
97-03, CA, USA. Berkeley, CA: Computer and Structures, Inc.
Bhatti, M. A., & Pister, K. S. (1981). A dual criteria Federal Emergence Management Agency. (2000).
approach for optimal design of earthquake-resis- Prestandard and commentary for the seismic
tant structural systems. Earthquake Engineering rehabilitation of buildings. FEMA-356. Wash-
& Structural Dynamics, 9, 557572. ington, DC.
Chan, C. M. (1997). How to optimize tall steel Federal Emergence Management Agency. (2005).
building frameworks. Guide to structural optimi- Improvement of nonlinear static seismic analysis
zation (pp. 165-195). ASCE Manuals and Reports procedure. FEMA-440. Washington, DC.
on Engineering Practice No.90. Federal Emergence Management Agency. (2006).
Chan, C. M., & Zou, X. K. (2004). Elastic and Next-generation performance-based seismic
inelastic drift performance optimization for rein- design guidelines. FEMA-445. Washington, DC.
forced concrete building under earthquake loads.
33(8), 929950. doi:10.1002/eqe.385
229
Feng, T. T., Arora, J. S., & Huang, E. J. (1977). Marler, R. T., & Arora, J. S. (2004). Survey
Optimal structural design under dynamic loads. of multi-objective optimization methods for
International Journal for Numerical Methods engineering. Structural and Multidisciplinary
in Engineering, 11(1), 3952. doi:10.1002/ Optimization, 26(6), 369395. doi:10.1007/
nme.1620110106 s00158-003-0368-6
Foley, C. M. (2002). Optimized performance- Moehle, J. P., & Mahin, S. A. (1991). Observations
based design for buildings. In Burns, S. A. (Ed.), on the behavior of reinforced concrete buildings
Recent advances in optimal structural design (pp. during earthquakes. American Concrete Institute
169240). American Society of Civil Engineers. SP-127. In Ghosh, S. K. (Ed.), Earthquake-
resistant concrete structures Inelastic response
Gao, X. W., & Bao, A. B. (1985). Probabilistic
and design.
model and its statistical parameters for seismic
load. Earthquake Engineering and Engineering National Standard of the Peoples Republic of
Vibration, 5(1), 1322. China. (2001). Chinese code for seismic design
buildings (GB50011-2001). Beijing, China: New
International Conference of Building Officials.
World Press.
(1997). Uniform building code (UBC). Whittier,
California, USA. Park, Y. J., & Ang, A. H.-S. (1985). Mecha-
nistic seismic damage model for reinforced
Kaisa, M. (1999). Nonlinear multiobjective op-
concrete. Journal of Structural Engineering,
timization. USA: Kluwer Academic Publishers.
111(4), 722739. doi:10.1061/(ASCE)0733-
Krawinker, H. (1994). Static pushover analysis. 9445(1985)111:4(722)
Proceedings SEAONC 1994 Fall Seminar on
Park, Y. J., Ang, A. H.-S., & Wen, Y. K.
the Developing Art of Seismic Engineering. San
(1987). Damage-limiting aseismic design of
Francisco, CA.
buildings. Earthquake Spectra, 3(1), 126.
Li, G. (1998). Reliability and performance based doi:10.1193/1.1585416
optimal design for seismic high-rising structures.
Vision, S. E. A. O. C. (2000). Committee. (1995).
Ph.D. Dissertation, Dalian University of Technol-
Performance based seismic engineering of
ogy, China.
buildings, part 2: Conceptual framework. Sac-
Li, G., Zhou, R.-G., Duan, L., & Chen, W.-F. ramento, CA: Structural Engineers Association
(1999). Multiobjective and multilevel optimization of California.
for steel frames. Engineering Structures, 21(6),
Wang, Q., Fang, H., & Zou, X. K. (2010). Applica-
519529. doi:10.1016/S0141-0296(97)00226-5
tion of micro-GA for optimal cost base isolation
Liu, M., Burns, S. A., & Wen, Y. K. (2003). design of bridges subject to transient earthquake
Optimal seismic design of steel frame buildings loads. Structural and Multidisciplinary Optimiza-
based on life cycle cost considerations. Earth- tion, 36(5), 493507.
quake Engineering & Structural Dynamics, 32(9),
Wen, Y. K., & Kang, Y. J. (1997). Optimal seismic
13131332. doi:10.1002/eqe.273
design based on life-cycle cost. Proceedings of
the International Workshop on Optimal Perfor-
mance of Civil Infrastructure Systems 1997, (pp.
194-210). New York, NY: ASCE.
230
Wen, Y. K., & Kang, Y. J. (2001). Minimum Zou, X. K., & Chan, C. M. (2005b). Optimal
building life-cycle cost design criteria. I: Meth- seismic performance-based design of reinforced
odology. Journal of Structural Engineering, concrete buildings using nonlinear pushover
127(3), 330337. doi:10.1061/(ASCE)0733- analysis. Engineering Structures, 27, 12891302.
9445(2001)127:3(330) doi:10.1016/j.engstruct.2005.04.001
Zou, X. K. (2002). Optimal seismic performance- Zou, X. K., Chan, C. M., Li, G., & Wang,
based design of reinforced concrete buildings. Q. (2007b). Multiobjective optimization for
Ph.D. Dissertation, Hong Kong University of performance-based design of concrete struc-
Science and Technology. tures. Journal of Structural Engineering,
133(10), 14621474. doi:10.1061/(ASCE)0733-
Zou, X. K. (2008). Integrated seismic design opti-
9445(2007)133:10(1462)
mization of nonlinear base-isolated RC buildings.
Structural and Multidisciplinary Optimization, Zou, X. K., Teng, J. G., Xia, S. H., & Wong, Y.
36(5), 493507. doi:10.1007/s00158-007-0184-5 L. (2007a). Optimal performance-based seismic
retrofit design of FRP-confined concrete buildings.
Zou, X. K., & Chan, C. M. (2001). Optimal
Composite Part B: Engineering International
drift performance design for nonlinear pushover
Journal, 38, 584597. doi:10.1016/j.compos-
response of concrete structures. WCSMO-4:
itesb.2006.07.016
Proceedings of the Fourth Would Congress of
Structural and Multidisciplinary Optimization, Zou, X. K., Wang, Q., Li, G., & Chan, C. M.
June 4-8, Dalian, China. (2010). Integrated reliability-based seismic drift
design optimization of base-isolated concrete
Zou, X. K., & Chan, C. M. (2005a). An optimal
buildings. Journal of Structural Engineer-
resizing technique for seismic drift design of
ing, 136(10), 12821295. doi:10.1061/(ASCE)
concrete buildings subjected to response spec-
ST.1943-541X.0000216
trum and time history loadings. Computers &
Structures, 83, 16891704. doi:10.1016/j.comp-
struc.2004.10.002
231
232
Chapter 10
Applications of Topology
Optimization Techniques in
Seismic Design of Structure
Kazem Ghabraie
University of Southern Queensland, Australia
ABSTRACT
During the last two decades, topology optimization techniques have been successfully applied to a wide
range of problems including seismic design of structures. This chapter aims to provide an introduction
engineering. Two well-established topology optimization techniques are introduced. Several problems
including eigenfrequency control of structures, compliance minimization under periodic loading, and
maximizing energy absorption of passive dampers will be addressed. Numerical instabilities and ap-
proaches to overcome them will be discussed. The application of the presented approaches and methods
will be illustrated using numerical examples. It will be shown that in seismic design of structures, topol-
ogy optimization methods can be useful in providing conceptual design for structural systems as well as
detailed design of structural members.
1. INTRODUCTION function subject to some constraints. In struc-

tural optimization, depending on the nature of
Solving optimization problems is an inherent part the design variables, three different optimization
of engineering design where one seeks the best categories can be recognized. Sizing optimization
design to minimize or maximize an objective arises when the design variables are connected to
the dimensions of the elements. It can be useful
DOI: 10.4018/978-1-4666-1640-0.ch010 where the layout and the shapes of the members
Applications of Topology Optimization Techniques
Figure 1. The three levels of structural optimization: top) sizing optimization; middle) shape optimiza-
tion; bottom) topology optimization
are known and it is desired to find the optimum history of topology optimization. Then we focus
dimensions. On another level, one can choose on two general optimization problems in seismic
the design variables to control the shape of the design of structures, the eigenvalue optimization
boundaries of the members. Such selection will problem and the problem of maximizing the en-
lead to shape optimization. If the overall layout ergy absorption.
of the members is known and it is already decided
where to put each member, in order to find the
best shapes of the members, one can use shape 2. TOPOLOGY OPTIMIZATION
optimization. In order to optimize the topology,
connectivity, or layout of a system, topology Initially addressed by Culmann (1866), the lay-
optimization techniques should be used. In topol- out optimization problem is not quite new. The
ogy optimization the design variables control the interesting work of Michell (1904) laid down the
topology and connectivity of the design. Figure principles of topology optimization of structures
1 schematically illustrates these three categories more than a century ago. After that, the field re-
of structural optimization. mained untouched for nearly seven decades until
Starting from topology optimization and feed- Prager and Rozvany improved and generalized the
ing the results to shape and sizing optimization Michells theory (e.g. refer to Prager 1969, 1974
routines will generally result in far greater savings and Rozvany 1972a,b). Yet the field didnt attract
than merely using shape and sizing optimization. much attention until Bendse and Kikuchi (1988)
Topology optimization techniques can thus be proposed a finite element-based numerical method
considered as important and powerful tools in for topology optimization of continuum structures.
hand of design engineers. Usually referred to as the homogenization method,
In this chapter we review the application of to- this approach soon became a basis upon which
pology optimization techniques in seismic design other topology optimization techniques have been
of structures. We start with a brief review of the developed.
233
In their approach, Bendse and Kikuchi consid- 3. STRUCTURAL RESPONSES

ered special microstructures as the building cells UNDER DYNAMIC LOADS
of the structure and employed the homogenization
method to find the macro-scale properties of the Consider the equation of motion for a finite ele-
cells in terms of their micro-scale dimensions. ment discretized linear system
By considering the dimensional properties of the
microstructures as design variables, they reduced + Cu + Ku = p
Mu (1.1)
the topology optimization of the structure to sizing
optimization of its microstructures. where M, C and K are mass, damping and stiff-
Using the idea behind the homogenization ness matrices respectively and u and p are time-
method, Bendse (1989) introduced a simpler dependent vectors of nodal displacement and nodal
approach to optimize the topology of structures. force respectively, i.e., u u(t) and p p(t). We
In this new approach instead of using the micro- assume a classical damping (Chopra 1995), for
structures and homogenization, Bendse proposed example Reyliegh damping of the form
an artificial material interpolation scheme relating
the material properties of the elements to their C = aM M + aK K (1.2)
relative density. After Rozvany et al. (1992), this
approach is referred to as Solid Isotropic Micro-
structures with Penalization (SIMP). The SIMP where aM and aK are constants.
approach is now one of the most established and We now expand the displacements in terms of
popular methods in topology optimization. modal contributions
A simple FE-based topology optimization Nd
technique was later proposed by Xie and Steven u(t ) = qr (t )r (1.3)
(1993). Named Evolutionary Structural Optimiza- r =1
tion (ESO), the technique was based on the idea

of evolving the structure towards an optimum where Nd is the number of degrees of freedom and
design by progressively removing its inefficient
elements. The Bi-directional ESO (BESO) was qr (t ) = C r cos r t + Sr sin r t, r = 1, , N d
the main successor of the ESO method. Initially (1.4)
introduced by Querin (1997), Querin et al. (1998)
and Yang et al. (1999a), the BESO algorithm was are harmonic functions and Cr and Sr are constants
capable of adding as well as removing elements. of integration associated with the rth degree of
This method is now a well-known topology op- freedom. The natural frequencies r and natural
timization technique which is widely used due to modes r are solutions of the following eigenvalue
its clear topology results and ease of application. problem
The SIMP and BESO techniques will be de-
tailed and used in later sections of this chapter. Kr = r2 Mr (1.5)
In the next section we investigate the equation of
motion of a structural system to find out which
parameters shape the responses of structures under which represents the free vibration of the un-
dynamic loads. damped system. For simplicity, we further require
that the modes are M-orthonormal, i.e.,
234
Tn Mr = nr , n, r = 1, , N d (1.6) ing (or minimizing) any combination of natural

frequencies. A practically useful example of such
where nr is the Kronekers delta which equals 1 extensions will be briefly addressed in section 6.
for n = r and 0 otherwise. Premultiplying Equa- It is worth noting that maximizing fundamental
tion (1.5) by Tn and using Equation (1.6) we get frequency results in structures with a reasonable
stiffness against static loads in general (Bendse
Tn Kr = r2nr , n, r = 1, , N d (1.7) and Sigmund 2003).
Damping effects are ignored and linear elastic
material behavior is assumed in this section. Also
which means the modes are also K-orthogonal. all material parameters are taken as deterministic
Using Equation (1.3) in Equation (1.1) and pre- quantities. Random variability of material strength
multiplying by Tn we obtain parameters can significantly affect the ductility
and energy absorption capacities of structures
Nd Nd Nd
Tn Mrqr + Tn Crqr + Tn Krqr = Tn p subjected to seismic loading (Kuwamura and Kato

r =1 r =1 r =1 1989, Elnashai and Chryssanthopoulos 1991).
(1.8) Uncertainties of variables can be considered
in structural optimization by integrating Reli-
We now make use of M-orthonormality of the ability Analysis (see e.g. Kharmanda et al. 2004
modes and the classical damping Equation (1.2) and Papadrakakis et al. 2005) or through Robust
to simplify Equation (1.8) to Optimization (see e.g. Beyer and Sendhoff 2007).
Using the finite element discretization and the
qn + 2n nqn + n2qn = Tn p (1.9) SIMP approach we introduce the following mate-
rial interpolation scheme to express the Youngs
aM 1 a modulus Ee of element e in terms of its relative
where n = + K n is the damping ratio density xe as
2 n 2
of the n-th mode (Chopra 1995).
According to Equation (1.9), the response of Ee (xe ) = xep E (1.10)
a structure under a dynamic load depends on its
natural frequencies n and damping ratios n. where E is the Youngs modulus of the base
isotropic material. The power p > 1 is known as
the penalty factor and is introduced to push the
4. MAXIMIZING EIGENFREQUENCIES solutions towards a solid-void topology. A typical
IN FREE VIBRATION value for the penalty factor is p = 3 (Bendse and
Sigmund 1999). The relative densities are chang-
As seen in the previous section, controlling the ing in the range 0 xe 1 in which xe = 1 represents
response of structures can involve eigenfrequency solids and xe = 0 represents void areas. In order
optimization. In this section we address the prob- to avoid singularities in the stiffness matrix of the
lem of maximizing the fundamental frequency of system, Equation (1.10) may be replaced by
a structure in free vibration. This problem was
initially addressed by Daz and Kikuchi (1992) Ee (xe ) = E + xep (E E ) (1.11)
using the homogenization method. Here, we for-
mulate the problem using the SIMP approach. This
in which E is a small elastic modulus assigned
formulation can be simply extended to maximiz-
to voids. Based on Equation (1.10), the (local
235
level) stiffness matrix of the element e can be where j = j2 , j = 1, , N d and N is the number
expressed as of elements. The second constraint restricts the
volume of the design to an upper limit denoted
Ke (xe ) = Ke + xep (Ke Ke ) (1.12) by v . In this statement ve is the volume of the
element e.
in which Ke and Ke are the stiffness matrices of
4.1. Sensitivity Analysis
the element e when it is made of the base mate-
rial and void (in its solid and void states) respec-
Solving Problem (1.15) requires finding the sen-
tively. Similarly for the density and the mass
sitivities of 1 with respect to design variables xe.
matrix of the element e we can write
Differentiating Kj = jMj we can write
e (xe ) = xe (1.13) j j j
K M
j + K = M j + j j + j M
x e x e x e x e xe
and (1.16)
Me (xe ) = xe Me (1.14) Premultiplying by Tj and rearranging the

terms we obtain
Where is the density of the base material and
K M
Me is the mass matrix of the element e in its Tj j j = j Tj M j Tj (K j M) j
xe xe xe xe
solid state. (1.17)
Using these material interpolation schemes
one can change element e from solid to void and Using the symmetry of K and M, we can read-
backwards by changing the value of xe. Thus by ily conclude that
choosing xe-s as design variables, one can produce
different topologies without altering the finite T
element mesh. Tj (K j M) = (K j M) j = 0 .

We can now formulate the optimization prob-
lem. The fundamental frequency optimization We also use Equation (1.6) in Equation (1.17)
problem can be stated as finding the best topol- to finally express the sensitivities as
ogy of a structure to maximize its fundamental
frequency given a fixed amount of material. The j K M
= Tj j j (1.18)
problem can thus be formulated as xe xe xe
max
x 1 ,x 2 ,,x N
{ = 1 min j
j =1,2,,N d } The stiffness and mass derivatives in Equa-
tion (1.18) can be calculated using Eqs. (1.12)
such that K j = j M j , j = 1, 2, , N d and (1.14).
N

x v
e =1
e e v
4.2. Solution Method
0 xe 1, e = 1, 2, , N
(1.15) Having the sensitivities in Equation (1.18), the
optimization problem (1.15) can be solved using
236
suitable gradient-based techniques. Noting that the 1

De = ve = e
number of design variables (number of elements) x e
can be very large, one should adopt a solution N N
method capable of solving large-scale problems. v xeve = 0; v xeve 0; 0

e =1 e =1
The method of moving asymptotes (MMA)
xe = 0 e 0
proposed by Svanberg (1987) is a well-known
0 < xe < 1 e = 0
solution method used in topology optimization
problems. Another common approach is using xe = 1 e 0
optimality criteria (OC) based algorithms. In the e = 1, , N
following, after deriving the optimality criteria (1.21)
for the optimization problem (1.15), we propose
a heuristic iterative fixed-point algorithm to solve To increase the fundamental frequency, we add
the optimization problem based on the optimality a vector of increments x = (x1, x2,, xN) T
criteria. to the design variables x = (x1, x2,, xN)T. The
The eigenvalue equation, Kj = jMj can be subsequent change in the fundamental frequency
satisfied separately using finite element analysis. and the design volume can then be evaluated as
Excluding this equation, the Lagrangian of Prob-
T
lem (1.15) takes the form 1 = (1 ) x (1.22)
N N
= 1 + v xeve + e (1 xe ) + e xe v = vT x (1.23)
e =1
e =1
(1.19)
( )
T
1
where 1 = x 1
, x 12 , , x N1 is the gradient
where , e and e are Lagrange multipliers. vector of 1 and v = (v1 , v 2 , , v N )T .
Using Karush-Kuhn-Tucker results (Karush 1939; Let us now define the increments of design
Kuhn and Tucker 1951), the necessary optimal- variables as
ity conditions for Problem (1.15) can be expressed
as follows x = D = 1 v (1.24)

= 1 ve e + e = 0
xe x e If the volume constraint is inactive, we will
N N have = 0 and thus x = 1 which results in
v xeve = 0; v xeve 0; 0
T
e =1 e =1 1 = (1 ) 1 0 after substituting in
e (1 xe ) = 0; 1 xe 0; e 0, e = 1, N
Equation (1.22).
e xe = 0; xe 0; e 0, e = 1,N
If the volume constraint is active, on the other
(1.20) hand, we will have
If we define e = e e and use it in Equa- v = vT x = 0 (1.25)

tion (1.20), we can rewrite the optimality criteria
as
Using Equation (1.24) in Equation (1.25) and
solving for we obtain
237
vT 1 (Sigmund and Petersson 1998). Checkerboard

= (1.26)
vT v problem refers to the formation of alternating
solid and void elements in a checkerboard-like
If the boxing conditions are all inactive, i.e. if pattern resulting in artificially high stiffness. Mesh
0 < xe < 1, we can use Equation (1.26) in Equa- dependency refers to obtaining different optimal
tion (1.24) and then Equation (1.24) in Equation topologies for the same problem using different
(1.22) to write mesh sizes. Local minima refers to the problem
of obtaining different optimal topologies using the
T vT 1 v same mesh but different algorithmic parameters
1 = (1 ) 1 (1.27) and/or initial design.
vT v
One of the simplest yet effective approaches
to overcome checkerboard and mesh dependence
It can be easily verified that the right hand side problems is filtering sensitivities (Sigmund and
of Equation (1.27) is a form of CauchyBunya- Petersson 1998). In this approach the calculated
kovskySchwarz inequality and thus 1 0 . sensitivities are replaced by filtered sensitivities
Based on this discussion, we propose the fol- which are calculated as a weighted average of
lowing update scheme to solve Problem (1.15) the sensitivities of the neighboring elements. A
simple linear filter takes the form

(
0, x ,
(k ))

x(k +1) = max

N

min 1, x +

x
1 p
( ) , x(k ) + D(k )
p
D(k ) w

(k )

j
x j ij

=
e =1
(1.29)
(1.28) x i N
x i wij
j =1
Here the subscripts denote the iteration num-
ber and is a tuning parameter defining the move
in which wij = max{0, R dij}. R is known as the
limit. The vector D is defined in Equation (1.24).
filtering radius and dij denotes the distance between
Note that p used here is the previously defined
the centers of the elements i and j. The filtering
penalty power for stiffness. The value of the
scheme (1.29) can be activated by choosing the
Lagrange multiplier can be calculated using
filtering radius R bigger than the size of elements
bisection method in an inner loop. In finding ,
h. This can eliminate the checkerboard problem.
one should note that v < 0.
In this scheme, the filtering radius R imposes a
Note that the same algorithm can be used
local minimum length scale to the solutions. More
to maximize any of the natural frequencies. To
precisely, using this sensitivity filter, the width of
maximize the kth eigenvalue, for example, one
bars appearing in the resulting topologies could
needs to replace 1 and 1 by k and k respectively.
not be smaller than 2R. This property is useful in
achieving mesh independency. By defining R as
4.3. Numerical Instabilities
a ratio of the actual length of the design domain,
the mesh dependence problem can be rectified.
Most of the material distribution techniques,
Unlike the first two types of numerical insta-
including homogenization, SIMP and BESO
bilities, the local minima problem is mostly due to
methods, are known to be prone to three major
the use of gradient-based optimization algorithm
numerical instabilities, namely checkerboard
which can be trapped in local minima of usually
problem, mesh dependency, and local minima
non-convex objective functions. On the other
238
hand, the extremely large size of the problems in 4.4. Flowchart and
topology optimization is a great barrier in using Numerical Examples
non-gradient-based optimization techniques such
as Genetic Algorithm (GA) and Neural Networks. A flowchart of the proposed solution algorithm
The continuation method is a simple approach is depicted in Figure 2. The solution starts from
used and suggested by many researchers to over- an initial guess design. A uniform distribution of
come this problem in gradient-based optimization material defined as xe = v / vt , e = 1, , N with
methods (Sigmund and Petersson 1998). In this vt denoting the total volume of the design domain
approach, one would start solving the problem in a is usually used as an initial design. The main loop
more relaxed form and gradually apply restrictions. starts by analyzing the current design using finite
For example, one can start the solution considering element analysis. Based on FE results, the sensi-
no penalty factor (p = 1) and gradually increase tivities are calculated using Equation (1.18). The
p upon convergence of the solution. sensitivities are then filtered using Equation (1.29).
Apart from these three problems which are The updated variables are calculated in an inner
common in all types of topology optimization loop. In the inner loop, starting with a positive
problems some numerical artifacts are unique value for the Lagrange multiplier , the updated
to eigenvalue problems. The artificial modes variables are calculated using Equation (1.28).
problem comes under this category. These are The volume of this new design is checked and
localized modes appearing in regions with rela- the new value of is adjusted using the bisection
tively high mass to stiffness ratio. Pedersen (2000) approach. The inner loop continues until the
points that due to the interpolation schemes for value of converges. The updated design is then
stiffness and mass (Eqs. (1.12) and (1.14) respec- replaces the old one and the procedure is re-
tively), the mass to stiffness ratio rises steeply for peated until a convergence criterion is satisfied.
small values of x which ultimately results in local- The convergence criterion used here is defined
ized modes. To overcome this problem, a modi- as
fication in the stiffness interpolation scheme is
suggested by Pedersen (2000) to limit the mass k k 1
to stiffness ratio in low density areas (typically

i =k l +1
(i )
i =k l
(i )
x < 0.1 ). Following the same principle, Du and k k 1

(1.31)
Olhoff (2007) proposed a different approach by (i ) + (i )
modifying the mass interpolation scheme. The i =k l +1 i =k l
latter approach is adopted here. To this end, we

replace the original mass interpolation scheme, where (i) is the value of the objective function at
Equation (1.14), by the i-th iteration and k denotes the last iteration.
This condition compares the value of the objective
x M , x e > 0. 1 function in the last and second last l iterations and
Me (xe ) = e q 1e q assumes convergence is achieved when the rela-
10 xe Me ,
x e 0. 1 (1.30)
tive error is smaller than a predefined tolerance
0 < 1. In all examples reported here l = 5
and = 0.001 were used.
with q = 2p. This ensures that the mass to stiffness
ratio cannot exceed 10p-1.
239
Figure 2. The flowchart of the proposed solution algorithm
Example 1: Reinforcement The 312m domain is descretized using a

of a Planar Frame 30120 mesh of 4 node square bi-linear elements.
A consistent mass matrix formulation has been
As the first example we consider a frame in plane used (see e.g. Zienkiewicz et al. 2005). The stiff-
stress. The maximum volume of used material ness and density of the base material are assumed
should be limited to half of the volume of the as E = 2 105 MPa and = 8000 kg/m 3 re-
whole frame ( v / vt = 50% ). The initial design spectively. A penalty power of p = 4, a filtering
is depicted in Figure 3 (leftmost). The outer frame radius of R = 30cm = 3h, and move limit of =
is fixed to be solid and is non-designable. Note 0.2 have been used. The ratio between the stiffness
that if we do not consider the non-designable of solid and void areas is selected as E : E = 109 .
outer frame, the optimization program will obvi- The obtained topologies at different iterations
ously shorten the frame. are shown in Figure 3.
240
Figure 3. The initial design and topologies of the

p = 1, the initial and final eigenfrequencies will
first example at different iteration numbers
change to 105.2 rad/s and 174.0 rad/s respectively.
5. MULTIPLE EIGENFREQUENCIES
By steadily increasing (or decreasing) an eigenfre-

quency, it is possible that its value reaches adjacent
eigenfrequencies resulting in multiple eigenfre-
quencies. The problem of multiple eigenvalues
in structural optimization was first addressed by
Olhoff and Rasmussen (1977). It is shown by Haug
and Rousselet (1980) that the multiple eigenvalues
are not differentiable in the Frchet sense and can
Figure 4 shows the values of the fundamental only be expected to be Gteaux (directionally) dif-
frequency and the first eigenmode. The funda- ferentiable. This finding rules out the validity of
mental frequency has increased from 60.4 rad/s sensitivities calculated by Equation (1.18) in case
in initial design to 154.1 rad/s after 85 iterations of multiple eigenfrequencies. Ignoring this fact in
showing 155% increase. Note that the values of topology optimization will result in oscillation of
the fundamental frequency also depend on the
penalty factor. If we analyze these designs with
Figure 4. Evolution history of the fundamental frequency (a) and the first eigenmode of the optimal
design (b) in example 1
241
the objective function and suboptimal solutions modal sensitivities of Equation (1.18), decreases
as shown in the following example. the objective function instead of increasing it.
This produces oscillation after the point of coalesce
Example 2: Clamped-Clamped Beam and the algorithm converges at a suboptimal solu-
tion (Figure 5b).
Consider the problem of maximization of the
fundamental frequency of a clamped-clamped 5.1. Simple Approaches to Avoid
beam with volume fraction of 50% and material Multiple Eigenfrequencies
properties similar to example 1. All algorithmic
parameters are similar to example 1. The structure A number of simple approaches can be used to
has been discretized into 30240 identical 4-node avoid multiple eigenfrequencies. Kosaka and
bi-linear square elements. The initial and the fi- Swan (1999) proposed a symmetry reduction ap-
nal solutions and the evolution of the first three proach in which a symmetry condition is imposed
eigenfrequencies are shown in Figure 5. on the design variables to ensure a symmetric
It can be seen that after 29 iterations, the first solution. In their paper, Kosaka and Swan (1999)
two eigenfrequencies coalesce. Using the single noted that in a symmetric structure, the multiple
Figure 5. Example 2 without treatment: initial design (a), final solution (b), and evolution of the eigen-
frequencies (c)
242
eigenvalues are differentiable (in the Frchet max

x 1 ,x 2 ,,x N
sense). Note that in this approach the symmetry
such that j j , j = 1, 2, , N d
reduction is not applied in the analysis and the
analysis is based on the full structure. Using K j = j M j , j = 1, 2, , N d
symmetry reduced structure for analysis cannot N
be validated in eigenfrequency optimization since x v

e =1
e e v
the eigenmodes are not necessarily symmetric
0 xe 1, e = 1, 2, , N
even for a symmetric structure.
(1.32)
Another approach to avoid repeated eigenfre-
quencies is to add an extra condition to the optimi-
with < 1, for example = 0.95. This formulation
zation problem ensuring that the eigenfrequencies
is known as bound formulation. Note that with =
are distant from each other. For example one may
1, Problem (1.32) is equivalent to Problem (1.15).
reformulate problem (1.15) as follows (Bendse
It is also possible to turn around the multiple
and Sigmund 2003).
eigenvalue problem by including adjacent eigen-
frequencies in the objective function. For example
Yang et al. (1999b) used the arithmetic mean of
the eigenvalues as the objective function when
Figure 6. Solving example 2, using mean eigenfrequency as objective function: final solution (a) and
evolution of the objective function and the first three eigenfrequencies (b)
243
they fell within a small distance from each other. all of its components being zero except for its i-th
A more generalized mean eigenvalue objective component which is 1.
function has been considered by Ma et al. (1995). Due to this perturbation, the stiffness and mass
For example considering the harmonic mean of matrices will change to
the first two eigenvalues, we can define the fol-
lowing objective function K
K(x + xi ) = K + + o(),
x i
1 (1.36)
= 2 (11 + 21 ) (1.33) M
M(x + xi ) = M + + o()
x i
Using this objective function in example 2
results in a smooth increase for the first two ei- and the eigenvalues and the eigenvectors will
genvalues. Figure 6 shows the final solution and change to
the evolution history of the first three eigenvalues
considering the objective function defined in j (x + xi ) = + j + o(), j = 1, , m
Equation (1.33) in example 2. (1.37)
5.2. Sensitivity Analysis of Multiple j + j + o(),

j (x + xi ) = j = 1, , m
Eigenvalues (1.38)
It is also possible to solve the multimodal eigen- where j and j are the unknown sensitivities
frequency optimization problems directly. Using of the multiple eigenvalues and eigenvectors
a perturbation technique, Bratus and Seyranian respectively. o() indicates higher order terms.
(1983) calculated the sensitivities of multiple Note that j j(x, xi) and j j(x, xi), i.e. the
eigenvalues. Here we follow Seyranian et al. sensitivities depend on xi.
(1994) and Lund (1994) to present the sensitivity Using the perturbed values of Eqs. (1.36),
analysis of multiple eigenfrequencies. (1.37), and (1.38) in the main eigenvalue problem,
Assume an m-fold multiple eigenvalue after ignoring the higher terms, one obtains
j = , j = 1, , m (1.34) K
M
x x j + (K M) j = j M j

i i
Due to multiplicity, any linear combination (1.39)

of the corresponding eigenvectors will satisfy
the main eigenvalue problem Kj = jMj. Now Premultiplying Equation (1.39) by
assume the following linear combination Ts , s = 1, , m , the second term in the left-hand
side will cancel out and one obtains the following
m
m equations
j = jk k ,
j = 1, , m (1.35)
k =1
K M
Ts
j = j Ts M
j, s = 1, , m
where the coefficients jk are unknown. If we x i x i
apply a perturbation to the i-th optimization (1.40)
variable the vector of design variables changes
to x + xi where xi denotes a vector of size N,
244
We now substitute Equation (1.35) in Equation 5.3. Solution Algorithm

(1.40) and use Equation (1.6) to write
Assume that is an Ndm matrix whose columns
K are the eigenvectors 1,, m. Also consider the
M
m
jk Ts
x x k j sk = 0, s = 1, , m
k =1 i i vector d = (d1,d2,,dN)T defined as
(1.41)
K M
de = , e = 1, , N (1.45)
This linear system can be solved to calculate the xe xe
coefficients jk. A non-trivial solution only exists if
K Then the subeigenvalue problem of Equation

det Ts M s, k = 1, ,m (1.43) can be stated in the following matrix form
x x k j sk = 0,
i i
(1.42) A I = 0 (1.46)
This subeigenvalue problem can be solved to

calculate the sensitivities j, j = 1,,m due to the where A = TdTe is a symmetric mm matrix
increment of the i-th optimization variable. and Imm is the unity matrix.
If the vector of design variables undergo an Following Cox and Overton (1992) and Over-
increment of the form e with e = (x1, x2,, ton (1992), the necessary optimality conditions to
xN)T and ||e|| = 1, one can easily generalize Equa- solve problem (1.15) is that there exists an mm
tion (1.42) to symmetric positive semidefinite matrix with
trace() = 1 such that
det fskT e sk = 0, s, k = 1, , m (1.43)
= : (T d ) v =
De e e e
N N
where v xeve = 0; v xeve 0; 0
e =1
e =1

( (
fskT = Ts K
)
xM1 k , Ts ( K
)
xM2 k , , Ts ( K
) )
xMN k
T
, xe = 0 e 0
x 1 x 2 x N
s, k = 1, , m 0 < xe < 1 e = 0
(1.44) xe = 1 e 0
e = 1, , N
are known as the generalized gradient vectors (1.47)
(Seyranian et al. 1994). Note that fsk are vectors
of length N, thus fskT e are scalars. Also note that where the Frobenius matrix inner product is de-
due to symmetry of the stiffness and mass matri- fined as A:B = trace(ATB).
ces fsk = fks. The proof of optimality conditions in (1.47)
The solutions of subeigenvalue problem of will not be presented here. Enthusiast reader is
Equation (1.43) are the sensitivities of the multiple referred to Cox and Overton (1992), Overton
eigenvalues. This equation was initially introduced (1992), and Seyranian et al. (1994).
by Bratus and Seyranian (1983). Note that for the case of simple eigenvalues (m
= 1), one should have = 1 and optimality condi-
tions in (1.47) reduce to (1.21). Comparing (1.47)
with (1.21), one may note that the only difference
is that the sensitivities /xe in (1.21) have been
245
replaced by :(Tde) in (1.47). The only issue following quadratic equation which is emerged
here is to find a suitable matrix . from Equation (1.43)
Similar to Equation (1.24), we consider the
following increment vector f T x f12T x
det 11 T = 0 (1.54)
f12x T
f22 x
= : (T d ) v ,
xe = D e = 1, , N
e e e
(1.48)
We are interested in finding ij such that
To illustrate the calculation of , we consider Equation (1.54) results in two positive eigenval-
the simplest multiple case of m = 2. We assume ues 1 > 0 and 2 > 0. There are several ways
that the two eigenvalues are repeated if to achieve this. Here we assume the following
2 1
T
1 = f11 x = 1 (1.55)
(1.49)
1
2 1
T
2 = f22 x = 1 1
(1.56)
with being a small positive tolerance. In this
case we have T
f12 x = 0 (1.57)
= (1 , 2 ) (1.50)
Equation (1.57) implies that the matrix in Equa-
tion (1.54) is diagonal, hence the eigenvalues are
We also introduce the following positive equivalent to the diagonal terms as stated in Eqs.
semidefinite symmetric matrix (1.55) and (1.56). In Equation (1.55) we consid-
ered an increase of 1 for the lowest eigenvalue.

12 If the two eigenvalues are different, the increase
= 11
(1.51)
12 22 assigned to the second eigenvalue in Equation
(1.56) will be slightly lower than 1. This is to
reduce the difference between the two repeated
based on which we define
eigenvalues.
Substituting Equation (1.53) in Eqs. (1.55) to

= (1.52) (1.57), we obtain a set of three equations which
trace( ) can be solved to yield the three unknown coef-
ficients 11

, 22 , and 22 . These equations can
to ensure that trace() = 1. be summarized as
Substituting Equation (1.51) in Equation (1.48)
and using Eqs. (1.45) and (1.44), we may rewrite fT f T
f11T f12
1 + f11T v
11 11 f11 f22

11

Equation (1.48) in the following form T
f22 f22 T 2 1 T
f22 f12 22 = 1 1 + f22 v

symm. T
f12

f12 212

T

f12 v

x = 11
f11 + 212 f12 + 22 f22 v (1.53)
(1.58)
The change in the multiple eigenvalues 1 By solving Equation (1.58) one finds *
and 2 due to x can be calculated by solving the which is used in Equation (1.52) to find and
246
subsequently x from Equation (1.53). Like the than increasing them. For an objective function
single modal case, to find the value of the Lagrange defined as a combination of natural frequencies,
multiplier an inner bisection loop can be used. the optimization problem can usually be addressed
Figure 7 shows the final solution and the evolu- by minimal modification of the eigenfrequency
tion history of the first three eigenfrequencies of the maximization problem. Typically one just needs
problem of example 2 using the above approach. to update the sensitivities. One example of such
It can be seen that the multiple eigenvalues evolve objective functions was defined in Equation (1.33)
smoothly and a better solution is achieved. and dealt with in the preceding section. Other
Table 1 compares the final value of the first examples include maximizing the gap between
three eigenfrequencies of example 2 obtained two natural frequencies (see e.g. Du and Olhoff
using the three approaches considered here. As 2007 and Zhao et al. 1997) or designing structures
expected, using the multiple eigenvalue sensi- with a specified set of frequencies or eigenmode
tivities yields the best result. shapes (see e.g. Xie and Steven 1996, Yang et al.
1999b, Maeda et al. 2006) among others.
A common practical case is where the excitation
6. CONTROLLING THE NATURAL frequency is known and it is desired to move the
FREQUENCIES natural frequencies as far away as possible from
the excitation frequency. A suitable objective
In dynamic design of structures one usually re- function can be defined as
quires to control the natural frequencies rather
Figure 7. Solving example 2, using multiple eigenvalue sensitivities: final solution (a) and evolution of
the first three eigenfrequencies (b)
247
f = j2 2 (1.59) Example 3: Planar Frame

j J
with Non-Structural Mass
in which is the excitation frequency and A three-level planar frame is considered with 6
J {1, , N d } is a set of natural frequencies non-structural masses of 20,000kg each attached
considered. If we only consider the closest natu- to it as shown in Figure 8a. The ratio between the
ral frequencies to , the problem reduces to stiffness of solid and void areas is reduced to
maximizing the gap between the adjacent natural E : E = 100 . All other parameters are similar to
frequencies. Maximizing the fundamental fre- example 1.
quency is a special case of this problem with We fist maximize the fundamental frequency
= 0. of the frame. The evolution history of the funda-
Maximizing the gap between two natural fre- mental frequency and the final solution are shown
quencies can lead to multiple eigenfrequencies in Figures 8b and 8c respectively. The fundamen-
(Du and Olhoff 2007). The following example tal frequency has increased by 85% from 40.2 to
illustrates this. 74.6rad/s.
Table 1. Comparison of the results of the three approaches used to solve example 2
Approach 1 (rad/s) 2 (rad/s) 3 (rad/s)

Using single eigenvalue sensitivities (1.18) 257.1 267.1 330.9
Using the mean eigenvalue (1.33) as objective function 248.0 424.4 540.2
Using multiple eigenvalue sensitivities (1.43) 273.5 284.0 337.6
Figure 8. Example 3: initial design (a), final solution (b), and evolution of the fundamental frequency (c)
248
The first four natural frequencies of the initial this multiple eigenfrequency has been handled
structure are 1 = 40.2, 2 = 163, 3 = 215, and well by the algorithm.
4 = 326 (rad/s). We now assume an excitation
frequency of 175 rad/s which falls between the
second and the third natural frequencies, and try 7. FORCED VIBRATION
to move the natural frequencies of the structure
away from this frequency. The problem can be In previous sections we have focused on free
simplified to maximizing the gap between the vibration and did not include external forces in
second and the third natural frequencies, i.e., our formulations. This section deals with topol-
maximizing the following objective function ogy optimization in forced vibration where the
external dynamic forces are explicitly considered
f = 32 22 (1.60) in the problem formulation. Similar to previous
sections we ignore damping effects, uncertainties
of the forces, geometry, and materials and consider
The solution procedure explained in section
linear elastic material behavior.
5.3 has been adopted. The optimal topology and
Applications of topology optimization in
the evolution history of the first four natural fre-
forced vibration have been initially studied by
quencies are depicted in Figure 9. The optimiza-
Ma et al. (1993) and Ma et al. (1995) using a ho-
tion algorithm, tries to increase the third natural
mogenization approach. Ma et al. (1995) briefly
frequency while decreasing the second natural
discussed the forced vibration problem under pe-
frequency. After nearly 15 iterations, the third
riodic loads and defined the dynamic compliance
eigenfrequency coalesced with the fourth one but
Figure 9. Maximizing the gap between the third and the second eigenfrequencies of the frame of example
3: final solution (a) and evolution of the first four eigenfrequencies (b). The first four eigenfrequencies
of the optimal structure are 1 = 74.6, 2 = 190, 3 = 397, and 4 = 408 (rad/s).
249
as the objective function to be minimized. Min et 7.1. Objective Function and

al. (1999) minimized the dynamic compliance for Problem Formulation
structures under impulsive loads. The topology
design of structures under periodic loads has been Under static loads, the compliance defined as
studied more extensively by Jog (2002) where he
proposed a new positive-definite definition of c = pT u (1.65)
dynamic compliance as the average input power
over a cycle. Jog (2002) also studied the problem is proportional to the strain energy of the struc-
of minimizing the vibration amplitude at certain ture and is a typical objective function used in
control points. Topology design of structures topology optimization of structures. Minimizing
subjected to design-dependent dynamic loads the compliance maximizes the stiffness of the
(e.g. hydrodynamic pressure loading) has been structure. Under dynamic loads, the value of c
addressed by Olhoff and Du (2005). In more recent defined in Equation (1.65) varies with time. For
publications in this area, alternative approaches structures subjected to periodic loads, we consider
in using topology optimization to control the the average of c over a cycle, i.e.
structural responses in a frequency interval has
been studied by Jensen (2007) and Yoon (2010b). 2 /
Consider a harmonic external force of form c =
2 0
pT udt (1.66)
p = pC cos t (1.61)
as the objective function to be minimized. Here T
= 2/ is the time period. Note that this measure
where pC does not depend on time. The equation is not always positive, and thus, in problem for-
of motion for a discretized undamped system in mulation, one should consider the absolute value
forced vibration takes the form (or square) of average compliance as the objective
function. Otherwise, for c < 0 , the optimization
+ Ku = pC cos t
Mu (1.62) algorithm will push the structure towards resonance.
In absence of damping, u = uC cost. Using this
To solve this problem, we consider u = uC cost and p = pC cost in Equation (1.66), we can write
using which in Equation (1.62) gives
pTC uC 2 / 1 T
(K M ) u
2
C = pC (1.63)
c =
2 0
cos2 t dt =
2
pC uC
(1.67)
or The average compliance minimization problem

can now be formulated as follows
(K M ) u = p
2
(1.64)
min cm = 2 c = pTC uC
x 1 ,x 2 ,,x N
Comparing with equilibrium equation in
static state, K 2 M can be termed as dy-
such that (K M ) u 2
C = pC
N
namic stiffness. Note that unlike the static stiff-
ness, the dynamic stiffness matrix is not neces-
x e ve v
e =1
sarily positive definite. 0 < x min xe 1, e = 1, 2, , N

(1.68)
250
7.2. Sensitivity Analysis cm

De = ve = e , e = 1, , N
x e
In order to calculate the sensitivities, we rewrite (1.74)
the dynamic compliance by adding an (arbitrary)
adjoint vector multiplied by a zero function with additional conditions similar to (1.21). The
negative sign for cm x in Equation (1.74) is
T (K 2 M) uC pC
cm = pTC uC + u added because it relates to a minimization problem.

(1.69)
7.3. Examples
Differentiating with respect to the design
variables and rearranging the terms, we obtain The frame in example 3 is considered under
forced vibration. It is assumed that 6 identical
cm u
T (K 2 M) C + u
K M horizontal periodic loads of magnitude p1 = 500
= sign(c )pTC + u T 2 uC
xe x x x kN and frequency of are applied at locations of
(1.70) the concentrated masses as shown in Figure 10a.
Note that for linear elastic materials, changing the
where sign() is the sign function. Sensitivities of force magnitude will only change the values of
the dynamic compliance can now be written as the objective function but the evolution pattern of
the objective function values and the final topol-
cm K
M ogy will remain unchanged. This is not true for
T
=u 2 uC (1.71)
xe x x nonlinear problems.
Deformed shapes of optimum topologies ob-
in which the adjoint vector is selected such that tained for different load frequencies are shown in
Figure 10b-e. It can be seen that for high frequen-
cies, the optimum material distribution tends to
(K M) u = sign(c )p
2
C (1.72)
form damping masses.
When the input frequency is greater than the
Comparing Equation (1.72) with Equation fundamental frequency of the initial structure, the
(1.63), the adjoint vector is found to be optimization algorithm reduces the fundamental
= sign(pTC uC )uC which can be substituted
u frequency. This increases the static compliance
in Equation (1.71) to simplify the latter to and can lead to disintegrated designs (clearly
observable in Figure 10e). To prevent this disin-
cm K
M tegration, one can introduce an upper bound
= sign(c )uTC 2 uC (1.73)
xe x x condition on static compliance in the problem
formulation (Olhoff and Du 2005). One can also
include the static compliance in the objective
Having the sensitivities calculated, an ap-
function as shown in Figure 10.
propriate solution method such as the method of
When the structure is subjected to a number
moving asymptotes (MMA) can be employed to
of loads with different frequencies, one can define
solve the minimization problem. One can also
a multi-objective optimization problem to handle
use the OC-based solution procedure proposed
the case. This is of practical importance, for ex-
in section 4.2. The optimality criteria to solve
ample when one approximates a periodic load
Problem (1.68) can be expressed as
using Fourier series. In the following example,
251
Figure 10. Minimizing the dynamic compliance of the frame of example 3 under periodic loading: ex-
ternal periodic loads (a), and deformed shape of final topologies for different input frequencies; = 0
(static loading) (b), = 60 rad/s (c), = 175 rad/s (d), and = 330 rad/s (e). The first four natural
frequencies of the initial structure are 1 = 40.2, 2 = 163, 3 = 215, and 4 = 326 (rad/s).
we consider the structure to be subjected to a The previous example is considered under

periodic load p1 = pC1 cos t and a static load the dynamic loads (p1) in Figure 10a and the
p2 (with frequency of zero). The objective function static loads (p2) shown in Figure 11a. The ratio
is considered as between the magnitude of the loads is p2:p1 = 5.
Various optimal topologies resulted from using
cm = pTC1 uC1 + pT2 u 2 (1.75) various input frequencies of p1 are illustrated in
Figure 11b-e.
In a similar manner one can optimize the topol-
where ogy of a structure under any combination of static
and periodic loads with different frequencies.
(K M ) u
2
C1 = pC1 and Ku 2 = p2
(1.76)
8. OTHER CONSIDERATIONS
Sensitivities of this objective function can be
calculated as So far we have simplified the optimization prob-
lems by ignoring the damping effects and non-
cm K M T K linear material behavior. We have also limited
= sign(pTC1 uC1 )uTC1 2 uC u 2 u2
x x x 1 x our study to small deformations. Considering
(1.77) large deformations will lead to geometrically
non-linear problems.
252
Figure 11. Minimizing the dynamic compliance of the frame of example 3 under a combination of periodic
loading of Figure 10a and static loading: external static loads (a), and final topologies for different input
frequencies of the periodic load; = 0 (b), = 60 rad/s (c), = 175 rad/s (d), and = 330 rad/s (e)
Including either geometrical or material non- proves to be useful in topology optimization of

linearity in problems causes the stiffness matrix non-linear structures under dynamic loads (Yoon
to be load-dependent. The equilibrium equations 2010a, 2011). Another approach to solve this issue
in such problems are commonly solved by the is eliminating the void elements (see Section 9.1).
Newton-Raphson method. Finding eigenfrequen- Considering damping effects will also change
cies will require a subsequent modal analysis. the sensitivities of the objective functions consid-
Sensitivity analysis of a general displacement- ered. In the following we update the previously
based functional in a geometrically and materially derived sensitivities in presence of damping.
non-linear system has been formulated by Jung
and Gea (2004). This formulation can be used to 8.1. Forced Vibration with Damping
calculate sensitivities of compliance-like objective
functions. Similar procedure of deriving the sen- The equation of motion under a periodic load
sitivities of an energy functional for a non-linear takes the form
system is presented in Section 9.1.
In geometrically non-linear problems, the ex- + Cu + Ku = p = pC cos t
Mu (1.78)
tremely soft void elements of the SIMP material
model can be troublesome showing zero or even Introducing the complex displacement
negative tangent stiffness and causing serious z = (zC + izS )e it , and assuming u to be the real
convergence problems (see e.g. Buhl et al. 2000).
part of z, we rewrite this equation in the complex
A new approach called Element Connectivity
space as
Parameterization has been proposed by Yoon and
Kim (2005) to address this problem. This approach
253
Mz + Cz + Kz = pC e it (1.79) But from Equation (1.82),

Separating the real and imaginary parts, we K zC = 0
zC + K zC = K z
K C
obtain x e x e x e x e
(1.87)
BzC CzS = pc
(1.80) which can be used in Equation (1.86) to yield
BzS + CzC = 0
c 1 z 1 zC 1 K
where B = K 2M. Calculating zS from the = pTC C = zTC K = zTC zC
xe 2 xe 2 xe 2 xe
second equation in (1.80) yields (1.88)
zS = B1CzC (1.81) / x . Before

We thus need to calculate K e
proceeding we note that
After substituting in the first equation in
(1.80), we get B1B = I
B1
B + B1
B
=0
B1
= B1
B 1
B
xe xe xe xe
(1.89)
KzC = pC (1.82)
in which we used the fact that B is symmetric,
where i.e. BT = B. Differentiating Equation (1.83) and
making use of Equation (1.89) and symmetry of
K = B + 2CB1C (1.83) B and C, we obtain
The displacement vector u is the real part of z, so K B B 1 C 1

= 2CB1 B C + 22 B C
xe xe xe xe
u = ((zC + izS )e it ) = (cos t ) zC (sin t ) zS (1.90)
(1.84)
Pre- and post-multiplying by zC and using
Using this and p = pC cost in Equation (1.66), Equation (1.81), we achieve
we can write

K B B C
zTC zC = zTC zC zTS zS 2zTC zS
2 / 1 T xe xe xe xe
c = (p
T
z cos2 t pTC zs cos t sin t )dt = pC zC
2 0
C C
2 (1.91)
(1.85)
Substituting in Equation (1.88) and replacing
Differentiating with respect to design variable B by K 2M, we obtain
xe, we obtain
c 1 K
M K M
c 1 z = zTS 2 zS zTC 2 zC
= pTC C (1.86) xe 2 xe xe xe xe
xe 2 xe C
+ zTC zS
x e
(1.92)
254
If we now define the objective function as zS = B1CzC (1.97)

cm = 2 c , for the sensitivities of this function
we can write which after substituting in the first equation of
(1.95), gives
K
zT 2 M

S x x zS = 0, = B + 2CB1C
cm e e Kz K (1.98)
= sign(c ) C
x e T K M C
zC 2 zC + 2zTC zS
xe xe xe
Using Equation (1.98), we can now write
(1.93)

K zC
zC = K zTC
K zC = 0
zC = zTC K
Ignoring damping effects, we have C = 0, zS xe xe xe xe
= 0, and zC = uC, and thus the above equation (1.99)
reduces to Equation (1.73).
The vectors zC and zS in Equation (1.93) can Differentiating Equation (1.98), we get
be obtained by solving Equation (1.80). Given
the damping matrix C, the term C/xe is also K B C 1
calculable. For example, assuming the Reyliegh = + 2 2 B C
xe xe xe
damping formulation of Equation (1.2), we have
B 1
C / xe = aM M / xe + aK K / xe .Having 2
CB 1
B C + 2 1
CB C
xe xe
the sensitivities, one can solve the optimization
(1.100)
problem using a suitable solution algorithm.
Pre- and post-multiplying this equation by zC

8.2. Free Vibration with Damping
and using Equation (1.97), we obtain
Using a similar approach followed in the previous

K B C
section, in case of free vibration, Eqs. (1.79) and zTC zC = zTC zC 2zC zS
(1.80) need to be changed to xe xe xe

T B T
z zS 2 zC CzS = 0
Mz + Cz + Kz = 0, z = (zC + izS )e i t , S
xe xe
(1.94) (1.101)
BzC CzS = 0 Differentiating Equation (1.96), we have

(1.95)
BzS + CzC = 0
B K M
= 2 2 M
xe xe xe xe
respectively where is a natural frequency and
(1.102)
B = K 2M (1.96)
Substituting this equation in Equation (1.101)
and rearranging yields
From the second equation in (1.95), for zS we
can write
255
K M K
zTC 2
z zTS 2 M

C (Oh et al. 2009) as depicted in Figure 12a. In
xe xe C x x zS 2zC x zS
= e e e
these installations the device will deform mainly
xe 2 (zTC CzS + zTC MzC + zTS MzS )
in shear (Figure 12a). The design domain is the
(1.103)
inner part of the web as shown in Figure 12b. The
two 15 mm strips on the boundaries of the web
If we normalize the vectors zC and zS with re-
are non-designable.
spect to M, for sensitivities of = 2, we can write
We use the BESO method here and introduce
K M K 2 M
C a simple technique to solve shape optimization
zTC 2 z zTS
x x zS 2zC x zS
xe xe C e e e problems using BESO. Restricting the topology
= 2 =
xe x e zTC CzS

+2 of the design and performing shape optimization
(1.104) instead of topology optimization is useful when
the fabrication cost is an important factor. We also
It is easy to verify that in the without damping address a simple approach to obtain periodic
this equation reduces to Equation(1.18). Having designs which are produced by repeating a fixed
the sensitivities, one can solve the optimization pattern.
problem using a suitable solution algorithm. Considering a volume constraint, the energy
absorption maximization problem can be stated as
9. MAXIMIZING ENERGY max p

x 1 ,x 2 ,,x N
ABSORPTION
such that r = pp
=0
N
Apart from controlling natural frequencies and dy-
namic compliance, improving energy absorption
x v
e =1
e e =v
characteristics is also of significant importance 0 xe 1, e = 1, 2, , N

in seismic design of structures. In recent years, (1.105)
active and passive energy dissipating devices
have been widely studied and utilized to increase
energy absorption of structural systems (Soong and where p is the total plastic dissipation. Because
Spencer 2002). Topology optimization can be used the problem involves plastic behavior, one needs to
to maximize energy absorption of these devices. solve a non-linear equilibrium system of the form
In this section we consider the problem of maxi-
mizing the energy absorption of passive energy = 0
r = pp (1.106)
dissipaters which make use of yield deformation
of metals to mitigate the excitation energy. These
This requires an iterative solver to eliminate
kinds of energy dissipating devices are popular
the residual force vector r defined as the differ-
due to low cost of fabrication and maintenance
ence between the external and internal force
and easy installation (Ghabraie et al. 2010).
vectors, p and p respectively.
We consider an energy damping device which
The internal force vector is defined as
is made of a 100 mm-long cut of a standard struc-
tural wide-flange section with depth, flange width, N N
web thickness, and flange thickness of 161.8, = GTe BTe e d v = GTe qe
p
152.2, 8, and 11.5 mm respectively. This device e =1 v
e
e =1
can be installed in braces connections (Chan and (1.107)

Albermani 2008) or beam-column connections
256
Figure 12. A passive energy dissipating device: (a) Installation and deformation; (b) designable and
non-designable domains and loading. Dimensions are in mm
where Ge is the matrix that transforms the local We consider the energy damping device to
nodal values of element e to the global nodal undergo a full loading cycle consisting of an
values, B is the strain-displacement matrix and upward displacement of 10 mm followed by a
represents the local element stress vector. The 20 mm downward displacement and finally a 10
stress vector can be expressed as mm upward displacement to its original position
(Figure 12b). Due to this loading, the total plastic
e = De e (1.108) dissipation would be equal to the total strain energy
N
with being the strain vector and D the stress-strain pdu =
matrix. Similar to Eqs. (1.10) and (1.12) we can
p = = pe d ue (1.110)
e =1
define the relationship between D and the design

variables in the following form Using the trapezoidal rule for integration,
(1.110) can be written in the form
De (xe ) = xe De (1.109)
1 n
p = lim
n 2
(uT(i ) uT(i 1) ) (p(i ) + p(i 1) )
in which De is the stress-strain matrix of element i =1
(1.111)
e in its solid state.
257
where the subscripts in parentheses indicate the or calculated rigorously. Either way, they are
integration divisions. defined such that a higher sensitivity number
represents higher efficiency.
9.1. Bidirectional Evolutionary Considering the definition of the objective
Structural Optimization (BESO) function in Equation (1.111), the sensitivities of
this function can be calculated as
The BESO method has been used in eigenfre-
quency optimization (see e.g. Yang et al. 1999b, p 1 n u(i ) u(i 1)
T T
p(i ) p(i 1)
= lim T T
(p(i ) + p(i 1) ) + (u(i ) u(i 1) ) +
Huang et al. 2010) as well as in maximizing xe n 2 i =1 xe xe xe xe

energy absorption (see e.g. Huang et al. 2007, (1.112)
Ghabraie et al. 2010). This method solves the
topology optimization problem in two steps. At The first term in the right hand side cancels out
first, the optimization algorithm finds the most because on the boundaries with essential boundary
and least efficient elements to be added and re- conditions u/x = 0 and elsewhere p = 0. Hence
moved respectively. Then, it adjusts the number the above equation reduces to
of adding and removing elements based on the
volume constraint. p 1 n p(i ) p(i 1)
= lim ( uT(i ) uT(i 1) ) +
This method is capable of totally removing the xe n 2
i =1 xe xe
elements, hence one does not require to represent (1.113)
voids with a very soft material. This approach,
usually known as hard kill approach, results in On the other hand, differentiating Equation
faster solutions for only solid elements remain in (1.106) and using Eqs. (1.107) to (1.109), we obtain
the finite element model. Moreover, this approach
works well in geometrically non-linear problems p p

= = GTe qe , e = 1, , N (1.114)
as it is not prone to the instabilities caused by soft xe xe
elements in the SIMP material model (Buhl et al.
2000, Yoon and Kim 2005). Zhou and Rozvany
Substituting Equation (1.114) in Equation
(2001) showed that in certain cases the hard kill
(1.113), we can write
approach may result in non-optimal solutions and
thus this approach need to be applied with care. In
p 1 n
shape optimization, however, using the hard kill
xe
= lim
n 2
(uT(i ) uT(i 1) ) GTe (qe(i ) + qe(i 1) )
approach will not cause such problems. i =1
Another advantage of the BESO method is (1.115)

that the solutions will not contain any intermedi-
ate design variables (0 < x < 1) or grey areas. In Using the trapezoidal numerical integration
this method the boxing constraints 0 xe 1, e = scheme and recalling the definition of the strain
1,,N change to binary constraints of the form energy in Eqs. (1.110) and (1.111), the above
xe {0,1}, e = 1,,N. This is particularly help- equation reduces to
ful if one wants to impose shape restrictions as
the boundaries of solids and voids can be clearly p n
= lim (e(i ) e(i 1) ) = e (1.116)
defined in the black-white solutions of BESO. xe n
i =1
In BESO, the so-called sensitivity numbers are

used to evaluate the efficiency of the elements.
Sensitivity numbers might be assigned intuitively
258
where e is the total strain energy of element e numbers and the algorithm tries to move the
upon completion of the load cycle. Huang and Xie volume towards the volume limit gradually. Thus,
(2008) derived these sensitivities using the adjoint if the current volume is bigger than v the algorithm
method and verified it using a simple example. will increase the number of removing elements
We may now define the sensitivity numbers and vice versa. The procedure continues until no
for problem (1.105) as follows further significant improvement can be achieved.
The algorithms to update the solutions in the
p BESO method have been improved over time.
e = = e , e = 1, 2, , N (1.117)
xe One of the most recent algorithms is proposed by
Huang and Xie (2007). In this algorithm, at each
iteration k, the target volume of the next iteration
Based on this definition and using first order
is calculated using a small positive controlling
approximation, we can write
parameter called the evolutionary volume ratio (Rv)
N
p = e xe (1.118) v (k +1) = v (k ) (1 + sign(v v (k ) )Rv )
e =1
(1.120)
Note that in Equation (1.118) adding element where superscripts enclosed in parentheses in-
e will be reflected by xe = 1 0 = 1 and remov- dicate the iteration number. Then the number of
ing it results in xe = 1. Thus during the solution adding and removing elements are calculated such
procedure, if one introduces the ath element and that the volume of the next design becomes equal
removes the rth element, the change in the objec- to v(k+1) and the total number of added elements
tive function can be estimated as do not exceed vadd = vt Ra in which Ra is another
controlling parameter known as the maximum
p p allowable admission ratio.
p = = a r (1.119)
x a x r If one starts BESO with an initial design vol-
ume equal to v , the volume will be kept constant
As we are interested in maximizing p it is during the optimization procedure and the number
clear from Equation (1.119) that the elements with of adding and removing elements at each iteration
highest sensitivity numbers should be added to would be equal to each other. In this case Rv will
the design domain while the elements with lowest have no effect on the optimization procedure and
sensitivity numbers should be removed. the maximum number of adding and removing
elements is only controlled by Ra. Thus with a
9.2. Adding and Removing fixed volume, the effect of the Ra factor is similar
the Elements to the move limit in Equation (1.28).
After ranking the efficiency of the elements, the 9.3. Mirroring and Filtering
algorithm should select the number of elements Sensitivity Numbers
to be added and removed such that the volume
constraint is satisfied. Generally in BESO, one Due to the nonlinear nature of Problem (1.105),
starts the solution with an initial volume which the loading sequence affects the mechanical re-
is not necessarily equal to the volume limit v . sponses. As a result, the optimal shape flips by
The design is then updated using the sensitivity mirroring the loading sequence. In real life, how-
259
ever, the direction of the load is uncertain. Thus 9.4. Restricting the Topology
one needs to consider two displacement cycles: a
cycle as well as a cycle. But knowing that Topology optimization techniques like BESO
the results of these two displacement cycles are can naturally introduce new holes or fill the
mirrored images of each other, it is not necessary current holes in the design domain. However
to analyze the model under both of these loading this behavior might produce complicated shapes
conditions. Instead, one can add the sensitivities which might be costly to fabricate. To prevent the
of the mirrored elements together to account for BESO algorithm from introducing new holes, we
both displacement cycles. The sensitivity numbers restrict the designable domain to the elements at
are thus corrected as the boundaries of the shape at each iteration. The
designable domain at each iteration is defined as
e = e + e (1.121)
D = {e i, j B : i, j e i j }
where e is the corrected sensitivity number of (1.123)

element e and e is the element which is located
at the same location as e in the mirrored model. where is the set of boundary nodes defined as
Like the SIMP method, the BESO method is
also prone to the formation of checkerboard pat- B = {i em S, ev V : i em ev }
terns and mesh dependency. A filtering technique, (1.124)
similar to Equation (1.29) can be employed to
overcome these problems in the BESO algorithm with and denoting the sets of solid and void
elements respectively.
N
w j ij 9.5. Numerical Examples

=
e =1
(1.122)
e N
w
j =1
ij In the following numerical examples we fix the
volume to simplify the approach. Another benefit
of using fixed volume is that the results of different
Another significance of filtering in the BESO iterations are comparable to each other.
method is extrapolating the sensitivity numbers The modulus of elasticity of the material is
to void elements. If one uses a hard kill approach, considered as 206.1 GPa and the tensile yield
void elements are removed from the structure and stress is assumed to be 334 MPa. The material
their sensitivities cannot be evaluated directly. model considered is depicted in Figure 13.
In other words, all void elements will have a
sensitivity number of zero. The filtering scheme Example 4: Simple Damper
in Equation (1.122) extrapolates the sensitivities
to the void elements in the neighborhood of the We use the proposed BESO algorithm to optimize
solid elements. This extrapolation leads the BESO the shape of a simple damper. The target volume
algorithm to add the elements in the vicinity of the is 82% of the design domain. The initial and the
elements with high sensitivity numbers. final solutions are depicted in Figures 14a and 14b
respectively. The evolution history of the objective
function is plotted in Figure 14c. It can be verified
that the energy absorption of the optimal solution
260
Figure 13. The idealized material model used for

smooth the jagged boundaries of the solutions.
modeling the energy dissipating device
The smoothed versions of the initial and optimal
solution are analyzed and their stress distributions
are compared in Figure 15. It can be seen that the
optimal solution provides an even stress distribu-
tion and the stress concentration areas visible in
the initial design have been dissipated. This even
stress distribution improves the responses of this
design against low cycle fatigue. This has been
verified through experimental tests by Ghabraie
et al. (2010).
Figure 16 compares the force-displacement
curves of the initial and final solutions. The op-
is considerably higher (37.4% improvement) than timal solution shows a stiffer response than the
the initial design. initial design.
Example 5. Slit Damper

10. CONCLUSION
We now consider shape optimization of a slit
damper. The initial design is depicted in Figure This chapter reviewed the application of topol-
14d. In order to preserve the periodicity of the ogy optimization techniques in seismic design of
design, one needs to impose an additional con- structures. Two established topology optimization
straint to the optimization algorithm. To this end, methods, namely SIMP and BESO, have been in-
we partition the design domain into four cells. To troduced and their application has been illustrated
make the cells identical, the sensitivity numbers using numerical examples.
of elements are replaced by the mean value of Eigenfrequency optimization of linear elastic
sensitivity numbers of corresponding elements structures in free vibration has been addressed
in all cells. Putting this mathematically, we write using the SIMP method. Sensitivity analysis of
eigenfrequencies has been explained and a simple
1 N cell solution procedure has been presented based on
i =
N cell

j =1
i,j (1.125) optimality criteria. Possible numerical instabili-
ties have been mentioned and possible treatments
have been discussed.
where i is the average (corrected) sensitivity The problems involving multiple eigenfrequen-
number of element i in all cells, Ncell is the number cies have been considered and simple approaches
of cells, and i,j is the (original) sensitivity number to bypass these problems have been discussed.
of the element i in cell j. The sensitivities of multiple eigenfrequencies
The obtained solution is illustrated in Figure have been calculated and the optimality criteria
14e. The evolution history of the objective func- have been presented. A simple approach to solve
tion is plotted in Figure 14f. Again a significant these problems has been proposed and success-
improvement (64.3%) in energy absorption is fully applied to a simple problem. The problem of
observable. maximizing the gap between two eigenfrequen-
Ghabraie et al. (2010) have used a smooth- cies has also been addressed. This problem is of
ing postprocessor based on Bzier curves to practical significance when it is desired to push
261
Figure 14. Solutions of examples 4 and 5. The initial designs (a,d), the final designs (b,e), and the evolu-
tion history of the objective functions (c,f). Dark grey elements are non-designable
262
Figure 15. Comparing stress distribution in initial (a) and final (b) smoothed designs. Stresses are in MPa.
Figure 16. Comparing force-displacement graphs in initial and final smoothed designs
eigenfrequencies of a structure away from a given method. Topology design of structures under a
excitation frequency. combination of periodic (and static) loads with
Topology optimization of structures under different frequencies has been addressed. This
periodic loads has been discussed. An average is of practical importance, for example when a
dynamic compliance measure has been consid- periodic load is expanded using Fourier series.
ered as the objective function to be minimized. Issues and difficulties of considering the
Sensitivities of this objective function have been non-linear material and geometry behavior of
computed. Numerical examples have been solved the system have been briefly discussed. The
to illustrate the applications of the optimization sensitivities of eigenfrequencies and dynamic
263
compliance of structures in presence of damping Beyer, H.-G., & Sendhoff, B. (2007). Robust
have been derived. optimization A comprehensive survey. Com-
The BESO technique has been modified and puter Methods in Applied Mechanics and Engi-
used to maximize the energy absorption of a pas- neering, 196(3334), 31903218. doi:10.1016/j.
sive metallic damper. Sensitivity analysis of the cma.2007.03.003
non-linear system has been presented. Simple
Bratus, A. S., & Seyranian, A. P. (1983). Bi-
approaches to achieve shape optimization and
modal solutions in eigenvalue optimization
periodic solutions have been addressed. It has been
problems. Journal of Applied Mathematics and
illustrated that the optimized solution not only
Mechanics, 47(4), 451457. doi:10.1016/0021-
provides higher energy absorption capacities but
8928(83)90081-3
also offers smoother stress distribution resulting
in better fatigue resistance. Buhl, T., Pedersen, C. B. W., & Sigmund, O.
It has been shown that in seismic design of (2000). Stiffness design of geometrically nonlinear
structures, topology optimization techniques can structures using topology optimization. Struc-
be useful in both conceptual design of structural tural and Multidisciplinary Optimization, 19(2),
systems (e.g. maximization of fundamental fre- 93104. doi:10.1007/s001580050089
quency of a frame) and detailed design of structural
Chan, R. W. K., & Albermani, F. (2008). Ex-
members (e.g. maximization of energy absorption
perimental study of steel slit damper for passive
of passive dampers). These techniques are capable
energy dissipation. Engineering Structures, 30(4),
of dealing with different objective functions and
10581066. doi:10.1016/j.engstruct.2007.07.005
different material models.
Chopra, A. K. (1995). Dynamics of structures
Theory and applications to earthquake engi-
REFERENCES neering. Upper Saddle River, NJ: Prentice Hall
International.
Bendse, M. P. (1989). Optimal shape design as a
material distribution problem. Structural Optimi- Cox, S. J., & Overton, M. L. (1992). On the
zation, 1(4), 193202. doi:10.1007/BF01650949 optimal design of columns against buckling.
SIAM Journal on Mathematical Analysis, 23(2),
Bendse, M. P., & Kikuchi, N. (1988). Generat- 287325. doi:10.1137/0523015
ing optimal topologies in structural design using
a homogenization method. Computer Methods Culmann, K. (1866). Die graphische Statik.
in Applied Mechanics and Engineering, 71(2), Zrich, Switzerland: Mayer und Zeller.
197224. doi:10.1016/0045-7825(88)90086-2 Daz, A. R., & Kikuchi, N. (1992). Solutions
Bendse, M. P., & Sigmund, O. (1999). Material to shape and topology eigenvalue optimization
interpolation schemes in topology optimization. problems using a homogenization method. In-
Archive of Applied Mechanics, 69(910), 635654. ternational Journal for Numerical Methods in
Engineering, 35(7), 14871502. doi:10.1002/
Bendse, M. P., & Sigmund, O. (2003). Topology nme.1620350707
optimization - Theory, methods, and applications.
Berlin, Germany: Springer. Du, J., & Olhoff, N. (2007). Topological design
of freely vibrating continuum structures for
maximum values of simple and multiple eigen-
frequencies and frequency gaps. Structural and
doi:10.1007/s00158-007-0101-y
264
Elnashai, A. S., & Chryssanthopoulos, M. (1991). Jog, C. S. (2002). Topology design of structures
Effect of random material variability on seismic subjected to periodic loading. Journal of Sound
design parameters of steel frames. Earthquake and Vibration, 253(3), 687709. doi:10.1006/
Engineering & Structural Dynamics, 20(2), jsvi.2001.4075
101114. doi:10.1002/eqe.4290200202
Jung, D., & Gea, H. C. (2004). Topology optimi-
Ghabraie, K., Chan, R., Huang, X., & Xie, Y. M. zation of nonlinear structures. Finite Elements
(2010). Shape optimization of metallic yielding in Analysis and Design, 40(11), 14171427.
devices for passive mitigation of seismic energy. doi:10.1016/j.finel.2003.08.011
Engineering Structures, 32(8), 22582267.
Karush, W. (1939). Minima of functions of several
doi:10.1016/j.engstruct.2010.03.028
variables with inequalities as side constraints.
Haug, E. J., & Rousselet, B. (1980). Design sensi- Masters thesis, Department of Mathematics,
tivity analysis in structural mechanics. II: Eigen- University of Chicago, Chicago, Illinois.
value variations. Journal of Structural Mechanics,
Kharmanda, G., Olhoff, N., Mohamed, A., &
8(2), 161186. doi:10.1080/03601218008907358
Lemaire, M. (2004). Reliability-based topology
Huang, X., & Xie, Y. M. (2007). Convergent and optimization. Structural and Multidisciplinary
mesh-independent solutions for the bi-directional Optimization, 26(5), 295307. doi:10.1007/
evolutionary structural optimization method. s00158-003-0322-7
Kosaka, I., & Swan, C. C. (1999). A symmetry
10391049. doi:10.1016/j.finel.2007.06.006
reduction method for continuum structural topol-
Huang, X., & Xie, Y. M. (2008). Topology opti- ogy optimization. Computers & Structures, 70(1),
mization of nonlinear structures under displace- 4761. doi:10.1016/S0045-7949(98)00158-8
ment loading. Engineering Structures, 30(7),
Kuhn, H. W., & Tucker, A. W. (1951). Nonlinear
20572068. doi:10.1016/j.engstruct.2008.01.009
programming. In Proceedings of 2nd Berkeley
Huang, X., Xie, Y. M., & Lu, G. (2007). Topol- Symposium (pp. 481492). Berkeley, CA: Uni-
ogy optimization of energy-absorbing structures. versity of California Press.
International Journal of Crashworthiness, 12(6),
Kuwamura, H., & Kato, B. (1989). Effects of
663675. doi:10.1080/13588260701497862
randomness in structural members yield strength
Huang, X., Zuo, Z. H., & Xie, Y. M. (2010). on the structural systems ductility. Journal of
Evolutionary topological optimization of vibrat- Constructional Steel Research, 13(23), 7993.
ing continuum structures for natural frequencies. doi:10.1016/0143-974X(89)90007-2
Lund, E. (1994). Finite element based design
doi:10.1016/j.compstruc.2009.11.011
sensitivity analysis and optimization. Unpublished
Jensen, J. S. (2007). Topology optimization of doctoral dissertation, Institute of Mechanical En-
dynamics problems with Pad approximants. gineering, Aalborg University, Denmark.
International Journal for Numerical Methods in
Ma, Z. D., Cheng, H. C., & Kikuchi, N. (1995).
Engineering, 72(13), 16051630. doi:10.1002/
Topological design for vibrating structures.
nme.2065
gineering, 121(14), 259280. doi:10.1016/0045-
7825(94)00714-X
265
Ma, Z. D., Kikuchi, N., & Hagiwara, I. (1993). Papadrakakis, M., Lagaros, N. D., & Plevris,
Structural topology and shape optimization V. (2005). Design optimization of steel struc-
for a frequency response problem. Computa- tures considering uncertainties. Engineering
tional Mechanics, 13(3), 157174. doi:10.1007/ Structures, 27(9), 14081418. doi:10.1016/j.
BF00370133 engstruct.2005.04.002
Maeda, Y., Nishiwaki, S., Izui, K., Yoshimura, Pedersen, N. L. (2000). Maximization of eigen-
M., Matsui, K., & Terada, K. (2006). Structural values using topology optimization. Structural
topology optimization of vibrating structures with and Multidisciplinary Optimization, 20(1), 211.
specified eigenfrequencies and eigenmode shapes. doi:10.1007/s001580050130
International Journal for Numerical Methods
Prager, W. (1969). Optimality criteria derived
in Engineering, 67(5), 597628. doi:10.1002/
from classical extermum principles. Technical
nme.1626
report, SM Studies Series. Ontario, Canada: Solid
Michell, A. G. M. (1904). The limits of economy Mechanics Division, University of Waterloo.
of material in frame structures. Philosophical
Prager, W. (1974). A note on discretized Mi-
Magazine, 6(8), 589597.
chell structures. Computer Methods in Applied
Min, S., Kikuchi, N., Park, Y. C., Kim, S., & Chang, Mechanics and Engineering, 3(3), 349355.
S. (1999). Optimal topology design of structures doi:10.1016/0045-7825(74)90019-X
under dynamic loads. Structural Optimization,
Querin, O. M. (1997). Evolutionary structural
17(23), 208218.
optimisation stress based formulation and imple-
Oh, S. H., Kimb, Y. J., & Ryu, H. S. (2009). Seismic mentation. Unpublished Doctoral dissertation,
performance of steel structures with slit damp- Department of Aeronautical Engineering, Uni-
ers. Engineering Structures, 31(9), 19972008. versity of Sydney, Sydney, Australia.
doi:10.1016/j.engstruct.2009.03.003
Querin, O. M., Steven, G. P., & Xie, Y. M.
Olhoff, N., & Du, J. (2005). Topological design (1998). Evolutionary structural optimisation
of structures subjected to forced vibration. In J. (ESO) using a bidirectional algorithm. En-
Herskovits (Ed.), Proceedings of the 6th World gineering Computations, 15(8), 10311048.
Congress of Structural and Multidisciplinary Opti- doi:10.1108/02644409810244129
mization, Rio de Janeiro, Brazil, 30 May03 June.
Rozvany, G. I. N. (1972a). Grillages of maximum
Olhoff, N., & Rasmussen, S. H. (1977). On single strength and maximum stiffness. International
and bimodal optimum buckling loads of clamped Journal of Mechanical Sciences, 14(10), 651666.
columns. International Journal of Solids and doi:10.1016/0020-7403(72)90023-9
Structures, 13(7), 605614. doi:10.1016/0020-
Rozvany, G. I. N. (1972b). Optimal load transmis-
7683(77)90043-9
sion by flexure. Computer Methods in Applied
Overton, M. L. (1992). Large-scale optimization Mechanics and Engineering, 1(3), 253263.
of eigenvalues. SIAM Journal on Optimization, doi:10.1016/0045-7825(72)90007-2
2(1), 88120. doi:10.1137/0802007
Rozvany, G. I. N., Zhou, M., & Birker, T. (1992).
Generalized shape optimization without ho-
mogenization. Structural Optimization, 4(34),
250252. doi:10.1007/BF01742754
266
Seyranian, A. P., Lund, E., & Olhoff, N. (1994). Yoon, G. H. (2010a). Maximizing the fundamen-
Multiple eigenvalues in structural optimization tal eigenfrequency of geometrically nonlinear
problems. Structural and Multidisciplinary Op- structures by topology optimization based on ele-
timization, 8(4), 207227. ment connectivity parameterization. Computers
& Structures, 88(12), 120133. doi:10.1016/j.
Sigmund, O., & Petersson, J. (1998). Numerical
compstruc.2009.07.006
instabilities in topology optimization: A survey
on procedures dealing with checkerboards, mesh Yoon, G. H. (2010b). Structural topology opti-
dependencies and local minima. Structural and mization for frequency response problem using
Multidisciplinary Optimization, 16(1), 6875. model reduction schemes. Computer Methods in
Applied Mechanics and Engineering, 199(2528),
Soong, T. T., & Spencer, B. F. Jr. (2002).
17441763. doi:10.1016/j.cma.2010.02.002
Supplemental energy dissipation: State-of-
the-art and state-of-the-practice. Engineering Yoon, G. H. (2011). Topology optimization for
Structures, 24(3), 243259. doi:10.1016/S0141- nonlinear dynamic problem with multiple materi-
0296(01)00092-X als and material-dependent boundary condition.
Svanberg, K. (1987). The method of moving
753763. doi:10.1016/j.finel.2011.02.006
asymptotesA new method for structural op-
timization. International Journal for Numeri- Yoon, G. H., & Kim, Y. Y. (2005). Element connec-
cal Methods in Engineering, 24(2), 359373. tivity parameterization for topology optimization
doi:10.1002/nme.1620240207 of geometrically nonlinear structures. Interna-
tional Journal of Solids and Structures, 42(7),
Xie, Y. M., & Steven, G. P. (1993). A simple
19832009. doi:10.1016/j.ijsolstr.2004.09.005
evolutionary procedure for structural optimiza-
tion. Computers & Structures, 49(5), 885896. Zhao, C. B., Steven, G. P., & Xie, Y. M. (1997).
doi:10.1016/0045-7949(93)90035-C Evolutionary optimization of maximizing the
difference between two natural frequencies of
Xie, Y. M., & Steven, G. P. (1996). Evolutionary
a vibrating structure. Structural Optimization,
structural optimization for dynamic problems.
13(23), 148154. doi:10.1007/BF01199234
doi:10.1016/0045-7949(95)00235-9 Zhou, M., & Rozvany, G. I. N. (2001). On the valid-
ity of ESO type methods in topology optimization.
Yang, X. Y., Xie, Y. M., Steven, G. P., & Querin,
Structural and Multidisciplinary Optimization,
O. M. (1999a). Bidirectional evolutionary method
21(1), 8083. doi:10.1007/s001580050170
for stiffness optimization. American Institute of
Aeronautics and Astronautics Journal, 37(11), Zienkiewicz, O. C., Taylor, R. L., & Zhu, J. Z.
14831488. (2005). The finite element method: Its basis and
fundamentals. Oxford, UK: Elsevier Butterworth-
Yang, X. Y., Xie, Y. M., Steven, G. P., & Querin, O.
Heinemann.
M. (1999b). Topology optimization for frequencies
using an evolutionary method. Journal of Structur-
al Engineering, 125(12), 14321438. doi:10.1061/
(ASCE)0733-9445(1999)125:12(1432)
267
KEY TERMS AND DEFINITIONS Shape Optimization: Finding the best shape
of the elements of a structural system in which
Compliance Minimization: In structural the overall layout, topology and connectivity of
optimization under static loads, minimizing the the elements is previously determined is known
mean compliance is equivalent to maximizing as shape optimization.
the structural stiffness. Under dynamic loads, the Sizing Optimization: Finding the optimum
average of mean compliance over a cycle can be dimensions of the elements in a structural system
considered as an objective function. when the overall layout, connectivity and shape
Eigenfrequency Control: Topology optimiza- of the elements are fixed is termed as sizing
tion techniques can be used to increase or decrease optimization.
any of natural frequencies of a structural system. Topology Optimization: The problem of
Passive Dampers: These devices are used finding the best topology and layout of the ele-
to enhance the energy dissipation capability of ments of a structural system is known as topology
structural systems to mitigate seismic hazard. optimization.
Sensitivity Analysis: In solving an optimiza-
tion problem, finding the derivatives of the objec-
tive function with respect to the design variables
is sometimes referred to as sensitivity analysis.
268
269
Chapter 11
Overall Conceptual Seismic
Design and Local Seismic
Capacity Design for
Components of Bridges
Wan-Cheng Yuan
Tongji University, China
Yu-Guo Zheng
Tongji University and Hunan University of Science and Technology, China
Pak-Chiu Cheung
Tongji University, China
ABSTRACT
From the perspective of overall conceptual seismic design, four design strategies are presented to
decrease and balance the seismic force and displacement demands for some bridges working in a lin-
ear and elastic state: the adjustment of the layout and detail of piers and expansion joints for a typical
long span continuous girder bridge, the adoption of a new-type spatial bridge tower for a long span
cable-stayed bridge, the study on the isolation mechanism of an elastic cable seismic isolation device
for another cable-stayed bridge, and the study on the seismic potential and performance for long span
SCC (steel-concrete composite) bridges. From the perspective of local seismic capacity design, three
earthquake resistant strategies are presented to achieve economical, applicable, and valid seismic design
of local components of bridges working in a nonlinear state: the adoption and the study on a new cable
sliding friction aseismic bearing, the study on the seismic capacities of single-column bridge piers wholly
and locally reinforced with steel fiber reinforced concrete (SFRC), the study on the seismic capacities,
and the hysteretic performance and energy dissipation capabilities of bridge pile group foundations
strengthened with the steel protective pipes (SPPs). Research results show that these seismic design
strategies are effective to improve the seismic performance of bridges.
DOI: 10.4018/978-1-4666-1640-0.ch011
Overall Conceptual Seismic Design and Local Seismic Capacity Design
INTRODUCTION 1. Taking a long span continuous girder bridge

as an example, an optimal design for the
To give clear and correct directions for seismic layout and detail of the bridge components,
design optimization of bridges, the overall con- including geometry of piers, arrangement of
ceptual seismic design and the local seismic piers, location of expansion joint or braking
capacity design methods are proposed to obtain pier, is carried out so as to reduce the seismic
uniform and rational seismic demands and im- demands in the transverse direction as much
proved seismic capacities of structural components as possible. The seismic performance of the
in seismic design of bridges. bridges with different adjustments in differ-
There are many methods to realize seismic ent site conditions is calculated respectively.
design optimization so as to improve the seismic Based on the comparison of the results, the
performance of structures (Gong, 2003; Li, 1997; forms of some piers and the locations of
Liu, 2003; Vagelis, 2009; Zou, 2002). However, expansion joints are required to be adjusted
the structural seismic design optimization will be to improve the bending stiffness distribution
in wrong direction and invalid unless the initial for the bridge.
structural seismic design itself is appropriate and 2. Taking a long span cable-stayed bridge as an
rational. It is hoped that the strategies and methods instance, the original design proposal makes
proposed in the chapter are able to give clear and use of the inverted Y shape bridge tower,
correct directions for seismic design optimization while the seismic design dominated by the
of bridges. first longitudinal vibration mode may lead
There are two paths to improve the seismic to overlarge relative displacement between
design of bridges so as to make the seismic the girder and the tower. In order to solve the
demands more uniform and rational along struc- critical problem, a new-type spatial bridge
tural components. One is to reduce the seismic tower is proposed by integrated analyses
demands as much as possible, and the other is to of the structural dynamic characteristics,
increase the seismic capacities. Correspondingly design displacement and seismic responses.
there are two design methods, the overall linear Compared with the original tower, the new
seismic conceptual design and the local nonlinear spatial tower improves the seismic perfor-
seismic capacity design. From overall to local mance of the bridge significantly.
design strategies, some new ideas and strategies 3. With respect to the elastic cable seismic
of seismic design which can effectively improve isolation device installed between the girder
seismic performance of bridges are proposed and the lower horizontal beam of the tower
separately in this chapter. to mitigate excessive seismic effects, the
Overall conceptual seismic design is a con- influences of the elastic cable stiffness on
ceptual design method applicable for the whole the dynamic characteristics and the seismic
structure based on the linear seismic analysis. demands are investigated by parametric fi-
When the components of bridges work in an elas- nite element analyses of a real cable-stayed
tic state during the earthquakes, the conceptual bridge. The seismic isolation mechanism of
design will be an effective and efficient strategy. the elastic cables is discussed.
From the perspective of overall conceptual seismic 4. Girders of the SCC bridge are composed
design, four conceptual seismic design strategies of the structural steel and the concrete.
are proposed focusing on the whole structure. Compared with a conventional concrete
girder bridge with the same span, the su-
perstructure height and weight are gener-
270
ally less. Therefore the substructure can be determined for the new bearing by theoretical
made of hollow or more flexible piers under analyses, experimental testing and numeri-
earthquakes so that the total cost can be re- cal simulations. A constructed girder-arch
duced greatly. Response spectrum analyses composite bridge installed with the new
are carried out for two design proposals of bearing is studied and evaluated.
a typical four-span continuous bridge, the 2. It is well known that the potential plastic
reinforced concrete bridge with solid col- hinge regions are vulnerable and crucial in
umn piers and the SCC bridge with hollow bridge piers under earthquakes, while a struc-
column piers. Seismic responses of the two ture made of steel fiber reinforced concrete
design proposals are compared with each (SFRC) has good ductility performance and
other to discuss the seismic potential of the energy dissipation capabilities. Therefore a
SCC bridge. The seismic performance of a new idea is put forward in that the SFRC is
long span SCC arch bridge is also studied locally applied in the potential plastic hinge
by response spectrum analyses. region of a bridge pier in an earthquake re-
gion both for saving cost and increasing the
Because of the stochastic characteristics of seismic capacities. Then a nonlinear material
earthquakes, keeping bridges work in an elastic constitutive model is established for SFRC
state permanently is impossible and unreason- and applied to analyze the seismic capacities
able. When the components of a bridge locally of single-column bridge piers reinforced with
become plastic, the overall conceptual seismic SFRC. Compared with the single-column
design method is not applicable because of the bridge pier wholly reinforced with SFRC,
limitations of overall linear analyses. Compared the plastic hinge length and the reasonable
with the overall design, the local seismic capac- range locally reinforced with SFRC are
ity design is a method of great benefit for its determined for the pier.
economy, applicability and validity. It is a detail 3. Steel protective pipe (SPP) is a widely
design strategy for local regions of components adopted measure to make the construction
based on the nonlinear seismic analysis. From of pile group foundations convenient. But
this perspective, three design strategies according the influences of the SPPs on the seismic
to the local method are presented in the chapter. performance of the foundations need
further studies. The seismic capacities of
1. Excessive displacements and forces often a pile group foundation strengthened by
cause damage or collapse of bridges under the SPPs are analyzed and compared with
earthquakes. Taking a currently used fric- the normal pile group foundation with RC
tional isolation bearing as an example, the piles by the pushover method. By using the
seismic forces can be reduced under severe modified Park-Ang damage index (Park &
earthquakes but the seismic displacements Ang, 1985; Park, Ang, & Wen, 1985) and
induced by lateral forces are generally un- the fiber beam-column finite element model,
constrained. Therefore, a new aseismic bear- the hysteretic performance and energy dis-
ing system named as cable sliding friction sipation capabilities are studied based on
aseismic bearing is developed for bridges. numerical simulations and low cycle loading
It combines the advantages of pot bearings experiments for the foundation.
with the excellent displacement restraining
capabilities of cables. The performance
criteria and several design parameters are
271
BACKGROUND Generally the ductility approach and the seis-

mic mitigation and isolation approach are two
For the bridge engineers, the conceptual seismic common strategies in the conceptual design of
design is the first but most important stage in a a bridge. The ductility method requires detailed
bridge design that involves numerous complex reinforced concrete design for shear resistances
and time-consuming tasks. Any optimization in and confinement design at the potential plastic
the conceptual design phase will lead to greater hinges of piers in order to resist earthquake ac-
cost savings than those at the detail design stage tions (Fan & Zhuo, 2001). The seismic mitigation
when decisions become more restricted. Though and isolation approach protects the substructure
in strict terms, the structural seismic design opti- through additional means of energy dissipation
mization follows rigorous formulations and needs by shifting the structural fundamental frequency
definite algorithms, the choices and adjustments away from the dominant frequencies of the ground
of structural systems from the perspective of motions. However, the above two methods are
overall conceptual seismic design will inno- not satisfactory for a long span floating cable-
vationally give meaningful enlightenments and stayed bridge in view of the significant excessive
correct directions for seismic design optimization displacement demands at the tower top and the
of structures. Instead of deferring seismic analyses girder ends under earthquakes. Hence a new con-
to the detail design stage, it is emphasized that ceptual design strategy is formulated to improve
adequate seismic planning should be implemented the seismic performance of a long span floating
early in the conceptual design. cable-stayed bridge through rational configuration
For continuous girder bridges, past conceptual of bridge tower forms. Accordingly a new-type
design used to focus on the longitudinal seismic spatial tower is proposed to replace the generally
behavior, perhaps because most previous damage used inverted Y shape tower for an optimal bal-
reports referred to longitudinal support unseating ance between the seismic force and displacement
or structural collapse (Yang, 1999). However, demands of the bridge.
some transverse damages have been caused in Cable-stayed bridges are classified into differ-
Wenchuan Earthquake because of the lack of ent categories by the connection types between the
transverse earthquake resistances (Yuan & Sun, tower and the girder, such as the floating system
2008). Piers vertical torsion under earthquakes (no restraint between the girder and towers), the
also increases the transverse deformation of the semi-floating system (partial restraint between the
girders. Although a bidirectional ductility design girder and towers), the fixed system (full restraint
will be effective to increase piers torsional stiff- between the girder and towers), and so on. They
ness and transverse ductility capacity so as to are widely constructed in seismic zones for their
improve the transverse seismic performance, but good seismic performance (Walther, Houriet,
it may decrease the shear resistance significantly Isler, Moa, & Klein, 1999). However, the floating
(Fan, 1997). Therefore it is necessary to study the connection may cause excessive seismic displace-
transverse seismic performance of a continuous ment demands at the girder ends while the fixed
girder bridge to select the best structural form connection may lead to excessive force demands
under different site situations. Accordingly the at the tower bottom (Ye, Hu & Fan, 2004). So the
structures dynamic characteristics should adapt to seismic isolation connection types between the
the site conditions and the demands should be uni- tower and the girder, especially the elastic cable
form and rational along the structural components connection, are adopted by more cable-stayed
so as to decrease the overall seismic demands. bridges. It is important to study the seismic iso-
lation mechanism of the elastic cables, since the
272
seismic performance is directly correlated with studied and compared. Also current seismic de-
the cable stiffness for the floating cable-stayed sign guidelines of various countries give different
bridge installed with them. proposals for estimating the plastic hinge length
The SCC bridge combines the mechanical of a single-column pier reinforced with SFRC
advantages of steel and concrete, making the (Zhu, Fu, Wang, & Yuan, 2010). There is a need
bridge more reasonable and durable. Since 1950s, therefore to clarify these differences to facilitate
much research has been carried out to promote the application of SFRC to bridge piers.
the development of SCC bridges. Many countries Pile group foundations are widely used in long
such as USA and Japan developed the design span cable-stayed bridges and suspension bridges.
guidelines and specifications (Brozzetti, 2000; Many post-earthquake damage investigations
Hayward, 1988). Recently, a growing concern and theoretical studies show that the pile group
is mainly about the seismic performance of this foundations are usually vulnerable components
kind of bridges during earthquakes. So a typical under earthquakes (Zhou, 2008). Generally, they
SCC continuous girder bridge and an SCC arch are considered to be strengthened with the steel
bridge are mainly studied to discuss the seismic protective pipes (SPPs) for the construction con-
potential and performance of SCC bridges. venience. But the seismic performance is still not
Since the seismic mitigation and isolation de- fully determined. Based on numerical simulations
sign has received growing attention and become and scale-model experiments, seismic capacities,
a major research topic, various isolation devices hysteretic behavior and energy dissipation capa-
including rubber bearings, frictional bearings and bilities are reviewed in this chapter.
roller bearings have been developed for bridges
(Kunde & Jangid, 2003). Pot bearings are widely
used in long span railway and highway bridges for OVERALL CONCEPTUAL SEISMIC
their higher vertical load bearing capacities and DESIGN FOR BRIDGES
better deformation ability. It can be assigned to be
fixed or frictional in horizontal directions depend- In order to decrease the seismic demands in the
ing on the design demands of the bridges. How- conceptual design phase, the main concern of the
ever, the pot bearing cannot restrain the excessive engineers is to make an optimal arrangement of
seismic displacements when sliding, and cannot structural components to let the structures dy-
reduce the seismic forces when fixed. Considering namic characteristics adapt to the site conditions
the restraint measures of the pot bearing, a new and the seismic demands become more uniform
seismic isolation bearing, known as cable sliding and rational along structural components. Over-
friction aseismic bearing is introduced. This type all conceptual seismic design is applicable for
of bearings can be regarded as the combination of the global structure design based on the linear
pot bearings and restraining cables and has both seismic analysis. In the following sections, four
advantages of them. design strategies are reviewed from the overall
Considering the seismic vulnerability of conceptual seismic design viewpoint: (i) optimal
potential plastic hinge regions of bridge piers, design for layout and detail of continuous girder
more and more bridge piers are reinforced with bridge; (ii) design of towers for long span floating
SFRC in earthquake regions for its better ductil- cable-stayed bridge; (iii) seismic isolation mecha-
ity performance and energy dissipation capaci- nism of elastic cables in cable-stayed bridge; (iv)
ties (Gao, 2009). For the simplest single-column seismic potential and performance for long span
bridge pier, the seismic performance of the piers SCC bridges.
globally and locally reinforced with SFRC are
273
Optimal Design for Layout and portance coefficient and the synthetic influence
Detail of Continuous Girder Bridge coefficient are both assumed to be 1.
For an example continuous girder bridge with
The continuous girder bridge is one of the most 25 spans, each span being 29 m long, the finite
commonly used bridge structural forms. The element model is shown in Figure 2. The structure
substructure has major effects on the bridges is separated into 5 parts by 4 expansion joints.
dynamic properties and the seismic demands There are three types of piers, the solid single-
in different site conditions. Considering both column pier (SSP), the hollow single-column pier
the longitudinal and transverse responses of the (HSP) and the solid twin-column pier (STP), with
bridge, many efforts of the optimal design are put heights ranging from 5.5 m to 48 m. The bridge is
on the substructures, especially on piers. In view analyzed based on response spectrum method by
of the design acceleration response spectrum for a combination of transverse and vertical seismic
normal bridges in Chinese code shown in Figure 1 action. The peak ground acceleration equals 0.41g.
(Ministry of transport of the Peoples Republic of For the original bridge, the response spectrum
China, 1989), the acceleration response spectrum analysis results including the pier top displace-
varies according to the structural fundamental ments, the pier bottom shear forces and normal
period and the type of site. So in designing a stresses are displayed in Table 1 for both soil Type
continuous girder bridge as a complete structure, I (rock and stiff soils) and soil Type III (soft to
the stiffness distribution of some piers can be ad- medium clays and sands). It can be seen that both
justed to decrease seismic forces and improve the the response magnitudes and variations along
seismic performance for different site conditions. these piers are relatively large due to the uneven
In the design acceleration response spectrum, distribution of the piers stiffness. New strategies
the amplifications of ground acceleration are are suggested in the following sections to adjust
respectively 2.25 (0.2 / T ) and 2.25 (0.45 / T )
0.95
the piers stiffness distribution adapt to the site
for soil type I and soil type III according to Chi- conditions and to avoid the uneven seismic force
nese code (Ministry of transport of the Peoples demands caused by the transverse rotation of
Republic of China, 1989), which is only used as girders so as to make the demands more uniform
seismic inputs for the bridge. So the seismic im- and rational along the structural components.
Figure 1. Design response spectrum (Ministry of

Conceptual Seismic Design Strategies
transport of the Peoples Republic of China, 1989)
for Soil Type I
From the code design response spectrum (Figure

1), the specific period T = 0.2 s for soil Type I
g
is relatively small, beyond which the acceleration

amplification decreases rapidly. This means that
any increase in the structural period can decrease
the seismic force demands significantly, if the
structural fundamental period exceed the spe-
cific period. Therefore, reducing the lateral flex-
ural stiffness (i.e. shifting to larger fundamental
period) can efficiently enhance the seismic per-
formance in the transverse direction. Based on
274
Figure 2. Finite element model of the continuous girder bridge
Table 1. Seismic demands of the original continuous girder bridge
Soil type I condition Soil type III condition

Bending
stiffness(103 Disp. Shear stress Normal stress Disp. Shear stress Normal stress
Pier No. Pier type kN/m) (m) (MPa) (MPa) (m) (MPa) (MPa)
1 STP 521.0 0.024 4.630 4.340 0.047 9.290 8.710
2 STP 52.60 0.047 0.960 1.170 0.095 1.890 2.320
3 SSP 7.640 0.071 0.390 7.630 0.140 1.460 15.800
4 HSP 3.170 0.066 0.071 4.740 0.130 0.220 8.960
5 HSP 2.800 0.061 0.079 4.380 0.120 0.220 7.760
6 HSP 2.490 0.055 0.110 4.890 0.110 0.280 7.470
7 HSP 2.800 0.050 0.082 3.840 0.100 0.200 6.180
8 SSP 6.320 0.047 0.260 4.160 0.095 0.660 7.410
9 SSP 11.70 0.043 0.260 4.820 0.086 0.880 9.850
10 SSP 29.20 0.039 0.650 9.160 0.078 2.590 18.290
this consideration, a new-type pier, named as close be replaced by the CTP and the response spectrum
twin-column pier (CTP), is proposed to upgrade analysis results are shown in Table 2.
some piers of the bridge (Figure 3). The CTP is Comparing the results in Table 2 with Table
built by vertically dividing the SSP into two parts 1, it can be seen that all the stress demands and
along the axial symmetrical plane but close to some displacement demands are greatly miti-
each other. The piers transverse stiffness can be gated after some of the bridge piers are changed
reduced by 45% to 50%, but the longitudinal to the CTPs. Although the pier top displacement
flexural stiffness is still maintained. Hence the demands at pier No. 5 to pier No. 10 increase in
corresponding acceleration amplification is re- some degree, the values themselves are still small
duced greatly in the transverse direction. Accord- and do not exceed the requirements of the code
ingly the piers No. 3, 8, 9 and 10 are selected to (Ministry of transport of the Peoples Republic
275
Figure 3. Conceptual design strategy for bridge pier in soil type I condition
Table 2. Seismic demands of the upgraded bridge for soil type I condition
Pier No. Pier type Bending stiffness Disp. (m) Shear stress (MPa) Normal stress (MPa)
(103 kN/m)
1 STP 521.0 0.022 4.340 4.070
2 STP 52.60 0.044 0.910 1.100
3 CTP 4.170 0.066 0.210 2.460
4 HSP 3.170 0.065 0.066 4.850
5 HSP 2.800 0.063 0.065 4.030
6 HSP 2.490 0.060 0.080 4.410
7 HSP 2.800 0.059 0.063 3.760
8 CTP 3.450 0.058 0.270 2.740
9 CTP 61.30 0.056 0.200 2.310
10 CTP 16.00 0.054 0.370 4.110
of China, 1989). Specially, the seismic demand Conceptual Seismic Design

variations along these piers decrease and this Strategies for Soil Type III
conceptual design strategy specifically makes the
seismic demands more uniform and rational along Compared with the soil Type I, the specific pe-
the structural components and enhances the seis- riod T = 0.45 s for soil Type III is relatively large.
g
mic performance in the transverse direction ef- It is obvious that the large deformation becomes
ficiently. the critical problem for long span bridges in this
soil type. So the above proposed upgrading mea-
sures for soil Type I would not be applicable.
276
Figure 4. Conceptual design strategy for bridge

From the comparison between the results in
pier in soil type III condition
Table 3 and those in Table 1, it can be seen that
the displacements, the shear and normal stresses
are all reduced, indicating that the proposed con-
ceptual design strategies can improve the seismic
performance of the bridge.
Design of Towers for Long

Span Cable-Stayed Bridge
From the perspective of overall conceptual seismic

design, both force and displacement demands
should be taken into consideration. However the
force and displacement responses are usually mu-
tually contradictive. As can be seen from the design
response spectra (Figure 5) for a special bridge,
the Sutong Yangtze River Bridge in China (Yan
Therefore for soil Type III, the conceptual design
& Yuan, 2004), in the long period range the ac-
strategies are proposed as follows: (i) relocate the
celeration amplification is relatively small and yet
expansion joint at pier No. 10 to No. 8 in order
the displacement amplification may be too large
to adjust the stiffness distribution of the super-
to be tolerated. So for the flexible cable-stayed
structure and to avoid the uneven seismic force
bridge, a structural period lengthening would be
demands caused by the transverse rotation of
effective to reduce the force demands and yet
girders, (ii) separate the SSPs No. 8, 9 and 10 into
ineffective to decrease the excessive displacement
STPs (Figure 4) to modify the transverse stiffness
demands at the tower top and the deck ends. It is
thus to restrict the seismic displacements for these
obvious that the flexural stiffness of the tower has
piers. The seismic demands are listed in Table 3
a major influence on the fundamental period of
for the upgraded bridge.
the cable-stayed bridge. Therefore it is preferable
Table 3. Seismic demands of the upgraded bridge for soil type III condition
Pier No. Pier type Bending stiffness Disp. (m) Shear stress (MPa) Normal stress (MPa)
(103 kN/m)
1 STP 521.0 0.030 5.840 5.480
2 STP 52.60 0.060 1.200 1.460
3 SSP 4.170 0.089 0.920 9.360
4 HSP 3.170 0.098 0.180 6.790
5 HSP 2.800 0.110 0.210 7.320
6 HSP 2.490 0.120 0.270 8.310
7 HSP 2.800 0.130 0.250 8.680
8 STP 24.60 0.140 2.780 1.770
9 STP 30.50 0.120 2.840 2.310
10 STP 43.20 0.094 3.150 1.900
277
Figure 5. Design response spectra for the Sutong

tions to the towers. The design acceleration and
Yangtze River Bridge (Yan & Yuan, 2004)
displacement response spectra are shown in Fig-
ure 5 and the peak value of the design horizontal
ground acceleration is 0.197g, which is about
twice of the vertical acceleration input.
Conceptual Seismic Design
Flexible structures always have relatively long

fundamental periods. From dynamic analysis,
the fundamental period of the above cable-stayed
bridge is estimated as 13.40 s, a typical long period.
Since the fundamental period is much affected by
the flexural stiffness of the tower, a new-type spa-
tial tower scheme is proposed as shown in Figure
7b after a series of response spectrum analyses
to adopt an optimal tower type for the bridges and comparisons. The bending stiffness of the
fundamental period within a proper range so that proposed alternative tower (65.5103 kN/m) is
acceptable force and displacement demands can be 9.66 times of the original tower (6.78103 kN/m),
achieved in the seismic conceptual design phase. while the weight (1.17106 kN) is just 1.83 times
The Sutong Yangtze River Bridge (Yan & of the original (0.638106 kN). In conclusion the
Yuan, 2004), whose finite element model is shown proposed spatial tower can significantly increase
in Figure 6, is a long span floating cable-stayed the flexural stiffness yet with relatively small
bridge originally with inverted Y shape bridge additional material cost. The fundamental period
towers. The spans are 100 m, 100 m, 300 m, 1088 of the bridge with the proposed spatial towers is
m, 300 m, 100 m, 100 m and the superstructure reduced from 13.40 s to 5.87 s.
is designed as a floating system without connec-
Figure 6. Finite element model of the Sutong Yangtze River Bridge
278
Figure 7. Original and alternative spatial bridge towers
Comparison of Seismic Responses the tower top increases from 0.20 m to 0.41 m, this
0.41 m figure is considered small and practically
By using the design response spectra (Figure 5) acceptable for the long span cable-stayed bridge.
which are from the geological safety evaluation The force demand comparisons are shown in
in the field and fit for the site situation for the Table 5 between the original inverted Y shape and
bridge, the response spectrum analyses are con- the proposed spatial tower models. The total force
ducted respectively in the longitudinal direction demands at the spatial tower bottoms are slightly
and in the transverse direction under correspond- larger than those at the inverted Y tower, while
ing horizontal and vertical ground motions. The the demands in each column No. 1, 2, 3 and 4
displacement demand comparisons are shown in (Figure 7) of the spatial tower model are less than
Table 4 for the original design tower model and the or almost equal to those of the original one. It can
alternative spatial tower model. It can be seen that be concluded that the conceptual design strategy
the maximum longitudinal seismic displacements for the cable-stayed bridge, which adopts the
at the deck end, the tower top and the midspan are spatial tower to replace the original inverted Y
reduced significantly for the spatial tower model. shape tower, enables the seismic demands more
Since the spatial towers constraint to the deck uniform along the structural components and
becomes stronger, the transverse displacement obtain a balance between the force and displace-
at the midspan of the alternative model is also ment demands.
reduced. Although the transverse displacement of
279
Table 4. Seismic displacement comparisons

once the bridge installed with them, which may
Model Location Longitudinal Transverse change the overall seismic force distribution of
displacement displacement the bridge.
(m) (m)
As a case study, consider a long span float-
Original Girder 1.030 -
inverted end
ing cable-stayed bridge with spans 55 m, 165 m,
Y shape 165 m and 55 m and the finite element model is
Tower top 1.110 0.200
tower shown in Figure. 8. There are sliding bearings
model Span 1.040 0.950
center at the subsidiary and side piers and no bearings
Alterna- Girder 0.440 - between the tower and the deck girder from the
tive spa- end original design scheme of the bridge.
tial tower
Tower top 0.390 0.410
model
Span 0.450 0.870 Parametric Analyses
center
Based on dynamic analyses and parametric simula-

tions, in which the elastic cables are simulated as
Seismic Isolation Mechanism of elastic bearing connection elements, the variations
Elastic Cables in Cable-Stayed of the bridge fundamental period corresponding
Bridge to the elastic cable stiffness are shown in Figure 9
and the relationship between the vibration modes
The excessive seismic displacement of the girder and the stiffness is shown in Table 6.
end is one of the challenges to the cable-stayed It can be seen that the fundamental period
bridge designers, since it may exceed the deforma- decreases rapidly when the elastic cable stiffness
tion capacities of expansion joints. There may also increases from 0 to 10105 kN/m. At the stiffness
be excessive seismic bending moment at the tower values larger than 10105 kN/m, the period
bottom. Isolation devices are usually adopted to variation is small. It is also shown in Table 6 that
avoid excessive demands, including the system of the vibration modes do not change when the cable
elastic cables installed between the tower and the stiffness increases from 0 to 5105 kN/m. on the
deck. For a floating cable-stayed bridge without contrary, at stiffness of 30105 kN/m, the first
elastic cables, the seismic force is transmitted mode of vibration changes from a girder longitu-
from the deck to the tower mainly depending on dinal drift to a tower lateral bending in the same
the inclined cables. However, the elastic cables direction. This may be due to the rational elastic
become one of the main force transmission paths cable stiffness leading to a change in the dy-
Table 5. Seismic force comparisons at the tower bottom
Model Column Longitudinal direction Transverse direction

number
Shear force Bending moment Shear force Bending moment
(Figure 7)
(103 kN) (106 kN.m) (103 kN) (106 kN.m)
(i)Original inverted Y 1 (or 2) 30.90 2.290 47.60 1.670
shape tower model
(ii)Spatial tower 1 (or 2) 32.20 2.030 42.60 1.300
model
3 (or 4) 32.20 2.030 32.00 1.220
((ii)-(i))/(i) (%) 1 (or 2) 4.2 -11.4 -10.5 -22.2
3 (or 4) 4.2 -11.4 -32.8 -26.9
280
Figure 8. Finite element model of the cable-stayed bridge with elastic cables
Figure 9. Fundamental period versus elastic cable stiffness
Table 6. Vibration mode versus elastic cable stiffness
Mode Elastic cable stiffness (kN/m)

order
0 5105 30105
1 Girder longitudinal drift Girder longitudinal drift Tower lateral bending in same direction
2 Tower lateral bending in same direction Tower lateral bending in same direction Reversely symmetric vertical bending
3 Reversely symmetric vertical bending Reversely symmetric vertical bending Tower lateral bending in different direc-
tion
4 Tower lateral bending in different direc- Tower lateral bending in different direc- Girder longitudinal drift
tion tion
281
Figure 10. Girder end displacement versus elastic cable stiffness
namic characteristics, thus being able to reduce for the elastic cables, in which the displacement
the seismic demands and improve the seismic demands are reduced significantly while the mo-
performance of the bridge. ment and shear force demands have little chang-
The variations of the girder end displacement es. Thus the rational elastic cable stiffness adjust-
demands are shown in Figure 10 for two differ- ing strategy may achieve the seismic demand
ent site situations (Ministry of Transport of the uniformity along the structural components and
Peoples Republic of China, 2008), based on a balance between the force and displacement
results from elastic response spectrum analyses. demands during the conceptual design.
The relationships are shown in Figure 11 between
the tower-girder relative displacement and the stiff- Seismic Isolation Mechanism
ness. It can be seen that the displacement demands
decrease rapidly as the cable stiffness increases From the parametric analyses, it can be seen that
from 0 to 30105 kN/m and their decreasing rate if the cable stiffness increases the fundamental
becomes insignificant after the stiffness exceeds period of the bridge will decrease rapidly, the
30105 kN/m. vibration modes will change, the displacements
On the other hand, the variations of the tower will decrease rapidly, the bending moments will
bottom seismic force demands are shown in Fig- fluctuate in small amplitude and the shear force
ure 12 for the same two different site conditions. will monotonically increase.
It can be seen that the bending moment demands The above analyses reveal the core of the seis-
fluctuate in relatively small amplitude when the mic isolation mechanism of the elastic cables, the
cable stiffness increases, while the shear force change of structure system and the transformation
demands monotonously increase. of the force transmission path in the bridge tower.
Therefore it is concluded that the elastic cables Because of the limiting function of the elastic
change the seismic force transmission and distri- cables, a floating cable-stayed bridge gradually
bution significantly. The stiffness range from turns into a semi-floating system, and finally a
20105 to 70105 kN/m is practically acceptable tower-girder rigid connection system when the
282
Figure 11. Tower-girder relative displacement versus elastic cable stiffness
Figure 12. Seismic force demands versus elastic cable stiffness
283
Table 7. Seismic response comparisons of RC and SCC continuous girder bridges
Design Superstructure Pier area Pier stiffness Period Shear Moment Disp. (m)
scheme mass (106 kg) (m2) (106 kN/m) (s) (103 kN) (103 kN.m)
(i) RC 3.900 4.050 0.199 0.880 3.430 24.00 0.017
(ii) SCC 2.118 2.200 0.189 0.670 2.470 17.30 0.013
(ii)/(i) (%) 54.3 54.3 95.0 76.1 72.0 72.1 76.5
cable stiffness increases. It is obvious that the with the same span and deck width has a lighter
change of the system, namely the increase of the superstructure, less pier stiffness and fundamental
bridge stiffness will cause a decrease of the fun- period. The seismic shear force and bending mo-
damental period and an increase of the horizontal ment at the pier bottom are 72.0% and 72.1% of
brake force of the girder. The transformation of the those from the initial reinforced concrete scheme.
force transmission path is the main reason why the Moreover, the pier top displacement is only 76.5%
seismic demands change along the curves shown of that in design scheme (i). The seismic potential
in Figure 10 to 12. For a floating cable-stayed of the SCC bridge system is significant.
bridge without elastic cables under earthquakes, A major highway bridge located in the city of
the large horizontal forces produced by the girder, Hangzhou, China, the Jiubao Bridge consists of
whose acting point is usually at upper zones of an SCC continuous girder system for the northern
the tower, are all transmitted to the tower by the approach (55 m, 85 m, 85 m, 90 m), an arch with
stayed-cables. But the elastic cables can share SCC girder system for the main bridge (three 210
most of the seismic forces and transmit them to m spans) and another SCC continuous girder
lower zones of the tower for a cable-stayed bridge system for the southern approach (90 m, nine 85
installed with them. m, 55 m). The SCC deck is a reinforced concrete
slab on top of steel box girders. The superstructure
Seismic Potential and Performance is supported by sliding and pot bearings, both
for Long Span SCC Bridges types being fixed in the transverse direction. In
the longitudinal direction the bearings are designed
To illustrate the advantage of steel-concrete com- to be fixed at pier PN4, PS1, PS5, PS6 and PS7
posite (SCC) system, a typical four span bridge as shown in the finite element model in Figure
with spans of 25 m, 30 m, 30 m and 25 m, deck 13.
18 m wide, and piers 7 m high was ever studied It was suggested that the girders of the SCC
(Turkington, Carr, Cooke, & Moss, 1989). In bridge might be replaced by equivalent concrete
this section, there are two design schemes for the girders with weight twice as heavy as the SCC
bridge: (i) a conventional reinforced concrete (RC) ones. Response spectrum analyses were performed
continuous girder bridge supported by solid form (Cao, 2009) and the results are shown in Table 8
piers and (ii) an SCC continuous girder bridge sup- for the bridge, and it is clear that the long span
ported by hollow form piers, all basic properties SCC bridge has lower seismic force and displace-
being shown in Table 7. The seismic responses ment demands than the RC one. Hence, the SCC
of both design schemes from elastic response system has higher seismic potential and better
spectrum analyses (Ministry of Transport of the seismic performance.
Peoples Republic of China, 2008) are also shown
in Table 7. It can be seen that the SCC scheme
284
Figure 13. Finite element model of the Jiubao Bridge
Table 8. Seismic response comparisons of the arch bridges with RC and SCC girders
Girder Southern approach Main bridge

Shear-PS6 Moment-PS6 Displacement of Shear-PS1 Moment-PS1 Displacement of
(103 kN) (103 kN.m) bearing-PS2 (m) (103 kN) (103 kN.m) bearing-PN1(m)
(i) RC 8.450 156.10 0.290 10.492 138.85 0.240
(ii) SCC 5.195 95.53 0.190 7.894 102.89 0.190
(ii)/(i) (%) 61.5 61.2 65.5 75.2 74.1 79.2
Solutions and Recommendations For a long span floating cable-stayed bridge,

a new-type spatial bridge tower is proposed to
From the perspective of overall linear conceptual replace the originally designed inverted Y shape
seismic design, several new design strategies have tower. Analysis results show that the longitudinal
been presented. For a typical long span continuous seismic displacements can be reduced significantly
girder bridge, the forms of piers and the locations for a bridge with the spatial tower. The transverse
of expansion joints are adjusted respectively for displacement at the midspan of the spatial tower
different site conditions to improve the stiffness model is also reduced. Although the transverse
distribution. Based on code design response displacement of the tower top increases to some
spectra, theoretical analyses and numerical simula- extent, the magnitude of the displacement is
tions show that both the displacement and stress relatively small and practically acceptable for
demands are greatly mitigated, and the demand the cable-stayed bridge. Moreover, the seismic
variations along the piers are decreased as well force demands in each column of the spatial tower
after adopting the strategies. model are less than or almost equal to those of
the original model.
285
With respect to the elastic cable seismic When the components of bridge locally become
isolation device, the isolation mechanism, the plastic, the overall conceptual seismic design is
influences of the elastic cable stiffness on the not applicable because of the limitation of overall
bridge structural dynamic characteristics, the linear analyses. The local nonlinear seismic capac-
force transmission paths and the displacement and ity design is therefore beneficial for its economy,
force demands have been discussed for another applicability and validity.
floating cable-stayed bridge. Numerical analysis
results show that the fundamental period of the Innovation of Cable Sliding Friction
bridge decreases rapidly when the cable stiffness Aseismic Bearing for Bridges
increases. The vibration modes will change if the
cable stiffness increases to some extent, and the The girders of continuous bridges are generally
displacement demands decrease rapidly when the supported by bearings which transmit loads to
stiffness increases. The bending moment demands underlying piers and abutments. Bearing system
fluctuate in small amplitude, while the seismic generally comprises a fixed bearing on one pier and
shear force demands monotonically increase. several frictional bearings on other piers. Under
For the new-type SCC bridge, the seismic minor or moderate horizontal earthquake actions,
potential is further confirmed based on response almost all the superstructures earthquake-induced
spectrum analyses of a four span continuous forces are transmitted to the fixed pier which may
girder bridge and a long span SCC arch bridge. thus be damaged because of the excessive shear
Analysis results show that the SCC continuous force or bending moment. If the fixed bearing is
girder bridge with same span and deck width has designed to fail under severe earthquakes so as
lighter superstructure, less pier stiffness and fun- to protect the pier system, all the other frictional
damental period. Also the shear force and bending bearings will be mobilized to slide so that the bridge
moment at the pier bottom and the displacement system will be difficult to maintain equilibrium.
at pier top all decrease significantly. For the long A new cable sliding friction aseismic bearing
span SCC arch bridge, it also has lower seismic (Figure 14) is invented and presented, which is
force and displacement demands than a reinforced composed of the pot bearing, some cables and a
concrete bridge. shear bolt. Through reasonable design, the bearing
may perform as same as a fixed bearing system,
namely keep in service under minor or moderate
LOCAL SEISMIC CAPACITY DESIGN earthquakes. Moreover, the shear bolt of the new
FOR COMPONENTS OF BRIDGES bearing will fail during severe earthquakes, and
then a bearing originally used as a fixed bearing
Earthquakes have a habit of identifying structural will be converted into a frictional one. Hence, the
weaknesses and concentrating damage at these seismic force can be shared by all the piers not
locations (Priestley, Seible & Calvi, 1992). Lots only by the fixed pier, while the excessive seismic
of investigations on earthquake disaster show displacement can be restricted by the cables at-
that bridge damage caused by earthquakes mainly tached to the base of the bearing.
focuses on local regions and weak points of bridge As an improved type of frictional isolation
components, such as bearing failures, pier failures, devices, the new bearing is characterized by its
pile foundation failures, expansion joint failures, insensitivity to the frequency content of earthquake
and so on. Considering the stochastic character- excitations. It can be used in conventional girder-
istics of earthquakes, keeping bridges in elastic system bridges, cable-stayed bridges, suspension
state permanently is impossible and unreasonable. bridges and other similar structures. Compared
286
Figure 14. Schematic diagram of cable sliding friction aseismic bearing
to other isolation bearings, this bearing has the the superstructure and the pier should be
advantages of simplicity in principle, wide ap- restrained by the cables.
plication, mature technique, stable performance
and low cost (Yuan, Cao, Cheung, Wang & Rong, Design Parameters
2010).
Depending on the above performance objects,
Performance Criteria some important design parameters, such as the
characteristics of the cables, the shear strength of
Based on the above design strategies, three perfor- the bolt and the integral stiffness of the bearing
mance objectives are proposed for a cable sliding should be studied carefully as follows, except for
friction aseismic bearing as follows. the available characteristics of the pot bearing
from the manufacturers.
1. If the expected deformation of the bearing
is less than the design free displacement, 1. Characteristics of the cables
the cables should not be mobilized and the
new bearing functions as a conventional pot According to the geometry dimensions of the
bearing. bearing, each cables length L can be calculated
2. Under minor and moderate earthquakes, the as Eq. (1), where L is the projective distance of
xy
shear bolt in a fixed-type bearing should the cable in horizontal plane, H is the bearing
not break so that the bearing keeps service total height, A and B are length and width of the
and need not be replaced. With respect to upper plate, C and D are length and width of the
a sliding-type bearing, the friction capaci- lower plate, x and are the design displacement
y
ties between the stainless steel plate and in the bearings longitudinal and transverse direc-
the Teflon plate should bear the horizontal tion.
ground motions and dissipate the seismic
energy.

Lxy = [x + (A C ) / 2]2 + [y + (B D ) / 2]2
3. Under severe earthquakes with the breaking

of the shear bolt, the bearing should function
L = H 2
+ L 2

xy
as a frictional bearing to mitigate the seismic (1)

forces and dissipate the seismic energy, while
excessive relative displacement between
287
Figure 15. Stiffness models
The axial stiffness of a cable member is defined cables, u denotes the design free displacement
0
as K = EA1 L , where E is the elastic modulus of

2 when the bearing is in normal service load and
the cables, A1 is the cable section area and L is is the angle between the longitudinal direction
the cable length. of the cable when engages at a lateral displace-
ment of u and the direction of the horizontal
0
2. Shear strength of the bolt relative displacement of the friction bearing. So

combined with the characteristics of the fric-
Considering the requirements under service tional bearing and the cables, the integral stiffness
loads, the horizontal shear resisting capacity of model of the cable sliding friction aseismic bear-
the fixed-type pot bearing should not be less than ing can be shown in Figure 15c, where n is the
10% of the vertical load bearing capacity (Cao, number of cables and is a coefficient that
2009). Depending on the performance objects, represents the equivalent linearization of cable
the shear bolt should be out of service during stiffness under severe earthquakes.
severe earthquakes. So it is suggested that the Depending on pseudo-static cyclic loading
shear strength of the bolt be within 10% to 15% experiments and finite element analyses for a
of its vertical load capacity. prototype cable sliding friction aseismic bearing
(Figure 16) where the maximum vertical load is
3. Integral stiffness of the bearing 5000 kN, the actual values of the bearings char-
acteristics, such as the static and dynamic friction
The lateral stiffness of a frictional bearing can coefficients of the bearings sliding surface, the
be simplified as an ideal bilinear model as shown cable stiffness, the experimental hysteretic curves
in Figure 15a, where K denotes the elastic stiff-
1
of the bearing and the corresponding skeleton
ness of the bearing and F denotes the critical
s
curves, can be obtained (Cao, 2009; Yuan, Cao,
friction force which is calculated as F = N where Cheung, Wang & Rong, 2010). The experimental
s
is the friction coefficient and N is the vertical results show that the static and dynamic friction
force on the bearing. The cables can be regarded coefficients decrease as the vertical load increas-
as completely elastic material since they should es. Through the experiments and calculations, the
not yield, whose load-displacement curve is shown coefficient is suggested to be 0.373 for its ap-
in Figure 15b, where K is the stiffness of the plication in theoretical analysis models of the new
2
288
Figure 16. Prototype bearing

the fixed pier is PS1. For the new cable sliding
friction aseismic bearing used only on the fixed
pier PS1 in the bridge, the vertical load bearing
capacity is 35000 kN, the friction coefficient is
0.02 and the shear strength of the shear bolt is 3500
kN. The seismic responses are evaluated by using
an artificial synthetic ground motion (Figure 19),
whose excess probability is 3% in 100 years (Cao,
2009). The artificial synthetic ground motion is
applied in the longitudinal direction at the base of
the piles and abutments, while the vertical ground
motion is two thirds of it.
Since the stiffness of the cables is very impor-
tant for the new bearing, it is necessary to analyze
bearing. The results also show that the experi- the seismic demand variations of the bridge along
mental hysteretic curves agree with the numerical with the stiffness of the cables. Parametric analy-
ones well and the displacement restraining capac- ses are conducted and the results are shown in
ity of the cables is significant. Figure 20, in which the left vertical coordinate
indicates the seismic shear force at the bottom of
Application Study pier PS1 and the right one indicates the maximum
longitudinal seismic displacement of the deck
From the above theoretical assumptions and end. It can be seen that the seismic displacements
analyses, the new cable sliding friction aseismic of the deck end decrease rapidly, while the seismic
bearing seems sound and acceptable. After some shear forces at the bottom of pier PS1 do not in-
conformation improvements in order to be fit for crease too much, as the cable stiffness increases.
the requirements and convenience of manufac- Yet the shear forces increases a little and the deck
tures, the modified shop-manufactured bearing displacement changes slightly when the cable
shown in Figure 17 is applied in a girder-arch stiffness gradually increases from 1105 to 3105
composite bridge with spans of three 210 m, whose kN/m.
finite element model is shown in Figure 18 and
Figure 17. Modified shop-manufactured bearing
289
Figure 18. Finite element model of the girder-arch composite bridge
Figure 19. Synthetic seismic wave
Figure20. Seismic demands versus the cable stiffness
290
Table 9. Two different design schemes with dif-

demands are significantly reduced by 59%, when
ferent arrangements of bearings
the new bearing takes the place of the original
Design Pier Pier Pier PS1 Pier one. And other shear forces are also reduced at
scheme PN2 PN1 PS2 the pier bottoms of other piers except for the almost
Original Slid- Sliding Conventional fixed Slid- equal shear forces at the pier bottoms of the pier
design ing bearings bearings ing
bear- bear-
PN2. Furthermore, the possible maximum dis-
ings ings placements of the deck relative to the piers during
Alterna- Slid- Sliding Proposed cable slid- Slid- a severe earthquake that causes the conventional
tive ing bearings ing friction ing fixed bearings at pier PS1 to lose fixity may range
design bear- aseismic bearings bear-
ings ings from 268 mm to 280 mm. Yet the deck displace-
ments at all four piers are controlled to be less
than 160 mm by using the new bearing.
For the nonlinear time history analysis of the Moreover, the seismic capacity checks are
bridge, two design schemes are considered, (i) shown in Table 11 for the most unbeneficial pile
the original design with the conventional fixed of the bridge. It can be seen that the checks can-
bearings on the fixed pier PS1; (ii) the alternative not pass when the original bearing is used on the
design with the cable sliding friction aseismic fixed pier PS1. However, the checks can pass
bearings on the same pier. Besides, all other bear- when the new bearing replaces the original one.
ings are the same sliding ones for the two design So it can be concluded that the new bearing can
schemes and the arrangement of the bearings is decrease the seismic demands and be effective to
listed in Table 9. protect the bridge piers and foundations.
Considering the seismic demand variations of
the bridge along with the stiffness of the cables Seismic Performance of Bridge Piers
and many cables used in the new bearing resulting Reinforced with SFRC
in the larger supplied stiffness, so the cable stiff-
ness in the horizontal direction is taken as about There have been numerous studies on steel fiber
3.0105 kN/m in the application study. After the reinforced concrete (SFRC), most of which aimed
analyses, shear force demands at pier bottoms and at the constitutive behavior (Zollo, 1997). These
deck-pier relative displacements are shown in studies generally ignored an important fact that
Table 10. Note that the shear force is normalized the SFRC is more likely conducted in confined
with the deck weight. For the shear force com- conditions. Ramesh, Seshu and Prabhakar have
parisons, the bearings at PS1 are of conventional made experimental analyses of 90 specimens and
fixed type in the theoretical models of the bridge, pointed out that the SFRC confined by lateral ties
because the bearings in fixed conditions have has improved material properties (Ramesh, Seshu
relative more disadvantageous seismic force & Prabhakar, 2003). Then a constitutive model
demands. For the deck-pier relative displacement of SFRC, which has been widely accepted, is
comparisons, the fixed bearings are assumed to offered as Figure 21. The constitutive equations
have lost fixity and be free to move during a severe are also shown as Eq. (2), (3) and (4), in which
earthquake, focusing the performance of the C = (P P ) ( fv fc ) b s and RI is the rein-
'
i b bb
original and the new bearings. It can be seen that forced index.
the shear force demand at PS1 is much larger than
those at the other three piers by using the original P = fc (1 + 0.55C i ) (1.0228 + 0.1024 (RI )) Ag + fy Ag
'

bearing on the fixed pier PS1. Yet the excessive (2)
291
Table 10. Seismic demand comparisons of the bridge with different bearings
Shear force demands of piers Deck-pier relative displacement

Design scheme Pier location Base Shear (Wd) % of original Displacement (mm) % of original
PN2 0.0473 - 280 -
Original PN1 0.0545 - 273 -

design PS1 0.0914 - 268 -
PS2 0.0524 - 268 -
PN2 0.0491 104% 152 54%
Alternative PN1 0.0474 87% 142 52%

design PS1b 0.0537 59% 128 48%
PS2 0.0473 90% 113 42%
Table 11. Check comparisons of the unbeneficial pile for different bearings
Seismic Demand
Moment capacity
Design scheme Pier location Axis force (103 kN) Moment (103 kNm) (103 kNm) Check
PN2 9.633 18.45 22.15 Yes
Original PN1 10.872 15.39 22.59 Yes

design PS1 1.546 22.94 18.64 No
PS2 9.591 18.48 22.13 Yes
Alternative
PS1 4.081 15.23 19.90 Yes
design
Figure 21. Constitutive model of SFRC
292
Figure 22. Section of single-column pier
u = c (1 + 5.2C i ) (0.9899 + 0.2204 (RI ))

'
(3) Table 12. Typical single-column piers
Pier model Concentrated load at the Height of pier

0.85u u = 1.8847 + 0.121 (RI ) (4) top of pier (103 kN) (m)
M1 30 10
M2 30 15
M3 30 20
Seismic Capacity of Section
M4 30 25
Reinforced with SFRC
To study the seismic capacity of a section reinforce

For these typical single-column pier models,
with SFRC, a series of finite element models are
M1, M2, M3 and M4 reinforced with SFRC and
established for some typical single-column piers
CRC, pushover analyses are conducted for them
reinforced with SFRC and for some alternative
respectively. Then the capacity curve comparisons
piers with plain confined concrete (CRC), whose
are shown in Figure 23 and the curvature ductil-
section is shown in Figure 22 and design param-
ity comparisons are shown in Table 13. It can be
eters are shown in Table 12 for the pier model M1,
seen that the bending strength is improved if the
M2, M3 and M4. The longitudinal reinforcement
SFRC replaces the CRC for all the piers. It is also
ratio is 2.2% and the lateral reinforcement ratio is
shown in Table 13 that the curvature ductility of
0.8% for these piers. For the SFRC, the yielding
sections adopting SFRC is better than those adopt-
stress and strain are 50 MPa and 6.18810-3 and
ing CRC. The comparisons between the results
the ultimate stress and strain are 42.5 MPa and
in Table 13 and the ones in Figure 23 show that
13.8810-3. The yielding stress and strain are 38
the improvement extent of the ductility is much
MPa and 3.73010-3 and the ultimate stress and
higher than that of the bending strength after
strain are 32.3 MPa and 7.03010-3 for the CRC.
adopting the SFRC, which indicates that its a
293
Figure 23. Capacity curve comparisons between piers reinforced with SFRC and CRC
Table 13. Curvature ductility comparisons between piers reinforced with SFRC and CRC
Pier model Yield Ultimate curvature Curvature ductility Ratio of curvature ductility
curvature (10-3) (10-3) factor factor (ii)/(i) (%)
(i) M1(CRC) 0.615 8.097 13.166 201.7
(ii) M1(SFRC) 0.654 17.381 26.576
(i) M2(CRC) 0.915 6.277 6.860 206.4
(ii) M2(SFRC) 0.997 14.112 14.154
(i) M3(CRC) 0.912 4.733 5.190 288.5
(ii) M3(SFRC) 0.999 12.085 12.097
(i) M4(CRC) 0.899 4.401 4.895 263.8
(ii) M4(SFRC) 1.046 10.769 10.295
more effective way to improve the ductility earthquakes from the capacity design principles
rather than to enhance the bending strength. (Priestley, Seible & Calvi, 1992). It means that the
whole bridge pier reinforced with SFRC may be
Bridge Piers Reinforced with SFRC a tremendous waste. Therefore, it is necessary to
study the seismic capacities of bridge piers wholly
From the above analysis results, it is clear that and locally reinforced with SFRC.
the SFRC can efficiently improve the ductility For a group of piers M3 and M4 (Table 12),
capacity of bridge piers. However, only partial the lengths of local region reinforced with SFRC
regions of a bridge pier may enter plasticity under (LR-SFRC) are assumed to be 4 m and 5 m respec-
294
Figure 24. Capacity curve comparisons between piers wholly and locally reinforced with SFRC
Table 14. Curvature ductility comparisons between piers wholly and locally reinforced with SFRC
Pier model Yield Ultimate Curvature Ratio of curvature ductility

curvature (10-3) curvature (10-3) ductility factor factor (ii)/(i) (%)
(i) M3(WR-SFRC) 0.999 12.085 12.097 103.6
(ii) M3(LR-SFRC-4m) 1.047 13.111 12.522
(i) M4(WR-SFRC) 1.046 10.769 10.295 106.2
(ii) M4(LR-SFRC-5m) 1.038 11.344 10.929
tively at the pier bottoms. For another compared Plastic Hinge Length and Reasonable
group of piers M3 and M4 (Table 12), it is assumed Range Locally Reinforced with SFRC
that the piers are wholly reinforced with SFRC
(WR-SFRC). Based on pushover analyses, the For the third group of piers M3 and M4 (Table
capacity curve comparisons are shown in Figure 12), the lengths of the local region reinforced
24 and the curvature ductility comparisons are with SFRC are assumed to be 2.0 m and 2.5 m
shown in Table 14 between the piers wholly and respectively at the pier bottoms. Based on push-
locally reinforced with SFRC. It can be seen that over analyses, the results show that a local CRC
applying the SFRC locally in a reasonable region region above the CRC-SFRC interface has already
at the bottom of the single-column pier almost yielded, while the pier bottom range reinforced
has the same effect in improving the ductility and with SFRC has not yet yielded. It is obvious that
the bending strength as applying the SFRC in the the local region reinforced with SFRC has not
whole pier. So considering the economy, it is more attained the predicted aim to increase the sec-
advantageous to applying SFRC in a local region tion capacities for the third group of pier models
of the bridge pier rather than applying SFRC in because of the unreasonable range reinforced
the whole pier. with SFRC.
So, it is very important to determine the reason-
able range of the local region reinforced with SFRC
for bridge piers, namely the plastic hinge lengths.
While referring to the present seismic guidelines
295
Table 15. Plastic hinge lengths according to dif-

nently existence and be taken as a part of the piles
ferent guidelines
and a strengthening measure for the pile. But the
Pier model New Euro Code 8 AASHTO seismic performance is still not fully determined
Zealand for the pile group foundations with SPPs.
M3 3.750 1.840 1.890
M4 4.000 2.240 2.290 Seismic Capacity
In order to investigate the effect of the SPPs on

of several countries, it is shown in Table 15 that the seismic performance of the foundation, a real
the plastic hinge lengths differ from each other pile group foundation comprising of 9 reinforced
greatly for the pier M3 and M4 reinforced with concrete piles and a cap (Figure 25) is taken as
SFRC (Zhu, Fu, Wang & Yuan, 2010). an example. Four different scale-models are es-
Then for a single-column bridge pier, the tablished to study the seismic capacity of the pile
reasonable length lc of local region reinforced group foundation: (i) Model 1, all piles without
with SFRC is given as Eq. (5), in which f and SPPs, (ii) Model 2, only edge piles strengthened
cy
f are respectively the compressive yield stress- with SPPs in local regions from the pile top to the
second plastic hinge under the ground surface, (iii)
fy
es of the original reinforced concrete and the

Model 3, only the edge piles strengthened with
SFRC.
SPPs along the total pile height, and (iv) Model
lc = (1 fcy ffy )l (5) 4, all the piles strengthened with SPPs along the
total pile height. For latter three models, the area
of SPP accounts for 1.4% of the whole section
Considering the extra-strength of the original area. Other parameters of the models are listed
reinforced concrete, the equation can be modified in Table 16.
as the following Eq. (6), where is the extra- Depending on the results of pushover analyses,
strength coefficient. And the coefficient is the seismic capacity comparisons are shown in
suggested to be 1.2 to 1.4. Figure 26 for Model 1 to 4. The seismic capacity
variations for the edge pile in Model 3 are pre-
lc = (1 fcy ffy )l (6) sented in Figure 27 along with the steel protective
pipe ratio (SPPR) ranging from 0 to 2.5%. More-
over the section curvature variations in yielding
Seismic Performance of state and in ultimate state are presented in Figure
Pile Group Foundations 28 for edge piles in Model 3. It can be seen that
Strengthened with SPPs the seismic capacities, including the ductility and
the lateral resistance of the pile group foundation
The pile group foundations consisting of a group strengthened with SPPs, is better than those of
of piles and a cap supported by the pile group are the foundation without SPP. The lateral resistance
widely used in large bridge projects. The pile group of the edge piles increases, while the maximum
foundations are usually vulnerable components curvature decreases significantly along with the
under earthquakes. Steel protective pipes (SPPs) increasing of the SPPR. Therefore the potential
are always used in construction as formworks for is significant for the pile group foundation
the underwater in situ concrete piles. After the strengthened with SPPs.
bridge is completed, the SPPs can be of perma-
296
Figure 25. Pile group foundation system
Table 16. Design parameters of the pile group foundation models
Model Scale Diameter Height Free Pile Main Hoop Vertical

(cm) (cm) Segment space(cm) reinforcement reinforcement Force at
height(cm) ratio ratio the pier
top (kN)
1-8 1:20 9.0 72.0 49.0 44.5 0.871 0.379 75.0
Low Cycle Loading Experiments amplitude of the cycles was variable from 2 mm
to 80 mm and the frequency of the applied signal
From the above analyses results, it can be con- was equal to 0.02 Hz (Zhou, 2008).
cluded that the pile group foundation strengthened Considering the modified Park-Ang damage
with SPPs has better seismic capacities than the index (Park & Ang, 1985; Park, Ang, & Wen,
foundation without the SPP. So it is necessary to 1985), the fiber beam element finite element
further study the energy dissipation and hyster- model (Figure 30) was established for these mod-
etic performance under earthquakes in order to els and the low cycle loading experiments were
promote the application of the pile group foun- simulated by the finite element method.
dation strengthened with SPPs. Then another After the experiments and finite element
four scale-models, Model 5 to 8, are established, analyses, the numerical and experimental hyster-
whose design parameters are listed in Table 16 etic curves between the lateral force and the pier
too. Model 5 and 6 are normal reinforced concrete top displacement are shown in Figure 31. It can
(RC) pile group foundations and Model 7 and 8 be shown that the strength of the models strength-
are foundations strengthened with SPPs, whose ened with SPPs is about 1.77 times and the ultimate
pipe wall thickness is 0.1cm and length is 40cm displacement is about 1.66 times of the ones of
beginning from the pile top. Moreover, the low the RC models. Additionally, the hysteretic en-
cycle loading experiments (Figure 29) were con- ergy dissipation capabilities of the models
ducted for these models in the electro-hydraulic strengthened with SPPs are more than 2.15 times
servo loading system in Tongji University. The of the RC models, while the equivalent viscous
297
Figure 26. Seismic capacity comparisons
Figure 27. Seismic capacity variations of the edge pile in Model 3
Figure 28. Section curvature variations of the edge pile in Model 3
298
Figure 29. Low cycle loading experiments
Figure 30. Finite element model and fiber element sections
299
Figure 31. Hysteretic curves
damping ratio is reduced by 41.7%. The pinching important design parameters are determined by
effects of RC models are more obvious, while the theoretical analyses and experiment testing. The
hysteretic loops are plumper for models strength- application of the new bearing in a girder-arch
ened with SPPs. Comparisons between the ex- composite bridge shows that the relative seismic
perimental and numerical responses show that the displacements between the girder and piers are
numerical results agree with the experimental reduced significantly and the seismic forces at
trends well. almost all the pier bottoms are also reduced when
the new bearing replaces the original frictional
Solutions and Recommendations one on the fixed pier.
Based on the proposed nonlinear material
To realize more economical, applicable and valid constitutive model of SFRC, the seismic capaci-
seismic design of bridges, some local element ties of single-column bridge piers wholly and
seismic design strategies are presented to improve locally reinforced with SFRC are studied by the
the seismic capacities for bridges. Considering the pushover method. Analysis results show that the
disadvantages of some typical isolation devices, bending strength and the curvature ductility of
a new cable sliding friction aseismic bearing the piers are improved if the SFRC replaces the
is invented. Its performance criteria and some CRC. The improvement extent of the ductility is
300
much higher than the bending strength. Seismic tion. Undoubtedly, there are still many aspects
capacity comparisons between the piers wholly need to be studied to obtain more optimal seismic
and locally strengthened with SFRC show that performance for bridges, while this chapter mainly
applying SFRC locally in a reasonable area of focuses on two methods: (i) the overall conceptual
the single-column pier almost has the similar seismic design for whole bridges in a linear and
effect in improving the ductility and the bending elastic state; (ii) local seismic capacity design for
strength as applying the SFRC in the whole pier. components of bridges in a nonlinear and plastic
Then the plastic hinge length and the reasonable state. The seismic design strategies proposed in the
range locally strengthened with SFRC are clearly chapter consider the displacement requirements
determined, considering that the present seismic of bridges greatly, but they are still not enough to
guidelines of several countries have different attain the complete displacement-based design. So
definitions of the plastic hinge length for the the displacement-based design methods should be
single-column bridge pier reinforced with SFRC. further promoted in the future, since the limitations
For pile group foundations reinforce with SPPs, of the force-based design methods are more and
the seismic capacities, the hysteretic performance more obvious for bridges.
and energy dissipation capabilities are studied The overall conceptual seismic design mainly
by numerical simulations and low cycle loading focuses on typical bridges in this chapter. With
experiments. The pushover analysis results show the increase of irregular bridges constructed in
that the seismic capacity, including the ductility earthquake regions, more attention should be paid
and the lateral resistance of the pile group foun- to their seismic performance, and how to conduct
dation reinforced with SPPs, is better than the the overall conceptual seismic design and local
original one without SPPs. The lateral resistance seismic capacity design of them in the future.
of the edge piles increases, while the maximum Moreover, the stochastic seismic performance of
curvature decreases significantly along with the bridges with the cable sliding friction aseismic
increasing of the SPPR. The experiments and the bearing should be emphasized in future research,
finite element analysis comparisons show that since the determinate seismic performance has
the strength and the ultimate displacement of the been carefully studied in the chapter. For the pile
foundation models strengthened with SPPs are group foundation strengthened with SPPs, the
better those of the RC ones. The hysteretic energy erosion problem of steel pipes and its influences
dissipation capabilities of the models strengthened on the seismic design of foundations should be
with SPPs are much better than those of the RC further promoted.
models, while the equivalent viscous damping
ratio is reduced by 41.7%. The pinching effects of
RC models are more obvious, while the hysteretic CONCLUSION
loops are plumper for models strengthened with
SPPs. Comparisons between the experimental To give clear and correct directions for seismic
and numerical responses show that the numerical design optimization of bridges, some new seismic
results agree with the experimental trends well. design strategies are presented to obtain uniform
and rational seismic demands and improved
seismic capacities of structural components in
FUTURE RESEARCH DIRECTIONS seismic design of bridges, according to two design
methods: the overall conceptual seismic design
As an important lifeline project, seismic design and the local seismic capacity design.
of bridges should be paid more and more atten-
301
From the perspective of overall linear concep- invented and its seismic performance is studied by
tual seismic design for several typical bridges, the theoretical analysis and experimental testing.
four design strategies are mainly introduced in the The seismic capacities of bridge piers wholly and
chapter to make the structures dynamic character- locally reinforced with SFRC are studied based
istics adapt to the site conditions and let the seismic on the presented nonlinear material constitutive
demands become more uniform and rational along model of SFRC and the plastic hinge length and
structural components: optimal design for layout the reasonable range locally reinforced with SFRC
and detail of continuous girder bridge, design of are determined for the single-column bridge pier.
towers for long span floating cable-stayed bridge, For pile group foundations reinforced with SPPs,
seismic isolation mechanism of elastic cables in its seismic capacities, hysteretic performance
cable-stayed bridge, and seismic potential and and energy dissipation capabilities are studied
performance for long span SCC bridges. For a by numerical simulations and low cycle loading
long span continuous girder bridge, the optimal experiments. Theoretical analyses, experimental
adjustments of the piers form and locations of studies and some effective comparisons show
expansion joints are proposed to make the stiff- that the presented local seismic design strategies
ness distribution better suit to the site conditions. of components of bridges are economical, ap-
For a long span cable-stayed bridge, a new-type plicable and valid.
spatial bridge tower is proposed to replace the
original inverted Y shape tower based on elastic
response spectrum analyses. The influences of ACKNOWLEDGMENT
the cable stiffness on the dynamic characteristics,
seismic displacement and forces are investigated This research is supported by the Ministry of
for another floating system cable-stayed bridge Science and Technology of China (most recently
installed with the elastic cables seismic isolation under Grant No. SLDRCE 09-B-08), the National
device. For the new-type SCC bridge, the seismic Natural Science Foundation of China (most recent-
potential is further confirmed based on response ly under Grants No. 50978194 and No. 90915011),
spectrum analyses of a four span continuous girder and Kwang-Hua Foundation for College of Civil
bridge and a long span SCC arch bridge. Amounts Engineering at Tongji University. Such support is
of numerical analyses and result comparisons gratefully acknowledged by the authors.
show that the proposed overall conceptual seismic
design strategies can decrease the seismic demands
and achieve a balance between the seismic force REFERENCES
and displacement, and thus improve the seismic
performance for bridges. Brozzetti, J. (2000). Design development of steel-
From the perspective of local nonlinear seismic concrete composite bridges in France. Journal
capacity design for components, three strategies, of Constructional Steel Research, 55, 229243.
innovation of cable sliding friction aseismic doi:10.1016/S0143-974X(99)00087-5
bearing for bridges, bridge piers reinforced with Cao, X. (2009). Design strategy on aseismic
SFRC and pile group foundations strengthened capacity of large bridge. Doctoral dissertation,
with SPPs are proposed to improve the seismic Tongji University, Shanghai, China.
capacities for components of bridges. Based on the
advantages of typical frictional isolation devices Fan, L. (Ed.). (1997). Seismic resistance of bridge.
and displacement restraining capabilities of cables, Shanghai, China: Tongji University Press.
a new cable sliding friction aseismic bearing is
302
Fan, L., & Zhuo, W. (Eds.). (2001). Seismic duc- Ministry of Transport of the Peoples Republic of
tility design of bridge. Beijing, China: Peoples China. (2008). Guidelines for seismic design of
Communication Press. highway bridges (JTG/T B02-01-2008). Beijing,
China: Peoples Communication Press.
Gao, Y. (2009). Study on aseismic capacity of
elevated pile cap foundations of large bridge. Park, Y. J., & Ang, A. H. S. (1985). Seismic damage
Doctoral dissertation, Tongji University, Shang- model for reinforced concrete. Journal of Struc-
hai, China. tural Engineering, 111(4), 722739. doi:10.1061/
(ASCE)0733-9445(1985)111:4(722)
Gong, Y. (2003). Performance-based design of
steel building frameworks under seismic loading. Park, Y. J., Ang, A. H. S., & Wen, Y. K. (1985).
Doctoral dissertation, University of Waterloo, Seismic damage analysis of reinforced concrete
Waterloo, Canada. buildings. Journal of Structural Engineering,
111(4), 740757. doi:10.1061/(ASCE)0733-
Hayward, A. (1988). Cheaper steel-concrete com-
9445(1985)111:4(740)
posite bridges. In C. D. Buckner & I. Viest (Eds.),
Composite Construction in Steel and Concrete: Plevris, Y. (2009). Innovative computational
Proceedings of an Engineering Foundation Con- techniques for the optimum structural design
ference, New England College, Henniker, New considering uncertainties. Doctoral dissertation,
Hampshire, June 7-12, 1987 (pp. 194-206). New National Technical University of Athens, Athens,
York, NY: American Society of Civil Engineers. Greece.
Kunde, M. C., & Jangid, R. S. (2003). Seismic Priestley, M. J. N., Seible, F., & Calvi, G. M.
behavior of isolated bridges: A-state-of-the-art (1996). Seismic design and retrofit of bridges.
review. Electronic Journal of Structural Engi- New York, NY: John Wiley & Sons, Inc.
neering, 3, 140170. doi:10.1002/9780470172858
Li, D. (1997). Multi-objective optimum design Ramesh, K., Seshu, D. R., & Prabhakar, M.
of static and seismic-resistant structures with (2003). Constitutive behaviour of confined fibre
genetic algorithm, fuzzy logic and game theory. reinforced concrete under axial compression.
Doctoral dissertation, University of Missouri- Cement and Concrete Composites, 25, 343350.
Rolla, Rolla, USA. doi:10.1016/S0958-9465(02)00051-3
Liu, M. (2003). Development of multi-objective Turkington, D. H., Carr, A. J., Cooke, N., &
optimization procedures for seismic design of Moss, P. J. (1989). Seismic design of bridges
steel moment frame structures. Doctoral disserta- on lead-rubber bearings. Journal of Structural
tion, University of Illinois at Urbana Champaign, Engineering, 115(12), 30003016. doi:10.1061/
Urbana, USA. (ASCE)0733-9445(1989)115:12(3000)
Ministry of Transport of the Peoples Republic of Walther, R., Houriet, B., Isler, W., Moa, P., &
China. (1989). Specifications for earthquake resis- Klein, J. (1988). Cable stayed bridges. London,
tant design in highway Engineering (JTJ 004-89). UK: Thomas Telford, Ltd.
Beijing, China: Peoples Communication Press.
Yan, D., & Yuan, W. (2004). Conceptual seismic
design for long span cable-stayed bridges. Journal
of Tongji University, 32(10), 13441348.
303
Yang, Y. (1999). Seismic resistance of bridge and ADDITIONAL READING

fractal characteristics. Masters thesis, Tongji
University, Shanghai, China. Cao, X., Xiong, Y., Fu, Y., Guo, P., Han, P., & Yuan,
W. (2010, August). Seismic design of Hangzhou
Ye, A., Hu, S., & Fan, L. (2004). Seismic displace- Jiu-bao bridge. Paper presented at the 5th Civil
ment control for super-long-span cable-stayed Engineering Conference in the Asian Region and
bridges. China Civil Engineering Journal, 37(12), Australasian Structural Engineering Conference
3843. 2010, Sydney, Australia.
Yuan, W., Cao, X., Cheung, P., Wang, B., & Rong, Cao, X., Xiong, Y., Fu, Y., & Yuan, W. (2010).
Z. (2010, August). Development of cable-sliding Analysis of structure-water interaction of bridge
friction aseismic bearing for bridges. Paper pre- under earthquake excitation. Structural Engi-
sented at the 5th Civil Engineering Conference neers, 26(4), 6973.
in the Asian Region and Australasian Structural
Engineering Conference 2010, Sydney, Australia. Cao, X., Yuan, W., Gao, Y., & Wei, K. (2010).
Aseismic design of long-span continuous beam-
Yuan, Y., & Sun, B. (2008, October). General arch combination bridges. Earthquake Resistant
introduction of engineering damage of Wenchuan Engineering and Retrofitting, 32(3), 3035.
Ms 8.0 Earthquake. Journal of Earthquake En-
gineering and Engineering Vibration, 28, 5973. Gao, Y., Yuan, W., & Jin, X. (in press). Soil-
structure-water interaction of a cable-stayed bridge
Zhou, M. (2008). Experimental and theoretical under seismic excitation. Journal of Central South
studies on seismic performance of elevated pile University of Technology.
caps. Doctoral dissertation, Tongji University,
Shanghai, China. Wei, K., Guo, P., Han, P., Yuan, W., & Cao, X.
(2010). Seismic responses of a long multi-span
Zhu, Z., Fu, Y., Wang, B., & Yuan, W. (2010, composite bridge under traveling wave excitation.
August). Seismic study on bridge piers reinforced Structural Engineers, 26(2), 119125.
with SFRC in local region. Paper presented at the
5th Civil Engineering Conference in the Asian Wei, K., Yuan, W., & Cao, X. (2010). Study on
Region and Australasian Structural Engineering aseismic potential and performance for compos-
Conference 2010, Sydney, Australia. ite bridge. Engineering Mechanics, 27(Sup. I),
275-279.
Zollo, R. F. (1997). Fiber-reinforced concrete:
An overview after 30 years of development. Wei, K., Yuan, W., & Cao, X. (2010, August).
Cement and Concrete Composites, 19, 107122. Seismic analyses of long span steel-concrete
doi:10.1016/S0958-9465(96)00046-7 composite bridge. Paper presented at the 5th Civil
Engineering Conference in the Asian Region and
Zou, X. (2002). Optimal seismic performance- Australasian Structural Engineering Conference
based design of reinforced concrete buildings. 2010, Sydney, Australia.
Doctoral dissertation, Hong Kong University
of Science and Technology, Hong Kong, China. Yan, D., & Yuan, W. (2003). Conceptual design of
earthquake resistance for multi-span continuous
bridges in transverse direction. Journal of Tongji
University, 31(11), 12751279.
304
Yang, Y., Yuan, W., & Fan, L. (2000). Earthquake Cable-Sliding Friction Aseismic Bearing:
ground motions and site effects on seismic per- The new-type bearing which is composed of the
formance of bridges in shanghai zone. Journal pot bearing, some cables and a shear bolt.
of Tongji University, 28(3), 348352. Close Twin-Column Pier System: a pier
system made up of two symmetrical adjoining
Yuan, W., Zhou, M., & Yan, D. (2008, October).
parts (right and left part) which are half of the
Study and trend on conceptual seismic design
solid single-column pier divided along the verti-
for bridges in China. Paper presented at the 14th
cal mid-plane.
Conference on Earthquake Engineering, Beijing,
Elastic Cable Seismic Mitigation and Isola-
China.
tion Device: The elastic cables equipped between
Zhou, M., Yuan, W., & Zhang, Y. (2008, October). the tower and the girder which can adjust the
Analysis of pile equivalent anchorage length stiffness distribution of the cable-stayed bridges.
for elevated pile caps under lateral load. Paper Local Seismic Capacity Design: A seismic
presented at the 14th Conference on Earthquake design method which focuses on increasing the
Engineering, Beijing, China. seismic capacities of components or local elements
of a whole bridge in a nonlinear state.
Zhou, M., Zhang, Y., & Yuan, W. (2011). Fracture
Overall Conceptual Seismic Design: A con-
and damage mechanism of RC pile group founda-
ceptual seismic design method which focuses on
tion under low cycle loading. Key Engineering
the optimal design of whole structural system in
Materials, 452-453, 665668. doi:10.4028/www.
a linear and elastic state to make the structures
scientific.net/KEM.452-453.665
dynamic characteristics adapt to the site conditions
Zhu, Z., Fu, Y., Guo, P., & Yuan, W. (2010). Seismic and achieve more uniform and rational seismic
study on bridge columns reinforced with SFRC demands of structural components.
in plastic-hinged range. Structural Engineers, Pile Group Foundation Strengthened with
26(5), 7378. SPPs: The pile group foundation whose piles are
strengthened with the outer steel protective pipes
in the whole pile or in local regions.
Spatial Tower: The tower model which is
KEY TERMS AND DEFINITIONS
composed of four single columns in space to
replace the inverted Y shape tower in plane.
Bridge Pier Reinforced with SFRC: The pier
Steel-Concrete Composite Bridge: The
whose material is steel fiber reinforced concrete
bridge whose girder is composed of structural
in the whole pier or in local regions.
steel and reinforced concrete.
305
306
Chapter 12
Optimum Design of Structures
for Earthquake Loading by a
Cellular Evolutionary Algorithm
and Neural Networks
Saeed Gholizadeh
Urmia University, Iran
ABSTRACT
The present chapter deals with optimum design of structures for earthquake induced loads by taking
into account nonlinear time history structural response. As the structural seismic optimization is a time
consuming and computationally intensive task, in this chapter, a methodology is proposed to reduce the
computational burden. The proposed methodology consists of an efficient optimization algorithm and a
hybrid neural network system to effectively predict the nonlinear time history responses of structures.
The employed optimization algorithm is a modified cellular genetic algorithm which reduces the required
generation numbers compared with the standard genetic algorithm. Also, the hybrid neural network
system is a combination of probabilistic and generalized regression neural networks. Numerical results
demonstrate the computational merits of the proposed methodology for seismic design optimization of
structures.
INTRODUCTION been solved using gradient-based algorithms. As

the mathematical programming based methods
Optimum design of structures is a process of need gradient calculations, the considerable part
selecting the design variables such that an ob- of the optimization process is devoted to the sen-
jective function is minimized while all of the sitivity analysis and the computational work of
design constraints are satisfied. During the last these methods is usually high. Optimal design of
decades, structural optimization problems have real-world structures subject to seismic loading is
DOI: 10.4018/978-1-4666-1640-0.ch012
Optimum Design of Structures for Earthquake Loading
one of the major concerns in the field structural They are suitable particularly for problems too
engineering. When structures are subjected to complex to be modelled and solved by classical
sever earthquakes a huge amount of inertia loads is mathematics and traditional procedures.
imposed to the structures. In this case, considering The main objective of this chapter is to propose
linear elastic behavior and ignoring the nonlinear a computationally efficient methodology to op-
structural responses during the optimization pro- timum design of structures subject to earthquake
cess may lead to vulnerable structural systems. loading considering inelastic structural behaviour.
Therefore, seismic design codes suggest that, To achieve this task, an efficient genetic algorithm
under severe earthquake events, the structures (GA) based evolutionary optimization algorithm
should be designed to deform inelastically. To is employed to reduce the required analyses. Also,
achieve structural seismic design optimization a hybrid neural network system is employed to
it is necessary that the nonlinear structural time effectively predict the nonlinear time history
history analysis to be performed many times. In responses of structures during the optimization
this case, the computational burden of the opti- process.
mal seismic design process is so large that could
prevent designer from comprehensively exploring
the design space, and could ultimately result in BACKGROUND
unsuitable structures. Consequently, it is neces-
sary to employ efficient computational strategies During the last years, a number of researchers
to achieve optimal seismic design of structures have employed evolutionary algorithms to optimal
spending low computational costs. design of structures subject to dynamic loadings.
In the last decades, soft computing procedures Kocer and Arora (1999, 2002) employed GA for
have been widely used to solve massive and the optimal design of H-frame transition poles
complex engineering problems. Soft computing and latticed towers conducting nonlinear time-
includes many components and the most attrac- history analysis. They proposed the use of GA
tive ones are meta-heuristic optimization algo- and Simulated Annealing (SA) for the solution of
rithms and neural networks. As meta-heuristic discrete variable problems, although the computa-
or evolutionary optimization algorithms need not tional time required was excessive. Salajeghehand
gradient calculations they are more robust than Heidari (2005) incorporated wavelet transforms
the mathematical programming based techniques and neural networks into the GA-based optimiza-
and usually present better global behavior. Be- tion processes to predict linear structural responses
side the mentioned computational advantages, for a specific earthquake time history loading.
the disadvantage of these methods is a slow rate Lagaros et al. (2006) examined the influence of
of convergence towards the global optimum. various design procedures on the dynamic perfor-
A neural network is an interconnected network mance of real-scale steel buildings. Gholizadeh
of simple processing elements. The processing and Salajegheh (2009) employed meta-heuristic
elements interact along paths of variable con- particle swarm optimization (PSO) algorithm,
nection strengths which when suitably adapted fuzzy inference systems (FIS) and radial basis
can collectively produce complex overall desired function (RBF) neural network for optimizing
behavior. Neural networks operate as black box, linear structures subject to earthquake loading.
model-free, and adaptive tools to capture and Gholizadeh and Salajegheh (2010a) incorporated
learn significant structures in data. Their com- wavelet RBF neural network into a hybrid PSO-
puting abilities have been proven in the fields of GA optimization algorithm for seismic optimi-
prediction, pattern recognition, and optimization. zation of a real-scale steel building considering
307
linear behavior. Computational performance of process is repeated to achieve a new solution. The
GA and PSO examined by Gholizadeh (2010) process of creating the new population with elite
for dynamic design optimization of structures sites is continued until the method converges.
considering nonlinear responses while a sur- However, employing the CGA the number of the
rogate model employed to predict the structural required generations is considerably reduced dur-
responses during the optimization process. Also, ing the optimization process, but due to this fact
Gholizadeh and Samavati (2011) designed struc- that the seismic optimization process requires a
tures subject to earthquake for optimum weight great number of nonlinear time history analyses
considering linear responses by a combination the overall time of the optimization process is
of wavelet transforms, RBF neural networks still very long.
and an improved GA. Lagaros and Papadrakakis In the second strategy, in order to reduce the
(2011) employed back-propagation (BP) neural computational burden of the nonlinear time history
network to predict nonlinear seismic responses analysis, a hybrid neural network system (HNNS)
of 3D frame structures. They specified that the based on generalized regression neural network
predicted responses may be used in the frame (GRNN) (Wasserman, 1993) is employed. In this
work of performance based design of structures case, instead of performing nonlinear dynamic
to reduce the computational effort. The proposed analysis by finite element method (FEM) a neural
methodology in this chapter consists of two main network model is used to predict the necessary
computational strategies outlined as follows: nonlinear time history responses during the
In the first strategy, an efficient evolutionary optimization process. By employing the natural
optimization algorithm is employed. The main frequencies as the inputs of neural networks,
drawback of the evolutionary algorithms, espe- better performance generality can be obtained.
cially GA, is a slow rate of convergence. In the As the natural frequencies are required during
present chapter, a GA based structural optimiza- the optimization process, evaluating of these by
tion algorithm that can increase the probability of analytic methods can impose additional compu-
achieving the global optimum with accelerated tational burden to the process. In order to prevent
convergence emphasizing on structural nonlinear from this difficulty, another GRNN is employed
time history analysis reduction is employed. For to effectively predict the frequencies. During the
this purpose, the concepts of cellular automata optimization process many designs are examined
(CA) (Von Neumann, 1966) and GA are hybrid- and due to considering nonlinear response analysis
ized and the resulted hybrid algorithm is called it is probable that some of them lose their stabil-
cellular genetic algorithm (CGA). In the CGA, a ity during the ground motion and the nonlinear
small dimensioned grid is selected and the arti- dynamic analysis fails to converge. It is evident
ficial evolution is evolved by a novel crossover that such designs should be rejected. As during
and traditional mutation operations. In each itera- the optimization process, GRNN is employed
tion, cellular crossover operation produces a new to evaluate the responses instead of the exact
design at each site according to the fitness index nonlinear time history analysis, it is necessary to
of neighboring cells of each site. As the size of detect such instable structures. For this purpose, a
the population is small, the optimization process probabilistic neural network (PNN) (Wasserman,
converges to a pre-mature solution. In each pro- 1993) is used. Using the PNN, evaluating the
cess, the best solution is saved. For creating a new time history responses of the instable structures is
population the saved best solution is transformed ignored during the optimization process and this
to the new population and the remaining ones are makes the optimization process more efficient.
randomly selected. Thereafter, the optimization
308
As a test example, a 72-bar space steel tower The problem of structural optimization for
subjected to the El Centro earthquake is optimized earthquake induced loads can be formulated as
considering nonlinear time history responses. follows:
The numerical results indicate that the hybrid
methodology is a powerful and efficient tool for Minimize f ( X ) , X; X i R d ; i = 1, , n
optimal design of structures subjected to earth- (2)
quake loadings.
Subject to
g j (X , Z (t ), Z (t ), Z(t ), t ) 0 ; j = 1, , m
MAIN FOCUS OF THE CHAPTER (3)
Main focus of the present chapter is to provide X iL X i X iU ; i = 1, , n

an efficient soft computing based methodology to
achieve optimum design of structures subject to
earthquake. The methodology is a serial integra- where f, X, g, m and n are objective function,
tion of evolutionary algorithms for performing design variables vector, behavioral constraint, the
optimization and neural network for predicting number of constraints and the number of design
nonlinear time history responses of structures. variables, respectively. Also X iL and X iU are
The various issues of the proposed methodology lower and upper bounds on the ith design variable.
are described in the following subsections. A given set of discrete values is presented by Rd.
In this chapter, for the time-dependent nonlin-
Optimal Design Problem Formulation ear optimization problems, only the displacement
constraints are considered in Box 1.where dj and
The dynamic equilibrium for a finite element dj,all are the displacement of the jth node, and its
system subjected to earthquake loading can be allowable value, respectively.
written in the following usual form: To perform dynamic time history analysis con-
sidering geometrical and material nonlinearities,
( t ) + C Z ( t ) + K Z ( t ) = MI u
MZ g ( t ) ANSYS software (ANSYS, 2006) are employed. It
uses a step-by-step implicit numerical integration
(1)
procedure based on Newmarks method to solve
( t ) , Z ( t ) , Z ( t ) , u the dynamic equilibrium. In order to consider the
where M, C, K, I, Z g ( t )
transient nature of earthquake loading a simple
and t are mass matrix, damping matrix, stiffness bilinear stress-strain relationship with kinematic
matrix, unit matrix, acceleration vector, velocity hardening is adopted. As confirmed in (Lagaros
vector, displacement vector, ground acceleration et al., 2006) this law provides accurate results for
and the time, respectively. many practical applications.
Box 1.
d j (X , Z (t ), Z (t ), Z(t ), t )
g j (X , Z (t ), Z (t ), Z(t ), t ) = 1 0 ; j = 1, ..., m (4)
d j ,all
309
As all the constraints are time-dependent the space and time are discrete, and physical quanti-
consideration of all the constraints requires an ties are taken from a finite set of discrete values.
enormous amount of computational effort. Here, In its basic form, a cellular automaton consists of
the conventional method (Arora, 1999) is em- a regular uniform grid of sites or cells with a
ployed to deal with time-dependent constraints. discrete variable in each cell which can take on a
In this method the time interval is divided into ngp finite number of states. The state of the cellular
subintervals and the time-dependent constraints automaton is then completely specified by the
are imposed at each time grid point. Let the jth values si=si(t) of the variables at each cell i. Dur-
time-dependent constraint be written as: ing time, cellular automata evolve in discrete time
steps according to a parallel state transition de-
g j (X , Z (t ), Z (t ), Z(t ), t ) 0, 0 t ti termined by a set of local rules: the variables
(5) sik +1 = si (tk +1 ) at each site i at time tk+1 are up-
dated synchronously based on the values of the
where ti is time interval over which the constraints variablessnkc in their nc neighborhood at the pre-
need to be imposed. ceding time instant tk. The neighborhood nc of a
Because the total time interval is divided into cell i is typically taken to be the cell itself and a
ngp subintervals, the constraint (5) is replaced by set of adjacent cells within a given radius r. Thus,
the constraints at the ngp+1 time grid points as: the dynamics of a cellular automaton can be for-
mally represented as follows (Biondini et. al.,
g j (X , Z (t ), Z (t ), Z(t ), t ) 0; = 0, , ngp 2004):
(6)
sik +1 = (sik , snkc ), i r nc i + r (8)
The above constraint function can be evalu-
ated at each time grid point after the structure has
where the function is the evolutionary rule of
been analyzed. The objective function of con-
the automaton.
strained structural optimization problems is de-
and r are the feasible A proper choice of the neighborhood plays
fined in Box 2.where p a crucial role in determining the effectiveness
search space, and an adjusting penalty factor re-
of such a rule. In this chapter, the widely used
spectively.
Moore neighborhood (Von Neumann, 1966) of
interaction, by r=1, is adopted.
Cellular Genetic Algorithm
The CA technique can be combined with the
evolutionary algorithms to solve optimization
Cellular automata (CA) represents simple math-
problems. One of the most popular evolution-
ematical idealizations of physical systems in which
ary algorithms is GA. In the field of structural
Box 2.
f (X )

f (X ) = ngp m
if X
f (X ) + rP [ (max(g j (X , Z (t ), Z (t ), Z(t ), t ), 0)) ]
2
(7)
=0 j =1
otherwise
310
optimization some of researchers (Canyurt & X O = (1 )X d + X m (9)

Hajela, 2005; Rajasekaran, 2001; Gholizadeh &
Salajegheh, 2010b) have combined the concepts where is a random number on interval [0, 1],
of CA and GA to create cellular genetic algorithms Xd and Xm are variables in the father and mother
(CGA). The CGA presented in this chapter is a chromosomes, respectively.
modified version of the CGA proposed by Gholiza- These types of crossovers can not provide suf-
deh and Salajegheh (2010b) and is denoted as ficient information for a comprehensive search of
modified cellular genetic algorithm (MCGA). In the design space in the complex and large-scaled
the MCGA, as well as the CGA, individuals of a problems. Therefore the found solutions are
selected population are set on discrete locations usually local optima. In the MCGA a powerful
of a 2D grid. The state variables of each site are crossover operation is employed that provides
the design variables. In the MCGA, the evolution sufficient information for comprehensive and fast
process is accomplished locally, with probabilistic exploration of the design space.
interaction rules applied synchronously to each In the MCGA model, a population of potential
central site, and using information only from designs is structured in a 2D grid. In this case, each
members of its Moore neighborhood. When the site contains a real-valued string describing of a
population has been updated, the evolutionary design and therefore the state of the cellular au-
rules of the automaton are repeated until one of tomaton in each site is a vector of design variables.
the stopping criteria is met. In both the CGA and
MCGA, the objective function of the optimization si X i = {x 1 , x 2 , ..., x n }T , i = 1, 2, ..., nc
problem is employed to define the fitness of each (10)
design vector. In the MCGA the evolutionary rule
of the automaton includes the cellular crossover The CCO acts on the design variables and
operation (CCO) and the mutation operation (MO) combines the information available at the central
applied to the sites. sites and their immediate neighbors. In this case, a
In the GA type evolutionary algorithms cross- virtual individual is produced by using the fitness
over operation creates one or more offspring from indices of the individual in the immediate neigh-
the selected parents. Many different methods have bors of each central site as follows (Gholizadeh
been proposed for crossing over in the GA such as & Salajegheh, 2010b):
point crossover method. In this simple method one
or more points in the chromosome are randomly nc nc
selected as the crossover points. Then the vari- X i = X best + ( (X ibest X i , j )fj ) / fj
ables between these points are merely swapped j =1 j =1
(11)
between two parents. The problem with these point
crossover methods is that no new information is
where Xbest is the best solution found up to current
introduced. But the blending methods mitigate this
iteration. X ibest is the best individual in immediate
difficulty by finding ways to combine variables
values from the two parents into new variable neighbors of ith central cell. Xi,j is the jth indi-
values in the offspring. For instance in (Haupt & vidual in immediate neighbors of ith central cell
Haupt, 2004) such method has been proposed so and fj is its fitness value.
that a single offspring variable value XO is pro- In creation of X i the effect of individuals hav-
duced from combination of the two corresponding ing better fitness values is higher and vice versa.
parents variable values as follows: In each discrete time step, the CCO combines the
central cell with the virtual individual and pro-
311
duces a new design at each site according to the (2010b) the remaining cells are selected based on
following equation: a pure random selection but in the MCGA these
cells are selected by normal distribution about the
sik +1 = (sik , snkc ) X ik +1 = (1 )X ik + X ik elite individual of the previous stage. In this case,
(12) the MCGA compared with CGA can provide better
performance for controlling the balance between
The MO in the framework of the MCGA is exploration (global investigation of the search
similar to that used in the standard GA. In the place) and exploitation (the fine search around a
real-valued model of mutation, the value of the local optimum). Therefore, the MCGA increases
mutated design variable is replaced by a ran- the probability of founding better solutions spend-
domly selected value from the Rd . It has been ing lower computational cost. The flowchart of the
already demonstrated that the low values of mu- proposed MCGA algorithm is shown in Figure 1.
tation probability (0.001 to 0.004) is more effec-
tive, therefore, in this chapter also the value of Hybrid Neural Network System
0.004 is considered.
The MCGA is elitism based multi-stage evo- A hybrid neural network system (HNNS) is em-
lutionary algorithm. In the optimization process ployed to efficiently and accurately predict the
by MCGA, n individuals of a randomly selected nonlinear time history responses of the structures.
small initial population are set on locations of a In this neural system, GRNN and PNN, both from
2D grid and the search in the first process is com- the radial basis function neural networks family,
menced. As the size of the population is small, are serially integrated.
the optimization process rapidly converges and GRNN is a memory-based network that pro-
the best solution found in this stage, say 1X best , is vides estimates of continuous and discrete vari-
saved. In the next stage, a new elite population is ables and converges to the underlying regression
created based on the philosophy of giving more surface. GRNN has a one pass learning algorithm
chance to survive the elite individuals. In this with highly parallel structure. It does not require an
case, 1X best is copied to the n randomly se- iterative training procedure. The principal advan-
lected cells and the remaining cells are selected tages of GRNN are fast learning and convergence
as follows: to the optimal regression surface as the number
of samples becomes large. GRNN approximates
any arbitrary function between input and output
2
X j = N ( 1X best , 1X best ), j = 1, 2, ...,(n n )
vectors, drawing the function estimate directly
(13)
from the training data (Specht, 1990). GRNN is
often used for function approximation. It is a two
where N ( 1X best , 1X best ) represents a random layer feed forward network. The first layer of GR
number normally distributed with the mean of consists of RBF neurons with Gaussian activation
X and the variance of 1X best . Various values
1 best
functions while the output layer consists of linear
for are examined and the best results are obtained neurons. In the first layer it has as many neurons
by = 0.15 . as there are input-output vectors in training set.
In this case a new optimization process is Specifically, the first layer weight matrix is set to
achieved. The process of selecting elite popula- the transpose of the matrix containing the input
tions and achieving optimization processes are vectors. The second layer also has as many neurons
continued until the method converges. In fact in as input-output vectors, but here the weight matrix
the CGA proposed by Gholizadeh and Salajegheh is set to the matrix containing the output vectors.
312
Figure 1. Flowchart of MCGA
W1 = PT (14) identical to the count of the classes, is denoted by

NC. Based on these assignments, the PNN is cre-
W2 = T (15) ated with zero error on training samples (Wasser-
man, 1993). After training, a testing set of NT new
data samples are used to test the generalization
where W1, W2, P and T are the first layer weight, of the PNN. When a new input vector is pre-
the second layer weight, input and desired outputs sented, the RBF layer computes the distances
matrices, respectively. between it and the training samples. Then RBF
The PNN is mainly used for classification activation function is used to produce a vector
problems. To train the PNN a supervised train- whose elements indicate how close the input vec-
ing is accomplished. Typically, a PNN consists tor is to the training sample. Thus, the RBF layer
of an input layer, a RBF layer, and a competitive neurons with weight vectors quite far from the
(C) layer. input vector produce output values near zero,
During the training stage, a training set of NS while neurons with weight vectors quite close to
data samples is used. The number of the neurons it provide output values near one. Typically, sev-
in the first layer is identical to NS. Also, the weight eral neurons may be active to varying degrees.
matrix of this layer is set to the transpose of the The competitive layer sums these contributions
input matrix (Wasserman, 1993). The number of for each class of inputs to produce a vector of
the neurons in the competitive layer, which is probabilities. Finally, the second layer produces
313
a 1 corresponding to the largest element of the important task. All of the NS selected structures
RBF layer neurons and 0s elsewhere. Thus, the are considered in the classification phase. Natu-
network classifies the input vector into a specific rally, considering stable and instable structures
k (k = 1, 2, ..., NC ) class because that class has the number of classes, NC, is equal to 2. In the
the maximum probability of being correct. The training phase of the PNN, the inputs are Fi and
key advantage of PNN over the other networks the output is 1 for stability and 2 for instability of
is its rapid training. Since the number of layers the corresponding structure. Employing NS1 stable
in the PNN architecture is fixed and all the syn- and NS2 instable structures, the PNN is trained to
aptic weights are directly assigned using training detect stable and instable structures during the
samples, this procedure can be finished in only optimization process.
one epoch and no error correction procedure is The last stage in training the HNNS is to train
necessary. It has been proved that with enough a network to predict the nonlinear time history
training data a PNN is guaranteed to converge to responses of the NS1 stable structures. For this
a Bayesian classifier, which usually owns the purpose, another GRNN is considered. This net-
optimal classification capability (Wasserman, work is denoted as response predictor. The inputs
1993). and outputs of the GRNN are Fi and Ri of the NS1
The first step in training of HNNS is to select stable structures.
a data set. In the sampling process, NS structures In order to evaluate the accuracy of the ap-
based on their design variable vectors are selected. proximate nonlinear time history responses
The natural frequencies (Fi) and nonlinear time against their corresponding actual ones (obtained
history responses (Ri) of all the selected struc- by conventional FE analysis), two evaluation
tures are computed by the conventional FEM. In metrics are used.
(Gholizadeh et. al., 2009) it has been demonstrated
ngp ngp
that the best candidates as the inputs of the neural 1 1 1

networks for predicting the time history responses
RRMSE = ((
ngp - 1
(z i zi )2 ) /(
i =1 ngp
(z ) ))
i =1
i
2 2
of structures are natural frequencies. In this chapter (16)

also the natural frequencies are employed as the
inputs. During the optimization process evaluating ngp ngp
of the frequencies by analytic methods increases R 2 = 1 ( (z i zi )2 ) / ( (z i z )2 )

i =1 i =1
the computational effort of the process. In order (17)
to prevent from this, a GRNN is trained to predict
the natural frequencies. The inputs and outputs of where zi and zi are the ith component of the
this GRNN, denoted as frequency predictor, are exact and approximate responses, respectively.
design variables (Xi) and natural frequencies (Fi) The mean value of exact vectors component is
of the selected structures, respectively. During the expressed by z .
nonlinear time history analysis of a structure, it is In the normal mode when an unseen new
probable that the structure loses its overall stabil- sample (Xnew) is presented to the trained HNNS dur-
ity and the analysis procedure can not converge. ing the optimization process, at first the frequency
Thus, before training a neural network to predict predictor GRNN predicts its natural frequencies
the nonlinear responses, it is important to detect (Fnew). Then these frequencies are presented to
stable and instable structures. In this case, clas- the PNN to recognize the stability or instability
sifier neural networks can be employed. In the of the structure. If the structure is instable, it will
present chapter, a PNN is trained to achieve this be rejected else the response predictor GRNN
314
predicts the nonlinear time history responses of considered. The mass of 10000 kg is lumped at
the structure (Rnew). nodes of 1 to 4 of the structure. The 72 structural
The outline of the HNNS in training and normal members are divided into 9 groups, as follows: (1)
modes is shown in Figure 2. A1-A4, (2) A5-A12, (3) A13-A16, (4) A17-A24,
(5) A25-A28, (6) A29-A36 (7) A37-A40, (8) A41-
Fundamental Steps of the A48 and (9) A49-A72. The cross-sectional areas
Methodology of the elements can be chosen from the standard
Pipe profile list given as: (2.54, 11.2, 12.3, 13.9,
As explained, in the proposed methodology 15.2, 17.2, 18.9, 21.4, 25.7, 26.4, 32.1 and 33.1)
MCGA is employed to achieve optimization task 10-4m2. As the nonlinear structural behavior is
while HNNS is employed to predict the required considered only the displacement constraints are
structural responses. Fundamental steps of the included. In this case, the displacement of top
methodology are as follows: node of the structure is limited to 2 cm. The con-
straints are checked at 750 grid points with time
Step1: Data generation: A number of structures are step of 0.02 seconds. The computational time is
randomly selected and their natural frequen- measured in terms of CPU time required by a PC
cies and nonlinear time history responses for Pentium IV 3000 MHz.
earthquake loading are computed. To train and test the HNNS a training set of
Step2: Structural identification: a PNN is trained 161 samples are randomly selected and their
to detect stable and instable structures based natural frequencies and nonlinear time history
on their natural frequencies. responses are evaluated by the conventional FEM.
Step3: Frequency evaluation: a GRNN is trained During this process it is revealed that among all
to predict the natural frequencies of the the selected structures, in case of 11 ones the
structures during the optimization process. nonlinear dynamic analysis does not converge.
Step4: Nonlinear responses evaluation: another Therefore, in the data there are 150 stable struc-
GRNN is trained to predict the nonlinear time tures (NS1=150) and 11 instable ones (NS2=11).
history responses of the structures during the The time spent to FE analysis of 165 structures
optimization process. is 320 min.
Step5: Design optimization: optimal design A GRNN is trained to predict the natural fre-
process is achieved by MCGA incorporat- quencies of the structures during the optimization
ing HNNS. process. The inputs and outputs of the GRNN are
design variables (Xi, i=1,2,,161) and frequen-
Numerical Results cies (Fi, i=1,2,,161) of the selected structures,
respectively. Due to symmetry of the structure, its
To show the computational advantages of the pro- 1st, 3rd and 5th natural frequencies are considered.
posed methodology a 72-bar truss subjected to 15 From the 161 selected structures, 107 and 54 ones
seconds of the El Centro (S-E 1940) earthquake are randomly selected to train and test the network,
record is designed for optimal weight. The struc- respectively. In this case, the size of the GRNN is
ture is shown in Figure 3. The earthquake record 9-110-3. The results of testing the generalization
is applied in x direction. Youngs modulus, yield of the GRNN are given in Table 1. The time spent
stress and mass density are 2.11010 kg/m2, 2.4107 to train and test the GRNN is 0.6 min.
kg/m2 and 7850 kg/m3, respectively. A simple In order to train the PNN, 107 samples includ-
bilinear stress-strain relationship with kinematic ing 100 stable and 7 instable structures are con-
hardening by the modulus of 6.3108 kg/m2 is sidered. Also to test the PNN, 54 samples includ-
315
Figure 2. Outlines of the HNNS in (a) training and (b) normal modes
ing 50 stable and 4 instable ones are considered, corresponding natural frequencies as the inputs
respectively. The testing results of PNN, shown and the nonlinear time history displacements of
in Figure 4(a), indicate that there is good confor- the top node as the outputs. To train and test the
mance between exact results and predicted ones network 100 and 50 samples are considered, re-
by PNN. However by adding more instable struc- spectively. The size of the GRNN is 3-100-750.
tures to the data set the accuracy of the PNN can The results of testing the performance generality
be improved. The time spent to train and test the of the GRNN are shown in Figure 4(b). The time
PNN is 1.0 min. spent to train and test the network is 1.35 min.
In order to predict the top node displacement The results imply that the generalization of the
of the structure another GRNN is trained. As the GRNN is appropriate.
instable structures will be rejected during the The optimization task is achieved by the CGA
optimization process, therefore the GRNN is only and MCGA with 30 individuals using the exact
trained to predict the nonlinear responses of the nonlinear dynamic Analysis (ENDA) and approxi-
stable structures. In this case, the training set mate analysis by HNNS (HNNS). The maximum
includes the NS1 stable structures including their number of generations is limited to 100. In each
optimization case, ten independent optimal design
processes are performed and the best solution
Table 1. Testing errors of the GRNN found, the average number of generations and
the average time of optimization are given as the
Natural frequencies Mean error Maximum error final results. In the optimization process based
(%) (%)
on approximate analysis, to distinguish feasible
f1 0.7320 2.1042
and infeasible solutions, the criterion proposed
f3 0.1420 1.1308 by Vanderplaats (1999) is involved: if the sum
f5 0.4792 1.6149 of the violated constraints is less than 0.005, the
Average 0.4511 1.6166 corresponding solution is feasible, otherwise the
316
Figure 3. 72-Bar space steel truss
solution is infeasible. It should be noted that, in DISCUSSION

the optimization by CGA and MCGA incorporat-
ing HNNS, the necessary responses during the Training of the HNNS includes three steps. In the
optimization are predicted by the HNNS and to first step, a GRNN is trained to predict the first,
assess the feasibility of the final optimal solution third and fifth natural frequencies of the structures.
its responses are evaluated by the conventional FE The training task is achieved in 0.6 min and the
time history nonlinear analysis. The final results results show that the average error is 0.4511%.
of optimization are given in Table 2. This means that the generalization of the trained
It is demonstrated that the MCGA is superior GRNN is appropriate. In the second step, a PNN
to the CGA. It can be also observed that the best is trained to recognize stable structures from the
solution is obtained by MCGA using HNNS in instable ones. The time spent is 1 min and the nu-
terms of weight, time, and accuracy. The top node merical results of testing indicate that accuracy of
time history displacements of the solutions the PNN is good. As the final step, another GRNN
achieved by HNNS (CGA+HNNS and is trained to predict the time history displacement
MCGA+HNNS) are compared with their corre- the top node of the structure. In this case, the mean
sponding actual ones in Figure 5. Also, the stress- RRMSE and R2 of the predicted responses are
strain hysteretic curves corresponding to the ele- 0.0387 and 0.9971, respectively. The numerical
ment 14 of the optimal structures are shown in results of testing reveal that the generality of the
Figure 6. These curves imply that the structural HNNS is proper to provide accurate and reliable
behavior is highly nonlinear. results during the optimization phase.
317
Figure 4. (a) Testing results of PNN and (b) RRMSE and R2 of the predicted responses by GRNN
Table 2. Optimum designs
Element Groups No. CGA MCGA

ENDA HNNS ENDA HNNS
1 1 1 1 1
2 1 1 1 1
3 1 1 1 1
4 2 2 2 2
5 2 3 3 2
6 2 2 2 2
7 6 4 3 3
8 2 2 2 2
9 1 1 1 1
Weight (kg) 1099.6 1089.1 1081.4 1076.2
The average number of generations 74 78 66 63
The sum of the violated constraints 0.0000 0.0000 0.0000 0.0020
The average Optimization time (min) 4110.0 1.22 3600.0 0.95
Data generating time (min) - 320.0 - 320.0
Training time (min) - 3.0 - 3.0
Overall time (min) 4110.0 324.2 3600.0 324.0
RRMSE, R2 - 0.0732, 0.9951 - 0.0668, 0.9975
318
Figure 5. The top node displacements of the optimal design found by (a) CGA+HNNS and (b) MCGA+HNNS
compared with the corresponding actual ones
The optimization task is achieved by CGA FUTURE RESEARCH DIRECTIONS

and MCGA incorporating ENDA and HNNS. The
numerical results show that all of the solutions In this chapter, a single earthquake record is applied
are feasible and the stress-strain hysteretic curves to the structure. For the future researches, a bank
corresponding to the elements of the optimal struc- of natural or artificial records scaled according to a
tures imply that the structural behavior is highly seismic design code and considering hazard levels
nonlinear. The results obtained by MCGA are may be used. Also, other optimization algorithms
better than that of the CGA in terms of structural with higher performance may be investigated. To
weight, required generations and prediction ac- improve response prediction accuracy, other neural
curacy. The weight of the optimal structure found network models such as fuzzy network can be used.
by CGA using ENDA and its required generations As the computational work of the methodology
are 1099.6 and 74, respectively while these of the is low, it also can be used to solve the massive
MCGA are 1081.4 and 66, respectively. Also the seismic reliability based optimization problems.
corresponding weight and required generation
to the CGA+HNNS are 1089.1 and 78 while for
MCGA+HNNS process are 1076.2 and 63, respec- CONCLUSION
tively. Therefore it can be observed that MCGA is
superior to CGA and the best results are provided An efficient hybrid neural network based optimi-
by the MCGA+HNNS process. According to the zation algorithm is presented to achieve design
results the overall time of optimization using optimization of structures subjected to earthquake
HNNS is significantly reduced by a factor of 0.09 time history loading with inelastic behavior. A
compared with the optimization using ENDA. modified cellular genetic algorithm denoted as
319
Figure 6. The stress-strain hysteretic curves of the element 14 of the optimal structures (a) CGA+HNNS
and (b) MCGA+HNNS
MCGA is proposed for performing optimization using exact and predicted structural nonlinear
task. The MCGA is a multi-stage optimization responses. The numerical results imply that the
algorithm based on renewing the population in computational performance of the MCGA is better
each stage using elitism. In order to mitigate the than that of the CGA. Also it is observed that by
computational rigors of the nonlinear time history employing the HNNS the overall time of optimi-
analysis, a hybrid neural network system termed zation is about 0.09 times of the time required by
as HNNS is employed. The PNN and GRNN are exact optimization while the errors due to all the
serially integrated in the framework of the HNNS. approximations are small. Therefore, it can be
By using the HNNS the necessary nonlinear time finally concluded that the proposed methodology
history responses of the structures during the is a powerful tool to design optimization of struc-
optimization process are accurately predicted. tures subject to earthquake loading considering
Optimization is achieved by CGA and MCGA nonlinear behavior.
320
REFERENCES Gholizadeh, S., & Salajegheh, E. (2010b). A cel-

lular genetic algorithm for structural optimisation.
ANSYS Incorporated. (2006). ANSYS release In B. H. V. Topping, J. M. Adam, F. J. Pallars,
10.0, 2006. R. Bru, & M. L. Romero, (Ed.), Proceedings of
Arora, J. S. (1999). Optimization of structures the Tenth International Conference on Computa-
subjected to dynamic loads. In Leondes, C. T. tional Structures Technology, DATE, (pp. 1-14).
(Ed.), Structural dynamic systems computational Stirlingshire, UK: Civil-Comp Press.
techniques and optimization (pp. 173). Gordon Gholizadeh, S., Salajegheh, J., & Salajegheh, E.
and Breach Science Publishers. (2009). An intelligent neural system for predicting
Biondini, F., Bontempi, F., Frangopol, D. M., & structural response subject to earthquake. Ad-
Malerba, P. G. (2004). Cellular automata approach vances in Engineering Software, 40(8), 630639.
to durability analysis of concrete structures in doi:10.1016/j.advengsoft.2008.11.008
aggressive environments. Journal of Structural Gholizadeh, S., & Samavati, O. A. (2011). Struc-
Engineering, 130(11), 17241737. doi:10.1061/ tural optimization by wavelet transforms and neu-
(ASCE)0733-9445(2004)130:11(1724) ral networks. Applied Mathematical Modelling,
Canyurt, O. E., & Hajela, P. (2005). A cellular 35(2), 915929. doi:10.1016/j.apm.2010.07.046
framework for structural analysis and optimiza- Haupt, R. L., & Haupt, S. E. (2004). Practical ge-
tion. Computer Methods in Applied Mechan- netic algorithms. New Jersey: Wiley-Interscience
ics and Engineering, 194(30-33), 35163534. Publication.
doi:10.1016/j.cma.2005.01.014
Kocer, F. Y., & Arora, J. S. (1999). Optimal design
Gholizadeh, S. (2010). Structural optimization of H-frame transmission poles for earthquake
for earthquake loading with nonlinear responses loading. Journal of Structural Engineering,
by surrogate modeling based evolutionary al- 125(11), 12991308. doi:10.1061/(ASCE)0733-
gorithms. Asian Journal of Civil Engineering, 9445(1999)125:11(1299)
11(1), 2542.
Kocer, F. Y., & Arora, J. S. (2002). Optimal de-
Gholizadeh, S., & Salajegheh, E. (2009). Optimal sign of latticed towers subjected to earthquake
design of structures for time history loading by loading. Journal of Structural Engineering,
swarm intelligence and an advanced metamodel. 128(2), 197204. doi:10.1061/(ASCE)0733-
Computer Methods in Applied Mechanics and En- 9445(2002)128:2(197)
gineering, 198(37-40), 29362949. doi:10.1016/j.
cma.2009.04.010 Lagaros, N. D., Fragiadakis, M., Papadrakakis,
M., & Tsompanakis, Y. (2006). Structural opti-
Gholizadeh, S., & Salajegheh, E. (2010a). Opti- mization: A tool for evaluating dynamic design
mal seismic design of steel structures by an ef- procedures. Engineering Structures, 28(12),
ficient soft computing based algorithm. Journal 16231633. doi:10.1016/j.engstruct.2006.02.014
of Constructional Steel Research, 66(1), 8595.
doi:10.1016/j.jcsr.2009.07.006 Lagaros, N. D., & Papadrakakis, M. (2011). (in
press). Neural network based prediction schemes
of the non-linear seismic response of 3D buildings.
Advances in Engineering Software.
321
Rajasekaran, S. (2001). Optimization of large scale Specht, D. F. (1990). Probabilistic neural

three dimensional reticulated structures using cel- networks. Neural Networks, 3(1), 109118.
lular genetics and neural networks. International doi:10.1016/0893-6080(90)90049-Q
Journal of Space Structures, 16(4), 315324.
Vanderplaats, G. N. (1999). Numerical optimiza-
doi:10.1260/026635101760832244
tion techniques for engineering design. USA: VR
Salajegheh, E., & Heidari, A. (2005). Optimum and D, Inc.
design of structures against earthquake by wavelet
Von Neumann, J. (1966). Theory of self-repro-
neural network and filter banks. Earthquake En-
ducing automata (Burks, A. W., Ed.). University
gineering & Structural Dynamics, 34(1), 6782.
of Illinois Press.
doi:10.1002/eqe.417
322
323
Chapter 13
Fuzzy Identification
of Seismically Excited
Smart Systems
JinSeop Kim
Massachusetts Institute of Technology, USA
Yeesock Kim
Worcester Polytechnic Institute, USA
Tahar El-Korchi
Worcester Polytechnic Institute, USA
ABSTRACT
In this chapter, a nonlinear modeling framework to identify nonlinear behavior of smart structural
systems under seismic excitations is proposed. To this end, multi-input-multi-output (MIMO) autoregres-
sive exogenous (ARX) input models and Takagi-Sugeno (TS) fuzzy models are coalesced as the MIMO
ARX-TS fuzzy model. The premised part of the proposed MIMO ARX-TS fuzzy model is optimized using
the hierarchical clustering (HRC) algorithm, while its consequent parameters are optimized via the
weighted linear least squares estimation. The performance of the proposed model is investigated using
the dynamic response of a three-story shear planer frame structure equipped with a magnetorheological
(MR) damper subject to earthquake disturbances. Furthermore, the impact of the HRC algorithm on the
performance of the MIMO ARX-TS fuzzy model is compared with that of the subtractive and the fuzzy
C-means clustering algorithms. The equivalence of the original and identified data is numerically shown
to prove that the HRC MIMO ARX-TS fuzzy model introduced here is effective in estimating nonlinear
behavior of a seismically excited building-MR damper system.
DOI: 10.4018/978-1-4666-1640-0.ch013
Fuzzy Identification of Seismically Excited Smart Systems
1. INTRODUCTION systems (Hung et al. 2003). Among them, the

nonparametric SI approach is effective for the
The development of an accurate explicit math- complex nonlinear problems of large civil infra-
ematical model of dynamical systems is one structures, in particular, one of the nonparametric
of the most important tasks in structural health nonlinear SI methodologies that have been widely
monitoring and control system design for hazard used in the field of large civil structures is neural
prediction and mitigation of dynamical systems network (NN) because it is more readily useful
because precise mathematical information related than the parametric SI approach to identify in-
to the dynamic systems is used for damage predic- complete and incoherent measurements of large
tion and/or calculation of control forces (Kerber civil structures, in general (Smith & Chase 1993;
et al. 2007; Yen & Langari 1998; Lin et al. 2001; Masri et al. 2000; Hung et al. 2003), although
Bani-Hani 1999; Kim et al. 2011). However, it is conventional NN models have drawbacks of the
still challenging to derive a mathematical model slow convergence rate and the potential to local
of nonlinear dynamic systems (Moon & Aktan minima due to the characteristics of the black-box
2006). An example of such nonlinear dynamic model (Chassiakos & Masri 1996). On the other
systems can be sought when highly nonlinear hand, another popular nonparametric SI method
hysteretic actuators/dampers are applied to struc- for modeling complex nonlinear dynamic systems
tural systems for efficient energy dissipation: the is the fuzzy logic theory because it is effective to
structure integrated with the nonlinear control represent complex nonlinearities and uncertainties
devices behaves nonlinearly although the structure of dynamic systems in a more transparent way
itself is usually assumed to remain linear (Yi et al. (Langari 1999). Since Zadehs paper(1965), the
2001; Ramallo et al. 2004). Because the structure fuzzy logic has been widely applied to various SI
integrated with the nonlinear hysteretic control problems (Wang & Langari 1995). In particular,
device is intrinsically nonlinear, it is challenging there have been a number of studies on the TS
to develop an appropriate mathematical model fuzzy model in recent years, which provides an
for the integrated nonlinear system including the effective representation of nonlinear systems
interaction effects between the structural system with the aid of fuzzy sets, fuzzy rules, and a set
and the nonlinear control device while it plays a of local linear models (Filev 1991; Du & Zhang
key role in both structural health monitoring and 2008; Abonyi et al. 2000; Wang & Langari 1996;
controlling system (Dyke et al. 1998). Johansen &Babuska 2003; Takagi & Sugeno 1985;
Such a challenging nonlinear problem has Chen et al. 2007). On the other hand, the fuzzy
made a number of researchers pay a great deal of logic theory in the field of large scale civil infra-
attention to system identification (SI) approaches structures has been mainly used for nonlinear fuzzy
in recent years (Adeli & Jiang 2006). The non- control system design (Tani et al. 1998; Wang &
linear SI methodologies can be categorized into Lee 2002; Ahlawat & Ramaswamy 2004; Dounis
two parts: parametric and nonparametric SI ap- et al. 2007; Loh et al. 2003; Shook et al. 2008;
proaches (Bani-Hani et al. 1999). A parametric SI Casciati 1997; Yan & Zhou 2006; Choi &Schurter
method is to directly identify physical quantities & Roschke 2001; Zhou et al. 2003; Faravelli &
such as the stiffness and damping of the structural Rossi 2002; Al-Dawod et al. 2004; Battaini et
systems (Lin et al. 2001; Lin &Betti 2004; Yang al. 2004; Symans & Kelly 1999; Subramaniam
& Lin 2004); while a nonparametric SI method et al. 1996; Kim et al. 2004; Pourzeynali et al.
is to train the input-output map of the structural 2007; Kim et al. 2009; Nomura et al. 2007; Gu &
324
Oyadiji 2008). However, few investigations are MIMO ARX-TS fuzzy model can be summarized
found in the field of the nonlinear fuzzy SI for as follows: 1) The proposed modeling framework
civil building structures subjected to earthquake can be directly applied to the structure-MR damper
loadings (Adeli & Jiang 2006; Jiang & Adeli system without the decoupling process because it is
2005). Adeli and Jiang developed a fuzzy wavelet a nonlinear SI method; 2) it is more appropriate to
neural network (FWNN) model for nonlinear SI identify incomplete and incoherent measurements
of high-rise building structures. In their work, of large civil structures than typical parametric
the multi-input-single-output (MISO) FWNN SI approaches; and 3) it provides a systematic
was trained by a hybrid Levenberg-Marquardt design framework for the PDC-based nonlinear
least-squares algorithm. However, their approach TS fuzzy controller.
does not adopt a fuzzy model as an input-output In the rest of this chapter, first the HRC MIMO
mapping function, but uses a fuzzy C-mean ARX-TS fuzzy identification framework is de-
clustering technique only as one of data mining scribed, following by simulation results involving
methods for use in a neural network model. Fur- the time histories of the excitation input signals
thermore, no investigation has been conducted and the associated system output responses.
on a nonlinear Takagi-Sugeno (TS) fuzzy SI
for use with building structures equipped with a
highly nonlinear hysteretic control device such as 2. FUZZY LOGIC MODEL
magnetorheological (MR) damper. Without the
TS fuzzy model for the nonlinear building-MR In general, it is still challenging to develop a math-
damper systems, it is very difficult to design a ematical model for a dynamic model equipped
parallel distributed compensation (PDC)-based with the MR damper system because the nonlinear
TS fuzzy control system for damage mitigation dynamic system has multiple operating regions. In
of seismically excited civil structures equipped this section, a fuzzy modeling framework will be
with MR damper systems (Kim et al. 2009) and it presented to model nonlinear behavior of structure-
would not be easy to develop a damage detection MR damper systems: first, the fundamentals on
algorithm for the seismically excited structure-MR fuzzy logic models are summarized; and then
damper system. The reason is that the PDC-based HRCARXTS fuzzy model is introduced; and
TS fuzzy control design framework requires finally, simulation results are discussed.
that a TS fuzzy controller, i.e., a nonlinear fuzzy
controller, has the same premise parameters as 2.1. Membership Functions
a TS fuzzy model, i.e., dynamic model of the and Fuzzy Sets
building-MR damper system (Kim et al. 2009;
Kim et al. 2010). Therefore, this chapter proposes Membership functions (MFs) and fuzzy sets are
a fuzzy modeling framework based on the TS the cornerstone of a fuzzy logic-based system that
fuzzy model for identifying nonlinear behavior is appropriate for modeling complex nonlinear
of the building-MR damper systems subjected to systems with uncertain parameters. There exist
earthquake disturbances: hierarchical clustering- always a variety of uncertainties in engineering
based (HRC) multi-input, multi-output (MIMO) problems, e.g., the structural damage is very
autoregressive exogenous (ARX) Takagi-Sugeno large and the performance of an MR damper
(TS) fuzzy model: HRC MIMO ARX-TS fuzzy is sensitive to high temperature. However, ques-
model. The advantages of the proposed HRC tions would arise: How much damage would be
325
thought as very large quantity? or Which degree where Rj is the jth fuzzy rule; z FZ i
are premise
of temperature is high? In reality, it is impossible variables that can be either input or output values;
to model the uncertain variables in a conventional pi,j are fuzzy sets centered at the jth operating
way; while, MFs can be used for modeling such point; y = f (z FZ
1 i
, ..., z FZ ) can be any type of func-
variables as an element of a fuzzy set. Fuzzy sets
tion in terms of the premise vector
are constructed from MFs. For example, a fuzzy i
zFZ = z FZ
1
, , z FZ . The TS fuzzy model-based
set of the structural damage can be constructed as
three MFs, e.g., small, medium, and large. This reasoning is to simply compute weighted mean
fuzzy set is used for constructing the premise part values
of IF-THEN rules, i.e., IF STATEMENT.
Nr
2.2. Fuzzy Rules

w y
j =1
j j
y final = Nr
, (2)
A fuzzy rule base has a set of fuzzy IF-THEN rules; w
j =1
j
e.g., if a building structure has large damage, a

controller is operated such that an alarm is rung
twice, if the structural damage is medium, the
controller is operated such that the alarm is rung where wj is the fuzzy interpolation parameters; Nr
once, and if there is no damage in the building is the number of the fuzzy rules. To appropriate
structure, the controller is not operated. The set model nonlinear behavior of dynamic systems,
of IF-THEN rules is blended into an integrated ARX input models are applied to the consequence
system through fuzzy reasoning methods. part of the TS fuzzy model.
2.3. Fuzzy Reasoning

3. AUTOREGRESSIVE-
Fuzzy reasoning is a mechanism to perform the EXOGENOUS INPUT TAKAGI-
fuzzy inference system that derives conclusions SUGENO FUZZY MODEL
from a family of IF-THEN rules, i.e., fuzzy
reasoning is a methodology to organize a set of A nonlinear dynamic system can be described
the IF-THEN rules. Takagi and Sugeno (1985) by the following multivariable nonlinear model
developed a systematic methodology for a fuzzy
reasoning using linear functions in the consequent z = f (t, z, u), (3)
part. Because the TS-fuzzy model uses linear func-
tions in the consequent part, the defuzzification
procedure is not required. A typical fuzzy rule for where t is the time variable; z is a state vector; u
the TS fuzzy model has the form is an input vector; and f represents a multivariable
nonlinear dynamic system. The nonlinear dynamic
1
R j : If z FZ 2
is p1, j and z FZ i
is p2, j and and z FZ is pi , j model can be described by the multiple multi-input
Then y = f (z FZ
1 i
, ..., z FZ ), multi output (MIMO) ARX input-based TS fuzzy
(1) model (MIMO ARX-TS) fuzzy model
326
Box 1.
zFZ {y1 (k 1) , .., y1 (k n ) , .., y p (k n ) , u1 (k 1) , .., u1 (k m ) , .., u p (k m )} .
ni
1
R j : If z FZ 2
is p1, j and z FZ i
is p2, j and and z FZ is pi , j
i
i,j
i
(z FZ )
n m
w j (z ) = i =1
,
Then y (k ) = a y (k i ) + b u (k i ), FZ N r ni (6)

i,j i,j
i
i =1 i =1
(z )
j =1 i =1
i,j FZ
(4)
where n is the number of delay steps in the output where i , j (z FZ

i
) is the grades of membership of
signals; m is the number of delay steps in the i
z FZ ; N r is the number of local linear dynamic
input signals; y (k ) is the output; u (k ) is the input; models; and ni is the number of premise variables.
a i , j and bi , j are the consequent parameters (see Once the MIMO ARX-TS fuzzy model is set up,
Box 1). the premise parameters Pi,j and the consequent
Note that the number of the fuzzy rules cor- parameters ai,j and bi,j are determined such that
responds to the number of local linear MIMO the MIMO ARX-TS fuzzy model describes be-
ARX models, i.e., the mth local linear MIMO havior of a nonlinear dynamic system. In this
ARX dynamic model represents the mth fuzzy study, the premise parameters are determined via
rule that describes behavior of a nonlinear dy- clustering techniques, including the hierarchical,
namic system in a local linear operating region. the fuzzy C-means, and the subtractive clustering
However, a question would arise on how to blend algorithms, and the consequent part is optimized
the multiple local linear MIMO ARX dynamic using the weighted linear least squares estimation
models into an integrated nonlinear dynamic algorithm.
system model, i.e., how to construct a bridge
across the multiple MIMO ARX models. One
solution is found in the fuzzy logic-based inter- 4. OPTIMIZATION OF MIMO
polation (Yen & Langari 1998). The multiple ARX-TS FUZZY MODEL
local linear MIMO ARX models at the specific
operating point z FZ i
can be blended In the MIMO ARX-TS fuzzy identification model,
the parameters of the premise and consequent parts
) ai ,j y(k i ) + w j (z FZi ) bi ,j u(k i ), are optimized such that the MIMO ARX-TS fuzzy
n Nr n Nr
(k ) = w j (z FZ
y i
i =1 j =1 i =1 j =1
model effectively represents nonlinear behavior of
(5)
the physical system. In particular, it is desirable
to group a large data set into subsets of data with
similar patterns for efficient determination of the
where 0 w j (z FZ
i
) 1 is the normalized true premise part, i.e., the small number of MFs but
value of the jth rule, reasonable pattern recognition. Appropriate meth-
327
ods for the grouping are clustering techniques that of sets are evaluated to find the minimum distance
extract the center information of the subset of data pair, and the sets thus found are merged into a set.
within a large data set. In this study, the hierarchi- By repeating this, the number of sets decreases by
cal clustering algorithm is applied; and then, the 1 at each iteration, so the desired number Nq of
solution is compared with the other benchmark the sets are obtained after (N-Nq) iterations. The
clustering algorithms, such as the fuzzy C-means center of each cluster can be calculated from the
and the subtractive clustering algorithms. locations of data points composing it. Note that
while the procedure described above is called the
4.1. Premise Part agglomerative hierarchical clustering, a divisive
strategy can also be applied to obtain the same
The hierarchical clustering algorithm determines result. Note also that our choice of distance metrics
the membership of data to their clusters by con- produces the equivalent result as splitting the set at
structing the hierarchical organization of a given the (Nq-1) locations where the distances between
set of data, revealing the membership of each adjacent data points are farthest, since we are
datum differently at distinct hierarchy levels (Ward concerned in the clustering of one dimensional
1963; Clauset et al. 2008). Once the hierarchy data in this study.
is established, one of the levels can be chosen
either to yield a desired number of clusters or to 4.2. Consequent Part
optimize an objective function, depending on the
problem at hand. To describe the algorithm, the Once the premise part is optimized, the conse-
dissimilarity or distance between a pair of data quent part parameters can be optimized with the
up is first considered as weighted linear least squares algorithm. Based on
Gausss celebrated principle of least square (Gauss
dIJ = u p (I ) u p (J ) , (7) 1963), the linear least squares algorithm can be
formulated as a quadratic optimization problem
that minimizes the error between true values and
estimated model outputs
where the definition of distance can take any
well-behaved metric for distance. In addition, the 1
distance between a pair of sets can be defined as Min J = e(k )T e(k ), (9)
2
a function of the distances of all the pairs of data
which are extracted from respective sets, i.e.,
Duv = F {dIJ : I U p , J Vp }, (8) where e(k ) = y (k ); i.e., the error e(k ) is

(k ) y
the difference between the estimation model y (k )
and the true values y (k ). Note that the normal
linear least squares formulation can be easily
where F can be the minimum, the maximum, the extended into the weighted linear least squares
average, or any other functions of the elements, by introducing a factor of weight. Thus, in what
dIJ. In this study, we choose the Euclidean distance follows, the normal linear least squares estimator
for dIJ and the minimum for F. is derived first and then the weight factor is
The algorithm starts with allocating each data added into the normal least squares. A linear es-
point to its own set, so that each set contains only timation model for use with the linear least squares
one data point. Then, the distances of all the pairs algorithm is
328
(k ) = H(k ) j .
y (10) tion can be easily extended into a weighted least
squares estimator
1
where H(k ) is a set of independent specified j = H(k )T w ej H(k ) (k ),
H(k )T w ej y (16)
basis and j is the matrices of parameters to be
estimated. On the other hand, the true model can
be thought as a contaminated estimation model where w ej is the appropriate weighting parameter
(k ) = H(k ) j + e(k ).
y (11) and
H(k ) = [ y(k 1)T , ..., y(k n )T , u(k 1)T , ..., u(k m )T ]

From Eq. (11), the error dynamics is given by (17)
(k ) H(k ) j .
e(k ) = y (12)
and
Using the fact that a scalar equals its transpose,
substituting Eq. (12) into Eq. (9) leads to the fol- j = [a1, j , ..., a n , j , b1, j , ..., bm , j ]. (18)
lowing objective function
The nonlinear HRC MIMO ARX-TS fuzzy
(k )T H(k ) j + Tj H(k )T H(k ) j .
1
J = J ( j ) = (k )T y
y (k ) 2y modeling approach proposed in this chapter
2
(13) can be summarized: 1) nonlinear behavior of a
building-MR damper system are represented by
In this problem, the goal is to find j such that a family of multiple MIMO ARX input models
that are integrated into a nonlinear time-varying
the objective function J is minimized. For mini-
model through fuzzy rules; 2) the premise part
mization of the quadratic function of Eq. (13), the
of the multiple MIMO ARX-TS fuzzy model is
following necessary condition can be derived
determined by using the HRC algorithm; 3) the
consequent part parameters are optimized by a
j J = H(k )T H(k ) j H(k )T y
(k ) = 0,
family of weighted linear least squares. Finally,
(14) the effectiveness of the HRC MIMO ARX-TS
fuzzy model is demonstrated from a benchmark
building structure in the following section. Note
where j J is a Jacobian matrix, the first partial that the proposed HRC MIMO ARX-TS fuzzy
derivative of J with respect to j . For solution of model was used for a semiactive nonlinear fuzzy
this equation, the following analytical least squares control system design (Kim et al. 2009).
estimator is available
j = H(k )T H(k )
1
(k ).
H(k )T y (15) 5. CASE STUDY
5.1. MR Damper
By simply adding an appropriate weighting
parameter of w ej , the linear least squares estima- In recent years, smart control systems have been
considered for large civil structures because it
329
combines the best features of both active and

passive control systems. Smart control devices l 1 l
zBW = x BW y BW z BW z BW (x BW y BW ) z BW
include variable-orifice dampers, variable-
stiffness devices, variable-friction dampers, +ABW (x BW y BW ),
controllable-fluid dampers, shape memory alloy (20)
actuators, piezoelectrics, etc. (Hurlebaus & Gaul
(x BW y BW )} ,
1
2006). In particular, one of the controllable-fluid y BW = 0
(cBW 1
+ cBW )
{z BW + cBW
0
x BW + kBW
0
dampers, magnetorheological (MR) damper has (21)

attracted considerable attention in recent years
due to its appealing characteristics. In general, = a + bu BW , (22)
an MR damper consists of a hydraulic cylinder,
magnetic coils, and MR fluid comprising micron-
sized magnetically polarizable particles floating
1
cBW 1a
= cBW 1b
+ cBW u BW , (23)
within oil-type fluid. The MR damper is operated
as a passive damper; however, when a magnetic 0
cBW 0a
= cBW 0b
+ cBW u BW , (24)
field is applied to the MR fluid, it is changed into
a semi-solid state in a few milliseconds. This is
one of the most unique aspects of the MR damp- u BW = (u BW v BW ), (25)
ers compared to active systems: the malfunction
of the active control system might occur if some
control feedback components, e.g., wires and sen- where zBW and , called the evolutionary vari-
sors, are broken for some reasons during severe ables, describe the hysteretic behavior of the MR
earthquake events, while a semiactive system damper; cBW 0
is the viscous damping parameter at
operates as a passive damping system although high velocities; cBW 1
is the viscous damping pa-
the control feedback components are not function- rameter for the force roll-off at low velocities; a,
ing properly. Its characteristics are summarized b, cBW0a
, cBW
0b
, cBW
1a
, and cBW1b
are parameters that
in Kim et al. (2009)
account for the dependence of the MR damper
To fully use the best features of the MR
force on the voltage applied to the current driver;
damper, a mathematical model that portrays the 0
kBW controls the stiffness at large velocities; kBW 1
nonlinear behavior of the MR damper has to be

represents the accumulator stiffness; x BW 0
and x BW
developed first. However, this is challenging
because the MR damper is a highly nonlinear are the initial and arbitrary displacements of the
hysteretic device. In this study, a modified Bouc- spring stiffness kBW 1
, respectively; , , l and ABW
Wen model is used to predict the dynamic behav- are adjustable shape parameters of the hysteresis
ior of the MR damper force because it accurately loops, i.e., the linearity in the unloading and the
predicts the dynamic behaviors at both low and transition between pre-yielding and post-yielding
high velocity regions as shown in figure 1 (Spen- regions; v BW and u BW are input and output volt-
cer et al. 1997), where the MR damper force ages of a first-order filter, respectively; and is
fMR (t ) predicted by the modified Bouc-Wen the time constant of the first-order filter. Note that
model is governed by the following differential nonlinear phenomena occur when the highly
equations nonlinear MR dampers are applied to structural
systems for effective energy dissipation. Such an
1
fMR = cBW y BW + kBW
1 0
(x BW x BW ), (19) integrated structure-MR damper system behaves
330
Figure 1. Modified Bouc-Wen model for MR

is the mass matrix,
damper
c1 + c2 c2 0

C = c2 c2 + c3 c3 (28)
0 c3 c3

is the damping matrix,
k1 + k2 k2 0

K = k2 k2 + k3 k3 (29)
0 k3 k3

is the stiffness matrix,

nonlinearly although the structure itself is usu-
ally assumed to remain linear. fMR (t, x 1 , x1, v1 )

fMR (t, x 1 , x1 , v1 ) = 0
(30)
5.2. Building-MR Damper System
0
To demonstrate the effectiveness of the HRC
MIMO ARX-TS fuzzy model proposed in this
study, a 3-story shear planar frame structure em- is the MR damper force vector; wg denotes the
ploying an MR damper is investigated. An example ground acceleration, mi are the mass of the ith
of a building structure employing an MR damper floor, ki is the stiffness of the ith floor columns,
is depicted in Figure 2. The associated equation ci is the damping of the ith floor columns, the
of motion is given by vector x is the displacement relative to the ground,
x is the velocity, x
is the acceleration, x 1 and x1
+ Cx + Kx = fMR (t, x 1 , x1, v1 ) M w
Mx g , are the displacement and the velocity at the 1st
(26) floor level relative to the ground, respectively, v1
is the voltage level to be applied, and and
are location matrices of control forces and distur-
where the system matrices are; bance signal, respectively. The second order
differential equation can be converted into a set
m 0 0 of first order differential equations in state space
1
M = 0 m 2 0 (27) as
0 0 m 3

z 61 = A*61z61 + B*63 fMR (t, z 1 , z 4 , v1 )31 E*63 w
g 31

y91 = C z *
96 61 +D f*
93 MR (t, z1, z 4 , v1 )31 + n 91,
(31)
331
Figure 2. A 3-story shear planer frame structure employing an MR damper
where I 0

C = 0
*
I (34)
0 I M1K M1C
A* = (32)
M1K M1C

is the output matrix,

is the state matrix,
0
0
B* = 1 (33) D = 0
*
(35)
M F M1F

is the input matrix, is the feed-through matrix,
0
E* = (36)
F

332
is the disturbance location matrix, 5.4. Simulation Results
1 1 0 To demonstrate the effectiveness of the proposed

F = 0 1 1 (37) HRC MIMO ARX-TS fuzzy models, a three-story
0 building structure employing an MR damper is
0 1
investigated. Two input signals, which are a dis-
turbance signal and a control signal, are applied
to the benchmark three-story building structure to
is the location matrix that a Chevron brace is generate output data. Figure 3 shows the first input
located within the building structure, n is the signal, which is an artificial earthquake used as
noise vector, z1 and z4 are the displacement and a disturbance input signal such that the spectrum
the velocity at the 1st floor level of the three-story of the random signal includes the frequency char-
building structure, respectively, I is the identity acteristics of the earthquake ground acceleration.
matrix, and 0 is the zero matrix. Note that in the The second input is the MR damper force signal
earthquake engineering applications, the earth- as shown in Figure 4. On the other hand, the 3rd
quake disturbance excites all the floor levels floor acceleration and the 1st floor displacement
within the building structure as the inertia forces, responses are selected as output signals.
i.e. is a 3 1 vector with the same component of The model order of the MIMO ARX-TS fuzzy
the 1940 El-Centro earthquake. Properties of the model is chosen to be n = m = 2. Note that the
three story building structure are adopted from a number and the type of the MFs are determined
scaled model (Dyke et al. 1996) of a prototype via trial-and-errors. Although the architecture of
building structure that was developed by Chung et the MIMO ARX-TS fuzzy model can be optimized
al. (1989). The mass of each floor m1 = m2 = m3= via an optimization procedure (e.g., genetic algo-
98.3 kg; the stiffness of each story k1= 516,000 rithm), it is beyond the scope of the present
N/m, k2 = 684,000 N/m, and k3= 684,000 N/m; chapter. However, the authors intend to optimize
and the damping coefficients of each floor c1 = the architecture of the MIMO ARX-TS fuzzy
125 Ns/m, c2 = 50 Ns/m, and c3= 50Ns/m. In ad- model in near future. Note that the performance
dition, a SD-1000 MR damper whose parameters of the identified model can be improved by in-
are given in Table 1 is installed on the 1st floor creasing either the order of the MIMO ARX-TS
level using a Chevron brace, which leads to a fuzzy model or the number of the MFs, resulting
nonlinear dynamic model, i.e., a building-MR in the larger dimension of the fuzzy rule base.
damper system. Based on the physical model, a Figure 5 and Figure 6 compare the displacement
set of input-output data is generated for training and acceleration responses of the original simula-
the proposed HRC MIMO ARX-TS fuzzy system tion model with those of the identified HRC MIMO
identification procedures. Note that it is challeng- ARX-TS fuzzy model, respectively. As can be
ing to identify M, C, and K matrices through a seen from the figures, overall good agreements
linear time-invariant (LTI) model framework between the original values and the identified
because the building structures employing MR HRC MIMO ARX-TS fuzzy models are found in
dampers are nonlinear time-varying systems. the time histories of both displacement and ac-
Therefore, it is recommended to develop a nonlin- celeration responses.
ear time-varying model framework for modeling Table 2 shows the error of the identified MIMO
the building-MR damper system. ARX-TS fuzzy model: represents the mean
333
Table 1. Parameters for SD-1000 MR damper model
Parameter Value Parameter Value
0a
cBW 21.0 Nscm-1 a 140 Ncm-1
0b
cBW 3.50 Nscm-1 V-1 b 695 Ncm-1 V-1
0
kBW 46.9 Ncm-1 363 cm-2
1a
cBW 283 Nscm-1 363 cm-2
1b
cBW 2.95 Nscm-1 V-1 ABW 301
1
kBW 5.00 Ncm-1 l 2
0
x BW 14.3 cm 190 s-1
Table 2. Error quantities of the MIMO ARX-TS fuzzy model
Fuzzy C-means Clustering Hierarchical Clustering Subtractive Clustering
y1 y 2 y1 y 2 y1 y 2
E (y y)
2
3.194 10-4 0.468 3.685 104 0.460 5.354 104 0.666

1 y y 100
y y
81.540 78.881 80.492 78.947 78.756 76.379

Figure 3. Artificial earthquake
334
Figure 4. Magnetorheological (MR) damper force
Figure 5. Comparison of original data and MIMO ARX-TS fuzzy model: Displacement
Figure 6. Comparison of original data and MIMO ARX-TS fuzzy model: Acceleration
335
squared errors (MSE); is the fitting rate (FR), e.g., 7. CONCLUSION

if the proposed fuzzy models produce the same
responses as the simulation model, FR is 100; is In this chapter, a nonlinear fuzzy logic modeling
the output of the proposed MIMO ARX-TS fuzzy framework has been proposed to model nonlinear
models; and and are the output data from the behavior of structural systems employing smart
simulation model and the associated mean value, control devices: the hierarchical clustering-based
respectively. The performance of the proposed (HRC) multi-input, multi-output (MIMO) Au-
HRC approach is compared with benchmark toregressive eXogenous (ARX) Takagi-Sugeno
clustering algorithms, the fuzzy C-means and the (TS) fuzzy model (HRC MIMO ARX-TS fuzzy
subtractive clustering techniques. It is found from model). The HRC MIMO ARX-TS fuzzy model
Table 2 that the proposed HRC MIMO ARX-TS is developed through the integration of the HRC
fuzzy model outperforms over the other methods technique, a family of local linear ARX input
in terms of acceleration responses (i.e., ). The models, TS fuzzy model and weighted least squares
performance of identifying displacement (i.e., ) estimator. The proposed modeling framework can
and acceleration responses using the proposed be directly applied to the structure-MR damper
HRC MIMO ARX-TS fuzzy model is much bet- system without the decoupling process because
ter than the one of the subtractive clustering-based it is inherently a nonlinear system identification
approach. It is, however, shown from the table (SI) method; 2) it is more appropriate to iden-
that the identified displacement response of the tify incomplete and incoherent measurements
fuzzy C-means clustering-based approach is bet- of large civil structures than typical parametric
ter than that of the proposed HRC MIMO ARX- SI approaches; and 3) it provides a systematic
TS fuzzy model. In other words, the proposed design framework for the parallel distributed
HRC-approach effectively identifies the accel- compensation (PDC)-based nonlinear TS fuzzy
eration responses while the fuzzy C-means clus- controller. To demonstrate the effectiveness of the
tering approach better captures the behavior of proposed HRC MIMO ARX-TS fuzzy model, a
displacement responses. Note that accelerometers seismically excited building-MR damper system
are selected in the field of large-scale civil infra- is investigated. It is demonstrated from the time
structures in general in order to implement struc- history response analysis that the proposed fuzzy
tural control systems and/or structural health model is effective in identifying nonlinear behav-
monitoring systems because the acceleration re- ior of the building-MR damper system subjected
sponses are readily available, compared to other to earthquakes.
quantities such as displacements and velocities.
REFERENCES
6. FUTURE RESEARCH DIRECTIONS
Abonyi, J., Babuska, R., Verbruggen, H. B., &
In near future, the authors intend to optimize the Szeifert, F. (2000). Incorporating prior knowl-
architecture of the proposed HRC MIMO ARX- edge in fuzzy model identification. Interna-
TS fuzzy model. Also, it is recommended that the tional Journal of Systems Science, 31, 657667.
proposed models be validated using a variety of doi:10.1080/002077200290966
other disturbance signals.
336
Adeli, H., & Jiang, X. (2006). Dynamic fuzzy Chen, Y., Yang, B., Abraham, A., & Peng, L.
wavelet neural network model for structural (2007). Automatic design of hierarchical Takagi-
system identification. Journal of Structural Engi- Sugeno type fuzzy systems using evolutionary
neering, 132, 102111. doi:10.1061/(ASCE)0733- algorithms. IEEE Transactions on Fuzzy Systems,
9445(2006)132:1(102) 15, 385397. doi:10.1109/TFUZZ.2006.882472
Ahlawat, A. S., & Ramaswamy, A. (2004). Mul- Choi, K. M., Cho, S. W., Jung, H. J., & Lee, I.
tiobjective optimal fuzzy logic control system for W. (2004). Semi-active fuzzy control for seismic
response control of wind-excited tall buildings. response reduction using magnetorheological
Journal of Engineering Mechanics, 130, 524530. dampers. Earthquake Engineering & Structural
doi:10.1061/(ASCE)0733-9399(2004)130:4(524) Dynamics, 33, 723736. doi:10.1002/eqe.372
Al-Dawod, M., Samali, B., Kwok, K. C. S., & Chung, L. L., Lin, R. C., Soong, T. T., & Reinhorn,
Naghdy, F. (2004). Fuzzy controller for seismi- A. M. (1989). Experiments on active control for
cally excited nonlinear buildings. Journal of En- MDOF seismic structures. Journal of Engineer-
gineering Mechanics, 130, 407415. doi:10.1061/ ing Mechanics, 115, 16091627. doi:10.1061/
(ASCE)0733-9399(2004)130:4(407) (ASCE)0733-9399(1989)115:8(1609)
strm, K. J., & Eykhoff, P. (1971). System Clauset, A., Moore, C., & Newman, M. E. J.
identification-A survey. Automatica, 7, 123162. (2008). Hierarchical structure and the prediction of
doi:10.1016/0005-1098(71)90059-8 missing links in networks. Nature, 453, 98101.
doi:10.1038/nature06830
Bani-Hani, K., Ghaboussi, J., & Schneider, S. P.
(1999). Experimental study of identification and Dounis, A. I., Tiropanis, P., Syrcos, G. P., & Tseles,
control of structures using neural network Part D. (2007). Evolutionary fuzzy logic control of
1: Identification. Earthquake Engineering & base-isolated structures in response to earthquake
Structural Dynamics, 28, 9951018. doi:10.1002/ activity. Structural Control and Health Monitor-
(SICI)1096-9845(199909)28:9<995::AID- ing, 14, 6282. doi:10.1002/stc.83
EQE851>3.0.CO;2-8
Du, H., & Zhang, N. (2008). Application of
Battaini, M., Casciati, F., & Faravelli, L. (2004). evolving Takagi-Sugeno fuzzy model to nonlinear
Controlling wind response through a fuzzy system identification. Applied Soft Computing, 8,
controller. Journal of Engineering Mechan- 676686. doi:10.1016/j.asoc.2007.05.006
ics, 130, 486491. doi:10.1061/(ASCE)0733-
Dyke, S. J., Spencer, B. F. Jr, Sain, M. K., &
9399(2004)130:4(486)
Carlson, J. D. (1996). Modeling and control of
Casciati, F. (1997). Checking the stability of a magnetorheological dampers for seismic response
fuzzy controller for nonlinear structures. Micro- reduction. Smart Materials and Structures, 5,
computers in Civil Engineering, 12, 205215. 565575. doi:10.1088/0964-1726/5/5/006
doi:10.1111/0885-9507.00057
Dyke, S. J., Spencer, B. F. Jr, Sain, M. K., & Carl-
Chassiakos, A. G., & Masri, S. F. (1996). Identi- son, J. D. (1998). An experimental study of MR
fication of structural systems by neural networks. dampers for seismic protection. Smart Materials
Mathematics and Computers in Simulation, 40, and Structures, 7, 693703. doi:10.1088/0964-
637656. doi:10.1016/0378-4754(95)00012-7 1726/7/5/012
337
Faravelli, L., & Rossi, R. (2002). Adaptive fuzzy Johansen, T. A. (1994). Fuzzy model based con-
control: Theory versus implementation. Journal trol: Stability, robustness, and performance issues.
of Structural Control, 9, 5973. doi:10.1002/stc.5 IEEE Transactions on Fuzzy Systems, 2, 221234.
doi:10.1109/91.298450
Filev, D. P. (1991). Fuzzy modeling of complex
systems. International Journal of Approximate Johansen, T. A., & Babuka, R. (2003). Multi-
Reasoning, 5, 281290. doi:10.1016/0888- objective identification of Takagi-Sugeno fuzzy
613X(91)90013-C models. IEEE Transactions on Fuzzy Systems,
11, 847860. doi:10.1109/TFUZZ.2003.819824
Gauss, K. F. (1963). Theory of the motion of the
heavenly bodies moving about the sun in conic Kerber, F., Hurlebaus, S., Beadle, B. M., &
sections: A Translation of Theoria Motus. New Stbener, U. (2007). Control concepts for an
York, NY: Dover Publications. active vibration isolation system. Mechanical
Systems and Signal Processing, 21, 30423059.
Gu, Z. Q., & Oyadiji, O. (2008). Application of
doi:10.1016/j.ymssp.2007.04.003
MR damper in structural control using ANFIS
method. Computers & Structures, 86, 427436. Kim, H. S., & Roschke, P. N. (2006). Design of
doi:10.1016/j.compstruc.2007.02.024 fuzzy logic controller for smart base isolation
system using genetic algorithm. Engineer-
Hung, S. L., Huang, C. S., Wen, C. M., & Hsu,
ing Structures, 28, 8496. doi:10.1016/j.eng-
Y. C. (2003). Nonparametric identification of a
struct.2005.07.006
building structure from experimental data using
wavelet neural network. Computer-Aided Civil Kim, S. B., Yun, C. B., & Spencer, B. F. Jr. (2004).
and Infrastructure Engineering, 18, 356368. Vibration control of wind-excited tall buildings
doi:10.1111/1467-8667.t01-1-00313 using sliding mode fuzzy control. Journal of En-
gineering Mechanics, 130, 505510. doi:10.1061/
Hurlebaus, S., & Gaul, L. (2006). Smart struc-
(ASCE)0733-9399(2004)130:4(505)
ture dynamics. Mechanical Systems and Sig-
nal Processing, 20, 255281. doi:10.1016/j. Kim, Y., Hurlebaus, S., & Langari, R. (2010). Con-
ymssp.2005.08.025 trol of a seismically excited benchmark building
using linear matrix inequality-based semiactive
Jang, J. S. R., Sun, C. T., & Mizutani, E. (1997).
nonlinear fuzzy control. Journal of Structural En-
Neuro-fuzzy and soft computing: A computational
gineering, 136, 10231026. doi:10.1061/(ASCE)
approach to learning and machine intelligence.
ST.1943-541X.0000192
Upper Saddle River, NJ: Prentice Hall.
Kim, Y., Langari, R., & Hurlebaus, S. (2009).
Jang, J. S. R., Sun, C. T., & Mizutani, E. (1997).
Semiactive nonlinear control of a building using
Neuro-fuzzy and soft computing. Upper Saddle
a magnetorheological damper system. Mechani-
River, NJ: Prentice Hall.
cal Systems and Signal Processing, 23, 300315.
Jiang, X., & Adeli, H. (2005). Dynamic wave- doi:10.1016/j.ymssp.2008.06.006
let neural network for nonlinear identification
Kim, Y., Langari, R., & Hurlebaus, S. (2010). Mod-
of highrise buildings. Computer-Aided Civil
el-based multi-input, multi-output supervisory
and Infrastructure Engineering, 20, 316330.
semiactive nonlinear fuzzy controller. Computer-
doi:10.1111/j.1467-8667.2005.00399.x
Aided Civil and Infrastructure Engineering, 25,
387393. doi:10.1111/j.1467-8667.2009.00649.x
338
Langari, R. (1999). Past, present and future of Nomura, Y., Furuta, H., & Hirokane, M. (2007). An
fuzzy control: A case for application of fuzzy integrated fuzzy control system for structural vi-
logic in hierarchical control. Proceedings, 18th bration. Computer-Aided Civil and Infrastructure
International Conference of the North American Engineering, 22, 306316. doi:10.1111/j.1467-
Fuzzy Information Processing Society-NAFIPS, 8667.2007.00487.x
New York City, New York, (pp.760-765).
Pedrycz, W. (1984). An identification algorithm in
Lin, J. W., & Betti, R. (2004). On-line identification fuzzy relational systems. Fuzzy Sets and Systems,
and damage detection in non-linear structural sys- 13, 153167. doi:10.1016/0165-0114(84)90015-0
tems using a variable forgetting factor approach.
Pourzeynali, S., Lavasani, H. H., & Modarayi,
A. H. (2007). Active control of high rise build-
33, 419444. doi:10.1002/eqe.350
ing structures using fuzzy logic and genetic al-
Lin, J. W., Betti, R., Smyth, A. W., & Longman, gorithms. Engineering Structures, 29, 346357.
R. W. (2001). On-line identification of non-linear doi:10.1016/j.engstruct.2006.04.015
hysteretic structural systems using a variable trace
Ramallo, J. C., Yoshioka, H., & Spencer, B. F. Jr.
approach. Earthquake Engineering & Structural
(2004). A two-step identification technique for
Dynamics, 30, 12791303. doi:10.1002/eqe.63
semiactive control systems. Structural Control
Liu, W. Y., Xiao, C. J., Wang, B. W., Shi, Y., & and Health Monitoring, 11, 273289. doi:10.1002/
Fang, S. F. (2003). Study on combining subtrac- stc.43
tive clustering with fuzzy c-means clustering.
Schurter, K. C., & Roschke, P. N. (2001). Neuro-
Proceedings of Second International Conference
fuzzy control of structures using magnetorheologi-
on Machine Learning and Cybernetics, Xian,
cal dampers. Proceedings of the American Control
China, (pp. 2659-2662).
Conference, Arlington, Virginia, (pp. 1097-1102).
Loh, C. H., Wu, W. Y., & Lin, P. Y. (2003).
Sharifi, R., Kim, Y., & Langari, R. (2010). Sen-
Displacement control of isolated structures with
sor fault isolation and detection of smart struc-
semi-active control devices. Journal of Structural
tures. Smart Materials and Structures, 19, 115.
Control, 10, 77100. doi:10.1002/stc.18
doi:10.1088/0964-1726/19/10/105001
Masri, S. F., Smyth, A. W., Chassiakos, A. G.,
Shook, D. A., Roschke, P. N., Lin, P. Y., & Loh,
Caughey, T. K., & Hunter, N. F. (2000). Applica-
C. H. (2008). GA-optimized fuzzy logic control
tion of neural networks for detection of changes in
of a large-scale building for seismic loads. Engi-
nonlinear systems. Journal of Engineering Me-
neering Structures, 30, 436449. doi:10.1016/j.
chanics, 126, 666676. doi:10.1061/(ASCE)0733-
9399(2000)126:7(666)
Smith, A. H., & Chase, J. G. (1993). Identifica-
Moon, F. L., & Aktan, A. E. (2006). Impacts
tion of structural system parameters using the
of epistemic (bias) uncertainty on structural
cascade-correlation neural network. Journal of
identification of constructed (civil) systems.
Dynamic Systems, Measurement, and Control,
The Shock and Vibration Digest, 38, 399420.
116, 790792. doi:10.1115/1.2899280
doi:10.1177/0583102406068068
339
Spencer, B. F. Jr, Dyke, S. J., Sain, M. K., & Wang, L., & Langari, R. (1996). Complex systems
Carlson, J. D. (1997). Phenomenological model modeling via fuzzy logic. IEEE Transactions on
for magnetorheological dampers. Journal of En- Systems, Man, and Cybernetics. Part B, Cyber-
gineering Mechanics, 123, 230238. doi:10.1061/ netics, 26, 100106. doi:10.1109/3477.484441
(ASCE)0733-9399(1997)123:3(230)
Ward, J. H. Jr. (1963). Hierarchical grouping to
Subramaniam, R. S., Reinhorn, A. M., Ri- optimize an objective function. Journal of the
ley, M. A., & Nagarajaiah, S. (1996). Hybrid American Statistical Association, 58, 236244.
control of structures using fuzzy logic. Mi- doi:10.2307/2282967
crocomputers in Civil Engineering, 11, 117.
Yager, R. R., & Filev, D. P. (1993). Unified struc-
doi:10.1111/j.1467-8667.1996.tb00305.x
ture and parameter identification of fuzzy models.
Symans, M., & Kelly, S. W. (1999). Fuzzy IEEE Transactions on Systems, Man, and Cyber-
logic control of bridge structures using intel- netics, 23, 11981205. doi:10.1109/21.247902
ligent semi-active seismic isolation systems.
Yan, G., & Zhou, L. L. (2006). Integrated fuzzy
Earthquake Engineering & Structural Dy-
logic and genetic algorithms for multi-objective
namics, 28, 3760. doi:10.1002/(SICI)1096-
control of structures using MR dampers. Journal of
9845(199901)28:1<37::AID-EQE803>3.0.CO;2-
Sound and Vibration, 296, 368382. doi:10.1016/j.
Z
jsv.2006.03.011
Takagi, T., & Sugeno, M. (1985). Fuzzy identifica-
Yang, G., Spencer, B. F. Jr, Carlson, J. D., & Sain,
tion of systems and its applications to modeling
M. K. (2002). Large-scale MR fluid dampers:
and control. IEEE Transactions on Systems, Man,
Modeling and dynamic performance consider-
and Cybernetics, 15, 116132.
ations. Engineering Structures, 24, 309323.
Tanaka, K., & Sugeno, M. (1992). Stability analy- doi:10.1016/S0141-0296(01)00097-9
sis and design of fuzzy control systems. Fuzzy
Yang, Y. N., & Lin, S. (2004). On-line identifi-
Sets and Systems, 45, 135156. doi:10.1016/0165-
cation of non-linear hysteretic structures using
0114(92)90113-I
an adaptive tracking technique. International
Tani, A., Kawamura, H., & Ryu, S. (1998). Journal of Non-linear Mechanics, 39, 14811491.
Intelligent fuzzy optimal control of building doi:10.1016/j.ijnonlinmec.2004.02.010
structures. Engineering Structures, 20, 184192.
Yang, Y. N., & Lin, S. (2005). Identification
doi:10.1016/S0141-0296(97)00077-1
of parametric variations of structures based on
Wang, A. P., & Lee, C. D. (2002). Fuzzy sliding least squares estimation and adaptive tracking
mode control for a building structure based on technique. Journal of Engineering Mechan-
genetic algorithms. Earthquake Engineering & ics, 131, 290298. doi:10.1061/(ASCE)0733-
Structural Dynamics, 31, 881895. doi:10.1002/ 9399(2005)131:3(290)
eqe.127
Yen, J., & Langari, R. (1998). Fuzzy logic-intel-
Wang, L., & Langari, R. (1995). Decomposi- ligence, control, and information. Upper Saddle
tion approach for fuzzy systems identification. River, NJ: Prentice Hall.
Proceedings of the 34thIEEE Conference on
Decision and Control, New Orleans, LA, USA,
(pp. 261-265).
340
Yi, F., Dyke, S. J., Caicedo, J. M., & Carlson, Zadeh, L. A. (1965). Fuzzy sets. Information
J. D. (2001). Experimental verification of and Control, 8, 338353. doi:10.1016/S0019-
multinput seismic control strategies for smart 9958(65)90241-X
dampers. Journal of Engineering Mechanics,
127, 11521164. doi:10.1061/(ASCE)0733-
9399(2001)127:11(1152)
341
342
Chapter 14
Health Assessment of
Engineering Structures
Using Graphical Models
Abbas Moustafa
Minia University, Egypt
Sankaran Mahadevan
Vanderbilt University, USA
ABSTRACT
A hybrid qualitative-quantitative health assessment of structures using the bond graph theory is presented
in this chapter. Bond graph (BG) is an energy-based graphical-modeling tool for physical dynamic
systems, actuators, and sensors. BG provides domain-independent framework for modeling dynamic
systems with interacting components from multiple domains. Discrete structures are modeled using one-
to-one bond graph elements, while continuous structures are modeled using finite-mode bond graphs.
BG facilitates the construction of temporal causal graph (TCG) that links the system response to the
damaged component or faulty sensor. TCG provides qualitative damage isolation, which is not possible
using most existing system identification techniques. This leads to rapid isolation of damage and sig-
nificant reduction in computations. Quantitative identification of damage size is performed by analyzing
the substructure containing the damaged component, using the nonlinear least-squares optimization
technique, thus reducing the computations. The health assessment algorithm developed in this chapter
combines the Generic Modeling Environment (GME), the Fault Adaptive Control Technology (FACT)
software, and Matlab Simulink. Numerical illustrations on BG modeling of a hydraulic actuator and
system identification of a fifteen-story shear building and a high-rise structure under earthquake loads
are provided.
DOI: 10.4018/978-1-4666-1640-0.ch014
Health Assessment of Engineering Structures Using Graphical Models
1. INTRODUCTION 2005). In general, the parametric methods can

detect the damage location but require complete
This chapter proposes a graphical, domain- measurements and extensive computations for
independent, energy-based framework that is large structures. The non-parametric methods, on
capable of modeling multidisciplinary systems the other hand, require less measurement and have
with interacting components from structural, me- better adaptability to large structures but provide
chanical, electrical, and hydraulic domains. This a global assessment on the health status of the
framework is based on the bond graph (BG) theory structure (Gonzalez & Zapico, 2008).
introduced by Paynter (1961) and developed by Sensor performance also degrades with time
Karnopp, Rosenberg and Margolis (Rosenberg & under varying environmental conditions (Koh
Karnopp 1983, Karnopp & Margolis, 2006). For et al, 2003, De Oliveira et al, 2004, Elouedi et
example, an electrical circuit and a mechanical al, 2004, Blackshire et al, 2006, Glisic & In-
system can be described with the same bond graph audi 2007). Different degradation mechanisms
model. The use of bond graphs in electrical and have been observed in different types of sensors
mechanical engineering is well established. This (surface-bonded or fully-embedded) under various
method, however, has not received significant environmental effects such as temperature- and
research attention in civil engineering. The BG moisture-cycling (Elouedi et al, 2004, Blackshire
model of a dynamic system represents the system et al, 2006, Glisic & Inaudi 2007). Sensor perfor-
equations of motion implicitly in a graphical form mance is particularly relevant in the field of road
using bond graph elements. These elements model infrastructure where the loading conditions affect-
inertial, stiffness, damping and external forces. BG ing the main structure (traffic loads, temperature
elements include serial and parallel junctions that cycling, etc.) also affect the sensor measurements.
govern the dynamic equilibrium of the structure Details on recent sensor technologies can be found
subsystems. in (Ansari 2005, Manders et al, 2006). While sen-
Civil structures deteriorate over time and sor faults are difficult to be handled using existing
experience damage due to natural events such SI techniques, bond graphs are capable of model-
as earthquakes and wind. Structural health ing both the system components and the sensors.
monitoring (SHM) is a process that aims at pro- This enables damage detection in both structural
viding accurate and in-time information of the components and sensors. Bond graphs facilitate
structural health condition of existing structures. also the extraction of damage signatures off-line
A comprehensive review on recent advances in before sensor data collection thus providing rapid
health assessment of structures can be found in identification of the damage location through
Doebling et al (1998), Alvin et al (2003), Chang qualitative comparison of predicted and observed
et al (2003), Koh et al (2003), Lui & Ge (2005), signatures. The quantification of the damage size is
Gonzalez & Zapico (2008) and Moustafa et al performed by analyzing the substructure contain-
(2010). System identification (SI) techniques can ing the damaged component only, thus, reducing
be grouped into parametric and non-parametric the computational costs. The idea of using BG
methods. The parametric methods identify changes in system identification of frame structures was
in the structure global parameters (e.g. natural introduced by these authors (Moustafa et al, 2010).
frequencies, mode shapes and modal damping) The bond graph technique is different from
or in the local parameters (e.g. members stiff- most existing SI methods since it provides: (1)
nesses and damping) to characterize the structural graphical-modeling tool for dynamic systems
damage (Doebling et al 1998, Alvin et al. 2003, under time-varying loads, (2) domain-independent
Chang et al, 2003, Koh et al, 2003, Lui & Ge, modeling tool for dynamic analysis and health
343
assessment of physical systems across multiple sure and volume flow, and for thermodynamic
domains, (3) rapid qualitative identification of systems, they are temperature and entropy flow.
damage locations, (4) the ability to identify sensor Bond graph elements consist of five primitives
faults, (5) the ability to perform online diagnosis (passive elements), two ideal sources (active ele-
based on continuous monitoring, (6) reduction of ments), and two junctions (see Figure 1). The five
data processing errors due to absence of transfor- primitives are the Capacitance (C), the Inertance
mation to frequency domain or approximations in (I), the Resistor (R), the transformer (TF) and the
feature selection, and (7) rapid quantification of Gyrator (GY). The source elements are the effort
damage size since only the substructure containing source Se and the flow source S f , which produce
damage is analyzed. Finally, it may be noted that energy. The two junctions are the serial junction
the research carried out in this chapter represents (1-J) and the parallel junction (0-J), which connect
a novel application of bond graphs and structural various elements together. Bonds are energy
optimization to the health assessment of frame transfer pathways that connect elements and junc-
structures under earthquake loads. The next section tions and are represented as half arrows. This half
provides a brief overview on bond graph theory. arrow defines the positive direction of energy
flow. Effort and flow signals are the information
transferred through these pathways. Figure 2
2. BOND GRAPH THEORY shows the tetrahedron state demonstrating the
AND TERMINOLOGY relations related by C-, I-, and R-elements in
structural engineering setting (Paynter 1961). The
Bond graph is a graphical domain-independent vertices represent the pair, effort and momentum
framework capable of modeling dynamic systems and the pair, flow and displacement. The associ-
across multiple domains (e.g. structural, electrical, ated directed arrows imply that momentum and
mechanical, and hydraulic), thus providing a uni- displacement are computed by integrating effort
fied framework for dynamic analysis and SI of and flow, respectively. I- and C-elements represent
multidisciplinary systems. For instance, a hydrau- the state variables of the system that accumulate
lic actuator composed of electrical, hydraulic and net flow (I-element stores momentum and C-el-
mechanical components can be modeled easily ement stores displacement). Brief descriptions of
using bond graphs. Bond graph is a domain-in- bond graph elements are provided below with
dependent framework. For instance, a SDOF emphasis on translating concepts to structural
mechanical system and an electrical circuit can engineering (see Table 1 and Figure 1).
be represented using the same bond graph model
(an example is given later). The BG theory is 1. C-element (Figure a(1)): C-element relates
based on the energy conservation and energy effort to time integral of flow (i.e., displace-
exchange among system components. Irrespective ment) and is used to model linear or rotational
of the domain, power is the product of the two springs, or stiffness of a structural member.
conjugate variables effort e and flow f. In struc- The constitutive relation for a spring mod-
tural and mechanical systems, e and f describe t
the force and velocity (for translation) or torque eled as a C-element is e = 1 / C f dt ,

and angular velocity (for rotation) at a point in
where e is the force in the spring, f is the
the system. In each domain there is such a com-
bination of variables for which a physical inter- velocity and C is a constant
pretation exists. In electrical networks, e and f are ( k = 1 / C = stiffness ). The C-element
voltage and current. In hydraulics, they are pres- receives flow (cause) and produces effort
344
Figure 1. Bond graph elements: (a) passive elements (b) active sources (c) junctions
(effect). The small vertical line in Figure damper, e and f are related through a linear
a(1) is the causal stroke representing the relation (e = R f ). R-elements can have
integral causality relation (constitutive equa- the causal stroke at either ends.
tion). The half arrow defines the direction 4. TF-element (Figure 1(a4)): TF-element
of power flow ( power = e f ). can represent an ideal electrical transformer
2. I-element (Figure 1(a2)): I-element models or a massless lever. It can also represent the
inertial effects in structural systems, and mode shape at a point in the structure. TF-
inductance effects in electrical or fluid sys- element does not store or absorb energy. It
tems. For a mass m subjected to a force p , conserves power and transmits the factors
t t
of power with power scaling as defined by
f = (1 / m ) e dt = (1 / m) p dt . the transformer modulus r. The balance rela-

tion between the power variables for the
This relation represents the Newtons second
transformer of Figure 1(a4) is e1 f1 = e2 f2 ,
law of motion. Herein, the effort history is
integrated to generate flow thus I-element which leads to f2 = rf1 or e2 = (1 / r ) e1 .
receives effort (cause) and generates flow 5. GY-element (Figure 1(a5)): The gyrator
(effect). establishes the relationship between flow to
3. R-element (Figure 1(a3)): R-element mod- effort and effort to flow, gain keeping the
els energy dissipation components, such as, power on the ports the same. The simplest
dampers, dashpots, electrical resistors and gyrator is a mechanical gyroscope or an ideal
valves or losses in fluid lines. For a linear DC motor. Gyrators are used when power
345
Figure 2. Tetrahedron of state showing the relations of state variables and the constitutive relations in
structural engineering (Paynter 1961)
from one energy domain is transferred to tural systems, Se represents an external

another domain. The Gyrator modulus force, and Sf represents an input velocity to
defines the relation between effort and flow. the system. The causal stroke implies that
For the GY of Figure 1(a5), e2 = f1 and either the effort (force) or flow (velocity) is
e1 = f2 . generated by the source and is imposed on
6. Effort and flow sources (Se, Sf) (Figures the system. In either case, the half arrow is
1(b1) & 1(b2)): These are active BG ele- pointing away from the source indicating
ments since they produce energy. In struc-
Table 1. Conversion of Bond graph elements and terminology to structural engineering
Bond graph term Corresponding term in structural engineering Units

Capacitance (C) N/m
Inductance (I)
Models member or spring stiffness k = 1 /C N s2 / m
Resistance (R) Models inertia effects, mass m = I Ns/m
Transformer (TF) Models damping, friction or resistance forces D = R --
Gyrator (GY) Massless lever or participation factor in modal analysis --
Effort ( e ) Mechanical gyrator or ideal DC motor N
Flow ( f ) Force or torque F (t ) = e(t ) m/s
Displacement m
Effort source Se
Velocity or angular velocityu (t ) = f (t ) --
Flow source Sf
Power
f dt --
Nm/s
Energy E A source that exerts a force on the system (e.g. an actuator) Nm
Momentum P A source that exerts a velocity on the system Ns
Power = effort flow = e f
E = e f dt
P = e dt
346
that power flows from the source into the sum to zero. In a 0-J, the strong bond determines
system. the effort at the junction and is stroked nearer to the
7. Serial and parallel junctions (Figures junction. The constitutive relations for the 0-J are:
1(c1) & 1(c2)): A serial or 1-junction has
equality of flows (velocities), and efforts e1 = e2 = e3 = e4 = e5 ; f1 f2 f3 f4 f5 = 0
(forces) sum up to zero. The equality of (3)
velocities or displacements represents the
continuity condition between substructures Equations (1, 2 and 3 and the constitutive
that is known in FEM analysis. The summa- equations for BG elements define the relations
tion represents the D Alemberts principle among the system parameters and responses,
that implies dynamic equilibrium of forces which represent the basis for the SI technique
acting on the system or a substructure of it. developed in this chapter.
This junction represents a common veloc- This section provided a brief description of
ity point in a structural system. Thus, a 1-J BG theory and terminology. The modeling of
can model the velocity at a floor level in a discrete systems, including SDOF and MDOF
multi-story building. systems and an actuator composed of hydraulic
and mechanical components, using bond graphs
In a 1-junction only one bond, referred to as is developed in section 3. Continuous structures
strong bond, brings the information of flow (ve- are tackled in Sections 3 and 4, respectively. The
locity) and other bonds receive it (i.e., only one derivation of temporal causal graphs from bond
bond is open ended and other bonds are stroked graphs is explained in Section 5. Section 6 devel-
away from the junction). According to Figure ops the modeling of sensors using the bond graph
1(c1), bond 3 is the strong bond and thus the flow theory. The development of the health assessment
is governed by bond 3 and is imposed on other methodology is presented in Sections 7 and 8.
bonds. Conversely, the effort is determined by Numerical illustrations are provided in Section 9.
other bonds and is imposed on bond 3. Strong
bonds are used in constructing TCG from BG as
demonstrated in the next section. 3. MODELING DISCRETE
The constitutive relation of power for the DYNAMIC STRUCTURES
1-junction of Figure 1(c1) is given by: USING BOND GRAPHS
e1 f1 e2 f2 e3 f3 e4 f4 e5 f5 = 0 (1) This section demonstrates the construction of

bond graphs for discrete dynamic structures.
Since the 1-junction has equality of flows, First, we consider SDOF and MDOF structures.
Eq. (1) leads to: Subsequently, we demonstrate the modeling
procedure for an actuator composed of hydraulic
e1 e2 e3 e4 e5 = 0 (2) and mechanical components. Discrete structures
are those structures having lumped properties
(inertia, damping and stiffness). Verifications of
Eq. (2) represents the D Alemberts dynamic bond graphs are provided by deriving the equa-
equilibrium principle. The 0-junction (0-J) is the tions of motion of the structure from the bond
counterpart of the 1-junction and represents com- graph model which are shown to be identical to
mon effort (force) points in the system (Figure those obtainable from structural dynamics theory.
1(c2)). Thus a 0-J has equal efforts, and the flows
347
Figure 3. Single-domain representation of mechanical and electrical systems using bond graphs
3.1 Single-Degree-of- m x(t ) + D x(t ) + k x (t ) = p(t ) (5)

Freedom (SDOF) System
The free-body diagram of the forces acting
Figure 3(a) shows a SDOF system driven by an on the SDOF system (dynamic equilibrium of
external dynamic load p(t).The system is charac- the system using D Alembert principle) leads to
terized by a single flow f3 (velocity x ) and is Eq. (5) obtained from the BG model of Figure
thus represented by a 1-junction with I-, C-, R- and 3(c). The electrical circuit of Figure 3(b) is also
Se-elements representing the inertial, spring, represented using the same bond graph model.
damping and external forces, respectively. Note This example demonstrates use of bond graphs
that the 1-J has equal flows (f1 = f2 = f3 = f4) and as a unified tool for modeling dynamic systems
efforts sum to zero (e1 e2 e3 e4 = 0). The BG from different energy domains. Note that Figure
model of Figure 3(c) represents the equation of 3(d) represents the Temporal Causal Graph (TCG)
motion of the system in an implicit form. To which will be discussed later.
derive this equation from the BG of Figure 3(c),
consider the inertial force e3 given by the consti- 3.2 Multi-Degree-of-Freedom
tutive relation of the I-element as: (MDOF) Systems
mf3 = e3 =e1 e2 e4 (4) 3.2.1 Construction of

Bond Graph Model
Substituting
Figure 4(a) depicts an n-DOF system with each
mass driven by a dynamic load pi (t ) and charac-
f3 = x(t ) , e1 = p(t ) ,
terized by a distinct flow or velocity xi (t ) . Refer-
ring to section 3.1 and Figure 4(a), the steps of
e2 = (1 / C ) f2dt = kx (t ), and
constructing bond graphs for MDOF systems can
be summarized as follows:
e4 = R f4 = R f3 = D x(t ),
1. Model each mass velocity as 1-J. Insert I-
where D is the damping coefficient, in the above
and Se-elements to model inertial and ex-
equation, we get:
ternal forces.
348
Figure 4. BG and TCG representation of MDOF discrete dynamic systems and sensors
349
2. Insert 0-J between 1-J to accommodate force- energy storage elements. For the I-element, the
generating elements (stiffness and damping state variable is flow and for the C-element the
forces). state variable is effort. The n-DOF system has
3. Simplify the resulting bond graph by remov- 2n state variables (e.g., e2, f3, e7, f11, e14, f18, ).
ing the zero velocity at the support points. Each flow variable represents the velocity of the
4. Assign numbers, and directions to bonds associated mass and efforts represent the stiffness
using the half arrow (defines the direction forces in the springs. The inertial force of the first
of power). mass is given as (bond 3) in Box 1.
5. Assign causality to bond graph elements and Note the substitutions of equality of flows and
bonds (details provided in next subsection). balance of efforts at 1-junctions and equality of
efforts and balance of flows at 0-junctions.
These procedures result in the lumped pa- Similarly, the inertial force in bond 11, (see Box
rameters of damping (D1, D2, , Dn), stiffness 2).
(k1, k2, , kn) and inertia (m1, m2, , mn) being Considering the inertial force of the (n-1)th
interconnected using energy conserving junctions mass, one gets: (see Box 3).
producing the topological bond graph model Finally, the equilibrium equation of the forces
shown in Figure 4(b). Each mass is represented acting on the n th mass can be shown to be shown
as a separate block connected to neighboring in Box 4.
blocks through bonds. Sensors are also modeled as Combining Eqs. (6- 9), the equations of motion
separate blocks for measuring displacements and of the n-DOF structure is given in a matrix form
are discussed in Section 6. Note that Figure 4(c) in Box 5.
represents the TCG which will be discussed later. Eq. (10) represents the well known equation
of motion of the n-DOF structure derivable using
3.2.2 Causality Assignment and structural dynamics principles (Clough & Penzien
Bond Graph Verification 2003, Chopra 2007). Note that, if the structure is
subjected to ground acceleration y(t ) , the forces
To assign causality to BGs we assign fixed cau- at the floor levels are given as
salities to sources. Integral causality is assigned pi (t ) = mi y(t ); i = 1, 2, ..., n .
to storage elements (I- and C-elements) and is Thus the BG is simply a topological represen-
propagated through junctions. I- and C-elements tation of the system equations of motion, using
have integral causality since cause is integrated elements that exchange efforts and flows. The BG
to provide effect. Causality is then assigned to model is correct only if the correct equations of
R-elements and bonds connecting junctions. The motion of the system can be derived from it. The
procedure is continued until the entire BG is as- above example derived the equations of motion for
signed causality. an n-DOF system by hand for sake of explanation.
The BG model of Figure 4(b) represents the This derivation can be also automated (Manders
equations of motion of the n-DOF system in an et al, 2006).
implicit form. The verification of the BG model
is performed by deriving the system equations 3.3 Hydraulic Actuator System
of motion from the BG and comparing them
with those from structural dynamics theory. The This section illustrates the modeling of multidis-
derivation of the system equations from the bond ciplinary dynamic systems using BG. Figure 5(a)
graph is systematic and can be easily coded. The shows a simplified model of a hydraulic actuator
system state variables are associated with the
350
Box 1.
m1 f3 = e3 = e1 e2 e4 e5 = p1 (1 / C 1 ) f2dt D1 f4 e6 = p1 k1x 1 D1 f3 e7 e8
= p1 k1x 1 D1x1 (1 / C 2 ) f7dt D2 f8 = p1 k1x 1 D1x1 k2 f6dt D2 f6

(6)
= p1 k1x 1 D1x1 k2 ( f5 f9 )dt D2 ( f5 f9 )
= p1 k1x 1 D1x1 k2 (x 1 x 2 ) D2 (x1 x 2 )
m1x1 + (D1 + D2 )x1 D2x 2 + (k1 + k2 )x 1 k2x 2 = p1 (t )
Box 2.
m 2 f11 = e11 = e9 + e10 e12 = e6 + p2 e13 = e7 + e8 + p2 e14 e15

= (1 / C 2 ) f7dt + D2 f8 + p2 (1 / C 3 ) f14dt D3 f15 (7)
= k2 (x 1 x 2 ) + D2 (x1 x 2 ) + p2 k3 (x 2 x 3 ) D3 (x 2 x 3 )
m 2x2 D2x1 + (D2 + D3 )x 2 D3x 3 k1x 1 + (k2 + k3 )x 2 k3x 3 = p2 (t )
Box 3.
mn 1xn 1 Dn 1xn 1 + (Dn 1 + Dn )xn 1 Dn xn kn 1x n 1 + (kn 1 + kn )x n 1 kn x n = pn 1 (t ) (8)
Box 4.
mn xn Dn xn 1 + Dn xn kn x n 1 + kn x n = pn (t ) (9)
Box 5.
m1 0 0 ... 0 0 x1 (t ) D1 + D2 D2 0 ... 0 0 0 x1 (t )

0 m2 0 ... 0 0 x 2 (t ) D2 D2 + D 3 D3 ... 0 0 0 x 2 (t )
... ... ... ... +
... ... ...
... ... ... ... ... ... ...
... +
0 0 0 ... mn 1 0 xn 1 (t ) 0 0 0 ... Dn 1 Dn 1 + Dn Dn x n 1 (t )
0 0 0 ... 0 mn xn (t ) 0 0 0 ... 0 Dn Dn x n (t )
(10)
k1 + k2 k 2 0 ... 0 0 x 1 (t )
0 p1 (t )

kk2 k2 + k 3 k 3 ... 0 0 0 x 2 (t ) p2 (t )
... ... ... ... ... ... ... ... = ...

0 0 0 ... kn 1 kn 1 + kn kn x n 1 (t ) pn 1 (t )
0 kn x n (t )
p (t )
0 0 ... 0 kn n
351
(used in aircrafts) with interacting components variables (two capacitors and one inductance)
from mechanical and hydraulic domains. The namely e4, e8 and f7. Thus the equation of motion
actuator is composed of a single chamber and a is of a third order. To verify the BG model, the
piston of cross-sectional area A. The amount of effort e4 can be calculated as the force in the first
fluid entering the chamber qin through a hydraulic spring (bond 4) in Box 6.
pump is controlled by the control valve. The pres- Similarly, the equation governing the second
sure difference between right and left sides of the capacitor is given as:
chamber (pr - pl) results in a force that controls
the movement of the piston a distance u(t). The e8 = (1 / C )f8 = kf7 (12)
2
piston is attached by a rigid rod to a mass M that is
connected to a spring and a damper which model
The equation for the inductance element is
the aerodynamic effects. The displacement is con-
given as:
trolled either manually or automatically based on
a feedback system. The description of the actuator
parameters are given in Table 2. Mf7 = e7 = e6 e8 e9 = Ae4 e8 e9
The bond graph model for the actuator of (13)
Figure 5(a) is shown in Figure 5(b). Note that
Figure 5(c) represents the TCG which will be Substituting Eqs. (11) and (12) into Eq. (13)
discussed later. The reservoir is represented by and making use of e9 = Bf7 , we get:
an effort source Se and the pump is represented
by a transformer TF1 with efficiency parameter Mf7 = (1 / C )(Aqin A2 f7 ) kf7 Bf7 (14)
PE. The valve is represented by a resistor R1 = RV.
The difference in the chambers pressure pr - pl Substituting f7 = u (t ), f7 = u(t ), f7 = u(t )
(i.e. e3) is represented by a 0-Junction that ensures
into Eq. (14) and simplifying leads to:
equal pressures (e3 = e4 = e5). The fluid com-
pressibility is represented by a C-element
3u(t ) 2u(t ) u(t )
C 1 = V0 . The pressure difference in bond 5 (e5 MC 1 3
+ BC 1 + (kC 1 + A2 ) = Aqin
t t 2 t
= pr pl = e3 = e4) is multiplied by the cross- (15)
sectional area of the piston A to provide the force
acting on the mass M. This is represented by the Equation (15) represents the equation of motion
transformer element TF2, connecting bonds 5 and governing the dynamics of the hydraulic actuator
6, with the transformer parameter A. The inertial that can be obtained using traditional mechanical
effect of the mass M is modeled by an I-element. system derivation. Therefore, the mathematical
The spring is represented by a C-element and the basis of bond graphs is the same as the classical
damper is represented by an R-element R2 = B. dynamic theory. One should not, however, over-
The displacement u(t) is the time integral of the look the advantages of bond graphs (e.g., topologi-
flow (velocity) in bond 7. There are three state cal modeling tool, modeling sensors, qualitative
Box 6.
e4 = (1 / C )f4 = (1 / C )( f3 f5 ) = (1 / C )(S f Af7 ) = (1 / C )(qin Af7 ) (11)

1 1 1 1
352
Figure 5. Single domain representation of a hydraulic actuator having interacting components from
mechanical and hydraulic domains
353
Table 2. Description of the actuator parameters

Figure 6(a) depicts the modeling of continuous
Parameter Description dynamic systems using the finite-mode bond
A Cross sectional area of the piston
graphs. The transformers ensure that each mode
V0 Volume of the chamber is excited by the corresponding modal force. The
Bulk density of the fluid
number of modes to be retained in the dynamic
Rv Resistance of the valve
Se Effort source analysis varies based on the boundary conditions
pr Chambers pressure at right side and the dynamic characteristics of the structure
pl Chambers pressure at left side
pr - pl Pressure difference (Clough & Penzien 2003, Chopra 2007). The as-
M Mass sociated TCG is shown in Figure 6(b) and will be
k Spring stiffness
B Damping coefficient discussed later.
u(t) Displacement of the mass The dynamic analysis of continuous struc-
tures using the finite-mode bond graph approach
requires performing free vibration analysis to
determine the natural frequencies and mode shapes
damage identification, etc.) which are not provided
of the structure, see, e.g., Clough & Penzien, 2003
by most existing system identifications methods.
and Chopra, 2007. The free vibration analysis is
The next section demonstrates the modeling of
carried out based on the boundary conditions of
continuous structures using bond graphs.
the structure and can be performed analytically or
numerically (e.g. using the finite element method).
The natural frequencies and mode shapes are
4. MODELING CONTINUOUS
then used to quantify the parameters of the bond
STRUCTURES USING FINITE-
graph elements (I, C, R and TF) for each mode of
MODE BOND GRAPHS
vibration. In general, the first few modes usually
provide fairly acceptable results.
This section demonstrates the modeling of con-
tinuous structures such as beams and frames using
bond graphs. The modeling methodology is based
on the normal mode approach (Clough & Penzien
5. CONSTRUCTING
2003, Chopra 2007) and is known as finite-mode
TEMPORAL CAUSAL GRAPH
bond graph (Karnopp et al, 2006). To do this, the
FROM BOND GRAPH
external force (represented by an effort source Se)
The step that follows the construction of the bond
is connected to a 0-junction that is connected to
graph model is to use this model to develop a
n 1-junctions using transformers (Figure 6(a)).
framework that facilitates damage detection and
The 0-J decomposes the external forces into their
isolation. This is achieved by deriving the tem-
modal components. The 1-junctions represent the
poral causal graph (TCG) from the bond graph
vibration modes retained and the transformers
(Mosterman & Biswas, 1999). The TCG repre-
represent the mode shape evaluated at the point
sents the constitutive relations of BG elements
of application of the external force. The displace-
and junctions in graphical form. More precisely,
ment of the ith 1-J represents the contribution to
the TCG represents causal relations among the
the structure response in the generalized coordi-
system variables and parameters. The TCG is a
nates. The I-, C- and R-elements (inertia, stiffness
directed graph in which the vertices represent the
and damping forces) connected to the 1-junction
system variables and the directed edges express
represent the modal forces.
the relation between the vertices. The labels on
the edges determine the type of causal relation
354
between the variables. These relations are similar mass and then the connecting bonds (e.g. bonds 5
to the forms of Eqs. (1, 2 and 3 but in a qualita- and 12). Note that the TCG for the second, third,
tive form (=, +1, -1). For instance, Eq. (3) reveals , nth masses are identical and that each bond
that the efforts e1 and e2 are equal, the flows f1 graph block is represented by an associated TCG
and f2 have +1 relation (i.e., increase in f1 leads to block. The procedures are systematic and can be
increase in f2 and vice-versa) and f2 and f3 have -1 coded (Manders et al, 2006).
relation (i.e., increase in f1 leads to decrease in f2 The temporal causal graph for the actuator of
and vice-versa). The directions of the arrows in the Figure 5(a) is shown in Figure 5(c). The energy
TCG are defined based on the strong bond. Thus, source Se defines the effort direction from e0 to e1
a strong bond in a 1-junction governs the flow of operated by the pump efficiency PE. The direction
the strong bond with arrows directed towards the of the flow takes the opposite direction (i.e. from
flows of other bonds. The efforts of other bonds f1 to f0). Bond 2 is the strong bond (indicated by
will have arrows directed towards the effort of the the causal stroke or the perpendicular line away
strong bond. A strong bond in a 0-J governs the from the junction) at the 1-junction (common flow
effort at the junction with arrows directed towards junction f1 = f2 = f3) and thus f2 defines the flow
efforts of other bonds. at the junction. Accordingly f2 is connected with
To construct the TCG for the SDOF system f1 and f3 with arrows starting from f2 and pointing
of Figure 3(a) from the BG model, it is first towards f1 and f3. Since f1 = f2 = f3, therefore = sign
noted that the 1-J has equal flows and efforts sum is shown in these arrows. Conversely, the effort
to zero. We start with the external force p(t), be- is determined by e1 and e3 and thus the arrows are
ing imposed on the system, directed towards e1. directed from e1 and e3 towards e2 due to the rela-
Since bond 3 is the strong bond, thus f3 determines tion e1 - e2 e3 = 0 (or e2 = e1 e3) at the 1-junc-
the flow and e1, e2, e4 determine the effort at the tion, +1 and -1 signs are shown above the arrows
junction. Therefore, the flow arrows are drawn that connect e2 with e1 and e3, respectively. Bond
directed from f3 to f2, f4 with equal signs (equal 4 is the strong bond for the 0-junction (common
flow joint) and the efforts arrows are directed effort junction e3 = e4 = e5). Therefore, the effort
from e1, e2, e4 to e3. The labels on the arrows con- is defined by bond 4. Accordingly, e4 is con-
necting efforts are determined from the equation nected with e3 and e5 by arrows pointing to e3 and
e3 = e1 - e2 e4. Thus, the arrow connecting e3 and e5 with equality sign. The flow is determined by
e2 carries the label -1 indicating the sign between f3 and f5 (f3 f4 f5 = 0 or f4 = f3 f5) and thus the
e3 and e2. To complete the TCG model we trans- arrows connect f3 and f5 to f4 with +1 and -1 signs,
late the constitutive relations for I-, C- and R- respectively. These signs are determined from the
elements. For the I-element e3 = mf3 or relation f4 = f3 f5. The efforts e5 and e6 are re-
f3 = 1 / m e3dt thus the label is dt/m (i.e. effort lated by the area of the piston with an arrow
pointing from e5 to e6. The arrows for the flows
is integrated to provide flow). For the R-element, f5 and f6 take the opposite direction. The strong
the equation is e4 = Df4 and thus the label is D bond of the last 1-junction (f6 = f7 = f8 = f9) is bond
(being multiplied by f4 to give e4). For the C- 7 and thus the flow at this junction is defined by
element, e2 = k f2dt , thus the arrow carries the f7. Accordingly f7 is connected to f6, f8 and f9 by
label k. This completes the TCG of the SDOF arrows pointing from f7 towards f6, f8 and f9 with
system. equality sign. The effort at the junction is defined
To build the TCG for the MDOF system of by e6, e8 and e9 and is shown by arrows connect-
Figure4, we start by constructing the TCG for each ing these efforts to e7. The effort relation at the
1-junction is e6 e7 e8 e9 = 0 or e7 = e6 - e8 - e9.
355
Figure 6. Finite-mode BG and TCG representation of continuous dynamic systems and sensors
356
Accordingly, the arrows carry the signs 1, -1 and such as, temperature, humidity, and pore pressure,
-1, respectively. The constitutive relation for the and (c) Chemical, such as, chloride penetration,
resistor R1 is e2 = Rv f2 or f2 = 1 / Rv e2 which sulfate penetration, pH, carbonatation penetration,
is represented by an arrow connecting e2 to f2 rebar oxidation, steel oxidation, and timber decay.
and carries the constant 1/Rv. The constitutive Conventional sensors based on mechanical and
relation for the C-element C1 is electrical transducers are able to measure most
e4 = 1 / C 1 f4 dt = (1 / V0 ) f4dt and is rep- of these parameters. Fiber-optic sensors offer
superior performance compared with conventional
resented by an arrow connecting f4 to e4 and
sensors and have been used during the last few
carries the constant (1 /V0 ) . The constitutive
years in SHM of civil structures. More details on
relation for the I-element is e = Mf or7 7 recent sensor technologies can be found in Ansari
f7 = 1 / M e dt which is represented by an
7
(2005) and Manders et al (2006).
arrow connecting e7 and f7 pointing towards f7 In general, sensor performance and reliability
and carries the term 1/M dt. The constitutive rela- degrade with time under varying environmen-
tion for the second C-element is tal conditions and externally applied loadings
(Ansari, 2005, Blackshire et al, 2006, Manders
e8 = 1 / C 2 f8dt = k f8dt and is represented
et al, 2006, Glisic & Inaudi 2007). Different
by an arrow connecting f8 to e8 and carries the degradation mechanisms have been observed in
term k dt. The flow f7 is integrated to compute the different types of sensors (e.g. surface-bonded
displacement u(t). This completes the TCG or fully-embedded) under various environmental
model for the actuator. effects such as temperature- and moisture-cycling
The TCG model of the continuous systems (Blackshire et al, 2006, Glisic & Inaudi 2007).
derived from the bond graph model of Figure Deviations in measurements could result from
6(a) is shown in Figure 6(b). The derivation of damage in structural components or due to faults
the TCG follows the same procedures explained in the sensors. Sensor performance is particularly
above. Again, each modal mass is represented by relevant in the field of road infrastructure where
a separate TCG block. The modeling of a sensor the loading conditions affecting the main structure
that measures the structure tip displacement is (traffic loads, temperature cyclying, etc.) also
also included and is explained in the next section. affect the sensor measurements. In such situa-
tions, the use of traditional system identification
methods may not provide accurate identification
6. MODELING SENSOR FAULTS of actual damage cause since these methods do
USING BOND GRAPHS not account for sensor faults. On the other hand,
bond graphs can easily model both the physical
A typical health monitoring system consists of system and sensors (Daigle et al, 2006).
a network of sensors that collect measurements Faults in sensors could result in bias or drift
data periodically or continuously during short or in measured values. A biased measurement devi-
long terms. For civil structures, such as bridges, ates from its true value by a constant B. Drift is
tunnels, dams, power plants, high-rise buildings the growth of deviations over time. For a sensor
and historical monuments, the most relevant measuring the displacement response, a bias fault
measurement parameters are (Ansari, 2005, implies that the actual measured displacement
Manders et al, 2006): (a) Mechanical, such as, dm is the sum of the true displacement dt and the
strain, deformation, displacement, crack opening, bias B . The sensor is modeled as a modulated
stress and forces, (b) Physical or environmental,
357
source of flow (MSf) representing the relation the BP of positive deviation in x(t) leads identify-
dm (t ) = dt (t ) + B (Figure 4(b)). Thus, for the ing the damage causes as reduction in either k or
displacement sensor measuring the displacement D. The FP for reductions in k and D leads to the
of the first mass e7n-2= dm, e7n-1 = dt and e7n damage signature 00+ (see table 3).
= B1. The C-element (C = 1 ) integrates the true Consider the MDOF structure of Figure4. An
velocity to give the true displacement. This leads example is shown in Figure 7(a) for a deviant
to e7n 2 = d7n 1 + B1 or dm (t ) = dt (t ) + B1 . In displacement x1(t) being above normal. From the
case of drift, the bias quantity B1 is replaced TCG of Figure 4(c), an increase in x1(t) initiates
with a time drift function (i.e., backward propagation along f3 dt
x 1 and
dm (t ) = dt (t ) + ddrift (t ) ). Each sensor fault is implies that the first derivative of f3 is above
represented by separate BG and TCG blocks. normal (f3+). The step along f3 +1
e3 implies
e3 above normal (e3+), and e3
1
e2 implies
e2 below normal and along e2 k1
f2 implies
7. EXTRACTION OF DAMAGE k1 below normal, and so on. Eventually, we arrive
SIGNATURES FROM TEMPORAL at k1, k2, k3, D1, D2, D3 below normal due to x1(t)
CAUSAL GRAPH being above normal.
The forward propagation starts from one of
This section describes the extraction of the damage the damage causes identified in the previous step
signature matrix that lists the qualitative effects by propagating the change in this parameter and
of changes in system components on system tracing its signature on the system measurements.
response measurements. This step is performed An example of k1- (k1 below normal) is shown in
off-line using the TCG model before the collection Figure 7(b). The signature of k1- on x1 is derived
of sensor data. The TCG is derived as described by assuming that k1 decreases and following the
in Section 5. The TCG links damages to their path shown in Figure 7(b), extracted from Figure
causal effects on measurements, called damage 4(c). A decrease in k1 implies a decrease in e2
signatures, which represent 0th through kth order
from the relation e2 = k1 f2dt and is indicated
derivative changes on a measurement residual at
the instant of damage occurrence. They provide with a downward arrow. This implies an increase
the discriminatory power in the damage isolation in e3 since e2 = e1 e3 e4 e5 (shown with an
approach. upward arrow). The increase in e3 implies an
The damage signature derivation consists of increase in the first derivative of f3 from the rela-
two steps: backward propagation (BP) and forward t
propagation (FP). First, the TCG is used to per- tion f3 = (1 / m ) e3dt (shown with two upward

form a BP from possible deviant measurements to arrows), and so on. Finally, this forward propaga-
generate possible damage causes. This identifies a tion implies an increase in the second derivative
set of parameters that could be the reason for the of x1 which is indicated by three upward arrows
damage (e.g. k and D in the SDOF example). It also and the signature is 00+. The signatures of k2, k3,
determines whether the cause is above normal (+) D1, D2, D3, B1, B2, B3, dr1, dr2 and dr3 on the sys-
or below normal (-). In the FP step we use each of tem responses x1, x2 and x3 are extracted following
these damage causes to derive its signatures on the the same procedure and are shown in Table 4.
system measurements (e.g. effect of decrease in In Table 4, damage signatures are shown up
k or D on x). The damage signature of the SDOF to the first nonzero direction of change. Damage
structure of Figure 4 is given in Table 3. Herein, scenarios which produce discontinuities on the
358
Table 3. Damage signatures for the SDOF building up to the first non zero direction of change
Damage type Damage in structural components Sensor faults

k-
D -
k+
D +
B -
B +
dr- dr+
Damage signature (x) 00+ 00+ 00- 00- - + 0- 0+
measurements (0th order changes) provide ad- signature of damage in k1 is the same as the sig-
ditional discriminatory power. Higher order effects nature of damage in D1 (same observation is
eventually manifest as first order effects, and since valid for k2 and D2). Considering that the MDOF
we can only measure magnitude and slope reliably, system models a multi-story building, this obser-
only the first change (and whether it was discon- vation is consistent since the reduction in either
tinuous) is useful. If we measure displacement, the stiffness or the damping coefficient implies
the damage signatures represent a qualitative damage to the columns of the same floor. When
measure that reflects how a displacement observa- multiple causes are identified in the qualitative
tion could be affected by a change in one of the step, then all those causes are included in the
structure parameters. It is observed from Table 4 quantification step which helps to discriminate
that the damage in the stiffness parameter of k1, between them.
k2 and k3 have different signatures on the mea-
surements x 1 , x 2 and x 3 . On the other hand, the
Figure 7. Damage signature extraction
359
Table 4. Damage signatures for the MDOF building up to the first non zero direction of change
Response Damage in structural components Sensor faults

quantity
k1- k2- k3- D1- D2- D3- B1+ B2+ B3+ dr1+ dr2+ dr3+
x1 00+ 00+ 000- 00+ 00+ 000- + 0 0 0+ 00 00
x2 000+ 00- 00- 000+ 00- 00- 0 + 0 00 0+ 00
x3 0000+ 000- 00- 0000+ 000- 00- 0 0 + 00 00 0+
8. HYBRID QUALITATIVE-
QUANTITATIVE SYSTEM
[r (ti ) N 1 (t j )]
j j
1 1
IDENTIFICATION TECHNIQUE N 2 (t j ) =
N2

i = j N 2
r (ti ) ; N2 1 (t j ) =
N1

i = j N 1
(16)
The health assessment method developed here
consists of damage detection, qualitative damage The symbol generation step follows the
isolation, and quantitative damage identification comparison of the measurements from healthy
(Figure 8). Here, p is the external load and r is and damaged structures. If the measurement of
the residual, the difference between the current the damaged structure at a given point of time is
structural response x , obtained from sensor above normal, a (+) symbol is assigned, and if
measurements, and the undamaged structural below normal, a () symbol is assigned.
response x , determined using the BG model.
Note that the BG and TCG models and the dam- 8.2 Qualitative Damage Isolation
age signatures are derived before the sensor data
collection. This makes the damage detection, The damage isolation step begins once damage is
isolation and quantification efficient and inexpen- detected. Having developed the damage signatures
sive. The steps to be discussed in the next subsec- matrix in Section 6, this step is a monitoring of
tions are done after the collection of the sensor system response measurements over time and
health monitoring data. comparison with the damage signatures matrix to
identify possible damage causes. For the MDOF
8.1 Damage Detection and Symbol structure of Figure 4, the system measurements
Generation are the displacements x1, x2,, xn. These measure-
ments are compared with those for the undamaged
The detection scheme is based on comparison of structure. The damage that matches the damage
measurements from damaged and undamaged signature is the cause of damage. If multiple causes
structures. This leads to estimating the residuals, are identified, the quantitative step in next sec-
which, when statistically significant, imply dam- tion is used to discriminate between them. This
age occurrence. Since measurements contain qualitative approach is very fast, due to off-line
noise, a statistical test (Z-test) is used to examine development of the damage signature beforehand,
if the structure is damaged or not (Biswas et al, and thus enables on-line health assessment.
2003). The deviation of the residual r(t) from zero
is defined for the last N2 samples and the variance
of r(t) is estimated for a larger data points N1,
where N 1 >> N 2 as follows:
360
Figure 8. Damage identification for dynamic systems
8.3 Quantitative Damage squares method to the substructure containing the

Identification damaged floor only. This offers substantial savings
in the computational effort. For instance, if the
The qualitative damage isolation described above damage is isolated in the columns of the first or
identifies the damaged component without quan- the second floor we consider the substructure con-
tifying the damage size. For discrete structures taining the first two floors. Thus, the equations of
(i.e. the MDOF system that models a multi-story motion for the first two floors are given in Box 7.
building), the qualitative isolation identifies which These two equations can be recast in a matrix
floor has damage. The damage size is quantified form in Box 8.
by applying nonlinear optimization using the least-
Box 7.
m1x1 (t ) + [D1 + D2 ] x1 (t ) D2x 2 (t ) + [k1 + k2 ]x 1 (t ) k2x 2 (t ) = p1 (t )

(17)
m 2x2 (t ) D2x1 (t ) + [D2 + D3 ]x 2 (t ) D3x 3 (t ) k2x 1(t ) + [k2 + k3 ]x 2 (t ) k3x 3 (t ) = p2 (t )
Box 8.
D1

D2

x1 x1 x 2 0 x1 x1 x 2 0 D3 p1 (t ) m1x1 (t ) (18)
=
0 x 2 x1 x 2 x 3
0 x 2 x1 x 2 x 3 k1 p2 (t ) m 2x2 (t )
k
2

k3
361
Box 8.
Dn 1

xn 1 xn 2 xn 1 xn x n 1 x n 2 x n 1 x n Dn pn 1 (t ) mn 1xn 1 (t )

(19)
=

0 xn xn 1 0 x n x n 1 kn 1 pn (t ) mn xn (t )
k
n
Box 9.
Di2

Di1

xi2 xi 3 xi 2 xi 1 0 x i 2 x i 3 x i 2 x i 1 0 Di pi 2 (t ) mi 2x i 2 (t )
=
0 xi 1 xi 2 xi 1 xi 0 x i 1 x i 2 x i 1 x i ki 2
pi 1(t ) mi 1x i 1(t )

ki 1

ki
(20)
The number of unknown system parameters is the discrete number of time instants and t is
here is s = 6. If damage is qualitatively isolated the time step, Eq. (21) can be rewritten as:
in one of the columns of the top two floors, the
equations governing the motion of the substructure [ A(t )](l2)s [R ]s1 = [ p(t )](l2)1 (22)
containing the top two floors are given in Box 8.
In this case, s reduces to 4. If the damaged
columns are located between the third floor and Alternatively, Eq. (22) can be expressed as:
the (n-3)th floor, the equations of motion that s
describe the damaged substructure are given in A R ji i = p j ; j = 1, 2,...,s l (23)
Box 9. i =1
Herein, s = 6 and 3 i n 2 . Equations

(18), (19) or (20) can be written in matrix form Assuming Ri is the predictor of the ith param-
as: eter Ri , the total error in this estimate is given
by:
[ A(t )]2s [R ]s1 = [ p(t )]21 (21)
sl s 2
= [p j Aji R
] (24)
where, A(t) is a matrix of system responses, R is j =1
i
i =1
a vector of the unknown system stiffness and
damping parameters, p(t) is a vector of the exter-
To minimize the total error, is differentiated
nal excitation and the inertia forces and s = 4 or
with respect to Rj and is set to zero which leads
6. Considering that the structural response is
measured for a time duration T = l t where l to:
362
sl s sl
graph damage detection, diagnosis and identifica-
[ A R ]A
j =1 i =1
ji i jq = p j Ajq ; q = 1, 2, ...,s
j =1
tion algorithm is executed using Matlab.
(25)
9.1 Health Assessment of a
Eq. (25) gives s (4 or 6) simultaneous algebraic Fifteen-Story Shear Building
equations which when solved provide the system under Earthquake Load
parameters including the stiffness and damping
coefficient of the damaged columns. This example studies dynamic analysis and health
Thus, the quantitative damage identification assessment of a 15-story shear building under the
is performed for the substructure that contains first horizontal acceleration of the 1995 Kobe
the damaged component by minimizing the er- earthquake (Takatori station) scaled to 0.30 g
ror in the identified structural parameters. This PGA (PEER, 2005). We investigate the dynamic
offers tremendous savings in the computations. analysis of the structure first. The numerical
Damage quantification of continuous systems values of floor masses, columns stiffnesses and
follows the same procedures and is discussed in damping used in the BG simulation analyses are
the next section. given in Table 5. This structure was modeled using
a fifteen degree-of-freedom mechanical system
with linear behavior.
9. NUMERICAL ILLUSTRATIONS
9.1.1 Health Assessment of the
Two numerical examples are considered to Structure
demonstrate the use of bond graphs for dynamic
analysis and health assessment of discrete and Damage in structural components is simulated
continuous systems. The first example considers as reduction of 20% in the stiffness parameters
a 15-story building under earthquake load. The of the fourteenth floor (case 1) and the first floor
second example demonstrates the health assess- (case 2). Damage in the sensors is modeled as
ment of a high-rise building subjected to simulated bias or drift in the fourteenth floor or first floor
Kanai-Tajimi acceleration using the finite-mode sensors. The procedures described in Sections
bond graphs. 3 and 5 are used to generate the BG and TCG
The bond graph models of both structures are models of the structure. The BG and TCG mod-
created using the Fault Adaptive Control Technol- els are constructed from those shown in Figure
ogy (FACT) software (Mosterman & Biswas, 4 by duplicating the blocks of the third floor and
1999) in the Generic Modeling Environment third sensor. The damage signatures are derived
(GME-6, 2006) to generate a simulation model as described in Section 7.
(Matlab Simulink model) (Mathworks, 1999) that
is used to simulate the structural response. FACT 9.1.1.1 Damage in Structural Components\
contains a library of bond graph elements (i.e., I-, We first study damage in the structural compo-
C-, R-, Se-, Sf-, TF- and GY-elements and con- nents. For noise-free measurements, a reduction
necting bonds) that are used to assemble the bond of 20% is introduced to the stiffness of the four-
graph model in GME. The displacement, velocity, teenth floor. The damage criterion is taken as
and acceleration responses for the building are deviation in response measurements by more than
computed using the simulation model. The bond 5% from corresponding values of healthy structure.
The damage was successfully detected to be in
363
Table 5. Numerical values for mass, stiffness and damping coefficient for the 15-story building
Floor No. 1 2 3 4 5 6 7 8
Mass (kg) 103.2 97.2 97.2 95.3 100.2 108.1 76.6 80.0
3 64.5 63.4 72.7 72.1 68.7 52.3 48.5 51.5
Stiffness 10 (N/m) 645 634 727 721 687 523 600 650
Damping (N s/m)
Floor No. 9 10 11 12 13 14 15
Mass (kg) 90.0 90.0 108.1 76.6 80.0 90.0 70.0
3 48.1 68.7 52.3 48.5 51.5 48.1 16.8
Stiffness 10 (N/m) 700 687 523 600 650 700 168
Damping (N s/m)
the columns of the fourteenth floor due to devia- 9.1.1.2 Faults in Sensors
tion of displacement measurements x14. Similar Damage in sensors is performed by simulating
analyses with 2%, 4% and 6% noise level are also sensor measurements of the fourteenth floor
carried out and the identification algorithm detects displacement as bias (0.02 m) or drift (0.002 t)
and isolates the damage to be in the fourteenth functions from the true measurements at t = 5.0
floor. Herein, the Z-test is used to detect the dam- s (Figure 9). The diagnosis algorithm correctly
age. The parameters N 1 , N 2 , were taken as 50, detects the fault to be in the sensor at the four-
5 and 1.0, respectively. The quantitative identifi- teenth floor, since no other damage cause provides
cation scheme described in Section 7 leads to similar damage signatures. To quantify the bias
identifying the stiffness and damping parameters value, the two measurements at the fourteenth
of the last two floors, see Table 6. For noise-free floor are subtracted. Notwithstanding that the
measurements, the identified stiffness and damp- bias function contains noise, the average value,
ing parameter of the last two floors contain very however, was found to be about 0.019 m which
small errors. For 2% noise measurements, the is close to the correct value. Alternatively, the
reduction in the stiffness was identified close 19%. true and measured data can be denoised before
The accuracy of the identified parameters was computing the bias value. For the noise-free case,
seen to decrease for higher values of noise (see the difference leads to correctly quantifying the
Table 6). Denoising of sensor measurements will bias value to be 0.02 m.
definitely improve the accuracy of the identified Further numerical analyses with faults simu-
parameters. lated in the sensor measuring the displacement
It was found that the time step of the ground response of the first floor were also conducted.
acceleration of 0.01 s provides inaccurate results. The health assessment algorithm successfully
Therefore, interpolation of the ground acceleration detects, isolates, and quantifies the correct damage
at smaller time period of 0.005 s was performed. to be in the first floor sensor.
Similar analyses with damage simulated as reduc-
tion in the stiffness of the first floor columns were 9.2 Health Assessment of a High-
also conducted and the health assessment algo- Rise Building under Simulated
rithm was found to detect and isolate the damage Kanai-Tajimi Acceleration
successfully. Some of these results are provided
in Table 6. Note that s = 6 in this case. In the previous example discrete structures were
modeled using one-to-one bond graph elements.
364
Table 6. Actual and identified parameters for the 15-story shear building due to structural damage
Parameter Actual Identified values

values
Noise-free 2% noise 4% noise 6% noise
k14* 38480 38469 (0.03%) 38415 (0.17%) 38291 (0.49%) 38118 (0.94%)
D14 700 695.36(0.66%) 690.72 (1.33%) 676.87 (3.30%) 698.75 (4.57%)
k15 16800 16797 (0.02%) 16761 (0.23%) 16694 (0.63%) 16610 (1.13%)
D15 168 166.34(0.99%) 164.68 (1.98%) 159.71 (4.94%) 155.89 (7.21%)
k1** 51600 51596 (0.01%) 51223 (0.73%) 50652 (1.84%) 50272 (2.57%)
D1 645 643 (0.31%) 660.80 (2.46%) 678.68 (5.22%) 687.95 (6.66%)
k2 63400 63392 (0.02%) 62935 (0.73%) 62230 (1.85%) 61760 (2.59%)
D2 634 632 (0.32%) 656.15 (3.49%) 682.05 (7.58%) 696.09 (9.79%)
k3 72700 72688 (0.02%) 72157 (0.75%) 71335 (1.88%) 70786 (2.63%)
D3 727 724 (0.41%) 754.73 (3.81%) 787.71 (8.35%) 805.91 (10.85%)
Note: The numbers in parentheses indicates % error in identified parameter.
* Case 1: Damage in columns of fourteenth floor.
** Case 2: Damage in columns of first floor.
The modeling of continuous dynamic systems us- assessment analysis. The free vibration analysis
ing bond graphs is carried out using finite-mode was carried out analytically (Blevins, 1979).
bond graphs as described before. This example Damage in the structural components is mod-
demonstrates the health assessment of a reinforced eled as a reduction of 20% in the stiffness of the
concrete high-rise building using the finite-mode structure. The sensor fault is modeled as either bias
bond graphs. It may be recalled here that the of 0.02 m or drift function of 0.002 t, at t = 5.0
modal analysis technique is used to model the s. Figure 10 shows the input load and simulated
continuous system with an equivalent discretized noisy measurements (5% noise level). Possible
multi-degree-of-freedom system. The bond graph damage scenarios are taken as changes in modal
and temporal causal graph models of the building stiffness and damping in all modes. It should be
are shown in Figure 6. The bond graph and TCG noted here that the transformer index (TF = the
graph model the structural components and the participation factor) is a positive quantity for
sensor as well. The structure is taken to be sub- odd modes and is negative for even modes. For
jected to a simulated Kanai-Tajimi random ground the case of structural damage, the identification
acceleration at the support points and the response algorithm detects damage and isolates it to be in
measurement is taken as the displacement at the the modal stiffness or modal damping. Similarly,
tip point of the structure. The input acceleration the identification algorithm successfully isolates
and simulated displacement measurements with the sensor fault (bias or drift) to be the sensor.
5% noise are shown in Figure 10. In this example, damage detection and
The structure properties used in simulating the qualitative damage isolation are carried out. It is
sensor measurements are taken as h = 75.00 m, possible to perform damage quantification using
E = 2.0 1010 N/m2 and mass density of 2500 the least-squares method to quantify the modal
kg/m3. The structure is taken to have a uniform parameters. To do this, additional sensors should
cross sectional area of 8.00 m outer diameter and be placed for collecting required measurements
7.20 m inner diameter. A modal viscous damping for the quantification process. Given that changes
of 5% damping ratio is assumed for all modes. in the modal parameters of the structure provide
The first 5 modes are considered in the health a measure of the structural damage at the global
365
Figure 9. Example 1: Sensor measurements with 5% noise (a) Fourteenth floor displacement with dam-
age in structural components (b) Fourteenth floor displacement with sensor bias
Figure 10. Example 2: Input-output measurements with 5% noise (a) Simulated Kanai-Tajimi accelera-
tion (b) Tip displacement with damage in structural components (15% reduction in 1 ) (c) Tip displace-
ment with sensor bias 0.02 m (d) Tip displacement with sensor drift 0.002 t
366
level and not at the local level, this step was not linear material behavior need to be studied. The
considered in this chapter. study also considered a single damage to occur
at a time. The inclusion of multiple damages will
be considered in a future work. Notwithstanding
10. CONCLUDING REMARKS these shortcomings, it is believed that bond graphs
AND FUTURE CHALLENGES will open the door for new useful applications
in system identification and structural control
A new graphical energy-based framework for of engineering structures. The issue of sensor
dynamic analysis and health assessment of dy- performance is particularly relevant in the field
namic systems is developed in this chapter. This of road infrastructure where the loading condi-
framework is based on the bond graph theory, the tions affecting the main structure (traffic load,
associated temporal causal graphs, the derived temperature cycling, etc.) affect also the sensor
damage signatures and the least-squares optimiza- performance. It is difficult, however, to consider
tion method. The framework is different from other all these concerns in this chapter.
frameworks in the sense that it enables performing
qualitative system identification first, thus reduc-
ing the computations significantly. Subsequently, ACKNOWLEDGMENT
damage quantification is carried out if necessary.
BG is shown to represent the system equations of This study was partly supported by funds from the
motion in an implicit graphical form. The TCG U. S. Air Force Research Laboratory (Grant No.
links the system response to changes in the sys- USAF- 0060-43-0001). This support is greatly
tem components and sensor measurements which acknowledged.
facilitates deriving qualitative effects of changes
in system response on system components. The
advantages of the BG framework are: (1) Simple REFERENCES
graphical-modeling tool for dynamic systems,
(2) Unified (domain-independent) framework Alvin, K. F., Robertson, A. N., Reich, G. W.,
for dynamic analysis and health assessment of & Park, K. C. (2003). Structural system iden-
dynamic systems across multiple domains (e.g., tification: from reality to models. Computers
structural, mechanical, electrical, hydraulic, etc.), & Structures, 81(12), 11491176. doi:10.1016/
(3) Rapid qualitative identification of the damage S0045-7949(03)00034-8
location, thus reducing the computations, (4) The Ansari, F. (2005). Sensing issues in civil structural
ability to identify sensor faults, (5) The ability to health monitoring. Springer. doi:10.1007/1-4020-
perform online diagnosis, based on continuous 3661-2
monitoring, (6) Reduction of data processing,
online computations and associated errors due Biswas, G., Simon, G., Mahadevan, N., Narasim-
to absence of transformation to other domain han, S., Ramirez, J., & Karsai, G. (2003). A robust
or approximations in feature selection, and (7) method for hybrid diagnosis of complex systems.
Rapid quantification of damage size since only Proceedings of 5th Symposium on Fault Detection,
the substructure containing damage is analyzed. Supervision and Safety for Technical Processes
In this chapter, linear structural behavior and (pp. 1125-1131). Washington, DC.
relatively simple structures are considered. More
complex systems, continuum elements and non-
367
Blackshire, J. L., Martin, S., & Cooney, A. Glisic, B., & Inaudi, D. (2007). Fibre optic methods
(2006). Characterization and modeling of bonded for structural health monitoring. Chichester, UK:
piezoelectric sensor performance and durability John Wiley & Sons. doi:10.1002/9780470517819
in simulated aircraft environments. Proceedings
GME-6. (2006). A generic modeling environment,
of 3rd European Workshop on Structural Health
GME 6 users manual. Vanderbilt University.
Monitoring, Granada, Spain.
Retrieved from http://www.isis.vanderbilt.edu
Blevins, R. D. (1979). Formulas for natural fre-
Gonzalez, M. P., & Zapico, J. L. (2008). Seis-
quency and mode shape. New York, NY: Litton
mic damage detection in buildings using neural
Education Publishing Inc.
networks and modal data. Computers & Struc-
Chang, P. C., Flatau, A., & Liu, S. C. (2003). tures, 86(3-5), 416426. doi:10.1016/j.comp-
Review paper: Health monitoring of civil infrastruc.2007.02.021
structure. Structural Health Monitoring, 2(3),
Karnopp, D. C., Margolis, D. L., & Rosenberg, R.
257267. doi:10.1177/1475921703036169
C. (2006). System dynamics, modeling and simula-
Chopra, A. K. (2007). Dynamics of structures tion of mechatronic systems. John Wiley & Sons.
(3rd ed.). Englewood Cliffs, NJ: Prentice-Hall.
Koh, C. G., Chen, Y. F., & Liaw, C.-Y. (2003).
Clough, R. W., & Penzien, J. (2003). Dynamics of A hybrid computational strategy for identifica-
structures (3rd ed.). Computers & Structures Inc. tion of structural parameters. Computers &
Structures, 81(2), 107117. doi:10.1016/S0045-
Daigle, M., Koutsoukos, X., & Biswas, G. (2006).
7949(02)00344-9
Distributed diagnosis of coupled mobile robots.
IEEE International Conference on Robotics and Lui, E. M., & Ge, M. (2005). Structural damage
Automation, (pp. 3787-3794). identification using system dynamic properties.
De Oliveira, R., Frazao, O., Santos, J. L., &
doi:10.1016/j.compstruc.2005.05.002
Marques, A. T. (2004). Optic fibre sensor for
real-time damage detection in smart composite. Manders, E. J., Biswas, G., Mahadevan, N., &
Computers & Structures, 82(17-19), 13151321. Karsai, G. (2006). Component-oriented model-
doi:10.1016/j.compstruc.2004.03.028 ing of hybrid dynamic systems using the generic
modeling environment. Proceedings of 4th Work-
Doebling, S. W., Farrar, C. R., & Prime, M.
shop on Model-Based Development of Computer
B. (1998). A summary review of vibration-
Based Systems.
based damage identification methods. The
Shock and Vibration Digest, 30(2), 91105. Mathworks. (1999). Simulink: Dynamic system
doi:10.1177/058310249803000201 simulation for Matlab. Mathworks Inc.
Elouedi, Z., Mellouli, K., & Smets, P. (2004). Mosterman, P. J., & Biswas, G. (1999). Di-
Assessing sensor reliability for multisensor data agnosis of continuous valued systems in tran-
fusion within the transferable belief model. IEEE sient operating regions. IEEE Transactions,
Transactions on Systems, Man, and Cybernetics. Man, and Cybernetics-Part A, 29(6), 554565.
Part B, Cybernetics, 34(1), 782787. doi:10.1109/ doi:10.1109/3468.798059
TSMCB.2003.817056
368
Moustafa, A., Mahadevan, S., Daigle, M., & Rosenberg, R. C., & Karnopp, D. C. (1983).
Biswas, D. (2010). Structural and sensor system Introduction to physical system dynamics.
damage identification using the bond graph ap- McGraw-Hill.
proach. Structural Control and Health Monitor-
Sankararaman, S., & Mahadevan, S. (2011). (in
ing, 17, 178197. doi:10.1002/stc.285
press). Uncertainty quantification in structural
Pacific Earthquake Engineering Research damage diagnosis. Structural Control and Health
(PEER) Center. (2005). Retrieved from http:// Monitoring. doi:10.1002/stc.400
peer.berkeley.edu
Paynter, H. M. (1961). Analysis and design of
engineering systems. MIT Press.
369
370
Chapter 15
Optimal Design of Nonlinear
Viscous Dampers for Protection
of Isolated Bridges
Alexandros A. Taflanidis
University of Notre Dame, USA
Ioannis G. Gidaris
University of Notre Dame, USA
ABSTRACT
A probabilistic framework based on stochastic simulation is presented in this chapter for optimal design
of supplemental dampers for multi - span bridge systems supported on abutments and intermediate
piers through isolation bearings. The bridge model explicitly addresses nonlinear characteristics of the
isolators and the dampers, the dynamic behavior of the abutments, and the effect of pounding between
the neighboring spans to each other as well as to the abutments. A probabilistic framework is used to
address the various sources of structural and excitation uncertainties and characterize the seismic risk
for the bridge. Stochastic simulation is utilized for evaluating this seismic risk and performing the as-
sociated optimization when selecting the most favorable damper characteristics. An illustrative example
is presented that considers the design of nonlinear viscous dampers for protection of a two-span bridge.
INTRODUCTION Wang, Fang, & Zou, 2010). Lead-rubber bearings

or friction pendulum systems are selected for this
Applications of seismic isolation techniques to purpose in order to isolate the bridge deck from
bridges (Figure 1) have gained significant atten- its support, at the abutments and potentially at the
tion over the last decade (Jangid, 2008; Jnsson, locations of intermediate piers. This configura-
Bessason, & Haflidason, 2010; Makris & Zhang, tion provides enhanced capabilities for energy
2004; Perros & Papadimitriou, 2009; Tsopelas, dissipation during earthquake events while also
Constantinou, Okamoto, Fujii, & Ozaki, 1996; accommodating thermal movements during the
life-cycle of operation of the bridge. It is associ-
DOI: 10.4018/978-1-4666-1640-0.ch015 ated though with large displacement for the bridge
Optimal Design of Nonlinear Viscous Dampers for Protection of Isolated Bridges
Figure 1. Two-span base-isolated bridge
deck relative to its supports, especially under near lated bridges (Hwang & Tseng, 2005; Makris &
fault earthquake ground motions (Dicleli, 2006; Zhang, 2004; Ruangrassamee & Kawashima,
Dimitrakopoulos, Makris, & Kappos, 2010; Shen, 2003). The explicit consideration of the hyster-
Tsai, Chang, & Lee, 2004). These motions include etic behavior of the isolators and the highly
frequently a strong, longer period component nonlinear behavior of the dampers in the design
(pulse) that has important implications for flexible, process, as well as the proper modeling of soil-
isolated structures (Bray & Rodriguez-Marek, structure interaction at the foundations of the
2004; Mavroeidis & Papageorgiou, 2003). For bridge are some of the main challenges encoun-
base-isolated bridge systems, such large displace- tered in the design of such dampers. Linearization
ments under strong ground motions may lead to techniques are frequently adopted for modeling
(i) large inelastic deformations and formation of the bridge system (Hwang & Tseng, 2005); this
plastic hinges at the piers and abutments and (ii) simplifies the analysis, but there is great doubt if
pounding of the deck between adjacent spans or to it can accurately predict the combined effect of
intermediate seismic stoppers or to the abutments the non-linear viscous damping, provided by the
supporting the ends of the bridge (Dimitrakopou- dampers, along with the non-linear hysteretic
los, Makris, & Kappos, 2009; Dimitrakopoulos et damping, provided by the isolators, while ap-
al., 2010; Ruangrassamee & Kawashima, 2003). propriately addressing the soil-structure interac-
Such pounding will then lead to high impact tion characteristics and the nonlinearities intro-
stresses and increased shear forces for both the duced by pounding effects. Another challenge is
bridge deck and its supports (abutments and piers). the explicit consideration of the variability of
This overall behavior associated with excessive future excitations and of the properties of the
vibrations will ultimately lead to significant dam- structural system since a significant degree of
ages that affect not only the serviceability but also sensitivity has been reported between these char-
the structural integrity of the bridge system. For acteristics and the overall system performance
controlling such vibrations, application of seismic (Dimitrakopoulos et al., 2009; Perros & Papad-
dampers has been proposed and applied to iso- imitriou, 2009).
371
Figure 2. Schematic of augmented system and seismic risk description
This chapter presents a methodology that earities into the model used at the design stage. A
addresses all aforementioned challenges for the realistic model is also discussed for description of
design of supplemental dampers for seismically near-fault ground motions. This model establishes
isolated, short-spanned bridges. A probabilistic a direct link, in a probabilistic sense, between our
framework is proposed for addressing the various knowledge about the characteristics of the seismic
sources of uncertainty and quantifying the overall hazard in the structural site and future ground mo-
performance. This is established by characterizing tions. An efficient probabilistic sensitivity analysis
the relative plausibility of different properties is also discussed for investigating the influence
of the system and its environment (representing of each of the uncertain model parameters to the
future excitations) by appropriate probability overall seismic risk.
models. Seismic risk is then defined as the ex-
pected value of the system performance over these
models. Stochastic simulation is implemented for PROBABILISTIC DESIGN
evaluation of the multidimensional probabilistic FRAMEWORK
integral representing seismic risk and an efficient
algorithm (Kleinmann, Spall, & Naiman, 1999; Evaluation of seismic risk for isolated bridges
Spall, 1998) is adopted for performing the as- requires adoption of appropriate models for (i)
sociated optimization and selecting the optimal the bridge system itself, (ii) the excitation (ground
damper parameters. This establishes a versatile, motion), and (iii) the system performance (Figure
simulation-based framework for detailed charac- 2). The combination of the first two models pro-
terization of seismic risk that puts no restrictions in vides the structural response. The performance
the complexity of the modeling approach adopted. evaluation model assesses, then, the favorability
Thus, the bridge response is evaluated through of this response based on the selected performance
nonlinear dynamic analysis allowing for direct criteria.
incorporation of all important sources of nonlin-
372
The characteristics of the models for the seis- signer criteria (an example is discussed in the
mically isolated bridge and for future earthquake illustrative application considered later). The
excitations, described in the next two sections are conventions that lower values for h(,) corre-
not known with absolute certainty. Uncertainties spond to better performance, i.e. it is associated
may pertain to (i) the properties of the bridge with risk and ultimately corresponds to a risk
system, for example, related to stiffness or damp- measure, is used herein.
ing characteristics and traffic loads; to (ii) the For addressing the uncertainty in a prob-
variability of future seismic events, i.e., the mo- ability density function (PDF) p(), is assigned to
ment magnitude or the epicentral distance; to (iii) it, quantifying the relative likelihood of different
the predictive relationships about the character- model parameter values. This PDF incorporates
istics of the excitation given a specific seismic our available knowledge about the structural
event, for example duration of strong ground system and its environment into the respective
motion or peak acceleration; or to (iv) parameters knowledge, and should be selected based on this
related to the performance of the system, for knowledge (Jaynes, 2003). In this setting, the
example, thresholds defining fragility of system overall performance, i.e. seismic risk, is described
components. A probability logic approach pro- by the following probabilistic integral that cor-
vides a rational and consistent framework for responds to the expected value of h(,)
quantifying all these uncertainties and explicitly
incorporating them into the system description. C() = h(, ) p()d (1)

In this approach, probability can be interpreted
as a means of describing the incomplete or miss-
ing information (Jaynes, 2003) about the system Different selections for h() lead to different
under consideration and its environment, repre- characterization for seismic risk; for example, if
senting the seismic hazard, through the entire h(,)=Cin(,)+Clif(,), where Cin(,) cor-
life-cycle (Taflanidis & Beck, 2009a). responds to the initial cost and Clif(,) to the
To formalize these ideas denote the vector of additional cost over the lifetime due to repairs or
controllable system parameters, referred to downtime, then risk corresponds to life cycle cost
n
herein as design variables, be , where (Taflanidis & Beck, 2009a), if h(,)=IF(,),
where IF(,) is the indicator function for some
denotes the admissible design space with vol-
event F (one if F has occurred and zero if not),
ume V. These variables are related to the adjust-
then risk corresponds to the probability of unac-
able characteristics of the damper application, for
ceptable performance (Taflanidis & Beck, 2009b).
example viscosity properties or damper size. Let
n The probabilistically-robust stochastic design
, denote the augmented vector of
(Taflanidis & Beck, 2008) is then established
model parameters where represents the space
by selecting the design variables that minimizes
of possible model parameter values. Vector is
seismic risk C()
composed of all the model parameters for the
individual structural system, excitation, and per-
formance evaluation models indicated in Figure
{
* = arg min C() | f c () 0 } (2)
2. The seismic performance of the bridge, for
specific design and model description , is Where arg stands for argument that mini-
characterized by the performance measure mizes and fc() is a vector of deterministic
n xn
h(, ) : + , which ultimately quan- constraints, related, for example, to location or
tifies seismic utility or risk according to the de- space constraints for the dampers. Note that in this
373
formulation, all performance requirements against cases, for example, skewed bridges, for which a
future natural hazards are directly incorporated in complete three dimensional model is required to
the objective function. Finally, the constraints in accurately capture their response under seismic
Equation 2 may be incorporated into the definition excitation. Each span of the bridge is modeled
of admissible design space , which leads to the here as a rigid body. For appropriately address-
simplified expression: ing at the design stage the pounding between
adjacent spans, soil-structure interaction charac-
* = arg min C() (3) teristics and the nonlinear behavior of isolators
and dampers, nonlinear dynamic analysis is
used (Makris & Zhang, 2004; Saadeghvaziri &
For evaluation of Equation 1 and the optimiza- Yazdani-Motlagh, 2008; Zhang, Makris, & Delis,
tion in Equation 3 an approach based on stochastic- 2004). The pounding with the abutment and the
simulation will be discussed in detail later. Along dynamic characteristics of the latter are incorpo-
with the comprehensive risk quantification and rated in the analysis by modeling the abutment
the bridge and excitation models described next as a mass (Saadeghvaziri & Yazdani-Motlagh,
that establishes a versatile, end-to-end simulation 2008) connected to the ground by a spring and
based framework for detailed characterization and a dashpot, with stiffness and damping properties
optimization of seismic risk. This framework puts that are related to the local soil conditions (Zhang
no restrictions in the complexity of the models et al., 2004). The vibration behavior of the isola-
used allowing for incorporation of all important tors, the dampers and the pier is incorporated by
sources of nonlinearities and adoption of advanced appropriate nonlinear models (Ruangrassamee &
models for characterization of the ground motion. Kawashima, 2003; Taflanidis, 2009; Taflanidis &
Thus, it allows an efficient, accurate estimation Beck, 2009a; Zhang et al., 2004). The pounding
of the seismic risk by appropriate selection of the between adjacent spans, or to the abutments, is
characteristics of the stochastic analysis. Next we approximated here as a Hertz contact force with
will discuss in detail the bridge and excitation an additional non-linear damper to account for
models used in this study, and then proceed to energy losses (Muthukumar & DesRoches, 2006).
the computational framework for performing the A schematic of the bridge model with two
optimization in Equation 3. spans, supported by seismic isolators to the abut-
ments and to the intermediate pier is illustrated
in Figure 3. The two spans and abutments are
BRIDGE MODEL distinguished by using the convention right and
left for each of them. The gap between the two
For simplicity of the analysis, we will assume a spans is denoted by xo and the gap between the
two-span (as in Figure 1), straight bridge, whose left or right span and the corresponding abutment
fundamental behavior in the longitudinal direc- by xol or xor, respectively. Let also xp, xsl, xsr, xal,
tion can be adequately characterized with a planar xar, denote, respectively, the displacement relative
(two-dimensional) model. It is noted that the to the ground of the pier, the left and right span
bridge is assumed to be parallel to the direction of the bridge and the left and right abutment. The
of the seismic wave propagation and also that it total mass for the pier, the left and right span of
does not fall in the category of long-span bridges. the bridge and the left and right abutment are
The ideas discussed in the following sections, denoted, respectively, by mp, msl, msr, mal, mar.
related to probabilistic modeling and design, can This total mass includes both the weight of each
be directly extended, though, to more complex component as well as the live loads due to ve-
374
Figure 3. Schematic model for two-span bridge system with seismic isolators and supplemental dampers
hicle traffic. Additionally let Cp, Cal, Car denote, where xg is the ground acceleration; fc, fcl and fcr
respectively, the damping coefficient for the pier are the impact forces due to pounding between
and the left and right abutment. The equation for the two spans or between the spans and the left
the seismically isolated system is then derived by or right, respectively, abutments; fdl and fdr are the
equilibrium conditions as: left and right damper forces; and fial, fipl or fiar, fipr
are the forces of the isolators that support the left
msl xsl + ( fial + fcl + fdl ) + fipl + fc = 0 or right span of the bridge to the abutment and
msr xsr + ( fiar + fcr + fdr ) + fipr fc = 0 the pier, respectively; fp corresponds to the restor-
m p xp + C p x p + fp fipl fipr = m p xg ing force for the pier which is modeled as hyster-
mal xal + C al xal + fal ( fial + fcl + fdl ) = mal xg etic bilinear force with ultimate strength as de-
mar xar + C ar xar + far ( fiar + fcr + fdr ) = mar xg picted in Figure 3. Strength and stiffness
(4) deterioration may be additionally incorporated
375
into this model (Taflanidis & Beck, 2009a). All 0 ; if x sl x sr < x o

fc =
these forces are discussed in more detail next. kcc (x sl x sr x o )c +ccc (xsl xsr ) ; if x sl x sr x o

The hysteretic behavior of each isolator is (8)
modeled by a Bouc-Wen model which is described
by the following basic differential equation where kcc is the contact stiffness, ccc the non-linear
(Taflanidis, 2009): damper coefficient and c the contact exponent,
taken here with the nominal value for Hertz type
yi z = is xb z 2 (is sgn(xb z ) + is )xb (5) of impact, i.e. c=1.5 (Muthukumar & DesRoches,
2006). The damper coefficient may be expressed
in terms of the ratio of relative velocities of the
where xb is the displacement of the isolator, which pounding bodies before and after the contact,
corresponds to the relative displacement of the called coefficient of restitution ecc, as (Muthuku-
span relative to its support; z is a dimensionless mar & DesRoches, 2006):
hysteretic state-variable that is constrained by val-
ues 1; yi is the yield displacement; and is, is, and c
x sl x sr x o
is are dimensionless quantities that characterize ccc = 0.75kcc (1 ecc2 ) (9)
the properties of the hysteretic behavior. Typical vcon
values for these parameters are used here, taken
as is=1, is=0.1, and is=0.9 (Taflanidis, 2009). where vcon is the relative velocity at the initiation
The isolator forces fi may be then described based of contact. The contact stiffness is a function of
on the state-variable z and the relative isolator the elastic properties and the geometry of the
displacement xb. For friction-pendulum isolators colliding bodies. For elastic contact between two
and lead-rubber bearings, these forces are given isotropic spheres with radius R1 and R2 the follow-
respectively, by (Taflanidis, 2009): ing relationship holds (Werner, 1960)
fi = kp x b + N t z (6)
4 R1R2
kc = (10)
3(1 + 2 ) R1 + R2
fi = kp x b + (ke kp )yi z (7)
where i, i=1,2 is a material parameter for the ith

where kp is the post yield stiffness, Nt the average body given by
normal force at the bearing, is the coefficient
of friction, ke the pre yield stiffness. For the left 1 vi2
span xb is given by xb=xsl-xal and xb=xsl-xp for the i = (11)
E i
isolators supporting it to the abutment and the pier,
respectively. For the right span the corresponding
quantities are xb=xsr-xar, and xb=xsr-xp, respectively. with vi and Ei representing its Poissons ratio
The force due to pounding between the adjacent and modulus of elasticity, respectively. To cal-
spans is modelled as a single-sided Hertz contact culate(10) each of the spans can be roughly ap-
force with an additional damper that incorporates proximated as a sphere with radius (Muthukumar
in the analysis the energy dissipated during the & DesRoches, 2006)
contact (Muthukumar & DesRoches, 2006)
3mi
Ri = 3 (12)
4
376
Box 1.
0 ; if x sl x al > x ol
fcl = c (13)
kcl x sl x al + x ol +ccl (xsl xal ) ; if x sl x al x ol

Box 2.
0 ; if x sr x ar < x or
fcr = c (14)
kcr (x sr x ar x or ) +ccr (xsr xar ) ; if x sr x ar x or

ad
where is the material density and mi corresponds fdl = cd sgn(xsl xal ) xsl xal (15)
to the mass of the span. Ri and i are calculated
based on the total mass and material properties
where cd is the damping coefficient and ad an
of each span and then the contact stiffness kcc is
exponent parameter. These adjustable charac-
equal to kc given by Equation 10.
teristics are the controllable damper parameters
The force due to pounding of the left span to
to de selected at the design stage. Note that for
the left abutment is modeled as left-sided contact
ad=1 relationship 15 corresponds to a linear
force in Box 1.and similar modeling holds for the
viscous damper. Maximum forcing capabilities
pounding of the right span to the right abutment
for the damper, related to cost constraints, may
(see Box 2).
be incorporated into the model as a saturation of
The contact parameters kcl, kcr and ccl, ccr are
the damper force to fmax leading to (see Box 4).
defined similarly to the case of contact between
For the damper connecting the right span to
the spans, using the properties (mass, velocity)
the abutment the damper force is similarly ex-
of the colliding bodies of interest. The restitution
pressed in Box 4.
coefficients for collision of the left or right span
to their neighboring abutments are denoted by ecl
and ecr respectively.
EXCITATION MODEL
The damper forces depend on the type of
application. For simplicity only the case of
The analysis and design of any seismic structural
passive viscous dampers is discussed here, but
system needs to be performed considering poten-
the framework presented in the next sections is
tial damaging future ground motions. For base-
directly extendable to the design of other type
isolated bridges this translates to consideration
of dampers as well. For fluid viscous dampers
of near-fault ground motions.
the damper forces are a function of the relative
The last two decades numerous models have
velocity across the end points of the damper. For
been proposed for ground motion modeling
the damper connecting the left span to the cor-
(Atkinson & Silva, 2000; Shinozuka, Deodatis,
responding abutment (Makris & Zhang, 2004)
Zhang, & Papageorgiou, 1999) that incorporate
they may be described in Box 3.
important characteristics of the seismic source
as well as of the earth medium. Other studies
377
Box 3.
fmax sgn(xsl xal ) ad

; if cd xsl xal > fmax
fdl = ad (16)
cd sgn(xsl xal ) xsl xal ; if cd xsl xal
ad
fmax

Box 4.
fmax sgn(xsr xar ) ; if cd xsr xar

ad
> fmax
fdr = ad (17)
cd sgn(xsr xar ) xsr xar ; if cd xsr xar
ad
fmax

have focused on the estimation of the spatial uncertainty. Such a model was proposed recently in
variability of the seismic ground motions which (Taflanidis, Scruggs, & Beck, 2008) and is briefly
is a result of the transmission of the waves from discussed next. According to it the high-frequency
the source through the different earth strata to the and long-period components of the motion are
ground surface (Zerva, 2008). The latter effect independently modeled and then combined to
is considered to be very crucial in the design of form the acceleration time history. It is pointed
elongated structures such as long span bridges. out that the spatial variation of the seismic input
An important factor of the spatial variability is is not been taken into account, because the bridge
the site response effect (i.e. the difference at the systems we focus on are relatively short-spanned.
local soil conditions at two stations), especially
for structures situated in regions with rapidly High Frequency Component
changing local soil conditions and for bridges with
short or medium length spans (Der Kiureghian, The fairly general, point source stochastic
1996). Moreover, other researchers have devel- method (Boore, 2003) is selected for modeling
oped models that address the nonstationarity of the the higher-frequency (>0.10.2 Hz) component
ground motions by accounting the time variation of ground motions. This approach corresponds
of both intensity and frequency content (Conte & to source based stochastic ground motion
Peng, 1997; Rezaeian & Der Kiureghian, 2008). models, developed by considering the physics
An alternative tool for modeling uncertainty of of the fault rupture at the source as well as the
future earthquake loads is the critical excitation propagation of seismic waves through the entire
method in which the ground motion is represented ground medium till the structural site. It is based
as a product of a Fourier series and an envelope on a parametric description of the ground motions
function and in general involves estimating the radiation spectrum A(f;M,r), which is expressed as
excitation producing the maximum response a function of the frequency, f, for specific values
from a class of allowable inputs (Abbas, 2011; of the earthquake magnitude, M, and epicentral
Takewaki, 2002). distance, r. This spectrum consists of many factors
For the proposed probabilistic framework, a which account for the spectral effects from the
complete stochastic model for characterizing the source (source spectrum) as well as propagation
acceleration time-history of near-fault excitations through the earths crust up to the structural site.
is required, that addresses all important sources of The duration of the ground motion is addressed
378
through an envelope function e(t;M,r), which record-based stochastic ground motion could be
again depends on M and r. These frequency and used for modelling the high frequency component
time domain functions, A(f;M,r) and e(t;M,r), of the ground motion. An example of such a model
completely describe the model and their charac- is the one developed recently by Rezaeian & Der
teristics are provided by predictive relationships Kiureghian (2008), which has the advantage that
that relate them directly to the seismic hazard, i.e., additionally addresses spectral nonstationarities.
to M and r. More details on them are provided in
(Boore, 2003; Taflanidis et al., 2008). Particularly, Long Period Pulse
the two-corner point-source model by Atkinson
& Silva (2000) can be selected for the source For describing the pulse characteristic of near-
spectrum because of its equivalence to finite fault fault ground motions, the simple analytical model
models. This equivalence is important because developed by Mavroeidis & Papageorgiou (2003)
of the desire to realistically describe near-fault is selected. This model is based on an empirical
motions and adaptation of a point-source model description of near-fault ground motions and has
might not efficiently address the proximity of the been calibrated using actual near-field ground mo-
site to the source (Mavroeidis & Papageorgiou, tion records from all over the world. According
2003). The spectrum developed by Atkinson & to it, the pulse component of near-fault motions
Silva (2000) has been reported in their studies to is described through the following expression for
efficiently address this characteristic. the ground motion velocity pulse:
The time-history (output) for a specific event
magnitude, M, and source distance, r, is obtained Ap 2 f
V (t ) = 1 + cos p
(t to ) cos 2 fp (t to ) + p if t to p , to + p
according to this model by modulating a white- 2 p
2 fp 2 fp

=0 otherwise
noise sequence Zw=[Zw(it): i=1,2,, NT] by (18)
e(t;M,r) and subsequently by A(f;M,r) through
the following steps: the sequence Zw is multiplied where Ap, fp, p, p, and to describe the signal ampli-
by the envelope function e(t;M,r) and then trans- tude, prevailing frequency, phase angle, oscillatory
formed to the frequency domain; it is normalized character (i.e., number of half cycles), and time
by the square root of the mean square of the ampli- shift to specify the envelopes peak, respectively.
tude spectrum and then multiplied by the radiation Note that all parameters have an unambiguous
spectrum A(f;M,r); finally it is transformed back physical meaning. A number of studies (Alavi
to the time domain to yield the desired accelera- & Krawinkler, 2000; Bray & Rodriguez-Marek,
tion time history. The model parameters consist of 2004; Mavroeidis & Papageorgiou, 2003) have
the seismological parameters, M and r, describing been directed towards developing predictive rela-
the seismic hazard, the white-noise sequence Zw tionships that connect these pulse characteristics
and the predictive relationships for A(f;M,r) and to the seismic hazard of a site. These studies link
e(t;M,r). Figure 4 shows functions A(f;M,r) and the amplitude and frequency of near-fault pulse
e(t;M,r) for different values of M and r. It can be to the moment magnitude and epicentral distance
seen that as the moment magnitude increases the of seismic events, but they also acknowledge that
duration of the envelope function also increases significant uncertainty exists in such relationships.
and the spectral amplitude becomes larger at all This indicates that a probabilistic characterization
frequencies with a shift of dominant frequency is more appropriate. For example, according to
content towards the lower-frequency regime. the study by Bray and Rodriguez-Marek (2004)
Reduction of r primarily contributes to an overall the following relationships hold for the pulse
increase of the spectral amplitude. Alternatively, a frequency fp and the peak ground velocity Av
379
Figure 4. Radiation spectrum and envelope function, for various M, r
ln fp = ln fp + e f (19) correlation between the existence of the pulse and

the proximity of the structural site to the seismic
+e source/fault. For epicentral distances larger than
lnAv = lnA (20)
v v
15-20 km the pulse component according to this
equation is negligible (Bray & Rodriguez-Marek,
where fp and Av are median values (provided 2004).
through a regression analysis), and p and v are For the rest of the pulse parameters, the phase
prediction errors following a Gaussian distribution and the number of half cycles, no clear link to the
with zero mean and large standard deviation. For seismic hazard of the structural site has been yet
rock sites the predictive relationships for the identified. They need to be considered as indepen-
median values are (Bray & Rodriguez-Marek, dent model parameters. The study (Mavroeidis &
2004) Papageorgiou, 2003) provides relevant values for
these parameters by fitting Equation 18 to a suite
ln fp = 8.60 1.32M (21) of recorded near-fault ground motions.
Near-Fault Excitation Model

= 4.46 + 0.34M 0.58ln(r 2 + 7 2 )
lnAv
(22) The stochastic model for near-fault motions is
finally established by combining the above two
whereas the standard deviation for the predic- components through the methodology initially
tion errors ef and ev is estimated as 0.4 and 0.39, developed in (Mavroeidis & Papageorgiou, 2003).
respectively. Note that according to Equation 22 The model parameters consist of the seismological
the pulse amplitude, which is directly related to parameters M and r, the additional parameters for
the peak ground velocity (Mavroeidis & Papageor- the velocity pulse, p, p, the white noise sequence
giou, 2003), varies almost inversely proportional Zw, and the predictive relationships for fp, Av, and
to the epicentral distance. This indicates a strong
380
Figure 5. Sample near-fault ground motion; acceleration and velocity time histories
for the characteristics of A(f;M,r) and e(t;M,r). The coincides with the phase of the time history
following procedure describes the final model: generated in step 2.
6. Finally superimpose the time histories gener-
1. Generate the high-frequency component of ated in steps 2 and step 5.
the acceleration time history by the stochastic
method. Figure 5 illustrate a near-fault ground motion
2. Generate a velocity time history for the near- sample for an earthquake M=6.7, r=5km and pa-
field pulse using Equation 18. The pulse is rameters for the near-fault pulse p=1.7, p=/6.
shifted in time to coincide with the peak of Both the acceleration and velocity time histories
the envelope e(t;M,r). This defines the value of the synthetic ground-motion are presented. The
of the time shift parameter to. Differentiate existence of the near-fault pulse is evident when
the velocity time series to obtain an accel- looking at the velocity time history.
eration time series.
3. Calculate the Fourier transform of the ac-
celeration time histories generated in steps STOCHASTIC ANALYSIS,
1 and 2. OPTIMIZATION, AND SENSITIVITY
4. Subtract the Fourier amplitude of the time
series generated in step 2 from the spectrum Stochastic Analysis
of the series generated in 1.
5. Construct a synthetic acceleration time his- Since the nonlinear models considered for the
tory so that its Fourier amplitude is the one bridge and the excitation are complex and include
calculated in step 4 and its Fourier phase a large number of uncertain model parameters the
381
multi-dimensional integral in Equation 1 defin- selection of importance sampling densities may

ing seismic risk and involved in optimization preliminary focus on them.
in Equation 3 cannot be calculated, or even ac-
curately approximated, analytically. An efficient Optimization for the Design
alternative approach is to estimate the integral by Variables under Uncertainty
stochastic simulation (Taflanidis & Beck, 2008).
Using a finite number, N, of samples of drawn The optimal design choice is finally given by the
from some importance sampling density pis(), an optimization under uncertainty problem (Spall,
estimate for (1) is given by the stochastic analysis: 2003):
1 N p(i ) * = arg min C() (24)

C() = C(, ) =
N
i =1 h(, i ) pis (i )

(23)
The estimate of the objective function for this
where vector i denotes the sample of the uncer- optimization involves an unavoidable estimation
tain parameters used in the ith simulation and error and significant computational cost since N
denotes the entire samples set {i} used in the evaluations of the model response are needed for
evaluation. As N , then C but even for each stochastic analysis. These features that make
finite, large enough N, Equation 23 gives a good this optimization problem challenging. Many nu-
approximation for the integral in Equation 1. The merical techniques and optimization algorithms
importance sampling density pis() may be used have been developed to address such challenges
to improve the efficiency of this estimation. This in design optimization under uncertainty (Royset
is established by focusing on regions of the & Polak, 2004; Ruszczynski & Shapiro, 2003;
space that contribute more to the integrand of the Spall, 2003). Such approaches may involve one
probabilistic integral in Equation 1 (Taflanidis & or more of the following strategies: (i) use of
Beck, 2008), i.e. by selecting a proposal density common random numbers to reduce the relative
that resembles that integrand. If pis()=p() then importance of the estimation error when com-
the evaluation in Equation 23 corresponds to paring two design choices that are close in the
direct Monte Carlo integration. For problems with design space, (ii) application of exterior sampling
large number of model parameters, as the applica- techniques which adopt the same stream of random
tion discussed here, choosing efficient importance numbers throughout all iterations in the optimi-
sampling densities for all components of is zation process, thus transforming the stochastic
challenging and can lead to convergence problems problem into a deterministic one, (iii) simultaneous
for the estimator of Equation 23; thus it is prefer- perturbation stochastic search techniques, which
able to formulate importance sampling densities approximate at each iteration the gradient vector
only for the important components of , i.e. the by performing only two evaluations of the objec-
ones that have biggest influence on the seismic tive function in a random search direction, and
risk, and use q(.)=p(.) for the rest (Taflanidis & (iv) gradient-free algorithms (such as evolution-
Beck, 2008). For seismic risk applications the ary algorithms) which do not require derivative
characteristics of the hazard, especially the mo- information. References (Spall, 2003; Taflanidis
ment magnitude or epicentral distance are gener- & Beck, 2008) provide reviews of appropriate
ally expected to have the strongest impact on the techniques and algorithms for such optimiza-
calculated risk (Taflanidis & Beck, 2009a), so tion problems. The very efficient Simultaneous
Perturbation Stochastic Approximation (SPSA)
382
with Common Random Numbers (CRN) will be where k is the sample set used at iteration k and
employed in this study and is briefly reviewed next. n
k is a vector of mutually independent
random variables that defines the random direction
Simultaneous Perturbation of simultaneous perturbation for k and that satis-
Stochastic Approximation fies the statistical properties given in (Spall, 2003).
A symmetric Bernoulli 1 distribution is typi-
Simultaneous-perturbation stochastic approxi- cally chosen for the components of k. The selec-
mation (SPSA) (Kleinmann et al., 1999; Spall, tion for the sequences {ck} and {k} is discussed
2003) is an efficient stochastic search method. in detail in (Kleinmann et al., 1999). A choice that
It is based on the observation that one properly guarantees asymptotic convergence to * is
chosen simultaneous random perturbation in all k=/(k+w) and ck=c1/k, where 4>0,
components of provides as much information 22>1, with w,>0 and 0 < 1 . This
for optimization purposes in the long run as a full selection leads to a rate of convergence that as-
set of one at a time changes of each component. ymptotically approaches k /2 when CRN is used
Thus, it uses only two evaluations of the objective (Kleinmann et al., 1999). The asymptotically
function, in a direction randomly chosen at each optimal choice for is, thus, 1. In applications
iteration, to form an approximation to the gradi- where efficiency using a small number of itera-
ent vector. It also uses the notion of stochastic tions is sought after, use of smaller values for
approximation (Kushner & Yin, 2003) which are suggested in (Spall, 2003). For complex design
can significantly improve the computational ef- optimizations, where the computational cost for
ficiency of stochastic search applications (Spall, each iteration of the algorithm is high, the latter
2003). The latter approximation is performed by suggestion should be adopted. Note that the same
establishing (through proper recursive formulas) set of random numbers is selected for the two
an equivalent averaging across the iterations of the estimates of the objective function at each itera-
optimization algorithm, instead of getting higher tion used to calculate the gradient in Equation 26.
accuracy estimates for the objective function at This reduces the variance of the difference of
each iteration, that is averaging over one single these estimates thus creating a consistent estima-
iteration. tion error (Glasserman & Yao, 1992; Taflanidis
The implementation of SPSA takes the itera- & Beck, 2008) and improving the efficiency of
tive form: the estimate in Equation 26 (reduces relative
importance of estimation error). Across the itera-
k +1 = k k g k ( k , k ) (25) tions of the optimization algorithm, though, this
sample set is not the same; thus there is no de-
where 1 is the chosen point to initiate the pendence of the optimal solution on it (i.e. the
algorithm and the jth component for the Common approach here does not correspond to an exterior
Random Number (CRN) simultaneous perturba- sampling technique).
tion approximation to the gradient vector in the Regarding the rest of the parameters for the
kth iteration, gk(k,k), is given by: sequences {ck} and {k}: w is typically set to 10%
of the number of iterations selected for the algo-
C ( k + ck k , k ) C ( k ck k , k ) rithm and the initial step c1 is chosen close to
gk, j = the standard deviation of the prediction error for
2ck , j
the stochastic estimate in Equation 23. The value
(26)
of can be determined based on the estimate of
383
Box 5.
1
(i | ) = ( | )dx i
= h(, ) p()dx i
p(i ) h(, ) p(x i | i )dx i (28)

h(, ) p()d
g1 and the desired step size for the first iteration. dicate greater importance of in affecting the
Some initial trials are generally needed in order system performance, since they ultimately cor-
to make a good selection for , especially when respond to higher values for h(,). More impor-
little prior insight is available for the sensitivity tantly, though, this idea can be implemented to
of the objective function to each of the design each specific model parameter i (or even to group
variables. Convergence of the iterative process is of them), by looking at the marginal distribution
judged based on the value k +1 k in the last (i|). This distribution is given by (see Box 5).
few steps, for an appropriate selected vector norm. where xi corresponds to the rest of the components
Blocking rules can also be applied in order to of , excluding i. Comparison between this
avoid potential divergence of the algorithm, es- marginal distribution (i) and the prior distribu-
pecially in the first iterations (Spall, 2003). tion p(i) expresses the probabilistic sensitivity
of the seismic risk with respect to i. Uncertainty
Probabilistic Sensitivity Analysis in all other model parameters and stochastic ex-
for Model Parameters citation is explicitly considered by appropriate
integration of the joint probability distribution
An extension of the optimization framework may (|) to calculate the marginal probability dis-
be established to additionally quantify the impor- tribution (i|).
tance of each of the model parameters in affecting A quantitative metric to characterize this sen-
the overall risk. Foundation of this methodology sitivity is the relative information entropy, which
is the definition of an auxiliary probability density is a measure of the difference between probability
function that is proportional to the integrand of the distributions (i|) and p(i) (Jaynes, 2003) over
seismic risk integral for each design configuration the entire space of possible model parameter values
h(, ) p() ( | )
D ((i | ) || p(i )) = i d
( | ) = h(, ) p() (27) ( | )log
i
p(i ) i

h(, ) p()d
(29)
where denotes proportionality and the de-
The importance of each parameter (or sets of
nominator in (|) is simply a normalization
them) is then directly investigated by comparison
constant that will not be explicitly needed. The
of the relative information entropy value for each
sensitivity analysis is established by comparing
of them; larger values for the relative entropy
this auxiliary distribution (|) and the prior
indicate bigger importance.
probability model p(); based on the definition
An analytical expression, though, is not read-
of (|) in Equation 27 such a comparison pro-
ily available for the marginal distribution (i|)
vide information for h(,). Thus, bigger discrep-
since evaluation of Equation 28 is challenging.
ancies between distributions (|) and p() in-
An alternative stochastic-sampling approach is
384
Box 6.
( i ik )2

1 n
2 li2
(i | ) = e ; li = 1.06 n1/5 si (30)
nli 2 k =1
discussed next. This approach is based on exploit- For establishing better consistency in the
ing the information in samples of from the joint relative information entropy calculation p(i) may
distribution (|). Such samples may be obtained also be approximated by Equation 30 based on
by any appropriate stochastic sampling algorithm, samples, even when an analytical expression is
for example by the accept-reject method (Robert available for it. This way, any type of error intro-
& Casella, 2004). Furthermore this task may be duced by the Kernel density estimation is similar
integrated within the stochastic analysis: each of for both of the densities compared. The approxi-
the samples i used for the estimation in Equation mation for the relative information entropy is then
23 can be selected as a candidate sample c in the (Beirlant et al., 1997)
context of the Accept-Reject algorithm. The re-
quired samples from (|) are obtained at a small bi , u
( | )
additional computational effort over the risk as- D ((i | ) || p(i )) (i | )log i d
bi , l
p(i ) i
sessment task, since they require no new simula-
(31)
tions for the bridge model response. Projection,
now, of the samples from (|) to the space of
where the last scalar integral can be numerically
each of the model parameters provides samples
evaluated, for example using the trapezoidal rule,
for the marginal distributions (i|) for each of
and [bi,l, bi,u] is the region for which samples for
them separately. Thus using the same sample set
(i| ) and q(i) are available. This approach
this approach provides simultaneously informa-
ultimately leads to an efficient sampling-based
tion for all model parameters. For scalar quantities,
methodology for calculating the relative informa-
as in the case of interest here, the relative entropy
tion entropy for different parameters, which can
may be efficiently calculated (Beirlant, Dudewicz,
be performed concurrently with the risk assess-
Gyorfi, & Van der Meulen, 1997; Mokkadem,
ment, exploiting the readily available system
1989) by establishing an analytically approxima-
model evaluations for minimizing the computa-
tion for (i|), based on the available samples,
tional burden. Comparing the values for this en-
through Kernel density estimation. This estimate
tropy between the various model parameters leads
will not necessarily have high accuracy, but it can
then to a direct identification of the importance
still provide an adequate approximation for com-
of each of them in affecting risk. Parameters with
puting the information entropy integral. A Gauss-
higher value for the relative entropy will have
ian Kernel density estimator may be used for this
greater importance. Furthermore direct compari-
purpose (Martinez & Martinez, 2002). Using the
son of samples from the distributions (i| ) and
n available samples for i, with ik denoting the
p(i) could provide additional insight about what
k-th sample and si the standard deviation for these
specific values for each parameter contribute more
samples, the approximation for (i|) would be
to the risk (Taflanidis & Beck, 2009c).
(see Box 6).
385
ILLUSTRATIVE EXAMPLE to period of 0.079 sec for the left and 0.089 sec
for the right abutment. Both damping coefficients
Models and Uncertainty Description Cal and Car are selected based on modal damping
assumption. The damping ratios for each abutment
The design of nonlinear viscous dampers for a al, ar are treated as correlated uncertain variables
two-span seismically isolated concrete bridge is following a log-normal distribution with median
considered as illustrative example. The model 8%, coefficient of variation 20% and correlation
described previously is adopted for the bridge and coefficient 50%. The high correlation in the abut-
probability distributions are established, based ment model parameter characteristics is attributed
on our available knowledge, for all parameters to the common soil properties.
that are considered as uncertain. All probability For the left and right span of the bridge the
models are described here in terms of the mean self-weight of the deck is taken as 1000 and 1200
(or median) value, characterizing most probable ton respectively. Vehicle traffic is modeled as
model, and the coefficient of variation, charac- additional loads mtl and mtr for the left and right
terizing the potential spread for each parameter. span, respectively, that follow independent expo-
For variables whose values are expected to be nential distributions with mean value 20 ton. The
correlated, such dependency is expressed through isolators connecting each span to its supports are
their correlation coefficient. lead-rubber bearings modeled by Equation 7. All
The length of the span of the bridges is equal isolators have same properties; post-yield stiffness
to 30 and 34 m, respectively. The mass of the pier ke=3.0 kN/mm, pre-yield stiffness kp=30.0 kN/mm,
is taken as mp=100 ton. For the pier restoring force and yield displacement yi=2.5cm. These choices
fp as illustrated in Figure 3, the initial stiffness kp, correspond to a natural period of 2.57 sec and 2.81
post-yield stiffness coefficient p, over-strength sec for the right and left span, respectively, based
factor p and yield displacement yp, have mean val- on the post-yield stiffness and no vehicle traffic.
ues 70 k/mm, 10%, 30% and 0.04 m, respectively. The respective gap dimensions xo, xol, xor, whose
All these parameters are treated as independent potential variability is influenced by common
Gaussian variables with coefficient of variation weather conditions, are modeled as correlated
10%. For the nominal, i.e. most probable, model log-normal variables with median 10 cm, coef-
this assumption corresponds to period of 0.237 ficient of variation 20%, and large correlation
sec based on the initial stiffness. The damping coefficient, 70%.
coefficient Cp is selected based on modal damping The contact stiffnesses kcc, kcl, kcr for the Hertz
assumption, using the initial period of the pier. The impact forces between the spans or between each
damping ratio p is treated as uncertain variable span and the neighboring abutment are taken as
following a log-normal distribution with median independent log-normal variables with median
3% and coefficient of variation 25%. value given by 10 and large coefficient of variation
The mass of the left and right abutments are 40%, again reflecting our limited knowledge. For
taken, respectively, as mal=400 ton, mar=500 ton. calculating the median value of kc, Poissons ratio
For the right and left abutment restoring forces far, is taken as 0.15, the modulus of elasticity as 30
fal the stiffnesss kar and kal, respectively, are mod- GPa and the density 2.4 ton/m3. The equivalent
elled as correlated Gaussian variables with mean sphere radius 12 for each span is calculated by
value 2500 kN/mm (Saadeghvaziri & Yazdani- considering their total mass, including vehicle traf-
Motlagh, 2008), coefficient of variation 15%, and fic. The coefficients for restitution for the energy
correlation coefficient 50%. For the nominal, i.e. dissipated during contact for each span are mod-
most probable, model this assumption corresponds eled as independent, truncated in [0 1] Gaussian
386
variables with mean value 0.6 and coefficient of moment magnitude and the epicentral distance,
variation 20%. also uncertain parameters, through the conditional
The time-history for future near-fault ground median values given by Equations 21 and 22.
motions is described by the excitation model pre- The second set corresponds to a more abstract
sented earlier. The details for the functions A(f;M,r) representation but is independent of any other
and e(t;M,r) are the same as in (Taflanidis et al., model parameters. Throughout the discussion in
2008). The uncertainty in moment magnitude for this paper both representations will be utilized,
seismic events, M, is modeled by the Gutenberg- according to convenience, but the risk integral
Richter relationship truncated to the interval [Mmin, was actually formulated with respect to ef and ev.
Mmax]=[6, 8]. Smaller than Mmin earthquakes are The probability models for the rest of the pulse
not expected to have significant impact on the model parameters, the number of half cycles and
base isolated structure [and also do not frequently phase, are chosen, respectively, as truncated in [1,
contain a near-fault pulse component (Bray & ) Gaussian with mean value 2 and coefficient
Rodriguez-Marek, 2004)] whereas it is assumed of variation 15%, and uniform in the range [-/2,
that the regional faults cannot produce seismic /2]. These probability models are based on the
events with larger than 8 magnitude. This choice values reported in (Mavroeidis & Papageorgiou,
leads to a truncated exponential PDF (Kramer, 2003) when tuning the analytical relationship in
2003): Equation 18 to a wide range of recorded near-fault
ground motions.
bM exp(bM M ) The uncertain model parameter vector in this
p(M ) =
exp(bM M min ) exp(bM M max ) design problem consists of the bridge model pa-
(32) rameters s=[kp p p yp p kal kar al ar mtl mtr xo, xol
xor ecc ecl ecr kcc kcl kcr] the seismological parameter
The selection for the regional seismicity fac- g=[M r], the additional parameters for the near-
tor is b=0.9loge(10), corresponding to fairly fault pulse p=[ev ef p vp] (with the former two
significant seismic activity. For the uncertainty related to Av and fp) and the white-noise sequence,
in the event location, the epicentral distance, r, Zw, so =[s g p Zw]T. Table 1 reviews these
for the earthquake events is assumed to be a log- parameters along with their probability models.
normal variable with median rm=10 km and large
coefficient of variation 50%. For the near-field Damper Configuration and
pulse, the pulse frequency fp and the peak ground Performance Characteristics
velocity Av are selected according to the probabi-
listic models for the characteristics of near-field Nonlinear viscous dampers are applied to the con-
pulses in rock sites given earlier by Equations nection of each of the two spans to its correspond-
19 and 20. This is equivalent to fp and Av being ing abutment, as illustrated in Figures 1 and 3.
log-normal variables with median value the ones They are modeled by Equation 16 with maximum
in Equations 21 and 22 and coefficient of varia- force capability for each damper, representing cost
tion 0.4 and 0.39, respectively. Note that either constraints (Taflanidis & Beck, 2009a), selected
fp and Av, or ef and ev can be used as the uncertain as 4000 kN. Coefficients ad and cd correspond to
parameters addressing the uncertainty in the pulse the design variables for the problem and for cost
frequency and ground velocity amplitude; the first reduction (bulk ordering) are chosen the same
set corresponds to a representation with an unam- for all dampers. The initial design space for each
biguous physical meaning but is correlated to the of them is defined as ad[0.3,2] and cd[0.1, 30]
MN (sec/m)ad.
387
Table 1. Probability models for uncertain model parameters. In this table c.o.v=coefficient of variation,
= correlation coefficient, = mean value
Gaussian Correlated log-normal Log-normal

al
kp =70 kN/mm median = 8%, r Median = 10km
ar
c.o.v = 10% c.o.v = 20%, = 50% c.o.v = 50%
Log-normal Correlated Gaussian Truncated exponential in [6 8] with

kal M
p median = 3% =2500 kN/mm parameter 0.9loge(10)
kar
c.o.v = 25% c.o.v = 15%, = 50% vp Uniform in [0, ]
Gaussian Gaussian
p xo Correlated log-normal p
= 0.3 c.o.v = 10% = 0.1 c.o.v = 20%
xol median = 10 cm
Gaussian xor c.o.v = 15%, = 70% Gaussian
yp p
=0.03m c.o.v=10% =2.0 c.o.v=15%
kcc Independent log-normal Lognormal
mtl Independent Exponential
kcl median given by 10 fp Median Equation 21
mtr =20 ton
kcr c.o.v = 40%, c.o.v = 40%
Correlated Gaussian ecc Independent Gaussian truncated in Log-normal
kal
=2500 kN/mm ecl [0 1] Av Median Equation 22
kar
c.o.v = 15%, 50% ecr = 0.6 c.o.v = 15% c.o.v = 39%
The optimization problem is defined by Equa- each of these quantities, i.e. the probability that
tion 24, in which the objective function is the the response will exceed some acceptable perfor-
seismic risk given by Equation 23. The admis- mance , is used for describing the system per-
sible design space is defined in Box 7.where fdl formance. These fragilities are assumed to have
and fdr correspond to the damper force for the left a lognormal distribution with median and coef-
and right abutment, respectively. ficient of variation . The probabilistic description
A simplified design problem is also considered for the fragility incorporates into the stochastic
where ad is set to 1, corresponding to a linear analysis model prediction errors (Goulet et al.,
viscous damper. This will allow for investiga- 2007; Taflanidis & Beck, 2010). The seismic-risk
tion of the performance improvement gained by occurrence measure is the average of these fra-
considering nonlinear, rather than linear, damper gilities, see Box 8.where g is the standard Gauss-
implementation. ian cumulative distribution function. The charac-
The bridge performance measure assumed in teristics for the median of the fragility curves are
this study addresses potential seismic damages p=2200 kN for the pier shear, z=6 mm for the
for all components of the bridge: the pier, the abutment displacement and v=10 cm/sec for the
abutments, and the deck. The failure criteria used impact velocity, whereas the coefficient of varia-
are: (i) the maximum pier shear Vp, associated tion is set for all of them equal to p=z=v=0.4.
with yielding and inelastic deformations for the Seismic risk is finally given by the integral in
pier, (ii) the maximum displacement for the left Equation 1 which in this case corresponds to
and right abutment zr and zl, respectively, associ- probability of unacceptable performance for the
ated with permanent deformations for the ground, base-isolated bridge. The maximum value for
and (iii) the maximum velocity for impact between C() is one while the minimum value for C(),
the two spans vo or between each of the spans and corresponding to best possible performance, is
the left or right abutment vl and vr, respectively, zero.
associated with the damages that occur during
pounding (Davis, 1992). The fragility related to
388
Box 7.
{ a
: cdi [0.1, 30]MN (sec / m) ; adi [0.3, 2]; f dl , f dr 4000 KN
d
} (33)
Box 8.
1 ln(V ) ln( ) ln( z ) ln( ) ln( z ) ln( )

p p z z
h(, ) = [ g + g l
+ g
r

6 p z z (34)
ln(v ) ln( ) ln(v ) ln( ) ln(v ) ln( )
+ g o v
+ g
l v
+ g
r v
]
v
v v
Optimal Design design configuration for both nonlinear and linear

damper implementation, the overall objective
The bridge model, described by system of Equa- function C() as well as the expected fragility
tions 4, is created and analyzed in SIMULINK for each of the components. The results illustrate
(Klee, 2007). The computations are performed that the addition of the viscous dampers leads to
for this study on the Persephone cluster at the significant improvement of fragility of the bridge;
HIgh Performance system Analysis and Design there is a big difference between the optimal
(HIPAD) laboratory at the University of Notre C(*) and the uncontrolled performance C(0).
Dame (www.nd.edu/~hipad). The cluster is com- All six components contributing to the overall
posed of 42 nodes, each having eight 2.56GHz fragility are characterized by a considerable
Nehalem cores. All stochastic simulations are reduction, with the maximum pounding veloci-
performed in parallel mode, taking advantage of ties having by far the largest one. This illustrates
the multi-core capacities of the high-performance that the viscous dampers can significantly reduce
computing cluster. the undesirable collisions between the different
SPSA is used for the design optimization spans, which can have devastating effects for the
with selection of N=2000 for the estimate of serviceability of the bridge, while simultaneously
Equation 23 and importance sampling densities efficiently controlling other modes of failure for
established only for the moment magnitude, M, the bridge, as the pier shear or the abutment dis-
which is anticipated to be the most influential placements. It should be also pointed out that the
parameter within the ones included in vector , optimal linear damper configuration provides still
based on results from previous studies on the risk a significant improvement over the uncontrolled
assessment of flexible structures under near-fault bridge performance. Implementation, though, of
excitations (Taflanidis & Jia, 2011). A truncated nonlinear dampers provides a further reduction
in [6 8] Gaussian with mean 6.8 and standard in the seismic risk for the bridge, especially with
deviation 0.6 is selected for M. respect to the pier and abutment failure criteria.
Cumulative results from the optimization are The exponent coefficient for the dampers under
presented in Table 2 which includes the optimal optimal design is 0.66, which corresponds to a
389
Table 2. Optimization results; units for cd are MN (sec/m)ad
Case * C(*) Probabilities of failure for individual components

Vp zl zr vo vl vr
Nonlinear cd 9.7 0.029 0.047 0.034 0.055 0.006 0.0205 0.0093
Damper
ad 0.66
Linear Damper cd 19.5 0.034 0.059 0.037 0.071 0.006 0.0208 0.0102
ad 1.00
No Damper cd 0 0.256 0.284 0.179 0.247 0.253 0.332 0.244
ad -
Table 3. Sensitivity analysis results (normalised entropy) for total risk
No dampers (maximum relative entropy Dm=0.565)

M r ev ef v Av fp mtl mtr xol xor xo kcl
0.410 0.270 0.243 0.055 0.010 0.001 1.000 0.653 0.007 0.028 0.013 0.016 0.011 0.011
kcr kc ecl ecr ecc kal kar al ar p kp Ap p yp
0.007 0.006 0.005 0.004 0.002 0.003 0.002 0.003 0.003 0.001 0.003 0.005 0.004 0.005
Optimal nonlinear dampers (maximum relative entropy Dm=1.08)

M r ev ef v Av fp mtl mtr xol xor xo kcl
0.089 0.402 0.491 0.086 0.012 0.009 1.000 0.016 0.009 0.026 0.003 0.004 0.002 0.005
kcr kc ecl ecr ecc kal kar al ar p kp Ap p yp
0.006 0.007 0.003 0.001 0.001 0.031 0.026 0.003 0.002 0.005 0.004 0.003 0.004 0.002
significant degree of nonlinearity. This illustrates The results in Table 3 illustrate that the pulse
the importance of a design methodology that can amplitude Av is the most influential parameter in
efficiently address damper nonlinearities, so that affecting the overall seismic risk. The uncer-
a truly optimal design is identified. tainty in the pulse amplitude, ev, the moment
magnitude, M, and the epicentral distance, r, have
Stochastic Sensitivity Analysis also a significant impact. As pointed out earlier
Av is actually influenced by all three previous ones,
Results for the sensitivity analysis are finally so the higher importance for it is anticipated.
discussed. Table 3 includes the maximum relative Similar remark holds for fp, which is shown, though,
entropy Dm and the normalised entropy for each to have considerable influence in the seismic risk
parameter for the uncontrolled or optimal damper only for the bridge without the dampers. Since
configuration. The normalised entropy, Dn, is de- both Av and fp are directly related to the impact of
fined by scaling the entropy for each parameter by the near-fault pulse on the structure (amplitude
the maximum, over all parameters, entropy. This of pulse and unison conditions, respectively),
is an explicit measure of the relative importance these results, especially the high importance of
of each parameter to the seismic risk, with values Av and ev, indicate that seismic risk does indeed
closer to unity indicating higher importance. have a big correlation to the near-fault character-
390
istics of the excitation for base-isolated bridge. only importance for the system with optimized
All other excitation and structural parameters have viscous dampers. This is an interesting outcome
negligible importance in affecting seismic risk. and can be attributed to resonance characteristics;
Only the gap sizes, contact stiffnesses, and live the rigid type of connection to the abutments
loads from the trucks seem to have some, small, introduced through the dampers influences the
significance. vibration properties of the isolated-bridge and
The sensitivity pattern does change between thus alters the frequencies of the near fault-pulse
the no-damper and optimal-damper-configuration that have higher impact on it.
cases. This is another important outcome of this Sensitivity analysis results for individual com-
study; it shows that addition of the dampers not ponents of the total performance are presented in
only alters the seismic risk but also influences the Table 4. In particular the maximum pier shear Vp,
risk factors that contribute more to this risk. Since the maximum displacement for right abutment zr,
addition of the dampers provides an overall reduc- and the maximum velocity for impact between the
tion of seismic pounding occurrences (identified left span and abutment vl are reported. Only the
from the results of Table 2) the characteristics interesting model parameters are included in this
that have a stronger connection to this pound- table. For obtaining these results the risk occur-
ing are impacted more. It is interesting that the rence measure in Equation 34 is substituted by the
second most important model parameter for the fragility of the corresponding component. Since
no-damper case, the pulse frequency, has small this fragility has been already computed when
Table 4. Sensitivity analysis results for different response quantities
No Dampers Optimal Nonlinear Dampers

V zr vl Vp zr vl
Dm 0.227 0.701 0.494 1.179 0.672 2.740
Dn M 0.48 0.404 0.551 0.144 0.071 0.125
r 0.30 0.251 0.214 0.372 0.540 0.245
ev 0.19 0.246 0.176 0.407 0.453 0.525
ef 0.15 0.042 0.080 0.034 0.155 0.053
0.01 0.007 0.005 0.012 0.010 0.006
0.03 0.001 0.007 0.004 0.008 0.021
Av 1.00 1.000 1.000 1.000 1.000 1.000
fp 0.81 0.646 0.919 0.039 0.022 0.070
mtl 0.031 0.007 0.010 0.004 0.009 0.003
mtr 0.073 0.023 0.016 0.012 0.011 0.005
kp 0.02 0.002 0.003 0.006 0.004 0.001
xol 0.011 0.011 0.024 0.004 0.005 0.008
xor 0.012 0.016 0.028 0.002 0.006 0.018
xo 0.04 0.011 0.021 0.001 0.002 0.010
kcl 0.027 0.009 0.007 0.011 0.007 0.001
kcr 0.040 0.005 0.008 0.005 0.006 0.002
kc 0.09 0.004 0.006 0.005 0.010 0.003
391
calculating the initial risk occurrence measure, descriptions for the seismic hazard, exploiting an
no new simulations are needed, and the results end-to-end simulation-based approach.
in Table 4 are obtained at a small additional
computational cost (requiring only approxima-
tion of the densities through Kernel estimation CONCLUSION
and calculation of the relative entropy integral).
These results demonstrate some significant vari- The design of supplemental dampers for seismic
ability, indicating that there is a big dependence risk reduction of isolated multi-span bridges was
of the sensitivity analysis and the importance of the focus of this Chapter. The basis of the sug-
the various risk factors, on the exact performance gested approach is a probabilistic framework that
quantification chosen. Especially interesting is explicitly addresses all sources of uncertainty,
the fact that for the bridge with the dampers, the related either to future excitations or to the struc-
epicentral distance and the pulse amplitude are tural configuration, by appropriate probability
the only important risk factors for all different models. In this setting, seismic risk is expressed
quantifications. by a multidimensional integral, corresponding to
the expected value of the risk occurrence measure
over the space of the uncertain model parameters.
FUTURE RESEARCH DIRECTIONS Through appropriate definition of the risk occur-
rence measure this approach facilitates diverse
The design of structural systems for seismic risk risk quantifications. Stochastic simulation was
mitigation requires explicit consideration of the suggested for evaluation of the multidimensional
uncertainties for the characteristics of future ex- integral describing risk and an efficient algorithm
citations as well as for the system properties. The was discussed for performing the associated
constant advances in computer and computational design optimization and selecting the optimal
science, especially the widespread implementa- parameters for the damper implementation. An
tion of distributed computing, have created a new efficient sampling-based probabilistic importance
era for computer simulation and it is currently analysis was also presented, based on information
acknowledged that Simulation-Based Engineer- entropy principles, for investigating the influence
ing Science constitutes a critical new paradigm of each of the model parameters on the overall
for uncertainty quantification and propagation seismic risk.
and is providing great new potentials for detailed Due to fact that the framework is based on
modeling and solution of problems that were until stochastic simulation, consideration of complex
recently considered computationally intractable. nonlinear models for the bridge system and the
Future research efforts need to exploit these char- excitation was feasible. The adopted bridge model
acteristics and focus on modeling approaches for explicitly addressed nonlinear characteristics of
quantification or seismic risk and computational the isolators and the dampers, the dynamic behav-
frameworks for estimation of this risk that will ior of the abutments and the effect of pounding
explicitly address all important characteristics of between the neighboring spans to each other as well
the built systems and its environment, with no need to the abutments. A realistic probabilistic model for
to establish any type of approximations for com- future near-fault excitations was also considered.
putational simplicity. This will only be established An illustrative example was presented that con-
through development of high fidelity numerical sidered the design of nonlinear viscous dampers
models for the dynamic behavior of structural for protection of a two-span bridge. The fragility
systems and adoption of complex probabilistic of the bridge system related to seismic pounding
392
but also to failure modes associated with the pier Beirlant, J., Dudewicz, E. J., Gyorfi, L., & Van
shear and the abutment deformations were adopted der Meulen, E. C. (1997). Nonparametric entropy
as the risk occurrence measure, with seismic risk estimation: An overview. International journal
ultimately corresponding to the probability of of Mathematical and Statistical Sciences, 6(1),
unacceptable performance. The addition of the 17-40.
dampers was shown to provide considerable risk
Boore, D. M. (2003). Simulation of ground mo-
reduction, especially with respect to the vulner-
tion using the stochastic method. Pure and Ap-
abilities associated with seismic pounding. Results
plied Geophysics, 160, 635676. doi:10.1007/
from the sensitivity analysis demonstrated that the
PL00012553
excitation properties, especially the amplitude of
the pulse component of the ground motion, have Bray, J. D., & Rodriguez-Marek, A. (2004).
the highest importance in affecting seismic risk Characterization of forward-directivity ground
and that the inclusion of the dampers did alter these motions in the near-fault region. Soil Dynam-
sensitivity characteristics to a significant degree. ics and Earthquake Engineering, 24, 815828.
In closing, it is noted that the proposed method- doi:10.1016/j.soildyn.2004.05.001
ology offers also significant practical advantages
Conte, J. P., & Peng, B. F. (1997). Fully nonstation-
in real seismic design, since the optimal parameters
ary analytical earthquake ground-motion model.
of the viscous dampers can be efficiently esti-
Journal of Engineering Mechanics, 123, 1534.
mated through a probabilistic framework which
doi:10.1061/(ASCE)0733-9399(1997)123:1(15)
incorporates the various sources of uncertainty, as
well as, the various nonlinearities related with the Davis, R. O. (1992). Pounding of buildings mod-
different bridge components. It is stressed that by eled by an impact oscillator. Earthquake Engi-
considering the nonlinear damper implementation, neering & Structural Dynamics, 21, 253274.
significant performance improvement is achieved doi:10.1002/eqe.4290210305
which in general results to more economic design.
Der Kiureghian, A. (1996). A coherency
model for spatially varying ground motions.
Earthquake Engineering & Structural Dynam-
REFERENCES
ics, 25(1), 99111. doi:10.1002/(SICI)1096-
Abbas, M. (2011). Damage-based design earth- 9845(199601)25:1<99::AID-EQE540>3.0.CO;2-
quake loads for single-degree-of-freedom inelastic C
structures. Journal of Structural Engineering, Dicleli, M. (2006). Performance of seismic-
137(3), 456467. doi:10.1061/(ASCE)ST.1943- isolated bridges in relation to near-fault ground-
541X.0000074 motion and isolator characteristics. Earthquake
Alavi, B., & Krawinkler, H. (2000, January Spectra, 22(4), 887907. doi:10.1193/1.2359715
30-February 4). Consideration of near-fault Dimitrakopoulos, E., Makris, N., & Kappos, A. J.
ground motion effects in seismic design. Paper (2009). Dimensional analysis of the earthquake-
presented at the 12th World Conference on Earth- induced pounding between adjacent structures.
quake Engineering, Auckland, New Zealand. Earthquake Engineering & Structural Dynamics,
Atkinson, G. M., & Silva, W. (2000). Stochastic 38(7), 867886. doi:10.1002/eqe.872
modeling of California ground motions. Bulletin
of the Seismological Society of America, 90(2),
255274. doi:10.1785/0119990064
393
Dimitrakopoulos, E., Makris, N., & Kappos, Kleinmann, N. L., Spall, J. C., & Naiman, D.
A. J. (2010). Dimensional analysis of the earth- C. (1999). Simulation-based optimization with
quake response of a pounding oscillator. Journal stochastic approximation using common ran-
of Engineering Mechanics, 136(3), 299310. dom numbers. Management Science, 45(11),
doi:10.1061/(ASCE)0733-9399(2010)136:3(299) 15701578. doi:10.1287/mnsc.45.11.1570
Glasserman, P., & Yao, D. D. (1992). Some Kramer, S. L. (2003). Geotechnical earthquake
guidelines and guarantees for common random engineering. Prentice Hall.
numbers. Management Science, 38(6), 884908.
Kushner, H. J., & Yin, G. G. (2003). Stochastic
doi:10.1287/mnsc.38.6.884
approximation and recursive algorithms and ap-
Goulet, C. A., Haselton, C. B., Mitrani-Reiser, plications. New York, NY: Springer Verlag.
J., Beck, J. L., Deierlein, G., & Porter, K. A.
Makris, N., & Zhang, J. (2004). Seismic response
(2007). Evaluation of the seismic performance
analysis of a highway overcrossing equipped with
of code-conforming reinforced-concrete frame
elastomeric bearings and fluid dampers. Journal
building-From seismic hazard to collapse safety
of Structural Engineering, 130(6), 830845.
and economic losses. Earthquake Engineering
doi:10.1061/(ASCE)0733-9445(2004)130:6(830)
& Structural Dynamics, 36(13), 19731997.
doi:10.1002/eqe.694 Martinez, W. L., & Martinez, A. R. (2002). Com-
putational Statistics Handbook with MATLAB.
Hwang, J.-S., & Tseng, Y.-S. (2005). Design
Boca Raton, FL: Chapman & Hall/CRC.
formulations for supplemental viscous dampers
to highway bridges. Earthquake Engineering Mavroeidis, G. P., & Papageorgiou, A. P.
& Structural Dynamics, 34(13), 16271642. (2003). A mathematical representation of near-
doi:10.1002/eqe.508 fault ground motions. Bulletin of the Seismo-
logical Society of America, 93(3), 10991131.
Jangid, R. S. (2008). Stochastic response of bridges
doi:10.1785/0120020100
seismically isolated by friction pendulum system.
Journal of Bridge Engineering, 13(4), 319330. Mokkadem, A. (1989). Estimation of the entropy
doi:10.1061/(ASCE)1084-0702(2008)13:4(319) and information for absolutely continuous random
variables. IEEE Transactions on Information
Jaynes, E. T. (2003). Probability theory: The logic
Theory, 35, 193196. doi:10.1109/18.42194
of science. Cambridge, UK: Cambridge University
Press. doi:10.1017/CBO9780511790423 Muthukumar, S., & DesRoches, R. (2006). A
Hertz contact model with non-linear damping for
Jnsson, M. H., Bessason, B., & Haflidason, E.
pounding simulation. Earthquake Engineering &
(2010). Earthquake response of a base-isolated
Structural Dynamics, 35, 811828. doi:10.1002/
bridge subjected to strong near-fault ground
eqe.557
motion. Soil Dynamics and Earthquake En-
gineering, 30(6), 447455. doi:10.1016/j.soil- Perros, K., & Papadimitriou, C. (2009). Reliabil-
dyn.2010.01.001 ity analysis of bridge models with elastomeric
bearings and seismic stoppers under stochastic
Klee, H. (2007). Simulation of dynamic systems
earthquake excitations. Paper presented at the
with MATLAB and SIMULINK. Boca Raton, FL:
2009 ECCOMAS Thematic Conference on Com-
CRC Press.
putational Methods in Structural Dynamics and
Earthquake Engineering.
394
Rezaeian, S., & Der Kiureghian, A. (2008). A Spall, J. C. (1998). Implementation of the
stochastic ground motion model with separable simultaneous perturbation algorithm for sto-
temporal and spectral nonstationatiries. Earth- chastic optimization. IEEE Transactions on
quake Engineering & Structural Dynamics, 37, Aerospace and Electronic Systems, 34, 817823.
15651584. doi:10.1002/eqe.831 doi:10.1109/7.705889
Robert, C. P., & Casella, G. (2004). Monte Carlo Spall, J. C. (2003). Introduction to stochastic
statistical methods (2nd ed.). New York, NY: search and optimization. New York, NY: Wiley-
Springer. Interscience. doi:10.1002/0471722138
Royset, J. O., & Polak, E. (2004). Reliability- Taflanidis, A. A. (2009, 2224 June). Robust
based optimal design using sample average stochastic design of viscous dampers for base
approximations. Probabilistic Engineering isolation applications. Paper presented at the
Mechanics, 19, 331343. doi:10.1016/j.proben- Computational Methods in Structural Dynamics
gmech.2004.03.001 and Earthquake Engineering, Rhodes, Greece.
Ruangrassamee, A., & Kawashima, K. (2003). Taflanidis, A. A., & Beck, J. L. (2008). An efficient
Control of nonlinear bridge response with pound- framework for optimal robust stochastic system
ing effect by variable dampers. Engineering design using stochastic simulation. Computer
Structures, 25(5), 593606. doi:10.1016/S0141- Methods in Applied Mechanics and Engineering,
0296(02)00169-4 198(1), 88101. doi:10.1016/j.cma.2008.03.029
Ruszczynski, A., & Shapiro, A. (2003). Stochastic Taflanidis, A. A., & Beck, J. L. (2009a). Life-cycle
programming. New York, NY: Elsevier. cost optimal design of passive dissipative devices.
Structural Safety, 31(6), 508522. doi:10.1016/j.
Saadeghvaziri, M. A., & Yazdani-Motlagh, A. R.
strusafe.2009.06.010
(2008). Seismic behavior and capacity/demand
analyses of three multi-span simply supported Taflanidis, A. A., & Beck, J. L. (2009b). Stochastic
bridges. Engineering Structures, 30(1), 5466. subset optimization for reliability optimization
doi:10.1016/j.engstruct.2007.02.017 and sensitivity analysis in system design. Com-
puters & Structures, 87, 318331. doi:10.1016/j.
Shen, J., Tsai, M.-H., Chang, K.-C., & Lee, G.
C. (2004). Performance of a seismically iso-
lated bridge under near-fault earthquake ground Taflanidis, A. A., & Beck, J. L. (2009c). Stochastic
motions. Journal of Structural Engineering, subset optimization for reliability optimization and
130(6), 861868. doi:10.1061/(ASCE)0733- sensitivity analysis in system design. Computers
9445(2004)130:6(861) & Structures, 87(5-6), 318331. doi:10.1016/j.
Shinozuka, M., Deodatis, G., Zhang, R. C., &
Papageorgiou, A. S. (1999). Modeling, synthetics Taflanidis, A. A., & Beck, J. L. (2010). Reliability-
and engineering applications of strong earthquake based design using two-stage stochastic optimiza-
wave motion. Soil Dynamics and Earthquake tion with a treatment of model prediction errors.
Engineering, 18(3), 209228. doi:10.1016/S0267- Journal of Engineering Mechanics, 136(12),
7261(98)00045-1 14601473. doi:10.1061/(ASCE)EM.1943-
7889.0000189
395
Taflanidis, A. A., & Jia, G. (2011). A simula- ADDITIONAL READING

tion-based framework for risk assessment and
probabilistic sensitivity analysis of base-isolated Bahi, J. M., Contassot-Vinier, S., & Couturier,
structures. Earthquake Engineering & Structural R. (2008). Parallel iterative algorithms. From
Dynamics, 40(14). doi:10.1002/eqe.1113 sequential to grid computing. Boca Raton, FL:
Chapman & Hall/ CRC.
Taflanidis, A. A., Scruggs, J. T., & Beck, J. L.
(2008). Probabilistically robust nonlinear design Christopoulos, C., & Filiatrault, A. (2006).
of control systems for base-isolated structures. Principles of passive supplemental damping and
Journal of Structural Control and Health Monitor- seismic isolation. Pavia, Italy: IUSS Press.
ing, 15(3), 697719. doi:10.1002/stc.275 Clough, R. W., & Penzien, J. (1993). Dynamics
Takewaki, I. (2002). Seismic critical excitation of structures. New York, NY: Mc Graw-Hill Inc.
method for robust design: A review. Journal Cornell, C. A. (1968). Engineering seismic risk
of Structural Engineering, 128(5), 665672. analysis. Bulletin of the Seismological Society of
doi:10.1061/(ASCE)0733-9445(2002)128:5(665) America, 58(5), 15831606.
Tsopelas, P., Constantinou, M. C., Okamoto, Cox, R. T. (1961). Algebra or probable inference.
S., Fujii, S., & Ozaki, D. (1996). Experimental Baltimore, MD: The Johns Hopkins University
study of bridge seismic sliding isolation sys- Press.
tems. Engineering Structures, 18(4), 301310.
doi:10.1016/0141-0296(95)00147-6 Elnashai, A., & Di Sarnio, L. (2008). Fundamen-
tals of earthquake engineering. West Sussex, UK:
Wang, Q., Fang, H., & Zou, X.-K. (2010). Ap- John Wiley & Sons. doi:10.1002/9780470024867
plication of Micro-GA for optimal cost base
isolation design of bridges subject to transient Fujimoto, R. M. (2001). Parallel simulation:
earthquake loads. Structural and Multidisci- Parallel and distributed simulation systems. Paper
plinary Optimization, 41, 765777. doi:10.1007/ presented at the 33rd Winter Simulation Confer-
s00158-009-0470-5 ence, Arlington, Virginia.
Werner, G. (1960). Impact: The theory and physi- Gavin, H. P., & Zaicenco, A. (2007). Performance
cal behaviour of colliding solids. London, UK: and reliability of semi-active equipment isolation.
Edward Arnold. Journal of Sound and Vibration, 306(1-2), 7490.
doi:10.1016/j.jsv.2007.05.039
Zerva, A. (2008). Spatial variation of seismic
ground motions modeling and engineering ap- Gosavi, A. (2003). Simulation-based optimiza-
plications. Boca Raton, FL: CRC Press. tion: Parametric optimization techniques and
reinforcement learning. Norwell, MA: Kluwer
Zhang, J., Makris, N., & Delis, T. (2004). Academic Publishers.
Structural characterization of modern highway
overcrossings-Case study. Journal of Struc- Haukaas, T., & Der Kiureghian, A. (2007).
tural Engineering, 130(6), 846860. doi:10.1061/ Methods and object-oriented software for FE re-
(ASCE)0733-9445(2004)130:6(846) liability and sensitivity analysis with application
to a bridge structure. Journal of Computing in
Civil Engineering, 21(3), 151163. doi:10.1061/
(ASCE)0887-3801(2007)21:3(151)
396
Jangid, R. S. (2007). Optimum lead-rubber isola- Narasimhan, S., Nagarajaiah, S., & Johnson, E. A.
tion bearings for near-fault motions. Engineer- (2008). Smart base isolated benchmark building
ing Structures, 29, 25032513. doi:10.1016/j. Part IV: Phase II sample controllers for nonlinear
engstruct.2006.12.010 isolation systems. Journal of Structural Control
and Health Monitoring, 13(2-3).
Katsanos, E. I., Sextos, A. G., & Manolis, G. D.
(2010). Selection of earthquake ground motion Nickolls, J., Buck, I., Garland, M., & Skad-
records: A state-of-the-art review from a struc- ron, K. (2008). Scalable parallel program-
tural engineering perspective. Soil Dynamics ming with CUDA. Queue, 6(2), 4053.
and Earthquake Engineering, 30(4), 157169. doi:10.1145/1365490.1365500
doi:10.1016/j.soildyn.2009.10.005
Papadimitriou, C., Beck, J., & Katafygiotis,
Komodromos, P., Polycarpoy, P. C., Papaloizou, L. (1997). Asymptotic expansions for reli-
L., & Phocas, M. C. (2007). Response of seismi- abilities and moments of uncertain dynamic
cally isolated buildings considering poundings. systems. Journal of Engineering Mechanics,
Earthquake Engineering & Structural Dynamics, 123, 12191229. doi:10.1061/(ASCE)0733-
36, 16051622. doi:10.1002/eqe.692 9399(1997)123:12(1219)
Lutes, L. D., & Sarkani, S. (1997). Stochastic Papadimitriou, K. (1990). Stochastic character-
analysis of structural and mechanical vibrations. ization of strong ground motion and application
Upper Saddle River, NJ: Prentice Hall. to structural response. Pasadena, CA: Report No.
EERL 90-03, California Institute of Technology.
Marti, K. (2005). Stochastic optimization methods.
Berlin, Germany: Springer. Politopoulos, I., & Pham, H., K. (2009). Sensitiv-
ity of seismically isolated structures. Earthquake
Naeim, F., & Kelly, J. M. (1999). Design of seismic
Engineering & Structural Dynamics, 38(8),
isolated structures: From theory to practise. New
9891007. doi:10.1002/eqe.879
York, NY: Wiley. doi:10.1002/9780470172742
Priestley, M. J. N., Seible, F., & Calvi, G. M.
Nagarajaiah, S., & Narasimhan, S. (2006). Smart
(1996). Seismic design and retrofit of bridges.
base isolated benchmark building. Part II - phase
New York, NY: John Wiley & Sons, INC.
I sample controllers for linear isolation system.
doi:10.1002/9780470172858
Journal of Structural Control and Health Moni-
toring, 13(2-3), 589604. doi:10.1002/stc.100 Rubinstein, R. Y., & Kroese, D. P. (2008). Simu-
lation and the Monte Carlo method (2nd ed.).
Narasimhan, S., Nagarajaiah, S., Gavin, H. P., &
Hoboken, NJ: Wiley.
Johnson, E. A. (2006). Smart base isolated bench-
mark building part I: Problem definition. Journal Suchard, M. A., Wang, Q., Chan, C., Frelinger, J.,
of Structural Control and Health Monitoring, Cron, A., & West, M. (2009). Understanding GPU
13(2-3), 573588. doi:10.1002/stc.99 programming for statistical computation: studies
in massively parallel massive mixtures. Journal
of Computational and Graphical Statistics, 19(2),
419438. doi:10.1198/jcgs.2010.10016
397
Taflanidis, A. A. (2010). Reliability-based op- Optimization Under Uncertainty: Optimi-

timal design of linear dynamical systems under zation algorithms appropriate for problems for
stochastic stationary excitation and model uncer- which the objective function or the constraints are
tainty. Engineering Structures, 32(5), 14461458. computed through stochastic simulation.
doi:10.1016/j.engstruct.2010.01.023 Probabilistic Design (Robust Design):
Design of systems that explicitly addresses the
Zhang, Y. F., & Iwan, W. D. (2002). Protecting
uncertainties in their model description.
base isolated structures from near-field ground
Risk Quantification and Assessment: Meth-
motion by tuned interaction damper. Journal
odologies for mathematical description of risk and
of Engineering Mechanics, 128(3), 287295.
for its evaluation.
doi:10.1061/(ASCE)0733-9399(2002)128:3(287)
Seismic Dampers: Devices used for energy
dissipation (through mechanical or electrical
means) in the vibration of structural systems due
KEY TERMS AND DEFINITIONS to seismic excitation
Seismic Isolation: Decoupling of structures
Near-Fault Ground Motion: Seismic ground (building, bridges) from the ground through use
motions at small distances from epicenter of an of flexible isolators, commonly lead-rubber or
earthquake that include a strong pulse compo- friction pendulum bearings.
nent, evident in the velocity time history of the Stochastic Simulation: Methodologies and
excitation. algorithms that entail random sampling charac-
teristics.
398
399
Compilation of References
Abbas, A. M. (2006). Critical seismic load inputs for Adeli, H., & Jiang, X. (2006). Dynamic fuzzy wavelet
simple inelastic structures. Journal of Sound and Vibra- neural network model for structural system identifica-
tion, 296, 949967. doi:10.1016/j.jsv.2006.03.021 tion. Journal of Structural Engineering, 132, 102111.
doi:10.1061/(ASCE)0733-9445(2006)132:1(102)
Abbas, A. M., & Manohar, C. S. (2002). Investigations
into critical earthquake load models within deterministic Ahlawat, A. S., & Ramaswamy, A. (2004). Multiobjective
and probabilistic frameworks. Earthquake Engineering & optimal fuzzy logic control system for response control
Structural Dynamics, 31, 813832. doi:10.1002/eqe.124 of wind-excited tall buildings. Journal of Engineering
Mechanics, 130, 524530. doi:10.1061/(ASCE)0733-
Abbas, A. M., & Manohar, C. S. (2005). Reliability-
9399(2004)130:4(524)
based critical earthquake load models. Part 2: Nonlinear
structures. Journal of Sound and Vibration, 287, 883900. Akiyama, H. (1985). Earthquake-resistant limit-state
doi:10.1016/j.jsv.2004.12.003 design for buildings. Tokyo, Japan: University of Tokyo
Press.
Abbas, A. M., & Manohar, C. S. (2007). Reliability-based
vector nonstationary random critical earthquake excita- Alavi, B., & Krawinkler, H. (2000, January 30-February
tions for parametrically excited systems. Structural Safety, 4). Consideration of near-fault ground motion effects in
29, 3248. doi:10.1016/j.strusafe.2005.11.003 seismic design. Paper presented at the 12th World Confer-
ence on Earthquake Engineering, Auckland, New Zealand.
Abbas, M. (2011). Damage-based design earthquake loads
for single-degree-of-freedom inelastic structures. Journal Al-Dawod, M., Samali, B., Kwok, K. C. S., & Naghdy, F.
of Structural Engineering, 137(3), 456467. doi:10.1061/ (2004). Fuzzy controller for seismically excited nonlinear
(ASCE)ST.1943-541X.0000074 buildings. Journal of Engineering Mechanics, 130, 407
415. doi:10.1061/(ASCE)0733-9399(2004)130:4(407)
Abonyi, J., Babuska, R., Verbruggen, H. B., & Szeifert,
F. (2000). Incorporating prior knowledge in fuzzy model Alexandrov, N., Dennis, J. Jr, Lewis, R., & Torczon, V.
identification. International Journal of Systems Science, (1998). A trust-region framework for managing the use
31, 657667. doi:10.1080/002077200290966 of approximation models in optimization. Structural
Optimization, 15(1), 1623. doi:10.1007/BF01197433
Abrahamson, N. A. (1993). Non-stationary spectral
matching program RSPMatch: users manual. Altay, N., & Greene, W. G. (2006). OR/MS research
in disaster operations management. European Journal
Adamu, A., Karihaloo, B. L., & Rozvany, G. I. N. (1994).
of Operational Research, 175, 475493. doi:10.1016/j.
Minimum cost design of reinforced concrete beams us-
ejor.2005.05.016
ing continuum-type optimality criteria. Structural and
Alvin, K. F., Robertson, A. N., Reich, G. W., & Park, K. Arora, J. S. (1999). Optimization of structures subjected to
C. (2003). Structural system identification: from reality dynamic loads. In Leondes, C. T. (Ed.), Structural dynamic
to models. Computers & Structures, 81(12), 11491176. systems computational techniques and optimization (pp.
doi:10.1016/S0045-7949(03)00034-8 173). Gordon and Breach Science Publishers.
Al-Widyan, K., & Angeles, J. (2005). Model-based for- Arora, J. S. (2004). Introduction to optimum design. San
mulation of robust design. Journal of Applied Mechanics Diego, CA: Elsevier Academic Press.
Transaction, 127(3), 388396. doi:10.1115/1.1829728
Arora, J. S., & Haug, E. J. (1979). Methods of design
American Concrete Institute. (2008). ACI-318: Building sensitivity analysis in structural optimization. AIAA
code requirements for structural concrete and commen- Journal, 17(9), 970974. doi:10.2514/3.61260
tary. ACI.
Asadi, P. (2006) Optimization of distribution of reinforce-
Amin, M., & Ang, A. H.-S. (1968). Non-stationary ment in the seismic design of R/C frames. MSc. Thesis
stochastic models of earthquake motions. Proceedings Civil Eng. Dept., Sharif University of Technology.
ASCE, Journal of Engineering Mechanics Division, 94.
ASCE. (2007). Seismic rehabilitation of existing buildings
Anderson, J., & Hough, S. (1984). A model for the shape (pp. 4106). Reston, VA: ASCE Publications.
of the Fourier amplitude spectrum of acceleration at high
Aschheim, M. (1999). Yield point spectra: A simple al-
frequencies. Bulletin of the Seismological Society of
ternative to the capacity spectrum method. SEAOC 1999
America, 74(5), 19691993.
Convention, (pp. 373-379).
Andrade, C., Alonso, C., Garcia, D., & Rodriguez, J.
Aslani, H., & Miranda, E. (2005). Probability-based
(1991). Remaining life time of reinforced concrete struc-
seismic response analysis. Engineering Structures, 27(8),
tures: effect of corrosion in mechanical properties of the
11511163. doi:10.1016/j.engstruct.2005.02.015
steel, life prediction of corrodible structures. Proceedings
of the International Symposium of the National Association strm, K. J., & Eykhoff, P. (1971). System identification-
of Corrosion Engineers, 12/1-12/11, Cambridge, U. K. A survey. Automatica, 7, 123162. doi:10.1016/0005-
1098(71)90059-8
Andrade, C., Alonso, C., & Molina, F. J. (1993). Cover
cracking as a function of rebar corrosion: Part I: Experi- Atkinson, G. M., & Silva, W. (2000). Stochastic mod-
mental test. Materials and Structures, 26(3), 345353. eling of California ground motions. Bulletin of the
Seismological Society of America, 90(2), 255274.
Ang, A. H.-S., Lee, J.-C., & Pires, J. A. (1997). Cost-
doi:10.1785/0119990064
effectiveness evaluation of design criteria. Proceedings
of the International Workshop on Optimal Performance Atkinson, G., & Silva, W. (2000). Stochastic model-
of Civil Infrastructure Systems (pp.1-16). New York, ing of California ground motions. Bulletin of the
NY: ASCE. Seismological Society of America, 90(2), 255274.
doi:10.1785/0119990064
Ansari, F. (2005). Sensing issues in civil structural health
monitoring. Springer. doi:10.1007/1-4020-3661-2 Au, S., & Beck, J. (2001). Estimation of small failure
probabilities in high dimensions by subset simulation.
ANSYS Incorporated. (2006). ANSYS release 10.0, 2006.
Probabilistic Engineering Mechanics, 16(4), 263277.
Applied Technology Council. (1985). ATC-13: Earth- doi:10.1016/S0266-8920(01)00019-4
quake damage evaluation data for California. Redwood
Baber, T., & Noori, M. N. (1985). Random vibration of
City, CA: ATC.
degrading, pinching systems. Journal of Engineering Me-
Arias, A. (1970). A measure of earthquake intensity: chanics, 111(8), 10101026. doi:10.1061/(ASCE)0733-
Seismic design of nuclear power plants (pp. 438468). 9399(1985)111:8(1010)
Cambridge, MA: MIT Press.
400
Baber, T., & Wen, Y. (1981). Random vibration hysteretic, Baykasoglu, A., Owen, S., & Gindy, N. (1999b). A taboo
degrading systems. Journal of the Engineering Mechanics search based approach to find the Pareto optimal set in
Division, 107(6), 10691087. multiple objective optimization. Engineering Optimiza-
tion, 31(6), 731748. doi:10.1080/03052159908941394
Baker, J. W., Lin, T., Shahi, S. K., & Jayaram, N. (2011).
New ground motion selection procedures and selected Bazant, Z. P., & Najjar, L. J. (1972). Nonlinear water dif-
motions for the PEER transportation research program. fusion in unsaturated concrete. [Paris, France.]. Materials
PEER Technical Report 2011/03. 106p. and Structures, 5(25), 320.
Baker, J. W. (2011). Conditional mean spectrum: Tool Beck, J. L., Papadimitriou, C., Chan, E., & Irfanoglu, A.
for ground motion selection. Journal of Structural (1998). A performance-based optimal structural design
Engineering, 137(3), 322331. doi:10.1061/(ASCE) methodology. Report No. EERL 97-03, CA, USA.
ST.1943-541X.0000215
Beirlant, J., Dudewicz, E. J., Gyorfi, L., & Van der Meu-
Baker, J. W., & Cornell, C. A. (2006). Spectral shape, len, E. C. (1997). Nonparametric entropy estimation: An
epsilon and record selection. Earthquake Engineering overview. International journal of Mathematical and
& Structural Dynamics, 35, 10771095. doi:10.1002/ Statistical Sciences, 6(1), 17-40.
eqe.571
Bendse, M. P. (1989). Optimal shape design as a mate-
Balling, R. J., Pister, K. S., & Ciampi, V. (1983). Optimal rial distribution problem. Structural Optimization, 1(4),
seismic-resistant design of a planar steel frame. Earth- 193202. doi:10.1007/BF01650949
quake Engineering & Structural Dynamics, 11, 541556.
Bendse, M. P., & Kikuchi, N. (1988). Generating opti-
doi:10.1002/eqe.4290110407
mal topologies in structural design using a homogeniza-
Bamforth, P. B., & Price, W. F. (1996). An international tion method. Computer Methods in Applied Mechanics
review of chloride ingress into structural concrete. Rep. and Engineering, 71(2), 197224. doi:10.1016/0045-
No. 1303/96/9092. Middlesex, UK: Taywood Engineering 7825(88)90086-2
Ltd., Technology Division.
Bendse, M. P., & Sigmund, O. (1999). Material inter-
Bani-Hani, K., Ghaboussi, J., & Schneider, S. P. (1999). polation schemes in topology optimization. Archive of
Experimental study of identification and control of struc- Applied Mechanics, 69(910), 635654.
tures using neural network Part 1: Identification. Earth-
Bendse, M. P., & Sigmund, O. (2003). Topology opti-
quake Engineering & Structural Dynamics, 28, 9951018.
mization - Theory, methods, and applications. Berlin,
doi:10.1002/(SICI)1096-9845(199909)28:9<995::AID-
Germany: Springer.
EQE851>3.0.CO;2-8
Ben-Tal, A., & Nemirovski, A. (2002). Robust optimiza-
Basoz, N., & Kiremidjian, A. (1999). Development of
tion-methodology and applications. Mathematical Pro-
empirical fragility curves for bridges. Proceedings of
gramming, 92(3), 453480. doi:10.1007/s101070100286
5th US Conference on Lifeline Earthquake Engineering,
ASCE, New York, USA. Bertero, V. V., & Zagajeski, S. W. (1979). Optimal inelastic
design of seismic-resistant reinforced concrete framed
Battaini, M., Casciati, F., & Faravelli, L. (2004). Control-
structures. Paper presented at the Nonlinear Design of
ling wind response through a fuzzy controller. Journal
Concrete Structures, CSCE-ASCE-ACI-CEB Interna-
of Engineering Mechanics, 130, 486491. doi:10.1061/
tional Symposium, Ontario, Canada.
(ASCE)0733-9399(2004)130:4(486)
Bertolini, L. (2008). Steel corrosion and service
Baykasoglu, A., Owen, S., & Gindy, N. (1999a). Solution
life of reinforced concrete structures. Structure
of goal programming models using a basic taboo search
and Infrastructure Engineering, 4(2), 123137.
algorithm. The Journal of the Operational Research
doi:10.1080/15732470601155490
Society, 50(9), 960973.
401
Bertsimas, D., & Sim, M. (2006). Tractable approxima- Blevins, R. D. (1979). Formulas for natural frequency
tions to robust conic optimization problems. Mathematical and mode shape. New York, NY: Litton Education Pub-
Programming, 107(1-2), 536. doi:10.1007/s10107-005- lishing Inc.
0677-1
Bommer, J. J., & Acevedo, A. B. (2004). The use of
Besset, S., & Jezequel, L. (2007). Optimization of real earthquake accelerograms as input to dynamic
structural dynamic behaviour based on effective modal analysis. Journal of Earthquake Engineering, 1, 4391.
parameters. International Journal for Numerical Methods doi:10.1080/13632460409350521
in Engineering, 70(5), 523542. doi:10.1002/nme.1890
Boore, D. (2003). Simulation of ground motion using the
Beyer, H.-G., & Sendhoff, B. (2007). Robust optimization stochastic method. Pure and Applied Geophysics, 160(3),
A comprehensive survey. Computer Methods in Applied 635676. doi:10.1007/PL00012553
Mechanics and Engineering, 196(3334), 31903218.
Boore, D. M. (1983). Stochastic simulation of high-
doi:10.1016/j.cma.2007.03.003
frequency ground motions based on seismological models
Bhattacharjya, S., & Chakraborty, S. (2009). Robust opti- of the radiated spectra. Bulletin of the Seismological
mization of linear dynamic system with random parameters Society of America, 73, 18651894.
under stochastic earthquake excitation. International
Boore, D. M. (2003). Simulation of ground motion using
Journal of Reliability Quality and Safety Engineering,
the stochastic method. Pure and Applied Geophysics, 160,
16(3), 261279. doi:10.1142/S0218539309003393
635676. doi:10.1007/PL00012553
Bhatti, M. A., & Pister, K. S. (1981). A dual criteria
Boore, D., Joyner, W., & Fumal, T. (1997). Equations
approach for optimal design of earthquake-resistant
for estimating horizontal response spectra and peak ac-
structural systems. Earthquake Engineering & Structural
celeration from western North American earthquakes: A
Dynamics, 9, 557572.
summary of recent work. Seismological Research Letters,
Biondini, F., Bontempi, F., Frangopol, D. M., & Malerba, 68(1), 128153. doi:10.1785/gssrl.68.1.128
P. G. (2004). Cellular automata approach to durability
Bouazizi, M.-L., Ghanmi, S., & Bouhaddi, N. (2009).
analysis of concrete structures in aggressive environments.
Multi-objective optimization in dynamics of the structures
Journal of Structural Engineering, 130(11), 17241737.
with nonlinear behavior: Contributions of the metamodels.
doi:10.1061/(ASCE)0733-9445(2004)130:11(1724)
Finite Elements in Analysis and Design, 45(10), 612623.
Bjerager, P., & Krenk, S. (1989). Parametric sensitivity in doi:10.1016/j.finel.2009.03.007
first order reliability theory. Journal of Engineering Me-
Bouc, R. (1967). Forced vibration of mechanical systems
chanics, 115(7), 15771582. doi:10.1061/(ASCE)0733-
with hysteresis. Proceedings of the 4th Conference on Non-
9399(1989)115:7(1577)
linear Oscillations, Prague, Czechoslovakia, Sept. 5-9.
Blackshire, J. L., Martin, S., & Cooney, A. (2006).
Bozorgenia, Y., & Bertero, V. V. (2003). Damage spectra:
Characterization and modeling of bonded piezoelectric
Characteristics and applications to seismic risk reduction.
sensor performance and durability in simulated aircraft
environments. Proceedings of 3rd European Workshop
doi:10.1061/(ASCE)0733-9445(2003)129:10(1330)
on Structural Health Monitoring, Granada, Spain.
Bozorgnia, Y., & Bertero, V. V. (Eds.). (2004). Earth-
Bland, J. (1998). Structural design optimization with reli-
quake engineering. New York, NY: CRC Press.
ability constraints using tabu search. Engineering Optimi-
doi:10.1201/9780203486245
zation, 30(1), 5574. doi:10.1080/03052159808941238
402
Bratus, A. S., & Seyranian, A. P. (1983). Bimodal solu- California Department of Transportation. (2008). Caltrans
tions in eigenvalue optimization problems. Journal of contract cost data. Sacramento, CA: Author.
Applied Mathematics and Mechanics, 47(4), 451457.
Canyurt, O. E., & Hajela, P. (2005). A cellular framework
doi:10.1016/0021-8928(83)90081-3
for structural analysis and optimization. Computer Meth-
Bray, J. D., & Rodriguez-Marek, A. (2004). Characteriza- ods in Applied Mechanics and Engineering, 194(30-33),
tion of forward-directivity ground motions in the near-fault 35163534. doi:10.1016/j.cma.2005.01.014
region. Soil Dynamics and Earthquake Engineering, 24,
Cao, X. (2009). Design strategy on aseismic capacity of
815828. doi:10.1016/j.soildyn.2004.05.001
large bridge. Doctoral dissertation, Tongji University,
Brozzetti, J. (2000). Design development of steel- Shanghai, China.
concrete composite bridges in France. Journal of Con-
Casciati, F. (1997). Checking the stability of a fuzzy con-
structional Steel Research, 55, 229243. doi:10.1016/
troller for nonlinear structures. Microcomputers in Civil
S0143-974X(99)00087-5
Engineering, 12, 205215. doi:10.1111/0885-9507.00057
Brune, J. N. (1970). Tectonic stress and the spectra
Chai, Y. H., Romstad, K. M., & Bird, S. M. (1995). Energy-
of seismic shear waves from earthquakes. Journal of
based linear damage model for high-intensity seismic load-
Geophysical Research, 75, 49975009. doi:10.1029/
ing. Journal of Structural Engineering, 121(5), 857864.
JB075i026p04997
doi:10.1061/(ASCE)0733-9445(1995)121:5(857)
Bucher, C., & Bourgund, U. (1990). A fast and efficient
Chan, C. M. (1997). How to optimize tall steel build-
response surface approach for structural reliability prob-
ing frameworks. Guide to structural optimization (pp.
lems. Structural Safety, 7(1), 5766. doi:10.1016/0167-
165-195). ASCE Manuals and Reports on Engineering
4730(90)90012-E
Practice No.90.
Bucher, C., & Macke, M. (2005). Response surfaces
Chan, C. M., & Zou, X. K. (2004). Elastic and inelastic
for reliability assessment. In Nikolaidis, E., Ghiocel, D.
drift performance optimization for reinforced concrete
M., & Singhal, S. (Eds.), Engineering design reliability
building under earthquake loads. Earthquake Engineering
handbook. Boca Raton, FL: CRC Press.
& Structural Dynamics, 33(8), 929950. doi:10.1002/
Buhl, T., Pedersen, C. B. W., & Sigmund, O. (2000). eqe.385
Stiffness design of geometrically nonlinear structures
Chang, P. C., Flatau, A., & Liu, S. C. (2003). Re-
using topology optimization. Structural and Multidis-
view paper: Health monitoring of civil infrastruc-
ciplinary Optimization, 19(2), 93104. doi:10.1007/
ture. Structural Health Monitoring, 2(3), 257267.
s001580050089
doi:10.1177/1475921703036169
Buratti, N., Stafford, P. J., & Bommer, J. J. (2011). Earth-
Chang, S., & Shinozuka, M. (1996). Life cycle cost
quake accelerogram selection and scaling procedures for
analysis with natural hazard risk. Journal of Infrastruc-
estimating the distribution of drift response. Journal of
ture Systems, 2(3), 118126. doi:10.1061/(ASCE)1076-
Structural Engineering, 137(3), 345357. doi:10.1061/
0342(1996)2:3(118)
(ASCE)ST.1943-541X.0000217
Chan, R. W. K., & Albermani, F. (2008). Experimental
Caleman, T., Branch, M. A., & Grace, A. (1999). Opti-
study of steel slit damper for passive energy dissipation.
mization toolbox for the use with Matlab. Users guide.
Engineering Structures, 30(4), 10581066. doi:10.1016/j.
USA: The MATH WORKS Inc.
California Department of Transportation. (2006). Caltrans
seismic design criteria. Sacramento, CA: Author.
403
Charney, F. A. (2000). Needs in the development of a Chen, Y., Yang, B., Abraham, A., & Peng, L. (2007).
comprehensive performance based design optimization Automatic design of hierarchical Takagi-Sugeno type
process. In M. Elgaaly (Ed.), Advanced Technology in fuzzy systems using evolutionary algorithms. IEEE
Structural Engineering - Structural Congress 2000, ASCE, Transactions on Fuzzy Systems, 15, 385397. doi:10.1109/
May 8-10, (CD-ROM). TFUZZ.2006.882472
Chassiakos, A. G., & Masri, S. F. (1996). Identification of Choi, E., DesRoches, R., & Nielson, B. (2004). Seismic
structural systems by neural networks. Mathematics and fragility of typical bridges in moderate seismic zones.
Computers in Simulation, 40, 637656. doi:10.1016/0378- Journal of Engineering Structures, 26(2), 187199.
4754(95)00012-7 doi:10.1016/j.engstruct.2003.09.006
Chaudhuri, A., & Chakraborty, S. (2004). Sensitivity Choi, H., & Kim, J. (2006). Energy-based design of
evaluation in seismic reliability analysis of structures. buckling-restrained braced frames using hysteretic en-
Computer Methods in Applied Mechanics and Engineer- ergy spectrum. Engineering Structures, 28, 304311.
ing, 193(1-2), 5968. doi:10.1016/j.cma.2003.09.007 doi:10.1016/j.engstruct.2005.08.008
Chaudhuri, A., & Chakraborty, S. (2006). Reliability of Choi, K. M., Cho, S. W., Jung, H. J., & Lee, I. W. (2004).
linear structures with parameter uncertainty under non- Semi-active fuzzy control for seismic response reduction
stationary earthquake. Structural Safety, 28(3), 231246. using magnetorheological dampers. Earthquake Engineer-
doi:10.1016/j.strusafe.2005.07.001 ing & Structural Dynamics, 33, 723736. doi:10.1002/
eqe.372
Cheng, F. Y., & Chang, C. C. (1988). Safety-based optimum
design of nondeterministic structures subjected to various Chopra, A. K. (1995). Dynamics of structures Theory
types of seismic loads. NSF Report, U.S. Department of and applications to earthquake engineering. Upper Saddle
Commerce, VA, NTIS No. PB90-133489/AS. River, NJ: Prentice Hall International.
Cheng, F. Y., & Botkin, M. E. (1976). Nonlinear opti- Chopra, A. K. (2007). Dynamics of structures (3rd ed.).
mum design of dynamic damped frames. Journal of the Englewood Cliffs, NJ: Prentice-Hall.
Structural Division, 102, 609628.
Chung, L. L., Lin, R. C., Soong, T. T., & Reinhorn, A. M.
Cheng, F. Y., & Li, D. (1997). Multiobjective optimiza- (1989). Experiments on active control for MDOF seismic
tion design with Pareto genetic algorithm. Journal of structures. Journal of Engineering Mechanics, 115, 1609
Structural Engineering, 123(9), 12521261. doi:10.1061/ 1627. doi:10.1061/(ASCE)0733-9399(1989)115:8(1609)
(ASCE)0733-9445(1997)123:9(1252)
Clauset, A., Moore, C., & Newman, M. E. J. (2008). Hi-
Cheng, F. Y., & Truman, K. Z. (1982). Optimization erarchical structure and the prediction of missing links in
algorithm of 3-D building systems for static and seismic networks. Nature, 453, 98101. doi:10.1038/nature06830
loading. In Ames, W. F. (Ed.), Modeling and simulation
Clough, R. W., & Penzien, J. (1975). Dynamics of struc-
in engineering (pp. 315326). North-Holland Pub. Co.
tures. McGraw-Hill.
Chen, J., Che, J., Sun, H., Ma, H., & Cui, M. (2002).
Clough, R. W., & Penzien, J. (2003). Dynamics of struc-
Structural dynamic optimization with probability con-
tures (3rd ed.). Computers & Structures Inc.
straints of frequency and mode. Structural Engineering
& Mechanics, 13(5), 479490. Coello, C. A. (2000). An updated survey of GA-based
multi-objective optimization techniques. ACM Computing
Chen, S. H., Song, M., & Chen, Y. D. (2007). Robustness
Surveys, 32(2), 109143. doi:10.1145/358923.358929
analysis of vibration control structures with uncertain
parameters using interval algorithm. Structural Safety,
29(2), 94111. doi:10.1016/j.strusafe.2006.03.001
404
Collins, K. R., Wen, Y. K., & Foutch, D. A. (1996). COSMOS. (2005). Consortium organizations for strong-
Dual-level seismic design: A reliability-based meth- motion observation systems. Retrieved from http://
odology. Earthquake Engineering & Structural Dy- db.cosmos-eq.org/scripts/default.plx
namics, 25(12), 14331467. doi:10.1002/(SICI)1096-
Cox, S. J., & Overton, M. L. (1992). On the optimal design
9845(199612)25:12<1433::AID-EQE629>3.0.CO;2-M
of columns against buckling. SIAM Journal on Math-
Colorni, A., Dorigo, M., & Maniezzo, V. (1992). An ematical Analysis, 23(2), 287325. doi:10.1137/0523015
investigation of some properties of an ant algorithm.
Crisfield, M. A. (2000). Non-linear finite element analy-
In R. Manner, & B. Manderick (Eds.), Proceedings of
sis of solids and structures, Vol. 1: Essentials; Vol. 2:
the Parallel Problem Solving from Nature Conference
Advanced topics. New York, NY: John Wiley & Sons.
(PPSN 92), (pp. 509-520). Brussels, Belgium: Elsevier
Publishing. Culmann, K. (1866). Die graphische Statik. Zrich,
Switzerland: Mayer und Zeller.
Colorni, A., Dorigo, M., & Maniezzo, V. (1992). Dis-
tributed optimization by ant colonies. In F. Varela & Daigle, M., Koutsoukos, X., & Biswas, G. (2006).
P. Bourgine (Eds.), Proceedings of the First European Distributed diagnosis of coupled mobile robots. IEEE
Conference on Artificial Life, (pp. 134-142). Paris, France: International Conference on Robotics and Automation,
Elsevier Publishing. (pp. 3787-3794).
Comartin, C., Brvez, S., Naeim, F., Greene, M., Blondet, Datta, T. K. (2010). Seismic analysis of structures. Singa-
M., & Cherry, S. (2004). A challenge to the earthquake pore: John Wiley & Sons. doi:10.1002/9780470824634
engineering professionals. Earthquake Spectra, 20(4),
10491056. doi:10.1193/1.1809130 Davis, R. O. (1992). Pounding of buildings modeled by an
impact oscillator. Earthquake Engineering & Structural
Computer and Structures, Inc. (CSI). (2000). SAP2000/ Dynamics, 21, 253274. doi:10.1002/eqe.4290210305
NL-PUSH software, Version 7.40. Berkeley, CA: Com-
puter and Structures, Inc. De Donato, O. (1977). Fundamentals of elastic-plastic
analysis. In Cohn, M. Z., & Maier, G. (Eds.), Engineering
Conte, J. P., & Peng, B. F. (1997). Fully nonstationary plasticity by mathematical programming (pp. 325349).
analytical earthquake ground motion model. Journal New York, NY: Pergamon Press.
of Engineering Mechanics, 123, 1524. doi:10.1061/
(ASCE)0733-9399(1997)123:1(15) De la Llera, J., Esguerra, C., & Almazn, J. (2004).
Earthquake behavior of structures with copper energy
Cornell, A. C. (2008). Uncertainty propagation in proba- dissipators. Earthquake Engineering & Structural Dy-
bilistic seismic loss estimation. Structural Safety, 30(3), namics, 33(3), 329358. doi:10.1002/eqe.354
236252. doi:10.1016/j.strusafe.2006.11.003
De Oliveira, R., Frazao, O., Santos, J. L., & Marques, A. T.
Cornell, C. A., Jalayer, F., Hamburger, R. O., & Foutch, D. (2004). Optic fibre sensor for real-time damage detection
A. (2002). Probabilistic basis for 2000 SAC federal emer- in smart composite. Computers & Structures, 82(17-19),
gency management agency steel moment frame guidelines. 13151321. doi:10.1016/j.compstruc.2004.03.028
doi:10.1061/(ASCE)0733-9445(2002)128:4(526) Deb, K. (2001). Multi-objective optimization using evo-
lutionary algorithms. New York, NY: John Wiley & sons.
Cosenza, C., Manfredi, G., & Ramasco, R. (1993). The
use of damage functionals in earthquake engineering: Decanini, L. D., & Mollaioli, F. (2001). An energy-based
A comparison between different methods. Earthquake methodology for the assessment of seismic demand. Soil
Engineering & Structural Dynamics, 22, 855868. Dynamics and Earthquake Engineering, 21, 113137.
doi:10.1002/eqe.4290221003 doi:10.1016/S0267-7261(00)00102-0
405
Der Kiureghian, A. (1996). A coherency model for spatially Drenick, R. F. (1977). The critical excitation of nonlinear
varying ground motions. Earthquake Engineering & Struc- systems. Journal of Applied Mechanics, 18, 333336.
tural Dynamics, 25(1), 99111. doi:10.1002/(SICI)1096- doi:10.1115/1.3424047
9845(199601)25:1<99::AID-EQE540>3.0.CO;2-C
Du, H., & Zhang, N. (2008). Application of evolving
Daaz, A. R., & Kikuchi, N. (1992). Solutions to shape Takagi-Sugeno fuzzy model to nonlinear system identifica-
and topology eigenvalue optimization problems using tion. Applied Soft Computing, 8, 676686. doi:10.1016/j.
a homogenization method. International Journal for asoc.2007.05.006
Numerical Methods in Engineering, 35(7), 14871502.
Du, J., & Olhoff, N. (2007). Topological design of freely
doi:10.1002/nme.1620350707
vibrating continuum structures for maximum values of
Dicleli, M. (2006). Performance of seismic-isolated simple and multiple eigenfrequencies and frequency gaps.
bridges in relation to near-fault ground-motion and isola- Structural and Multidisciplinary Optimization, 34(2),
tor characteristics. Earthquake Spectra, 22(4), 887907. 91110. doi:10.1007/s00158-007-0101-y
doi:10.1193/1.2359715
Du, X., & Chen, W. (2000). Towards a better understanding
Dimitrakopoulos, E., Makris, N., & Kappos, A. J. of modelling feasibility robustness in engineering design.
(2010). Dimensional analysis of the earthquake response ASME Journal of Mechanical Design, 122(4), 385394.
of a pounding oscillator. Journal of Engineering Me- doi:10.1115/1.1290247
chanics, 136(3), 299310. doi:10.1061/(ASCE)0733-
Du, X., Sudijianto, A., & Chen, W. (2004). An interval
9399(2010)136:3(299)
framework for optimization under uncertainty using in-
Doebling, S. W., Farrar, C. R., & Prime, M. B. (1998). A verse reliability strategy. ASME Journal of Mechanical
summary review of vibration-based damage identification Design, 126(4), 562570. doi:10.1115/1.1759358
methods. The Shock and Vibration Digest, 30(2), 91105.
Du, Y. G., Clark, L. A., & Chan, A. H. C. (2005a). Re-
doi:10.1177/058310249803000201
sidual capacity of corroded reinforcing bars. Magazine
Dolsek, M. (2009). Incremental dynamic analysis with of Concrete Research, 57(3), 135147. doi:10.1680/
consideration of modelling uncertainties. Earthquake macr.2005.57.3.135
Engineering & Structural Dynamics, 38(6), 805825.
Du, Y. G., Clark, L. A., & Chan, A. H. C. (2005b). Effect
doi:10.1002/eqe.869
of corrosion on ductility of reinforcing bars. Magazine
Dong, W. M., Chiang, W. L., & Shah, H. C. (1987). Fuzzy of Concrete Research, 57(7), 407419. doi:10.1680/
information processing in seismic hazard analysis and de- macr.2005.57.7.407
cision making. Soil Dynamics and Earthquake Engineer-
Dyke, S. J., Spencer, B. F. Jr, Sain, M. K., & Carlson, J.
ing, 6(4), 202226. doi:10.1016/0267-7261(87)90003-0
D. (1996). Modeling and control of magnetorheological
Dorigo, M. (1992). Optimization, learning and natural dampers for seismic response reduction. Smart Mate-
algorithms. Milano, Italy: Politecnico di Milano. rials and Structures, 5, 565575. doi:10.1088/0964-
1726/5/5/006
Dorigo, M., & Sttzle, T. (2004). Ant colony optimization.
The MIT Press. Dyke, S. J., Spencer, B. F. Jr, Sain, M. K., & Carlson,
J. D. (1998). An experimental study of MR dampers for
Dounis, A. I., Tiropanis, P., Syrcos, G. P., & Tseles, D.
seismic protection. Smart Materials and Structures, 7,
(2007). Evolutionary fuzzy logic control of base-isolated
693703. doi:10.1088/0964-1726/7/5/012
structures in response to earthquake activity. Structural
Control and Health Monitoring, 14, 6282. doi:10.1002/
stc.83
406
EC2. (2004). Eurocode 2. Design of concrete struc- Fajfar, P., & Krawinklere, H. (1997). Seismic design meth-
turesPart 1: General rules and rules for buildings. (The odologies for the next generation of codes. Rotterdam,
European Standard EN 1992-1-1). Brussels, Belgium: The Netherlands: Balkema.
European Committee for Standardisation.
Fan, L. (Ed.). (1997). Seismic resistance of bridge. Shang-
EC8. (2004). Eurocode 8: Design of structures for hai, China: Tongji University Press.
earthquake resistance. (The European Standard EN
Fan, L., & Zhuo, W. (Eds.). (2001). Seismic ductility design
1998-1). Brussels, Belgium: European Committee for
of bridge. Beijing, China: Peoples Communication Press.
Standardisation.
Faravelli, L., & Rossi, R. (2002). Adaptive fuzzy control:
El Maaddawy, T., & Soudki, K. (2007). A model for
Theory versus implementation. Journal of Structural
prediction of time from corrosion initiation to corrosion
Control, 9, 5973. doi:10.1002/stc.5
cracking. Journal of Cement and Concrete Composites,
29(3), 168175. doi:10.1016/j.cemconcomp.2006.11.004 Fardis, M. N. (Ed.). (2010). Advances in performance-
based earthquake engineering. New York, NY: Springer.
Elishakoff, I., & Ohsaki, M. (2010). Optimization and anti-
optimization of structures under uncertainty. Singapore: Federal Emergence Management Agency. (2000). Pres-
Imperial College Press. doi:10.1142/9781848164789 tandard and commentary for the seismic rehabilitation of
buildings. FEMA-356. Washington, DC.
Elnashai, A. S., & Chryssanthopoulos, M. (1991). Effect of
random material variability on seismic design parameters Federal Emergence Management Agency. (2005). Im-
of steel frames. Earthquake Engineering & Structural provement of nonlinear static seismic analysis procedure.
Dynamics, 20(2), 101114. doi:10.1002/eqe.4290200202 FEMA-440. Washington, DC.
Elouedi, Z., Mellouli, K., & Smets, P. (2004). Assessing Federal Emergence Management Agency. (2006). Next-
sensor reliability for multisensor data fusion within the generation performance-based seismic design guidelines.
transferable belief model. IEEE Transactions on Sys- FEMA-445. Washington, DC.
tems, Man, and Cybernetics. Part B, Cybernetics, 34(1),
782787. doi:10.1109/TSMCB.2003.817056 FEMA. (1997). NEHRP guidelines for the seismic re-
habilitation of buildings, FEMA 273. Washington, DC:
Enevoldsen, I., & Srensen, J. (1994). Reliability-based Federal Emergency Management Agency.
optimization in structural engineering. Structural Safety,
15(3), 169196. doi:10.1016/0167-4730(94)90039-6 FEMA. (2003). NEHRP recommended provisions for
seismic regulations for new buildings and other structures,
Esteva, L., Lpez, O. D., & Hernndez, E. I. (2010). FEMA 450, Part 1: Provisions. Washington, DC: Federal
Seismic vulnerability functions of multi-story buildings: Emergency Management Agency.
estimation and applications. Structure and Infrastructure
Engineering Special Issue: Vulnerability Assessment of FEMA. 356. (2000). Prestandard and Commentary for
Structures and Infrastructures, 6(1-2), 316. the seismic rehabilitation of buildings. Reston, VA: ASCE
Publications.
Fajfar, P. (1992). Equivalent ductility factors, taking
into account low-cyclic fatigue. Earthquake Engineer- Feng, T. T., Arora, J. S., & Huang, E. J. (1977). Optimal
ing & Structural Dynamics, 21, 837848. doi:10.1002/ structural design under dynamic loads. International
eqe.4290211001 Journal for Numerical Methods in Engineering, 11(1),
3952. doi:10.1002/nme.1620110106
Fajfar, P. (2000). A nonlinear analysis method for perfor-
mance-based seismic design. Earthquake Spectra, 16(3), Filev, D. P. (1991). Fuzzy modeling of complex systems.
573599. doi:10.1193/1.1586128 International Journal of Approximate Reasoning, 5,
281290. doi:10.1016/0888-613X(91)90013-C
407
Fleury, C. (1983). Dual methods for optimizing finite Frangopol, D. M. (2011). Life-cycle performance,
element flexural systems. Computer Methods in Ap- management, and optimisation of structural systems
plied Mechanics and Engineering, 37(3), 249275. under uncertainty: Accomplishments and challenges.
doi:10.1016/0045-7825(83)90078-6 Structure and Infrastructure Engineering, 7(6), 389413.
doi:10.1080/15732471003594427
Fleury, C. (1989). CONLIN: An efficient dual optimizer
based on convex approximation concepts. Structural Fujita, K., Moustafa, A., & Takewaki, I. (2010). Optimal
Optimization, 1(2), 8189. doi:10.1007/BF01637664 placement of viscoelastic dampers and supporting mem-
bers under variable critical excitations. Earthquakes and
Fleury, C., & Braibant, V. (1986). Structural optimization:
Structures, 1(1), 4367.
A new dual method using mixed variables. International
Journal for Numerical Methods in Engineering, 23(3), Ganzerli, S., Pantelides, C. P., & Reaveley, L. D.
409428. doi:10.1002/nme.1620230307 (2000). Performance-based design using structural
optimization. Earthquake Engineering & Structural
Foley, C. M. (2002). Optimized performance-based design
Dynamics, 29(11), 16771690. doi:10.1002/1096-
for buildings. In Burns, S. A. (Ed.), Recent advances in
9845(200011)29:11<1677::AID-EQE986>3.0.CO;2-N
optimal structural design (pp. 169240). American Society
of Civil Engineers. Gao, Y. (2009). Study on aseismic capacity of elevated
pile cap foundations of large bridge. Doctoral disserta-
Foley, C. M., & Schinler, D. (2003). Automated design of
tion, Tongji University, Shanghai, China.
steel frames using advanced analysis and object-oriented
evolutionary computation. Journal of Structural Engi- Gao, X. W., & Bao, A. B. (1985). Probabilistic model
neering, 129(5), 648660. doi:10.1061/(ASCE)0733- and its statistical parameters for seismic load. Earthquake
9445(2003)129:5(648) Engineering and Engineering Vibration, 5(1), 1322.
Foschi, R. O. (2000). Modeling the Hysteretic response of Gauss, K. F. (1963). Theory of the motion of the heavenly
mechanical connectors for wood structures. Proceedings, bodies moving about the sun in conic sections: A Transla-
World Timber Engineering Conference, Whistler, B.C. tion of Theoria Motus. New York, NY: Dover Publications.
Fragiadakis, M., & Lagaros, N. D. (2011). An overview to Ghabraie, K., Chan, R., Huang, X., & Xie, Y. M. (2010).
structural seismic design optimisation frameworks. Com- Shape optimization of metallic yielding devices for pas-
puters & Structures, 89(11-12), 11551165. doi:10.1016/j. sive mitigation of seismic energy. Engineering Structures,
compstruc.2010.10.021 32(8), 22582267. doi:10.1016/j.engstruct.2010.03.028
Fragiadakis, M., Lagaros, N. D., & Papadrakakis, M. Ghanem, R., & Spanos, P. D. (1990). Stochastic finite
(2006a). Performance-based earthquake engineering us- element analysis: A spectral approach. Berlin, Germany:
ing structural optimisation tools. International Journal Springer.
of Reliability and Safety, 1(1/2), 5976. doi:10.1504/
Ghobara, A., Abou-Elfath, H., & Biddah, A. (1999).
IJRS.2006.010690
Response-based damage assessment of structures. Earth-
Fragiadakis, M., Lagaros, N. D., & Papadrakakis, M. quake Engineering & Structural Dynamics, 28, 79104.
(2006b). Performance-based multiobjective optimum doi:10.1002/(SICI)1096-9845(199901)28:1<79::AID-
design of steel structures considering life-cycle cost. EQE805>3.0.CO;2-J
Structural and Multidisciplinary Optimization, 32(1),
Ghobarah, A. (2001). Performance-based design in
111. doi:10.1007/s00158-006-0009-y
earthquake engineering: State of development. Journal
Fragiadakis, M., & Papadrakakis, M. (2008). Performance- of Engineering Structures, 23(8), 878884. doi:10.1016/
based optimum seismic design of reinforced concrete S0141-0296(01)00036-0
structures. Earthquake Engineering & Structural Dynam-
ics, 37(6), 825844. doi:10.1002/eqe.786
408
Ghobarah, A., Abou-Elfath, H., & Biddah, A. (1999). Glisic, B., & Inaudi, D. (2007). Fibre optic methods for
Response-based damage assessment of structures. Earth- structural health monitoring. Chichester, UK: John Wiley
quake Engineering & Structural Dynamics, 28(1), 79104. & Sons. doi:10.1002/9780470517819
doi:10.1002/(SICI)1096-9845(199901)28:1<79::AID-
Glover, F. (1989). Tabu search - Part I. ORSA Journal on
EQE805>3.0.CO;2-J
Computing, 1(3), 190-206.
Gholizadeh, S., & Salajegheh, E. (2010b). A cellular
Glover, F. (1990). Tabu search - Part II. ORSA Journal
genetic algorithm for structural optimisation. In B. H.
on computing, 2(1), 4-32.
V. Topping, J. M. Adam, F. J. Pallars, R. Bru, & M. L.
Romero, (Ed.), Proceedings of the Tenth International GME-6. (2006). A generic modeling environment, GME
Conference on Computational Structures Technology, 6 users manual. Vanderbilt University. Retrieved from
DATE, (pp. 1-14). Stirlingshire, UK: Civil-Comp Press. http://www.isis.vanderbilt.edu
Gholizadeh, S. (2010). Structural optimization for earth- Goel, R. K. (1997). Seismic response of asymmetric
quake loading with nonlinear responses by surrogate systems: Energy-based approach. Journal of Struc-
modeling based evolutionary algorithms. Asian Journal tural Engineering, 123(11), 14441453. doi:10.1061/
of Civil Engineering, 11(1), 2542. (ASCE)0733-9445(1997)123:11(1444)
Gholizadeh, S., & Salajegheh, E. (2009). Optimal design Goldberg, D. (1989). Genetic algorithms in search,
of structures for time history loading by swarm intelli- optimization, and machine learning. Reading, MA: Ad-
gence and an advanced metamodel. Computer Methods dison Wesley.
in Applied Mechanics and Engineering, 198(37-40),
29362949. doi:10.1016/j.cma.2009.04.010 Gong, Y. (2003). Performance-based design of steel
building frameworks under seismic loading. Doctoral
Gholizadeh, S., & Salajegheh, E. (2010a). Optimal seismic dissertation, University of Waterloo, Waterloo, Canada.
design of steel structures by an efficient soft computing
based algorithm. Journal of Constructional Steel Research, Gonzalez, M. P., & Zapico, J. L. (2008). Seismic dam-
66(1), 8595. doi:10.1016/j.jcsr.2009.07.006 age detection in buildings using neural networks and
modal data. Computers & Structures, 86(3-5), 416426.
Gholizadeh, S., Salajegheh, J., & Salajegheh, E. (2009). An doi:10.1016/j.compstruc.2007.02.021
intelligent neural system for predicting structural response
subject to earthquake. Advances in Engineering Software, Groenwold, A., Etman, L., Snyman, J., & Rooda, J. (2007).
40(8), 630639. doi:10.1016/j.advengsoft.2008.11.008 Incomplete series expansion for function approximation.
Structural and Multidisciplinary Optimization, 34(1),
Gholizadeh, S., & Samavati, O. A. (2011). Structural 2140. doi:10.1007/s00158-006-0070-6
optimization by wavelet transforms and neural networks.
Applied Mathematical Modelling, 35(2), 915929. Groenwold, A., & Wood, D. W. (2009). Globally conver-
doi:10.1016/j.apm.2010.07.046 gent optimization algorithm using conservative convex
separable diagonal quadratic approximations. AIAA
Glasserman, P., & Yao, D. D. (1992). Some guidelines and Journal, 47(11), 26492657. doi:10.2514/1.41975
guarantees for common random numbers. Management
Science, 38(6), 884908. doi:10.1287/mnsc.38.6.884 Guedri, M., Ghanmi, S., Majed, R., & Bouhaddi, N.
(2009). Robust tools for prediction of variability and
Glass, G. K., & Buenfeld, N. R. (2000). The influence of optimization in structural dynamics. Mechanical Systems
chloride binding on the chloride induced corrosion risk in and Signal Processing, 23(4), 11231133. doi:10.1016/j.
reinforced concrete. Corrosion Science, 42(2), 329344. ymssp.2008.08.013
doi:10.1016/S0010-938X(99)00083-9
409
Gunawan, S., & Azarm, S. (2005). Multi-objective robust Hasan, R., Xu, L., & Grierson, D. E. (2002). Push-over
optimization using a sensitivity region concept. Struc- analysis for performance-based seismic design. Com-
tural and Multidisciplenary Optimization, 29(1), 5060. puters & Structures, 80(31), 24832493. doi:10.1016/
doi:10.1007/s00158-004-0450-8 S0045-7949(02)00212-2
Gunawan, S., & Papalambros, P. Y. (2007). Reliability Haselton, C., Baker, J. W., Liel, A. B., & Deierlein,
optimization with mixed continuous-discrete random G. G. (2011). Accounting for ground motion spectral
variables and parameters. Journal of Mechanical Design, shape characteristics in structural collapse assessment
129(2), 158165. doi:10.1115/1.2406085 through an adjustment for epsilon. Journal of Structural
Engineering, 137(3), 332344. doi:10.1061/(ASCE)
Guo, X., Bai, W., Zhang, W., & Gao, X. (2009). Confidence
ST.1943-541X.0000103
structural robust design and optimization under stiffness
and load uncertainties. Computer Methods in Applied Hassan, R., & Crossley, W. (2008). Reliability-based
Mechanics and Engineering, 198(41-44), 33783399. optimization under uncertainty including discrete vari-
doi:10.1016/j.cma.2009.06.018 ables. Journal of Spacecraft and Rockets, 45(2), 394405.
doi:10.2514/1.28827
Gu, Z. Q., & Oyadiji, O. (2008). Application of MR
damper in structural control using ANFIS method. Haug, E. J., & Rousselet, B. (1980). Design sensitivity
Computers & Structures, 86, 427436. doi:10.1016/j. analysis in structural mechanics. II: Eigenvalue varia-
compstruc.2007.02.024 tions. Journal of Structural Mechanics, 8(2), 161186.
doi:10.1080/03601218008907358
Habibi, A. (2008). Optimal seismic performance-based
design of 2D reinforced concrete frames. Doctoral dis- Haupt, R. L., & Haupt, S. E. (2004). Practical genetic
sertation, Tarbiat Modarres University, Tehran, Iran. algorithms. New Jersey: Wiley-Interscience Publication.
Habibi, A., & Moharrami, H. (2010). Nonlinear sensitivity Hayward, A. (1988). Cheaper steel-concrete composite
analysis of reinforced concrete frames. Finite Elements bridges. In C. D. Buckner & I. Viest (Eds.), Composite
in Analysis and Design, 46(7), 571584. doi:10.1016/j. Construction in Steel and Concrete: Proceedings of an
finel.2010.02.005 Engineering Foundation Conference, New England Col-
lege, Henniker, New Hampshire, June 7-12, 1987 (pp.
Haftka, R. T., & Gurdal, Z. (1992). Elements of structural
194-206). New York, NY: American Society of Civil
optimization. Norwell, MA: Kluwer Academic Publishers.
Engineers.
Haldar, A. (Ed.). (2006). Recent developments in reliabil-
HAZUS-MH/MR3. (2007). FEMAs software program
ity-based civil engineering. Singapore: World Scientific
for estimating potential losses from disasters.
Publishing Co. Pte. Ltd.doi:10.1142/9789812707222
He, W. L., & Agrawal, A. K. (2008). Analytical model of
Haldar, A., & Mahadevan, S. (2000). Reliability assess-
ground motion pulses for the design and assessment of
ment using stochastic finite element analysis. USA: John
seismic protective systems. Journal of Structural Engi-
Wiley and Sons.
neering, 134(7), 11771188. doi:10.1061/(ASCE)0733-
Hanks, T. G., & McGuire, R. K. (1981). The character of 9445(2008)134:7(1177)
high frequency ground motions based on seismic shear
Housner, G. W., & Hudson, D. E. (1958). The Port
waves. Bulletin of the Seismological Society of America,
Hueneme earthquake of March 18, 1957. Bulletin of the
71, 20712095.
Seismological Society of America, 48, 163168.
Hart, G. C., & Wong, K. (2000). Structural dynamics for
Huang, B., & Du, X. (2007). Analytical robustness
structural engineers. New York, NY: John Wiley & Sons.
assessment for robust design. Structural and Multidis-
ciplinary Optimization, 34(2), 123127. doi:10.1007/
s00158-006-0068-0
410
Huang, X., & Xie, Y. M. (2007). Convergent and mesh- INPRES. (1995). Microzonificacin Ssmica del Gran
independent solutions for the bi-directional evolution- Mendoza. Instituto Nacional de Prevencin Ssmica,
ary structural optimization method. Finite Elements in Publicacin Tcnica N 19.
Analysis and Design, 43(14), 10391049. doi:10.1016/j.
INPRES-CIRSOC 103. (1995). Reglamento Argentino
finel.2007.06.006
para construcciones sismorresistentes, Parte II: Con-
Huang, X., & Xie, Y. M. (2008). Topology optimization strucciones de Hormign Armado, INTI.
of nonlinear structures under displacement loading. En-
International Building Code. (2006).
gineering Structures, 30(7), 20572068. doi:10.1016/j.
engstruct.2008.01.009 International Conference of Building Officials. (1997).
Uniform building code (UBC). Whittier, California, USA.
Huang, X., Xie, Y. M., & Lu, G. (2007). Topology
optimization of energy-absorbing structures. Interna- Iyengar, R. N. (1972). Worst inputs and a bound on the high-
tional Journal of Crashworthiness, 12(6), 663675. est peak statistics of a class of non-linear systems. Journal
doi:10.1080/13588260701497862 of Sound and Vibration, 25, 2937. doi:10.1016/0022-
460X(72)90593-7
Huang, X., Zuo, Z. H., & Xie, Y. M. (2010). Evolutionary
topological optimization of vibrating continuum structures Jalayer, F., & Cornell, C. A. (2009). Alternative non-linear
for natural frequencies. Computers & Structures, 88(56), demand estimation methods for probability-based seis-
357364. doi:10.1016/j.compstruc.2009.11.011 mic assessments. Earthquake Engineering & Structural
Dynamics, 38(8), 951972. doi:10.1002/eqe.876
Hung, S. L., Huang, C. S., Wen, C. M., & Hsu, Y. C.
(2003). Nonparametric identification of a building struc- Jangid, R. S. (2008). Stochastic response of bridges seis-
ture from experimental data using wavelet neural network. mically isolated by friction pendulum system. Journal
Computer-Aided Civil and Infrastructure Engineering, 18, of Bridge Engineering, 13(4), 319330. doi:10.1061/
356368. doi:10.1111/1467-8667.t01-1-00313 (ASCE)1084-0702(2008)13:4(319)
Hurlebaus, S., & Gaul, L. (2006). Smart structure dy- Jang, J. S. R., Sun, C. T., & Mizutani, E. (1997). Neuro-
namics. Mechanical Systems and Signal Processing, 20, fuzzy and soft computing. Upper Saddle River, NJ:
255281. doi:10.1016/j.ymssp.2005.08.025 Prentice Hall.
Hurtado, J. (2004). Structural reliability Statistical Jaynes, E. T. (2003). Probability theory: The logic of
learning perspectives. Lecture Notes in Applied and science. Cambridge, UK: Cambridge University Press.
Computational Mechanics, 17. Springer Verlag. doi:10.1017/CBO9780511790423
Hwang, H., Liu, J. B., & Chiu, Y.-H. (2001). Seismic Jensen, H. (2005). Structural optimization of linear dy-
fragility analysis of highway bridges. Tennessee, USA: namical systems under stochastic excitation: A moving re-
Center of Earthquake Research and Information, The liability database approach. Computer Methods in Applied
University of Memphis. Mechanics and Engineering, 194(12-16), 1757-1778.
Hwang, J.-S., & Tseng, Y.-S. (2005). Design formulations Jensen, H. (2002). Reliability-based optimization of
for supplemental viscous dampers to highway bridges. uncertain systems in structural dynamic. AIAA Journal,
Earthquake Engineering & Structural Dynamics, 34(13), 40(4), 731738. doi:10.2514/2.1705
16271642. doi:10.1002/eqe.508
Jensen, H. (2006). Structural optimization of non-linear
Hwang, K.-H., Lee, K.-W., & Park, G.-J. (2001). Robust systems under stochastic excitation. Probabilistic En-
optimization of an automobile rearview mirror for vibra- gineering Mechanics, 21(4), 397409. doi:10.1016/j.
tion reduction. Structural and Multidisciplinary Optimiza- probengmech.2006.02.002
tion, 21(4), 300308. doi:10.1007/s001580100107
411
Jensen, H., & Beer, M. (2010). Discrete-continuous vari- Jurecka, F., Ganser, M., & Bletzinger, K. U. (2007). Update
able structural optimization of systems under stochastic scheme for sequential spatial correlation approximations
loading. Structural Safety, 32(5), 293304. doi:10.1016/j. in robust design optimization. Computers & Structures,
strusafe.2010.03.007 85(10), 606614. doi:10.1016/j.compstruc.2006.08.075
Jensen, H., Valdebenito, M., Schuller, G., & Kusanovic, Kaisa, M. (1999). Nonlinear multiobjective optimization.
D. (2009). Reliability-based optimization of stochastic USA: Kluwer Academic Publishers.
systems using line search. Computer Methods in Applied
Kalkan, E., & Kunnath, S. K. (2008). Relevance of ab-
Mechanics and Engineering, 198(49-52), 39153924.
solute and relative energy content in seismic evaluation
doi:10.1016/j.cma.2009.08.016
of structures. Advances in Structural Engineering, 11(1),
Jensen, J. S. (2007). Topology optimization of dynam- 118. doi:10.1260/136943308784069469
ics problems with Pad approximants. International
Kalkan, E., & Luco, N. (Eds.). (2011). Earthquake
Journal for Numerical Methods in Engineering, 72(13),
ground motion selection and modification for nonlinear
16051630. doi:10.1002/nme.2065
dynamic analysis of structures. Journal of Structural
Jiang, X., & Adeli, H. (2005). Dynamic wavelet neural Engineering, 137(3), 277467. doi:10.1061/(ASCE)
network for nonlinear identification of highrise buildings. ST.1943-541X.0000355
Computer-Aided Civil and Infrastructure Engineering, 20,
Kang, B., Park, G. J., & Arora, J. S. (2006). A review of
316330. doi:10.1111/j.1467-8667.2005.00399.x
optimization of structures subjected to transient loads.
Jin, R., Chen, W., & Simpson, T. (2001). Comparative stud- Structural and Multidisciplinary Optimization, 31(2),
ies of metamodelling techniques under multiple modeling 8195. doi:10.1007/s00158-005-0575-4
criteria. Structural and Multidisciplinary Optimization,
Kang, S.-C., Koh, H.-M., & Choo, J. F. (2010). An
23(1), 113. doi:10.1007/s00158-001-0160-4
efficient response surface method using moving least
Jog, C. S. (2002). Topology design of structures subjected squares approximation for structural reliability analysis.
to periodic loading. Journal of Sound and Vibration, Probabilistic Engineering Mechanics, 25(4), 365371.
253(3), 687709. doi:10.1006/jsvi.2001.4075 doi:10.1016/j.probengmech.2010.04.002
Johansen, T. A. (1994). Fuzzy model based control: Stabil- Karlaftis, M. G., Kepaptsoglou, K. L., & Lampropoulos,
ity, robustness, and performance issues. IEEE Transactions S. (2007). Fund allocation for transportation network
on Fuzzy Systems, 2, 221234. doi:10.1109/91.298450 recovery following natural disasters. Journal of Urban
Planning and Development, 133(1), 8289. doi:10.1061/
Johansen, T. A., & Babuka, R. (2003). Multiobjective
(ASCE)0733-9488(2007)133:1(82)
identification of Takagi-Sugeno fuzzy models. IEEE
Transactions on Fuzzy Systems, 11, 847860. doi:10.1109/ Karnopp, D. C., Margolis, D. L., & Rosenberg, R. C.
TFUZZ.2003.819824 (2006). System dynamics, modeling and simulation of
mechatronic systems. John Wiley & Sons.
Jnsson, M. H., Bessason, B., & Haflidason, E. (2010).
Earthquake response of a base-isolated bridge subjected Karush, W. (1939). Minima of functions of several
to strong near-fault ground motion. Soil Dynamics and variables with inequalities as side constraints. Masters
Earthquake Engineering, 30(6), 447455. doi:10.1016/j. thesis, Department of Mathematics, University of Chicago,
soildyn.2010.01.001 Chicago, Illinois.
Jung, D., & Gea, H. C. (2004). Topology optimization of Kennedy, J., & Eberhart, R. (1995), Particle swarm
nonlinear structures. Finite Elements in Analysis and De- optimization. IEEE International Conference on Neural
sign, 40(11), 14171427. doi:10.1016/j.finel.2003.08.011 Networks, Piscataway, NJ, (pp. 19421948).
412
Kerber, F., Hurlebaus, S., Beadle, B. M., & Stbener, U. Kim, Y., Langari, R., & Hurlebaus, S. (2010). Model-
(2007). Control concepts for an active vibration isolation based multi-input, multi-output supervisory semiactive
system. Mechanical Systems and Signal Processing, 21, nonlinear fuzzy controller. Computer-Aided Civil and
30423059. doi:10.1016/j.ymssp.2007.04.003 Infrastructure Engineering, 25, 387393. doi:10.1111/
j.1467-8667.2009.00649.x
Kharmanda, G., Olhoff, N., Mohamed, A., & Lemaire,
M. (2004). Reliability-based topology optimization. Kiureghian, A. D., & Crempien, J. (1989). An evolutionary
Structural and Multidisciplinary Optimization, 26(5), model for earthquake ground motion. Structural Safety,
295307. doi:10.1007/s00158-003-0322-7 6, 235246. doi:10.1016/0167-4730(89)90024-6
Khashaee, P. (2004). Damage-based seismic design Klee, H. (2007). Simulation of dynamic systems with
of structures. Earthquake Spectra, 21(2), 371387. MATLAB and SIMULINK. Boca Raton, FL: CRC Press.
doi:10.1193/1.1896366
Kleiber, M., & Hien, T. D. (1992). The stochastic finite
Kim, C., Wang, S., & Choi, K. K. (2005). Efficient re- element method. Chichester, UK: John Wiley and Sons.
sponse surface modelling by using moving least-squares
Kleinmann, N. L., Spall, J. C., & Naiman, D. C. (1999).
method and sensitivity. AIAA Journal, 43(1), 24042411.
Simulation-based optimization with stochastic approxima-
doi:10.2514/1.12366
tion using common random numbers. Management Sci-
Kim, H. S., & Roschke, P. N. (2006). Design of fuzzy ence, 45(11), 15701578. doi:10.1287/mnsc.45.11.1570
logic controller for smart base isolation system using
Kocer, F. Y., & Arora, J. S. (1999). Optimal design of
genetic algorithm. Engineering Structures, 28, 8496.
H-frame transmission poles for earthquake loading.
doi:10.1016/j.engstruct.2005.07.006
Kim, S. B., Yun, C. B., & Spencer, B. F. Jr. (2004). doi:10.1061/(ASCE)0733-9445(1999)125:11(1299)
Vibration control of wind-excited tall buildings using
Kocer, F. Y., & Arora, J. S. (2002). Optimal design of
sliding mode fuzzy control. Journal of Engineering
latticed towers subjected to earthquake loading. Journal
of Structural Engineering, 128(2), 197204. doi:10.1061/
9399(2004)130:4(505)
(ASCE)0733-9445(2002)128:2(197)
Kim, Y., Hurlebaus, S., & Langari, R. (2010). Control of
Koh, C. G., Chen, Y. F., & Liaw, C.-Y. (2003). A hybrid
a seismically excited benchmark building using linear
computational strategy for identification of structural
matrix inequality-based semiactive nonlinear fuzzy con-
parameters. Computers & Structures, 81(2), 107117.
trol. Journal of Structural Engineering, 136, 10231026.
doi:10.1016/S0045-7949(02)00344-9
doi:10.1061/(ASCE)ST.1943-541X.0000192
Kong, J. S., Ababneh, A. N., Frangopol, D. M., & Xi,
Kim, Y.-I., & Park, G. J. (2010). Nonlinear dynamic
Y. (2002). Reliability analysis of chloride penetration in
response structural optimization using equivalent
saturated concrete. Journal of Probabilistic Engineer-
static loads. Computer Methods in Applied Mechanics
ing Mechanics, 17(3), 302315. doi:10.1016/S0266-
and Engineering, 199(9-12), 660676. doi:10.1016/j.
8920(02)00014-0
cma.2009.10.014
Kosaka, I., & Swan, C. C. (1999). A symmetry reduction
Kim, Y., Langari, R., & Hurlebaus, S. (2009). Semiactive
method for continuum structural topology optimization.
nonlinear control of a building using a magnetorheological
Computers & Structures, 70(1), 4761. doi:10.1016/
damper system. Mechanical Systems and Signal Process-
S0045-7949(98)00158-8
ing, 23, 300315. doi:10.1016/j.ymssp.2008.06.006
Kovcs, L. (1980). Combinatorial methods of discrete
programming. Mathematical Methods of Operation
Research, 2.
413
Kramer, S. L. (2003). Geotechnical earthquake engineer- Lagaros, N. D., Naziris, I. A., & Papadrakakis, M. (2010).
ing. Prentice Hall. The influence of masonry infill walla in the Framework of
the performance-based design. Journal of Earthquake En-
Krawinker, H. (1994). Static pushover analysis. Proceed-
gineering, 14, 5779. doi:10.1080/13632460902988976
ings SEAONC 1994 Fall Seminar on the Developing Art
of Seismic Engineering. San Francisco, CA. Lagaros, N. D., & Papadrakakis, M. (2007). Robust
seismic design optimization of steel structures. Structural
Kuhn, H. W., & Tucker, A. W. (1951). Nonlinear program-
ming. In Proceedings of 2nd Berkeley Symposium (pp.
doi:10.1007/s00158-006-0047-5
481492). Berkeley, CA: University of California Press.
Lagaros, N. D., & Papadrakakis, M. (2007). Seismic
Kunde, M. C., & Jangid, R. S. (2003). Seismic behavior
design of RC structures: A critical assessment in the
of isolated bridges: A-state-of-the-art review. Electronic
framework of multi-objective optimization. Earthquake
Journal of Structural Engineering, 3, 140170.
Engineering & Structural Dynamics, 36(12), 16231639.
Kurdi, M. H., Haftka, R. T., Schmitz, T. L., & Mann, doi:10.1002/eqe.707
B. P. (2008). A robust semi-analytical method for cal-
Lagaros, N. D., & Papadrakakis, M. (2011). (in press).
culating the response sensitivity of a time delay sys-
Neural network based prediction schemes of the non-
tem. Journal of Vibration and Acoustics, 130(6), 16.
linear seismic response of 3D buildings. Advances in
doi:10.1115/1.2981093
Engineering Software.
Kushner, H. J., & Yin, G. G. (2003). Stochastic approxi-
Lagaros, N., Garavela, A. T., & Papadrakakis, M. (2008).
mation and recursive algorithms and applications. New
Innovative seismic design optimization with reliability
York, NY: Springer Verlag.
constraints. Computer Methods in Applied Mechanics and
Kuwamura, H., & Kato, B. (1989). Effects of randomness Engineering, 198, 2841. doi:10.1016/j.cma.2007.12.025
in structural members yield strength on the structural sys-
Langari, R. (1999). Past, present and future of fuzzy con-
tems ductility. Journal of Constructional Steel Research,
trol: A case for application of fuzzy logic in hierarchical
13(23), 7993. doi:10.1016/0143-974X(89)90007-2
control. Proceedings, 18th International Conference of the
Lagaros, N. D., Fragiadakis, M., & Papadrakakis, M. North American Fuzzy Information Processing Society-
(2004). Optimum design of shell structures with stiffening NAFIPS, New York City, New York, (pp.760-765).
beams. AIAA Journal, 42(1), 175184. doi:10.2514/1.9041
Lawler, E. L., Lenstra, J. K., Rinnooy Kan, A. H. G., &
Lagaros, N. D., Fragiadakis, M., Papadrakakis, M., Shmoys, D. B. (1985). The traveling salesman problem:
& Tsompanakis, Y. (2006). Structural optimization: A A guided tour of combinatorial optimization. New York,
tool for evaluating dynamic design procedures. Engi- NY: Wiley.
neering Structures, 28(12), 16231633. doi:10.1016/j.
Lee, H. S., Tomosawa, F., & Noguchi, T. (1996). Effect
of rebar corrosion on the structural performance of single
Lagaros, N. D., Garavelas, A. T., & Papadrakakis, M. reinforced beams. In Lacasse, M. A., & Vanier, D. J.
(2008). Innovative seismic design optimization with (Eds.), Durability of building material and components
reliability constraints. Computer Methods in Applied (pp. 571580). London, UK: E & FN Spon.
Mechanics and Engineering, 198(1), 2841. doi:10.1016/j.
Lee, K., & Park, G. (2001). Robust optimization consider-
cma.2007.12.025
ing tolerances of design variable. Computers & Structures,
Lagaros, N. D., & Karlaftis, M. G. (2011). A critical as- 79(1), 7786. doi:10.1016/S0045-7949(00)00117-6
sessment of metaheuristics for scheduling emergency in-
Lewis, A. S. (2002). Robust regularization. Technical
frastructure inspections. Swarm and Evolutionary Compu-
Report. Columbia: Simon Fraser University.
tation, 1(3), 147163. doi:10.1016/j.swevo.2011.06.002
414
Li, D. (1997). Multi-objective optimum design of static and Liu, M., Burns, S. A., & Wen, Y. K. (2005). Multiobjective
seismic-resistant structures with genetic algorithm, fuzzy optimization for performance-based seismic design of
logic and game theory. Doctoral dissertation, University steel moment frame structures. Earthquake Engineering
of Missouri-Rolla, Rolla, USA. & Structural Dynamics, 34(3), 289306. doi:10.1002/
eqe.426
Li, G. (1998). Reliability and performance based optimal
design for seismic high-rising structures. Ph.D. Disserta- Liu, M., Burns, S. A., & Wen, Y. K. (2006). Genetic al-
tion, Dalian University of Technology, China. gorithm based construction-conscious minimum weight
design of seismic steel moment-resisting frames. Journal
Liang, Q. Q. (2005). Performance-based optimi-
of Structural Engineering, 132(1), 5058. doi:10.1061/
zation of structures. New York, NY: Spon Press.
(ASCE)0733-9445(2006)132:1(50)
doi:10.4324/9780203334713
Liu, M., Wen, Y. K., & Burns, S. A. (2004). Life cycle
Li, G., Zhou, R.-G., Duan, L., & Chen, W.-F. (1999). Mul-
cost oriented seismic design optimization of steel mo-
tiobjective and multilevel optimization for steel frames.
ment frame structures with risk-taking preference. En-
Engineering Structures, 21(6), 519529. doi:10.1016/
gineering Structures, 26(10), 14071421. doi:10.1016/j.
S0141-0296(97)00226-5
Lin, J. W., & Betti, R. (2004). On-line identification
Loh, C. H., Wu, W. Y., & Lin, P. Y. (2003). Displacement
and damage detection in non-linear structural systems
control of isolated structures with semi-active control
using a variable forgetting factor approach. Earthquake
devices. Journal of Structural Control, 10, 77100.
Engineering & Structural Dynamics, 33, 419444.
doi:10.1002/stc.18
doi:10.1002/eqe.350
Lui, E. M., & Ge, M. (2005). Structural damage iden-
Lin, J. W., Betti, R., Smyth, A. W., & Longman, R.
tification using system dynamic properties. Computers
W. (2001). On-line identification of non-linear hyster-
& Structures, 83(27), 21852196. doi:10.1016/j.comp-
etic structural systems using a variable trace approach.
struc.2005.05.002
Earthquake Engineering & Structural Dynamics, 30,
12791303. doi:10.1002/eqe.63 Lund, E. (1994). Finite element based design sensitivity
analysis and optimization. Unpublished doctoral dis-
Liu, M. (2003). Development of multi-objective optimiza-
sertation, Institute of Mechanical Engineering, Aalborg
tion procedures for seismic design of steel moment frame
University, Denmark.
structures. Doctoral dissertation, University of Illinois at
Urbana Champaign, Urbana, USA. Lutes, L. D., & Sarkani, S. (1997). Stochastic analysis
of structural and mechanical vibrations. Upper Saddle
Liu, W. Y., Xiao, C. J., Wang, B. W., Shi, Y., & Fang,
River, NJ: Prentice Hall.
S. F. (2003). Study on combining subtractive clustering
with fuzzy c-means clustering. Proceedings of Second Maeda, Y., Nishiwaki, S., Izui, K., Yoshimura, M., Matsui,
International Conference on Machine Learning and K., & Terada, K. (2006). Structural topology optimization
Cybernetics, Xian, China, (pp. 2659-2662). of vibrating structures with specified eigenfrequencies and
eigenmode shapes. International Journal for Numerical
Liu, M. (2005). Seismic design of steel moment-resisting
Methods in Engineering, 67(5), 597628. doi:10.1002/
frame structures using multiobjective optimization. Earth-
nme.1626
quake Spectra, 21(2), 389414. doi:10.1193/1.1902952
Mahin, S. A., & Bertero, V. V. (1981). An evaluation of
Liu, M., Burns, S. A., & Wen, Y. K. (2003). Optimal
inelastic seismic design spectra. Journal of the Structural
seismic design of steel frame buildings based on life cycle
Division, 107(ST9), 17771795.
cost considerations. Earthquake Engineering & Structural
Dynamics, 32(9), 13131332. doi:10.1002/eqe.273
415
Makris, N., & Zhang, J. (2004). Seismic response analysis Martinez, W. L., & Martinez, A. R. (2002). Computational
of a highway overcrossing equipped with elastomeric Statistics Handbook with MATLAB. Boca Raton, FL:
bearings and fluid dampers. Journal of Structural En- Chapman & Hall/CRC.
gineering, 130(6), 830845. doi:10.1061/(ASCE)0733-
Martin-Perez, B., Pantzopoulou, S. J., & Thomas, M. D. A.
9445(2004)130:6(830)
(2001). Numerical solution of mass transport equations in
Mander, J. B., & Basoz, N. (1999). Seismic fragility concrete structures. Journal of Computers and Structures,
curve theory for highway bridges. Proceedings of the 79(13), 12511264. doi:10.1016/S0045-7949(01)00018-9
5th US Conference on Earthquake Engineering TCLEE
Masri, S. F., Smyth, A. W., Chassiakos, A. G., Caughey, T.
No.16, (pp. 31-40).
K., & Hunter, N. F. (2000). Application of neural networks
Mander, J. B., Priestly, M. J. N., & Park, R. (1988). for detection of changes in nonlinear systems. Journal
Theoretical stress-strain model for confined concrete. of Engineering Mechanics, 126, 666676. doi:10.1061/
Journal of the Structural Division, 114(8), 18041826. (ASCE)0733-9399(2000)126:7(666)
doi:10.1061/(ASCE)0733-9445(1988)114:8(1804)
Masson, G., Cogan, S., Bouhaddi, N., Lombard, J. P.,
Manders, E. J., Biswas, G., Mahadevan, N., & Karsai, G. & Bonini, J. (2002). Parameterized reduced models for
(2006). Component-oriented modeling of hybrid dynamic efficient optimization of structural dynamic behavior.
systems using the generic modeling environment. Pro- Collection of Technical Papers - AIAA/ASME/ASCE/
ceedings of 4th Workshop on Model-Based Development AHS/ASC Structures, Structural Dynamics and Materials
of Computer Based Systems. Conference, (pp. 1434-1440).
Manoharan, S., & Shanmuganathan, S. (1999). A com- Mathworks. (1999). Simulink: Dynamic system simulation
parison of search mechanisms for structural optimization. for Matlab. Mathworks Inc.
Computers & Structures, 73(1-5), 363-372.
Mavroeidis, G. P., & Papageorgiou, A. P. (2003). A math-
Marano, G. C., Greco, R., & Sgobba, S. (2010). A compari- ematical representation of near-fault ground motions.
son between different robust optimal design approaches: Bulletin of the Seismological Society of America, 93(3),
Application to tuned mass dampers. Probabilistic En- 10991131. doi:10.1785/0120020100
gineering Mechanics, 25(1), 108118. doi:10.1016/j.
Mavroeidis, G., & Papageorgiou, A. (2003). A mathemati-
probengmech.2009.08.004
cal representation of near-fault ground motions. Bulletin of
Marano, G. C., Sgobba, S., Greco, R., & Mezzina, M. the Seismological Society of America, 93(3), 10991131.
(2008). Robust optimum design of tuned mass damp- doi:10.1785/0120020100
ers devices in random vibrations mitigation. Journal of
Ma, Z. D., Cheng, H. C., & Kikuchi, N. (1995). Topologi-
Sound and Vibration, 313(3-5), 472492. doi:10.1016/j.
cal design for vibrating structures. Computer Methods in
jsv.2007.12.020
Applied Mechanics and Engineering, 121(14), 259280.
Marano, G. C., Trentadue, F., & Greco, R. (2006). Optimal doi:10.1016/0045-7825(94)00714-X
design criteria for elastic structures subject to random dy-
Ma, Z. D., Kikuchi, N., & Hagiwara, I. (1993). Structural
namic loads. Engineering Optimization, 38(7), 853871.
topology and shape optimization for a frequency response
doi:10.1080/03052150600913028
problem. Computational Mechanics, 13(3), 157174.
Marler, R. T., & Arora, J. S. (2004). Survey of multi- doi:10.1007/BF00370133
objective optimization methods for engineering. Structural
McGee, R. (1999). Modeling of durability performance of
Tasmanian bridges. Proceedings of the ICASP8 Confer-
doi:10.1007/s00158-003-0368-6
ence, (pp. 297-306). Sydney, Australia.
416
McGuire, R. K. (1995). Probabilistic seismic hazard Mitropoulou, C., Lagaros, N. D., & Papadrakakis, M.
analysis and design earthquake: Closing the loop. Bulletin (2010). Building design based on energy dissipation: A
of the Seismological Society of America, 85, 12751284. critical assessment. Bulletin of Earthquake Engineering,
8(6), 13751396. doi:10.1007/s10518-010-9182-x
Mehanny, S. S., & Deierlein, G. G. (2000). Modeling of
assessment of seismic performance of composite frames Moehle, J. P., & Mahin, S. A. (1991). Observations on the
with reinforced concrete columns and steel beams. Report behavior of reinforced concrete buildings during earth-
No. 135, The John Blume Earthquake Research Center, quakes. American Concrete Institute SP-127. In Ghosh,
Stanford University. S. K. (Ed.), Earthquake-resistant concrete structures
Inelastic response and design.
Memari, A. M., & Madhkhan, M. (1999). Optimal design
of steel frames subject to gravity and seismic codes pre- Moharrami, H. (1993). Design optimization of reinforced
scribed lateral forces. Structural and Multidisciplinary concrete frameworks. Ph.D. Thesis, The University of
Optimization, 18(1), 5666. Waterloo, Waterloo, Ontario, Canada
Mendonca, D., Beroggi, G. E. G., van Gent, D., & Moharrami, H. (2006). Optimality criteria: A robust
Wallace, W. A. (2006). Designing gaming simulations nonlinear optimization algorithm. The 7th International
for the assessment of Group decision support Systems Congress on Civil Engineering, Tarbiat Modarres Uni-
in emergency response. Safety Science, 44, 523535. versity, Tehran, Iran.
doi:10.1016/j.ssci.2005.12.006
Moharrami, H., & Alavinasab, A. (2006). An optimiza-
Mendonca, D., Beroggi, G. E. G., & Wallace, W. A. tion procedure for automated design of seismic resistant
(2001). Decision support for improvisation during steel frames. International Journal of Civil Engineering,
emergency response operations. International Journal 4(2), 86105.
of Emergency Management, 1(1), 3038. doi:10.1504/
Moharrami, H., & Grierson, D. E. (1993). Computer-
IJEM.2001.000507
automated design of reinforced concrete frameworks.
Michell, A. G. M. (1904). The limits of economy of Journal of Structural Engineering, 119(7), 20362058.
material in frame structures. Philosophical Magazine, doi:10.1061/(ASCE)0733-9445(1993)119:7(2036)
6(8), 589597.
Mokkadem, A. (1989). Estimation of the entropy and
Mills-Curran, W. C., & Schmit, L. A. (1985). Structural information for absolutely continuous random variables.
optimization with dynamic behaviour constraints. AIAA IEEE Transactions on Information Theory, 35, 193196.
Journal, 23(1), 132138. doi:10.2514/3.8881 doi:10.1109/18.42194
Ministry of Transport of the Peoples Republic of China. Mller, O. (2001). Metodologa para evaluacin de la
(1989). Specifications for earthquake resistant design probabilidad de falla de estructuras sismorresistentes y
in highway Engineering (JTJ 004-89). Beijing, China: calibracin de cdigos. Tesis de Doctorado en Ingeniera,
Peoples Communication Press. Universidad Nacional de Rosario.
Ministry of Transport of the Peoples Republic of China. Mller, O., & Foschi, R. (2003). Reliability evaluation in
(2008). Guidelines for seismic design of highway bridges seismic design: A response surface methodology. Earth-
(JTG/T B02-01-2008). Beijing, China: Peoples Com- quake Spectra, 19(3), 579603. doi:10.1193/1.1598200
munication Press.
Mller, O., Foschi, R., Quiroz, L., & Rubinstein, M. (2008).
Min, S., Kikuchi, N., Park, Y. C., Kim, S., & Chang, S. Optimizacin de prticos con acciones ssmicas: diferen-
(1999). Optimal topology design of structures under dy- tes estrategias numricas utilizando redes neuronales.
namic loads. Structural Optimization, 17(23), 208218. [AMCA.]. Mecnica Computacional, 28, 25832603.
417
Mller, O., Foschi, R., Quiroz, L., & Rubinstein, M. Moustafa, A. (2002). Deterministic/reliability-based
(2009a). Performance-based seismic optimization imple- critical earthquake load models for linear/nonlinear
menting neural networks. In Frangopol, D. M. (Ed.), engineering structures. Doctoral dissertation, Depart-
Computational structural dynamics and earthquake ment of Civil Engineering, Indian Institute of Science,
engineering, structures and infrastructures series (Vol. Bangalore, India.
II, pp. 547564). Taylor & Francis Group.
Moustafa, A., Takewaki, I., & Wijeyewickrema, A. K.
Mller, O., Foschi, R., Quiroz, L., & Rubinstein, M. (2010). Selection of earthquake records for time-history
(2009c). Structural optimization for performance- analysis of structures. Joint 7th International Conference
based design in earthquake engineering: Applications on Urban Earthquake Engineering and 5th International
of neural networks. Structural Safety, 31(6), 490499. Conference on Earthquake Engineering, 3-5 March,
doi:10.1016/j.strusafe.2009.06.007 Tokyo, Japan.
Mller, O., Foschi, R., Rubinstein, M., & Quiroz, L. Moustafa, A. (2009). Critical earthquake load inputs for
(2006). Momento-curvatura de secciones de hormign multi-degree-of-freedom inelastic structures. Journal
armado sismorresistentes utilizando redes neuronales. of Sound and Vibration, 325, 532544. doi:10.1016/j.
[AMCA.]. Mecnica Computacional, 25, 21452162. jsv.2009.03.022
Mller, O., Foschi, R., Rubinstein, M., & Quiroz, L. (2007). Moustafa, A. (2009). Discussion of a new approach of
Optimizacin de prticos sismorresistentes utilizando selecting real input ground motions for seismic design:
redes neuronales y algoritmo sin clculo de gradientes. The most unfavourable real seismic design ground mo-
[AMCA.]. Mecnica Computacional, 26, 18241839. tions. Earthquake Engineering & Structural Dynamics,
38, 11431149. doi:10.1002/eqe.885
Mller, O., Foschi, R., Rubinstein, M., & Quiroz, L.
(2009b). Seismic structural reliability using different Moustafa, A. (2010). Closure to discussion of critical
nonlinear dynamic response surface approximations. earthquake load inputs for multi-degree-of-freedom in-
Structural Safety, 31(5), 432442. doi:10.1016/j.stru- elastic structures. Journal of Sound and Vibration, 330,
safe.2008.12.001 356360. doi:10.1016/j.jsv.2010.09.002
Moon, F. L., & Aktan, A. E. (2006). Impacts of epistemic Moustafa, A. (2010). Discussion of analytical model
(bias) uncertainty on structural identification of con- of ground motion pulses for the design and assessment
structed (civil) systems. The Shock and Vibration Digest, of seismic protective systems. Journal of Structural
38, 399420. doi:10.1177/0583102406068068 Engineering, 137(1), 229230. doi:10.1061/(ASCE)
ST.1943-541X.134
Moore, R. E. (1979). Methods and applications of interval
analysis. Philadelphia, PA: SIAM Bookmart, PA. Moustafa, A. (2010). Identification of resonant earth-
quake ground motion. Indian Academy of Sciences,
Morinaga, S. (1996). Remaining life of reinforced concrete
35(3), 355371.
structures after corrosion cracking. Durability of Build-
ing Material and Components (pp. 127136). London, Moustafa, A. (2011). Damage-based design earthquake
UK: E & FN Spon. loads for SDOF inelastic structures. Journal of Structural
Mosterman, P. J., & Biswas, G. (1999). Diagnosis of
ST.1943-541X.0000074
continuous valued systems in transient operating regions.
IEEE Transactions, Man, and Cybernetics-Part A, 29(6), Moustafa, A., & Mahadevan, S. (2010). Probabilistic
554565. doi:10.1109/3468.798059 critical earthquake excitations using earthquake response
spectra. [Building and Housing]. Asian Journal of Civil
Engineering, 11(3), 295319.
418
Moustafa, A., Mahadevan, S., Daigle, M., & Biswas, D. Nigam, N. C. (1972). Structural optimization in random
(2010). Structural and sensor system damage identifica- vibration environment. AIAA Journal, 10(4), 551553.
tion using the bond graph approach. Structural Control doi:10.2514/3.50151
and Health Monitoring, 17, 178197. doi:10.1002/stc.285
Oh, S. H., Kimb, Y. J., & Ryu, H. S. (2009). Seismic
Moustafa, A., & Takewaki, I. (2010). Characterization and performance of steel structures with slit dampers. En-
modeling of near-fault pulse-like strong ground motion gineering Structures, 31(9), 19972008. doi:10.1016/j.
via damage-based critical excitation method. Structural engstruct.2009.03.003
Engineering & Mechanics, 34(6), 755778.
Ohsaki, M., Kinoshita, T., & Pan, P. (2007). Multiobjective
Moustafa, A., & Takewaki, I. (2010). Deterministic and heuristic approaches to seismic design of steel frames with
probabilistic representation of near-field pulse-like ground standard sections. Earthquake Engineering & Structural
motion. Soil Dynamics and Earthquake Engineering, 30, Dynamics, 36(11), 14811495. doi:10.1002/eqe.690
412422. doi:10.1016/j.soildyn.2009.12.013
Olhoff, N., & Du, J. (2005). Topological design of struc-
Moustafa, A., & Takewaki, I. (2010). Modeling critical tures subjected to forced vibration. In J. Herskovits (Ed.),
strong ground motion sequences on inelastic structures. Proceedings of the 6th World Congress of Structural and
Advances in Structural Engineering, 13(4), 665679. Multidisciplinary Optimization, Rio de Janeiro, Brazil,
doi:10.1260/1369-4332.13.4.665 30 May03 June.
Moustafa, A., & Takewaki, I. (2011). Response of nonlin- Olhoff, N., & Rasmussen, S. H. (1977). On single and
ear SDOF structures to random acceleration sequences. bimodal optimum buckling loads of clamped columns.
Engineering Structures, 33(4), 12511258. doi:10.1016/j. International Journal of Solids and Structures, 13(7),
engstruct.2011.01.002 605614. doi:10.1016/0020-7683(77)90043-9
Moustafa, A., Ueno, K., & Takewaki, I. (2010). Critical OpenSees Development Team. (2009). OpenSees: Open
earthquake loads for SDOF inelastic structures considering system for earthquake engineering simulation. Berkeley,
evolution of seismic waves. Earthquakes and Structures, CA, USA.
1(2), 147162.
Otani, S. (1981). Hysteretic models of reinforced concrete
Muthukumar, S., & DesRoches, R. (2006). A Hertz contact for earthquake response analysis. Journal of the Faculty
model with non-linear damping for pounding simulation. of Engineering. University of Tokyo, 36(2), 407441.
Overton, M. L. (1992). Large-scale optimization of ei-
811828. doi:10.1002/eqe.557
genvalues. SIAM Journal on Optimization, 2(1), 88120.
Myers, R. H., & Montogmery, D. C. (1995). Response doi:10.1137/0802007
surface methodology: Process and product optimization
Owen, D. R. J., & Hinton, E. (1980). Finite elements in
using designed experiments. USA: John Wiley and Sons.
plasticity- Theory and practice. Swansea, UK: Pineridge
Nakashima, M., Saburi, K., & Tsuji, B. (1996). En- Press.
ergy input and dissipation behavior of structures with
Pacific Earthquake Research (PEER) Center. (2011).
hysteretic dampers. Earthquake Engineering & Struc-
Retrieved from http://peer.berkeley.edu/peer_ground_mo-
tural Dynamics, 25, 483496. doi:10.1002/(SICI)1096-
tion_database
9845(199605)25:5<483::AID-EQE564>3.0.CO;2-K
Papadrakakis, M., Charmpis, D. C., Lagaros, N. D., &
National Standard of the Peoples Republic of China.
Tsompanakis, Y. (Eds.). (2009). Computational structural
(2001). Chinese code for seismic design buildings
dynamics and earthquake engineering. London, UK:
(GB50011-2001). Beijing, China: New World Press.
Taylor & Francis.
Newmark, N. M., & Hall, W. J. (1982). Earthquake
spectra and design. Monograph, Earthquake Engineering
Research Institute. EERI.
419
Papadrakakis, M., & Lagaros, N. (2002). Reliability-based Park, Y. J., Ang, A. H.-S., & Wen, Y. K. (1985). Seismic
structural optimization using neural networks and Monte damage analysis of reinforced concrete buildings. Journal
Carlo simulation. Computer Methods in Applied Me- of Structural Engineering, 111(4), 740757. doi:10.1061/
chanics and Engineering, 191, 34913507. doi:10.1016/ (ASCE)0733-9445(1985)111:4(740)
S0045-7825(02)00287-6
Park, Y. J., Ang, A. H.-S., & Wen, Y. K. (1987). Damage-
Papadrakakis, M., Lagaros, N. D., & Plevris, V. (2005). limiting aseismic design of buildings. Earthquake Spectra,
Design optimization of steel structures considering un- 3(1), 126. doi:10.1193/1.1585416
certainties. Engineering Structures, 27(9), 14081418.
Paulay, T., & Priestley, M. J. N. (1992). Seismic design of
doi:10.1016/j.engstruct.2005.04.002
reinforced concrete and masonry buildings. John Wiley
Park, Y. J., Reinhorn, A. M., & Kunnath, S. K. (1987). & Sons, Inc. doi:10.1002/9780470172841
IDARC: Inelastic damage analysis of frame sheare-wall
Paynter, H. M. (1961). Analysis and design of engineering
structures. Technical Report NCEER-87-0008, National
systems. MIT Press.
Center for Earthquake Engineering Research, State Uni-
versity of New York at Buffalo. Pecker, A. (Ed.). (2007). Advanced earthquake engineer-
ing analysis. Udine, Italy: Springer. doi:10.1007/978-3-
Park, G. J., & Kang, B. S. (2003). Validation of a
211-74214-3
structural optimization algorithm transforming dy-
namic loads into equivalent static loads. Journal of Pedersen, N. L. (2000). Maximization of eigenvalues using
Optimization Theory and Applications, 118(1), 191200. topology optimization. Structural and Multidisciplinary
doi:10.1023/A:1024799727258 Optimization, 20(1), 211. doi:10.1007/s001580050130
Park, G. J., Lee, T. H., Lee, K., & Hwang, K. H. (2006). Pedrycz, W. (1984). An identification algorithm in fuzzy
Robust design: An overview. AIAA Journal, 44(1), relational systems. Fuzzy Sets and Systems, 13, 153167.
181191. doi:10.2514/1.13639 doi:10.1016/0165-0114(84)90015-0
Park, K. J., Lee, J. N., & Park, G. J. (2005). Structural PEER. (2005). Pacific Earthquake Engineering Research
shape optimization using equivalent static loads trans- (PEER) center: NGA database. Retrieved January 1, 2009,
formed from dynamic loads. International Journal for from http://peer.berkeley.edu/nga/
Numerical Methods in Engineering, 63(4), 589602.
doi:10.1002/nme.1295 Peizhuangm, W., Xihui, L., & Sanchez, E. (1986).
Set-valued statistics and its application to earthquake
Park, Y. J., & Ang, A. H. S. (1985). Mechanistic seis- engineering. Fuzzy Sets and Systems, 18(3), 347356.
mic damage model for reinforced concrete. Journal of doi:10.1016/0165-0114(86)90011-4
Structural Engineering, 111(4), 722739. doi:10.1061/
(ASCE)0733-9445(1985)111:4(722) Penelis, G. G., & Kappos, A. J. (1997). Earthquake-
resistant concrete structures. London, UK: E & FN SPON.
Park, Y. J., Ang, A. H. S., & Wen, Y. K. (1985). Seismic
damage analysis of reinforced concrete buildings. Journal Prez Lpez, J. R. (2005). Contribucin a los mtodos
of Structural Engineering, 111(4), 740757. doi:10.1061/ de optimizacin basados en procesos naturales y su
(ASCE)0733-9445(1985)111:4(740) aplicacin a la medida de antenas en campo prximo.
Retrieved from http://www.tesisenred.net/TDR-0305107-
Park, Y. J., & Ang, A. H.-S. (1985). Mechanistic seis- 180847
mic damage model for reinforced concrete. Journal of
Structural Engineering, 111(4), 722739. doi:10.1061/ Perez, R. E., & Behdinan, K. (2007). Particle swarm
(ASCE)0733-9445(1985)111:4(722) approach for structural design optimization. Comput-
ers & Structures, 85, 15791588. doi:10.1016/j.comp-
struc.2006.10.013
420
Perros, K., & Papadimitriou, C. (2009). Reliability analysis Prager, W. (1969). Optimality criteria derived from clas-
of bridge models with elastomeric bearings and seismic sical extermum principles. Technical report, SM Studies
stoppers under stochastic earthquake excitations. Paper Series. Ontario, Canada: Solid Mechanics Division,
presented at the 2009 ECCOMAS Thematic Conference University of Waterloo.
on Computational Methods in Structural Dynamics and
Prager, W. (1974). A note on discretized Michell structures.
Earthquake Engineering.
Computer Methods in Applied Mechanics and Engineer-
Pezeshk, S. (1998). Design of framed structures: An inte- ing, 3(3), 349355. doi:10.1016/0045-7825(74)90019-X
grated non-linear analysis and optimal minimum weight
Prasad, B. (1983). Approximation, adaptation and
design. International Journal for Numerical Methods in
automation concepts for large scale structural opti-
Engineering, 41(3), 459471. doi:10.1002/(SICI)1097-
mization. Engineering Optimization, 6(3), 129140.
0207(19980215)41:3<459::AID-NME293>3.0.CO;2-D
doi:10.1080/03052158308902462
Philippacopoulos, A. J., & Wang, P. C. (1984). Seismic
Priestely, M. J. N., Calvi, G. M., & Kowalsky, M. J.
inputs for nonlinear structures. Journal of Engineering
(2007). Displacement-based seismic design of structures.
Pavia, Italy: IUSS Press.
9399(1984)110:5(828)
Priestley, M. J. N., Seible, F., & Calvi, G. M. (1996).
Pinho, R. (2007). Non linear dynamic analysis of struc-
Seismic design and retrofit of bridges. New York, NY:
tures subjected to seismic action. In Pecker, A. (Ed.),
John Wiley & Sons, Inc.doi:10.1002/9780470172858
Advanced earthquake engineering analysis. New York,
NY: Springer Wien. doi:10.1007/978-3-211-74214-3_5 Priestly, M. J. N., Seible, F., & Calvi, G. M. (1996).
Seismic design and retrofit of bridges. USA: John Wiley
Plevris, V. (2009). Innovative computational techniques
and Sons Inc.doi:10.1002/9780470172858
for the optimum structural design considering uncertain-
ties. Ph. D. dissertation, Institute of Structural Analysis Quek, S. T., Teo, Y. P., & Balendra, T. (1990). Non-
and Seismic Research, National Technical University of stationary structural response with evolutionary spectra
Athens, June 2009. using seismological input model. Earthquake Engineer-
ing & Structural Dynamics, 19, 275288. doi:10.1002/
Plevris, V., Karlaftis, M. G., & Lagaros, N. D. (2010). A
eqe.4290190210
swarm intelligence approach for emergency infrastructure
inspection scheduling. In Gopalakrishnan, K., & Peeta, Querin, O. M. (1997). Evolutionary structural optimisa-
S. (Eds.), Sustainable and resilient critical infrastructure tion stress based formulation and implementation. Unpub-
systems: Simulation, modeling, and intelligent engineer- lished Doctoral dissertation, Department of Aeronautical
ing. Springer. doi:10.1007/978-3-642-11405-2_8 Engineering, University of Sydney, Sydney, Australia.
Pourzeynali, S., Lavasani, H. H., & Modarayi, A. H. Querin, O. M., Steven, G. P., & Xie, Y. M. (1998). Evolu-
(2007). Active control of high rise building structures using tionary structural optimisation (ESO) using a bidirectional
fuzzy logic and genetic algorithms. Engineering Struc- algorithm. Engineering Computations, 15(8), 10311048.
tures, 29, 346357. doi:10.1016/j.engstruct.2006.04.015 doi:10.1108/02644409810244129
Pradlwarter, H. J., & Schuller, G. I. (1993). Equivalent Rajasekaran, S. (2001). Optimization of large scale three
linearization - A suitable tool for analyzing MDOF- dimensional reticulated structures using cellular genetics
systems. Probabilistic Engineering Mechanics, 8(2), and neural networks. International Journal of Space Struc-
115126. doi:10.1016/0266-8920(93)90005-G tures, 16(4), 315324. doi:10.1260/026635101760832244
421
Ramallo, J. C., Yoshioka, H., & Spencer, B. F. Jr. (2004). Rozvany, G. I. N., Zhou, M., & Birker, T. (1992). Gen-
A two-step identification technique for semiactive control eralized shape optimization without homogenization.
systems. Structural Control and Health Monitoring, 11, Structural Optimization, 4(34), 250252. doi:10.1007/
273289. doi:10.1002/stc.43 BF01742754
Ramesh, K., Seshu, D. R., & Prabhakar, M. (2003). Con- Ruangrassamee, A., & Kawashima, K. (2003). Control of
stitutive behaviour of confined fibre reinforced concrete nonlinear bridge response with pounding effect by vari-
under axial compression. Cement and Concrete Compos- able dampers. Engineering Structures, 25(5), 593606.
ites, 25, 343350. doi:10.1016/S0958-9465(02)00051-3 doi:10.1016/S0141-0296(02)00169-4
Reiter, L. (1990). Earthquake hazard analysis. New York, Rubinstein, M., Giuliano, A., & Mller, O. (2006). Diseo
NY: Columbia University Press. preliminar de estructuras sismorresistentes: un tratamiento
unificado de los efectos traslacionales y rotacionales.
Rezaeian, S., & Der Kiureghian, A. (2008). A stochastic
Memorias XIX Jornadas Argentinas de Ingeniera Es-
ground motion model with separable temporal and spectral
tructural, 49.
nonstationatiries. Earthquake Engineering & Structural
Dynamics, 37, 15651584. doi:10.1002/eqe.831 Ruszczynski, A., & Shapiro, A. (2003). Stochastic pro-
gramming. New York, NY: Elsevier.
Riddell, F. (1995). Inelastic design spectra accounting for
soil conditions. Earthquake Engineering & Structural Saadeghvaziri, M. A., & Yazdani-Motlagh, A. R. (2008).
Dynamics, 24, 14911510. doi:10.1002/eqe.4290241106 Seismic behavior and capacity/demand analyses of three
multi-span simply supported bridges. Engineering Struc-
Robert, C. P., & Casella, G. (2004). Monte Carlo statistical
tures, 30(1), 5466. doi:10.1016/j.engstruct.2007.02.017
methods (2nd ed.). New York, NY: Springer.
Saetta, A. V., Schrefler, B. A., & Vitaliani, R. V. (1993).
Rodriguez, J., & Ortega, L. M. (1994). Assessment of
The carbonation of the concrete and the mechanism of
structural elements with corroded reinforcement. Cor-
moisture, heat and carbon dioxide flow through porous
rosion and Corrosion Protection in Concrete. Sheffield,
material. Journal of Cement and Concrete Research,
UK: Sheffield Academic press.
23(4), 761772. doi:10.1016/0008-8846(93)90030-D
Rosenberg, R. C., & Karnopp, D. C. (1983). Introduction
Sahab, M. G., Ashour, A. F., & Toropov, V. V. (2005). Cost
to physical system dynamics. McGraw-Hill.
optimisation of reinforced concrete flat slab buildings.
Royset, J. O., & Polak, E. (2004). Reliability-based Engineering Structures, 27(3), 313322. doi:10.1016/j.
optimal design using sample average approximations. engstruct.2004.10.002
Probabilistic Engineering Mechanics, 19, 331343.
Saikat, S., & Manohar, C. S. (2005). Inverse reliability
doi:10.1016/j.probengmech.2004.03.001
based structural design for system dependent critical
Rozvany, G. I. N. (1972a). Grillages of maximum strength earthquake loads. Probabilistic Engineering Mechan-
and maximum stiffness. International Journal of Me- ics, 20, 1931. doi:10.1016/j.probengmech.2004.03.006
chanical Sciences, 14(10), 651666. doi:10.1016/0020-
Saka, M. P. (1997). The theorems of structural variation
7403(72)90023-9
for grillage systems. In Topping, B. H. V., & Leeming,
Rozvany, G. I. N. (1972b). Optimal load transmission M. B. (Eds.), Innovation in computer methods for civil
by exure. Computer Methods in Applied Mechanics and structural engineering (pp. 101111). Edinburgh,
and Engineering, 1(3), 253263. doi:10.1016/0045- UK: Civil Comp Press. doi:10.4203/ccp.50.8.2
7825(72)90007-2
Salajegheh, E., & Heidari, A. (2005). Optimum design of
structures against earthquake by wavelet neural network
and filter banks. Earthquake Engineering & Structural
Dynamics, 34(1), 6782. doi:10.1002/eqe.417
422
Sankararaman, S., & Mahadevan, S. (2011). (in press). Shi, Y., & Eberhart, R. (1998). A modified particle swarm
Uncertainty quantification in structural damage diagnosis. optimizer. IEEE World Congress on Computational Intel-
Structural Control and Health Monitoring. doi:10.1002/ ligence, Anchorage, AK, USA, (pp. 69-73).
stc.400
Shinozuka, M. (1970). Maximum structural response to
Saragoni, G., & Hart, G. (1974). Simulation of artificial seismic excitations. Journal of Engineering Mechanics,
earthquakes. Earthquake Engineering & Structural 96, 729738.
Dynamics, 2(3), 249267. doi:10.1002/eqe.4290020305
Shinozuka, M., Deodatis, G., Zhang, R. C., & Papageor-
Sarkar, A. (2003). Linear stochastic dynamical system giou, A. S. (1999). Modeling, synthetics and engineering
under uncertain load: Inverse reliability analysis. Journal applications of strong earthquake wave motion. Soil
of Engineering Mechanics, 129(6), 665671. doi:10.1061/ Dynamics and Earthquake Engineering, 18(3), 209228.
(ASCE)0733-9399(2003)129:6(665) doi:10.1016/S0267-7261(98)00045-1
Scharage, L. (1989). Optimization modeling with LINGO. Shinozuka, M., Feng, M. Q., Kim, H.-K., & Kim, S.-
Technical report, LINDO Systems inc., Chicago. H. (2000b). Nonlinear static procedure for fragility
curve development. Journal of Engineering Mechan-
Schuller, G., & Jensen, H. (2008). Computational meth-
ics, 126(12), 12871296. doi:10.1061/(ASCE)0733-
ods in optimization considering uncertainties- An over-
9399(2000)126:12(1287)
view. Computer Methods in Applied Mechanics and En-
gineering, 198(1), 213. doi:10.1016/j.cma.2008.05.004 Shinozuka, M., Feng, M. Q., Lee, J., & Naanuma, T.
(2000a). Statistical analysis of fragility curves. Jour-
Schurter, K. C., & Roschke, P. N. (2001). Neuro-fuzzy
nal of Engineering Mechanics, 126(12), 12241231.
control of structures using magnetorheological damp-
doi:10.1061/(ASCE)0733-9399(2000)126:12(1224)
ers. Proceedings of the American Control Conference,
Arlington, Virginia, (pp. 1097-1102). Shinozuka, M., & Henry, L. (1965). Random vibration
of a beam column. Journal of Engineering Mechanics,
SEAOC. (1995). Vision 2000, performance based seismic
91, 123143.
engineering of buildings (Vol. I and II). Sacramento, CA:
Structural Engineers Association of California. Shinozuka, M., & Sato, Y. (1967). Simulation of non-
stationary random processes. Journal of Engineering
SEAOC. Vision Committee. (2002). Performance based
Mechanics, 93(1), 1140.
seismic design engineering. Sacramento, CA: Structural
Engineers Association of California (SEAOC) Report. Shook, D. A., Roschke, P. N., Lin, P. Y., & Loh, C. H.
(2008). GA-optimized fuzzy logic control of a large-scale
Seyranian, A. P., Lund, E., & Olhoff, N. (1994). Multiple
building for seismic loads. Engineering Structures, 30,
eigenvalues in structural optimization problems. Struc-
436449. doi:10.1016/j.engstruct.2007.04.008
tural and Multidisciplinary Optimization, 8(4), 207227.
Sigmund, O., & Petersson, J. (1998). Numerical instabili-
Sharifi, R., Kim, Y., & Langari, R. (2010). Sensor
ties in topology optimization: A survey on procedures
fault isolation and detection of smart structures. Smart
dealing with checkerboards, mesh dependencies and local
Materials and Structures, 19, 115. doi:10.1088/0964-
minima. Structural and Multidisciplinary Optimization,
1726/19/10/105001
16(1), 6875.
Shen, J., Tsai, M.-H., Chang, K.-C., & Lee, G. C. (2004).
Smith, A. H., & Chase, J. G. (1993). Identification of
Performance of a seismically isolated bridge under near-
structural system parameters using the cascade-correlation
fault earthquake ground motions. Journal of Structural
neural network. Journal of Dynamic Systems, Measure-
Engineering, 130(6), 861868. doi:10.1061/(ASCE)0733-
ment, and Control, 116, 790792. doi:10.1115/1.2899280
9445(2004)130:6(861)
423
Somerville, P., Smith, N., Punyamurthula, S., & Sun, J. Svanberg, K. (2002). A class of globally convergent opti-
(1997). Development of ground motion time histories for mization methods based on conservative convex separable
Phase 2 of the FEMA/SAC steel project. SAC joint ven- approximations. SIAM Journal on Optimization, 12(2),
ture, California, USA. U.S. Geological Survey. (USGS). 555573. doi:10.1137/S1052623499362822
Retrieved from www.usgs.gov
Swisher, J. R., Hyden, P. D., Jacobson, S. H., & Schru-
Son, Y. K., & Savage, G. J. (2007). Optimal probabilistic ben, L. W. (2000). A survey of simulation optimization
design of the dynamic performance of a vibration ab- techniques and procedures. In J. A. Joines, R. R. Barton,
sorber. Journal of Sound and Vibration, 307(1-2), 2037. K. Kang, & P. A. Fishwick (Eds.), 2000 Winter Simula-
doi:10.1016/j.jsv.2007.06.032 tion Conference.
Soong, T. T., & Spencer, B. F. Jr. (2002). Supplemental en- Symans, M., & Kelly, S. W. (1999). Fuzzy logic con-
ergy dissipation: State-of-the-art and state-of-the-practice. trol of bridge structures using intelligent semi-active
Engineering Structures, 24(3), 243259. doi:10.1016/ seismic isolation systems. Earthquake Engineering &
S0141-0296(01)00092-X Structural Dynamics, 28, 3760. doi:10.1002/(SICI)1096-
9845(199901)28:1<37::AID-EQE803>3.0.CO;2-Z
Spall, J. C. (1998). Implementation of the simultaneous
perturbation algorithm for stochastic optimization. IEEE Taflanidis, A. A. (2009, 2224 June). Robust stochastic
Transactions on Aerospace and Electronic Systems, 34, design of viscous dampers for base isolation applica-
817823. doi:10.1109/7.705889 tions. Paper presented at the Computational Methods
in Structural Dynamics and Earthquake Engineering,
Spall, J. C. (2003). Introduction to stochastic search
Rhodes, Greece.
and optimization. New York, NY: Wiley-Interscience.
doi:10.1002/0471722138 Taflanidis, A. A. (2010). Reliability-based optimal design
of linear dynamical systems under stochastic stationary
Specht, D. F. (1990). Probabilistic neural networks.
excitation and model. Engineering Structures, 32(5),
Neural Networks, 3(1), 109118. doi:10.1016/0893-
14461458. doi:10.1016/j.engstruct.2010.01.023
6080(90)90049-Q
Taflanidis, A. A., & Beck, B. L. (2008). An efficient
Spencer, B. F. Jr, Dyke, S. J., Sain, M. K., & Carlson, J. D.
framework for optimal robust stochastic system design
(1997). Phenomenological model for magnetorheological
using stochastic simulation. Computer Methods in Ap-
dampers. Journal of Engineering Mechanics, 123, 230
plied Mechanics and Engineering, 198(1), 88101.
238. doi:10.1061/(ASCE)0733-9399(1997)123:3(230)
doi:10.1016/j.cma.2008.03.029
Strasser, F. O., & Bommer, J. J. (2009). Review: Strong
Taflanidis, A. A., & Beck, J. L. (2009a). Life-cycle
ground motions Have we seen the worst? Bulletin of
cost optimal design of passive dissipative devices.
the Seismological Society of America, 99(5), 26132637.
Structural Safety, 31(6), 508522. doi:10.1016/j.stru-
doi:10.1785/0120080300
safe.2009.06.010
Subramaniam, R. S., Reinhorn, A. M., Riley, M. A., &
Taflanidis, A. A., & Beck, J. L. (2009b). Stochastic subset
Nagarajaiah, S. (1996). Hybrid control of structures using
optimization for reliability optimization and sensitivity
fuzzy logic. Microcomputers in Civil Engineering, 11,
analysis in system design. Computers & Structures, 87,
117. doi:10.1111/j.1467-8667.1996.tb00305.x
318331. doi:10.1016/j.compstruc.2008.12.015
Svanberg, K. (1987). The method of moving asymptotes
Taflanidis, A. A., & Beck, J. L. (2009c). Stochastic subset
A new method for structural optimization. International
optimization for reliability optimization and sensitiv-
Journal for Numerical Methods in Engineering, 24(2),
ity analysis in system design. Computers & Structures,
359373. doi:10.1002/nme.1620240207
87(5-6), 318331. doi:10.1016/j.compstruc.2008.12.015
424
Taflanidis, A. A., & Beck, J. L. (2010). Reliability-based Takewaki, I. (2002b). Robust building stiffness design for
design using two-stage stochastic optimization with a treat- variable critical excitations. Journal of Structural Engi-
ment of model prediction errors. Journal of Engineering neering, 128(12), 15651574. doi:10.1061/(ASCE)0733-
Mechanics, 136(12), 14601473. doi:10.1061/(ASCE) 9445(2002)128:12(1565)
EM.1943-7889.0000189
Takewaki, I. (2004). Bound of earthquake input energy.
Taflanidis, A. A., & Jia, G. (2011). A simulation-based Journal of Structural Engineering, 130, 12891297.
framework for risk assessment and probabilistic sensitivity doi:10.1061/(ASCE)0733-9445(2004)130:9(1289)
analysis of base-isolated structures. Earthquake Engineer-
Takewaki, I. (2007). Critical excitation methods in earth-
ing & Structural Dynamics, 40(14). doi:10.1002/eqe.1113
quake engineering. Elsevier Science.
Taflanidis, A. A., Scruggs, J. T., & Beck, J. L. (2008).
Takewaki, I. (2009). Building control with pas-
Probabilistically robust nonlinear design of control
sive dampers. Singapore: John Wiley & Sons.
systems for base-isolated structures. Journal of Struc-
doi:10.1002/9780470824931
tural Control and Health Monitoring, 15(3), 697719.
doi:10.1002/stc.275 Takewaki, I., Murakami, S., Fujita, K., Yoshitomi, S., &
Tsuji, M. (2011). The 2011 off the Pacific coast of Tohoku
Tajimi, H. (1960). A statistical method of determining the
earthquake and response of high-rise buildings under
maximum response of a building during earthquake. In
long-period ground motions. Soil Dynamics and Earth-
G. W. Housner (Ed.), Proceedings of the Second World
quake Engineering, 31(11), 15111528. doi:10.1016/j.
Conference on Earthquake Engineering, 11-18 July 1960
soildyn.2011.06.001
Tokyo and Kyoto, Japan, (pp. 781-797). Tehran, Iran:
University press. Tamura, H., Yamamoto, K., Tomiyama, S., & Hatono,
I. (2000). Modelling and analysis of decision making
Takagi, T., & Sugeno, M. (1985). Fuzzy identification
problem for mitigating natural disaster risks. European
of systems and its applications to modeling and control.
Journal of Operational Research, 122(2), 461468.
IEEE Transactions on Systems, Man, and Cybernetics,
doi:10.1016/S0377-2217(99)00247-7
15, 116132.
Tanaka, K., & Sugeno, M. (1992). Stability analysis and
Takeda, T., Sozen, M. A., & Nielsen, N. (1970). Reinforced
design of fuzzy control systems. Fuzzy Sets and Systems,
concrete response to simulated earthquakes. Journal of
45, 135156. doi:10.1016/0165-0114(92)90113-I
the Structural Division, 96(ST12), 25572573.
Tangaramvong, S., & Tin Loi, F. (2011). Mathematical
Takewaki, I. (2001). Probabilistic critical excitation for
programming approaches for the safety assessment of
MDOF elastic-plastic structures on compliant ground.
semirigid elastoplastic frames. International Journal of
Solids and Structures, 48(6), 10111023. doi:10.1016/j.
13451360. doi:10.1002/eqe.66
ijsolstr.2010.12.003
Takewaki, I. (2002). Seismic critical excitation method
Tani, A., Kawamura, H., & Ryu, S. (1998). Intelligent fuzzy
for robust design: A review. Journal of Structural Engi-
optimal control of building structures. Engineering Struc-
neering, 128(5), 665672. doi:10.1061/(ASCE)0733-
tures, 20, 184192. doi:10.1016/S0141-0296(97)00077-1
9445(2002)128:5(665)
Tolson, B. A., Maier, H. R., Simpson, A. R., & Lence,
Takewaki, I. (2002a). Seismic critical excitation method
B. J. (2004). Genetic algorithms for reliability-based
for robust design: A review. Journal of Structural En-
optimization of water distribution systems. Journal of
gineering, 128, 665672. doi:10.1061/(ASCE)0733-
Water Resources Planning and Management, 130(1),
9445(2002)128:5(665)
6372. doi:10.1061/(ASCE)0733-9496(2004)130:1(63)
425
Tomlin, J. (1970). Integer and non-linear programming. Vanderplaats, G. N. (1999). Numerical optimization
In Branch and bound methods for integer and non-convex techniques for engineering design. USA: VR and D, Inc.
programming (pp. 437450). Amsterdam, The Nether-
Vanmarcke, E. H., Shinozuka, M., Nakagiri, S., Schuller,
lands: North Holland.
G. I., & Grigoriu, M. (1986). Random fields and sto-
Tsompanakis, Y., Lagaros, N. D., & Papadrakakis, M. chastic finite elements. Structural Safety, 3(3), 143166.
(Eds.). (2008). Structural design optimization considering doi:10.1016/0167-4730(86)90002-0
uncertainties. London, UK: Taylor & Francis.
Vidal, T., Catel, A., & Francois, R. (2004). Analyzing crack
Tsopelas, P., Constantinou, M. C., Okamoto, S., Fujii, S., width to predict corrosion in reinforced concrete. Cement
& Ozaki, D. (1996). Experimental study of bridge seismic and Concrete Research, 34(1), 165174. doi:10.1016/
sliding isolation systems. Engineering Structures, 18(4), S0008-8846(03)00246-1
301310. doi:10.1016/0141-0296(95)00147-6
Vision, S. E. A. O. C. (2000). Committee. (1995). Per-
Turkington, D. H., Carr, A. J., Cooke, N., & Moss, P. J. formance based seismic engineering of buildings, part
(1989). Seismic design of bridges on lead-rubber bearings. 2: Conceptual framework. Sacramento, CA: Structural
Journal of Structural Engineering, 115(12), 30003016. Engineers Association of California.
doi:10.1061/(ASCE)0733-9445(1989)115:12(3000)
Von Neumann, J. (1966). Theory of self-reproducing
Uang, C.-M., & Bertero, V. V. (1990). Evaluation of seismic automata (Burks, A. W., Ed.). University of Illinois Press.
energy in structures. Earthquake Engineering & Structural
Walther, R., Houriet, B., Isler, W., Moa, P., & Klein,
Dynamics, 19, 7790. doi:10.1002/eqe.4290190108
J. (1988). Cable stayed bridges. London, UK: Thomas
Uniform Building Code (1997). Telford, Ltd.
USGS. (2008). Documentation for the 2008 update of Wang, L., & Langari, R. (1995). Decomposition ap-
the United States national seismic hazard maps (No. proach for fuzzy systems identification. Proceedings of
Open-File Report 2008-1128). Reston, VA: United States the 34thIEEE Conference on Decision and Control, New
Geological Survey. Orleans, LA, USA, (pp. 261-265).
USGS/EERI Advance Reconnaissance Team. (2010). The Wang, A. P., & Lee, C. D. (2002). Fuzzy sliding mode
Mw 7.0 Haiti earthquake of January 12, 2010. control for a building structure based on genetic algo-
rithms. Earthquake Engineering & Structural Dynamics,
Val, D. V. (2004). Aspects of corrosion in reinforced
31, 881895. doi:10.1002/eqe.127
concrete structures and its influence on structural safety.
Report No. 2002950, National Building Research Institute. Wang, L., & Langari, R. (1996). Complex systems mod-
eling via fuzzy logic. IEEE Transactions on Systems,
Val, D., & Stewart, M. G. (2003). Life-cycle cost analysis
Man, and Cybernetics. Part B, Cybernetics, 26, 100106.
of reinforced concrete structures in marine environments.
doi:10.1109/3477.484441
Journal of Structural Safety, 25(4), 343362. doi:10.1016/
S0167-4730(03)00014-6 Wang, Q., Fang, H., & Zou, X. K. (2010). Application
of micro-GA for optimal cost base isolation design of
Valdebenito, M., & Schuller, G. (2011). Efficient strat-
bridges subject to transient earthquake loads. Structural
egies for reliability-based optimization involving non
linear, dynamical structures. Computers & Structures,
89(19-29), 1797-1811. Wang, W. M., Peng, Y. H., Hu, J., & Cao, Z. M. (2009).
Collaborative robust optimization under uncertainty based
Vanderplaats, G. M. (1997). DOT users manual. Colorado
on generalized dynamic constraints network. Structural
Springs, CO: VMA.
doi:10.1007/s00158-008-0271-2
426
Ward, J. H. Jr. (1963). Hierarchical grouping to optimize Xi, Y., & Bazant, Z. P. (1999). Modeling chloride penetra-
an objective function. Journal of the American Statistical tion in saturated concrete. Journal of Materials in Civil
Association, 58, 236244. doi:10.2307/2282967 Engineering, 11(1), 5865. doi:10.1061/(ASCE)0899-
1561(1999)11:1(58)
Wen, Y. K., & Kang, Y. J. (1997). Optimal seismic design
based on life-cycle cost. Proceedings of the International Xu, L., Gong, Y., & Grierson, D. E. (2006). Seismic design
Workshop on Optimal Performance of Civil Infrastructure optimization of steel building frameworks. Journal of
Systems 1997, (pp. 194-210). New York, NY: ASCE. Structural Engineering, 132(2), 277286. doi:10.1061/
(ASCE)0733-9445(2006)132:2(277)
Wen, Y. K., Ellingwood, B. R., & Bracci, J. M. (2004).
Vulnerability function framework for consequence-based Yager, R. R., & Filev, D. P. (1993). Unified structure and
engineering (No. Project DS-4 Report). Urbana, IL: Mid- parameter identification of fuzzy models. IEEE Transac-
America Earthquake (MAE) Center. tions on Systems, Man, and Cybernetics, 23, 11981205.
doi:10.1109/21.247902
Wen, Y. K., & Kang, Y. J. (2001). Minimum building
life-cycle cost design criteria. I: Methodology. Journal Yamaguchi, H., & El-Abd, A. (2003). Effect of energy
of Structural Engineering, 127(3), 330337. doi:10.1061/ input characteristics on hysteretic damper efficiency.
(ASCE)0733-9445(2001)127:3(330) Earthquake Engineering & Structural Dynamics, 32,
827843. doi:10.1002/eqe.250
Wen, Y. K., & Kang, Y. J. (2001a). Minimum building
life-cycle cost design criteria, I: Methodology. Journal of Yan, D., & Yuan, W. (2004). Conceptual seismic design
Structural Engineering, 127(3), 330337. doi:10.1061/ for long span cable-stayed bridges. Journal of Tongji
(ASCE)0733-9445(2001)127:3(330) University, 32(10), 13441348.
Wen, Y. K., & Kang, Y. J. (2001b). Minimum building Yang, C. S., DesRoches, R., & Padgett, J. E. (2009).
life-cycle cost design criteria, II: Applications. Journal Fragility curves for a typical California box girder bridge.
of Structural Engineering, 127(3), 338346. doi:10.1061/ Proceedings of the 2009 ASCE Technical Council on Life-
(ASCE)0733-9445(2001)127:3(338) line Earthquake Engineering Conference, San Francisco,
California, USA.
Werner, G. (1960). Impact: The theory and physical be-
haviour of colliding solids. London, UK: Edward Arnold. Yang, Y. (1999). Seismic resistance of bridge and frac-
tal characteristics. Masters thesis, Tongji University,
Westermo, B. D. (1985). The critical excitation and
Shanghai, China.
response of simple dynamic systems. Journal of Sound
and Vibration, 100, 233242. doi:10.1016/0022- Yan, G., & Zhou, L. L. (2006). Integrated fuzzy logic and
460X(85)90417-1 genetic algorithms for multi-objective control of structures
using MR dampers. Journal of Sound and Vibration, 296,
Wong, K. K. F., & Yong, R. (2002). Earthquake response
368382. doi:10.1016/j.jsv.2006.03.011
and energy evaluation of inelastic structures. Journal of
Engineering Mechanics, 128(3), 308317. doi:10.1061/ Yang, G., Spencer, B. F. Jr, Carlson, J. D., & Sain, M. K.
(ASCE)0733-9399(2002)128:3(308) (2002). Large-scale MR fluid dampers: Modeling and
dynamic performance considerations. Engineering Struc-
Xie, Y. M., & Steven, G. P. (1993). A simple evolution-
tures, 24, 309323. doi:10.1016/S0141-0296(01)00097-9
ary procedure for structural optimization. Computers
& Structures, 49(5), 885896. doi:10.1016/0045- Yang, X. Y., Xie, Y. M., Steven, G. P., & Querin, O. M.
7949(93)90035-C (1999a). Bidirectional evolutionary method for stiffness
optimization. American Institute of Aeronautics and
Xie, Y. M., & Steven, G. P. (1996). Evolutionary struc-
Astronautics Journal, 37(11), 14831488.
tural optimization for dynamic problems. Computers
& Structures, 58(6), 10671073. doi:10.1016/0045-
7949(95)00235-9
427
Yang, X. Y., Xie, Y. M., Steven, G. P., & Querin, O. M. Yoon, G. H. (2011). Topology optimization for non-
(1999b). Topology optimization for frequencies using an linear dynamic problem with multiple materials and
evolutionary method. Journal of Structural Engineer- material-dependent boundary condition. Finite Elements
ing, 125(12), 14321438. doi:10.1061/(ASCE)0733- in Analysis and Design, 47(7), 753763. doi:10.1016/j.
9445(1999)125:12(1432) finel.2011.02.006
Yang, Y. N., & Lin, S. (2004). On-line identification of Yoon, G. H., & Kim, Y. Y. (2005). Element connectiv-
non-linear hysteretic structures using an adaptive tracking ity parameterization for topology optimization of geo-
technique. International Journal of Non-linear Mechanics, metrically nonlinear structures. International Journal of
39, 14811491. doi:10.1016/j.ijnonlinmec.2004.02.010 Solids and Structures, 42(7), 19832009. doi:10.1016/j.
ijsolstr.2004.09.005
Yang, Y. N., & Lin, S. (2005). Identification of parametric
variations of structures based on least squares estimation Yuan, W., Cao, X., Cheung, P., Wang, B., & Rong, Z.
and adaptive tracking technique. Journal of Engineering (2010, August). Development of cable-sliding friction
Mechanics, 131, 290298. doi:10.1061/(ASCE)0733- aseismic bearing for bridges. Paper presented at the 5th
9399(2005)131:3(290) Civil Engineering Conference in the Asian Region and
Australasian Structural Engineering Conference 2010,
Yan, L., & Byrne, P. M. (1992). Lateral pile response to
Sydney, Australia.
monotonic loads. Canadian Geotechnical Journal, 29,
955970. doi:10.1139/t92-106 Yuan, Y., & Sun, B. (2008, October). General introduction
of engineering damage of Wenchuan Ms 8.0 Earthquake.
Ye, A., Hu, S., & Fan, L. (2004). Seismic displacement
Journal of Earthquake Engineering and Engineering
control for super-long-span cable-stayed bridges. China
Vibration, 28, 5973.
Civil Engineering Journal, 37(12), 3843.
Zadeh, L. A. (1965). Fuzzy sets. Information and Control,
Yen, J., & Langari, R. (1998). Fuzzy logic-intelligence,
8, 338353. doi:10.1016/S0019-9958(65)90241-X
control, and information. Upper Saddle River, NJ:
Prentice Hall. Zahrah, T. F., & Hall, W. J. (1984). Earthquake energy
absorption in sdof structures. Journal of Structural En-
Yi, F., Dyke, S. J., Caicedo, J. M., & Carlson, J. D. (2001).
gineering, 110, 17571772. doi:10.1061/(ASCE)0733-
Experimental verification of multinput seismic control
9445(1984)110:8(1757)
strategies for smart dampers. Journal of Engineering
Mechanics, 127, 11521164. doi:10.1061/(ASCE)0733- Zerva, A. (2008). Spatial variation of seismic ground
9399(2001)127:11(1152) motions modeling and engineering applications. Boca
Raton, FL: CRC Press.
Yoon, G. H. (2010a). Maximizing the fundamental ei-
genfrequency of geometrically nonlinear structures by Zhang, J. (2003). Performance-based seismic design
topology optimization based on element connectivity using designed experiments and neural networks. PhD
parameterization. Computers & Structures, 88(12), Thesis, Department of Civil Engineering, University of
120133. doi:10.1016/j.compstruc.2009.07.006 British Columbia, Canada.
Yoon, G. H. (2010b). Structural topology optimization Zhang, J., Makris, N., & Delis, T. (2004). Structural
for frequency response problem using model reduction characterization of modern highway overcrossings-Case
schemes. Computer Methods in Applied Mechanics and study. Journal of Structural Engineering, 130(6), 846860.
Engineering, 199(2528), 17441763. doi:10.1016/j. doi:10.1061/(ASCE)0733-9445(2004)130:6(846)
cma.2010.02.002
Zhao, C. B., Steven, G. P., & Xie, Y. M. (1997). Evo-
lutionary optimization of maximizing the difference
between two natural frequencies of a vibrating structure.
Structural Optimization, 13(23), 148154. doi:10.1007/
BF01199234
428
Zhou, M. (2008). Experimental and theoretical studies Zou, X. K., & Chan, C. M. (2005). An optimal resizing
on seismic performance of elevated pile caps. Doctoral technique for seismic drift design of concrete buildings
dissertation, Tongji University, Shanghai, China. subjected to response spectrum and time history load-
ings. Computers & Structures, 83(19-20), 16891704.
Zhou, M., & Rozvany, G. I. N. (2001). On the validity
doi:10.1016/j.compstruc.2004.10.002
of ESO type methods in topology optimization. Struc-
tural and Multidisciplinary Optimization, 21(1), 8083. Zou, X. K., & Chan, C. M. (2005a). An optimal resizing
doi:10.1007/s001580050170 technique for seismic drift design of concrete buildings
subjected to response spectrum and time history loadings.
Zhu, Z., Fu, Y., Wang, B., & Yuan, W. (2010, August).
Computers & Structures, 83, 16891704. doi:10.1016/j.
Seismic study on bridge piers reinforced with SFRC in
local region. Paper presented at the 5th Civil Engineering
Conference in the Asian Region and Australasian Struc- Zou, X. K., & Chan, C. M. (2005b). Optimal seismic perfor-
tural Engineering Conference 2010, Sydney, Australia. mance-based design of reinforced concrete buildings using
nonlinear pushover analysis. Engineering Structures, 27,
Zienkiewicz, O. C., Taylor, R. L., & Zhu, J. Z. (2005).
12891302. doi:10.1016/j.engstruct.2005.04.001
The finite element method: Its basis and fundamentals.
Oxford, UK: Elsevier Butterworth-Heinemann. Zou, X. K., Chan, C. M., Li, G., & Wang, Q. (2007b).
Multiobjective optimization for performance-based design
Zollo, R. F. (1997). Fiber-reinforced concrete: An
of concrete structures. Journal of Structural Engineer-
overview after 30 years of development. Cement and
ing, 133(10), 14621474. doi:10.1061/(ASCE)0733-
Concrete Composites, 19, 107122. doi:10.1016/S0958-
9445(2007)133:10(1462)
9465(96)00046-7
Zou, X. K., Teng, J. G., Xia, S. H., & Wong, Y. L.
Zou, X. (2002). Optimal seismic performance-based
(2007a). Optimal performance-based seismic retrofit
design of reinforced concrete buildings. Doctoral disserta-
design of FRP-confined concrete buildings. Composite
tion, Hong Kong University of Science and Technology,
Part B: Engineering International Journal, 38, 584597.
Hong Kong, China.
doi:10.1016/j.compositesb.2006.07.016
Zou, X. K., & Chan, C. M. (2001). Optimal drift perfor-
Zou, X. K., Wang, Q., Li, G., & Chan, C. M. (2010). Inte-
mance design for nonlinear pushover response of concrete
grated reliability-based seismic drift design optimization
structures. WCSMO-4: Proceedings of the Fourth Would
of base-isolated concrete buildings. Journal of Structural
Congress of Structural and Multidisciplinary Optimiza-
tion, June 4-8, Dalian, China.
ST.1943-541X.0000216
Zou, X. K. (2008). Integrated seismic design optimiza-
tion of nonlinear base-isolated RC buildings. Structural
doi:10.1007/s00158-007-0184-5
429
430
About the Contributors
Vagelis Plevris is an Assistant Professor at the Department of Civil and Structural Engineering
Educators, School of Pedagogical & Technological Education (ASPETE) in Athens, Greece and a Re-
search Associate at the School of Civil Engineering, National Technical University of Athens (NTUA).
He holds a Bachelor in Civil Engineering, an MSc in Earthquake Engineering, a PhD in Computational
Mechanics from NTUA and also a Master in Business Administration from the Athens University of
Economics and Business. His research work focuses mainly on computational earthquake engineering,
design optimization, reliability analysis of structures and neural networks applications in structural
engineering. His published research work includes more than 10 peer-reviewed journal papers, 4 con-
tributed books as editor, 4 chapters in international scientific books, and 22 papers in international
conferences.
Chara Ch. Mitropoulou is a post-doctoral researcher at the Institute of Structural Analysis and
Seismic Research of the School of Civil Engineering, National Technical University of Athens. She
holds BSc, MSc, and Ph.D. degrees in Civil Engineering from the National Technical University of
Athens. She has participated in a number of research projects and consultancies in the areas of structural
analysis and structural dynamics, shape, sizing and topology optimum design of structures, mathemati-
cal programming, performance based design of RC and steel structures as well as life cycle cost and
fragility assessment of structures. Her publications include 4 book chapters, 2 papers in non-referred
international scientific journals, 7 publications in referred international scientific journals, and 13 papers
in conference proceedings.
Nikos D. Lagaros is a Lecturer at the Institute of Structural Analysis and Seismic Research of the
School of Civil Engineering, National Technical University of Athens and at the School of Corps of
Engineers of the Hellenic Army since 2007. He is serving as Treasurer of the Greek Association for
Computational Mechanics, while he acts as NSF panellist at the Directorate for Engineering Civil,
Mechanical and Manufacturing Innovation, since 2010. Dr. Lagaros provides consulting, peer-review,
and expert-witness services to private companies and federal government agencies in Greece. A focus
of his consulting work is the assessment of buildings after earthquake events and the development of
technical software. The results of his research activity encompass a wide range of topics with an inter-
disciplinary character in the field of computational mechanics. His published research work contains:
70 peer-reviewed journal papers, 1 book, 6 contributed books, 2 international conference proceedings,
15 chapters in international books, and 135 papers in international conferences.
***
Azadeh Alipour, PhD, is an Assistant Professor at the Department of Civil and Environmental Engi-
neering of University of Massachusetts, Amherst. She received her Doctoral degree from the University
of California, Irvine in 2010. Prior to joining to the University of Massachusetts, she was a postdoctoral
scholar at University of California, Irvine. Dr. Alipour has served as a committee member in ASCE and
ACI. Furthermore, she is a reviewer of ASCE Journal of Bridge Engineering, ASCE Journal of Engi-
neering Mechanics, and Journal of Structure and Infrastructure Engineering. Her research interests
include probabilistic performance evaluation of civil infrastructure components under multi-hazards,
risk mitigation strategies for infrastructure systems under extreme conditions, and life-cycle engineering
and management of civil infrastructures.
Luis Becerra is a Masters student of Civil Engineering at Santa Maria University, Valparaiso, Chile.
The topic of his Masters thesis is the optimal design of structural systems subject to stochastic loading
using an interior-point method for optimization.
Soumya Bhattacharjya, born on 12th March, 1979, is an Assistant Professor in the Department of
Civil Engineering, Bengal Engineering and Science University, Shibpur, India. He completed his Bach-
elor of Engineering in Civil Engineering in 2002 from Bengal Engineering College, Shibpur, India and
Master of Engineering with specialization in Structural Engineering in 2004 from Bengal Engineering
and Science University, Shibpur, India. Subsequently, he obtained his PhD in Civil Engineering from
the same university in the field of Robust Optimization of Structures in 2011. In general, Dr. Bhattacha-
rjyas research interest lies in the field of reliability, optimization under uncertainty, steel structures, and
composite mechanics.
Subrata Chakraborty is currently a Professor at the Bengal Engineering and Science University,
Shibpur. He is a fellow of the Indian National Academy Engineering and the Institution of Engineers
(India). Prof. Chakraborty did his Bachelor of Engineering in Civil Engineering from Bengal Engineer-
ing College, Shibpur, M. Tech. and Ph.D. from Indian Institute of technology, Kharagpur. He was a
postdoctoral researcher at University of Cambridge, UK and University of Arizona, USA and Technical
University of Aachen, Germany. In general, Prof. Chakrabortys research interest lies in the field of
computational mechanics under uncertainty, structural health monitoring, vibration control, composite
mechanics, et cetera. He has published extensively in peer reviewed journals, authored textbook and
book chapters, and reviewed research articles for various national and international journals. As an in-
dependent researcher, Prof. Chakraborty has completed a number of research projects funded by various
agencies and he is also active in important industrial consultancy.
Pak-Chiu Cheung is a Kwang-Hua Foundation visiting Professor of Civil Engineering at Tongji

University, Shanghai, China. Also an adjunct Associate Professor at the Hong Kong Polytechnic Uni-
versity, he retired as a Structural Engineer from the Architectural Services Department of Hong Kong
SAR Government in 2010 and was President of American Society of Civil Engineers Hong Kong Section
for the session 2008-2009. He received his B.S. from the National Cheng-Kung University, Taiwan, in
1974, M.S. from Cornell University, New York, in 1976, and Ph.D. from the University of Canterbury,
New Zealand, in 1991. Dr. Cheungs research interest focuses on seismic structural design.
431
Tahar El-Korchi is a Professor and Chair of the Civil and Environmental Engineering Department
at Worcester Polytechnic Institute in Worcester, Mass. He has been at WPI since 1987. His teaching and
research interests are in the smart materials and structures, structural composites and control, construction
materials and processes, pavement engineering, and testing. He has authored and coauthored over 80
publications in this area, including a textbook on Pavement Engineering. He received funding through
federal and state agencies including the NSF PYI program. He is active in numerous organizations
including TRB, FHWA, ACI, and ASCE. He is very active in the global perspective program at WPI
where he directs the Panama and Morocco student project centers.
Amr Elnashai, is the Head of the Civil and Environmental Engineering Department, and Bill and
Elaine Hall Endowed Professor at the University of Illinois at Urbana-Champaign, USA. He was Direc-
tor of the NSF multi-institution interdisciplinary Mid-America Earthquake Center and NSF Network for
Earthquake Engineering Simulation (NEES) Laboratory at Illinois. Amr obtained his MSc and PhD from
Imperial College, University of London, UK and is a Fellow of the UK Royal Academy of Engineering.
Before joining the University of Illinois in June 2001, Amr was Professor of Earthquake Engineering
and Head of Section at Imperial College (1985-2001). He is founder and co-editor of the Journal of
Earthquake Engineering and editorial board member of several other journals. Amrs technical interests
are multi-resolution distributed analytical simulations, network analysis, large-scale hybrid testing and
field investigations of the response of complex networks and structures to earthquakes. He has produced
more than 250 research publications, including over 120 refereed journal papers, many conference, key-
note and prestige lectures (including the Nathan Newmark Distinguished Lecture), research reports, 2
books and several book chapters, magazine articles and earthquake field investigation reports. Amr has
successfully supervised 40 Doctoral and over 100 Master of Science theses.
Ricardo O. Foschi is Emeritus Professor, Civil Engineering, University of British Columbia, Vancou-
ver, Canada. Dr. Foschi received his Ph.D in Engineering Mechanics from Stanford University in 1966,
and has contributed many publications in different topics of applied mechanics, including behaviour of
wood structures, reliability, and application of general probabilistic methods to different problems in
engineering. His publications have appeared in national and international journals, as well as contributed
book chapters. His continuing areas of research include reliability applications in earthquake engineer-
ing, performance-based design, and optimization applications.
Bora Gencturk is an Assistant Professor at the Civil and Environmental Engineering Department
at the University of Houston. Bora obtained his MSc and PhD from University of Illinois at Urbana-
Champaign. He is a member of several professional organizations including the American Society of
Civil Engineers, American Concrete Institute and Earthquake Engineering Research Institute. Boras
technical interest are use of advanced materials for hazard mitigation and sustainable construction, life-
cycle analysis of new and existing buildings subjected multiple hazards, development of small-scale
structural testing methods, and vulnerability assessment of structures. Bora has produced more than 20
research publications, including two book chapters and five research reports.
432
Kazem Ghabraie completed a Bachelors and a Masters degree in Civil Engineering in University
of Tehran, Iran. He obtained his PhD in Civil Engineering from RMIT University in Melbourne, Aus-
tralia in 2009. Since 2010, he has been working as a Lecturer in University of Southern Queensland,
Toowoomba, Australia. His research interests include non-linear finite element method and numerical
structural optimization. Since 2002 he is working on topology optimization of structures as his main
research area.
Saeed Gholizadeh was born in Urmia, Iran, in 1978. He received his B.Sc. degree in Civil Engineer-
ing from Urmia University in 2001, and his M.Sc. and Ph.D. degrees in Structural Engineering from
University of Kerman as the first rank in 2003 and 2009, respectively. Dr. Gholizadeh has been with
the Department of Civil Engineering, Urmia University, Iran, since 2009. His research interests include
structural optimization, earthquake engineering, space structures, and soft computing. He has published
over 50 journal papers, conference articles and book chapters in the above areas. Dr. Gholizadeh is a
member of the Centre of Excellence for Fundamental Studies in Structural Engineering of Iran Univer-
sity of Science and Technology, member of the Iranian Institute of Spatial Structures and member of the
editorial board of the International Journal of Optimization in Civil Engineering.
Ioannis Gidaris received his Diploma in Civil Engineering with major in Structural Engineering from
the Civil Engineering Department of the Aristotle University of Thessaloniki in 2008. He completed the
Masters program Earthquake Resistant Design of Structures in 2009 in the same department. He is
currently a PhD student at the University of Notre Dame and his research interests are performance based
seismic design of structures, assessment of the nonlinear response of structures subjected to earthquake
excitations, life-cycle cost assessment, and Bayesian updating of deteriorating bridge infrastructure
systems through monitoring data.
Hctor Jensen is Professor of Civil Engineering at Santa Maria University, Valparaiso, Chile. He
received his degree of Mathematician Civil Engineer from University of Chile, Santiago, and his Ph.D.
in Applied Mechanics from California Institute of Technology (Caltech), Pasadena, USA. Jensens
research expertise is in structural reliability based optimization, stochastic structural dynamics, and
structural reliability.
Jin Seop Kim is a postdoctoral researcher at the Department of Brain and Cognitive Sciences at
Massachusetts Institute of Technology (MIT), U. S. A. He studied for his Doctorate at the Department of
Physics and Astronomy at Seoul National University, Republic of Korea. His research interests include
critical phenomena and universality in statistical physics, structure and dynamics of complex systems,
biophysics, and computational models of neural systems. At MIT he is studying the topology of neuronal
systems at electron microscope resolution for macroscopic systems.
Yeesock Kim currently holds the position as the Assistant Professor of Structural Engineering at the
Department of Civil and Environment Engineering at Worcester Polytechnic Institute (WPI), Massachu-
setts, USA. He was a postdoctoral researcher of Control Engineering at the Department of Mechanical
Engineering at Texas A&M University, College Station, USA, in 2007-2010. He is the author of 6 book/
book chapters, including 42 technical papers in various fields of smart structures, vibration control, system
433
identification, and control theory. At WPI, he teaches Smart Structures, Smart Infrastructure Analysis
and Design, Structural Dynamics, and Structural Analysis and Design. He is also currently serving on
the IEEE Control Systems Society Editorial Board.
Sankaran Mahadevan is a Professor at the Department of Civil & Environmental Engineering,

Vanderbilt University, USA. He is also the Director of the NSF-IGERT Multidisciplinary Graduate
Program in Reliability and Risk Engineering and Management. Dr. Mahadevan is a member of the
American Society of Civil Engineers (ASCE) and has acted as an Associate Editor for the Journal of
Structural Engineering (ASCE). Professor Mahadevan co-authored two books on reliability and statisti-
cal methods in engineering design and on reliability assessment using stochastic finite element analysis.
He has more than 100 research papers published in reputable international journals and several book
chapters. His research interest includes probabilistic computational mechanics, reliability and risk as-
sessment, structural durability, fatigue and fracture, model verification and validation under uncertainty
and reliability-based design optimization.
Hamid Moharrami is a Senior Professor in Tarbiat Modares University, Tehran, Iran, since October
1993. He received his B.Sc. from Amirkabir Universities of Technology, and M.Sc., from Sharif Uni-
versities of Technology, both in Tehran Iran. He continued his education at the University of Waterloo,
Canada where, he got his Ph.D. in 1993 in Civil Engineering with a speciality in structural optimization.
Dr. Moharramis research interests include design optimization, nonlinear analysis, numerical analysis
and experimental methods. He has been doing research in three branches of design optimization. One
is developing new optimization algorithms or improving the existing algorithms. The second is the
use of the optimization algorithm in various fields of Civil Engineering design problems. The third is
the application of optimization algorithms, especially LP and QP in developing nonlinear structural
analysis algorithms. His research endeavour has led to many fascinating solution schemes for design
optimization of different types of structures including buildings, bridges, space structures, differential
surge tanks, et cetera.
Oscar Moller is Professor, School of Civil Engineering, University of Rosario, Argentina, and the
Institute of Applied Mechanics and Structures (IMAE), University of Rosario, Argentina. He is also
Researcher with the Research Council of the University of Rosario, Argentina. Dr. Moller received his
Doctoral degree in Engineering in 2001 from the School of Civil Engineering, University of Rosario,
Argentina. Dr. Mollers interests are structural mechanics and dynamics, nonlinear response of structures,
earthquake engineering, and issues of reliability in the context of seismic design. He has previously
published in international journals and has had collaborations in book chapters in the area of earthquake
engineering, focussing on reliability, performance-based design, and structural optimization algorithms.
Abbas Moustafa is an Associate Professor at the Department of Civil Engineering, Minia University,
Egypt. He is also the chairman of the Department of Civil Engineering, High Institute of Engineering,
Tamoh, Giza, Egypt. He is the editor of two books on earthquake resistant structures and geotechnical
earthquake engineering. Dr. Moustafa got his Doctoral degree in 2002 from the Indian Institute of Sci-
ence in earthquake engineering and structural reliability. He joined Vanderbilt University as a senior
research associate. He worked for a funded research project by the US-Air Force on structural health
434
monitoring using bond graphs. He joined also Nagasaki and Kyoto Universities as a JSPS research fel-
low. Dr. Moustafa has more than 25 research papers published in international journals, and he acts as a
reviewer for a number of regional and international journals. His research interest includes earthquake
engineering, seismic design, inelastic structures, optimum design, structural health monitoring, structural
safety, base-isolation, structural control, and uncertainty modeling.
Marcelo Rubinstein is Professor, School of Civil Engineering, University of Rosario, Argentina, and
the Institute of Applied Mechanics and Structures (IMAE), University of Rosario, Argentina, of which
Professor Rubinstein is the Director of the Structures Laboratory. Professor Rubinstein received his
Civil Engineering degree from the University of Rosario in 1962. His interests are structural mechanics
and dynamics, structural design, bridge analysis and design, earthquake engineering, and behavior of
structures with energy dissipation devices.
Fabin Savino is a Student and Research Assistant, School of Engineering and IMAE, University
of Rosario, Argentina. Future interests focus on structural analysis and design, including computer ap-
plications and optimization of structures.
Juan Seplveda received his Diploma and Masters Degree in Civil Engineering from Santa Maria
University, Valparaiso, Chile, in 2010. The topic of his Masters thesis is the optimal design of structural
systems subject to stochastic loading considering discrete design variables.
Behrouz Shafei, PhD, is an Assistant Professor at the Department of Civil and Environmental Engi-
neering of University of Massachusetts, Amherst. He received his Doctoral degree from the University
of California, Irvine in 2011. Prior to joining to the University of Massachusetts, he was a postdoctoral
scholar at University of California, Irvine. Dr. Shafei has been selected as ASCE 2012 New Face of Civil
Engineering. Furthermore, he is the recipient of the Public Impact Distinguished Fellowship and James
D. Cooper best paper award. Dr. Shafeis research addresses the deterioration of civil infrastructure
components subjected to multiple environmental stressors. His investigation also includes the effects of
spatial and temporal uncertainties on the estimation of the durability and state of deterioration.
Masanobu Shinozuka, PhD, PE, NAE, is a Distinguished Professor at the Department of Civil and
Environmental Engineering of University of California, Irvine. He received his Doctoral degree from
Columbia University in 1960. He is an Elected Member of the National Academy of Engineering, Dis-
tinguished Member of ASCE, and Fellow ASME. He has been the recipient of prestigious awards and
medals, including the ASCE Theodore von Karman medal (1994), ASCE Nathan M. Newmark medal
(1985), and ASCE Alfred M. Freudenthal medal (1978). Dr. Shinozuka is a world-renowned expert in
applied mechanics and structural engineering. He has published several technical papers on random
vibration, reliability of structural systems, structural control, and risk assessment of civil infrastructure
components.
Alexandros Taflanidis is the Rooney Family Assistant Professor in the Department of Civil Engi-
neering and Geological Sciences at the University of Notre Dame. Dr. Taflanidis received his diploma
in Civil Engineering (2002) and Masters in Earthquake Resistant Design of Structures (2003) from
435
the Aristotle University of Thessaloniki. He received his PhD in 2008 in Civil Engineering with a minor
in Control and Dynamical Systems from the California Institute of Technology. His research focuses
on the implementation of advanced computational and simulation methodologies for assessment and
mitigation of risk due to natural hazards (such as earthquakes, waves and hurricanes). He is also the
Director of the High Performance System Analysis and Design (HIPAD) Lab at Notre Dame, pursuing
fundamental research in the analysis and design of high-performance engineering systems in the pres-
ence of probabilistically characterized modeling uncertainties. The applications his group examines
extend to seismic resistant design of structural systems, assessment of risk due to hurricanes in support
of emergency responses, optimal monitoring and maintenance of bridge infrastructure systems, and
optimization of offshore energy (wind/wave) conversion devices.
Marcos Valdebenito is Assistant Professor of Civil Engineering at Santa Maria University, Valparaiso,
Chile. He received his Diploma and Masters Degree in Civil Engineering from Santa Maria University,
Valparaiso, Chile, in January 2006 and his Doctoral Degree in Civil Engineering from the University of
Innsbruck, Innsbruck, Austria, in February 2010. His fields of interest are simulation methods, reliability-
based optimization, and reliability sensitivity.
Wan-Cheng Yuan is a Professor of Civil Engineering in the State Key Laboratory of Disaster Re-
duction in Civil Engineering, Tongji University in Shanghai, China. He was Head of the Universitys
Department of Bridge Engineering from 1992 to 1996. He was a visiting Professor at the Hong Kong
University of Science and Technology in 1997 and at Darmstadt University of Technology in Germany
from 1993 to 1994. He received his B.S., M.S. and Ph.D. degrees from Tongji University in 1984, 1987,
and 1990 respectively. Dr. Yuans research interests cover various areas of bridge engineering, including
seismic structural design, performance and safety assessment, bridge vibration, ductility of high-strength
concrete, and bridge health monitoring. He received the First Class Prize of the National Competition
for Progress in Science and Technology of China in 2009, the First Level Technology Advances Award
from Shanghai City in 2008, and the Second Award in Science and Technology Advances Program from
the Ministry of Education of China in 1995. He has authored and co-authored more than a hundred
technical papers.
Yu-Guo Zheng is a Ph.D. Candidate in Bridge and Tunnel Engineering at Tongji University and
presently a Lecturer in Civil Engineering at the Hunan University of Science and Technology, Xiangtan,
China.
Xiao-Kang Zou obtained her Bachelor Degree and Master Degree of Civil Engineering in 1987 and
1990, respectively, in Northern Jiaotong University (currently called Beijing Jiaotong University) in
Beijing, P.R. China. She obtained her PhD degree in 2002 from the Department of Civil Engineering,
the Hong Kong University of Science and Technology. Her PhD topic is Optimal Seismic Performance-
Based Design of Reinforced Concrete Buildings. Currently, she works in AECOM Asia Company Ltd.
in Hong Kong as a principal Engineer. She is mainly responsible for developing and applying elastic
and inelastic optimization technique to design projects under wind load and seismic load.
436
437
Index
A chloride transport mechanism 77

capillary suction 77
acceleration-displacement response spectrum ionic diffusion 77
(ADRS) 182 permeation 77
aleatory uncertainties 129 close twin-column pier (CTP) 275
Ant Colony Optimization (ACO) 154, 165 Close Twin-Column Pier System 305
Autoregressive eXogenous (ARX) 323, 325, 336 coefficient of variation (CoV) 156
axial-moment hinge 216-217 collapse performance level 4, 9-10, 13, 23, 25, 34-
35, 37, 39, 129, 134, 141, 143, 148, 176-177,
B 212, 217, 224, 271-272, 394
interstory drift 3-4, 6-7, 10-12, 16-18, 35, 65-
back-propagation (BP) 308
66, 70, 214, 225
basic safety objective (BSO) 177
maximum global damage index 33, 35
Bi-directional ESO (BESO) 234
maximum local damage index 33, 35
bond graph (BG) 342-343
Collapse Prevention (CP) 10, 177, 224
bridge modelling
Combined Methods 184
abutments 84, 286, 289, 370-371, 374-375,
Common Random Numbers (CRN) 383
377, 386, 388, 391-392
complete quadratic combination (CQC) 212
pier columns 84
compliance minimization 250
superstructure model 84
concentration cycle 77-78
Bridge Piers Reinforced with SFRC 271, 291, 294,
continuation method 239
302, 304
continuous girder bridges 272
BWBM model 44
cumulative distribution function 92
C D
Cable-Sliding Friction Aseismic Bearing 304-305
damage detection 343
cable-stayed bridges 272
damage indices 128, 145
fixed system 272
damage signature derivation 358
floating system 272
backward propagation (BP) 358
semi-floating system 272
forward propagation (FP) 358
Capacitance (C) 344
deck drift ratio (DDR) 85
cellular automata (CA) 308, 310
depassivated 81
cellular crossover operation (CCO) 311
Design (decision) Variables 4
cellular genetic algorithm (CGA) 308
Design Parameters (DPs) 107
chloride binding capacity 78
deterministic and probabilistic seismic hazard analy-
chloride content profile 80
sis (DSHA and PSHA) 8
chloride-induced corrosion 77, 100
deterministic design optimization (DDO) 107
diffusion cycle 78
Index
discrete formulation 51 I
ductility 132
Immediate Occupancy (IO) 10, 177, 224
E Incremental Scheme 184, 190
Inertance (I) 344
earthquake loads 129, 131-132, 344 initial cost 31
earthquake-resistant design of structures 129 Initial Stiffness method 184
Eigenfrequency Control 232, 268 inspection and maintenance intervals 93
Elastic Cable Seismic Mitigation and Isolation inspection scheduling 152, 159, 168, 173
Device 305 inter-story drift responses 217
elastic dynamic 3 isolation design 273
evaporable water content 78
Evolutionary Algorithms (EA) 155 L
Evolutionary Structural Optimization (ESO) 234
exact nonlinear dynamic Analysis (ENDA) 316 life-cycle cost (LCC) 1, 20, 87, 99
excitation models 52 Life Safety (LS) 10, 177, 224
life safety performance level 4, 9-10, 13, 23, 25, 34-
F 35, 129, 153, 177, 217, 224
interstory drift 3-4, 6-7, 10-12, 16-18, 35, 65-
Fault Adaptive Control Technology (FACT) 342, 66, 70, 214, 225
363 maximim local damage 33, 35
Finite Difference (FD) 188 maximum global damage index 33, 35
fmincon algorithm 137 Linear dynamic analysis 150, 178-179
For single-objective seismic design optimization 2 Linear Programming (LP) 187
fragility analysis 86 Linear static analysis 175, 178-179
frictional bearings 273 linear time-invariant (LTI) 333
fully stressed design (FSD) 186 load and resistance factored design (LRFD) 175
fuzzy-decision theory 220 local capacity design 271
fuzzy inference systems (FIS) 307 local minima 238
local response quantities 217
G longitudinal support unseating 272
generalized regression neural network (GRNN) 308 lower bound Fourier amplitude spectra (LBFAS)
Generic Modeling Environment (GME) 342 134
genetic algorithms (GA) 6
gradient-based algorithms 5 M
ground motion uncertainities 24 magnetorheological (MR) 323, 325, 330, 335
Gyrator (GY) 344 Markov Chain Monte Carlo procedure 53
Material Properties 3, 8, 11-12, 20, 83-84, 101, 153,
H 155, 168, 176, 234, 242, 291, 377
hard kill approach 258 Matlab optimiztion toolbox 137
Hazard Levels 4, 7-10, 13, 17, 87, 153, 319 maximum allowable admission ratio 259
heuristic-based algorithms 5, 22 Membership functions (MFs) 325
hierarchical clustering-based (HRC) 325, 336 metaheuristic optimization 168
HIgh Performance system Analysis and Design meta-heuristic optimization algorithms 307
(HIPAD) 389 method of moving asymptotes (MMA) 237, 251
hollow single-column pier (HSP) 274 modified cellular genetic algorithm (MCGA) 311
homogenization method 233 moment hinge 215, 217
hybrid design space (HDS) 106 monotonic convergence 53, 72
Monte Carlo Simulation (MCS) 11, 115
438
Index
moving least-squares method (MLSM) 108 pheromone 165

multi degree of freedom (MDOF) 111 pheromone trail 165
multi-input, multi-output (MIMO) 325, 336 pile-cap mass 25
multiple-objective seismic design optimization 2 pile group foundation strengthened with SPPs 296
Multi-Stripe Dynamic Analysis (MSDA) 158 plastic hinges 196, 214
mutation operation (MO) 311 plastification 189
pot bearings 273
N probabilistic design 372
probabilistic neural network (PNN) 308
National Oceanic and Atmospheric Administration probability density function (PDF) 31, 110, 156, 373
(NOAA) 78 pushover analysis 209
near-fault ground motions 372, 379
neural networks 24 Q
neural network system (HNNS) 308, 312
Newton-Raphson method 184, 194, 253 Quadratic Programming (QP) 185, 187
Nonlinear dynamic analysis 26, 31, 33-34, 45-46,
129, 145, 148, 175, 178, 180-181, 196, 204, R
308, 315-316, 372, 374
nonlinear static analysis 3, 153, 178-179, 212 radial basis function (RBF) 307
nonlinear system identification 336 reinforced concrete (RC) 1-2, 77, 153, 208, 284, 297
nonlinear time-history analysis method 129 reliability based design optimization (RBDO) 106
non-stationary stochastic process 51, 55 reliability-based optimization (RBO) 52
repair costs 31
O Resistor (R) 344
response parameters 33
Objective (merit) Function 4 DIES 31-33, 35
OpenSees 83-84, 103 DILO 33, 35
Operational (OP) 177 DIST 33, 35-36
operational perfermance level 9, 20, 23, 25, 34, 171, UMAX 33, 35
173, 177, 211 response spectrum analysis 208
elastic displacement limit 34 response surface method (RSM) 105, 108
interstory drift 3-4, 6-7, 10-12, 16-18, 35, 65- risk quantification 374
66, 70, 214, 225 roller bearings 273
Optimality Criteria (OC) 187, 210, 237 root mean square (RMS) 112
optimization under uncertainty 382 rubber bearings 273
optimum path 165
optimum performance-based seismic design S
(OPBSD) 176
overall conceptual seismic design 270 Saturated Design (SD) 118
SCC bridge 273
P seismic dampers 371
seismic mitigation 273
parameter uncertainty 106 Sensitivity Analysis 51-52, 114, 174, 176, 187-191,
Pareto-Optimality 4-5 193-196, 199, 201, 204-205, 236, 244, 251,
Pareto optimal set 221 253, 261, 264-265, 268, 306, 372, 384, 390-
Particle Swarm Optimization (PSO) 154, 307 393, 395-396
passive dampers 151, 232, 264, 268 sensitivity based approach 107
performance based design optimization (PBDO) 106 sensitivity numbers 258
Performance-Based Design (PBD) 153, 175 sensor faults 343
performance-based seismic design (PBSD) 6 Sequential Linear Programming (SLP) 187
Performance Levels 4, 9-13, 23-25, 34, 37-39, 177, sequential quadratic programming (SQP) 121, 187
195, 200-202, 217, 219 service failure 92-94, 100
439
Index
shape optimization 233 subset simulation 66

Simulated Annealing (SA) 6, 307 system identification (SI) 324, 336, 343
simultaneous failure mode (SFM) 186
Simultaneous Perturbation Stochastic Approxima- T
tion (SPSA) 382
single-degree-of-freedom (SDOF) 6, 348 taboo search (TS) 6, 14
sizing optimization 232 Takagi-Sugeno (TS) 323, 325, 336
smart control 329 temporal causal graph (TCG) 342, 348, 354
Solid Isotropic Microstructures with Penalization the peak ground acceleration (PGA) 8
(SIMP) 234 time-history 55, 379, 387
solid twin-column pier (STP) 274 topology optimization 233
Solution (objective function) Space 4 total cost 7, 23-25, 29-30, 38-41, 43, 45, 48, 77,
Sorting Evolution Strategies II (NSES-II) 155 87-88, 271
Space of Design (decision) Variables or Search transformer (TF) 344
Space 4 Travelling Salesman Problem (TSP) 161
spatial tower 272, 278 tuned mass damper (TMD) 107
Steel-Concrete Composite Bridge 304-305
steel fiber reinforced concrete (SFRC) 269, 271, 291 U
steel protective pipes (SPPs) 269, 273, 296 uncertain but bounded (UBB) 107
stoachastic model 380 uniform hazard spectra (UHS) 8
stochastic earthquake load 108, 115, 124 upper bound Fourier amplitude spectra (UBFAS)
stochastic structural optimization (SSO) 106, 111 134
structural collapse 148, 176, 272 U-shaped flexural plates (UFP) 69
Structural health monitoring (SHM) 343
Structural optimization 1-6, 15, 21-23, 44, 50-53, V
73-75, 105-106, 111, 126-127, 164, 174-176,
204-206, 209, 229, 232-235, 241, 258, 264- virtual work 5, 210, 214, 228
268, 306, 308-310, 321, 344
shape optimization 1, 258 W
sizing optimization 1 weighted sum method (WSM) 109
topology optimization 1, 239, 256
Structural variation theory studies 185
440

Structural Seismic Design Optimization

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Structural Seismic Design Optimization

Hochgeladen von

Copyright:

Verfügbare Formate

Structural Seismic

Design Optimization and

Chara Ch. Mitropoulou

Published in the United States of America by

Library of Congress Cataloging-in-Publication Data

British Cataloguing in Publication Data

Compilation of References................................................................................................................ 399

About the Contributors..................................................................................................................... 430

Compilation of References................................................................................................................ 399

About the Contributors..................................................................................................................... 430

Chara Ch. Mitropoulou

Chara Ch. Mitropoulou

INTRODUCTION zation of the properties of individual elements to

BACKGROUND ON of a structure. Each hazard level is usually

Performance Levels: Predefined levels in the space of decision variables to a point,

Figure 2. Illustration of Pareto-optimality

Table 1. Earthquake design levels in Vision 2000

state probabilities are calculated the cost of repair (4)

respectively, of earthquake demand evaluated The Optimization Algorithm

Minimum Maximum Increment

Figure 9. Steps for performing LCC oriented seismic design optimization

Figure 10. RC frame structure used for the example application

Oscar Mller Fabin Savino

Marcelo Rubinstein Ricardo O. Foschi

INTRODUCTION objective (e.g.,the total cost), with constraints

1. THE OPTIMIZATION The optimization process requires the calcula-

Figure 1. Neural network architecture

A. Search for weights W trying to zero out the

2E The optimization process OPT described in 4.2

Figure 2. Schematic for the design parameters optimization process

5.1 Initial Cost

The nonlinear dynamic analysis should include (21)

Under the assumption that the structure is fully (28)

C 2 DIES = C (DIES ) ft2 (t ) dt b

6. 1 Random Variables, (31)

particular combination; and (b) the concrete and k =1

Figure 3. Portal frame

Table 1. Application 1, main random variables, and definitions

X(6) = span 0.00298 0.01389 Steel reinforcement ratio, beam span

X(8) = col 0.008 0.04286 Steel reinforcement ratio, columns

X(9) = fr / fc0 0 0.15 Normalized concrete confinement pressure

X(10) = aG (cm / s2) 25 1200 Ground peak acceleration

Interstory drift distribution over the sub-combinations. Thus, in

Life safety: F (X) F (X)

Figure 4. Neural networks agreement for the response DIST

Table 2. Random variables

Variable X X Type Variable X X Type

X(1) = m 5.1x10-4 5.1x10-5 Normal X(10) = aG 94 cm/s2 130cm/s2 Lognormal

X(2) = bb 20 cm 1 cm Normal X(11) = fg 2.50 Hz 0.375 Hz Normal

X(3) = hb ? cm 0.05 X Normal X(12) = aG 0 0.25 Normal

X(4) = bc 30 cm 1.5 cm Normal X(13) = aG X(13) = X(10) [1.0+X(12)]

X(5) = hc ? cm 0.05 X Normal X(14) = RN 1 0 1 Normal

X(6) = span ? 0.10 X Lognormal X(15) = U1 0 1 Uniform

X(7) = end ? 0.10 X Lognormal X(16) = XN 1 0 1 Normal

X(8) = col ? 0.10 X Lognormal X(17) = XN 2 0 1 Normal

X(9) = fr / fc0 0.10 0.01 Normal X(18) = XN 3 0 1 Normal

Table 3. Initial values for the design parameters

Table 4. Target minimum reliability levels (Paulay and Priestley, 1992)

Performance level Pf annual annual Pf

Table 5. Target minimum reliability levels proposed

Performance level Pf annual annual Pf

expanded, indicating that at least a local minimum 6.6 Optimization Sensitivity to

Table 6. Initial costs and reliabilities, preliminary design

Costs [$] Performance level