MSC Dissertation Zhang 2012

DIFFERENT APPROACHES TO DERIVE
ANALYTICAL FRAGILITY FUNCTIONS

OF BRIDGES

A Dissertation Submitted in Partial Fulfilment of the Requirements
for the Master of Science Degree in

Earthquake Engineering

By
Xiaoxuan ZHANG
..

Supervisors: Dr Rui Pinho
Dr Ricardo Monteiro

February, 2013

Istituto Universitario di Studi Superiori di Pavia
Universit degli Studi di Pavia

The dissertation entitled Different Approaches to Derive Analytical Fragility Functions of
Bridges, by Xiaoxuan Zhang, has been approved in partial fulfilment of the requirements for
the Master of Science Degree in Earthquake Engineering.

Name of Reviewer 1

Name of Reviewer 2

Abstract
i

ABSTRACT

Fragility functions play a critically significant role in the assessment of seismic loss, thus the
development of reliable procedures for their calculation has become increasingly popular. The
existence of different analytical methodologies, as well as structure modelling approaches, has an
important influence in the results of fragility functions. In addition it has been recognised that bridges
is one of the most vulnerable structural types during past seismic events. Depending on the seismic
conditions of local site, seismic vulnerability assessment of bridges can be carried out based on
fragility curves. In this study, 3D analytical RC bridge models are generated by OpenSees based on
different material and geometric parameters in order to take structural variability in consideration.
Nonlinear analyses are conducted for each bridge model under different ground motion records.
Curvature ductility is employed as the reference parameter in determining the bridge damage limit
states in terms of structural capacity. Several analytical procedures are used to derive fragility
functions for each simulated bridge. Seven different nonlinear static procedures (Capacity Spectrum
Method, N2, Displacement Coefficient Method, Modified Modal Pushover Analysis, Adaptive
Capacity Spectrum Method, Adaptive Modal Combination Procedures and Modified Adaptive Modal
Combination Procedure), which make use of conventional and adaptive pushover analysis, are
investigated using real ground motion records, as tools to derive fragility functions. A sufficiently
large, thus representative, ensemble of RC bridges is considered in order to duly account for the
uncertainty associated to the use of such methodologies. Furthermore, nonlinear dynamic analyses are
carried out as the baseline method for this parametric study and conclusions are obtained according to
direct comparison. Based on the results, N2 has been found as the method providing the best balance
between accuracy and complexity.

Keywords: fragility, bridges, nonlinear static procedures, dynamic analysis

Acknowledgements
ii

ACKNOWLEDGEMENTS

I would like to express my deepest appreciation to my supervisors, Dr. Ricardo Monteiro and Dr. Rui
Pinho, for their guidance, advice and kind help throughout the entire study. I would like to express my
appreciation to Vitor Silva for his support and help during this study. I also want to thank all of my
friends of the ROSE Programme at the UME SCHOOL for their kindness and friendship. Last but not
the least, I want to thank my family for their endless love and encouragement.

Index
iii

TABLE OF CONTENTS

Page
ABSTRACT ............................................................................................................................................. i
ACKNOWLEDGEMENTS .................................................................................................................... ii
TABLE OF CONTENTS ....................................................................................................................... iii
LIST OF FIGURES .............................................................................................................................. vii
LIST OF TABLES ................................................................................................................................. xi
1 General Introduction .......................................................................................................................... 1
1.1 Featuring introduction ................................................................................................................ 1
1.2 Object and goal .......................................................................................................................... 4
1.3 Thesis outline ............................................................................................................................. 4
2 State-of-the-art and literature review ................................................................................................. 6
3 Classification and Generation of Bridge Populations ...................................................................... 10
3.1 Bridge characterization ............................................................................................................ 10
3.2 Random generation of bridge models ...................................................................................... 11
4 Bridge Models ................................................................................................................................. 12
Index
iv
4.1 Selection of nonlinear analytical software ............................................................................... 13
4.2 Materials ................................................................................................................................... 14
4.2.1 Concrete Modal .............................................................................................................. 14
4.2.2 Steel Model .................................................................................................................... 17
4.3 Abutment .................................................................................................................................. 18
4.3.1 Equivalent linear spring abutments ................................................................................ 18
4.3.2 Diaphragm abutments .................................................................................................... 18
4.3.3 Seat abutments ............................................................................................................... 19
4.4 Fibre Elements ......................................................................................................................... 21
4.5 Deck ......................................................................................................................................... 23
4.6 Mass and Load ......................................................................................................................... 23
4.7 Damping ................................................................................................................................... 23
5 Ground Motion Records .................................................................................................................. 26
5.1 Italian seismicity ...................................................................................................................... 26
5.2 Considered Records ................................................................................................................. 27
5.2.1 Approach 1Non-scaled records .................................................................................. 28
5.2.2 Approach 2Scaled Records ........................................................................................ 29
6 Damage Limit States ........................................................................................................................ 34
6.1 Literature Review ..................................................................................................................... 34
6.2 Selection of limit states ............................................................................................................ 40
7 Nonlinear Static Procedures ............................................................................................................ 43
7.1 Pushover Analysis .................................................................................................................... 43
Index
v
7.1.1 Conventional Pushover .................................................................................................. 43
7.1.2 Adaptive Pushover ......................................................................................................... 44
7.2 Nonlinear Static Procedures ..................................................................................................... 46
7.2.1 Capacity Spectrum Method (CSM) ............................................................................... 47
7.2.2 N2 48
7.2.3 Displacement Coefficient Method (DCM) .................................................................... 50
7.2.4 Modified Modal Pushover Analysis (MMPA) .............................................................. 51
7.2.5 Adaptive Capacity Spectrum Method (ACSM) ............................................................. 53
7.2.6 Adaptive Modal Combination Procedures (AMCP) ...................................................... 55
7.2.7 Modified Adaptive Modal Combination Procedures (MAMCP) .................................. 56
7.3 Overall Procedures ................................................................................................................... 57
7.4 Comparison with nonlinear dynamic analysis ......................................................................... 57
8 Results 58
8.1 Introduction .............................................................................................................................. 58
8.2 Bridge Index (BI) based on individual records ........................................................................ 59
8.2.1 Capacity Spectrum Method ............................................................................................ 60
8.2.2 N2 Method ..................................................................................................................... 60
8.2.3 Displacement Coefficient Method ................................................................................. 61
8.2.4 Modified Modal Pushover Analysis .............................................................................. 61
8.2.5 Adaptive Capacity Spectrum Method ............................................................................ 62
8.2.6 Adaptive Modal Combination Procedure ...................................................................... 62
8.2.7 Modified Adaptive Modal Combination Procedure ...................................................... 63
Index
vi
8.2.8 Global Results ................................................................................................................ 63
8.3 Bridge Index (BI) based on spectrum-matching scaled records .............................................. 67
8.4 Fragility Curves ....................................................................................................................... 73
8.4.1 Nonlinear Static Procedures ........................................................................................... 73
8.4.2 Nonlinear Dynamic Analysis ......................................................................................... 76
8.4.3 Comparison of results .................................................................................................... 76
9 CLOSING REMARKS .................................................................................................................... 79
9.1 Conclusions .............................................................................................................................. 79
9.2 Future Recommendation .......................................................................................................... 81
10 REFERENCES ................................................................................................................................ 83

Index
vii

LIST OF FIGURES

Page
Figure 1.1. Damage to bridges during the San Fernando, 1971 and Loma Prieta, 1989
earthquakes ......................................................................................................................... 1
Figure 1.2. Sample fragility curves for different damage state[Avsar,2009] ............................. 2

Figure 4.1. Models for seismic bridges analysis[Priestley et al, 1996] .................................... 12
Figure 4.2. Transverse and longitudinal directions in bridges modelling ................................. 13
Figure 4.3. Configuration of bridge .......................................................................................... 13
Figure 4.4. Kent-Scott-Park concrete model for unconfined and confined concrete ................ 14
Figure 4.5. Mander concrete model .......................................................................................... 16
Figure 4.6. Conventional pushover using different concrete models ....................................... 16
Figure 4.7. Displacement profile for the two concrete models ................................................. 17
Figure 4.8. Giuffre-Menegotto-Pinto Steel model .................................................................... 17
Figure 4.9. Example of distributed plasticity ............................................................................ 21
Figure 4.10 Example of fibre elements section ....................................................................... 22
Figure 4. 11 Examples of fibre elements for bridge pier .......................................................... 22
Index
viii
Figure 4.12. Nonlinear Dynamic analysis with zero damping models ..................................... 24

Figure 5.1. Distribution of magnitude of Italian historic earthquake ....................................... 27
Figure 5.2. Distribution of the PGA of those records ............................................................... 29
Figure 5.3. Spectra of the 50 records and NEHRP design spectrum ........................................ 29
Figure 5.4 PGA distributions of the 10 records ........................................................................ 31
Figure 5.5. Spectra of the 10 records and design spectrum ...................................................... 32
Figure 5.6. Median spectra and design spectrum ...................................................................... 32
Figure 5.7. Comparison between two different sets of records ................................................ 33

Figure 6.1 Member limit state and structure limit state ............................................................ 38

Figure 7.1. Pushover curves ...................................................................................................... 45
Figure 7.2 Graphical procedures of CSM ................................................................................. 47
Figure 7.3. Inelastic Spectra by N2 ........................................................................................... 49
Figure 7.4. Bilinearization of the force-displacement curves ................................................... 51
Figure 7.5 Spectral reduction methods ..................................................................................... 54

Figure 8.1. Relationship between the relative error and size of sample ................................... 59
Figure 8.2. CSM median Bridge Index per intensity level ....................................................... 60
Figure 8.3. N2 median Bridge Index per intensity level ........................................................... 60
Figure 8.4. DCM median Bridge Index per intensity level ....................................................... 61
Figure 8.5. MMPA median Bridge Index per intensity level ................................................... 61
Index
ix
Figure 8.6. ACSM median Bridge Index per intensity level .................................................... 62
Figure 8.7. AMCP median Bridge Index per intensity level .................................................... 63
Figure 8.8. MAMCP median Bridge Index per intensity level ................................................. 63
Figure 8. 9. Median Bridge Index per intensity level for all seven methods ............................ 63
Figure 8.10. Median BI for each intensity level of CSM, N2 and DCM .................................. 64
Figure 8.11. Median BI for each intensity level of CSM, N2 and DCM .................................. 64
Figure 8.12. Median Bridge Index for each NSP ...................................................................... 65
Figure 8.13. Median Bridge Index for each PGA ..................................................................... 65
Figure 8.14. Median Brige Index for N2, MMAP and MAMCP ............................................. 66
Figure 8.15. Acceleration spectra ............................................................................................. 67
Figure 8.16. CSM median Bridge Index per intensity level based on scaled spectrum ........... 68
Figure 8.17. N2 median Bridge Index per intensity level based on scaled spectrum ............... 68
Figure 8.18. DCM median Bridge Index per intensity level based on scaled spectrum ........... 68
Figure 8.19. MMPA median Bridge Index per intensity level based on scaled spectrum ........ 69
Figure 8.20. ACSM median Bridge Index per intensity level based on scaled spectrum ......... 69
Figure 8.21. AMCP median Bridge Index per intensity level based on scaled spectrum ......... 69
Figure 8.22. MAMC median Bridge Index per intensity level based on scaled spectrum ....... 70
Figure 8.23. Median Bridge Index per intensity level based on scaled spectrum for each
method ............................................................................................................................... 70
Figure 8.24. Median BI for each intensity levels of CSM, N2 and DCM ................................ 71
Figure 8.25. Median BI for each intensity levels of MMPA, ACSM, AMCP and MAMCP ... 71
Index
x
Figure 8.26. Average values for each intensity ......................................................................... 71
Figure 8.27. Median Bridge Index for each NSPs .................................................................... 72
Figure 8.28. Median Bridge Index for each NSPs with two approaches .................................. 72
Figure 8.29. Fragility curves of NSPs ....................................................................................... 75
Figure 8.30. Fragility curves for nonlinear dynamic analysis .................................................. 76
Figure 8.31. General results of each method for each Limit states ........................................... 77

Index
xi

LIST OF TABLES

Page

Table 2.1. Description of three different bridge configurations by Banerjee and Shinozuka
[2008] .................................................................................................................................. 7
Table 2.2. Damage states limits for left most columns by Banerjee and Shinozuka [2008] ...... 8
Table 2.3. Structure attributes for generating bridge sample by Avsar et al [2011] ................... 8

Table 3.1. Distribution of material and geometry properties .................................................... 10

Table 4.1 Deck cross section geometric and material properties .............................................. 23
Table 4.2. Different damping factors in OpenSees ................................................................... 25
Table 4.3. Different damping factors in SeismoStruct ............................................................. 25

Table 5.1. Characteristics of scaled of ground motion records ................................................. 30

Table 6.1. Damage states given by HAZUS [2003] ................................................................. 34
Table 6.2. Damage states from Hwang et al [2001] ................................................................ 35
Table 6.3 Damage states based on ductility by Hwang et al [2001] ......................................... 35
Index
xii
Table 6. 4. Ductility limits for weak piers and strong bearings ................................................ 36
Table 6. 5. Drift limits from Basoz and Mander [1999] ........................................................... 36
Table 6. 6. Damage states defined by Choi [2004] ................................................................... 37
Table 6. 7. Limit states by J. Kowalsky [2000] ........................................................................ 38
Table 6. 8. Medians and dispersions for bridge component limit states using Baysian updating
........................................................................................................................................... 39
Table 6. 9. Damage states from Karakostas et al [2006] .......................................................... 40
Table 6. 10. Four damage limit states from Neilson et al [2005] ............................................. 41
Table 8.1. Median BI of seven approaches for per intensity level ........................................... 66
Table 8.2. Bridge Index for each intensity ................................................................................ 72
Table 8.3. Statistics of fragility functions by NSPs .................................................................. 74
Table 8.4 Statistics of fragility functions by Nonlinear Dynamic Analysis ............................ 76
Chapter 1 General Introduction
1

1 Introduction
1.1 General scope
Earthquakes have caused severe damage, ever since, to human populations, buildings and
society in general. During notable past earthquake events, which happened in seismic prone
regions all around the world, such as the 1971 San Fernando earthquake in the US, the 1994
Northridge earthquake in the US, the 1995 Kobe earthquake in Japan, the 1999 Marmara
earthquake in Turkey, the 1999 chichi earthquake in Taiwan, the 2008 Wenchuan earthquake
in China, the 2009 LAquila earthquake in Italy, the 2011 Christchurch earthquake in New
Zealand, etc. It has been witnessed how bridges are one of the most vulnerable components of
structural network in general, due to the gravity effect of its weight only or to inadequate
lateral forces based design. Figure 1.1 shows the severe damage to bridges in the earthquake
events. Consequently, many countries have revised their seismic hazard levels by increasing
the design peak ground acceleration and reliable seismic vulnerability assessment of bridges
and determination of their performance under seismic actions are more and more seen as
critical.

Figure 1.1. Damage to bridges during the San Fernando, 1971 and Loma Prieta, 1989 earthquakes
2
After characterizing the seismic hazard for a specific region by means of probabilistic or
deterministic approaches, vulnerability assessment can be carried out based on both seismic
information and proper methodologies, which aim at deriving reliable fragility functions.
Besides it is one of the most critical aspects in the seismic loss assessment studies, fragility
functions are also widely applied for global failure probability calculations and other
assessment studies. Fragility functions play a significant role in defining the failure
probability of bridges, however, notwithstanding the extensive available knowledge on the
matter, evidenced by a large number of past and recent studies, there are still many
uncertainties surrounding bridges fragility functions coming whether from the different
structural and material parameters of bridges or the procedures used for their derivation. In
order to minimize such uncertainties, the most reliable methods should be followed. Figure
1.2 shows a sample of fragility curves with different damage states.

Figure 1.2. Sample fragility curves for different damage state[Avsar,2009]
Fragility functions are a fundamental component to determine vulnerability, within risk
assessment procedures, and can be obtained mainly by empirical or analytical methodologies.
When following an empirical approach, information about bridge damage is collected from
reconnaissance reports or site surveys. According to the damage level information and bridge
classes, fragility curves can be derived by statistical analysis. Given that the damage states are
identified by means of real structural or non-structural damage information after an
earthquake, the empirical approach can be considered the most realistic method. However,
3
significant drawbacks can be associated to empirical methodologies. Since all the collected
data is based on observation, the subjectivity is another issue that can not be forgotten, so as
the potential lack of accuracy in the determination of the ground motion in a certain region.
Moreover, there are only a few places in the world where post-earthquake damage data has
been collected.
When there is no damage data or expert opinions, an empirical approach is not possible to
follow, and an analytical approach becomes the only method to derive fragility curves for
bridges. As analytical approach requires the definition of a single model, believed to be
representative of a class of bridges, or a set of randomly generated bridge model, usually
making use of finite element techniques. The accurateness of the analytically determined
fragility curves generally depends on three factors: bridge structures sampling and modelling,
ground motion selection and damage state definition.
The methodologies used to derive fragility functions within an analytical approach can be
categorized in two main groups: the ones that make use of Nonlinear Static Procedures (NSP)
or Nonlinear Dynamic Analysis to estimate the structural response. It is well known that the
dynamic based procedures are typically seen as the ones that can provide the most accurate
response estimate by applying a real acceleration time history at the base of the structure.
However, the complexity of the methodology, which includes the definition of the damping
model, the post-yield behaviour model, as well as the expected large amount of computational
work, is often regarded as impractical. On the other hand, NSPs have some merits when
comparing with the complicated nonlinear dynamic analysis, such as low computational
burden, simple definition of post-yield behaviour model, etc. When considering NSPs,
pushover analysis is carried out, which is then incorporated into a nonlinear static procedure
to obtain an estimate the response displacement (or other parameters) experienced by the
structures for a given ground motion record. The critical drawbacks of this methodology lay
under the assumption that the structural performance is estimated from a structure statically
and laterally loaded.

4
1.2 Object and goal
Many tools are currently available for calculating fragility curves, especially based on
analytical approaches, which has become increasingly popular in earthquake engineering
community due to its scientific soundness and because they overcome the lack of data in
empirical approaches. Both static and dynamic nonlinear analysis can be employed for
deriving fragility curves within an analytical approach and both of them feature numerous
different possibilities of application as well as many sources of uncertainty.
In this study, different methodologies (NSPs) are considered to develop analytical fragility
functions of RC bridges. The performance of the different NSPs, when applied to a large
number of randomly generated bridge configurations, is compared with results of an extensive
number of nonlinear dynamic analyses, considered as reference, in order to identify the most
accurate and efficient method for calculating fragility functions.

1.3 Thesis outline
In order to reach the main goal of the study, each chapter will correspond to every aspect
considered to be relevant within seismic fragility curves of bridges hence this thesis is
composed of seven main chapters, defined as follows:
Chapter 1: General introduction and the outline of the thesis
Chapter 2: State-of-art and literature review
Chapter 3: Classification of bridges based on statistical analysis.
Chapter 4: Selection of analysis tools and generation of 3D nonlinear models based on the
previous studies.
Chapter 5: Selection of earthquake ground motion records. In this part, two different sets of
records were used representing scaling records and unscaling records
respectively.
Chapter 6: Based on the previous studies, choose the proper definition of damage limit
states for calculating bridge fragility functions.
5
Chapter 7: Calculation of seismic performance of bridges using both nonlinear static (NSPs)
and dynamic analysis. Statistical comparison of different engineering demand
parameters and fragility curves obtained through NSPs or nonlinear dynamic
analysis.
Chapter 8: Results of all those analytical approaches.
Chapter 9: Conclusions and future developments.
Chapter 2 State-of-the-art and Literature Review
6
2 State-of-the-art and Literature Review
In the past few decades there has been a significant research effort regarding fragility
functions for buildings whereas, clearly, less thorough investigations have been carried out for
the study of bridges. Fragility functions of bridges is thus a topic that still undergoes
significant development and improvement, particularly in what concerns the classes of
bridges under scrutiny and subsequent loss assessment studies.
Hwang et al [2001] presented an analytical approach for deriving fragility curves for highway
bridges. In this study, a set of earthquake-site-bridge sample were set up to take the
uncertainties from ground motion records, site conditions and bridges into account. The
failure probabilistic characteristics of structural demand were presented as a function of the
correlated ground motion records in terms of spectra acceleration or peak ground acceleration.
In order to calculate the fragility curves, quantitative bridge damage states were defined
according to qualified description given by HAZUS.
Karim and Yamazaki [2003] proposed a simplified method for the definition of fragility
curves for highway RC bridges. Four typical bridge piers and two bridge configurations were
considered and a set of 250 strong ground motion records were selected for nonlinear
dynamic analysis. Fragility curves were obtained using damage indices and ground motion
parameters based on lognormal distribution. Based on the observed relationship between the
fragility curves and the over-strength ratio, a simplified mathematic equation was proposed
for calculating fragility curves by such simplified method.
In a study of Nielson [2005], an expanded methodology for calculating fragility curves of
highway bridges was carried out, which directly estimated the bridge system fragility curves
from component fragilities. OpenSees were used to build the 3D bridge finite element models,
and the bridge classification was based on different bearings and abutments conditions.
Damage limit states for each damage state were defined for different elements according to
different criteria. A Bayesian approach was also applied for dealing with the uncertainties of
the damage states. The conclusion drawn from this study was that the global fragility of
bridge system is higher than any of its individual components.
Kappos et al [2006] introduced an advanced inelastic analysis tools for calculating fragility
curves for bridges. The main steps of the newly developed method included modal pushover
7
analysis based on appropriately selected monitoring points for drawing the pushover curves
and utilizing the capacity spectrum method, and nonlinear dynamic analysis accounting for
the influence of spatial variability of earthquake ground motion on both straight and curved
bridges.
In a study of Karakostas et al [2006], the methodology proposed by Kappos et at [2006],
which took use of pushover analysis and inelastic demand spectrum, for evaluating fragility
curves of bridges is employed for the case of Northern Greece. In this study, a specific bridge
was chosen and the related 3D model was built with finite element program Sap2000. A
pushover analysis was performed for both longitudinal and transverse directions using 1
st

mode shape load distribution. Five damage states were defined in terms of damage ratio
y
D ! ! / = , where ! is the response displacement calculated from the nonlinear static method,
and !
y
is correlated to the yield displacement. The mean expected peak ground acceleration
correspond to each damage limit state was assessed as the point of intersection between the
capacity curve and the demand spectrum. Finally, fragility curves were computed based on
lognormal probability distribution functions.
In a study of Jeong and Elnashai [2007], analytical approaches for deriving fragility functions
were carried out based on the fundamental response quantities of stiffness, strength and
ductility. Single degree of freedom systems were generated to construct a Response Database
of coefficients for defining fragility curves. The uncertainties of modelling simplification
were checked with the more accurate multi-degree of freedom systems.
In a study of Banerjee and Shinozuka [2008], fragility curves for RC bridges were derived
based on mechanistic quantification of damage states. Table 2.1 shows the geometric
parameters of the different three bridges considered in this study.
Table 2.1. Description of three different bridge configurations by Banerjee and Shinozuka [2008]

8
Time history response of the different tested structures was measured in terms of rotational
ductility at the end of columns. Five different damage states were defined by terms of yield
and ultimate rotations obtained from bi-linear moment-rotation plot. Table 2.2 displays the
damage limit states used in this study.
Table 2.2. Damage states limits for left most columns by Banerjee and Shinozuka [2008]

In a study from Avsar et al [2011], analytical fragility curves for ordinary highway bridges
were calculated. Four major bridge classes were considered based on the skew angle, number
of columns per bent and span numbers. Nonlinear dynamic analysis was employed for each
3D bridge model prepared with OpenSees platform. Damage states criterion for structure
critical components was defined for deriving fragility functions. Table 2.3 shows the
geometric and material parameters considered in the 3D bridge modelling.
Table 2.3. Structure attributes for generating bridge sample by Avsar et al [2011]

9
Generally, there are several methodologies for calculating fragility functions and the resulting
curves are conditional on the assumptions and approaches followed in the process, which
includes the parameters used in modelling, definition of the damage states, computational
approaches, etc. These difference caused by different methodologies will lead to critical
discrepancies in the seismic assessment, even when considering the same region. Besides the
limitations of the previous studies are that only unique methodologies was applied in one
study and the lack of comparison among those methodologies. In this study, in order to
improve those drawbacks, several analytical methodologies are employed to calculate fragility
functions for the same bridge type in the same region. Nonlinear dynamic analysis is
calculated as the baseline for the sensitive study, seven different NSPs are applied to make
comparison and to yield the conclusion of the difference among those different methodologies.
Chapter 3 Classification and Generation of Bridge Populations
10
3 Classification and Generation of Bridge Populations
3.1 Bridge characterization
The first step to adopt when calculating fragility curves for a seismic loss assessment study of
a population of bridges can be the classification of the bridges into typological classes, based
on simple parameters that characterize their configuration and influence their seismic
response. In their study, Moschonas et al [2009] categorized all the bridges based on the
characteristics of the piers, bearings and pier-to-bearing connections. In other studies, such as
the one of Avsar [2009], other parameters with significant influence on seismic response of
bridges have been proposed for the classification, which include skew angle, number of spans
and number of bent columns. Afterwards, for each chosen parameter, a statistical probabilistic
distribution can be defined and used for random generation of populations of bridges. The
most typically used distributions are normal, log-normal, gamma, beta or exponential
distributions.
In this study, in order to keep the computational time and effort in a sustainable range, a
single geometrical configuration of bridges with four bays was considered. Nevertheless, the
length, height and configuration of the piers were all submitted to variation, according to pre-
defined distributions. In addition, uncertainty was also considered to come from the material
and other geometrical properties of the bridges, according to the distributions presented in
Table 3.1.
Table 3.1. Distribution of material and geometrical properties

The distributions assumed for the different variables intend to represent a scenario of typical
RC bridges and viaducts in Italian. The initial idea behind this study was to use distribution
characterization obtained by statistical analysis of data of real Italian bridges, which for
several bureaucratic/administrative reasons, was not obtained in time for this work.
Parameters Mean Cov lower bound upper bound Type of distribution
Steel modulus(Gpa) 200 3% - - Normal
Steel yield strength(MPa) 371.1 11% 250 - Normal
Concrete strength(MPa) 40 20% 20 70 Gamma
Column height(m) 15 15% 8 25 Normal
Column diameter(m) 2 14% 1.2 4.79 Lognormal
bay length(m) 50 20% 30 80 Normal
beamcap width(m) 5 4% 2 20 Lognormal
beamcap height(m) 1 2% 1 4 Lognormal
Chapter 3 Classification and Generation of Bridge Populations
11
3.2 Random generation of bridge models
Once the statistical distributions for all the relevant parameters have been defined, the bridge
models can be sampled either following Monte Carlo or Latin Hypercube approaches.
In order to get a reliable sample, which is able to represent the probability distribution as good
as possible, a proper simulation procedure should be applied. As we all know it is the size of
the sample that the appropriateness of the simulation procedure mainly depends on. Among
the available solutions, the most commonly used method is the Monte Carlo approach based
on pure random simulation. However, a more recent and innovative technique is the Latin
Hypercube, which has been initially proposed by Mckay et al [1979], and further developed
by Iman et al [1981]. One of the most important claims of the Latin Hypercube approach is
that it reduces the size of the sampling efficiently.
In this study, an automatic framework was created to randomly simulate the selected
parameters (material and geometrical) for each individual bridge. As a first step, parameters
are randomly generated based on the distributions defined above using Monte Carlo
simulation scheme.
Chapter 4 Bridge Models
12
4 Bridge Models
Proper analytical modelling of bridges is of great importance when calculating seismic
fragility curves and it still encloses a number of challenging issues, among which some need
to be further developed and others call for reasonable assumptions. Special attention should
be paid when making simplified assumptions since any simplification will have a direct
impact on the analysis and, consequently, on the reliability of resulting fragility curves. On
the other hand, an excessively refined or complex model will lead to higher computational
demand or even unreachable results. A model should therefore be as simple as possible, while
retaining its soundness.
Priestley et al [1996] suggested various models with different complexity levels for seismic
analysis of bridges, ranging from lumped mass models to finite element models, as shown in
Figure 4.1.

Figure 4.1. Models for seismic bridges analysis [Priestley et al, 1996]
Given that seismic action generally causes much higher damage along the transverse
direction, rather than the longitudinal one, of bridges, 3D models are used for the analysis of
the transverse deformation under different ground motion levels. Figure 4.2 gives the example
of transverse and longitudinal direction of bridges.
13

Figure 4.2. Transverse and longitudinal directions in bridges modelling [Priestley et al, 1996]
Several key aspects, with respect to the modelling task, will be discussed in this chapter, e.g.,
concrete and steel material models, abutment stiffness calculation, fibre elements, lumped
masses and loads, amongst others.
4.1 Selection of nonlinear analytical software
In this study, OpenSees[1] has been employed to carry out all the static and dynamic
nonlinear analysis required by the analytical procedures. OpenSees, that is Open System for
Earthquake Engineering Simulation, which is developed by U.C. Berkeley and funded by
Pacific Earthquake Engineering Research Center (PEER), is widely applied in the structural
or soil seismic response study. At a preliminary stage, its estimates have been compared and
double checked with a different, equally widespread, software program: SeismoStruct. Both
of the software tools enable 3D bridge modelling using force-based fibre elements with
distributed nonlinearity. Figure 4.3 shows the configuration of the bridge model used for the
initial calibration study.

Figure 4.3. Configuration of a bridge model
14
The reason for choosing OpenSees to perform this study is that OpenSees could provide an
automatic computation platform, which allows the calculations to be performed according to a
script automatically without changing parameters manually. That is a clearly fundamental
feature in a random simulation based work.

4.2 Materials
4.2.1 Concrete Modal
Several concrete models are available in literature, with different accuracy levels. Aktan and
Ersoy [1979] pointed out that the concrete model is not the most significant aspect, when
considering the moment-curvature diagram of a reinforced concrete section, thus stating that it
should not be too important to choose a more sophisticated model. It is typically the steel
model, rather than the concrete, that plays a critical role in determining the ductility behaviour
for a given section.
When analysing a reinforced concrete section, there are two different types of concrete to be
considered : confined and unconfined concrete.
The section core concrete, which is confined by transverse reinforcement, is different from the
cover concrete, in terms of stress-strain relationship. OpenSees Model Concrete01
represents a uniaxial Kent-Scott-Park concrete material object with degraded linear
unloading/reloading stiffness according to the work of Karsan-Jirsa [1969] and no tensile
strength, which can be used for both confined and unconfined concrete. The strain-stress
relationships for both confined and unconfined concrete model are illustrated in Figure 4.4.

Figure 4.4. Kent-Scott-Park concrete model for unconfined and confined concrete
15
According to the plots in the figure above, the parameters needed to characterize the confined
concrete behaviour can be calculated using equations 4.1 to 4.3,
co
yh s
f
K
!
"
+ = 1

(4.1)
K
cco
! = 002 . 0 " and K
co cco
! =" "

(4.2)
cco
h
s
co
co
s
h
Z
! "
#
#
$ +
$
+
=
'
75 . 0
1000 145
29 . 0 3
5 . 0

(4.3)
where K is a parameter related to the limit strain and increasing compression strength due to
confinement. Z is the strain softening slope.
The slope corresponding to the initial stiffness ,E
c
, can be calculated through equation 4.4,
co pc c
f E ! / 2" =

(4.4)
Assuming the maximum concrete strain as !
c0
=0.002, E
c
can be calculated directly based on
the given compression strength.
On the other hand, SeismoStruct make use of Mander [1988] concrete model, for which the
strain-stress relationship is depicted in Figure 4.5. The modulus of elasticity associated to the
initial stiffness branch of the material behaviour is defined by Equation 4.5, with units in MPa.
'
5000
c
c
f E = (4.5)
16

Figure 4.5. Mander concrete model
Again, a brief comparison between the two selected software tools, considering the nonlinear
behaviour of the concrete material, has been carried out, in terms of pushover curves of a
bridge with a simple geometrical configuration. The results obtained with OpenSees and
SeismoStruct, employing their corresponding concrete models, are shown in Figure 4.6,
which denotes how the stiffness associated to Manders concrete model degrades slower than
when applying Kent-Scott-Park model.

Figure 4.6. Conventional pushover using different concrete models
In Figure 4.6, the uniform distributed load pushover analyses give much higher base shear
than the parabolic distributed load pushover analyses in both SeismoStruct and OpenSees
results. Moreover the pushover curves from the two software are followed the same changing
17
trend under two different load patterns. However, as the two software make use of two
different concrete types, there are discrepancies between the pushover curves derived from the
two software even under the same distributed load pattern. In order to check whether the
different concrete material will lead huge difference in displacement profile, Figure 4.7 is
generated to show the bridge displacement profile at different performance stage.

Figure 4.7. Displacement profile for the two concrete models
From the observation of figure 4.7, one can easily realize how the displacement profiles
obtained with different concrete models show much more acceptable differences between
them, when compared to the force-displacement curves in Figure 4.6. Given that OpenSees
provides a sophisticated and stable analysis algorithm for Kent-Scott-Park model, the
Concrete01 model has been chosen for all the models and analyses to be carried out with
OpenSees.
4.2.2 Steel Model
Reinforcement bars are modelled with Steel01 in OpenSees, which makes use of the well-
known Giuffre-Menegotto-Pinto model with isotropic strain hardening. The hysteretic
behaviour of the steel model is depicted in Figure 4.8.

Figure 4.8. Giuffre-Menegotto-Pinto Steel model
18
Loading and unloading paths are involved in the monotonic loading bi-linear behaviour law
and the stress-strain relationship expression is defined by Equation 4.6.
*
/ 1 *
*
*
) 1 (
) 1 ( !
!
!
" b b
R R
+
+
# =

(4.6)
4.3 Abutments
Abutments are usually detailed and designed for service loads and then checked for seismic
performance. There are different kinds of abutments applied to the bridges structures and a
short description of the most commonly employed types is carried out in what follows.
4.3.1 Equivalent linear spring abutments
Commonly, equivalent linear springs are used in structural models to simulate the deck
restrains provided by the abutments. The characterization of the equivalent springs should
represent the dynamic behaviour of the soil behind the abutments, the structure components of
the abutments and the interaction between them.
In design applications, stiffness values of these springs are usually determined based either on
simplified rules and iterative process or from abutment capacity and expected deformation
during the earthquake.
4.3.2 Diaphragm abutments
Diaphragm abutment is one of the most popular bridge abutments types, which is good at
absorbing energy during an earthquake. The longitudinal resistance for the seismic analysis of
a diaphragm abutment should be based on mobilizing the backfill equal to the depth of the
superstructure plus the shear capacity of the abutment diaphragm. The ultimate passive
resistance can be assumed by CALTRANS (California Department of Transportation) as
370KPa. The transverse resistance for seismic analysis of a diaphragm abutment should be
based on the ultimate shear capacity of one wingwall and all piles. When Class 400, Class 625
or standard 400mm CIDH piles are used an ultimate shear capacity of 180KN per pile may be
used. The ultimate shear capacity of the piles is the limiting force for transverse keys for
diaphragm abutment. To reduce the possible damage to the piles, transverse keys should be
designed for 75% of the limiting force.
19
4.3.3 Seat abutments
Another commonly used type of abutments is seat abutment, which gives the designers more
control over the amount of forces the abutments could resist, but introduces the potential for
the superstructure becoming unseated leading to collapse of the end span. The longitudinal
stiffness assumed for the seismic analysis should be based on mobilizing only the soil equal to
the depth of the superstructure. The limiting force for the transverse shear keys may be
approximated by adding the ultimate shear capacity of one wingwall plus the ultimate shear
capacity of the piles.
Priestley et al [1996] proposed that all the aforementioned three types of abutments should
follow the same properties:
1. They are massive structures;
2. Mobilize and interact with large soil masses;
3. Based on their geometry, exhibit significantly higher stiffness values than do other bridge
bents and thus attract proportionally higher seismic forces;
4. Feature some or all of the following highly nonlinear elements and behaviour
characteristics: breakaway shear keys, expansion joint restrainers, sacrificial wing and
back walls and a potential for inelastic pile action.
In the CALTRANS procedure, the design value for abutment capacity in longitudinal
direction is computed as the sum of the resistance values provided by the foundation and the
soil behind the backwall, which utilize an abutment capacity based on a maximum effective
soil pressure of 239kPa, amplified by about 50% to 368kPa for dynamic seismic loads,
capacity levels recently verified by large-scale abutment tests. For the projected abutment area
A
eff
in the loading direction, the nominal dynamic abutment capacity, F
abutment
, can be
determined using Equation 4.7,
eff abutment
A kPa F = 368

(4.7)
The design value of abutment stiffness is calculated as a ratio of the abutment capacity and
acceptable deformation. The backwall in longitudinal and shear key in transverse direction are
20
generally designed as sacrificial element to protect the whole structures against the inelastic
action.
CALTRANS guidelines set the initial abutment stiffness estimate according to Equation 4.8,
p p s a
n B k k k 7 115 + = + =

(4.8)
where B represents the effective abutment width and n
p
is the number of piles. In the
transverse direction the effective width B is taken as the length of the wing walls multiplied
by a factor of 8/9 to account for differences in participation of both wing walls. The
displacement at the abutments may be underestimated by this approach unless the secant
stiffness or substitute structure approach is extended to the entire system including equivalent
bent characteristics.
Goel and Chopra [1997] concluded that the values proposed by CALTRANS procedures are
too high for both transverse and longitudinal directions. Reliable values to represent the
interaction of abutment and soil were instead estimated from the force-deformation relations
of the system soil-abutment by dynamic equilibrium of the road deck. Hysteretic loops were
applied for the analysis and provided the results of 7500 and 12000 kip/ft, which are equal to
109,454kN/m and 175,127kN/m for positive and negative deformation respectively. The same
study also proposed the stiffness of abutments sitting directly on piles to be 709,150kN/m and
229,970kN/m in longitudinal and transverse direction respectively.
In the study of Alvarez [2004] and J.C. Oritiz Restrepo [2007], 75000kN/m was suggested for
the abutment stiffness with limited displacement of 100mm.
Casarotti et al [2005] considered, in their study, two kinds of abutments. In one of them the
abutment is directly sitting on top of the piles, for which a bi-linear behaviour was assumed
for modelling purposes. The pre-yielding stiffness was defined as 229,970kN/m for transverse
direction and 0.5% post-yielding stiffness ratio, when the displacement exceeded the design
value of 25mm. The other considered type is when the abutment sits on top of pot bearings,
for which elastic behaviour was assumed as 26,329kN/m.
The work presented herein did not have the scope of going deep into abutment design but
rather choosing a reasonable model to define the abutments of the bridges used in the
extensive fragility curve parametric study. Indeed, no uncertainty related to the abutments has
21
been considered within the random simulation phase (see Table 2.1), although it should be
seen as one of the future developments (see chapter 9). Accordingly, for this study, the type of
abutments considered has been the one sitting on top of pot bearings, allowing a maximum
elastic deformation of 500mm. The stiffness of the abutments was set as 26,329kN/m,
according to what proposed in the study of Casarotti et al [2005].

4.4 Fibre Elements
According to the types of the plasticity formulation, nonlinear elements can be divided into
two major groups: lumped plasticity elements and distributed plasticity elements. The latter
one, shown in Figure 4.9, can be derived through displacement-based formulation, force-
based formulation and mathematic integration.

Figure 4.9. Example of distributed nonlinearity
OpenSees provides a powerful library of possible fibre elements shown in Figure 4.10, which
will have a significant importance in modelling properly the distributed material nonlinearity.
22

Figure 4.10 Example of fibre elements section
The displacement and flexibility matrix of the force-based elements can be calculated as per
Equations 4.9 and 4.10.
!
= "
L
s T
Q
e
dx xe N
0

(4.9)

!
=
L
Q
s T
Q
e
xdx xN xf N f
0

(4.10)
where s stands for section and e stands for element, N
Q
is the force shape function.
Those equations are the exact solution for beam theory independent of the material properties.
In this study, force-based elements (nonlinear beam-column element) have been adopted to
accurately represent the inelastic deformation of the analysed bridges. Figure 4.11 shows the
definition of the fibre elements section.

Figure 4. 11 Examples of fibre elements for bridge pier
23
4.5 Deck
Decks are typically modelled as linear elastic elements, which represent the expected
behaviour of this structural component under seismic actions. Given that it is often made up
of pre-stressed, rather than regular reinforced concrete, no damage or nonlinear deformation is
expected to occur.
In this study, the deck was modelled with a 3D elastic element, whose characteristics are
described in Table 4.1, capable of reproducing local geometric properties.
Table 4.1 Deck cross section geometric and material properties

4.6 Mass and Load
Masses and gravity loads have been defined considering the reinforcement and concrete parts
of the bridges as well as the relative density of asphalt, thus self-weight only. Indeed,
according to common design codes and guidelines, e.g. AASHTO LRFD [2007] and EC8
[2005], as far as bridge analysis is concerned, no traffic live loads are to be considered when
carrying out seismic response calculations.

4.7 Damping
Damping is known to be as controversial as important within nonlinear dynamic analysis
hence a damping model should be chosen and employed carefully.
Generally Rayleigh damping is widely used when defining system damping, which can be
calculated as Equation 4.11,
K
i
M
i
i
!
"
!
"
#
2 2
1
+ =

(4.11)
Where "
i
is the frequency of the correlated mode, #
M
and #
K
are the damping factors for mass
matrix and stiffness matrix respectively.
EI2(KNm2) EI3(KNm2) GJ(KNm2)
1.32E+08 1.32E+08 4.75E+09
24
However, as Joan F. Hall [2005] pointed out that Rayleigh damping typically using the initial
linear stiffness matrix for the stiffness-proportional part damping consistently overestimates
the damping ratio for the system, especially when the structure has softening nonlinearity, the
damping forces generated by such matrix can become unrealistically large compared to the
restoring forces. In order to overcome the drawbacks of Rayleigh damping, Priestley and
Grant [2005] suggested that the tangent-stiffness proportional damping could be applied as
Equation 4.12,
!
"
#
T
K
=

(4.12)
A first calibration of OpenSees and SeismoStruct, in terms of dynamic analysis, is carried out
using models with no damping and the same parameters, except for the concrete model,
although both concrete models have the same compression strength. Figure 4.12 displays the
nonlinear dynamic analysis results obtained for those two models. The peak displacement
occurred at the same time and the difference between the two curves can be disregarded.

Figure 4.12. Nonlinear dynamic analysis with zero damping models
In OpenSees, although the command for defining damping is called Rayleigh, damping
factors are defined separately for mass matrix and stiffness matrix, while stiffness matrix
affected damping plays a more important role.
Table 4.2 presents different maximum displacement observed using the same geometrically
simple bridge model as previous examples when using different damping factors, applied to
mass and stiffness matrices. The column named Ratio shows the ratio between the
25
maximum displacement deriving from each damping combination and the maximum
displacement measured when no damping is considered.
Table 4.2. Different damping factors in OpenSees

From Table 4.3, one can say that the mass matrix gives less critical influence in the structural
damping analysis, when compared to the zero damping model. The maximum displacement
with stiffness damping of 2% decreases to 95% while by increasing the mass damping of 1%,
the maximum displacement only drops by 1%. Moreover 5% stiffness damping will lead to
more than 10% decrease in maximum displacement.
By carrying out the same brief parametric study on damping with SeismoStruct, Table 4.3 is
obtained when choosing Rayleigh damping. The column Mode1 and Mode2 values are the
damping coefficients applied in Rayleigh damping when considering the same mass and
stiffness proportional damping listed in the second and third columns. The results show that
the displacement is smaller with respect to OpenSees, which means that the amount of
estimated damping is slightly larger but still comparable.
Table 4.3. Different damping factors in SeismoStruct

For this study, stiffness proportional damping was employed and the damping coefficient
used in Opensees was 2% for stiffness matrix in order to minimize the effect of damping and
keep it in a reasonable range.
Rayleigh Damping Mass Matrix Stiffness Matrix Max displ(m) Ratio
ok(%) 1 5 0.282 87%
ok(%) 0 5 0.2844 88%
ok(%) 1 2 0.3053 94%
ok(%) 0 2 0.3081 95%
ok(%) 0 0 0.3244 100%
OpenSees
Rayleigh Damping Mass Matrix Stiffness Matrix Max displ(m) Ratio Mode1 Mode2
ok(%) 1 5 0.2699 82% 2.6 3.1
ok(%) 0 5 0.2683 81% 2.6 0
ok(%) 1 2 0.3034 92% 1.05 1.26
ok(%) 0 2 0.3034 92% 1.05 0
ok(%) 0 0 0.3307 100% 0 0
SeismoStruct
Chapter 5 Ground Motion Records
26
5 Ground Motion Records
Ground motion records are generally recognized as the essential input for any seismic
assessment based on nonlinear dynamic analysis. Generally, there are two kinds of records
could be applied in the dynamic analysis: the artificial records and the real records. Given that
the former one makes use of computational techniques to amplify the original records, which
will lead to changes of frequency content and energy release. On the contrary, the real records
represent better the local seismic action. Thus, real records are more commonly employed in
dynamic analysis. If the real records are considered, it is recommended that a sufficiently
large number of different records is used, in order to duly represent an intended seismic
hazard scenario, characterized by predefined ranges of parameters such as magnitude and
peak ground acceleration (PGA), most common fault failure mechanism, frequency content,
duration and epicentre distance.
Moreover, they play an important role when calculating fragility curves, given that each
ground motion record is necessarily related to specific seismic hazard and geotechnical
conditions of the region where it took place, which will induce important uncertainties in
fragility functions calculation. Consequently, in order to minimize such uncertainties, proper
care should be paid when records are chosen for analysis.
5.1 Italian seismicity
The ground motion records usually are collected in terms of the level of peak ground motion
acceleration (PGA), as well as the distance and magnitude. The majority of the strongest
earthquakes that occurred in Italy in past featured dip-slip mechanism and some of those
events were associated to strike faults. The average depth of an Italian damaging earthquake is
10km.
Based on the data from Catalogue of Strong Earthquakes in Italy [2000] from 461 B.C. to
1997, the chart in Figure 5.1 shows the distribution of historical earthquakes, in terms of
magnitude, that occurred in Italy.
27

Figure 5.1. Distribution of magnitude of Italian historic earthquake
5.2 Considered records
The parametric study presented herein considered two approaches, with respect to the
selection of ground motion records. Such approaches take the uncertainty associated to the
seismic ground motion into account in two different ways.
The first approach considers non-scaled records, directly chosen from one of the available
ground motion databases, restricted to the predefined ranges of magnitude and epicenter
distance. A sufficiently large number of records (e.g. 50 or 100) are considered, so as to duly
incorporate the epistemic/aleatory uncertainties. In addition, a uniform distribution of the peak
ground acceleration, within a representative range (e.g. 0 1g), is sought. When following
this sort of approach, each record is used only once, given that the entire range of ground
motion intensity (measured in terms of peak ground acceleration) is covered by the different
time-histories. Given that there is no need for scaling, the fact that the records keep their
original form can be seen as a major advantage. On the other hand, each ground motion
intensity level is somewhat represented by a single accelerograms hence not taking into
account record-to-record variability.
The second approach considers a smaller number of real records, chosen under spectral
matching criteria. The selected records are chosen assuring that their median (or mean)
response spectrum matches (within a given tolerance) the PSHA-based response spectrum for
the region of interest, associated to a certain probability of exceedance for a given return
period. The median response spectrum is associated to a level of ground motion intensity, i.e.,
28
peak ground acceleration thus the selected records can be scaled until the entire range of
interest, of PGAs is covered. Under such conditions, for each ground motion intensity level,
the record-to-record variability is contemplated by use of the entire set of selected records.
The main drawback is the need for record scaling.
Scaled records (Approach 2) will be used to assess the individual performance of the different
nonlinear static procedures (NSPs), i.e. CSM, N2, MMPA, ACSM, AMCP and MAMCP,
through comparison with nonlinear dynamic analysis, whereas non-scaled records (Approach
1), which lead to a lower computational onus, will be used for the extension of the parametric
study to the computation of fragility curves.

5.2.1 Approach 1Non-scaled records
The selection of real earthquake records, under the first approach, was carried out using the
PEER database [2], which is short for Pacific Earthquake Engineering Research Center. In
agreement with the historical characteristic of the Italian territory seismicity, briefly described
in Section 4.1 and Figure 4.1, the following intervals for the different parameters were used
for the selection of the records. As those records were selected based on historical magnitude
distribution, 50 real records were chosen in order to avoid changes of frequency content or
event duration, which are inherent to the magnitude of the event.
1) Magnitude Range: 5.0 7.5 Mw.
2) Distance Range: 10 150 km.
3) Rupture mechanism: Strike-slip and Dip-slip.
The distribution of peak ground acceleration and the median response spectrum of the 50
selected records are respectively shown in Figure 5.2 and Figure 5.3.
29

Figure 5.2. Distribution of the PGA of the selected records

Figure 5.3. Spectra of the 50 records and the median spectrum

5.2.2 Approach 2Scaled Records
Within the second approach, for the sake of agreement with the first one, a response spectrum
associated to a given probability of exceedance, for the Italian territory, should be considered
for the matching and selection of records. Given that this is not a clear endeavour, if one
considers the Italian territory (as considered in the ranges defined in the first approach) and
0 5 10 15 20 25 30 35 40 45 50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
G
A
(
g
)
No. of ground motion records
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.5
1
1.5
2
2.5
3
S
a
(
g
)
Period(s)
30
because the study to be carried out is essentially parametric, an already existing set of 10
records, which was used in the aforementioned study by R. Monteiro [2011], chosen from Los
Angeles Strong Motion Database, was considered instead. The employed set of records are
selected from a suite of historical ground motion data [SAC1997] scaled to match 10%
exceeding probability in 50 years (475 years return period) uniform hazard spectrum for Los
Angeles, which is known as NEHRP.
Another reason for the consideration of a second approach is the fact that many guidelines
prescribing the use of Nonlinear Static Procedures, such as Capacity Spectrum Method or N2,
recommended, in agreement with the original formulation of the methods, the employment of
smoothed design spectra.
Table 5.1 shows the detailed information of each of the 10 scaled records, including
magnitude, duration and PGA. Keeping the SAC Project designation, the 10 records are
labelled as: LA02, LA04, LA06, LA08, LA10, LA12, LA14, LA16, LA18, LA20 and those
ten records contain very large energy and frequency complexity.
Table 5.1. Characteristics of scaled of ground motion records [SAC,1997]

31
Based on Table 5.1, the main characteristics of the records can be summarized as follows:
1) Magnitude range: 6.0 7.3 Mw.
2) Distance to fault range: 1.2 36 km.
3) Peak ground acceleration: 0.2 1.0g.

Figure 5.4, in tandem with Figure 5.2 for Approach 1, plots the distribution of the peak
ground acceleration of the 10 records.

Figure 5.4 PGA distributions of the 10 records
Figure 5.5 depicts the spectra of the 10 real ground motion records and the NEHRP design
spectrum to which they have been scaled to match.
0 1 2 3 4 5 6 7 8 9 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
P
G
A
(
g
)
No. of ground motion records
32

Figure 5.5. Spectra of the 10 records and design spectrum
Finally, Figure 5.6 illustrates the comparison between the median of the 10 ground motion
records and, again, the matched NEHRP design spectrum. Figure 5.7 displays the difference
of those two different approaches.

Figure 5.6. Median spectra and design spectrum
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.5
1
1.5
2
2.5
3
3.5
S
a
(
g
)
Period(s)
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
S
a
(
g
)
Period(s)

NEHRP
Median
33

Figure 5.7. Comparison between two different sets of records
From Figure 5.7, we can see that the non-scaled records present lower spectral ordinates, with
respect to the 10 scaled records thus the different sets of records will be useful to check
whether the nonlinear static procedures could provide good estimations or not under both less
and more demanding conditions.

0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
S
a
(
g
)
Period(s)

NEHRP
non-scaled Median
scaled Median
Chapter 6 Damage Limit States
34
6 Damage Limit States
The definition of the damage states is a fundamental step when calculating fragility curves. In
addition to quantitative information, each damage state lies also on functional and operational
interpretation of the status of the structure it refers to. A large number of studies have
addressed the definition of damage states for the specific seismic assessment of bridges. A
brief overview is presented next.
6.1 Literature review
In HAZUS [FEMA, 2003], five different limit states are defined, based on damage observed
in the bridges components. Those are: None, Slight/Minor, Moderate, Extensive and
Complete, for which a detailed qualitative description is provided but, unfortunately, no
corresponding parameter quantification is not carried out. The lack of quantitative
characterization of limit states is a significant limitation to the employment of the HAZUS
approach within analytical computation of fragility curves. Table 6.1 shows the quantitive
description of the five damage states given by HAZUS.
Table 6.1. Damage states given by HAZUS [2003]

35
Hwang et al [2001] proposed two different approaches for seismic damage limits based on the
description of five damage states by HAZUS. In the first approach, illustrated in Table 6.2,
four damage states were defined according to the flexural capacity of the columns.
Table 6.2. Damage states from Hwang et al [2001]

In Table 6.2, M
1
is the column moment at the first yielding of the longitudinal bar and M
y
is
the yield moment at the idealized moment curvature curve of the given column section. !
p
is
the plastic hinge rotation with !
c
equal to 0.002 and 0.004 respectively for the columns with or
without lap splices at the bottom of the column.
In the second approach, damage states were defined using the displacement ductility ratio of
the columns to derive the overall fragility functions, which is defined by Equation 6.1,
1 cy
d
!
!
=

(6.1)
In which " is the relative displacement at the top of a column obtained from seismic response
analysis, and "
cy1
is the relative displacement of a column when the longitudinal reinforcing
bars at the bottom of the column reach the first yield. Five damage states, described in Table
6.3 were defined using the parameter of displacement ductility ratio of columns,
d
.
Table 6.3 Damage states based on ductility by Hwang et al [2001]
36

As "
cy1
is defined from the first yield of the longitudinal bars at the bottom of a column,
cy1

is equal to 1.
cy
is the yield displacement ductility ratio of a column and
c2
is the
displacement ductility ratio when !
c
=0.002.
max
is the displacement ductility ratio defined
as
c2
+3.0.
Liao and Loh [2004] proposed 4 damage states using displacement and ductility illustrated in
Table 6.4. For each damage state, ductility limits were specified for weak piers and strong
bearings by considering the design type of bridges, which was either seismic or conventional
design.
Table 6. 4. Ductility limits for weak piers and strong bearings

In the work by Basoz and Mander [1999], five damage states were defined, as represented in
Table 6.5. Drift limits were used for defining damage states of seismically or non-seismically
design bridges, which are applicable for bridges with weak pier and strong bearings.
Displacement limits increase as the drift limits grow bigger.
Table 6. 5. Drift limits from Basoz and Mander [1999]
37

In the study of Choi et al [2004], damage states were defined based on column ductility
demand, steel fixed and expansion bearing deformations, and elastomeric bearing
deformations, which are summarized in Table 6.6
Table 6. 6. Damage states defined by Choi [2004]

Choi et al [2004] pointed out that the quantified data for the damage states were based on the
previous studies and tests results. The displacement at the complete damage limit state was
assumed to be #=255mm, which accounts for the unseating PSC-girders.
In a study of Kibboua et al [2011], the damage assessment of bridge piers is carried out using
the quantified Park-Ang damage index DI expressed as
u
h d
DI
! +
= ,where
d
is the
displacement ductility,
u
is the ultimate ductility of the bridge piers, $ is the cyclicloading
factor taken as 0.15 and
h
is the cumulative energy ductility.
Priestley et al [1996] proposed two kinds of limit states: section limit state and structure limit
state based on strain properties and ductility respectively, which is illustrated in Figure 6.1.
38

Figure 6.1. Member limit state and structure limit state
The pushover curves are converted into bilinear representation by the following procedures.
The yielding point of the longitudinal reinforcement is specified to determine linear elastic
portion and initial slope of the bilinear curve. The ultimate curve point is specified when
reinforced steel or confined concrete extreme fibre has reach its ultimate strain value or when
the moment capacity has decreased.
Priestley et al. proposed three limit states for the assessment of bridges: serviceability,
damage control and survival.
For the serviceability limit state, the typical value for displacement ductility factors is 2.
For the damage control limit state, generally the displacement ductility factors is from 3 to
6, while it is appropriate in Europe that it is equal to 4.
Although survival limit state is of critical significance, there is no specific proposed
corresponding ductility value.
J. Kowalsky [2000] proposed a method to determine damage limits for circular bridge
columns, which considered two damage states: serviceability and damage control. It is
assumed that those damage states are related to concrete compression and steel tension strain
limits, as detailed in Table 6.7.
Table 6. 7. Limit states by J. Kowalsky [2000]
39

After establishing the two damage states based on dimensionless curvature relationships,
expression for drift ratio can be derived easily, which is shown in equation 6.2,6.3,
D
L K
L
D
K K
Y
p
Y S
S
3
+
!
= " (6.2)
D
L K
L
D
K K
Y
p
Y DC
DC
3
+
!
= " (6.3)
Where L
p
represents the member plastic hinge length according to Priestley et al [1996].
Curvature ductility factors for serviceability and damage control limit states can be obtained
by dividing the design limit states curvature by the yield curvature.
In a study of Nielson et al [2005], four damage limit states were defined based on HAZUS
guidelines, using a component level approach, as described in Table 6.8. The proposed
framework includes both physics-based and descriptive features, integrated by means of a
Bayesian approach.
Table 6. 8. Medians and dispersions for bridge component limit states using Baysian updating

In a study of Karakostas et al [2006], five damage state criteria were used to characterize the
damage level the structure was under, for each ground motion record. Those five damage
states were defined by , as described in Table 5.9, which refers to ratio between the
performance displacement and yield displacement thus ductility in displacements.
y
! ! /
40
Table 6. 9. Damage states from Karakostas et al [2006]

Generally, damage states described by HAZUS are widely accepted in the development of
fragility curves. However, as different studies may focus on different regions with different
configuration of bridges, the detailed limit states and criteria may have big difference among
one another. On the other hand, most of those studies make use of global damage states when
carrying out the seismic assessment analysis. In this study, four limit states based on the
description of HAZUS are considered, which also concentrated on the global damage states of
the whole bridges.

6.2 Selection of limit states
According to Hwang et al [2001], element-level limit state criteria are more suitable when
detailed seismic damage assessment is required, such as the assessment of the seismic retrofit
scheme of a bridge. For other purposes, such as seismic fragility analysis of bridge
populations, other approaches, based on the assessment of global damage of a bridge, should
be applied.
Accordingly, in this particular study, given that a large number of bridges, defined with
randomly generated properties, it is more appropriate to use structure-level, rather than
element-level, limit state criteria to define damage states. The chosen engineering demand
parameter, used for quantification of damage of each pier, was ductility in displacements. On
a simplified fashion, the maximum value among the different bridge piers was taken as
representative of the behaviour of the entire bridge.
41
In this study, the definitions of each damage limit state is following the description of damage
states proposed by HAZUS [2003], thus four damage states were considered: slight damage,
moderate damage and extensive damage and collapse.
The use of strain measurements from nonlinear numerical analysis can be seen as of
questionable accurateness. Consequently, curvature, a parameter far more stable, has been
chosen to check damage limit states occurrence. Therefore, as Neilson et al [2005] used the
curvature ductility to check each limit states, four damage limit states were also defined based
on the HAZUS description and Bayesian approach. Thus, Neilsons study was chosen as the
definition for each damage states.
The ductility in curvatures corresponding to each of the four damage limit states is
presented in Table 6.10.

Table 6. 10. Four damage limit states from Neilson et al [2005]

Accordingly, displacement ductility can be calculated based on the corresponding observed
curvature ductility using equations 6.4 and 6.5, as proposed by Priestley et al [1996].
)] / ( 5 . 0 1 [ ) / ( ) 1 ( 3 1 l l l l
p p
! " ! ! " + =
# $
(6.4)
b p
d l l 9 08 . 0 + = (6.5)
Accordingly, incorporating the data from Table 5.10 in Equation 5.4, the limit states in terms
of displacement ductility for a pier of, e.g., height 15m, using 30mm reinforcement bars, will
result as follows.
m d l l
b p
47 . 1 03 . 0 9 15 08 . 0 9 08 . 0 = ! + ! = + =
For LS1, curvature ductility is equal to 1.29, thus the related displacement ductility is:
Slight Moderate Extensive collapse
Curvature
ductiltiy
1.29 2.10 3.52 5.24
42
081 . 1 )] 15 / 47 . 1 ( 5 . 0 1 [ ) 15 / 47 . 1 ( ) 1 29 . 1 ( 3 1 )] / ( 5 . 0 1 [ ) / ( ) 1 ( 3 1 = ! " ! ! " + = # " # # " + =
$
l l l l
p p %

308 . 1 )] 15 / 47 . 1 ( 5 . 0 1 [ ) 15 / 47 . 1 ( ) 1 01 . 2 ( 3 1 )] / ( 5 . 0 1 [ ) / ( ) 1 ( 3 1 = ! " ! ! " + = # " # # " + =
$
l l l l
p p %

705 . 1 )] 15 / 47 . 1 ( 5 . 0 1 [ ) 15 / 47 . 1 ( ) 1 52 . 3 ( 3 1 )] / ( 5 . 0 1 [ ) / ( ) 1 ( 3 1 = ! " ! ! " + = # " # # " + =
$
l l l l
p p %

185 . 2 )] 15 / 47 . 1 ( 5 . 0 1 [ ) 15 / 47 . 1 ( ) 1 24 . 5 ( 3 1 )] / ( 5 . 0 1 [ ) / ( ) 1 ( 3 1 = ! " ! ! " + = # " # # " + =
$
l l l l
p p %

When the different damage states are very distant from each other, in terms of the
corresponding values for ductility, the allowance range for each damage state will be larger,
and so will be the tolerance for differences between each method, which means that the
comparison between methods, in terms of fragility functions, could hide important differences.
However, from the 15m tall pier example above, one can see that the gap between two
different damage states is small, thus the comparison of corresponding fragility functions can
become sensitive enough, which will render the comparison between different NSPs an
dynamic analysis more clear.
Chapter 7 Nonlinear Static Procedures
43
7 Nonlinear Static Procedures
7.1 Pushover Analysis
The earthquake engineering community has been witnessing a significant development of the
performance-based seismic engineering concepts and methodologies in the last two decades.
New ideas and specific procedures, such as pushover analysis, have become increasingly
popular.
It is widely accepted that the most accurate method of analysis for evaluating the seismic
response of structures is nonlinear dynamic analysis. This sort of analysis is however
associated to critical issues, such as the selection of appropriate ground motion records or
intrinsic complexities, such as definition of damping model and post-elastic behaviour, and
typically requires considerable computational effort with time-step integration. As a
consequence, alternative nonlinear static procedures, based on pushover analysis, are
recognized as simplified, yet reliable, techniques for the assessment of the seismic response of
existing structures.
In this study, three different types of pushover analysis are considered for each bridge: 1)
adaptive displacement-based pushover analysis; 2) conventional pushover analysis with
uniformly distributed lateral load; and 3) conventional pushover analysis with 1
st
mode
proportional distributed lateral load.

7.1.1 Conventional Pushover
A pushover curve shows the relationship between the total base shear and the displacement of
a reference node of a MDOF system. The advantages of using pushover analysis, with respect
to nonlinear dynamic analysis, are mostly related to time and computational onus saving. On
the other hand, typical disadvantages pointed out to the employment of pushover analysis are
the eventually questionable level of accuracy of seismic performance estimates, especially in
irregular structures, including bridges. Such drawback has been particularly associated to
conventional pushover, which makes use of constant load patterns, such as uniform or 1
st

mode proportional shapes, which can hardly take higher modes and inelastic behaviour into
proper consideration.
44
A first attempt to overcome such inherent limitation was proposed in a study of Chopra and
Goel [2002], which presented a modal pushover analysis (MPA) procedure. In this procedure,
multiple pushover analyses are carried out using the load pattern of each relevant vibration
mode and the final response of a structure is calculated as the combination of the different
modal response estimates, which enables the higher modes contribution to be considered.

7.1.2 Adaptive Pushover
A different proposal to overcome the drawbacks of traditional pushover consists of using
adaptive or fully adaptive pushover analysis, developed Antoniou and Pinho [2004]. In this
approach, instead of applying an invariant load vector, the structural properties of the model
are evaluated at each step of the analysis and the loading pattern is updated accordingly thus
taking into account the influence of higher mode effects, degradation characters and spectral
amplifications due to ground motion frequency content.
1. The adaptive pushover logarithm, applied to th analysis of 3D bridges, can be described
as follows.Define nominal load vector P
0
and inertial mass. This step is only carried out
at the beginning of the analysis. In adaptive pushover, the load vectors are defined
automatically at each step therefore the nominal vector of load should follow a uniform
distribution. Moreover, the inertial masses of the structure should be modelled so that
eigenvalue analysis could be carried out at each step.
2. Compute load factors. This and the next two steps are repeated at each equilibrium
stage of the pushover analysis. The magnitude of the loading vector increment !P at
any step is given, in general terms, by the product of its nominal counterpart P
0
, defined
previously, and the load factor "# at that step. The latter is automatically increased till a
predetermined target is reached.
(6.1)
3. Calculate normalised scaling vector. The normalised modal scaling vector ,
computed at the start of each load increment, reflects the actual stiffness state of the
structure, the contribution of the different modes and the influence of the frequency of a
0
P P ! " = " #
D
45
particular input. In order to determine the scaling vector at each step, eigenvalue
analysis is carried out at the end of the previous load increment.
(6.2)
The lateral load profiles of each vibration mode are then combined using SRSS rule.
The displacement obtained at each location after their modal combination are
normalised so that the maximum displacement remains proportional to the load vector.
(6.3)
4. Update the loading vector. Once the normalised scaling vector and load factor have
been determined, as well as the value of initial nominal load vector, the loading force
vector at the given analysis step is updated by incremental updating technique.
The calculation of the updated loading vector is performed in two successive steps: the
updating of the loading vector per each structural mode of relevance and the
computation of the current normalised modal scaling vector.
(6.4)
A brief comparison between the different types of pushover curves, for a simple bridge
model, using OpenSees, is carried out in Figure 7.1

Figure 7.1. Pushover curves
j d ij j ij
s D
,
! " =
) max(
i
i
i
D
D
D =
0 1
P D P P
k k k k
! " + =
#
46
The observation of the different curves indicates that the adaptive pushover exhibits the
higher stiffness degradation after yielding, when compared to 1
st
proportional pushover
analysis.

7.2 Nonlinear Static Procedures
It is commonly accepted that deformations are of higher importance, with respect to seismic
performance thus it is believed that seismic design and/or assessment should rely on
displacements. Nonlinear static procedures (NSP) represent a simplified approach for the
assessment of the seismic behaviour of structures based on pushover analysis. Generally they
consist of the following steps: 1) define a SDOF model from pushover analysis of a MDOF
structure; 2) bi-linearize the capacity curve of the SDOF system; 3) intersect the capacity
with the response spectrum to find out the performance point of the structure.
NSPs can currently be found in codes and guidelines, such as the ATC-40 [ATC,1996],
FEMA-273 [ATC,1997] or the European Code [CEN, 2005a, 2006a]. The reason why NSP
have become such popular within the engineering community is that, when compared to
nonlinear dynamic analysis, NSPs have proved to provide reasonable estimates, requiring less
time and computational effort.
There is a group of pioneering methods, corresponding to the first proposals of nonlinear
static analysis based procedures, which have led to reasonably accurate results. Capacity
Spectrum Method (CSM), introduced by Freeman et al. [1975] and implemented in ATC-40
guidelines, is one of those. Similarly, the N2 method, has been proposed by Fajfar and
Fischinger [1988] and included afterwards in the recommended simplified procedures in
European Code [CEN,2005a]. These first methodologies are mainly focused on simplicity and
consider either the first mode or uniform load distribution for the pushover computation.
Recently, an improved version of CSM has been presented in FEMA-440 guidelines
[ATC,2005], including updated empirical equations to estimate equivalent viscous damping
and spectral reduction factor. Displacement Coefficient Method (DCM) was initially
introduced in ATC-40 and provides a considerable simple empirical equation to calculate the
seismic response of structures using different specific coefficients. Given that all those first
methods fail to take higher modes into consideration, Modal Pushover Analysis was proposed
47
by Chopra and Goel [2002] and has been modified recently according to the seismic
behaviour of higher modes. On the other hand, with the development of the adaptive pushover
analysis, the methodologies based on adaptive or fully adaptive scope enjoy an increasing
development. Several methods were proposed by different scholars and some of them have
already been put into utilization, such as Adaptive Capacity Spectrum Method (ACSM) or
Adaptive Modal Combination Procedures (AMCP).

7.2.1 Capacity Spectrum Method (CSM)
The Capacity Spectrum Method (CSM) was initially proposed by Freeman et al. [1975],
which, by means of a graphical procedure, illustrated in Figure 7.2, iteratively compares the
structure capacity and the seismic demand, represented by a capacity curve and a response
spectrum, respectively. In order to account for the nonlinear behaviour of the structural
system, the ground motion spectrum is computed for a level of equivalent viscous damping at
each iteration.

Figure 7.2 Graphical procedures of CSM
According to ATC-40, the procedure for employing CSM is the following:
1. Carry out a conventional pushover analysis of the multi degree of freedom structural
48
model;
2. Define the capacity curve for the MDOF structure;
3. Transform the MDOF capacity curve into a SDOF capacity curve based on the
participation factor of the fundamental mode;
4. Approximate a bi-linear curve from the SDOF capacity curve. Based on ATC-40, the
stiffness before yielding is equal to the initial stiffness of the structure. In order to derive
the yielding point, the principle of equal areas is applied;
5. Intersect the capacity curve with the demand spectrum, initially damped of 5% and
determine the performance point. Iteration is carried out by reducing the demand spectrum
based on the updated equivalent viscous damping until the convergence in performance
point estimate is reached.
Recently, FEMA-440 proposed an update to this method, introducing new equations for
calculating equivalent viscous damping and periods. In this study, the procedure was applied
following the updated version.
In this study, CSM were carried out using two different lateral load patters for the
conventional pushover analysis: uniform distributed lateral load and 1
st
mode proportional
lateral load.

7.2.2 N2
The N2 method was formally proposed by Fajfar [2000], as a simplified procedure for
estimating the seismic response of structures. N stands for nonlinear analysis whereas 2
represents two individual mathematical models applied, which are pushover analysis and
spectrum approach separately.
N2 method is similar to CSM in some aspects, given that both of them carry out an analysis
based on capacity curves and response spectra. On the other hand, the major difference
between the methods is that N2 employs inelastic spectra, given in Figure 7.3, rather than
elastic spectra with equivalent viscous damping, which avoids the need for assumptions and
empirical expressions. Moreover, according to FEMA356 [ATC,2000], the post yielding
49
stiffness is equal to zero for the bi-linearization step. In addition, the performance point could
be derived without iteration.

Figure 7.3. Inelastic Spectra by N2
The most essential concept of N2 method is the time-independent lateral displacement profile,
which has an important influence over the pushover curves and the final performance point of
structures. In order to get reasonable pushover curves, at least two load patterns should be
applied, which are uniform distributed load and 1
st
mode shape proportional distributed load.
The procedure for application of N2 is defined by the steps below:
1. Carry our a conventional pushover analysis for the multi degree of freedom model;
2. Derive the capacity curve for the MDOF structure, based on the total base shear and
reference node displacement;
participation factor of the fundamental mode;
4. Calculate the bi-linear curve from the SDOF capacity curve. Build the elasto-perfectly
plastic capacity curves and calculate the corresponding period of the equivalent SDOF
systems;
50
5. Determine the target displacement by comparing the equivalent period of the system and
the given spectral corner periods, placing the structure in the proper period range;
Check if the target displacement equal to the displacement assumed from the bi-linear
curve; if not, iteration can be carried out to define a new bilinear curve, based on the target
displacement, until convergence is reached.

7.2.3 Displacement Coefficient Method (DCM)
The Displacement Coefficient Method (DCM) was initially implemented in ATC-40 [1996],
then revised and modified by FEMA440 recently. DCM consists of modifying the elastic
spectral displacement for the effective fundamental period (extracted from the capacity
curve), using four coefficients related to different key factors in determining the seismic
response of a structure, shown in equation 7.5.
g
T
S C C C C
e
a t
2
2
3 2 1 0
4!
" =

(7.5)
C
0
is the modification factors to spectral displacement of the equivalent SDOF system to the
top displacement of the MDOF system. In this study, it is equal to the first modal participation
factor at the level of the control node.
C
1
is the modification factor to expected maximum inelastic displacement, applied to the
displacement calculated for linear elastic response.
C
2
is the modification factor related to the effect of hysteric shape, stiffness degrading and
strength degrading on maximum displacement.
C
3
is the modification factor to P-delta effect.
The definition of the coefficients is mostly empirical, essentially developed by means of
statistical analysis of the dynamic behaviour of SDOF models.
The DCM procedure, according to FEMA256, and further improved by FEMA440, is
described next:
51
1. Carry out a conventional pushover analysis for a multi degree of freedom model;
2. Derive the capacity curve for the MDOF structure based on the total base shear and
reference node displacement;
participation factor of the fundamental mode.
4. Calculate the bi-linear curve from the SDOF capacity curve, illustrated in Figure 7.4.
According to FEMA356, the stiffness of the pre-yielding branch should be equal to the
secant stiffness calculated based on the base shear equal to 60% of the yield strength.

Figure 7.4. Bilinearization of the force-displacement curves
5. Calculate target displacement by a set of four coefficients. In this step, iterations may need
for re-bilinearization till the target displacement equal to the displacement assumed from
the bi-linear curve.
DCM has been developed and optimized with a view to application to buildings thus the
results for bridges rely on some simplifications and specific assumptions.

7.2.4 Modified Modal Pushover Analysis (MMPA)
The Modal Pushover Analysis was initially introduced by Chopra and Goel [2002],
suggesting repeated nonlinear static procedures for every significant mode of the structure.
52
This method has been widely used due to the the difficulty of the traditional procedurs in
taking higher modes and the redistribution of the inertia forces into consideration, which play
important roles in bridge analysis.
Recently, Chopra and Goel formulated that the response of higher modes during earthquakes
remains in the elastic range hence elastic response spectra should be used for higher modes
effects estimates. They proposed the Modified Modal Pushover Analysis (MMPA), based on
MPA to calculate the performance point of a specific structure. The novelty is that in MMPA,
inelastic response is only being considered for the fundamental mode while for higher modes,
the whole structure is assumed to exhibit elastic behaviour. Chopra and Goel [2004] have
addressed the comparison between the two methods showing that the MPA results
underestimate the seismic response of structures whilst the MMPA yields better estimates,
through the use of elastic SDOF systems.
The MMPA procedure can be summarized in the following steps:
1. Compute the n natural frequencies and modes for the linearly elastic vibration of the
structure;
2. For the first mode, develop the base shear-roof displacement pushover curve for force
distribution proportional to the mass and mode shape;
3. Idealize the pushover curve as a bilinear curve and convert it, computing the first mode
inelastic SDOF system quantities;
4. Compute the peak deformation of the first mode inelastic SDOF system defined
previously using nonlinear response history analysis, inelastic design spectrum or
empirical equations for the ratio of deformations of inelastic and elastic systems;
5. Compute the dynamic response due to the first mode combining the effects of lateral and
gravity loads;
6. Compute the dynamic response due to higher modes under the assumption that the system
remains elastic, performing a classical modal analysis of a linear MDOF system, skipping
the need for additional pushover analysis;
Determine the total response combining the peak modal responses using SRSS rule.
53

7.2.5 Adaptive Capacity Spectrum Method (ACSM)
The Adaptive Capacity Spectrum Method (ACSM) was proposed by Casarotti and Pinho
[2007] and allows the estimation of the seismic response of structures using a fully adaptive
perspective.
Rather than using any elastic or inelastic mode of vibration to convert the MDOF structure
into the equivalent SDOF system, the equivalent SDOF adaptive capacity curve is step by
step derived by calculating the equivalent system displacement and acceleration based on the
actual deformed shape at each analysis step. The developed adaptive capacity curve is
intersected with the demand spectrum, providing an estimate of the inelastic acceleration and
displacement demand (performance point) of the structure. An iterative procedure is then
required until convergence is reached in terms of the equivalent viscous damping to be used in
the reduction of the demand spectrum.
One of the key issues within ACSM is the definition of spectral Reduction Factors, which can
be estimated according to several approaches.
Reduction factors can be roughly categorized into two groups: damping based, computed
according to Equations (7.6), (7.7) and (7.8), and ductility based, as defined by Equations (7.9)
and (7.10). The former can be described as deriving reduction factors associated with
equivalent viscous damping and applying it to both acceleration and displacement spectrum
ordinates. The latter use 5%-damping spectra and then reduce the acceleration ordinate by a
factor defined as a function of ductility. The spectral reduction within ductility-based methods
is not exactly vertical, given that displacements are modified as well.
% 5 , , !
" =
el a damp a
S B S (7.6)
% 5 , , !
" =
el d damp d
S B S (7.7)
2
,
,
!
damp a
damp d
S
S = (7.8)
R
S
S
el a
duct a
% 5 ,
,
!
= (7.9)
54
% 5 , % 5 , , ! !
" = =
el d el a duct d
S C S
R
S

(7.10)

Figure 7.5 Spectral reduction methods
A recent study, by Casarotti et al. [2009], has compared the employment of a number of
approaches to estimate spectral reductions factors and concluded that, among the different
approaches, the damping-based model proposed by Priestley et al. [2007] was the most
accurate.
The ACSM procedure is summarized as follows:
1. Carry out an adaptive pushover analysis for the multi degree of freedom model of the
structure;
2. Derive the equivalent SDOF adaptive capacity curve;
3. Intersect the SDOF adaptive capacity curve with the demand spectrum, calculate the
performance point for an assumed damping;
4. Bi-linearize the capacity curve at the performance point and calculate the corresponding
damping;
Check if the calculated damping equal to the assumed one, if not iterate step 3 to 5 till
they are equal and get convergence.

55
7.2.6 Adaptive Modal Combination Procedures (AMCP)
With the main purpose of taking higher modes into consideration, Adaptive Modal
Combination (AMCP) was proposed by Kalkan and Kunnath [2006], based on adaptive
pushover analysis.
In this procedure, higher modes are taken into account by combining individual modal
pushover response with effects of varying dynamic properties, defining parameters for
calculating inelastic spectrum. Moreover, the procedure also makes use of Capacity Spectrum
Method and Modal Pushover Analysis concepts, and, at the same time, it eliminates the need
to pre-estimate the target displacement. When calculating the increment step for capacity
curve of the equivalent SDOF system, an energy-based method is applied. A key aspect of the
method is that a set of capacity spectra based on a series of predetermined ductility levels are
used for each mode to approximate displacement demand.
The steps to be followed when employing AMC are:
1. Compute modal properties of the multi degree of freedom model, such as frequencies,
mode shapes and modal participation factors;
2. For each considered mode, apply an adaptive pushover analysis using the corresponding
mode proportional lateral force, which, similarly to adaptive pushover analysis, could be
calculated after every load step;
3. Calculate the equivalent SDOF adaptive capacity curve for each mode;
4. Calculate the bi-linear curve from the SDOF capacity curve based on elasto-perfectly
plastic post-yielding behaviour.
5. Intersect the SDOF adaptive capacity curve with the demand spectrum and calculate the
6. Check if the calculated damping is equal to the assumed one; if not, iterate steps 4 to 6
until convergence is reached.

56
This method was proposed having building structures, instead of bridge ones, in mind thus its
application within this study required some simplification.

7.2.7 Modified Adaptive Modal Combination Procedures (MAMCP)
As Chopra pointed out, with respect to higher modes, the seismic response generally remain
in elastic range thus the Modified Adaptive Modal Combination Procedure is proposed by the
author. The procedure is mostly based on AMCP, with the only difference being that, in
MACMP, inelastic response is only considered for the fundamental mode whilst, for the
higher modes, the structural response is assumed to be elastic. An immediate consequence is
that MACMP requires less computational effort.
The procedure can be applied according to the steps below:
1. Carry out a 1
st
mode-based adaptive pushover analysis for the multi degree of freedom
model of the structure;
2. Derive the equivalent SDOF adaptive capacity curve;
3. Intersect the SDOF adaptive capacity curve with the demand spectrum and calculate the
4. Bi-linearize the capacity curve at the performance point and calculate the corresponding
damping;
5. Check if the calculated damping equals the assumed one; if not, iterate steps 3 to 5 until
convergence occurs;
6. Compute the dynamic response due to higher modes under the assumption that the system
remains elastic, performing a classical modal analysis of a linear MDOF system, skipping
the need for additional pushover analysis.

57
7.3 Overall Procedures
The overall procedure that has been followed to assess the performance of the different NSPs,
when calculating fragility curves, can be summarized in the following steps.
1. Random generation of a population of 3D bridge FE models using Monte Carlo simulation;
2. Pushover analysis of each bridge and conversion to the equivalent SDOF system curve;
3. Estimate nonlinear target displacement for each bridge, for each of the selected Nonlinear
Static Procedures, using a large set of ground motion records;
4. Identification of the global damage state based on the nonlinear response;
5. Representation of the cumulative percentage of bridges in each damage state versus the
value of the intensity measure corresponding to each record;
6. Regression analysis to calculate the parameters defining the fragility functions.

7.4 Comparison with nonlinear dynamic analysis
Nonlinear dynamic analysis is widely recognized as the most accurate and reliable tool to
estimate the seismic response of a structure. However, the requirements of this approach are
relatively more complicated, if compared to the previously described static procedures. Such
high complexity is frequently related to the level of the detail the model, the definition of
initial masses, the determination of system damping, etc., all leading to a significant increase
in the computational onus.
In this study, nonlinear dynamic analysis is carried out as the baseline for the sensitivity study
to yield the conclusion that which nonlinear static procedures gives the best compromising
between the accuracy and complexity. The closer the results derived from nonlinear static
analysis to the nonlinear dynamic analysis, the more accurate the methodology provides.
In this study, OpenSees was used to carry out the dynamic analysis of the randomly generated
portfolio bridges, with a specific set of ground motion records. Only the significant duration
of the records was considered based on 5% maximum PGA according to Bommer and Pereira
[1999].
Chapter 8 Results
58
8 Results
8.1 Introduction
The presentation of results has been divided in two main parts: one corresponds to the
validation of the different NSPs, through direct comparison by means of the parameter Bridge
Index, BI calculated as equation 8.1, which is defined as median of the ratios between the
response parameter quantities, at different bridge locations, estimated for the performance
point of the structure obtained through a specific NSP and the maximum response parameter
quantities estimated with nonlinear dynamic analysis, whereas the other regards the relative
performance of the different NSPs in terms of Fragility Curves.
Dynamic
NSP
BI
!
!
= (8.1)
The response of the bridge for a given intensity level can be measured at different locations
thus the parameter Bridge Index (BI) was chosen to directly compare the different methods,
based on the performance point (PP) response parameters, that is the maximum response
structural displacement of the deck.
In tandem with the two approaches described in Section 4.2, the comparison of the different
NSPs with dynamic analysis was carried out by calculating BI ratios in two different ways.
One was to derive performance points of each NSP using intersection
1
with individual real
spectra whereas the other was to compute performance points from median spectra matching a
predefined design spectrum, as prescribed by the majority of the tested NSPs and applied in
Monteiro [2011].
With respect to the computation of fragility functions using nonlinear static procedures,
although almost all available guidelines suggest the use of smoothed design spectra to
calculate performance points for NSPs this study used the ground motion records selected
from the PEER database, which are real, non-scaled records (in line with the approach
described in Section 5.2.1).

1
To be precise, not all of the NSPs use the intersection of capacity curve with response spectrum to find the performance
point.
Chapter 8 Results
59
In order to make sure that the size of sample would be high enough to assure soundness
results, a brief preliminary parametric study was carried out, testing statistical convergence. In
this study, 10000 RC bridges were generated to estimate the maximum displacement of the
middle pier and the mean value of those 10000 bridges was assumed as the exact solution.
Afterwards samples of different sizes were generated (from 5 to 300), and the corresponding
mean values were used to calculate the relative error, which is defined as the difference
between the mean value derived from each sample and the exact solution divided by the latter
one. In Figure 8.1, which illustrates such calibration procedure, one can see that when the
size of sample is larger than 170, the relative error remains below 4% thus, conservatively, the
size of the samples, to use as case study, has been chosen as 200.

Figure 8.1. Relationship between the relative error and size of sample
It is not always straightforward to employ NSPs that rely on capacity-demand interaction in
an automatic fashion, when using real accelerograms, given that the intersections of the
structure capacity curve with the response spectrum can occur more than one time. In such
case, according to Casarotti et al [2005], the first intersection point was chosen for the
analysis.

8.2 Bridge Index (BI) based on individual records
In this part of study, 200 bridges were generated randomly and the set of ground motion
records was the 10 LA records.

0 50 100 150 200 250 300
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
R
e
l
a
t
i
v
e

E
r
r
o
r
Size of Sample
Chapter 8 Results
60
8.2.1 Capacity Spectrum Method
Figure 8.2 illustrated the median BIs obtained with CSM-based estimation.

Figure 8.2. CSM median Bridge Index per intensity level
Generally, using pushover analysis that makes use of 1
st
mode shape proportional or uniform
load patterns yields similar results, within the employment of CSM. Furthermore, with
increasing intensity level, CSM produces more accurate seismic performance estimates, when
compared with nonlinear dynamic analysis. One of the possible reasons for the
underestimating trend may be the fact that CSM uses equivalent viscous damping, which may
lead to overestimation of damping.

8.2.2 N2 Method
Figure 8.3 illustrates the median BIs obtained from N2-based estimates.

Figure 8.3. N2 median Bridge Index per intensity level
The observation of the N2 results leads to similar conclusions to what has been found for
CSM: the 1
st
mode proportional load pattern yields slightly more accurate predictions. In
addition, generally speaking, the larger the intensity level, the closer the results get to
dynamic analysis ones.
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.6
0.8
1
1.2
1.4
1.6
M
e
d
i
a
n

B
I
Intensity Level:PGA(g)
BI CSM

CSMuni
CSM1st
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.6
0.8
1
1.2
1.4
1.6
M
e
d
i
a
n

B
I
BI N2

N2uni
N21st
Chapter 8 Results
61
8.2.3 Displacement Coefficient Method
Figure 8.4 illustrates the median BIs obtained from DCM-based estimates.

Figure 8.4. DCM median Bridge Index per intensity level
It seems quite evident that the 1
st
mode proportional load pattern leads to better predictions,
which clearly states for the significant influence of the choice for the lateral load patterns of
pushover analysis on the prediction of the bridge performance point.
On the other hand, as DCM was initially proposed for buildings and some of the coefficients
are not important or even relevant for bridges, the results derived from this procedure might
be compromised, and were indeed not as good as the ones obtained with other procedures.

8.2.4 Modified Modal Pushover Analysis
Figure 8.5 illustrates the median BIs obtained from MMPA-based estimates.

Figure 8.5. MMPA median Bridge Index per intensity level
Modified modal pushover analysis takes higher modes contribution into consideration, which,
according to the indices in Figure 8.5, do seems to have led to better predictions, especially
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.5
1
1.5
2
M
e
d
i
a
n

B
I
BI DCM

DCMuni
DCM1st
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.6
0.8
1
1.2
1.4
1.6
M
e
d
i
a
n

B
I
BI MMPA
Chapter 8 Results
62
when compared to other methods that make use of 1
st
mode based conventional pushover
analysis.

8.2.5 Adaptive Capacity Spectrum Method
Figure 8.6 shows the Bridge Index obtained with Adaptive Capacity Spectrum Method.

Figure 8.6. ACSM median Bridge Index per intensity level
Similarly to CSM, but contrarily to N2 and MMPA, ACSM is a method that relies
considerably on damping, estimating it to then again estimate the spectral reduction factor. In
spite of making use of an adaptive pushover analysis, ACSM did not yield particularly
accurate predictions, exhibiting a general underestimating trend. As has been observed for
other procedures, the accuracy of the ACSM predictions increases with the increasing of the
intensity level.

8.2.6 Adaptive Modal Combination Procedure
Figure 8.7 illustrates the comparison with nonlinear dynamic analysis, obtained when AMCP
is employed.

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.6
0.8
1
1.2
1.4
1.6
M
e
d
i
a
n

B
I
BI ACSM
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.4
0.6
0.8
1
1.2
1.4
1.6
M
e
d
i
a
n

B
I
BI AMCP
Chapter 8 Results
63
Figure 8.7. AMCP median Bridge Index per intensity level
Considering both higher modes effects and post-yield stiffness degradation, AMCP provides
fairly good estimates for lower intensity levels whilst for higher intensity levels, considering
the higher modes contribution through SRSS combination for the final response parameter
prediction, AMCP tends to slightly overestimate nonlinear dynamic analysis.

8.2.7 Modified Adaptive Modal Combination Procedure
MAMCP results in terms of Bridge Index are shown in Figure 8.8.

Figure 8.8. MAMCP median Bridge Index per intensity level
Similarly to the previous methods, MAMCP provides slightly overestimating indexes for high
levels of intensity. Nevertheless, the average Bridge Index for this procedure is around 0.9,
which somehow denotes global underestimation of the performance.

8.2.8 Global Results

Figure 8. 9. Median Bridge Index per intensity level for all seven methods
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
M
e
d
i
a
n

B
I
BI MAMCP
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.5
1
1.5
2
M
e
d
i
a
n

B
I
BI

CSM
N2
DCM
MMPA
ACSM
AMCP
MAMCP
Chapter 8 Results
64
Figure 8.9 plots the Bridge Index for all of the nonlinear static procedures applied in this
study. Generally, NSP-based response estimates of bridges get closer to the dynamic-based
ones, thus more accurate, with increasing intensity level.
In order to render the comparison more clear, the set of NSPs was divided in two groups, one
containing the pioneering and/or conventional ones and the other including the improved
and/or adaptive ones. The corresponding results are presented in Figures 8.10 and 8.11.

Figure 8.10. Median BI for each intensity level of CSM, N2 and DCM
From the observation of Figure 8.10, one can conclude that, generally, the classic
procedures tend to underestimate the response. In addition, it can be seen that DCM has a
very variable behaviour within the considered intensity range, whereas CSM can be seen as
the method with the steadiest behaviour, although it sometimes heavily underestimates the
response displacement estimates.

Figure 8.11. Median BI for each intensity level of CSM, N2 and DCM
Figure 8.11, on the other hand, shows that the so-called innovative NSPs, do bring in some
improvement, given that BIs are definitely closer to unity. Despite that MAMCP considers the
higher modes as elastic, ACSM and MAMCP yield somewhat similar results, which indicate
that within the use of adaptive modal pushover analysis, the 1
st
mode contribution is more
important than the one from higher modes.
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.5
1
1.5
2
M
e
d
i
a
n

B
I
BI

CSM
N2
DCM
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.5
1
1.5
2
M
e
d
i
a
n

B
I
BI

MMPA
ACSM
AMCP
MAMCP
Chapter 8 Results
65

Figure 8.12. Median Bridge Index for each NSP
Finally, Figure 8.12 shows overall comparison of the tested NSPs, by calculating the median
Bridge Index, across all the intensity levels. MMPA, ACSM and, particularly, N2 and
MAMCP stand as the best perfomring procedures, yielding global BI ratios pratically equal to
one., i.e., staitic-based estimates that globally match the dynamic ones.

Figure 8.13. Median Bridge Index for each PGA
Figure 8.13 shows the global average Bridge Index results from a different perspective, i.e.,
for each peak ground motion acceleration, the median BI across all the tested NSPs is plotted.
The results shows, again, that with increasing intensity level, the accuracy of the predictions
of the deifferent NSPs becomes higher. As a general conclusion one can say that the NSP
predictions are mostly within a -/+20% range, with respect to the dynamic analysis ones.
As a final output, Figure 8.14 represents the median BI values for N2, MMPA, ACSM and
MAMCP, which have been recognised as the best performing procedures among all the seven
NSPs. When eliminating the less accurate procedures from the representation, the range in
which the majority of the estimates fall within becomes -/+10%.
CSM N2 DCM MMPA ACSM AMCP MAMCP
0.5
1
1.5
M
e
d
i
a
n

B
I
Global BI
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity levels:PGA(g)
Displacement BI
Chapter 8 Results
66

Figure 8.14. Median Brige Index for N2, MMAP and MAMCP
Table 8.1 represents the quantitative version of the results, given by the exact BI values for
every considered NSP, per intensity level.
Table 8.1. Median BI of seven approaches for per intensity level

The observation of the numbers reinforces the main trend identified so far for nearly all of the
tested Nonlinear Static Procedures: with the increase of the intensity measure (peak ground
acceleration), the static-based displacement estimates become closer to dynamic analysis
based ones. Nevertheless, for some of the low intensity ground motion records, some NSPs
can still provide fair predicitons.
Based on the results presented so far, even though adaptive pushover analysis considers the
actual stiffness of the structure at each step, it still yielded underestimated predictions, which
might be due to a further needed calibration of the equivalent viscous damping model. On the
other hand, the N2 method, which makes use of a constant load pattern and inelastic spectrum,
provided fairly good estimates. A similar scenario was found for the application of MMPA,
which also make use of inelastic spectrum but takes higher modes into consideration. With
respect to other characteristics of the procedures, such as computional and time efforts, N2
would be preferable.
It is worth mentioning that the proposed MAMCP also provides comparetively good estimates,
coupled with considerable time saving by accounting for higher modes in elastic regime.
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity levels:PGA(g)
Displacement BI
PGA(g) 0.676 0.479 0.235 0.426 0.360 0.969 0.657 0.580 0.817 0.986 Median
BI_CSM1st 0.728 0.664 0.542 0.786 0.742 0.873 0.972 0.618 0.941 0.740 0.761
BI_N21st 1.186 0.525 0.772 1.249 1.066 1.375 1.031 0.988 0.889 0.900 0.998
BI_DCM1st 1.118 0.407 0.358 1.141 0.207 1.895 0.919 0.671 0.562 0.659 0.794
BI_MPA 1.162 0.540 0.744 1.220 1.061 1.335 0.950 0.951 0.867 0.875 0.970
BI_ACSM 0.746 0.744 0.544 1.164 0.715 0.882 0.900 0.796 1.359 1.000 0.885
BI_AMC 1.084 0.989 0.709 1.543 1.119 1.173 1.209 1.224 2.102 1.566 1.272
BI_MAMC 0.850 0.843 0.622 1.323 0.739 1.184 0.980 0.839 1.451 1.044 0.987
Median 0.982 0.673 0.613 1.204 0.807 1.245 0.994 0.870 1.167 0.969 0.952
Chapter 8 Results
67
8.3 Bridge Index (BI) based on spectrum-matching scaled records
In agreement with what was mentioned in Section 5.2.2, the records applied within this
approach are known to feature high energy and frequency content, which renders them
particularly demanding for some structures. Furthermore, some Nonlinear Static Procedures,
such as CSM, N2, DCM and MMPA, are originally formulated with a view to the use of
design smoothed, rather than real, response spectra. A second approach was thus followed, in
which design spectrum was used within the four mentioned NSP methodologies, whereas the
median real spectrum was used within ACSM, AMCP and MAMCP.
These two spectra, applied to the different NSPs, are compared in Figure 8.15. The period of
the population of bridges generally ranges between 0.4s and 1.2s. Within this range, according
to Figure 8.15, there is a relatively good matching between the design spectrum of NEHRP
and the median spectra.

Figure 8.15. Acceleration spectra
In order to minimize the influence of the uncertainties related to the different records, the
mean nonlinear dynamic result is calculated using the maximum displacements deriving from
the ten time-history analyses.
Four different scale factors are used for the consideration of five different intensity levels
(0.5,1.0,1.5,2.0 and 2.5), which means amplifying the median spectrum and design spectrum
using those scale factors.
0 0.5 1 1.5 2 2.5 3 3.5 4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
S
a
(
g
)
Period(s)

NEHRP
Median
Chapter 8 Results
68
Figure 8.16 provides the results obtained when employing Capacity Spectrum Method.

Figure 8.16. CSM median Bridge Index per intensity level based on scaled spectrum
It can be seen that with increasing of intensity level, the method provides more accurate
estimates, which is a similar trend to one observed when using individual record derived
spectra.
Furthermore, similar results are found when using N2, DCM and MMPA, as evidenced by
Figures 8.17 to 8.19.

Figure 8.17. N2 median Bridge Index per intensity level based on scaled spectrum

Figure 8.18. DCM median Bridge Index per intensity level based on scaled spectrum
0 0.5 1 1.5 2 2.5 3
0.4
0.6
0.8
1
1.2
1.4
1.6
M
e
d
i
a
n

B
I
Intensity Levels
BI CSM

CSMuni
CSM1st
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI N2

N2uni
N21st
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI DCM

DCMuni
DCM1st
Chapter 8 Results
69

Figure 8.19. MMPA median Bridge Index per intensity level based on scaled spectrum
Figure 8.17 to 8.19 present the performance of the procedures that make use of smoothed
design spectra. Among the four approaches, N2, closely followed by MMPA, provides the
most accurate estimates, when compared to dynamic analysis.
The latter presents a slight drop down branch for high intensity levels. A possible reason for
this might be the fact that when calculating the performance point, elastic behaviour is
assumed for the higher modes, which might not necessarily happen with considerably high
intensity levels.
Figures 8.20 to 8.22 represent the results obtained with the procedures that make use of
median response spectrum of the ten selected records.

Figure 8.20. ACSM median Bridge Index per intensity level based on scaled spectrum

Figure 8.21. AMCP median Bridge Index per intensity level based on scaled spectrum
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI MMPA
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI ACSM
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI AMCP
Chapter 8 Results
70

Figure 8.22. MAMC median Bridge Index per intensity level based on scaled spectrum
Results show that AMCP generally provides overestimated results, when compared to
dynamic analysis, whereas, on the other hand, by considering elastic behaviour for higher
modes, MAMCP yields fairly good estimates. On the other hand, ACSM yields similar
estimates to the ones obtained when using real ground motion spectrum, which utilizing the
equivalent viscous damping model and spectrum reduction factor, rather than inelastic
spectrum, have led to underestimated predictions.
Figure 8.23 plots the median BIs for the tested methods all-together.

Figure 8.23. Median Bridge Index per intensity level based on scaled spectrum for each method
When compared to Figure 8.11, Figure 8.23 shows that when using a median design response
spectrum, the dispersion among the different NSPs is considerably lower than the one
associated to the real records spectrum approach. Moreover, ACSM and MAMCP generally
provide the best estimates, whilst AMCP tends to overestimate the nonlinear dynamic
response predictions.
In order to render thr presentation of the results of the seven NSPs more clear, two different
sets of NSPs, traditional and improved, were again plotted in Figures 8.24 and 8.25.
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI MAMCP
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI

CSM
N2
DCM
MMPA
ACSM
AMCP
MAMCP
Chapter 8 Results
71

Figure 8.24. Median BI for each intensity levels of CSM, N2 and DCM

Figure 8.25. Median BI for each intensity levels of MMPA, ACSM, AMCP and MAMCP
It can be seen from Figure 8.24 and 8.25 that the previous conclusion can be confirmed that
the higher the seismic intensity, the closer the NSPs results get to the dynamic ones.
Moreover, the second set of the approaches provide better performance than the first one.
The global median results (i.e. across all the NSPs) for each intensity level, and for each NSP
are presented in Figures 8.26 and 8.27, respectively, which demonstrate that the conclusion
that the accuracy of the predicitons increases with intensity level is again confirmed.

Figure 8.26. Average values for each intensity
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI

CSM
N2
DCM
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
BI

MMPA
ACSM
AMCP
MAMCP
0 0.5 1 1.5 2 2.5 3
0.5
1
1.5
M
e
d
i
a
n

B
I
Intensity Levels
Displacement BI
Chapter 8 Results
72

Figure 8.27. Median Bridge Index for each NSPs
Finally, Figure 8.28 represents the comparison between the two different approaches, in terms
of global median Bridge Index for each NSP.

Figure 8.28. Median Bridge Index for each NSPs with two approaches
Figure 8.28 shows that N2, MMPA and ACSM yield similar and the most accurate estimates,
CSM and DCM underestimate and AMCP overestimates the nonlinear dynamic response in
terms of deck displacements. In addition, the use of two approaches allows the verification of
the relative performance of each NSP, together with the fact that the first approach, which
makes use of non-scaled records, provides predictions closer to the dynamic analysis.
Table 8.2 shown illustrates the numbers behind the different plots, for median Bridge Index
(BI) for each intensity.
Table 8.2. Bridge Index for each intensity

0.5
1
1.5
M
e
d
i
a
n

B
I
Gloabal BI
0
0.5
1
1.5
2
M
e
d
i
a
n

B
I
Displacement BI

real spectra
median spectra
Intensity Level 0.5 1 1.5 2 2.5 Median
BI_CSM1st 0.992 0.744 0.836 0.878 0.901 0.878
BI_N21st 0.743 0.815 0.877 0.926 0.955 0.877
BI_DCM1st 0.603 0.812 0.732 0.730 0.719 0.730
BI_MPA 0.859 0.774 0.893 0.948 0.867 0.867
BI_ACSM 0.934 0.818 0.817 0.936 0.901 0.901
BI_AMC 1.129 1.032 1.096 1.206 1.161 1.129
BI_MAMC 0.833 0.841 0.847 0.890 0.893 0.847
Chapter 8 Results
73
As one can see from the table, accurate estimates are obtained not only the higher intensity
levels but also, for some methods, at the intensity level of 0.5, to provide comparatively good
results. The better quality for some intensity levels, with respect to others, may have to do
with the regular shape of the used spectra around the intersection region, which thus features
lower uncertainty.
The next section presents the results obtained with the different NSPs, in terms of fragility
functions, which have been calculated using of non-scaled real spectra, given that, according
to Figure 6.33, such approach leads to more accurate predictions.

8.4 Fragility Curves
Response estimation for fragility curves calculations was based on statistic analysis and
lognormal distribution has been assumed for each fragility curve as a reasonable choice. The
procedure is described as follows:
1. Derive the linear regression function of the relationship between natural logarithm
distribution of Probability of exceedance and ln(PGA);
2. Calculate according to the linear function;
3. Calculate the lognormal distribution for each PGA and get the curve.
As mentioned in Chapter 6, the global displacement ductility for each bridge has been
assumed as the highest among the different piers.
The same set of 50 ground motion records described in Section 4.2 was used for calculating
fragility functions.
8.4.1 Nonlinear Static Procedures
The mean value and dispersion, lamda and kesi, for the lognormal distribution calculated
using the predictions of the different NSPs are presented in Table 8.3 to 8.9 and the
corresponding fragility functions, calculated for the different limit states, are plotted in Figure
8.29. Each NSP was applied in its optimized fashion, according to the results of Section 8.2.
! ! " = = b
m
,
1
Chapter 8 Results
74
Table 8.3. Statistics of fragility functions by NSPs
NSP LS Lamda Kesi
CSM
LS1 -1.167 0.260
LS2 -0.941 0.257
LS3 -0.501 0.293
LS4 -0.098 0.334
N2
LS1 -0.859 0.212
LS2 -0.731 0.222
LS3 -0.360 0.280
LS4 0.201 0.375
DCM
LS1 -1.389 0.333
LS2 -1.202 0.334
LS3 -0.972 0.293
LS4 -0.795 0.289
MMPA
LS1 -0.780 0.322
LS2 -0.594 0.329
LS3 -0.285 0.303
LS4 0.267 0.387
ACSM
LS1 -0.691 0.229
LS2 -0.342 0.273
LS3 0.010 0.359
LS4 0.987 0.490
AMCP
LS1 -1.090 0.210
LS2 -0.915 0.213
LS3 -0.800 0.214
LS4 -0.499 0.265
MAMCP
LS1 -1.132 0.179
LS2 -0.999 0.204
LS3 -0.790 0.221
LS4 -0.412 0.271

Chapter 8 Results
75

Figure 8.29. Fragility curves of NSPs

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility CSM

LS1
LS2
LS3
LS4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility N2

LS1
LS2
LS3
LS4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility DCM

LS1
LS2
LS3
LS4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility MMPA

LS1
LS2
LS3
LS4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility ACSM

LS1
LS2
LS3
LS4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility AMCP

LS1
LS2
LS3
LS4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility MAMCP

LS1
LS2
LS3
LS4
Chapter 8 Results
76
8.4.2 Nonlinear Dynamic Analysis
The mean value and dispersion, lamda and kesi, for the lognormal distribution calculated
using nonlinear dynamic analysis predictions are presented in Table 8.10 and the
corresponding fragility functions, calculated for the different limit states, are plotted in Figure
8.30.
Table 8.4 Statistics of fragility functions by Nonlinear Dynamic Analysis

Figure 8.30. Fragility curves for nonlinear dynamic analysis
8.4.3 Comparison of results
The comparison of static- and dynamic-based fragility curves, for each of the limit states, is
illustrated in Figure 8.31.
lamda kesi
LS1 -0.870 0.242
LS2 -0.626 0.271
LS3 -0.309 0.309
LS4 0.146 0.382
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility Dynamic

LS1
LS2
LS3
LS4
Chapter 8 Results
77

Figure 8.31. General results of each method for each Limit states
Among the seven selected nonlinear static procedures, N2 provides the closest to dynamic
analysis estimations, featuring also with the lower computational onus efforts, followed by
MMPA. Generally, the methods requiring the definition of an equivalent viscous damping
model, such as ACSM, AMCP and MAMCP, yield fragility curves farther from dynamic
estimates. On the contrary, the methods making use of inelastic spectra, i.e. N2 and MMPA,
did get closer to dynamic analysis.
Adaptive pushover based methods, such as ACSM, did not necessarily provide equally
improved predictions, as observed in previous studies [Pinho et al., 2009]. A possible reason
could be related to the equivalent viscous damping model used in the procedure, which may
be still leading to underestimation of the displacements when using ground motion records
with low peak ground accelerations, such as the ones applied in this study.
Given that many records feature low peak ground acceleration, the chosen intensity measure,
higher modes did not play a predominant role, since the structures did not go far beyond the
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility LS1

CSM
N2
DCM
MMPA
ACSM
AMCP
MAMCP
Dynamic
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility LS2

CSM
N2
DCM
MMPA
ACSM
AMCP
MAMCP
Dynamic
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility LS3

CSM
N2
DCM
MMPA
ACSM
AMCP
MAMCP
Dynamic
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
E
x
c
e
e
d
e
n
c
e

P
r
o
b
a
b
i
l
i
t
y
PGA(g)
Fragility LS4

CSM
N2
DCM
MMPA
ACSM
AMCP
MAMCP
Dynamic
Chapter 8 Results
78
elastic range thus AMCP and MAMCP provided similar results. Adding this to the
computational effort required for carrying out adaptive pushover analysis for each relevant
mode, such procedures can become rather elaborated while still not providing more accurate
estimates of fragility curves. The author would thus not recommended AMCP or MAMCP for
the prediction of seismic performance of bridges, by means of fragility curves, within loss
assessment studies.
CSM and DCM did not succeed as well to yield accurate estimates, when compared to
nonlinear dynamic analysis, in tandem with what had been verified before, in terms of Bridge
Index results.
Chapter 9 Closing Remarks
79
9 Closing Remarks
9.1 Conclusions
A number of past studies have addressed the ability of nonlinear static procedures to,
individually or comparatively to alternative counterparts, assess the structural performance of
single bridges. Much less attention has been paid, on the other hand, to such evaluation within
a given region context, comprising a population of bridges, assessing the performance of the
different approaches in terms of vulnerability models, loss estimations or seismic risk
calculations. In this study, seven different nonlinear static procedures were applied to
investigate the most suitable analytical approach(es) for deriving fragility functions of bridges
for use in loss assessment studies.
In order to fulfil such goal, a sufficient large number (200) of 3D bridge models, using a
representative bridge configuration of three piers and four spans, were randomly generated
using a fibre model based structural analysis software (OpenSees). The generation of different
bridges was based on the variation of a number of parameters, which included concrete
strength, steel strength, pier heights, column diameter and bay lengths. For all those
considered parameters a statistical distribution was defined.
After generating a detailed 3D model, two different pushover analyses were carried out,
conventional and adaptive. The former was tested with two different load distribution patterns:
uniform and first mode proportional. Based on the pushover curves derived from the
nonlinear static analysis, seven nonlinear static procedures, which include Capacity Spectrum
Method (CSM), Displacement Coefficient Method (DCM), N2, Modified Modal Pushover
Analysis (MMPA), Adaptive Capacity Spectrum Method (ACSM), Adaptive Modal
Combination Procedure (AMCP) and Modified Adaptive Modal Combination Procedure
(MAMCP), were employed for estimating response displacement for a number of ground
motion records.
As a tool to measure accuracy of the nonlinear static procedures, nonlinear dynamic analyses
were carried out, using two different sets of ground motion records. One consisted of non-
scaled real records selected from PEER database, which presented similar fault mechanisms
and potential seismic intensity to the Italy territory characteristics. The second consisted of
80
scaled records in order to match a smoothed design spectrum, associated to a predefined
exceedance probability.
Damage states were defined for different levels in terms of displacement ductility, which was
considered as the highest value amongst the different piers. Probability of exceeding a
particular damage limit state was determined for each bridge by checking whether the
maximum response displacement calculated through NSPs or Nonlinear dynamic analysis
went beyond the limit state threshold. Lognormal distribution functions were used to
characterize the fragility functions for each method.
The comparison of the response estimates derived from NSPs and nonlinear dynamic analysis
was carried out in terms of Bridge Index, a ratio that directly divides the response
displacement from each NSP by the maximum response displacement from nonlinear
dynamic analysis, and in terms of corresponding fragility functions, using different NSP
estimates.
The main conclusions from the results of this study can be summarized in the following
points:
Generally, Nonlinear Static Procedures were able to provide reasonable predictions in
terms of response displacements, when compared to the nonlinear dynamic estimates.
Notable exceptions include the Capacity Spectrum Method (CSM) and Displacement
Coefficient Method (DCM), which yielded poor estimates, mostly due to the fact that
they constitute pioneering approaches (CSM) or feature a strong empirical basis (DCM)
which can be hard to apply in bridge analysis. The Adaptive Modal Combination
Procedure (AMCP) has considerably overestimated the dynamic results, for high ground
motion intensity levels, whereas for low intensity levels it provided similar results to
Modified Adaptive Modal Combination Procedure (MAMCP). The Adaptive Capacity
Spectrum Method (ACSM) provided fairly good estimates for large peak ground
acceleration records, while, for low intensity ground motion records, it tends to
underestimate the response displacement. N2, followed by Modified Modal Pushover
Analysis (MMPA), provided quite accurate estimates, with respect to nonlinear dynamic
results, for both high and low intensity levels.
81
Globally, nonlinear static procedures provided better estimates with increasing ground
motion intensity. The improved methods, which make use of adaptive pushover analysis
and take higher modes contribution into account did not perform significantly better than
the traditional ones. Indeed, if a single procedure should be selected, by taking
computational effort and time in consideration, N2 would rank first among the seven
NSPs.
When comparing the results of fragility curves derived from different nonlinear static
approaches, one can see that the conclusion summarized by BI could be confirmed. i.e.,
N2 provides the closest results to the nonlinear dynamic ones, followed by MMPA. CSM
and DCM, initially proposed for building structures, failed to provide good estimates of
fragility functions of bridges. As the set of ground motion records considered in fragility
analysis did not send the structures highly into the nonlinear range, AMCP and MAMCP
yielded similar results due to the elastic behaviour of higher modes. ACSM, making use
of equivalent viscous damping, turned out to underestimate the results when applied with
low intensity ground motion records. Moreover, there is still significant dispersion among
the different methodologies, which, apart from the conceptual differences, might be
related to the definition of limit states or the lognormal distribution chosen.

9.2 Future Recommendations
The main goal of this study was to provide an overall perspective of the different available
analytical approaches for the derivation of fragility functions of bridges. Some details and
simplified assumptions could be further addressed more precisely thus the author would like
to refer to some possible future developments.
With respect to 3D modelling, the stiffness of the abutments plays an important role,
which has influence in the results of pushover analysis. However, in this study only one
kind of abutment was taken into consideration thus no variability was considered within
the randomly generated bridges. More attention should be paid to the stiffness of the
abutments without necessarily going through a full abutment design procedure, by, e.g.
establishing a simplified relationship between the stiffness of the abutments and the
bay/deck lengths;
82
In this study, only one bridge configuration was considered. However, different bridge
configurations could have a critical influence on the seismic assessment. Thus the
consideration of additional bridge configurations, possibly based on real case study will
be developed in the future;
Other intensity measures could be considered, when calculating fragility functions,
instead of PGA. Possible alternatives, such as spectral acceleration (Sa) or Arias Intensity
(ASI), could be considered in a short parametric study, to list the best performing
intensity measure;
Further refinement could be considered with respect to the type of distribution used to
characterize the fragility functions through nonlinear regression analysis;
The comparison of different NSPs should be extended to the calculation of vulnerability
functions and loss estimates.

References
83

10 REFERENCES

ATC [1996] Seismic Evaluation and Retrofit of Concrete Buildings, Volumes 1 and 2. Report No.
ATC-40, Applied Technology Council, Redwood City, CA.
ATC [1997] NEHRP Guidelines for the Seismic Rehabilitation of Buildings. Report No. FEMA-273,
Federal Emergency Management Agency, Washington, DC.
ATC [2000] Prestandard and Commentary for the Seismic Rehabilitation of Buildings. Report No.
FEMA-365, Federal Emergency Management Agency, Washington, DC.
ATC [2005] Improvement of Nonlinear Static Seismic Analysis Procedures. Report No. FEMA-440,
Federal Emergency Management Agency, Washington, DC.
Antoniou, S. and Pinho, R. [2004] Development and verification of a displacement-based adaptive
pushover procedure, Journal of Earthquake Engineering, Vol.8, No.5, pp. 643-661.
Aktan, A. E. and Ersoy, U. [1979] Analytical Study of R/C Material Hysteresis, Proceedings of the
AICAP-CEB Symposium on Structural Concrete Under Seismic Actions, CEB Bulletin
d'Information No. 132, pp. 97-104.
Avsar, O. [2009] Fragility based seismic vulnerability assessment of ordinary highway bridges in
Turkey, Thesis, Middle East Technical University, Turkey.
Bal, I.E., Crowley, H., Pinho, R., Gulay, F.G. [2008] Detailed assessment of structural characteristics
of Turkish RC building stock for loss assessment models, Soil Dynamics and Earthquake
Engineering, Vol. 28, pp. 914-932.
Bal, I.E., Crowley, H., Pinho, R. [2010] Displacement-based Earthquake Loss Assessment: Method
Development and Application to Turkish Building Stock, ROSE Research Report, IUSS Press,
Pavia, Italy.
Banerjee, S., Shinozuka, M. [2008] Mechanistic quantification of RC bridge damage states under
earthquake through fragility analysis, Probabilistic Engineering Mechanics, Vol.23, No.1, pp. 12-
22.
References
84
Bommer,J.J.,Martinez-Pereira,A. [1999] The effective duration of earthquake strong motion,
Journal of Earthquake Engineering, Vol.3, No. 2, pp. 127-172.
CALTRANS,(various dates), Bridge Memo To Designers,California Department of Transportation,
Sacramento, California.
Cardone, D., Perrone, G., Sofia, S. [2011] A performance based adaptive methodology for the
seismic evaluation of multi-span simply supported deck bridges, Bulletin of Earthquake
Engineering, Vol. 9, No. 5, pp. 1463-1498.
Casarotti, C., Pinho, R. and Calvi, G. M. [2005] Adaptive pushover-based methods for seismic
assessment and design of bridge structures. ROSE Research Report No. 2005/06, IUSS Press,
Pavia, Italy
Casarotti, C. and Pinho, R. [2006] Seismic response of continuous span bridges through fiber-based
finite element analysis, Earthquake Engineering and Engineering Vibration, Vol.5 No. 1, pp. 119-
131.
Casarroti, C., Pinho, R. [2007] An adaptive capacity spectrum method for assessment of bridges
subjected to earthquake action, Bulletin of Earthquake Engineering, Vol. 5, No. 3, pp. 377-390.
Casarotti, C., Monteiro, R. and Pinho, R. [2009] Verification of spectral reduction factors for seismic
assessment of bridges, Bulletin of the New Zealand Society for Earthquake Engineering, Vol. 42,
No. 2, pp. 111-121.
CEN [2005a] Eurocode 8: Design of Structures for Earthquake Resistance - Part 1: General rules,
seismic actions and rules for buildings. EN 1998-2, Comit Europen de Normalisation, Brussels,
Belgium.
CEN [2005b] Eurocode 8: Design of Structures for Earthquake Resistance - Part 2: Bridges. EN
1998-2, Comit Europen de Normalisation, Brussels, Belgium
Choi, E., DesRoches, R., Nielson, B. [2004] Seismic fragility of typical bridges in moderate seismic
zones, Engineering Structures, Vol. 26, No. 2, pp. 187-199.
Chopra, A. K. [1995] Dynamics of Structures: Theory and Applications to Earthquake Engineering.
Prentice-Hall, Englewood Cliffs, NJ.
Chopra, A. K. and Goel, R. K. [2001] A modal pushover analysis procedure to estimating seismic
demands for buildings: Theory and preliminary evaluation. PERR Report 2001/03, Pacific
Earthquake Engineering Research Center, University of California, Berkeley, CA.
Chopra, A. K. and Goel, R. K. [2002] A modal pushover analysis procedure for estimating seismic
demands for buildings, Earthquake Engineering & Structural Dynamics, Vol.31,No.3, pp. 561-
582.
Chopra, A.K., Geol, R.K., Chintanapakdee, C. [2004] Evaluation of a modified MPA procedure
assuming higher modes as elastic to estimate seismic demands, Earthquake Spectra, Vol. 20, No.
3, pp. 757-778.
Elnashai, A.S., Borzi, B. [2004] Deformation-based vulnerability functions for RC bridges,
Structural Engineering and Mechanics, Vol. 17, No. 2, pp. 215-244.
References
85
Fajfar, P. and Fischinger, M. [1988] N2 - A method for non-linear seismic analysis of regular
buildings, Proceedings of the 9th World Conference in Earthquake Engineering, August 2-9,
Tokyo-Kyoto, Japan.
Fajfar, P. [1999] Capacity spectrum method based on inelastic demand spectra, Earthquake
Engineering & Structural Dynamics, Vol. 28, No. 9, pp. 979-993.
Fajfar, P. [2000] A nonlinear analysis method for performance based seismic design, Earthquake
Spectra, Vol. 16, No. 3, pp. 573-592.
Freeman, S. A., Nicoletti, J. P. and Tyrell, J. V. [1975] Evaluation of Existing Buildings for seismic
risk - A case study of Puget Sound Naval Shipyard, Bremerton, Washington, Proceedings of the
1st U.S. National Conference on Earthquake Engineering, Berkley, USA.
Geol, R.K., Chopra, A.K. [1997] Evaluation of bridge abutment capacity and stiffness during
earthquakes, Earthquake Spectra, Vol. 13, No. 1.
Hall, J.F. [2005] Problems encountered from the use (or misuse) of Rayleigh damping, Earthquake
Engineering and Structural Dynamics, Vol. 35, No.5, pp. 525-545.
HAZUS [2003], Multi-hazard Loss Estimation Methodology Earthquake Model, Federal Emergency
Management Agency, Washington, D.C., USA.
Hwang, H., Liu,J.B., Chiu, Y. [2001] Seimic fragility analysis for highway bridges, Individual study,
Center for Earthquake Research and Information, University of Memphis, USA.
Iman, R. L., Helson, J. C. and Campbell, J. E. [1981] An Approach to Sensitivity Analysis of
Computer Models: Part I - Introduction, Input Variable Selection and Preliminary Variable
Assessment. Journal of Quality Technology, Vol.13, No.3, pp. 174-183.
Jeong, S., Elnashai, A.S. [2007] Probabilistic fragility analysis parameterized by fundamental
response quantities, Engineering Structures, Vol.29, No. 6, pp. 1238-1251.
Kalkan, E., Kunnath, S.K. [2006] Adaptive modal combination procedure for nonlinear static
analysis of building structures, Journal of Structural Engineering, Vol. 132, No. 11, pp. 1721-
1731.
Kappos, A.J., Chryssanthopoulos, M.K., Dymiotis, C. [1999] Uncertainties analysis of strength and
ductility of confined reinforced concrete members, Engineering Structures, Vol. 21, pp. 195-208.
Kappos, A.J., Moschonas, I., Paraskeva, T., Sextos, A. [2006] A methodology for derivation of
seismic fragility curves for bridges with the aid of advanced analysis tools, First European
Conference on Earthquake Engineering and Seismology, Geneva, Switzerland.
Karakostas, C., Makarios, T., Lekidis, V., Kappos, A. [2006] Evaluation of vulnerability curves for
bridges-a case study, First European Conference on Earthquake Engineering and Seismology,
Geneva, Switzerland.
Karim, K.R., Yamazaki, F. [2003] a simplified method of constructing fragility curves for highway
bridges, Earthquake Engineering and Structural Dynamics, Vol. 32, No.10, pp. 1603-1626.
Karsan, I.D, Jirsa, J.O. [1969] Behaviour of concrete under compressive loading, Structural
Division, ASCE, USA, 95(2).
References
86
Kibboua, A., Naili, M., Benouar, D., Kehila, F. [2011] Analytical fragility curves for typical Algerian
reinforced concrete bridge piers, Structural Engineering and Mechanics, Vol.39, No. 3, pp. 411-
425.
Kowaslsky, M.J. [2000] Deformation limit states for circular reinforced concrete bridge columns,
Journal of Structural Engineering, Vol. 126, No.8, pp. 869-878.
Liao, W., Loh, C.H. [2004] Preliminary study on the fragility curves for highway bridges in Taiwan,
Journal of the Chinese Institute of Engineers, Vol. 27, No. 3, pp. 367-375.
Lupoi, G., Franchin, P., Lupoi, A., Pinto, P.E. [2006] Seismic fragility analysis of structural systems,
Journal of Engineering Mechanics, Vol. 132, No. 4, pp. 385-395.
Mander, J.B. [1999] Fragility curves development for assessing the seismic vulnerability of highway
bridges, Individual Study, University at Buffalo, State University of New York, USA.
McKay, M. D., Beckman, R. J. and Conover, W. J. [1979] A Comparison of Three Methods for
Selecting Values of Input Variables in the Analysis of Output from a Computer Code.
Technometrics, Vol.21, No.2, pp. 239-245.
Monteiro, R. [2011] Probabilistic Seismic Assessment of Bridges, Thesis, Faculty of Engineering,
University of Porto, Portugal.
Moschonas, I.F., Kappos, A.J., Panetsos, P., Papadopoulos,V., Makarios, T., Thanopoulos, P. [2009]
Seismic fragility curves for greek bridges: methodology and case studies, Bulletin Earthquake
Engineering, Vol. 7, pp. 439-468.
Nielson, B.G. [2005] Analytical fragility curves for highway bridges in moderate seismic zones,
Thesis, School of Civil and Environment Engineering, Georgia Institute of Technology, USA.
Nielson, B.G., DesRoches, R. [2007] Seismic fragility methodology for highway bridges using a
component level approach, Earthquake Engineering and Structural Dynamics, Vol. 26, No. 6 pp.
823-839.
Papanikolaou, V.K., Elnashai, A.S. [2005] Evaluation of conventional and adaptive pushover
analysis I: Methodology, Journal of Earthquake Engineering, Vol. 9, No. 6, pp. 923-941.
Pinho, R. and Antoniou, S. [2005] A displacement-based adaptive pushover algorithm for assessment
of vertically irregular frames, Proceedings of the 4th European Workshop on the Seismic
Behaviour of Irregular and Complex Structures, August 26-27, Thessaloniki, Greece.
Pinho, R., Casarotti, C. and Monteiro, R. [2007] An Adaptive Capacity Spectrum Method and other
Nonlinear Static Procedures Applied to the Seismic Assessment of Bridges, Proceedings of the 1
st

US-Italy Seismic Bridge Workshop, April 19-20, Pavia, Italy.
Pinto, P.E. [2006] Seismic assessment of concrete bridges, Proceeding of the 2
nd
International
Congress, Naples, Italy.
Priestley, M.J.N., Seible, F., Calvi, G.M. [1996] Seismic Design and Retrofit of Bridges, Wiley-
Interscience, New York.
Priestley, M.J.N., Grant, D.N. [2005] Viscous damping in seismic design and analysis, Journal of
Earthquake Engineering, Vol. 9, No. 2, pp. 229-255.
References
87
Priestley, M. J. N., Calvi, G. M. and Kowalsky, M. J. [2007] Displacement-based Seismic Design of
Structures. IUSS Press, Pavia, Italy.
Rossetto, T., Elnashai, A. [2005] A new analytical procedure for derivation of displacement-based
vulnerability curves for populations of RC structures, Engineering Structures, Vol. 27, No. 3, pp.
397-409.
Shinozuka, M., Feng, M.Q., Kim, H.K., Kim, S.H. [2000] Nonlinear static procedure for fragility
curve development, Journal of Engineering Mechanics, Vol. 126, No. 12, pp. 1287-1295.

Web Reference:
[1]OpenSees: http://opensees.berkeley.edu/
[2]PEER strong motion database: http://peer.berkeley.edu/smcat/

MSC Dissertation Zhang 2012

Hochgeladen von

Dokumentinformationen

Originaltitel

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

MSC Dissertation Zhang 2012

Hochgeladen von

Copyright:

Verfügbare Formate

DIFFERENT APPROACHES TO DERIVE

ANALYTICAL FRAGILITY FUNCTIONS

Das könnte Ihnen auch gefallen