Sie sind auf Seite 1von 627

CILAMCE 2018

XXXIX Ibero-Latin American Congress on


Computational Methods in Engineering

November 11-14, 2018


Paris and Compiègne, France

Editors:
A. Ibrahimbegović, P.M. Pimenta
CILAMCE 2018 Proceedings of XXXIX
Ibero-Latin American Congress on
Computational Methods in Engineering
Adnan Ibrahimbegovic and Paulo de Mattos Pimenta
c Adnan Ibrahimbegovic and Paulo de Mattos Pimenta

Édition : Adnan Ibrahimbegovic and Paulo de Mattos Pimenta


17 rue Ferdinand Sarrazin, 60280 Margny les Compiegne, France

ISBN : 978-2-9565961-0-3

Dépôt Légal : Novembre 2018


ISBN 978-2-95-659610-3
0.00 EUR

9 782956 596103
CILAMCE 2018 : XXXIX Ibero-Latin American Congress on
Computational Methods in Engineering
November 11-14, 2018, Paris/Compiègne, France
EDITORIAL
Adnan Ibrahimbegovic , Paulo M. Pimenta
1 2

1
University of Technology Compiegne – Sorbonne Universités, 60200 Compiegne, France, adnan.ibrahimbegovic@utc.fr
2
Escola Polytechnica, University of Sao Paulo, 380-05508-010 Sao Paulo, Brazil, ppimenta@usp.br

Abstract: CILAMCE 2018 is the thirty ninth in the series of International Congresses focusing upon
Computational Methods in Engineering, held each year in Brazil. Previous meeting have all been
organized under umbrella of Brazilian association abmec, ‘Associação Brasileira de Métodos
Computacionais em Engenharia’, as the sponsor of CILAMCE Congress series. For this particular
Congress, we have additional two sponsors from France: CSMA, Association of Computational
Mechanics, and SMAI, Association of Applied and Industrial Mathematics. The institutions organizing
CILAMCE 2018 are University of Technology Compiegne, a member Sorbonne Universities, and Escola
Politecnica of University of Sao Paulo. The CILAMCE 2018 is held from 11 to 13 November 2018 in
Compiegne, with the final day of November 14 in Paris. The Congress seeks to provide a platform for
learning from some of the worlds' leading specialists in numerical methods, coming from engineering
disciplines and applied mathematics.

1 OBJECTIVES OF CILAMCE 2018 CONGRESS


The main idea of this CILAMCE Congress is to examine recent advances in numerical methods
in currently most active research domains, with applications to interface and/or interaction of
several engineering disciplines. The multi-physics models and methods of this kind are often
bridging the phenomena taking place at multiple scales in space and time, which ought to be
placed in interaction or accounted for simultaneously in order to provide the most reliable results
explanations. This class of problems calls for the development and combination of different
modeling tools and computational methods in order to advance the field towards currently
relevant industrial applications. A number of different schools have developed in various
domains, both in engineering sciences and mathematics, with sometimes very little or no
interaction between them. It is an explicit goal of this CILAMCE Congress to bring all the
different communities together, from Brazil and from France, in the truly open scientific spirit,
and thus provide a sound basis for a fruitful exchange and crossfertilization of ideas among them.

i
2 INTERNATIONAL SCIENTIFIC COMMITTEE
Abimael F. D. Loula (LNCC), Brazil
Adair Roberto Aguiar (EESC/USP), Brazil
Adnan Ibrahimbegovic, France
Alberto Cardona, Argentina
Alfredo Gay Neto, Brazil
Álvaro Luiz Gayoso de Azeredo Coutinho (UFRJ), Brazil
André Teófilo Beck (USP/São Carlos), Brazil
Anthony Gravouil, France
Bernardo Horowitz (UFPE), Brazil
Bert Sluys, Netherlands
Carlos Magluta (COPPE/UFRJ), Brazil
Carlos Manuel Tiago Tavares Fernandes (IST/Portugal), Portugal
Cheng Liang Yee (USP), Brazil
Christianne de Lyra Nogueira (UFOP), Brazil
Delphine Brancherie, France
Eduardo de Miranda Batista (UFRJ), Brazil
Eduardo Fancello (UFSC), Brazil
Eduardo M. B. Campello (USP), Brazil
Eduardo M. R. Fairbairn (COPPE/UFRJ), Brazil
Emílio Carlos Nelli Silva (USP), Brazil
Estevam Barbosa de Las Casas (UFMG), Brazil
Eugenio Oñate, Brazil
Evandro Parente Jr (UFC), Brazil
Felício Bruzzi Barros (UFMG), Brazil
Fernando Alves Rochinha (UFRJ), Brazil
Florian De Vuyst, France
Francisco Chinesta, France
Francisco Evangelista Jr (UnB),Brazil
Gilberto Gomes (UnB),Brazil
Glaucio H. Paulino (Georgia Tech), Brazil
Gray Farias Moita (CEFET/MG), Brazil
Hachmi Ben Dhia, France
Hélio José C. Barbosa (LNCC), Brazil
Henrique Campelo Gomes (USP), Brazil
Hermann Matthies, Germany
Jean-Philippe Ponthot, Belgium
João Costa Pantoja (UnB), Brazil
Jörg Schröder, Germany
Jorge Ambrósio, Portugal
José Guilherme Santos Da Silva (UERJ), Brazil
José Luís Drummond Alves (UFRJ), Brazil
Julien Yvonnet, France
Julio R. Meneghini, Brazil
Kam Liu Wing, U.S.A.
Leonardo da Silveira P. Inojosa (UnB), Brazil
Lineu José Pedroso (UnB), Brazil

ii
Luis Bitencourt, Brazil
Marcílio Alves, Brazil
Marcio Augusto Roma Buzar (UnB),Brazil
Márcio Muniz de Faria (UnB), Brazil
Marco Lúcio Bittencourt (UNICAMP), Brazil
Nelson F.F. Ebecken (UFRJ), Brazil
Ney Augusto Dumont (PUC/RJ), Brazil
Ney Roitman (COPPE/UFRJ), Brazil
Nicolas Moës, France
Pablo Javier Blanco (LNCC), Brazil
Patrícia de Oliveira Faria (UFSC), Brazil
Paulo B. Gonçalves (PUC-Rio), Brazil
Paulo M. Pimenta (USP), Brazil
Paulo R. M. Lyra (UFPE), Brazil
Peter Wriggers, Germany
Phillippe Devloo (UNICAMP), Brazil
Raul R. Silva (PUC-Rio), Brazil
Reyolando M.L.R.F. Brasil (UFABC), Brazil
Ricardo Azoubel da Mota Silveira (UFOP), Brazil
Roberto Dalledone Machado (PUC-PR), Brazil
Roger Ohayon, France
Ruy Marcelo Pauletti, Brazil
Sergio Idelsohn, Argentina
Sérgio Proença (USP/São Carlos), Brazil
Suzana Moreira Ávila (UnB), Brazil
Sylvia Regina Mesquita de Almeida (UFG), Brazil
Tadeusz Burczynski, Poland
Tod Laursen, U.A.E.
Yvon Maday, France
Zenon José Guzman Nuñez del Prado (UFG), Brazil

3 LOCAL ORGANIZING COMMITTEE


Sorbonne Universities-UTC : Prof. Adnan Ibrahimbegovic, Dr. Delphine Brancherie, Ms. Brigitte
Duch, Prof. Florian De Vuyst
University of Sao Paulo-EP : Prof. Paulo M. Pimenta, Prof. Eduardo M.B. Campello
Sorbonne Universities-UTC doctoral students : Mr. Ivan Rukavina, Mr. Pablo Moreno Navarro,
Ms. Tea Rukavina, Ms. Rosa Adela Mejia, Ms. Abir Boujelben, Mr. X.N. Do, Ms. S. Grbcic, Ms.
E. Hadzalic, Mr. I. Imamovic, Mr. E. Karavelic, Ms. S. Dobrilla

4 CILAMCE 2018 TOPICS


The Congress topics to be addressed concern not only ‘classical’ domains of Solid, Structural and
Fluid Mechanics, but also a number of currently ‘hot’ domains, such as: Heterogeneous
materials, Complex structures and systems, Material and structure failures, Adaptive modeling,
Mechanics of porous media, Fluid-structure interaction, Multi-phase flows, Model reduction,
Stochastic Processes, Uncertainty Propagation, Industrial Applications ...
The final day of the congress, scheduled at IUF location in Paris, will regroup invited plenary
lecturers giving overview of both current research in their fields and still open questions and
explorations to come. The keynote and standard lectures of each of first two congress days of this
congress at UT Compiegne, will address a number of issues concerning computational methods

iii
in wide variety of topics.
The following organized Mini-Symposia (containing 6 or more papers) are scheduled at the
congress:
Computational Methods for Image Processing and Analysis
Advanced Analysis in Steel and Steel-Concrete Composite Structures
Topology Optimization of Multifunctional Materials, Fluids and Structures
Modeling, Simulation and Control of Aerospace and Naval Structures Dynamics
Modelling, Design and Additive Manufacturing of Vibro-Acoustic Metamaterials
Boundary Element and Mesh-Reduced Methods
Numerical Modelling in Support to the Assessment and Strengthening of Existing Structures
Engineering Simulations using Finite Difference Methods
Structural Reliability Methods and Design Optimization Under Uncertainties
Analysis and Design of Offshore Systems
Numerical Simulation of Wave Propagation
Advances in the Lattice-Boltzmann Method in Computational Fluid Dynamics
Computational Fracture Mechanics (CFM)
Data Science Approaches and Applications
Experimental Investigation of Wave Propagation in Light Weight Structures Undergoing
Flexural Vibration
Health Monitoring and Numerical Modeling of Structures
Numerical Methods Applied to Structural Design of Civil Construction
Advances on the Design and Optimization of Membranes, Gridshells and Bending-active
Structures
Computational Plasticity: Monotonic and Cyclic Applications
Analysis of Stability and Deformation of Evaporites Cavity
On the Numerical Simulation of Discrete Fracture Networks
Ageing and material pathologies towards civil engineering structural response
Advances in Solid and Structural Mechanics
Advances Thermal Sciences and Heat Propagation Problems
Modern Industrial Applications
Computational Modelling of Damage in Materials, Solids and Structures
Fluid-Structure Interaction
Reduced Order Modeling in Computational Mechanics

5 CILAMCE 2018 - PROGRAM AT GLANCE


Sunday November 11, 2018, Compiègne, Welcome Event:
10 a.m. Armistice Day 100th anniversary: [Compiègne where the peace agreement between
France and Germany was signed to end The World War I, on November 11, 1918]

Monday/Tuesday November 12-13, 2018, UT Compiègne


9:30 a.m. – 5 p.m. Parallel Sessions MS : Université de Technologie Compiègne
Nov. 12, 6 p.m. Reception City Hall / Nov. 13, 5 p.m. Gen. Assembly abmec

Wednesday November 14, 2018, Paris


1:30 p.m. Plenary Session: IUF-Institut Universitaire de France, 1 rue Descartes, Paris
6 p.m. Guided Tour (Panthéon, Notre Dame, Musée Louvre …)
8 p.m. Banquet : Bateaux Mouches, [cruise on river Seine], Paris

iv
6 CILAMCE 2018 CONGRESS VENUES

FIG. 1 – COMPIEGNE / MONDAY 11 – TUESDAY 12 NOVEMBER 2018 (MINI-SYMPOSIA LECTURES)


UNIVERSITE DE TECHNOLOGIE COMPIEGNE, CENTRE DE RECHERCHE, France

location:
(WATCH FOR SIGN: CILAMCE 2018)
ADDRESS: 66, AVENUE DE LANDSHUT, 60201 COMPIEGNE, France

v
FIG. 2 – PARIS / WEDNESDAY, 14 NOVEMBER 2018 (PLENARY LECTURES)

IUF- MINISTERE DE L’ENSEIGNEMENT SUPRERIEUR, DE LA RECHERCHE ET DE L’INNOVATION
[ IUF - MINISTRY OF HIGHER EDUCATION, RESEARCH AND INNOVATION ]

location:
(WATCH FOR SIGN: CILAMCE 2018)
ADDRESS: 25, RUE DE LA MONTAGNE SAINTE GENEVIEVE, 75005 PARIS, FRANCE

vi
Contents

1 PLENARY LECTURES 1
1.1 ASPECTS OF MULTI-SCALE COUPLING OF INELASTIC PROCESSES IN SOLID MECHAN-
ICS
Hermann Matthies and Adnan Ibrahimbegovic . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 NUMERICAL MODELING OF FRACTURE IN HIGHLY HETEROGENEOUS MATERIALS
Julien Yvonnet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 FINITE ELEMENT ANALYSIS OF SLENDER BEAMS AND THIN SHELLS
Paulo de Mattos Pimenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 OLD AND NEW CONTRIBUTIONS ON REDUCED BASIS METHODS
Yvon Maday . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 SECTIONAL LECTURES 15
2.1 ENGINEERING STRUCTURES INTEGRITY AND DURABILITY VALIDATION
Adnan Ibrahimbegovic, Abir Boujelben, Xuan Nam Do, Simona Dobrilla, Emina Hadzalic, Sara
Grbčić, Ismar Imamovic, Emir Karavelić, Rosa Adela Mejia Nava, Pablo Moreno-Navarro, Mijo
Nikolic, Ivan Rukavina, Tea Rukavina . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 DISCRETE ELEMENT MODELING OF GRANULAR AND PARTICULATE MATERIALS
Eduardo M. B. Campello . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 A MIXED DATA-DRIVEN/TIME INTEGRATOR KERNEL-BASED APPROACH FOR MANI-
FOLD LEARNING OF DYNAMICAL SYSTEMS
Florian De Vuyst, Faker Ben Belgacem, Piotr Breitkopf, Georges Oppenheim and Pierre Villon . 25

3 OTHER LECTURES 29
3.1 EFFICIENT COMPUTATIONAL MODEL FOR FLUID – STRUCTURE INTERACTION IN
APPLICATION TO LARGE OVERALL MOTION OF GIANT WIND TURBINE FLEXIBLE
BLADES WITH LONG TERM RESPONSE
Abir Boujelben and Adnan Ibrahimbegovic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 STRAIN LOCALIZATIONS IN PERIDYNAMIC BARS WITH NON-CONVEX POTENTIAL
Adair Aguiar, Gianni Royer-Carfagni and Alan Seitenfuss . . . . . . . . . . . . . . . . . . . . . 35
3.3 EVALUATION OF TEMPERATURE AND DAMAGE OVER STRUCTURAL NATURAL FRE-
QUENCIES
Daniel Soares and Alexandre Cury . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 MODELING WHEEL-RAIL CONTACT INTERACTION AND VEHICLE DYNAMICS: OVERVIEW
OF ONGOING DEVELOPMENTS AND NEW ACHIEVEMENTS
Alfredo Gay Neto and Thiago Pereira . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.5 A NUMERICAL ASSESSMENT OF STIFFNESS ASSOCIATED WITH A SINGLE MOOR-
ING LINE: THE EFFECT OF SEA CURRENT AND APPLICATION FOR A FLOATING OFF-
SHORE WIND TURBINE
Giovanni Aiosa Do Amaral, Estevan Covari Isaak, Alfredo Gay Neto and Guilherme Rosa Franzini 42
3.6 IMPROVING PERFORMANCE OF FINITE DIFFERENCE 3D HETEROGENEOUS ACOUS-
TIC WAVE EQUATION SIMULATIONS
Carlos Barbosa, Schirley Jorge, Raphael Vilela, Luciano Leite, Jose Camata and Alvaro Coutinho 46
3.7 INFLUENCE OF PRECONDITIONING ON FINITE ELEMENT SIMULATION OF COLUMN
COLLAPSE USING A WELL-POSED µ(I)-RHEOLOGY
Linda Gesenhues, José J. Camata, Adriano M.A. Côrtes, Fernando A. Rochinha and Alvaro L.G.A.
Coutinho . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.8 NUMERICAL SIMULATION OF WIND ACTION ON SUPERSTRUCTURE OF A BRIDGE
AND DYNAMIC RESPONSE RATE FOR DIFFERENT DECK GEOMETRIES
Anaximandro Souza, Leonardo Lisboa and Rodrigo Melo . . . . . . . . . . . . . . . . . . . . . 52

vii
3.9 ANALYSIS AND STRUCTURAL CORRECTION OF A BELT CONVEYOR TR-315K- 03 FROM
VALE COMPANY IN BRAZIL USING COMPUTATIONAL MECHANICS TECHNIQUES
Anaximandro Souza, Carlos Magno and Luís Jorge de Mesquita . . . . . . . . . . . . . . . . . . 56
3.10 PROBABILISTIC PROGRESSIVE COLLAPSE OF A TRUSS: A BENCHMARK EXAMPLE
OF SYSTEM RBDO AND RISK-BASED OPTIMIZATION
André Teófilo Beck, Rodolfo Tessari and Henrique Kroetz . . . . . . . . . . . . . . . . . . . . . 59
3.11 UNCERTAINTY QUANTIFICATION & RISK ANALYSIS IN ENGINEERING, WITH APPLI-
CATIONS TO OIL & GAS EXPLORATION
André Teófilo Beck . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.12 DRYING OF CERAMIC BLOCKS IN AN INDUSTRIAL TUNNEL DRYER: MODELING AND
SIMULATION
Anderson Melchiades Vasconcelos Da Silva, Mariana Julie Do Nascimento Santos and Antonio
Gilson Barbosa de Lima . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.13 NON-ISOTHERMAL ABSORPTION OF WATER IN VEGETABLE FIBER REINFORCED POLY-
MER COMPOSITES: AN APPLICATION TO THE LANGMUIR-TYPE MASS TRANSPORT
MODEL AND THE FINITE-VOLUME METHOD
Rafaela Quinto Costa Melo, José Vieira Silva and Antonio Gilson Barbosa de Lima . . . . . . . 65
3.14 AN EFFICIENT DOMAIN DECOMPOSITION METHOD WITH CROSS-POINT TREATMENT
FOR HELMHOLTZ PROBLEMS
Axel Modave, Xavier Antoine and Christophe Geuzaine . . . . . . . . . . . . . . . . . . . . . . . 67
3.15 STOCK PRICE CHANGE PREDICTION USING NEWS TEXT MINING
Marcelo Beckmann, Nelson Ebecken and Beatriz De Lima . . . . . . . . . . . . . . . . . . . . . 71
3.16 VIGA BIENGASTADA DE CONCRETO SOBRE BASE ELÁSTICA SUBMETIDA A CAR-
REGAMENTO UNIFORMEMENTE DISTRIBUÍDO: ANÁLISE NUMÉRICA VIA FTOOL E
SAP2000
Breenda Lorrana Vieira Lima and Michell Macedo Alves . . . . . . . . . . . . . . . . . . . . . 77
3.17 TOWARDS A METHODOLOGY FOR THE OPTIMAL DESIGN OF A GROUP OF RISERS IN
LAZY-WAVE CONFIGURATION
Bruno Monteiro, Bruno Jacovazzo, Carl Albrecht and Breno Jacob . . . . . . . . . . . . . . . . 82
3.18 A COMPARATIVE STUDY OF NEIGHBOURING PARTICLE SEARCH METHODS IN COM-
PUTATIONAL FLUID MECHANICS: DIRECT SEARCH AND LINKED LIST TECHNIQUES
EVALUATION
Carlos Alberto Dutra Fraga Filho, Lucas Lustosa Schuina and Brenda Silva Porto . . . . . . . . 83
3.19 CONNECTION BETWEEN BERNOULLI-EULER RODS AND KIRCHHOFF-LOVE SHELLS
Cátia Silva, Sascha Maassen, Nils Viebahn, Paulo de Mattos Pimenta and Jörg Schröder . . . . . 87
3.20 NUMERICAL MODELING OF THE BRAZE WELDED ASSEMBLY OF A COPPER COATED
WITH A NICKEL-BASED ALLOY AND STEEL
Chawki Tahri, Christophe Bertoni, Eric Feulvarch, Helmut Klocker and Jean-Michel Bergheau . . 91
3.21 ON THE NUMERICAL SIMULATION OF DISCRETE FRACTURE NETWORKS
Philippe R. B. Devloo and Chensong Zhang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
3.22 SOME RESULTS ABOUT EXISTENCE AND UNIQUENESS SOLUTION FOR STOCHASTIC
BENDING BEAM
Claudio Roberto Ávila Da Silva Júnior and Pedro Danizete Damázio . . . . . . . . . . . . . . . 95
3.23 BUBNOV GALERKIN METHOD APPLIED EULER BERNOULLI BEAM STOCHASTIC BEND-
ING
Claudio Roberto Ávila Da Silva Júnior and Pedro Danizete Damázio . . . . . . . . . . . . . . . 102
3.24 A FULLY-CONSERVATIVE FINITE VOLUME FORMULATION FOR COUPLED POROE-
LASTIC PROBLEMS
Herminio T. Honorio, Felipe Giacomelli, Lucas G. T. Da Silva and Clovis R. Maliska . . . . . . . 111
3.25 A LAGRANGIAN TRANSPORT MODEL TO SIMULATE THE TRAJECTORY OF OCEANIC
FLOATS
Daiane Faller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.26 FINITE ELEMENT STRATEGIES FOR MODELING THE SIZE EFFECT IN NANO-REINFORCED
MATERIALS
Dang Phong Bach, Delphine Brancherie and Ludovic Cauvin . . . . . . . . . . . . . . . . . . . 117
3.27 CONCRETE QUANTITY MINIMIZATION FOR A SHALLOW FOUNDATION INTEGRAT-
ING PARTICLE SWARM OPTIMIZATION AND A SOIL STRUCTURE-INTERACTION ANAL-
YSIS.
David Sebastian Cotes Prieto, Cesar Andrés Mendez Poveda, William Giovanny Alfonso León and
Oscar Javier Begambre Carrillo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121

viii
3.28 ANALYTICAL-NUMERICAL STUDY OF VIBRATIONS IN CYLINDRICAL SHELL UNCOU-
PLED AND COUPLED WITH FLUID
Davidson De Oliveira França Júnior and Lineu José Pedroso . . . . . . . . . . . . . . . . . . . 125
3.29 STUDY OF FLUID-STRUCTURE INTERACTION THROUGH ADDITIONAL MASS IN CYLIN-
DRICAL TANKS UNDER FREE VIBRATIONS
Davidson De Oliveira França Júnior and Lineu José Pedroso . . . . . . . . . . . . . . . . . . . 130
3.30 THE STRONG DISCONTINUITY APPROACH FOR DUCTILE FAILURE WITH DAMAGE
Jeremie Bude and Delphine Brancherie . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
3.31 ULTRASONIC COMPUTED TOMOGRAPHY USING FULL-WAVEFORM INVERSION FOR
QUANTITATIVE IMAGING OF BONES IN THE VISCOELASTIC CASE
Dimitri Komatitsch, Simon Bernard, Vadim Monteiller, Philippe Lasaygues and Régine Guillermin 136
3.32 COMPARATIVO DE DIMENSIONAMNETO DOS SISTEMAS ESTRUTURAIS SLIM FLOR E
STEEL DECK
Djemerson Mateus de Andrade, Geraldo Donizetti de Paula and Rovadávia Aline de Jesus Ribas 139
3.33 NUMERICAL STUDY OF BALLAST-FLYING PROCESS CAUSED BY ACCRETING SNOW/ICE
ON HIGH-SPEED TRAINS BY USING THE DISCONTINUOUS DEFORMATION ANALYSIS
(DDA)
Dong Ding, Abdellatif Ouahsine and Peng Du . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
3.34 ON THE DYNAMIC BEHAVIOR OF RODS AND BEAMS WITH NONLINEAR BOUNDARY
STIFFNESS
Douglas Roca Santo, Jean-Mathieu Mencik, Bin Tang, Paulo José Paupitz Gonçalves and Michael
John Brennan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
3.35 EFFECT OF PARTICLE SPIN ON THE SPATIO-THERMAL FIRE HAZARD DISTRIBUTION
OF INCANDESCENT FRAGMENTS
Eduardo M. B. Campello . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
3.36 RELIABILITY ANALYSIS OF A CONCRETE FRAME
Eduardo Naccache and Valério Almeida . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
3.37 PARAMETER ESTIMATION FOR KINETIC MODELING OF 18F-FDG IN POSITRON EMIS-
SION TOMOGRAPHY(PET) IMAGE
Eliete Biasotto Hauser, Evandro Manica, Gianina Teribele Venturin, Samuel Greggio, Eduardo R.
Zimmer, Wyllians Vendramini Borelli and Jaderson Costa Da Costa . . . . . . . . . . . . . . . . 158
3.38 A FIBER OPTIMIZATION METHOD BASED ON NORMAL DISTRIBUTION FUNCTION
FOR HYPERELASTIC MATERIALS
André Luis Ferreira Da Silva, Ruben Andres Salas and Emílio Carlos Nelli Silva . . . . . . . . . 162
3.39 NUMERICAL INVESTIGATION OF STRONG ADDED MASS EFFECT FOR FLUID-STRUCTURE
CALCULATIONS APPLIED TO MOVING HYDROFOILS
Emmanuel Lefrançois . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
3.40 NOVEL METHOD FOR ACOUSTIC FLUID-STRUCTURE INTERACTION IN APPLICATION
TO OVERALL SAFETY OF STRUCTURES IN QUASI-STATIC SETTING
Emina Hadzalic, Adnan Ibrahimbegovic and Samir Dolarevic . . . . . . . . . . . . . . . . . . . 171
3.41 MULTI-SURFACE PLASTICITY MODEL WITH SOFTENING FOR SOLIDS WITH MARKED
DIFFERENCE OF FAILURE
Emir Karavelić and Adnan Ibrahimbegovic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
3.42 COLLAPSE OF PARABOLICALLY TAPERED CHS STEEL COLUMNS: FINITE ELEMENT
MODEL VALIDATION
Tiago Zampaolo, Ernesto Massaroppi Jr., Miguel Abambres and Tiago Ribeiro . . . . . . . . . . 179
3.43 REPEATED RICHARDSON EXTRAPOLATION WITH VERIFICATION OF THE ORDER OF
ACCURACY OF NUMERICAL SCHEMES AND THE FORMS OF APPLYING BOUNDARY
CONDITIONS OF THE FINITE VOLUME METHOD ON THE LAPLACE EQUATION
Fabiana De Fatima Giacomini and Ana Paula Da Silveira Vargas . . . . . . . . . . . . . . . . . 183
3.44 NUMÉRICAL SIMULATION OF THE DESCENT OF CASING PIPES OF OFFSHORE OIL
WELLS
Tiago Sten Freitas, Fabrício Nogueira Corrêa and Breno Pinheiro Jacob . . . . . . . . . . . . . 185
3.45 PARAMETRIC STUDY TO OPTIMIZE THE HYDRODYNAMIC EFFICIENCY OF A SUB-
MERGED PRESSURE DIFFERENTIAL WAVE ENERGY CONVERTER
Raí Quintas, Fabricio Corrêa and Carl Albrecht . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
3.46 3D DUCTILE CRACK SIMULATION BASED ON H-ADAPTIVE METHODOLOGY
Fangtao Yang, Alain Rassineux, Carl Labergere and Khemais Saanouni . . . . . . . . . . . . . . 190

ix
3.47 FLOW OVER A CYLINDER FIXED BY SPRING AND DAMPER SOLVED BY FOURIER
PSEUDOSPECTRAL AND IMMERSED BOUNDARY METHODS
Felipe Pamplona Mariano, Andreia Aoyagui Nascimento and Aristeu Da Silveira Neto . . . . . . 194
3.48 A NUMERICAL STUDY OF FLOW AROUND CIRCULAR AND RECTANGULAR CYLIN-
DERS
Elder Gualberto Alves, Felipe Pamplona Mariano and Andreia Aoyagui Nascimento . . . . . . . 198
3.49 ASSESSMENT OF NON-CONVENTIONAL CONSTRAINTS IMPOSED TO A HYPERELAS-
TIC MODEL
Felipe Tempel Stumpf and Rogério José Marczak . . . . . . . . . . . . . . . . . . . . . . . . . . 200
3.50 SOFTWARE EROSION ON A STRUCTURAL ANALYSIS SOFTWARE – A CASE STUDY
Flavio Torres Da Fonseca, Roque Luiz Da Silva Pitangueira and Samuel Silva Penna . . . . . . . 201
3.51 ENTROPIC CONSIDERATIONS ON THE LBGK MODEL FOR ADVECTION-DIFFUSION
Florian De Vuyst and Thomas Douillet-Grellier . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
3.52 COMPACT THIRD ORDER EXPANSION OF LATTICE BOLTZMANN SCHEMES
François Dubois . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
3.53 ON THE OPTIMIZATION OF POWER GENERATED BY COMBINED CYCLE HEAVY-DUTY
GAS TURBINE
Paulo Guilherme Inca, Gabriel Maidl, Eduardo Massashi Yamao, Renato De Arruda Penteado
Neto, Leandro Coelho and Cintia De Carvalho Toledo . . . . . . . . . . . . . . . . . . . . . . . 210
3.54 THREE-DIMENSIONAL FINITE ELEMENT MODEL TO PREDICT FLEXURAL BEHAVIOR
OF REINFORCED CONCRETE BEAMS
Gabriela Bandeira De Melo Lins De Albuquerque, Valdir Pignatta E Silva and João Paulo Cor-
reia Rodrigues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
3.55 MULTISCALE ANALYSIS OF MULTIPLE CRACKS IN AIRCRAFT FUSELAGE
Thiago Arnaud Abreu De Oliveira, Gilberto Gomes, Francisco Evangelista Junior and Alvaro
Martins Delgado Neto . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
3.56 IDENTIFICATION OF PRINCIPAL RESEARCH TOPICS STUDIED IN ENGINEERING IN
BRAZIL
Jether Gomes, Thiago Dias and Gray Moita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
3.57 THE FUZZY LOGIC TECHNIQUE FOR EMULATION OF THE DECISION-MAKING PRO-
CESS DURING HUMAN MOVEMENT SIMULATION
Henrique Braga, Gray Moita and Paulo E. M. De Almeida . . . . . . . . . . . . . . . . . . . . . 226
3.58 FE2 MULTI-SCALE METHOD APPLIED TO SIMULATE COMPOSITE MATERIALS STRUC-
TURES IN SUPERCOMPUTERS
Guido Giuntoli, Guillaume Houzeaux, Mariano Vázquez and Sergio Oller . . . . . . . . . . . . . 230
3.59 A LAGRANGE MULTIPLIER FORMULATION FOR CONSIDERING MEMBER DEFORMA-
TION CONSTRAINTS INTO MATRIX STRUCTURAL ANALYSIS
Guilherme Barros, Luiz Fernando Martha and Evandro Parente Junior . . . . . . . . . . . . . . 232
3.60 OBJECT ORIENTED FRAMEWORK FOR MULTIPHYSICS ANALYSIS
Guilherme Garcia Botelho, Humberto Alves Da Silveira Monteiro, Rodrigo Guerra Peixoto and
Roque Luiz Da Silva Pitangueira . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
3.61 OVERALL ADAPTIVE MODELING FOR THE STATIC AND DYNAMIC FINITE ELEMENT
ANALYSIS OF STRUCTURES
Guoqiang Wei, Pascal Lardeur and Frédéric Druesne . . . . . . . . . . . . . . . . . . . . . . . . 237
3.62 REPRESENTATION OF A PRODUCTION WELL USING A SERIES OF POINT SINKS IN 2D
BEM SIMULATIONS
Gustavo Gontijo, Éder Albuquerque and Eugênio Fortaleza . . . . . . . . . . . . . . . . . . . . . 241
3.63 DYNAMIC ANALYSIS OF FOUNDATIONS CONSIDERING SOIL-STRUCTURE INTERAC-
TION IN A LAYERED MEDIUM
Helena Fideles, Marco Santos, Sergio Santos and Webe Mansur . . . . . . . . . . . . . . . . . . 245
3.64 SEISMIC ANALYSIS OF WIND TURBINE
Henrique Ataide Nery De Castro Filho, Suzana Moreira Avila and José Luis Vital de Brito . . . . 248
3.65 POST-BUCKLING ANALYSIS OF PLANE FRAME STRUCTURES WITH VARIABLE RIGID-
ITY ELEMENTS
Iara Ribeiro, Francisco Célio de Araújo, Kátia Silva and Tatiane Maga Mendes . . . . . . . . . . 251
3.66 UM ESTUDO EM VIBRAÇÕES LIVRES ACOPLADAS BARRAGEM-RESERVATÓRIO PELO
MÉTODO DAS DIFERENÇAS FINITAS E ELEMENTOS FINITOS
Iarly Vanderlei Da Silveira, Mauricio Mendes Vitalli and Lineu José Pedroso . . . . . . . . . . . 255

x
3.67 TOPOLOGICAL OPTIMIZATION APPLIED IN FLEXIBLE ROBOTIC ARMS – BY SESO
Adriano Ribeiro Marinho, Hélio Luiz Simonetti, Valério S. Almeida, Isabel Jesus and Ramiro
Barbosa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
3.68 EXPERIMENTAL TESTING AND NUMERICAL RESEARCH ON STRUCTURAL CONNEC-
TION BEHAVIOR UNDER CYCLIC LOAD
Ismar Imamovic, Adnan Ibrahimbegovic and Esad Mesic . . . . . . . . . . . . . . . . . . . . . . 264
3.69 MULTISCALE COMPUTATION PROCEDURE FOR LOCALIZED FAILURE: ED-FEM VER-
SUS FE2 MICRO-MACRO REPRESENTATIONS
Ivan Rukavina and Adnan Ibrahimbegovic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
3.70 ON THE ANALYSIS OF AN INNOVATIVE ARCHITECTURAL PROJECT OF THE XIII CEN-
TURY CONSIDERING MODERN NUMERICAL TECHNIQUES: THE SPIRE OF THE SEN-
LIS CATHEDRAL
Eduard Antaluca, Raphael Rolin, Jean-Louis Batoz and Fabien Lamarque . . . . . . . . . . . . . 272
3.71 A NODAL INTEGRATION METHOD TO SIMULATE WELDING AND ROLLING PROCESSES
IN THE REFERENCE FRAME RELATED TO THE SOLICITATIONS
Yabo Jia, Jean-Christophe Roux, Eric Feulvarch, Jean-Baptiste Leblond and Jean-Michel Bergheau276
3.72 MODELLING OF INCOMPRESSIBLE FLOWS WITH MOVING BOUNDARIES EMPLOY-
ING THE ARLEQUIN FRAMEWORK
Jeferson Fernandes and Rodolfo Sanches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
3.73 A POSITIONAL FEM FORMULATION FOR CONTACT ANALYSIS OF 2D ELASTOPLAS-
TIC SOLIDS
Péricles Rafael Pavão Carvalho, Humberto Breves Coda, Rodolfo André Kuche Sanches and Je-
ferson Fernandes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
3.74 NUMERICAL SIMULATION OF SINGLE-PHASE TWO-COMPONENT FLOW IN NATURALLY
FRACTURED OIL RESERVOIRS
Joao Gabriel Souza Debossam, Juan Diego Dos Santos Heringer, Grazione de Souza and Helio
Pedro Amaral Souto . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
3.75 EFFICIENT LBM ON GPUS FOR DENSE MOVING OBJECTS USING IMMERSED BOUND-
ARY CONDITION
Joel Beny and Jonas Latt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
3.76 SIMULAÇÃO TRANSIENTE DE ESCOAMENTOS LAMINARES UTILIZANDO O MÉTODO
SPH
Almério Pamplona, Flávia Cardoso, Fabiano Dos Santos and Joel Vasco . . . . . . . . . . . . . . 295
3.77 CONTINUOUS MONITORING OF THE DYNAMIC BEHAVIOR OF A HIGH-RISE TELECOM-
MUNICATIONS TOWER
Jorge Leite, Diogo Ribeiro, Nuno Pinto and Rui Calçada . . . . . . . . . . . . . . . . . . . . . . 297
3.78 SIMULATING CRACK FRICTION INDUCED HYSTERESIS IN CERAMIC MATRIX COM-
POSITES (CMC)
Jorge Nunez Ramirez, Antoine Hurman and Fréderic Laurin . . . . . . . . . . . . . . . . . . . . 298
3.79 HIGH PERFORMANCE COMPUTING FOR ASSESSMENT OF THE INFLUENCE OF TOPO-
GRAPHIC FACTOR IN SEISMIC AMPLIFICATION IN OIL FIELDS
José Alves, Alvaro Coutinho, Renato Elias, Carlos Silva, Leandro Gazoni, Marcus Casagrande
and Ricardo Borges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
3.80 DYNAMIC ANALYSIS OF BUILDINGS CONSIDERING THE EFFECT OF MASONRY IN-
FILLS
José Guilherme Santos Da Silva, Leonardo Souza Bastos and Carolina Sanchez Guerrero . . . . 306
3.81 HUMAN COMFORT ASSESSMENT OF BUILDINGS SUBJECTED TO NONDETERMINIS-
TIC WIND LOADINGS
José Guilherme Santos Da Silva, Alan Barile and Leonardo Souza Bastos . . . . . . . . . . . . . 310
3.82 NUMERICAL SIMULATION OF NON-ISOTHERMAL FLOW IN OIL RESERVOIRS USING
A TWO-EQUATION MODEL
Juan Diego Dos Santos Heringer, João Gabriel de Souza Debossam, Grazione de Souza and
Helio Pedro Amaral Souto . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
3.83 APPLICATION OF SUPPORT VECTOR MACHINE AND FINITE ELEMENT METHOD IN
THE PREDICTIONS OF MECHANICAL PROPERTIES OF LIGHTWEIGHT AGGREGATE
CONCRETE
Aldemon Bonifacio, Julia Mendes, Michele Farage, Flavio Barbosa, Ciro Barbosa and Anne-Lise
Beaucour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 318

xi
3.84 INFLUENCE OF LIGHTWEIGHT AGGREGATE’S SHAPES ON THE NUMERICAL EVALU-
ATION OF LIGHTWEIGHT AGGREGATE CONCRETE
Aldemon Bonifacio, Julia Mendes, Michele Farage, Flavio Barbosa, Ciro Barbosa and Anne-Lise
Beaucour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
3.85 DAMAGE DETECTION IN BEAMS USING AN ADDITIONAL MOBILE MASS: A NUMER-
ICAL VALIDATION
Juliana Santos, Marcus Morais, Erwin Palechor, Marcela Machado, Ramon Silva and Luciano
Bezerra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
3.86 INTEROPERABILITY BETWEEN STRUCTURAL DESIGN SOFTWARE WITH BIM PLAT-
FORM
Kaio Ricardo Da Silva, Evangelos Dimitrios Christakou and Lenildo Santos Da Silva . . . . . . . 324
3.87 A NEW APPROACH TO CREATE PROFITABLE CUSTOMER PROFILES USING MACHINE
LEARNING AND RFMP MODEL
Leandro Da Silva Carvalho and Nelson Francisco Favilla Ebecken . . . . . . . . . . . . . . . . 327
3.88 3D SEISMIC MODELING WITH OCTREE MESHES
André Valente, Leandro Di Bartolo and Webe Mansur . . . . . . . . . . . . . . . . . . . . . . . 338
3.89 DEEP NEURAL NETWORK FOR VECTOR FIELD TOPOLOGY RECOGNITION WITH AP-
PLICATIONS TO FLUID FLOW SUMMARIZATION
Eliaquim Ramos, Leandro Tavares Da Silva, Gilson Giraldi and Jaime Cardoso . . . . . . . . . 342
3.90 ESTUDO EM OTIMIZAÇÃO DE TERÇAS PARA COBERTURAS DE GALPÕES LEVES
Kamila Nascimento, Marco Bessa and Leonardo Inojosa . . . . . . . . . . . . . . . . . . . . . . 346
3.91 ANÁLISE ESTRUTURAL DA BIBLIOTECA DA FACULDADE DE MEDICINA DO DISTRITO
FEDERAL
Marco Bessa, Leonardo Inojosa, José Humberto De Paula, Ronaldo Almeida and Marcio Buzar . 350
3.92 MODULAR MODELLING APPROACH FOR FDM PRINTED STRUCTURES AND PIEZO
DISKS FOR METAMATERIAL DESIGN
Gabriel Konda Rodrigues, Maíra Martins Da Silva and Leopoldo Pisanelli Rodrigues de Oliveira 354
3.93 MEMBRANE-TYPE ACOUSTIC METAMATERIALS: A MODULAR PERSPECTIVE
Lucas Yudi Moriya Sampaio and Leopoldo Pisanelli Rodrigues de Oliveira . . . . . . . . . . . . 368
3.94 EVALUATION OF THE RESISTANCE OF TRUSSED SLABS WITH STEEL FORMWORK IN
COLD FORMED U PROFILE
Lucas Favarato, Adenilcia Calenzani, Juliana Pires, Elisabeth Junges and Johann Ferrareto . . . 379
3.95 TOWARDS MICRO-MECHANICAL CHARACTERIZATION OF PLASTIC ANISOTROPY IN
HEXAGONAL MATERIALS USING SHAPE MANIFOLD LEARNING
Balaji Raghavan, Ludovic Cauvin, Salima Bouvier, Piotr Breitkopf and Fodil Meraghni . . . . . . 382
3.96 ESTUDO NUMÉRICO DO COMPORTAMENTO DINÂMICO DE UMA TORRE EÓLICA VIA
CFD
Adriana Silva, Lineu Pedroso and Luiz Oliveira . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
3.97 APPLICATION OF A MULTIOBJETIVE OPTIMIZATION PARETO APPROACH TO DESIGN
THE SDRE CONTROLLER FOR A RIGID-FLEXIBLE SATELLITE
Luiz Souza . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
3.98 RISK ANALYSIS OF REINFORCED CONCRETE STRUCTURES SUBJECTED TO INDUCED
SEISMICITY USING LUMPED DAMAGE MECHANICS MODELING
Andréia H. A. Silva, Gonzalo L. Pita, Julio Flórez-Lopez, Gustavo H. Siqueira, Thiago D. Santos,
Karolinne O. Coelho, Ricardo A. Picón and Luiz C. M. Vieira Jr. . . . . . . . . . . . . . . . . . 395
3.99 IDENTIFYING ELASTIC PROPERTIES OF GEOMECHANICAL RESERVOIR MODELS US-
ING METAHEURISTIC OPTIMIZATION TECHNIQUES
Marcello Congro, Rafael Abreu and Deane Roehl . . . . . . . . . . . . . . . . . . . . . . . . . . 396
3.100COMPARATIVE STUDY IN RESULTS AND COMPUTATIONAL PERFORMANCE OF THE
MAIN METHODS OF SOLUTION OF SPARSE AND SYMMETRIC LINEAR SYSTEMS THAT
RESULT FROM FINITE ELEMENT METHOD
Calebe Souza and Maria Clara Rameiro . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
3.101NUMERICAL ANALYSIS OF ALUMINUM PLANE PLATES WITH TWO COPLANAR HOLE-
EDGE CRACKS UTILIZING DUAL BOUNDARY ELEMENTS METHOD.
Matheus Lisboa and Carla Anflor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
3.102EFFECT OF THE MUDDY AREA ON THE SURFACE WAVE ATTENUATION AND THE
SHIP’S SQUAT
Mohamed Ali, Sami Kaidi and Emmanuel Lefrançois . . . . . . . . . . . . . . . . . . . . . . . . 405

xii
3.103FLUID-STRUCTURE INTERACTION MODELING OF A PEDIATRIC VENTRICULAR AS-
SIST DEVICE
Mohammad Malekan, Simão Bach and Idagene A. Cestari . . . . . . . . . . . . . . . . . . . . . 409
3.104IMPLEMENTATION OF EXTENDED FINITE ELEMENT FOR WEAK DISCONTINUITIES
USING PYTHON
Nasser Alkmim and Francisco Evangelista Junior . . . . . . . . . . . . . . . . . . . . . . . . . . 413
3.105COMPARATIVE ANALYSIS OF TURBULENCE MODELS IN HYDRAULIC JUMPS
David Oliveira Fonseca, Raquel Jahara Lobosco, Graziela Maria Faquim Jannuzzi and Necesio
Gomes Costa . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417
3.106SYNCHRONIZATION IN ECOLOGICAL NETWORKS
Gilberto Pereira and Nelson Ebecken . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
3.107SHAPED CHARGE OPTIMIZATION THROUGH EVOLUTIONARY METHODS
Alexandre de Assis Motta and Nelson Francisco Favilla Ebecken . . . . . . . . . . . . . . . . . 424
3.108STOCHASTIC APPROXIMATION TO HETEROGENEOUS DYNAMIC SYSTEMS
Nikolaos Limnios and Adnan Ibrahimbegovic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
3.109NUMERICAL SIMULATION OF A HISTORICAL MILITARY BUILDING PLACED IN MEX-
ICO CITY, SUBJECTED TO EARTHQUAKE AND SOIL SUBSIDENCE EFFECTS
Norberto Domínguez, Arturo Suarez and Juan-Edmundo Mayorga . . . . . . . . . . . . . . . . . 433
3.110ELECTROMAGNETIC MODELING USING NEDELEC ELEMENTS OF HIGH-ORDER AND
HPC
Octavio Castillo Reyes, Josep de La Puente and José María Cela . . . . . . . . . . . . . . . . . . 437
3.111ELECTROSTATIC AND MAGNETOSTATIC DUAL FORMULATIONS AND FINITE ELE-
MENT IMPLEMENTATION BASED ON DIFFERENTIAL FORMS
Pablo Moreno-Navarro, Adnan Ibrahimbegovic and Alejandro Ospina . . . . . . . . . . . . . . . 438
3.112IMPLEMENTATION OF NUMERICAL SOLUTIONS FOR BOND-SLIP EFFECT APPLIED IN
REINFORCED CONCRETE BEAMS CONSIDERING THE VARIABLE BOND MODEL
João Xavier, Robson Pereira, Lineu Pedroso and Paulo Martins . . . . . . . . . . . . . . . . . . 442
3.113AVALIAÇÃO DE MÉTODOS SIMPLIFICADOS E REFINADOS UTILIZADOS PARA A DE-
TERMINAÇÃO DAS FLECHAS IMEDIATAS EM ELEMENTOS FLETIDOS DE CONCRETO
(COERÊNCIA COM DADOS EXPERIMENTAIS)
Robson Pereira, João Paulo Xavier, Paulo Martins and Lineu Pedroso . . . . . . . . . . . . . . . 446
3.114BOUNDARY CONDITIONS IN REGULARIZED LATTICE-BOLTZMANN
Luiz Hegele, Keijo Mattila, Jonas Hegele and Paulo Philippi . . . . . . . . . . . . . . . . . . . . 450
3.115APPLICATION OF HIGH ORDER FLUX RECONSTRUCTION/CORRECTION PRO-CEDURE
VIA RECONSTRUCTION (FR/CPR) METHOD COUPLED TO A NON-ORTHODOX MUL-
TIPOINT FLUX APPROXIMATION METHOD (MPFA-D) IN THE SIMULATION OF TWO-
PHASE FLOWS IN PETROLEUM RESERVOIRS
Gustavo Ramirez, Fernando Contreras, Darlan Carvalho and Paulo Lyra . . . . . . . . . . . . . 452
3.116AN INTERACTIVE-GRAPHICS TOOL FOR MODELING AND ANALYZING STRUCTURES
IN MATLAB
Pedro Lopes, Rafael Rangel and Luiz Fernando Martha . . . . . . . . . . . . . . . . . . . . . . . 454
3.117GEOMETRICALLY EXACT TIMOSHENKO ROD THEORY WITH INCREMENTAL RODRIGUES
PARAMETERS
Pedro Virgolino, Cátia Costa E. Silva and Paulo de Mattos Pimenta . . . . . . . . . . . . . . . . 458
3.118SIMULATION OF SHIP MOTIONS IN STATIC AND DYNAMIC STATES
Peng Du and Abdellatif Ouahsine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
3.119NUMERICAL SIMULATION OF DISCRETE FRACTURE NETWORKS IN PETROLEUM RESER-
VOIR SIMULATION
Chensong Zhang and Philippe R. B. Devloo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
3.120A RANDOM SET BASED IDENTIFICATION STRATEGY
Pierre Feissel, Liqi Sui and Thierry Denoeux . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466
3.121MODEL ORDER REDUCTION IN STRUCTURAL RAPID DYNAMICS WITH APPLICATION
TO CRASHWORTINESS DESIGN
Pierre Phalippou, Salim Bouabdallah, Piotr Breitkopf and Pierre Villon . . . . . . . . . . . . . . 470
3.122EDUCATIONAL TOOL FOR THE ANALYSIS OF STRUCTURES WITH GEOMETRIC NON-
LINEARITY
Rafael Rangel and Luiz Fernando Martha . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
3.123INFLUENCE OF MATERIAL HETEROGENEITY ON THE STABILITY OF EXPLICIT HIGH-
ORDER SPECTRAL ELEMENT METHODS
Régis Cottereau and Ruben Sevilla . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474

xiii
3.124EVALUATION OF LOW-COST MEMS ACCELEROMETERS PERFORMANCE FOR STRUC-
TURAL HEALTH MONITORING MODAL ANALYSIS
Renan Rocha Ribeiro and Rodrigo Lameiras . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
3.125A NONLINEAR DYNAMIC MODEL OF A NON-IDEAL MOTOR SUPPORT STRUCTURE
Reyolando Brasil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
3.126UNSUPERVISED REAL-TIME SHM TECHNIQUE FOR NOVELTY DETECTION BASED ON
SINGLE-VALUED FEATURES
Rharã Cardoso, Alexandre Cury and Flávio Barbosa . . . . . . . . . . . . . . . . . . . . . . . . 484
3.127INFLUENCE OF PARTIAL SHEAR CONNECTION OF STEEL-CONCRETE COMPOSITE
BEAM IN THE MULTI-STOREY BUILDINGS ANALYSIS
Ígor J.M. Lemes, Luís E.S. Dias, Ricardo A.M. Silveira, Amilton R. Silva and Andréa R.D. Silva . 488
3.128ADVANCED INELASTIC ANALYSIS OF STEEL ARCHES WITH TUBULAR CROSS SEC-
TIONS
Ricardo Silveira, Lidiane de Deus, Jéssica Silva and Ígor Lemes . . . . . . . . . . . . . . . . . . 491
3.129UNCERTAINTY EVALUATION IN EULER-BERNOULLI AND TIMOSHENKO BENDING
STATICS PROBLEMS
Roberto Squarcio and Claudio Roberto Ávila . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
3.130VIBRATION CONTROL BY IMPLICIT DISSIPATIVE SCHEME IN NONLINEAR DYNAM-
ICS 2D BEAM STRUCTURES
Rosa Adela Mejia Nava, Adnan Ibrahimbegovic and Rogelio Lozano-Leal . . . . . . . . . . . . . 499
3.131NUMERICAL ANALYSIS OF THE MECHANICAL BEHAVIOR OF WELDED CONNECTIONS
BETWEEN SQUARE HOLLOW CORE SECTION COLUMN AND I-BEAM
Rosicley Rosa and Juliano Neto . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 503
3.132AN OUTLINE OF SATS – A SYSTEM FOR THE ANALYSIS OF TAUT STRUCTURES
Ruy Marcelo Pauletti . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
3.133NUMERICAL ANALYSIS OF 3D MICROPOLAR BOUNDARY VALUE PROBLEMS
Sara Grbčić, Adnan Ibrahimbegovic and Gordan Jelenic . . . . . . . . . . . . . . . . . . . . . . 508
3.134A GEOMETRICALLY EXACT EULER-BERNOULLI BEAM FORMULATION FOR NONLIN-
EAR 3D MATERIAL LAWS
Sascha Maassen, Paulo de Mattos Pimenta and Jörg Schröder . . . . . . . . . . . . . . . . . . . 512
3.135PORE-SCALE SIMULATION OF THE WETTABILITY INFLUENCE ON THE DISPLACE-
MENT OF IMMISCIBLE FLUIDS USING A LATTICE-BOLTZMANN METHOD
Ricardo Bazarin, Christian Naaktgeboren and Silvio Junqueira . . . . . . . . . . . . . . . . . . 516
3.136PROBABILISTIC ANALYSIS OF FAILURE OF FIBER REINFORCED CONCRETE
Simona Dobrilla, Noémi Friedman, Tea Rukavina, Hermann Matthies and Adnan Ibrahimbegovic 526
3.137ANISOTROPIC HIGH-ORDER ADAPTIVE BOUNDARY ELEMENT METHODS FOR 3D ACOUS-
TIC WAVE PROPAGATION
Faisal Amlani, Stéphanie Chaillat and Adrien Loseille . . . . . . . . . . . . . . . . . . . . . . . 530
3.138FORECAST DEMAND ON THE QUEBEC POWER GRID: CHALLENGES AHEAD
Stéphane Dellacherie, Olivier Milon, Cédric Poutré and Arnaud Zinflou . . . . . . . . . . . . . . 533
3.139MODELING COMPLETE FIBER PULL-OUT IN FIBER-REINFORCED COMPOSITES
Tea Rukavina, Adnan Ibrahimbegovic and Ivica Kozar . . . . . . . . . . . . . . . . . . . . . . . 534
3.140DEVELOPMENT OF A ROBOTIC PROSTHESIS CONTROLLED BY MYOELECTRIC SEN-
SORS
Kesley Silva, Matheus Delgado, Adriano Drumond and Thiago Dias . . . . . . . . . . . . . . . . 538
3.141A MODEL TO SOLVE ROTATION PROBLEMS BETWEEN SHELLS
Thiago Fernandes, Paulo de Mattos Pimenta and Cátia Silva . . . . . . . . . . . . . . . . . . . . 542
3.142EVALUATION OF CRACKED FUSELAGE SUBJECT TO CONTACT MECHANICS
Thiago Arnaud Abreu De Oliveira, Gilberto Gomes, Francisco Evangelista Junior and Alvaro
Martins Delgado Neto . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
3.143MODELLING OF ACTUATED 3D MECHANISMS BY A DYNAMIC GEOMETRIC NONLIN-
EAR FEM FORMULATION WITH APPLICATIONS IN AEROSPACE STRUCTURES
Tiago Siqueira and Humberto Coda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 550
3.144COMPARATIVE ANALYSIS OF THE PROGRESSIVE COLLAPSE IN A MULTISTOREY STEEL
BUILDING
Mariana Souza and Tulio Melo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554
3.145STRUT AND TIE ANALYSIS USING TOPOLOGICAL OPTIMIZATION WITH 3D TRUSS
FINITE ELEMENTS
Valério S. Almeida, Hélio L Simonetti and Edson B. Ferreira . . . . . . . . . . . . . . . . . . . . 556

xiv
3.146IMMERSED BOUNDARY – LATTICE BOLTZMANN METHOD FOR 2D PARTICLE SEDI-
MENTATION IN POWER-LAW FLUIDS
Vanessa Glück Nardi and Admilson T. Franco . . . . . . . . . . . . . . . . . . . . . . . . . . . . 560
3.147ANALYSIS OF INTERACTING FLUID-SOLID COUPLED MODELS CONSIDERING AN EF-
FICIENT ADAPTIVE TIME DOMAIN FORMULATION
Victor Bicalho Civinelli de Almeida and Delfim Soares Jr. . . . . . . . . . . . . . . . . . . . . . 564
3.148A SIMPLE ALTERNATIVE FOR HIGHER ORDER FINITE-ELEMENT SOLUTION OF BOUND-
ARY VALUE PROBLEMS POSED IN CURVED DOMAINS
Vitoriano Ruas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 581
3.149LOCAL MESHLESS METHOD WITH REDUCED INTEGRATION FOR HIGH IRREGULAR-
ITY OF THE NODAL ARRANGEMENT
Wilber Vélez, Thiago Araujo, Tiago Oliveira and Artur Portela . . . . . . . . . . . . . . . . . . . 585
3.150CRACK PROPAGATION IN DYNAMICS BY EMBEDDED STRONG DISCONTINUITY AP-
PROACH AND ENERGY BALANCE IN SIMULATIONS
Xuan Nam Do and Adnan Ibrahimbegovic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 588
3.151MOVING IMMERSED BOUNDARY METHOD FOR FSI PROBLEMS
Shang Gui Cai, Abdellatif Ouahsine and Yannick Hoarau . . . . . . . . . . . . . . . . . . . . . 592
3.152A PRAGMATIC STRATEGY FOR THE NUMERICAL SIMULATION OF THE FLUID FLOW
DURING WELDING PROCESSES
Yassine Saadlaoui, Eric Feulvarch, Alexandre Delache, Jean-Baptiste Leblond and Jean-Michel
Bergheau . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594
3.153PARALLEL LBM-DEM SIMULATION OF FLUID FLOW EROSION USING GPU
Zeyd Benseghier, Pablo Cuéllar, Li-Hua Luu, Stéphane Bonelli and Pierre Philippe . . . . . . . . 600

xv
1. PLENARY LECTURES

1
CILAMCE 2018 C ONGRESS
12-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A SPECTS OF M ULTI -S CALE C OUPLING OF


I NELASTIC P ROCESSES IN S OLID M ECHANICS∗
Hermann G. Matthies1 , Adnan Ibrahimbegović2
1 Institute of Scientific Computing, Technische Universität Braunschweig
38092 Braunschweig, Germany; e-mail: wire@tu-bs.de
2 Université de Technologie de Compiègne/Sorbonne Universités

Laboratoire Roberval de Mécanique, Centre de Recherche Royallieu


60200 Compiègne, France; e-mail: adnan.ibrahimbegovic@utc.fr

The realistic representation of the mechanical behaviour of heterogeneous materials at large


scales is a considerable challenge. Such materials can be represented at different scales, depend-
ing upon the objectives and the physical mechanisms that are important to account for. But
often one needs computations on scales which are far larger than the significant heterogeneities.
With limited computational resources for typical engineering applications, most frequently
the analysis has to be performed at the structure scale or macro-scale, where the description
of the heterogeneity can not be resolved and is uncertain having only a statistical description.
If only elastic deformations are considered, at this scale cement-based materials—in contrast
to most situations involving geological materials—can be considered as homogeneous, and their
properties can be obtained by homogenisation. Those models are well known for their robustness
and lead to relatively moderate computational cost. Similar things can be said about mildly
non-linear and irreversible behaviour, like incipient plasticity and damage mechanics. Here
one would like to have a phenomenological model of some simpler type, which for this range
of deformations captures the main features of the micro-structure, e.g. has similar amount of
stored and dissipated energies.
Such models are based on a set of ‘material parameters’ which need to be identified, by
minimising some kind of error measure between predicted and observed response, mainly from
experiments providing unique load paths and boundary conditions. In simple idealised situations
this homogenisation can be performed analytically.
However, such homogenisation approaches only, if at all, capture the average behaviour of the
material. They do not take into account the inherent uncertainties attached to heterogeneous
materials and structures. Here we will introduce a method to take account of the uncertainties
in in the heterogeneous materials, the meso-scale, and will also be able to reflect the effect of this
uncertainty on the macro-scale. Normally, homogenisation assumes that the small heterogeneous
scales are infinitesimally small compared to the macro-scale, and that only the mean response
has to be considered, and all response variability due to the variation on small- or meso-scales
has been averaged away. There assumptions are not always valid. One is when the scales are not
well separated, and geological materials and concrete are two examples where there is variability
typically over all scales of interest. The other situation occurs when variations on the meso-scale
may be the cause for large scale effects, like cracks and local material failures, which have severe
consequences on the macro-scale. The well-known size effect is such a feature [1], and we want
numerical methods which are able to capture this. Furthermore we want to be able to allow
Partly supported by the Deutsche Forschungsgemeinschaft (DFG) and

L’Agence Nationale de la Recherche (ANR)

3 1
models of different kind to be used on different scales, and be able to bind them together. This
will continue work which was reported on previously at different stages of development [5, 9].
Additionally, we want to have efficient computational procedures, and hence the ability to switch
the computation on different scales as need arises, one could think of this as a form of ‘zoom’
capability [3] for regions where interesting localised phenomena take place.
The variability at all scales which we are thinking of has a geometrical aspect, considering
the arrangement (positions, shapes) of the different phases or materials. The other variability
is in the actual properties of the different phases or materials which make up the heterogeneous
body. Examples of such models, which are actually discrete at the meso-scale, can be found in
[6, 2, 12, 4].
At the macro-scale,we are interested in a continuum model, which exhibits the salient fea-
tures of the meso-scale heterogeneous material. As the macro-scale can not represent the true
variability on the meso-scale, which is typically only known statistically anyway, and hence un-
certain, it is proposed to model this additional uncertainty, which comes from the upscaling and
possible model errors, also by a stochastic model. So we choose a stochastic macro-scale con-
tinuum model, where the unresolved variability of the meso-scale is modelled by the stochastic
content. As the salient features of the heterogeneous material we consider here its volume aver-
aged free energy, and the volume averaged dissipation. We shall restrict ourselves here to purely
isothermal processes for the sake of simplicity.
We consider two different stages resp. resolutions of the computation, which can be used
in conjunction in different parts of the computational domain. A true two-scale computation
in the areas where the macro-scale continuum model alluded to above is not possible do give
a response which is accurate enough. Here the interaction between macro- and meso-scale is
considered so strong that the meso-scale resolution is considered necessary. This may happen
in regions where severe irreversible material processes occur, such that the deformation path of
the macro-scale, which is imposed onto the meso-scale, interacts with the meso-scale response
in some profound way. To capture also the size effect, the so-called MIEL-technique is employed
[7, 9], which has a mesh in a frame of macro elements, see also [3, 4]. As the meso-scale is
heterogeneous and uncertain, this meso-scale model is by default a probabilistic model.
To lower the computational cost, the multi-scale approach just mentioned may be in certain
areas of the computation replaced by the macro-scale continuum model alluded to above. This
is a model which is only defined on the macro scale, and has to be pre-computed. In this
procedure, which may be termed an offline procedure, the results obtained at the fine scale are
used to identify and define the probabilistic variation of the phenomenological model parameters
used at the macro-scale. Both the macro- and meso-scale models are probabilistic models, and
hence probabilistic methods may be used to transfer information between the scales. By using
the fast computations possible with functional representations of stochastic properties, this can
be done with Bayesian methods [13, 11, 10, 8].
The macro-scale model is made such that it is capable of representing the salient features of
the meso-scale behaviour, such as elastic deformation combined with e.g. ductile, quasi-brittle,
or brittle behaviour [6, 2, 12, 4]. The variables which bind the models at the different scales
together are as input the overall deformation, and as output, which has to be matched, one can
take, as already mentioned, the volume averaged free energy density and the entropy production
density. Our latest work [15, 14, 16] demonstrates that it is possible to use these variables for
parameter identification. Finally, what is needed is a procedure to determine when to switch
between these different representations of the heterogeneous meso-scale. This will be based
on an error estimator, which is capable of estimating the error between the continuum model
and the multi-scale representation of the heterogeneous material, such that the computationally
expensive multi-scale model is only used where necessary in material ‘zoom’ fashion [3, 9].

References
[1] J.-B. Colliat, M. Hautefeuille, A. Ibrahimbegović, and H. G. Matthies, Stochastic approach to quasi-
brittle failure and size effect, Comptes Rend. Académie Science: Mech. (CRAS) 335 (2007), 430–435.

4 2
[2] X. N. Do, A. Ibrahimbegović, and D. Brancherie, Dynamics framework for 2D anisotropic continuum-
discrete damage model for progressive localized failure of massive structures, Computers and Struc-
tures 183 (2017), 14–26, doi:10.1016/j.compstruc.2017.01.011.
[3] M. Hautefeuille, J.-B. Colliat, A. Ibrahimbegović, and H. G. Matthies, Multiscale zoom capabil-
ities for damage assessment in structures, Damage Assessment and Reconstruction after Natural
Desasters and Previous Military Activities (A. Ibrahimbegović and M. Zlatar, eds.), NATO-ARW
series, Springer, 2008.
[4] M. Hautefeuille, J.-B. Colliat, A. Ibrahimbegović, H. G. Matthies, and P. Villon, A multi-scale
approach to model localized failure with softening, Computers and Structures 94–95 (2012), 83–95,
doi:10.1016/j.compstruc.2011.11.007.
[5] A. Ibrahimbegovic and H. G. Matthies, Probabilistic multiscale analysis of inelastic localized failure
in solid mechanics, Computer Assisted Methods in Engineering and Science 19 (2012), 277–304,
Available from: http://cames.ippt.gov.pl/pdf/CAMES_19_3_5.pdf.
[6] E. Karavelić, M. Nikolić, A. Ibrahimbegović, and A. Kurtović, Concrete meso-scale model with full
set of 3D failure modes with random distribution of aggregate and cement phase. Part I: Formulation
and numerical implementation, Computer Methods in Applied Mechanics and Engineering in Press
(2017), doi:10.1016/j.cma.2017.09.013.
[7] D. Markovič, R. Niekamp, A. Ibrahimbegović, H. G. Matthies, and R. L. Taylor, Multi-scale modeling
of heterogeneous structures with inelastic constitutive behavior: Part I — physical and mathematical
aspects, Engineering Computations 22 (2005), 664–683, doi:10.1108/02644400510603050.
[8] H. G. Matthies, Uncertainty quantification and Bayesian inversion, Encyclopaedia of Computational
Mechanics (E. Stein, R. de Borst, and T. J. R. Hughes, eds.), vol. 1, John Wiley & Sons, Chichester,
2nd ed., 2017, Part 1. Fundamentals. Encyclopaedia of Computational Mechanics, doi:10.1002/
9781119176817.ecm2071.
[9] H. G. Matthies and A. Ibrahimbegović, Stochastic multiscale coupling of inelastic processes in
solid mechanic, Multiscale Modelling and Uncertainty Quantification of Materials and Struc-
tures (M. Papadrakakis and G. Stefanou, eds.), 135–157, vol. 3, Springer, Berlin, 2014, doi:
10.1007/978-3-319-06331-7_9.
[10] H. G. Matthies, E. Zander, B. V. Rosić, and A. Litvinenko, Parameter estimation via conditional
expectation: a Bayesian inversion., Advanced Modeling and Simulation in Engineering Sciences 3
(2016), 24, doi:10.1186/s40323-016-0075-7. MR 2202913 (2006h:74023)
[11] H. G. Matthies, E. Zander, B. V. Rosić, A. Litvinenko, and O. Pajonk, Inverse problems in a
Bayesian setting, Computational Methods for Solids and Fluids — Multiscale Analysis, Probab-
ility Aspects and Model Reduction (A. Ibrahimbegović, ed.), Computational Methods in Applied
Sciences, vol. 41, Springer, Berlin, 2016, pp. 245–286, doi:10.1007/978-3-319-27996-1_10.
[12] M. Nikolić and A. Ibrahimbegović, Rock mechanics model capable of representing initial heterogeneit-
ies and full set of 3D failure mechanisms, Computer Methods in Applied Mechanics and Engineering
290 (2015), 209–227, doi:10.1016/j.cma.2015.02.024.
[13] B. V. Rosić, A. Kučerová, J. Sýkora, O. Pajonk, A. Litvinenko, and H. G. Matthies, Parameter
identification in a probabilistic setting, Engineering Structures 50 (2013), 179–196, doi:10.1016/j.
engstruct.2012.12.029.
[14] B. V. Rosić, M. S. Sarfaraz, H. G. Matthies, and A. Ibrahimbegović, Stochastic upscaling of random
microstructures, Proceedings in Applied Mathematics and Mechanics 17 (2017), 869–870, doi:
10.1002/pamm.201710401.
[15] M. S. Sarfaraz, B. V. Rosić, and H. G. Matthies, Stochastic upscaling of heterogeneous materi-
als, Proceedings in Applied Mathematics and Mechanics 16 (2016), 679–680, doi:10.1002/pamm.
201610328.
[16] M. S. Sarfaraz, B. V. Rosić, H. G. Matthies, and A. Ibrahimbegović, Stochastic upscaling via linear
Bayesian updating, Multiscale Modeling of Heterogeneous Structures (J. Sorić, P. Wriggers, and
O. Allix, eds.), Lecture Notes in Applied and Computational Mechanics, vol. 86, Springer, 2018,
pp. 163–181, doi:10.1007/978-3-319-65463-8_9.

5 3
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

NUMERICAL MODELING OF FRACTURE IN HIGHLY HETEROGENEOUS MATERIALS


Julien Yvonnet1
1
Université Paris-Est, Laboratoire de Modélisation et Simulation Multi Echelle, UMR 8208 CNRS,
5 Boulevard Descartes, 77454 Marne-la-Vallée Cedex 2, France, julien.yvonnet@univ-paris-est.fr

In recent years, the fast development of 3D imaging techniques such as micro tomography
combined with in-situ testing have offered new possibilities to investigate the microstructural damage
evolution due to microcracking of highly complex materials such as concrete [1] or regular lattices
obtained by additive manufacturing (3D printing). These new experimental and manufacturing
techniques open tough challenges for the modeling of damage in heterogeneous quasi brittle materials:
simulate the initiation and propagation of complex 3D micro crack networks in realistic geometries
arising from micro tomography, identify the microstructural damage models, construct damage models
at the scale of homogeneous materials, or even optimizing the local topology of constituents to
maximize the resistance of the material to fracture.
In a first part [2-6], we present recent results of microcracking numerical models in realistic
concrete microstructures obtained by combining micro tomography, 3D imagery, in-situ testing, and
the numerical phase field method to fracture [7,8]. We show that this method combines several
advantages in this context: (a) the possibility to simulate the initiation and propagation of complex
micro crack networks in realistic geometries of microstructures; (b) no mesh dependence, allowing
using directly regular meshes obtained from segmented images; (c) a very small number of parameters
to be identified; (d) a high robustness and efficiency of algorithms. We present the first, to our best
knowledge, direct comparisons between 3D complex evolving crack networks obtained from
experiments and simulations at both micro and macro scales in plaster and concrete samples [3]. An
identification of microstructural damage parameters based on inverse approaches combining in-situ
testing, 3D imaging, 3D image correlation and the numerical phase field method for fracture is
presented.
In a second part [9,10], we describe an approach to design complex 2-phase lattice materials
which can be obtained by recent 3D printing techniques to maximize their fracture resistance to
microcracking [7,8]. The method uses topology optimization combined with full simulations of micro
cracks initiation and propagations in heterogeneous periodic lattice structures. At each iteration of the
procedure, the topology of local geometries is optimized for maximizing the total fracture energy of
the material, by taking into account both bulk and interfacial damage between phases. Finally, a
method to construct simplified models of damage description at the scale of the homogeneous
materials in these regular lattice materials is proposed.

References

[1] C. Chateau, T.T. Nguyen, M. Bornert, J.Yvonnet, DVC-based image subtraction to detect cracking
in lightweight concrete, Strain, 2018, accepted.
[2] T.T. Nguyen, J. Yvonnet, M. Bornert, C. Chateau, F. Bilteryst, E. Steib, Large-scale simulations of
quasi-brittle microcracking in realistic highly heterogeneous microstructures obtained from micro CT
imaging, Extreme Mechanics Letters, 17:50-55, 2017

6
[3] T.T. Nguyen, J. Yvonnet, Q.-Z. Zhu, M. Bornert, C. Chateau, Initiation and propagation of
complex 3D networks of cracks in heterogeneous quasi-brittle materials: direct comparison between in
situ testing- microCT experiments and phase field simulations, Journal of the Mechanics and Physics
of Solids, 95:320-350, 2016.
[4] T.T. Nguyen, J. Yvonnet, M. Bornert, C. Chateau, K. Sab, R. Romani, R. Le Roy, On the choice of
numerical parameters in the phase field method for simulating crack initiation with experimental
validation, International Journal of Fracture, 197(2), 213-226, 2016.
[5] T.T. Nguyen, J. Yvonnet, Q.-Z. Zhu, M. Bornert, C. Chateau, A phase-field method for
computational modeling of interfacial damage interacting with crack propagation in realistic
microstructures obtained by microtomography, Computer Methods in Applied Mechanics and
Engineering, 312:567-595, 2016.
[6] T.T. Nguyen, J. Yvonnet, Q.-Z. Zhu, M. Bornert, C. Chateau, A phase field method to simulate
crack nucleation and propagation in strongly heterogeneous materials from direct imaging of their
microstructure, Engineering Fracture Mechanics, 139:18-39, 2015
[7] G.A. Francfort, J.J. Marigo, Revisiting brittle fracture as an energy minimization problem, J.
Mech. Phys. Solids 46(8), 1319–1342, 1998.
[8] C. Miehe, M. Hofacker, F. Welschinger, A phase field model for rate-independent crack
propagation: Robust algorithmic implementation based on operator splits, Comput. Methods Appl.
Mech. Engrg. 199, 2776–2778, 2010.
[9] D.Da, J. Yvonnet, L. Xia, G. Li, Topology optimization of particle-matrix composites for optimal,
fracture resistance taking into account interfacial damage, International Journal for Numerical
Methods in Engineering, 2018, accepted (doi: 10.1002/nme.5818)
[10] L. Xia, D. Da, J. Yvonnet, Topology optimization for maximizing the fracture resistance of quasi-
brittle composites, Computer Methods in Applied Mechanics and Engineering, 332:234-254,2018.

Acknowledgements

This work has benefited from a French government grant managed by ANR within the frame of the
national program Investments for the Future ANR-11-LABX-022–01.
The financial support from IUF (Institut Universitaire de France) is gratefully acknowledged.

7
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

FINITE ELEMENT ANALYSIS OF SLENDER BEAMS AND THIN SHELLS


Paulo de Mattos Pimenta1

1
Polytechnic School at University of São Paulo, Brazil, ppimenta@usp.br

ABSTRACT
This work reviews the theoretical and numerical results introduced by the author in the last fifteen years
[1-8]. As novelty a geometrically exact Bernoulli-Euler rod and a Kirchhoff-Love shell numerical
models are developed. The parameterization of the rotation field is done by the rotation tensor with the
Rodrigues formula that makes the updating of the rotational variables very simple. For the rods a Cubic
Hermitian interpolation for the displacements together with a quadratic Lagrange interpolation for the
torsion angle of the rod were employed within the usual Finite Element Method, leading to adequate C1
continuity. For the shell a consistent plane stress condition is incorporated at the constitutive level of the
model. A triangular finite element, with a quadratic interpolation for the displacements are applied for
the shell model. The connection between the two models is done by a Lagrange multiplier. This model
can be used in any case of rod and shell working together like with reinforced shells.

1. INTRODUCTION
The objective of this work is to present the theoretical and numerical results developed by the author in
[1-8]. As novelty we develop geometrically exact Bernoulli-Euler rod and a Kirchhoff-Love shell
numerical models. The parameterization of the rotation field is done by the rotation tensor with the
Rodrigues formula that makes the updating of the rotational variables very simple. For the rods a Cubic
Hermitian interpolation for the displacements together with a quadratic Lagrange interpolation for the
torsion angle of the rod were employed within the usual Finite Element Method, leading to adequate C1
continuity. For the shell a consistent plane stress condition is incorporated at the constitutive level of the
model. A triangular finite element, with a quadratic interpolation for the displacements are applied for
the shell model.

2. ROTATION RODRIGUES PARAMETERS


Let  denote the vector of Rodrigues parameters. The Rodrigues parameterization furnishes
4  
ˆ     
Q    1 2  , where   Skew    and 2     . (2.1)

2 2 
4
Vector  can be obtained from Q  ei  eir with the aid of

4
 axial  SkewQ  (2.2)
1  trQ

8
The curvature  is

  axial Q QT   Ξ  and r  QT   axial QTQ    ΞT  , (2.3)

and r is its back-rotated counterpart with

4  
Ξ   I  1   . (2.4)
4     2 
2.1. INCREMENTAL DESCRIPTION

Let   i and   i 1 denote a quantity at instants ti and ti , respectively. And let    be an incremental
quantity. Thus, one gets for the rotation tensor the following relations

Qi 1  QQi , where
ˆ  i 1  , Q  Q
ˆ    ˆ  i  . (2.5)
Qi 1  Q and Qi  Q

We recall the following result by Rodrigues

4  
i 1       1     . (2.6)
4  i    i
2
i 

In the incremental description we have
4   .
Ξ   I  1   (2.7)
4   2  2 

From [1] for the rod we have


2 1
  e3m e3i  e3i 1     e3m e3m , (2.8)
where  is the incremental torsion angle.

With aid of (2.2), (2.9) below is used for the shell

2
  e ij 1  e ij (2.9)
1  e ij 1  e ij

3.GEOMETRICALLY EXACT BERNOULLI-EULER ROD KINEMATICS


Let e1r , e2r , er3  be an orthogonal system placed at the reference configuration of the rod. The vectors
r
er ,   1, 2 , are placed on the cross section of the rod. Thus, e3 is orthogonal to this plane. The position
of the rod material points in the reference configuration is  and r the director
r

    r r ,   e3r , r r  er . (3.1)

We introduce the coordinate     e3r ,      0,   , where  is the rod length at reference


configuration. In the current configuration, the position of the material points is given by
1
x  z  r , r  Qr r , with e3  z  z  , and z   e3r  u  and z   u  . (3.2)

9
4. GEOMETRICALLY EXACT KIRCHHOFF-LOVE SHELL KINEMATICS
The middle surface of the shell is plane in the reference. Let e1r , e2r , e3r  be an orthogonal system with
the vectors er placed on the shell reference mid-plane and e3r normal to this plane. The position of the
rod material points in the reference configuration is  and r r is the director just like in the rod.
    r r ,   er , r r  3e3r . (4.1)

In the current configuration, the position of the material points is given by


x  z  r , r  Qr r , z    u , (4.2)

Note that

z, = er  u, and z,  u, with   ,  and (4.3)


1
e1  z,1 z,1 ,
1
e3  z,1  z,2 (z,1  z,2 ), (4.4)
e2  e3  e1 .

5. NUMERICAL EXAMPLES
Some numerical examples are presented during the presentation. A sample is displayed below.

Figure 1: pinched cylindrical shell

Figure 2: axially pulled hyberbolic anisotropic shell

10
Figure 3: inversion of creased shell

REFERENCES
[1] Silva, C.C., Maassen, S., Pimenta, P.M. & Schröder, J. “Geometrically exact analysis of Bernoulli-
Euler rods” submitted to Computer Methods In Applied Mechanics and Engineering, 2018
[2] Gay Neto, A., Pimenta, P.M. & Wriggers, P., Contact between spheres and general surfaces,
Comp. Meth. Appl. Mech. Egng. (2018) 686:716. https://doi.org/10.1016/j.cma.2017.09.016
[3] Viebahn, N., Pimenta, P.M. & Schröder, A simple triangular finite element for nonlinear thin
shells: statics, dynamics and anisotropy, Comput Mech (2017) 59: 281.
https://doi.org/10.1007/s00466-016-1343-6
[4] Gay Neto, A., Pimenta, P.M. & Wriggers, P., A master-surface to master-surface formulation for
beam to beam contact. Part II: Frictional interaction, Comp. Meth. Appl. Mech. Egng. (2017)
146:174. https://doi.org/10.1016/j.cma.2017.01.038
[5] Gay Neto, A., Pimenta, P.M. & Wriggers, P., A master-surface to master-surface formulation for
beam to beam contact. Part I: Frictionless interaction, Comp. Meth. Appl. Mech. Egng. (2016)
146:174. https://doi.org/10.1016/j.cma.2017.01.038
[6] Ota, N.S.N., Wilson, L., Gay Neto, A., Pellegrino, S. & Pimenta, P.M., Nonlinear dynamic
analysis of creased shells, Finite Elements in Analysis and Design (2016), 64:74.
https://doi.org/10.1016/j.finel.2016.07.008
[7] Gay Neto, A., Malta, E.R. & Pimenta, P.M., Catenary riser sliding and rolling on seabed during
induced lateral movement, Marine Structures (2015), 223:243.
https://doi.org/10.1016/j.marstruc.2015.02.001
[8] Generalization of the C1 TUBA plate finite elements to the geometrically exact Kirchhoff–Love
shell model, Comp. Meth. Appl. Mech. Egng. (2015), 210:244.
https://doi.org/10.1016/j.cma.2015.05.018

11
CILAMCE 2018 C ONGRESS
12-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

O LD AND N EW C ONTRIBUTIONS ON R EDUCED B ASIS M ETHODS


Yvon Maday 1
1 Sorbonne Université, Université Paris-Diderot SPC, CNRS,
Laboratoire Jacques-Louis Lions, LJLL, F-75005 Paris
Institut Universitaire de France
and Division of Applied Math, Brown University

Reduced basis methods (RBM) are a class of model order reduction (MOR) methods that allow to sim-
plify the approximation of the solution to a given problem by using very few information. The con-
struction can either be i) from explicit knowledge of informations on the solution, i.e. by point-wise
mesures of the solution or by acquisition of some moments on the solution, or ii) from knowledge of an
equation that the solution satisfies : a mathematical model that is provided by experts on the discipline
of the solution existence. A fundamental frame is that there must be some parameters around that allow
to characterize the set of all solutions when the parameter varies. This set constitute a manifold, the
structure of which (Kolmogorov width), is used to simplify the characterization of the solution, within
the manifold, with a small amount of information.
The idea, in the basic form of the reduced basis method, is then to identify few basis functions belong-
ing to the manifold and express a good approximation of the solution of interest as a linear combination
of these few basis functions. The approximation improving when the number of basis functions that are
used increases. This approach, in the context of parameter dependent partial differential equation (PDE)
— i.e. implicit definition of the solution — has led to Galerkin reduced basis methods that allow real
time simulations for optimization or inverse problems for instance. The PDE can be stationary or time
dependent, linear or non linear, leading to unique solutions or facing bifurcation phenomenon.
The addition of reliable a posteriori analysis and estimators allow in addition to certify the results in
the offline (restitution) stage.
In the explicit context, the approach is better known as the empirical interpolation method (EIM) that
allow to reconstruct a solution from point-wise measurements, this methods actually is at the basis of the
extension of the reduced basis method to nonlinear PDE’s.
The above statements, though still a hot subject in the numerical analysis community, is rather well
established now, at least in the pure context where either i) we consider that the explicit measurement
acquired on the solution are noise free, ii) the mathematical model written as a PDE is bias-free, iii) the
manifold of solution has indeed a simple structure that leads to a small Kolmogorov width.
But first, in real applications : i) there are always noises on the acquisition of data, ii) the model,
even very good, always has a bias with respect to the true phenomenon and iii) for convection dominated
problems, the linear nature of the manifold of all solutions is missing.
Second, in real word also, the basis functions are not known analytically, nor even exactly. They
are generally obtained from an underlying finite element code (in the implicit formulation), and thus
the Galerkin reduced element method generally requires to have access to the code to transform it and
implement an efficient reduced basis method. But this is not always possible leading to the invention of
“non invasive” reduced basis element methods.
In this talk, by summarizing some of the works with collaborators, I shall address how to cope with
these different difficulties together with recent aspects on the use of these technics in data assimilation
and data mining.

12 1
References
[1] Y Maday, T Anthony, JD Penn, M Yano. PBDW state estimation: Noisy observations; configuration-adaptive
background spaces; physical interpretations, ESAIM: Proceedings and Surveys 50, 144-168, (2015).
[2] Y Maday, O Mula, AT Patera, M Yano. The generalized Empirical Interpolation Method: stability theory
on Hilbert spaces with an application to the Stokes equation, Computer Methods in Applied Mechanics and
Engineering 287, 310-334, (2015).
[3] Y Maday, AT Patera, JD Penn, M Yanoi. A parameterized?background data?weak approach to variational
data assimilation: formulation, analysis, and application to acoustics, International Journal for Numerical
Methods in Engineering 102 (5), 933-965, (2015).
[4] Y Maday, A Manzoni, A Quarteroni. An online intrinsic stabilization strategy for the reduced basis approxi-
mation of parametrized advection-dominated problems, Comptes Rendus Mathematique 354 (12), 1188-1194,
(2016).
[5] Y Maday, O Mula, G Turinici. Convergence analysis of the generalized empirical interpolation method, SIAM
Journal on Numerical Analysis 54 (3), pp 1713-1731 (2016).
[6] Cagniart, N and Crisovan, R and Maday, Y and Abgrall, R. Model Order Reduction for Hyperbolic Problems:
a new framework, (2016).
[7] Ballarin, Francesco and Rozza, Gianluigi and Maday, Yvon Reduced-order semi-implicit schemes for fluid-
structure interaction problem, Model Reduction of Parametrized Systems, pp. 149–167, Springer, 2017.
[8] Maday, Yvon, and Tommaso Taddei Adaptive PBDW approach to state estimation: noisy observations; user-
defined update spaces, arXiv preprint arXiv:1712.09594 (2017).
[9] Chakir, R and Maday, Y and Parnaudeau, PA non-intrusive reduced basis approach for parametrized heat
transfer problem, 2018
[10] Gallinari, Patrick and Maday, Yvon and Sangnier, Maxime and Schwander, Olivier and Taddei, Tom-
masoReduced BasisÕ Acquisition by a Learning Process for Rapid On-line Approximation of Solution to
PDEÕs: laminar flow past a backstep, Archives of Computational Methods in Engineering, 25-1, pp. 131–
141, Springer, 2018
[11] Chakir, Rachida and Dapogny, Charles and Japhet, Caroline and Maday, Yvon and Montavon, Jean-Baptiste
and Pantz, Olivier and Patera, AnthonyComponent Mapping Automation for Parametric Component Reduced
Basis Techniques (RB-Component), ESAIM Proc.SMAI, 2018
[12] Hammond, Janelle K and Chakir, Rachida and Bourquin, Frédéric and Maday, YvonPBDW: a non-intrusive
Reduced Basis Data Assimilation Method and its application to outdoor Air Quality Models, 2018
[13] Fick, Lambert and Maday, Yvon and Patera, Anthony T and Taddei, Tommaso A stabilized POD model for
turbulent flows over a range of Reynolds numbers: optimal parameter sampling and constrained projection,
Journal of Computational Physics, Elsevier, 2018
[14] Argaud, J-P and Bouriquet, B and de Caso, F and Gong, H and Maday, Y and Mula, O. Sensor place-
ment in nuclear reactors based on the generalized empirical interpolation method, Journal of Computational
Physics,363, pp. 354–370, Elsevier, 2018
[15] Herrero, Henar and Maday, Yvon and Pla, Francisco. Reduced basis method applied to a convective stability
problem, Journal of Mathematics in Industry, 8-1, SpringerOpen, 2018

13 2
2. SECTIONAL LECTURES

15
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

ENGINEERING STRUCTURES INTEGRITY AND DURABILITY VALIDATION

A. Ibrahimbegovic1, A. Boujelben2, X.N. Do2, S.Dobrilla2,6, E. Hadzalic2,4, S. Grbcic2,3, I.


Imamovic2,4, E. Karavelic2,4, A. Mejia2, P. Moreno2, M. Nikolic5, I. Rukavina2, T.Rukavina2,3
1
UT Compiegne – Sorbonne Univ. & Institute Universitaire de France, adnan.ibrahimbegovic@utc.fr
2
UT Compiegne –Sorbonne Univ.,3 Univ. Rijeka, 4 Univ. Sarajevo, 5 Univ. Split, 6 TU-Braunschweig

Abstract. In this work we address the challenge pertinent to validating safety of large engineering structures, both in
terms of integrity to failure under extreme conditions and durability within their environment. Of particular interest are
industrial domains of excellence in France, such as energy-production (nuclear power plants, large offshore turbines) and
air- or land- transportation (large airplanes, high speed trains, cargo ships). The main obstacle to overcome pertains to our
inability to certify the structural safety by performing with real-size and real-time experiments, either due to excessive
structure size, to excessive cost due to irreplaceable structure component or due to inability to reproduce with high fidelity
the extreme conditions to which the structure is exposed. We seek to propose the state-of-the-art advances in
computational methods that can be brought to bear upon this class of problems, providing the full understanding of the
potential failure modes of the given system, along with the very detailed simulation of extreme conditions brought by
man-made and natural hazards (explosion, fire, stormy wind, earthquake, tsunami …). We also seek further
developments in recently proposed approach to coupled mechanics-probability computations that can be successfully
used to provide a detailed interpretation of structure tests under heterogeneous stress field and to identify both model
parameters and their probability distribution. Finally, we propose to use such a combined approach with probability
computations for uncertainty propagation, which can offer a clear explanation of the size effect influence on dominant
failure modes of massive composite structures.
Keywords: massive structures; multiscale computations; coupled mechanics-probability approach; size effect;

1. Introduction and Motivation

This paper deals with important challenge on validating the durability and lifetime integrity of massive
composite structures under extreme conditions. The illustrative examples (see Figure 1) come from the
application domains in energy production systems, with both currently dominant nuclear or renewable
energy sources (nuclear power plant, offshore wind-turbines or hydro-turbines), as well as in air- and
land-transportation (large airplanes, high speed trains or cargo ships). Special attention is given to
costly massive structures with ‘irreplaceable’ components, which are characterized by a number of
different failure modes that require the most detailed description and interaction across the scales. We
would like to significantly improve the currently dominant experimental approach, and thus accelerate
innovations in this domain.

2. Main Objectives
The main objective is development of novel Mesh-in-Element (MIEL) Multiscale Method capable of
representing strain field heterogeneities induced by evolution (and interaction) of localized failure
mechanisms in massive structure, pertaining to micro scale (FPZ-fracture process zone), macro scale
including softening (macro cracks) and non-local macro scale (bond-slip for long fiber reinforcement).
The objective of MIEL Multiscale Method is also to provide capabilities for quantifying the risk of
premature localized failure through probability description of initial defects (microstructure
heterogeneity) and uncertainty propagation through scales. The novel scientific concept to be explored
pertains to multiscale formulation and solution of coupled nonlinear mechanics-probability problem

17
replacing the standard homogenization approach that can only provide average (deterministic)
properties of heterogeneous composites. This concept is of interdisciplinary nature with Mechanics
(defining probability distribution) and Applied Mathematics (providing uncertainty propagation)
combined in order to capture the influence of heterogeneities and fine scale defects on premature
failure.

a) b) c)

Fig. 1 Durability of (costly) massive composite structures: a) nuclear power plant both existing PWR and
new EPR systems - stringent requirement on waterproof containment structure of CBFR composites;
b) (European answer to) Boeing 787 Dreamliner, the most fuel-efficient aircraft built of CFRP
composites - requirement on no-return-to-hub for crack reparation; c) large offshore wind-turbine with
CFRP composites blades and deep-sea CBRF composites support – requirement of operation
capabilities for extreme weather conditions

The most important challenge concerns the ability to provide the sound, probability-based explanation
of size effect, with different failure modes observed for different size specimens and real structure
built of the same composite materials (see Figure 2).


Fig. 2 Mesh-In-Element (MIEL) multiscale method – validation of structural integrity of giant offshore wind-
turbines with carbo-epoxy blades and reinforced concrete support: i)left-top row: scales for 3D carbon-
epoxy woven composite with fine scale for resin constituents, meso-scale warp and weft stuffers and
non-local scales for binder; left-bottom row: scale for cement, concrete and reinforced concrete; ii) right
top: off-line probability computation for identification by Bayesian updates; right-bottom: on-line
probability computation for probability-based interpretation of size effect.

3. Concluding remarks on current and future research

The biggest potential gain concerns changing the validation procedures for massive structures that are
beyond the size suitable for testing at present. The scientific gains concern providing the Mesh-in-
Element (MIEL) Multiscale Method that connects computations with design studies (optimization),
testing (identification) and safety verification (monitoring) of massive composite structures. The
scientific gains also concern further placing the proposed method within multiphysics framework,
along with the original use of goal oriented error estimates to provide sufficiently reliable
interpretation of extreme conditions (e.g. fluid or heat flow) and the code-coupling software
implementation to quickly integrate existing simulation codes within such a framework.

18
The main technological gain is in development of the open source computational tools that can speed-
up testing, innovation and decision-making in complex composite systems. Of special interest is the
strategy that allows to integrate the existing legacy software products that are used to verify and
validate safety of particular components assembled within such complex systems. There are multiple
challenges in solving any such problem pertaining to: theoretical formulation, discrete approximation,
algorithmic stability and robustness, and finally informatics developments capable of integrating
existing legacy codes. Two model problems of composites with great application potentials will be
examined. First pertains to cement-based fiber reinforced (CBFR) composites, which will allow for
validation of our method against recently completed experimental program in French excellence
project ECOBA. Second model of carbon fiber reinforced polymers (CFRP) that is validated in
collaboration with experimentalists at Université de Technologie de Compiègne.

Further details on point of departure in the current research and developments to follow are given in
our recent works (Ibrahimbegovic et al. 2003a, 2003b, 2003c, 2003d, 2004, 2005, 2007, 2008, 2009a,
2009b, 2009c, 2010a, 2010b, 2011, 2012, 2014, 2015, 2016, 2017, 2018).

Acknowledgments

The research described in this paper is financially supported by the Region Hauts-de-France, EU
(FEDER), ANR and DFG, French Ministry of Research and French Ministry of Foreign Affairs. We
also acknowledge many interesting discussions and collaborations with: D. Brancherie (UTC), E.
Lefrancois (UTC), S. Dolarevic (Univ. Sarajevo), G. Jelenic (Univ. Rijeka), A. Kurtovic (Univ.
Sarajevo), Z. Nikolic (Univ. Split), J.L. Perez-Aparicio (Univ. Politècnica Valencia), I. Kozar (Univ.
Rijeka), B. Brank (Univ. Ljubljana), H.G. Matthies (TU Braunschweig).

References

[1] Ibrahimbegovic A., D. Markovic, Strong coupling methods in multi-phase and multi-scale., Comp.
Meth. Appl. Mech. Eng., 192, 3089-3107, (2003)
[2] Ibrahimbegovic A., D. Brancherie, Combined hardening and softening constitutive model for
plasticity: precursor to localized failure, Comp. Mech., 31, 88-100, (2003)
[3] Ibrahimbegovic A., A. Delaplace, Microscale and mesoscale discrete models for dynamic fracture
of structures, Comp. Struc., 81, 1255-1265, (2003)
[4] Ibrahimbegovic A., D. Markovic, F. Gatuingt, Constitutive Model of Coupled Damage-Plasticity
and Its Numerical Implementation, Revue euro. élém. finis, 12, 381-405, (2003)
[5] Ibrahimbegovic A., C. Knopf-Lenoir, A. Kucerova, P. Villon, Optimal design and optimal control
of structures undergoing finite rotations, Int. J. Num. Meth. Eng., 61, 2428-2460, (2004)
[6] Ibrahimbegovic A., I. Gresovnik, D. Markovic, S. Melnyk, T. Rodic, Shape optimization of two-
phase material with microstructure, Eng. Comp., 22, 605-645, (2005)
[7] Ibrahimbegovic A., S. Melnyk, Embedded discontinuity finite element method for modeling of
localized failure in heterogeneous materials with structured mesh: an alternative to extended finite
element method, Comp. Mech., 40, 149-155, (2007).
[8] Ibrahimbegovic A., P. Jehel, L. Davenne, Coupled damage-plasticity model and direct stress
interpolation, Comp. Mech., 42, 1-11, (2008)
[9] Ibrahimbegovic A., G. Herve, P. Villon, Nonlinear impact dynamics and field transfer suitable for
parametric design studes, Eng. Comput., 26, 185-204, (2009)
[10] Ibrahimbegovic A., A. Kucerova, et al. ‘CE structures: multiscale damage representation,
identification..’, in ‘Damage assessment and quick reconstruction after wars and natural desasters’,
Springer, 1-28, (2009)
[11] Ibrahimbegovic A.. Nonlinear Solid Mechanics: Theoretical Formulations and Finite Element
Solution Methods, Springer, (ISBN 978-90-481-2330-8, E-book 978-1-4020-9793-5), pp. 1-571,
(2009)
[12] Ibrahimbegovic A., A. Boulkertous, L. Davenne, D. Brancherie, Modeling of reinforced-concrete
structures providing crack-spacing based on XFEM, ED-FEM and novel operator split solution
procedure., Int. J. Num. Meth. Eng., 83, 452-481, (2010)

19
[13] Ibrahimbegovic A., A. Boulkertous, L. Davenne, M. Muhasilovic, A. Pokrklic, ‘On modeling of
fire resistance tests on concrete and reinforced-concrete structures’, Computers and Concrete, vol
7. No. 4, 285-301, (2010)
[14] Ibrahimbegovic A., J.B. Colliat, M. Hautefeuille, D. Brancherie, S. Melnyk, , Probability based
size effect representation for failure of civil engineering structures built of heterogeneous materials,
in (eds. M. Papadrakakis, M. Fragiadakis, G. Stefanou), ‘Computational Methods in Stochastic
Dynamics’, Springer, Berlin, 289-311, (2011)
[15] Ibrahimbegovic A., H.G. Matthies, Probabilistic Multiscale Analysis of Inelastic Localized
Failure in Solid Mechanics, Comp. Assis. Meth. Eng. Sci., 19, 277–304, (2012)
[16] Ibrahimbegovic A., R. Niekamp, C. Kassiotis, D. Markovic, H. Matthies, Code-coupling strategy
for efficient development of computer software in multiscale and multiphysics nonlinear evolution
problems in computational mechanics, Advances Eng. Software, 72, 8-17, (2014)
[17] Ibrahimbegovic A., J-M. Ghidaglia, A. Serdarevic, E. Ilic, M. Hrasnica, S. Dolarevic, N.
Ademovic, ‘ECCOMAS MSF 2015 – Multiscale Computational Methods for Solids and Fluids’,
Univ. Sarajevo, (ISBN 978-9958-638-23-7), pp. 1-269, (2015)
[18] Ibrahimbegovic A., ‘Computational Methods for Solids and Fluids: Multiscale Analysis,
Probability Aspects and Model Reduction’, Springer, (ISBN 978-3-319-27994-7), pp. 1-493,
(2016)
[19] Ibrahimbegovic A., B. Brank, I. Kozar, ‘ECCOMAS MSF 2017 – Multiscale Computational
Methods for Solids and Fluids’, (ISBN 978-961-6884-49-5), Univ. Ljubljana, pp. 1-279, (2017)
[20] Ibrahimbegovic A., A. Boujelben, Long-term simulation of wind turbine structure for distributed
loading describing long-term wind loads for preliminary design, Int. J. Coupled Systems
Mechanics, 7, 233-254, (2018)

20
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

DISCRETE ELEMENT MODELING OF GRANULAR AND PARTICULATE MATERIALS

Eduardo M. B. Campello 1
1
Department of Structural and Geotechnical Engineering, University of São Paulo, Brazil
campello@usp.br

Granular and particulate materials can be found in a myriad of natural phenomena and human
applications, and across various length scales: from rock stacks and gravel piles to fine powders and
particulate flows; from heaps of nuts and beans and mounds of sand to compact aggregates of very high
added value in the pharmaceutical, chemical, food and microelectronics industries. Knowledge on the
physics of such materials has evolved significantly over the past decades, especially with the aid of
computational methods, but still there is a lot to pursue. From the point of view of their theoretical
representation, granular and particulate materials cannot be idealized as a continuum – unless the length
scale of the physical phenomenon one is interested in is very large in comparison to the size of the
grains, as happens, e.g., in the overall deformation of soil massifs in geotechnical engineering problems.
Instead, they must be understood a priori as a discrete medium, in which matter is concentrated in a very
large number of elementary solid entities (the grains or particles), each with finite size and well-defined
shape. The grains or particles typically experience numerous contact interactions, which in turn occur
with significant energy dissipation through both friction and localized plastic deformations. This
particular nature gives rise to a number of intriguing phenomena that are peculiar to these materials,
such as size segregation under shaking (and a corresponding resistance to mixing), occasional
fluidization, the formation of wave patterns at their free surfaces upon excitation (as observed in fluids),
the ability to form static piles that do not dismantle or dissolve spontaneously (as typical of solids), and
many others. These aspects illustrate well the fact that, though constituted of elementary solid entities –
i.e., by entities that have their individual behavior very well characterized –, these materials have their
own physics. The purpose of this talk is to present a discrete element model (DEM) for the simulation
of granular and particulate materials. The approach is intended to be simple, yet keeping as much
consistency as possible with classical mechanics principles. To this end, and unlike most of the DEM
formulations available in the literature, we separate from the outset the physics of the problem from the
numerical scheme devised to solve its equations. This allows us to concentrate on the physics and the
numerics independently, leading to a clearer and cleaner formulation – at least in our opinion. One
distinguished feature is the adoption of a vector type of parameterization for the description of the
rotations of the grains. This enables us to represent rotations in a manner that is formally identical to
displacements, i.e., by vectors, such that all degrees-of freedom of the model retain the same formal
structure. Complex entities such as quaternions or Euler angles are thereby circumvented, in sharp
contrast to all DEM models in the literature. We use phenomenological models to describe the various
forces involved at the level of the particle interactions. Particular attention is devoted to the modeling
of contact and adhesion (with stick-slip friction and rolling resistance), since inter-particle energy
exchange, occasional bonding and agglomeration are crucial aspects in many real-world applications
involving these types of materials. The consideration of temperature effects, which may be of utmost
importance in some applications, is also undertaken within a fully coupled thermo-mechanical

21
triangular piling

11.9 mm

300 mm

f = 7.8 Hz f = 12 Hz

λ Pmax λ
Pmax

Figure 1. Surface instabilities in a vertically vibrated granular medium. Problem definition (top) and sequences of snapshots
(from top to bottom) for two different values of the vibration frequency. Left sequence: f=7.8 Hz; right sequence: f=12 Hz.
Surface instabilities show up at the initial stages of the motion, giving rise to an array of waves separated by a typical
wavelength λ, whose magnitude varies with the imposed excitation frequency. The images above are in excellent visual
agreement with the experimental results reported by J. Duran in his famous book Sands, powders and grains: an introduction
to the physics of granular matter (Springer, 1997).

framework. Details of the model´s computational implementation, through a special time-stepping


(fixed-point, iterative) numerical scheme are briefly outlined. In the end, numerical simulations
depicting some of the intriguing phenomena of these materials, as well as covering some modern

22
engineering applications such as the deposition of particles onto surfaces for additive manufacturing and
3D printing technologies, and the infiltration of particle-laden fluids into porous media for material
functionalization, are shown to illustrate the potentialities of the scheme. We remark that this abstract
aims only to outline what is addressed in the talk; for a comprehensive technical description of the
model, as well as systematic derivations of its equations, we refer to previous works by the author in
Campello [1], [2], [3], [4], [5], [6], Campello and Zohdi [7], [8], [9], [10], Campello and Cassares [11],
[12], Gay Neto and Campello [13] and Fernandes et al. [14]. Figure 1 and Table 1 below show results
from a typical simulation, as a model example. We believe that simple, consistent computational models
may be a useful tool for the simulation of granular and particulate materials and, in a broader sense,
many other particle systems.

Table 1. Surface instabilities in an extended granular medium. Analyses results. Wavelengths and
amplitudes are measured at a time instant when the cell bottom touches the bottom layer of the stack,
exactly as in the physical experiment by J. Duran.

Wavelength Wave amplitude Pmax


Frequency
Experimental Present work Experimental Present work
6.3 Hz 101.0 mm 100.0 ± 5.2 mm 85.1 mm 87.6 ± 4.0 mm
7.8 Hz 73.4 mm 72.1 ± 4.7 mm 55.5 mm 57.0 ± 3.2 mm
10.0 Hz 53.1 mm 52.3 ± 4.1 mm 33.8 mm 34.2 ± 2.4 mm
12.0 Hz 43.5 mm 42.8 ± 3.1 mm 23.4 mm 21.0 ± 2.0 mm
14.1 Hz 37.4 mm 37.0 ± 3.0 mm 17.0 mm 16.5 ± 2.0 mm

References

[1] E. M. B. Campello, "A computational model for the simulation of dry granular materials,"
International Journal of Nonlinear Mechanics, vol. 106, pp. 89-107, 2018.
[2] E. M. B. Campello, "A description of rotations for DEM models of particle systems,"
Computational Particle Mechanics, vol. 2, p. 109–125, 2015.
[3] E. M. B. Campello, Um modelo computacional para o estudo de materiais granulares,
(Habilitation thesis). São Paulo: Escola Politécnica da Universidade de São Paulo, 2016.
[4] E. M. B. Campello, "Computational modeling of granular materials," in Proceedings of the
XXVII Iberian-Latin-American Congress on Computational Methods in Engineering, Brasília,
2016.
[5] E. M. B. Campello, "Computational modeling and simulation of rupture of membranes and thin
films," Journal of the Brazilian Society of Mechanical Sciences and Engineering, vol. 37, pp.
1793-1809, 2015.
[6] E. M. B. Campello, "Computational modeling of particle deposition processes," in Proceedings
of the XXXVIII Iberian Latin-American Congress on Computational Methods in Engineering,
Florianópolis, 2017.
[7] E. M. B. Campello and T. Zohdi, "A computational framework for simulation of the delivery of
substances into cells," International Journal for Numerical Methods in Biomedical Engineering,
vol. 30, p. 1132–1152, 2014.
[8] E. M. B. Campello and T. Zohdi, "Design evaluation of a particle bombardment system used to
deliver substances into cells," Computer Modeling in Engineering & Sciences, vol. 98, no. 2, pp.
221-245, 2014.

23
[9] E. M. B. Campello and T. I. Zohdi, "Effect of particle spin on the spatio-thermal distribution of
incandescent materials released from explosions," Computational Particle Mechanics, p. (in
print), 2019 (accepted).
[10] E. M. B. Campello and T. I. Zohdi, "On pressurized functionalized particle-laden fluid
infiltration into porous media," International Journal for Multiscale Computational Engineering,
p. (in print), 2018 (accepted).
[11] E. M. B. Campello and K. R. Cassares, "Rapid generation of particle packs at high packing
ratios for DEM simulations of granular compacts," Latin American Journal of Solids and
Structures, vol. 13, pp. 23-50, 2016.
[12] K. R. C. Seko and E. M. B. Campello, "Micromechanics of randomly-generated particle packs
for DEM simulations of granular materials using the stress-force-fabric relationship," in
Proceedings of the XXXVIII Iberian Latin-American Congress on Computational Methods in
Engineering, Florianópolis, 2017.
[13] A. G. Neto and E. M. B. Campello, "Granular materials interacting with thin flexible rods,"
Computational Particle Mechanics, vol. 4, no. 2, pp. 229-247, 2017.
[14] A. C. S. Fernandes, H. C. Gomes, E. M. B. Campello and P. M. Pimenta, "A fluid-particle
interaction method for the simulation of particle-laden fluid problems," in Proceedings of the
XXXVIII Iberian Latin-American Congress on Computational Methods in Engineering,
Florianópolis, 2017.

24
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A MIXED DATA - DRIVEN / TIME INTEGRATOR KERNEL - BASED APPROACH FOR


MANIFOLD LEARNING OF DYNAMICAL SYSTEMS

Florian De Vuyst 1 , Faker Ben Belgacem 2 , Piotr Breitkopf 3 , Georges Oppenheim 4 and Pierre Villon 5
1 LMAC, Sorbonne Universités, UTC, , Compiègne France, fdevuyst@utc.fr
2 LMAC, Sorbonne Universités, UTC, Compiègne France, faker.ben-belgacem@utc.fr
3 Laboratoire Roberval, Sorbonne Universités, UTC, CNRS, Compiègne France, piotr.breitkopf@utc.fr
4 Université Marne-la-Vallée, Université Paris-Sud, CNRS, France, georges.oppenheim@gmail.com
5 Laboratoire Roberval, Sorbonne Universités, UTC, CNRS, Compiègne France, pierre.villon@utc.fr

This paper deals with the identification of dynamical systems from data, taking into account a priori
knowledge in the form of a continuous differential model. We investigate Dynamic Mode Decomposition
methods and extensions (extended DMD, kernel DMD). It appears that approximate time integrators can
be used in the design of a suitable kernel. The resulting hybrid computational approach can be seen as a
time advance solver recalibrated by data.
Let us consider the following continuous autonomous dynamical system:

ẋ(t) = F (x(t)), t > 0, x(0) = x0 . (1)

where x0 ∈ Rd and F : Rd → Rd is a Lipschitz continuous mapping. From the differential problem (1)
and a time sampling parameter τ > 0, one can derive a discrete dynamical system

xk+1 = f (xk ), k ≥ 0, x0 = x0 (2)

where Z τ
f (x) = x + F (x(s)) ds (3)
0

The admissible states {x(t)}t≥0 solutions of (1) are supposed to belong to a sub-manifold M ⊂ Rd . This
is of course also true for the discrete states {xk }k≥0 . Remark that one can use a time integrator on F
to get a prediction f˜ of f . Let φ(x, τ) denote a given time integrator over the time increment τ with x
as initial data. So we can use f˜(xk ) = φ(xk , τ). For example, the simple case of explicit forward Euler
scheme
f˜(x) = x + τF (x)
returns a first-order accurate approximation of f (with an error of order o(τ)).
The system (1) is supposed to be a rather accurate model of a true physical system from wich one can
have measurements from sensors. We will assume that the sensors return states (or observable quantities)
at discrete times t k = τk, k ≥ 0. From the data we would like to identify (estimate) the true vector field fT
of the real discrete dynamical system xk+1 = fT (xk ), with the help of the a priori knowledge of the
model (1).
Dynamic Mode Decomposition (DMD) [2, 3, 4] is a rather recent data-driven computational ap-
proach that identifies the best linear model that maps vector-valued input states to propagated output
states. Consider the two d × N data matrices
   
| | | | | |
X = x1 x2 . . . xN  , Y = y 1 y 2 . . . y N 
| | | | | |

25 1
with input states xk and output states
y k = fT (xk ).
In standard DMD, we want to find the best least square linear model

xk+1 = fT (xk ) ≈ A xk k ≥ 0,

where A ∈ Md (R), i.e. we look for a minimizer

1
min kY − AXk2F .
A∈Md (R) 2

We straightforwardly get
A = Y X†
where X † denotes standard the Moore-Penrose pseudo-inverse of A. DMD has been proved to be a
powerful tool in different contexts like e.g. feature extraction [2], short-term prediction, prediction of
quasi-periodic solutions [5], or state estimation for feedback control.
Since the original linear DMD, one can now find today nonlinear DMD extensions in the literature. A
nonlinear discrete dynamical system like (2) can be formulated in a different way thanks to the Koopman
theory [1]. One can define an operator K acting on functions g : M → Rd defined by K g = g ◦ f , i.e.

(K g)(x) = g(f (x)), x∈M. (4)

The operator K is referred to as the Koopman operator, or compositional operator. It is a linear operator
acting on an infinite-dimensional functional space. If K is known, then actually the discrete dynamical
system (2) is completely determined. Indeed, applying the Koopman operator to the function g(x) =
x · ei gives the ith component of f : K g(x) = f (x) · ei . It is possible to find an approximation of the
Koopman operator in a finite-dimensional space. In the literature one can find Galerkin-based approaches
as well as collocation methods. In what follows, we deal with a collocation approach. Consider a basis
of M linearly independent functions {ψi }i=1,...,M , ψi : M → R. Let Ψ(x) ∈ RM denote the vector with
{ψi (x)}i as components:
T
Ψ(x) = ψ1 (x), ..., ψM (x) .
We will assume that M  d. Following Koopman’s ideas, rather than identifying the mapping between
the input states xk and the output states y k = f (xk ), we search for a linear mapping between the aug-
mented input data Ψ(xk ) and augmented output data Ψ(f (xk )), i.e. a matrix we look for A such that

Ψ(f (xk )) ≈ A Ψ(xk ) (5)

in a least square sense. Consider the two snapshot matrices


   
| | | | | |
Z = Ψ(x1 ) Ψ(x2 ) . . . Ψ(xN ) , Y = Ψ(y 1 ) Ψ(y 2 ) . . . Ψ(y N )
| | | | | |

with y k = f (xk ). Then the minimizer of the least square problem

1
min kY − AZk2F
A∈MM (R) 2

is clearly given by A = Y Z † . This is the idea of the so-called extended Dynamic Mode Decomposition
(EDMD) approach [4]. As a next step, suppose that the basis functions ψi can reproduce the full state
vector x itself, i.e. there are M constant vectors bi such that
M
x = ∑ bi ψi (x), (6)
i=1

26 2
which can be written in vector form
x = B Ψ(x) (7)
with B = col(bi )i . By combining (5) and (7), we have the approximation

y k = f (xk ) = BΨ(f (xk ))) ≈ BA Ψ(xk ). (8)

We get the data-driven model/estimator

ŷ = fˆ(x) = R Ψ(x) (9)

with a rectangular matrix R = BA. It is of course nonlinear by construction. In the time continuous case,
we would get a similar model in the form dtd (Ψ(x)) ≈ AΨ(x), x ≈ BΨ(x) so that
d
ẋ ≈ B (Ψ(x)) ≈ R Ψ(x), R = BA.
dt
Without neither any information on data location nor knowledge on f , one can choose standard basis
functions ψi (x) with universal approximation property like polynomials, Fourier functions, radial basis
functions, kernel-based approximation, etc as discussed in [5]. But the price to pay may be the large
amount of basis functions to return an accurate model. Sparse representation techniques and LASSO-
type algorithms [6] can be used to identify the active set of basis functions at the cost of a higher numer-
ical complexity.

Kernel-based interpolation and approximation


When some sample states {x1 , ..., xN ∈ Rd } at different sites are available, kernel-based techniques
[10] appear to be a nice framework of approximation. Consider a symmetric positive definite kernel
k : Rd ×Rd → R. From snapshot states (x1 , ..., xn ) and (f (x1 ), ..., f (xn )), we can define an interpolation
I f of f in the form
N
I f (x) = ∑ a j k(x, x j ) (10)
j=1

with a1 , ..., aN ∈ Rd to determine to get the interpolation property I f (xi ) = f (xi ) for all i ∈ {1, ..., N},
i.e.
N
f (xi ) = ∑ a j k(xi , x j ), i = 1, ..., N.
j=1

This leads to the system of linear symmetric positive semi-definite system KA = F to solve, where
Ki j = k(xi , x j ), A = col(a j ) and F = col(f (xi )). For a positive definite kernel matrix K, we get a
unique matrix A = K −1 F. In [7], Héas and Herzet link EDMD with kernel approximation. By choosing

ψ j (x) , k(x, x j ), j = 1, ..., N

as basis functions, we get a vector of kernel observables Ψ(x) = k(x):

k(x) = (k(x, x1 ), ..., k(x, xN ))T .

The Koopman operator approximation restricted to the vector space spanned by the functions k(., xi ),
i = 1, ..., N consists in finding a matrix A such that

k(f (x)) ≈ A k(x)

in a least square sense. If x can be itself accurately reproduced by the kernel functions, i.e. x ≈ Bk(x),
we get the approximation
fˆ(x) = R k(x), (11)
where R can be identified from the data. From time-continuous system, we would search for approxima-
tions in the form ẋ = R k(x). Now comes the question of a good choice of kernel function for a given
dynamical system.

27 3
Suitable kernels using approximate time integrators
Let us now discuss about a suitable choice of kernel k. A first trivial remark is that if f were known, then
the best choice of kernel would be
k? (x, y) = f (x) · f (y). (12)
The kernel (12) is a reproducing kernel [8, 10] and its associated feature map (denoted Φ in the sequel,
see [9]) identifies with f itself. Consider the approximation fˆ of f defined by
N N
fˆ(x) = ∑ a j k? (x, x j ) = ∑ a j f (x j ) · f (x).
j=1 j=1

This can be written in matrix form fˆ(x) = AF T f (x) with A = col(a j ) ∈ MdN (R) and F = col(f (x j )) ∈
MdN (R). If A is chosen such that AF T = I (i.e. A = (F T )† ), then fˆ identifies itself with f : fˆ = f .
The feature map Φ does not need to be high-dimensional for exact reconstruction of f . Moreover, is
it sufficient to get d linearly independent state measurements {f (x j )} j to identify the matrix A; if this
case, we would have A = F −T .
In the context of this paper, we do not know fT nor f . We assume that the underlying dynamical
system is continuous in time and we have access to state measurements (or suitable observables) at
discrete instants t k = kτ. We have an a priori knowledge of an accurate time-continuous differential
model of the system, characterized by the vector field F . The discrete vector field f is linked to F
by the integral formula (3). If it is able to approximate this formula by an explicit one, leading to an
intermediate fidelity model f˜ of f , then we are able to define an approximate kernel k̃ in the form

k̃(x, y) = f˜(x) · f˜(y). (13)

Results
In the paper different choices of approximate kernels (13) built from time integration schemes are studied.
Numerical results obtained on nonlinear discrete dynamical system show both the accuracy and the
predictability of the resulting model.

References
[1] I. Mezić, Analysis of fluid flows via spectral properties of the Koopman operator, Annual Review of Fluid
Flows, 45, pp. 357–378 (2013).
[2] P. Schmid, Dynamic mode decomposition of numerical and experimental data, J. Fluid Mech., 2010.
[3] J. Tu, C. Rowley, D. Luchterburh, S. Brunton, J. Kutz, On dynamic mode decomposition: Theory and appli-
cations;, J. Comp. Dynamics, 2014
[4] M. Williams, I. Kevredikis, C. Rowley, A data-driven approximation of the Koopman operator: Extending
dynamic mode decomposition, J. Nonlinear Science, 2015.
[5] J.N. Kutz, S.L. Brunton, B.W. Brunton, J.L. Proctor, Dynamic Mode Decomposition, Data-driven modeling of
complex systems, SIAM book (2016).
[6] S. Brunton, J. Proctor and J. N. Kutz, Discovering governing equations from data by sparse identification of
nonlinear dynamical systems, Proceedings of the National Academy of Sciences 113 (2016) 3932-3937.
[7] P. Héas, C. Herzet, Optimal kernel-based dynamic mode decomposition, ArXiv preprint (2017).
[8] N. Aronszajn, Theory of reproducing kernels, T. Am. Math. Soc., 68, 3, pp. 337–404 (1950).
[9] S. Bergman, The Kernel function and conformal mapping, 2nd edn. American Mathematical Society, Provi-
dence, RI (1950).
[10] A. Berlinet and C. Thomas-Agnan, Reproducing Kernel Hilbert spaces in Probability and Statistics, Kluwer
(2004).

28 4
3. OTHER LECTURES

29
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

E FFICIENT COMPUTATIONAL MODEL FOR FLUID - STRUCTURE INTERACTION IN


APPLICATION TO LARGE OVERALL MOTION OF GIANT WIND TURBINE FLEXIBLE
BLADES WITH LONG TERM RESPONSE

Abir Boujelben 1 , Adnan Ibrahimbegovic 2


1 Sorbonne Universités, Université de Technology de Compiègne, abir.boujelben@utc.fr
2 Sorbonne Universités, Université de Technology de Compiègne, adnan.ibrahimbegovic@utc.fr

1. Context and objectives


The current push for renewable energy resources is the main motivation for constructing large wind
turbines with flexible blades exceeding 100 meters of length. Increasing the blade dimensions leads to
serious problems of operation, stability and durability which must be addressed in the preliminary design
process. It is therefore necessary to carry out long term computations based on fluid-structure interaction
(FSI) model to better understand the blade behavior when structure flexibility and up-sizing effects are
taken into account. The first ingredient of FSI is the aerodynamic model. The computational fluid
dynamic (CFD) is considered to be the most accurate, but also by far the most costly model especially for
full-scale 3D FSI computations. It is impractical to use it for the case that concerns testing several design
configurations for comparison. Then, we prefer to use panel method (PM) which is sufficiently fast since
it is restricted only to the fluid-structure interface. It is yet reasonably accurate for computing the overall
thrust on the blade. The PM model developed in this paper is modified to account for flexible lifting
body in large overall motion respectively by introducing vorticities in fluid-structure interface and by
considering the relative flow velocity. The second ingredient of FSI is the structure model. In view of the
target size, the blades are subjected to large deformations, displacements and rotations. The finite element
model is based on the theory of deformable solid body overall large motion. In this paper, we choose
the 3D solid model which allows more rigorous deformation description, provides more accurate results
from beam and gives the local stresses within the blade which is of interest for fatigue failure, as the most
common problem for wind turbines. Compared with the standard 8-node solid element, the proposed
element is enhanced by a particular choice of the deformation measure for more accurate approximation
of bending and torsional deflections. These deflections are associated to high frequency modes leading
to stiff problems. The standard implicit time stepping scheme are not appropriate for analysis of such
problems. We propose a new robust time-stepping scheme based on mid-point scheme which improves
the numerical stability and allows us to carry out long term simulations. The FSI computation is achieved
by an iterative coupling algorithm. The model exchange in each time step the aerodynamic loads (panel
method code) and the structure displacements and velocities (FEM approximation) at the fluid-structure
interface until equilibrium. The appropriate choice of the aerodynamic and structure models leads to a
direct way to data transfer between fluid and structure parts.

2. Problem statement
2.1. Fluid flow computations by 3D vortex panel method
The flow field around the blade is assumed to be potential (∇ · v f = 0; ∇ × v f = 0), except for the layer
in contact with the blade where a vorticity field is introduced (ξ = ∇ × v f ) in order to obtain a non-zero

31 1
RR RR
circulation (Γ = S ∇ × v f · n dS = S ξ · n dS) and thus to generate the blade lift (L = ρ f |v∞ |Γ) [1].
By using the Helmholtz decomposition, the velocity field can be written as the sum of the gradient of a
scalar potential φ and the curl of a vector potential ψ
v f = ∇φ + ∇ × ψ (1)
By substituting such velocity definition into both the incompressibility condition and the vorticity ex-
pression, we obtain
∇2 φ = 0; ξ = −∇2 ψ (∇.ψ = 0) (2)
By means of Biot-Savart law, the velocity induced at a point P by vorticity filament dl with a constant
circulation Γ is derived from (2)2
Z
Γ (rξ − rP )
∇×ψ = × dl (3)
4π l |rξ − rP |3
In the discrete form, the blade surface is divided into a number of panels. A vortex ring of constant
strength Γ is placed in the center of each panel. By solving equation (2)1 , the fluid velocity at the center
of each panel is approximated by
f
v f = v∞ + vd (4)
f
where vd is the velocity induced by vortex singularities in the center of a panel derived from (3) by
superposing the contribution of the four straight filaments of the vortex rings. The zero normal flow
boundary condition and the Kutta condition are then imposed to compute the unknowns circulations
(v f − vs ) . n = 0; ΓT.E = 0 ⇒ Γw = Γu − Γl (5)
where vs is the structure velocity including the rotation velocity and the deformation velocity. The com-
puted fluid velocity is written in Eulerian formulation, when computing pressure, this velocity should be
transferred from the Eulerian to Lagrangian formulations v̄ f = v f − vs . By using the Bernoulli equation,
the pressure of the panel can be computed at its center according to
1
p = ρ ( |v∞ |2 − |v̄ f |2 ) (6)
2

2.2. Structure model based upon 3D enhanced solid element with energy conserving
time-stepping scheme
By choosing the Biot strain (H = RT (I + ∇u + d) − I) and its conjugate stress T, we introduce the finite
rotation tensor R, which further imposes the path-dependency. The follower pressure pϕ, f , output of
aerodynamic model, is parametrized in the initial configuation by using the Nanson formula and the
corresponding variational equations are written in Lagrangian description [2]
R s R R R
V δu · ρs üs dV + V symm[δH] · symm[T] + skew[δH] · skew[T] dV − s
V δu . f dV − S p δu
s · p f J(x)F−T ndS =0
R
V {symm[R
s,T δds ] · symm[T] + skew[Rs,T δP · ds ] · skew[T] − δds · P}dV = 0
(7)
To solve these equation, we propose a time-stepping scheme [4] based on the mid point approximation
with two modifications concerning an algorithmic constitutive equations and the kinematic variables
approximation in order to ensure the energy conservation via equation (7)
symm[T]alg
n+ 1
= 12 C(symm[H]n+1 + symm[H]n ); skew[T]alg
n+ 1
= 12 γ(skew[H]n+1 + skew[H]n )
2 2
(8)
2 1
ün+ 1 = (u̇n+1 + u̇n )/∆t; u̇n+1 = −u̇n + ∆t un+1 ; Rn+ 12 = 2 (Rn+1 + Rn )
2

Thus, the increments of respectively kinetic energy K and potential energy Π in free motion correspond
to
R
Kn+1 − Kn = V δus · ρs üsn+ 1 dV
R 2 R (9)
Πn+1 − Πn = V symm[δH]n+ 1 · symm[T]n+ 1 + skew[δH]n+ 1 · skew[T]n+ 1 dV − V δus . f dV
2 2 2 2

This is particularly important for nonlinear dynamic analysis with presence of high frequencies, ensuring
the robustness of the computation over very long time interval.

32 2
2.3. Fluid-structure interaction with an iterative coupling algorithm
The structural and fluid computations are performed separately and later coupled through data exchange
at the fluid-structure interface. The coupling algorithm is based on CSS algorithm, extended by a sub-
cycling between the structure and fluid solvers until reaching convergence. The coupling algorithm steps
are as follows:
f ,(k=0)
1)- Initialize the pressure distribution for the first iteration (k = 0) → Pn+1 = Pnf
2)- Repeat the iterative procedure until the convergence, advancing at each iteration (from (k − 1) to
(k) (k)
(k)) (a) The variables un+1 and vn+1 , obtained from the structure solver, are transferred to the panel
f ,(k)
method code in order to update the mesh. (b) The normal pressures vector Pn+1 is computed by panel
method code. (c) The pull-back operation is performed to transfer the pressure intensity to the initial
f ,(k)
configuration for each panel, then the convergence pressure criterion is checked by means of ||P̃n+1 −
f ,(k−1) f ,(k)
P̃n+1 || ≤ tol1 . (d) Pn+1 is transferred to the FEM code to be considered as a follower external load. (e)
(k+1) (k)
The second convergence criterion is checked pertinent to the structure displacement ||un+1 − un+1 || ≤
tol2 . If the convergence is not reached, the kinetic variables are transferred again to the fluid code to
continue with the next iteration. As the structure formulation involves finite rotations that are highly
nonlinear and require special update procedure, the computation at iteration (k + 1) should be carried out
from the configuration at tn which has already reached the convergence state.
3)- Otherwise, we advance to the next time step.

3. Results and discussion


For numerical simulation, we consider the nonlinear dynamic response of the NREL offshore 5-MW
wind turbine with a 126 m rotor diameter. It is subject to a steady wind velocity 11.4m/s and an initial
rotation velocity 12rpm. In figure 1(a, b), the aerodynamics loads normal (Fn ) and tangential (Ft ) to the
rotor plane, computed by our aerodynamic code, are in good agreement with standard aeroelastic codes
(FAST, MIRAS). The normal force which causes the blade bending in the direction of the wind is much
higher than the tangential force. This is reflected in the time histories of the in-plane and out-of-plane tip
blade deflections. Compared to results reported in [3], results obtained by the iterative algorithm coupling
are considerately more accurate than those obtained by the one-iteration algorithm (see Figure 1(c, d)).
Thus, the sub-cycling between structure and fluid part is needed in order to improve the approximation
quality and avoid overestimate computed response, particularly since the proposed FSI algorithm is in
rapid convergence as shown in Figure 1( f ) (less than 10 iterations for ∆t = 0.05s). We can notice in
Figure 1(d, f ) that the the computed in-plane displacement and the twist displacement show the presence
of high frequency modes, superposed on top of low frequencies modes which is the ultimate cause of
stiff problems. By using the proposed energy conserving time-stepping scheme, we can ensure long-term
numerical stability as shown in Figure1(e). In fact, for a time step equal to ∆ = 0.01s, the computation
using the standard Newmark scheme can no longer converge for time exceeding 12s. However, the energy
conserving scheme ensures the convergence over long time interval. After validation of the proposed
model, we have herein a numerical tool that ensures a good compromise between low computational
cost and accurate results. It is suitable for testing new blades configurations in the preliminary design
process. For example, in order to avoid peak material stresses and fatigue failures, the blade can be
pre-bent at an angle θ and changes its orientation with respect to the wind direction so that it receives
the wind in the back side of the turbine (see Figure 2). This new configuration provides a more uniform
distribution of stress and reduces significantly the stress values. However, the blade rotation becomes
slower due to the decrease of the tangential force for this configuration which can affect the wind turbine
productivity.

References
[1] J. Katz and A. Plotkin Low Speed Aerodynamics (Cambridge Aerospace Series), Cambridge university press,
2001.

33 3
Problem statement (a) 8000 (b) 1000
FAST
Miras 800

Tangential force (N)


Normal force (N)
6000 Panel method
600
4000
400
FAST
2000
200 Miras
Panel method
0 0
10 20 30 40 50 60 10 20 30 40 50 60
(c) (d) radius (m) (f) radius (m)

2 2 0.12
Out-plane deflection (m)

0.1

In-plane deflection (m)


0 1 0.08

Twist angle (rad)


0.06
-2
0 0.04
-4 0.02
-1 0
-6 -0.02 r=19.95 m
r=32.25 m
Iterative coupling -2 Iterative coupling -0.04 r=44.55 m
-8 -0.06 r=52.75 m
One-iteration coupling One-iteration coupling r=63 m
-0.08
0 1 2 3 4 5 0 1 2 3 4 5 0 1 2 3 4 5
80 time (s) time (s) time (s)
(e) (f)
Tip displacement in x direction (m)

standard newmark
60 10
energy conserving scheme
40
8

Number of iterations
20

0 6

-20
4
-40
2
-60

-80 0
0 5 10 15 20 23 30 35 0 1 2 3 4 5
time (s) time (s)

Figure 1: NREL wind turbine simulations with the proposed FSI model

t=1.2 s
Out-plane deflection (m)

6
4 Conventional upwind
2 Inclined dowind
0.00E+07
1.00E+07 0
2.00E+07 -2
3.00E+07 -4
4.00E+07 -6
5.00E+07 -8
6.00E+07 0 1 2 3 4 5 6
time (s)
7.00E+07 80
Tip displacement in

8.00E+07 60 Conventional upwind


9.00E+07 40 Inclined dowind
x direction (m)

1.00E+08 20
1.00E+10 0
-20
-40
Conventional Pre-bent -60
upwind downwind -80
0 1 2 3 4 5 6 Conventional upwind Inclined downwind
Von Mises stress time (s)

Figure 2: Comparaison between conventional upwind blade and inclined downwind blade

[2] A. Boujelben and A. Ibrahimbegovic Finite-strain three-dimensional solids with rotational degrees of freedom:
non-linear statics and dynamics, Advanced Modeling and Simulation in Engineering Sciences, 646-654, 2016.
[3] W. Musial G. Scott J. Jonkman and S. Butterfield Definition of a 5-MW reference wind turbine for offshore
system development,Technical Report NREL/TP-500-38060 National Renewable Energy Laboratory, 2009.
[4] A. Boujelben and A. Ibrahimbegovic Conserving and decaying energy for finite-strain three-dimensional
solids with rotational degrees of freedom in nonlinear dynamics, Comptes rendus Mecanique, 571-580, 2018.

34 4
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

STRAIN LOCALIZATIONS IN PERIDYNAMIC BARS WITH NON-CONVEX POTENTIAL

Adair R. Aguiar1, Gianni F. Royer-Carfagni2, Alan B. Seitenfuss3


1
Department of Structural Engineering, São Carlos School of Engineering, University of São Paulo,
Brazil
2
Department of Engineering and Architecture, University of Parma, Italy
3
Department of Structural Engineering, São Carlos School of Engineering, University of São Paulo,
Brazil

The static states of a peridynamic nonlinear elastic bar of finite length in a hard device are
investigated. The nonlocal character of the model requires that the edge conditions be defined on a
boundary layer of the same length of the horizon, affecting the solution in the interior. The approach is
variational and is based on the assumption that the energy is a function of the relative displacement
between particles within the horizon, not weighted by the inverse of their relative distance as in Dayal
and Bhattacharya [1]. This implies a weak interaction between neighboring points that allows for
discontinuities in the displacement field. In the simplest linear elastic case, the jumps depend upon the
length of the horizon and are concentrated at the bar ends. Here, we consider a more complex behavior
by assuming a non-monotone constitutive relation associated with a non-convex strain energy density,
which is analogous to assumptions made by Ericksen [2] in classical nonlinear elasticity. At a micro-
scale of the same order of the length of the horizon, the displacement field is characterized by the
orderly formation of undulations and discontinuities, while at the macroscopic level the stress vs.
elongation graph is a sequence of strain hardening and softening branches. The equilibrium path,
found numerically with a pseudo-arc-length continuation method, becomes unstable above a certain
elongation. Convexification of the strain energy density provides solutions that are very different from
this because, unlike the Ericksen's model, re-arrangements of material phases along the bar is
prevented by the nonlocal nature of peridynamics. Results of this work were published in Aguiar,
Royer-Carfagni and Seitenfuss [3]. They could be used to interpret the complex phenomena of
localization of plastic strain in slip bands experimentally observed in ductile metallic bars.

References

[1] Dayal, K., Bhattacharya, K.. Kinetics of phase transformations in the peri- dynamic formulation of
continuum mechanics. J. Mech. Phys. Solids 54 (9), 1811-1842, 2006.
[2] Ericksen, J.L.. Equilibrium of bars. Journal of Elasticity, 5(3-4):191-201, 1975.
[3] Aguiar, A.R., Royer-Carfagni, G.F., Seitenfuss, A.B.. Wiggly strain localizations in peridynamic
bars with non-convex potential. International Journal of Solids and Structures 138 (1), 1-12, 2018.

Keywords: Peridynamics; Nonlocal theory; Non-convex energy; Nonlinear elasticity; Phase


transitions.

35
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

EVALUATION OF TEMPERATURE AND DAMAGE OVER STRUCTURAL NATURAL


FREQUENCIES

Daniel Soares1, Alexandre Cury1


1
University of Juiz de Fora, dancardososoares@gmail.com
2
University of Juiz de Fora, alexandre.cury@engenharia.ufjf.br

Structural modal parameters i.e. natural frequencies, damping ratios and mode shapes are dynamic
features obtained either by measuring the vibration responses of a structure or by means of finite
elements models. Over the past two decades, modal parameters have been used to detect damage in
structures by observing its variations over time. However, such variations can also be caused by
environmental factors such as humidity, wind and, more importantly, temperature. In so doing, the use
of modal parameters as damage indicators can be seriously compromised if these effects are not
properly tackled. Many researchers around the world have found numerous methods to mitigate the
influence of such environmental factors from modal parameters and many advanced damage indicators
have been developed and proposed to improve the reliability of structural health monitoring. In this
paper, several vibration tests are performed on a simply supported steel beam subjected to different
damage scenarios and temperature conditions, aiming to describe the variation in modal parameters
due to temperature changes. Moreover, four methodologies are proposed to identify damage, such as
Robust Regression, Principal Component Analysis and Support Vector Regression. Results show a
slightly linear decrease in the modal parameters due to temperature increase, although it is not possible
to establish an empirical equation to describe this tendency.

Keywords: SHM; damage detection; temperature effects; modal identification

1. Introduction and Motivation

Structural Health Monitoring (SHM) is based on the premise that damage cause changes in the
structure’s physical properties (stiffness, mass and damping). Over the last decades, modal parameters
and other dynamic features obtained from vibration tests have been used to assess damage, since they
are functions of such structural properties. Research in vibration-based damage identification has been
rapidly expanding over the last few years, especially in applications involving bridges and buildings.
However, modal parameters are also sensitive to environmental factors such as humidity, wind and
temperature. The latter is especially responsible for modal variations that often are higher than those
caused by structural damage. This condition might compromise the reliability of SHM techniques, by
either masking the presence of damage or giving false positive alarms. To overcome this problem,
many researchers have studied the underlying relationships between modal parameters, environmental
factors and structural damage. Such studies were performed by means of numerical simulations and
experimental tests in laboratory or in situ, as described in references Peeters et al (2001), Sohn et al.
(1999), Meruane and Heylen (2012), Nguyen et al (2014), Wei (2015), Morales et al. (2018).

36
2. Materials and Methods

To provide a better understanding over such a phenomenon, this paper proposes an experimental study
of the variation of modal parameter estimates due to temperature and damage. This study is performed
on a simply supported steel beam subjected to 20 different temperature setups and 6 structural damage
scenarios, yielding 720 tests. To assess these three damage scenarios under the effect of temperature
variation, four statistical damage detection techniques are proposed: linear and nonlinear regression,
Principal Component Analysis (PCA) and Support Vector Regression (SVR). Results show that
among the four proposed techniques, only one is reliable for structural damage detection.
The methodology proposed in this paper consists in placing a 25mm x 6mm x 1500mm steel beam
into an electric oven and heating it up progressively while performing vibration tests under different
temperatures and damage conditions. Figure 1 depicts the general scheme of the experiment and the
Figure 2 shows the flowchart of the proposed methodology.
The minimum temperature for all tests was set to 18°C, since it was the minimum temperature
reachable using an air conditioning system. The maximum temperature for all tests was 56°C. This
maximum temperature was set to: i) do not damage the insulation of the accelerometer’s coaxial cables
(for which safe operational temperature must be under 70°C); ii) obtain an adequate sampling of
readings (20 samples obtained at intervals of 2°C); iii) obtain a temperature gradient of 38°C which is
wider than seasonal temperature variation in many tropical countries.

Figure 1. General scheme of the experiment

Figure 2. Methodology’s flowchart

2.1 Damage scenarios

Artificial damage scenarios were simulated by cutting the beam’s cross section to change local
stiffness and cause a reduction in the natural frequencies. Figure 3 details three damage scenarios. The
first one reduces the beam’s cross section area by 16.67% whereas the second by 33.34%. The fourth,
fifth and sixth damage scenarios consisted in adding a lumped mass of 10g, 20g and 50g at 375mm
from the left support, respectively.

37
Figure 3. Beam’s damage scenarios simulated through electrical saw cuts.

3. Preliminary results and discussion

Figures 4 and 5 present the results obtained by the four proposed statistical damage detection
techniques applied to 720 vibration tests performed on the beam (3 tests at 20 temperatures under 3
damage scenarios). The first results correspond to a linear regression model, followed by nonlinear
(second order) regression. Then, PCA are performed. Finally, one investigates the results obtained by
Support Vector Regression.

Figure 4. Results obtained using linear (left) and nonlinear (right) regression models.

Figure 5. Results obtained using PCA (left) and SVR (right) models.

38
4. Concluding remarks on current and future research

In general, both linear and nonlinear regressions showed a rather clear relationship between damage,
temperature and natural frequencies, but in some cases, the models did not fit well the entire dataset.
Thus, this method partially identified the damage scenarios and showed a limited relationship between
those physical quantities.
Support Vector Regression provided better results, as the damage levels could be more easily
identified. However, further research is necessary using PCA, since the relationship between natural
frequencies, temperature and damage could not be properly established.

Acknowledgments

The authors are grateful for the financial support of CNPq (National Council for Scientific and
Technological Development), FAPEMIG (Foundation of Support Research of the State of Minas
Gerais), CAPES (Coordination for the Improvement of Higher Education Personnel) and UFJF
(Federal University of Juiz de Fora).

References

[1] Peeters, B., De Roeck, G., (2001). “One-year monitoring of the Z-24 Bridge: Environmental
effects vs Damage events”. Earthquake engineering and structural dynamics, Volumen 30, pp. 149-
171.
[2] Sohn, et al., (1999). “An experimental study of temperature effect on modal parameters of the
Alamosa Canyon Bridge”. Earthquake engineering and structural dynamics, Issue 28, pp. 879-897.
[3] Meruane, V., Heylen, W., (2012). “Structural damage assessment under varying temperature
conditions”. Structural health monitoring, may, Issue 11, pp. 345-357.
[4] Nguyen, V. H., Mahowald, J., Golinval, J. C., Maas, S., (2014). “Damage detection in bridge
structures including environmental effects”. Proceedings of the 9th International Conference on
Structural Dynamics, EURODYN.
[5] Wei J.J., Lv Z.R (2015). “Structural damage detection including the temperature difference based
on response sensitivity analysis”. Structural engineering and mechanics, An Int'l Journal Vol. 53 No. 2
[6] F. Morales, A. Cury, “Analysis of thermal and damage effects over structural modal parameters”,
Structural Engineering and Mechanics 65 (1), 43-51, 2018.

39
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

MODELING WHEEL-RAIL CONTACT INTERACTION AND VEHICLE DYNAMICS: OVERVIEW


OF ONGOING DEVELOPMENTS AND NEW ACHIEVEMENTS

Alfredo Gay Neto 1, Thiago Felipe Medeiros Pereira 2


1
Polytechnic School at University of São Paulo, alfredo.gay@usp.br
2
Polytechnic School at University of São Paulo, fmp.thiago@gmail.com

A research cooperation on wheel-rail contact mechanics was established in 2016 between Vale
Company and University of São Paulo. The objective is to investigate, by employing numerical
simulations, the wheel-rail contact interaction problem, focusing on heavy haul applications. On this
context, the present work provides an overview of last developments, such as shows some new
achievements and practical case studies.
Wheel-rail contact interaction is a complex phenomenon. The geometry of wheels is usually based
on a conical surface, aiming at providing a self-guidance of the wheelset while experiencing rolling on
rails. This provides unique abilities for curving and for dealing with geometrical imperfections on the
track. On the other hand, lateral oscillations of the wheelset always take place as a natural
consequence of the geometry of wheels and rails. With that, contact location usually changes all the
time during the vehicle movement. Contact actions depend on positioning of the wheelset on the track,
varying along its length according to curve radius, super-elevation, local geometry of wheels and rails,
such as operational conditions, as the traveling speed. Thus, when concerning about wheel-rail
contact, it is impossible to treat it separately from the dynamics of the whole system, which rules the
vehicle behavior. With that, computational modeling of such complex system may be viewed as a
challenge.
Computational simulation of railway systems is usually performed by multibody software, based
on rigid bodies dynamics. The wheel-rail interaction may be included as unilateral constraints, by
assuming pointwise or distributed actions. On such context, special constitutive laws may be
established to treat contact between rigid bodies, incorporating local flexibility for the purpose of
establishing contact forces properly. This is usually done by establishing creepage coefficients,
employed in contact constitutive laws.
Present work employs a distinct philosophy from usual simulators of railway vehicle dynamics. As
proposed by [1], we considered rigid bodies as special implementations within a finite element
environment. With that, one would be able to combine rigid and flexible parts, according to modeling
interests. Particularly, geometrically-exact nonlinear finite element beams and shells may be used to
compound a complex structural system. All implementations were done in Giraffe solver [2]. A way to
incorporate the rigid body implementation was proposed in [3]. To time-integrate the system, we
employed the Newmark method, adapted for dealing with finite rotations, as proposed by [4].
The particular novelty of current developments is in incorporating the master-master contact
formulation, initially proposed for beam-to-beam contact in [5]-[6], for considering contact involving
rigid and flexible bodies. Particularly for the wheel-rail contact, this is welcome in order to represent
the interaction between a wheel and a flexible track system, composed by beams, springs and
dashpots. We expect advantages on employing the master-master contact technique, since it assumes
pointwise contact interaction with no pre-defined location on material points of contacting bodies.
Contrary, it involves a particular searching procedure performed along model evolution, aiming at

40
establishing moving contact actions in a convenient way. Thus, scenarios of rolling, may be naturally
considered. In this context, the possibility of using a contact formulation that automatically handles
such changes in contact location along time is welcome.
Another need is to establish special constraints, such as hinge joints, cardan joints and others. With
that, one may set-up a system representing the primary and secondary suspensions of a boogie
employed in heavy-haul wagons. Ideas of special constraints proposed in [7] were extended in present
work, by establishing a new translational joint, practical for establishing the suspension system.
The present work also provides some particular cases studies involving dynamic behavior of a
bogie under a condition of movement along a track with a sudden change in track stiffness. This may
represent a transition between a standard track and a tunnel or a bridge, were one may find distinct
equivalent stiffness. The bogie is modeled considering a collection of rigid bodies and joints. Primary
and secondary suspensions are considered, such as a point masses, representing cargo.
Ongoing developments aim at incorporating more robust methods to solve local contact detection
problems. This is a real need for the master-master contact formulation. Furthermore, current available
constitutive equations for contact employ only linear penalties (with Coulomb saturation on tangential
direction). No advance creepage models were employed yet, which are also a need for reasonable
physical results on realistic wheel-rail contact.
At the conference, an overview presentation of the project will be offered, such as discussions on
modeling issues and challenges of ongoing works. Practical examples will also be shown.

Acknowledgements
The authors acknowledge Vale S.A. for the support through Wheel-Rail Chair project. The first
author acknowledges CNPq (Conselho Nacional de Desenvolvimento Científico e Tecnológico) under
the Grant 308190/2015-7.

References

[1] Cardona, M.; Geradin; Doan, D. B. Rigid and flexible joint modelling in multibody dynamics using
finite elements. Comp. Methods in App. Mech. Eng., v. 89, p. 395-418, 1991.
[2] Gay Neto, A. Giraffe User’s Manual - Generic Interface Readily Accessible for Finite Elements,
2018. Available at: <http://sites.poli.usp.br/p/alfredo.gay/>.
[3] Campos, P.R.R., Gay Neto, A. Rigid body formulation in a finite element context with contact
interaction, Comp. Mechanics, online, 2018.
[4] Ibrahimbegovic, A.; Mamouri, S. Energy conserving/decaying implicit time-stepping scheme for
nonlinear dynamics of three-dimensional beams undergoing finite rotations. Comp. Methods in App.
Mech. Eng., v. 191, p. 4241-4258, 2002.
[5] Gay Neto, A.; Pimenta, P. M.; Wriggers, P. A Master-surface to Master-surface Formulation for
Beam to Beam Contact. Part I: Frictionless Interaction. Comput. Methods Appl. Mech. Engrg., v.
303, p. 400-429, 2016.
[6] Gay Neto, A.; Pimenta, P. M.; Wriggers, P. A Master-surface to Master-surface Formulation for
Beam to Beam Contact. Part II: Frictional Interaction. Comput. Methods Appl. Mech. Engrg., v. 319,
p. 146-174, 2017.
[7] Gay Neto, A. Simulation of Mechanisms Modeled by Geometrically-Exact Beams using Rodrigues
Rotation Parameters. Comp. Mechanics, v. 59 (3), p. 459-481, 2017.

41
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A NUMERICAL ASSESSMENT OF STIFFNESS ASSOCIATED WITH A SINGLE MOOR -


ING LINE : T HE EFFECT OF SEA CURRENT AND APPLICATION FOR A F LOATING
O FFSHORE W IND T URBINE

Giovanni A. Amaral 1 , Estevan C. IsaaK 2 , Alfredo G. Neto 2 , Guilherme R. Franzini 1

1 Offshore Mechanics Laboratory, Escola Politécnica, University of São Paulo


2 Computational Mechanics Laboratory, Escola Politécnica, University of São Paulo

Introduction
Floating Offshore Wind Turbines (FOWTs) increase the number of possible sites for wind energy farms.
It is remarkable that this type of solution allows the installation in regions with water depths superior to
100 m. As FOWTs are still in their early days, a series of numerical and experimental studies on this has
been carried out in the last years by different research groups. We emphasize that, even though their high
cost, experimental investigations on the different aspects of the FOWTs are of great importance. In this
context, numerical simulations are commonly adopted in the early design phases.
As in all types of floating units, the mooring system design is an important task in the FOWT project.
The equivalent stiffness due to the mooring system has to be specified in order to deal with a series of
static and dynamic criteria. Essentially, the mooring stiffness has to be able to either keep the floating
unit offset within operational limits or to detune its natural frequencies from those arisen from the waves
and the sea current loads.
The definition of equivalent stiffness of the mooring system may be a complex task, since it depends
on each mooring line configuration, its physical properties, sea current incidence and sometimes on the
contact between the mooring line and the seabed. Reference [1] presented an analytic methodology
to quantify the stiffness due to mooring line when the floating unit experiences surge, sway and yaw
movements..
Nowadays, the mooring systems of the FOWTs farms installed mainly employ slack chain catenary
mooring lines. In this context, the contribution of a particular line to the overall stiffness can be evalu-
ated either analytically or numerically. Each line could be seen as a nonlinear spring, conteracting the
floating unit offsets by applying restoring forces. Although being a classic-well-solved problem, the an-
alytical catenary approach has some limitations. For example, it is not an easy task to take into account
hydrodynamic drag forces from ocean current analytically; see [2] and [3]. This confirms that the use of
some numerical tools for obtaining the force-displacement relation are worth.
One of the main issues of the numerical approach to model a catenary-laying mooring line or flexible
risers systems is the difficulty of modeling high-flexible structures. As FOWT mooring lines are usually
composed by steel chains, their very low (or even null) bending stiffness could lead to very strong ge-
ometric non-linearities. Consequently, there is a challenge in defining the initial static; see [4]. Many
methods have been used for calculating the initial static shape of very flexible structures. An initial con-
trolled geometry near to the target equilibrium geometry is suggested in [5]. In [6], aiming to solve a
similar problem for a riser, the authors proposed a strategy based on imposing a pretension on it in order
to increase its geometric stiffness, improving the convergence.

42 1
Figure 1: Load-steps sequences for achieving the final mooring line configuration.

This paper investigates the influence of the sea current in the relationship between force on the top of
a mooring line and the corresponding displacement. Focus is placed on the stiffness on the horizontal
plane. This force-displacement curve can be obtained by imposing quasi-static motions on the floating
unit in the horizontal direction of the plane of the catenary. Firstly, the case without sea current is
compared with the one analytically obtained. Then, the effects of the sea current acting on the mooring
line are investigated. In order to exemplify the methodology, the international benchmark of the Offshore
Code Comparison Collaboration Continuation - OC4 - is taken as a case study.

Methodology
We employ a geometrically-exact beam model to numerically obtain the force-displacement relation-
ship. The environmental loads considered in the model are the line weight, the buoyancy force, the sea
current hydrodynamic forces and the contact between the line and the seabed. A complete description
of the static loading applied to a exactly beam model can be found at [6]. All models are studied using
the GIRAFFE finite element solver. This is an in-house software, coded using C++ language and proper
for receiving new finite elements and contact models. Further information regarding GIRAFFE can be
found in [7].
The mooring line is taken completely immersed in the sea water. The combination of the effects of
the structure self-weight and the buoyancy results in a "effective weight" force. More details about the
effective weight and effective force may be found in [8]. In turn, the hydrodynamic effect caused by the
sea current is taken into account by using the Morison’s model; see [9]. In this model, the drag force
is proportional to the square of the sea current velocity. The contact between the mooring line and the
seabed is a nonlinear boundary condition. Since the portion of the line that lays on the seabed is not
known and the position of the touchdown point (TDP) could vary during the simulation, it is a crucial
problem. Techniques to deal with this problem in statics and dynamics of geometrically-exact beam
models can be found in [10] and [11], respectively.
As mentioned before, one of the contributions of this paper is the proposal of a technique to deal with
finite element modeling of very flexible structures. The main goal of this methodology is to make the
structure increase its geometric stiffness in order to speed-up the convergence. The strategy is performed
in some load-steps. At the beginning, the mooring line is supposed unstressed laying on the seabed. At
this stage, the line is meshed and its extremities are constrained to be fixed. The gravity action starts to
be considered at this step. Then, prescribed displacements at the line nodes are established, making the
line assume an initial guess position. This nodal displacement was preliminary determined by using the
analytic solution of the catenary problem. The second load-step consists in release the displacements
of almost every mesh node, except by the first and the last ones. It is important to remark that these
two nodes are related, respectively, to the anchor and the fairlead points. As a result of the release,
the mooring line seeks for its equilibrium configuration. Once this guess is not far from the real static
position, the convergence at this load-step comes easily. Fig. 1 sketches the proposed procedure.
Once the static equilibrium configuration is reached, the actual study for evaluating the line stiffness
bahavior can start. It is important to highlight that any kind of analyses may take place here: statics,
quasi-statics or dynamics. Firstly, the sea current is imposed by giving an incidence angle α and a profile

43 2
(a) Front view. (b) Superior view.

Figure 2: Sea current incidence angle α at the mooring line.

of fluid speed. Then, a quasi-static analysis is performed, by imposing motions at the fairlead node at
the axial direction of the line. The horizontal force-displacement curve is obtained directly from this
approach.
The main goal of this work is to evaluate the effects of the sea current on the contribution of a single
mooring line to the equivalent stiffness. Although the friction force caused by the contact between the
line and the seabed could be significant, this is not evaluated at this paper. The sea current is herein
defined by its intensity and incidence angle α. Fig. 2 illustrates of the mooring line, indicating the
incidence the sea current.

Preliminary Results
As already mentioned, the international benchmark of the OC4 is taken as a case study. Tab. 1 presents
the mooring line proprieties, extracted from [12].

Table 1: Mooring line properties.

Depth to anchors below the free surface 200 m


Depth to fairleads below the free surface 14 m
Radius measured from the platform center to the anchor 837.6 m
Radius measured from the platform center to the fairleads 40.9 m
Unstretched mooring line length 835.5 m
Mooring line diameter 0.0766 m
Mass per unit length 113.35 kg/m
Axial stiffness 7.536E+8 N
Tangential drag coefficient 0.37
Normal drag coefficient 2.40

Aiming at verifing the proposed methodology, an initial test without the sea current is carried out. The
outcome for this simulation is the horizontal force-displacement. The comparison with the results for
different analytical models is showed in Fig. 3a. Two analytical models are proposed: the classic non-
extensible catenary line and an elastic cable line, as proposed by [2]. It is possible to note that the FEM
model very well agrees with the analytical solution for extensible cables.
As a preliminary test for the influence of the hydrodynamic forces for the horizontal force-displacement
curve, it has been taken into account two scenarios of sea current, characterized by incidence angles
α = 0o and α = 180o , both with a sea current speed of 5 m/s. The results are compared to the case
without sea current and showed in Fig. 3b. Even though the significant sea current velocity, there is no
significant difference between the conditions with and without sea current. This is associated with the
heavy mooring line herein investigated. The fact that sea currents direction took into account lies on the
plane of the mooring line could be another reason. Further works include the study of lighter and mixed
mooring lines and other sea currents speeds and incidence angles.

44 3
(a) Comparison between numerical and analytical mod- (b) Numerical results under sea current.
els. Without sea current.

Figure 3: Fairlead forces vs. Fairlead-anchor horizontal distance.

Acknowledgments
The first and the second authors are grateful to the Office Naval Research (ONR) for the financial
support. The third and the fourth authors acknowledge the Brazilian National Research Council (CNPq)
for the grants 308190/2015-7 and 310595/2015-0.

References
[1] C. P. Pesce, G. A. Amaral, and G. R. Franzini. Mooring system stiffness: A general analitycal formulation
with an application to floating offshore wind turbaines. In Proceeding of the ASME 2018 1st International
Offshore Wind Technical Conference [Submited]. American Society of Mechanical Engineers, 2018.
[2] O. M. Faltinsen. Sea Loads on ships and offshore structures. Cambridge Ocean Technology Series. Cam-
bridge university press, 1 edition, 1993.
[3] M. Hall, B. Buckham, and C. Crawford. Evaluating the importance of mooring line model fidelity in floating
offshore wind turbine simulations. Wind Energy, 17(12):1835–1853, 2013.
[4] R. L. Webster. On the static analysis of structures with strong geometric nonlinearity. Computers & Struc-
tures, 11(1-2):137–145, 1980.
[5] R. Antonutti, C. Peyrard, A. Incecik, D. Ingram, and L. Johanning. Dynamic mooring simulation with
Code_Aster with application to a floating wind turbine. Ocean Engineering, 151:366 – 377, 2017.
[6] A. Gay Neto, C. A.. Martins, and P. M.. Pimenta. Static analysis of offshore risers with a geometrically-exact
3d beam model subjected to unilateral contact. Computational Mechanics, 53(1):125–145, 2013.
[7] A. Gay Neto. Giraffe user’s manual v 1.0.200, 2017.
[8] A. G. Neto, P. M. Pimenta, and C. A. Martins. Hydrostatic pressure load in pipes modeled using beam finite
elements: theoretical discussions and applications. Journal of Engineering Mechanics, 143(4):04017003,
2017.
[9] J.R. Morison, J.W. Johnson, and S.A. Schaaf. The force exerted by surface waves on piles. Journal of
Petroleum Technology, 2(5):149–154, 1950.
[10] A. Gay Neto, E. R. Malta, and P.M. Pimenta. Catenary riser sliding and rolling on seabed during induced
lateral movement. Marine Structures, 41:223–243, 2015.
[11] A. Gay Neto. Dynamics of offshore risers using a geometrically-exact beam model with hydrodynamic loads
and contact with the seabed. Engineering Structures, 125:438–454, 2016.
[12] A. Robertson, J. Jonkman, F. Vorpahl, W. Popko, J. Qvist, L. Frøyd, X. Chen, J. Azcona, E. Uzunoglu,
C. Guedes Soares, C. Luan, H. Yutong, F. Pengcheng, A. Yde, T. Larsen, J. Nichols, R. Buils, L. Lei, T. An-
ders Nygard, D. Manolas, A. Heege, S. Ringdalen Vatne, H. Ormberg, T. Duarte, C. Godreau, H. Fabricius
Hansen, A. Wedel Nielsen, H. Riber, C. Le Cunff, R. Abele, F. Beyer, A. Yamaguchi, K. Jin Jung, H. Shin,
W. Shi, H. Park, M. Alves, and M. Guérinel. Offshore code comparison collaboration continuation within
iea wind task 30: Phase ii results regarding a floating semisubmersible wind system. In Proceedings of the
ASME 2014 33rd International Conference on Ocean, Offshore and Arctic Engineering. American Society
of Mechanical Engineers, 2014.

45 4
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

I MPROVING P ERFORMANCE OF F INITE D IFFERENCE 3D H ETEROGENEOUS


ACOUSTIC WAVE E QUATION S IMULATIONS
Carlos H. S. Barbosa1 , Schirley C. Jorge1 , Raphael F. Vilela1 , Luciano Leite1 , José J. Camata2 , Alvaro
L. G. A. Coutinho1
1 COPPE/Federal University of Rio de Janeiro, Brazil,
2 Computer Science, Federal University of Juiz de Fora.

Abstract: Geophysical imaging faces nowadays new challenges related to 3D data acquisition, which
means that we need to simulate full 3D volumes. Therefore, we present a hybrid MPI/OpenMP ap-
proach for modeling the 3D heterogeneous acoustic wave equation. Test cases provided by the High-
Performance Computing for Energy project are solved using our optimized numerical code. Results
show that a hybrid MPI/OpenMP strategy overlapping computation and communication can be very
efficient in standard multi-core machines.

1. Introduction
Finite difference (FD) algorithms are the most straightforward approach to simulate 3D waves. The
geophysical imaging industry has relied for years on highly specialized FD codes for their modeling
and migration algorithms. Nevertheless, despite their success, nowadays geophysical imaging faces
new challenges. First of all, 3D data acquisition are now commonplace, which means that we need to
model full 3D volumes. Furthermore, reservoirs are found in deeper and more geologically complex
areas, which requires more computing power and tweaks to the physical approximations used to model
waves. Also, computer architectures have changed dramatically over the years, mostly driven by the
commodity market. New computational architectures and parallel paradigms imply that codes require
constant maintenance to beat obsolescence. Today, FD algorithms are still as popular as ever, and the
applications where they are used have increased significantly. In this work, fast and effective optimization
and porting strategies are shown to boost the performance of a classical extrapolation scheme. A standard
eighth order in space and second order in time 3D acoustic FD code, parallelized with MPI/OpenMP and
running on Intel general-purpose CPUs is analyzed and optimized. Results for the seismic imaging
benchmark released by the High Performance Computing for Energy project are shown to demonstrate
the efficiency of our approach.

2. Wave Equation Formulation


A simple way to derive the heterogeneous acoustic wave equations, regarding particle velocity and pres-
sure, is using the Newton’s Law of momentum conservation and the continuity equation [1]. Thus, the
wave propagation in a 3D heterogeneous medium can be described as the following first-order system of
equations:

∂v(r,t)
ρ(r) + ∇p(r,t) = f(r,t), (1)
∂t
1 ∂p(r,t) ∂iv (r,t)
+ ∇ · v(r,t) = ,
κ(r) ∂t ∂t

46 1
where, p is the pressure, v = (vx , vy , vz ) the velocity field, r = (rx , ry , rz ) the position vector, t the time. ρ
and κ are functions of the position and define the density and the acoustic bulk modulus of the medium,
respectively. The density of external body force is represented by the vector f = ( fx , fy , fz ), and the
source, iv , is the density of volume injection, such as an air gun, for instance. Finally, ∂/ ∂t is the time
derivative, and the symbol ∇ is the vector differential operator.
The first-order system of equations 1 describes how the velocity and the acoustic pressure change in a
medium with properties κ and ρ submitted to external perturbations iv and f. The numerical discretization
used to solve the coupled system of equations 1 is based on the acoustic Standard Staggered Grid scheme
(SSG) as proposed by Virieux [2]. However, Virieux’s original formulation is second-order accurate in
time and space. Here we apply a high order discretization, such as an eighth-order accurate in space, to
reach better accuracy. Thus, the discretized coupled equations are given by:

n+ 1 n− 1 ∆t
vi+ 12, j,k = vi+ 12, j,k − bi+ 1 , j,k (Dpx ) ,
2 2 2 h
n+ 1 n− 1 ∆t
vi, j+2 1 ,k = vi, j+2 1 ,k − bi, j+ 1 ,k (Dpy ) , (2)
2 2 2 h
n+ 12 n− 12 ∆t
vi, j,k+ 1 = vi, j,k+ 1 − bi, j,k+ 1 (Dpz ) ,
2 2 2 h

∆t
pn+1 n
i, j,k = pi, j,k − κi, j,k (Dvx + Dvy + Dvz ),
h
where the operators Dpx , Dpy , Dpz , Dvx , Dvy , Dvz are:

Dpx = pni+1, j,k − pni, j,k + pni+2, j,k − pni−1, j,k + pni+3, j,k − pni−2, j,k + pni+4, j,k − pni−3, j,k ,
Dpy = pni, j+1,k − pni, j,k + pni, j+2,k − pni, j−1,k + pni, j+3,k − pni, j−2,k + pni, j+4,k − pni, j−3,k ,
Dpz = pni, j,k+1 − pni, j,k + pni, j,k+2 − pni, j,k−1 + pni, j,k+3 − pni, j,k−2 + pni, j,k+4 − pni, j,k−3 ,

n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1
Dvx = vi+ 12, j,k − vi− 12, j,k + vi+ 32, j,k − vi− 32, j,k + vi+ 52, j,k − vi− 52, j,k + vi+ 72, j,k − vi− 72, j,k ,
2 2 2 2 2 2 2 2
n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1
Dvy = vi, j+2 1 ,k − vi, j−2 1 ,k + vi, j+2 3 ,k − vi, j−2 3 ,k + vi, j+2 5 ,k − vi, j−2 5 ,k + vi, j+2 7 ,k − vi, j−2 7 ,k ,
2 2 2 2 2 2 2 2
n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1 n+ 1
Dvz = vi, j,k+
2
1 −v
2
i, j,k− 1
+ vi, j,k+
2
3 −v
2
i, j,k− 3
+ vi, j,k+
2
5 −v
2
i, j,k− 5
+ vi, j,k+
2
7 −v
2
i, j,k− 7
.
2 2 2 2 2 2 2 2

The indexes i, j, and k represent the directions (x, y, z), n is the temporal step, ∆t is the temporal sampling
rate, and h is the grid size. The velocity is given by v, p is the pressure, and b is the buoyancy (1/ρ).

3. Hybrid MPI/OpenMP Parallelism and Optimizations


The wave propagation kernel concentrates most of the program execution time (≈ 98% of total). The first
modification to the serial code use OpenMP directives. In addition to this, current processors allow oper-
ating more than one floating-point operation per loop iteration, which is known as a Single-Instruction-
Multiple-Data (SIMD) model. In the SIMD model, instead of applying an instruction multiple times to
each chunk of a vector, it employs just one single instruction to a whole vector (e.g., float, or, at least,
a portion more extensive than a single unit (a single float). In this work, the AVX2 (Advanced Vector
Extensions) technology, with 256 bits [3], is used for SIMD computations. The code contains preproces-
sor directives and uses memory alignment allocation to ensure vectorization. The processor cores pinned
OpenMP threads using the Intel environment variable to control thread affinity. The code uses first touch
approach to reduce NUMA imbalance.
For large 3D discretizations, domain decomposition is mandatory in multi-core architectures. An
MPI communication strategy is also applied to construct our Hybrid MPI/OpenMP parallel solution. It
uses non-blocking send/receiver MPI routines that allows overlap communication and computation. The

47 2
computational domain is a Cartesian grid, and it can be partitioned among the processes. Thus, each
MPI process can get all information about its domain/subdomain and neighbors relationship without any
message exchange. The subdomain stencil computation using non-blocking communication is performed
in four stages:(i) gather its subdomain boundary data and sends to appropriate neighbors; (ii) compute
the stencil only in regions that do not depend on data from neighboring processes; (iii) verify that it has
received data from some neighbor and, if true, copy the data to a ghost area; (iv) compute stencil in halo
region.
However, this scheme can result in load imbalance and performance degradation because the imple-
mentation of the send call requires the corresponding receiver to be posted on the target before initializing
the data transfer. Thus, data transfer occurs during the MPI Waitall() call even when there is enough com-
putation to overlap. To overcome this issue, we have implemented a scheme where a dedicated thread
is used to improve asynchronous communication progress. As a consequence, a new nested OpenMP
parallel region is created to compute the finite-difference stencil.

4. Results and Discussions


The first result presented is related to the scalability analysis of the discretized homogeneous acoustic
wave equation with 4096×4096×4096 grid points for different core numbers as shown in Table 1. Using
code profiling tools, we can get insights into the effectiveness of the domain decomposition strategy, use
of thread parallelism and the computational functional unit cores. We use the Tuning and Analysis
Utilities (TAU) [4] linked to PAPI to get hardware events (for instance, L1/L2 cache misses). OMPT tool
provides runtime state tracking, which enables a sampling-based performance tool to understand what
an OpenMP application thread is doing. These tools allow us to discover the most appropriate OpenMP
SCHEDULE strategy and also to perform stencil optimizations, which increased the performance of
the seismic modeling code. Besides, we have allocated one MPI process to each compute node and
use 16 OpenMP threads per node. Tests were carried out on Stampede2, a cluster with 4,200 Knights
Landing (KNL) nodes, the second generation of processors based on Intel’s Many Integrated Core (MIC)
architecture and 1,736 Intel Xeon Skylake nodes. Thus, the scalability analysis shows that our scheme
can reach 80% of efficiency up to 4096 cores.

Table 1: Acoustic wave propagation - scalability analysis - grid 4096 × 4096 × 4096, 500 time steps.
MPI Process Process Distribution Threads CPU Time Speed-up
64 4×4×4 1024 773.88 1.00
128 4×4×8 2048 494.83 1.56
256 4×8×8 4096 234.77 3.30

We also experiment our optimized code using the flat acoustic test at 20 Hz maximum frequency
[5] (codenamed AF-UNIT-20Hz) provided by the High-Performance Computing for Energy (HPC4E)
project1 . The AF-UNIT-20Hz test consists to run the modeling code in the 10x10x4.5 Km block with
minimum velocity 1684.08 m/s and maximum velocity 4000.0 m/s both for the P-waves. The model
includes a flat free-surface condition, where receivers record all three velocity components of motion
with a 5 milliseconds sampling. The shot is an explosive source (Ormsby wavelet with cut-off frequency
at 20Hz.) buried at a depth of 10 meters below the free surface. The second-order accurate in time and
eighth-order accurate in space discretized approach allows using 3.33 points per minimum wavelength,
i.e., a grid space of 25.0 meters and a temporal sampling rate of 0.001 seconds. Finally, to avoid spurious
reflection on the boundary of the domain the Absorbing Boundary Condition (ABC) developed by Cerjan
[6] is used. Considering a single shot (AF-UNIT-20Hz) located at sx = 5000m and sy = 5000m the
experiment is carried out on Lobo Carneiro (SGI ICE X), a cluster with 504 CPU’s Intel Xeon E5-
2670v3 (Haswell), where each node (252 nodes) has 24 cores. Thus, the optimized code is executed in
a dual-socket node, using all 24 available cores. OpenMP directives are inserted in the finite-difference
kernel. Besides, to ensure stencil vectorization, the code contains preprocessor directives (AVX2) and
1 hpc4e.eu

48 3
uses memory alignment allocation. The experiment takes only 720s of walltime in a single node with a
speedup of 6.3. Figure 1(b) shows the vertical velocity component (seismogram) extracted from inline
located at 5000 meters (Figure 1(a)) as specified in the AF-UNIT-20Hz experiment provided by the
HPC4E.

(a) (b)

Figure 1: Layer horizons (1(a)) related to the test cases. Vertical velocity (1(b)) extracted from inline
located at 5000 meters [5].

5. Conclusions
In this work we report improvements made in hybrid parallel high-order 3D finite difference simulation
of the heterogeneous acoustic wave equation in multi-core machines. We introduce a domain decomposi-
tion approach overlapping computation and communication. OpenMP directives and AVX2 instructions
speed-up kernel stencil computations, improving the overall performance. The code for modeling homo-
geneous acoustic wave equation with our hybrid parallel approach is freely available for download at the
High Performance Computing for Energy project website2 .

Acknowledgements
The research has received funding from CNPq, FAPERJ, the European Commission (HPC4E H2020
project) and the Brazilian Ministry of Science, Technology, Innovation and Communications. Com-
puter resources are provided by the High Performance Computer Center, COPPE/UFRJ and TACC, The
University of Texas at Austin, USA.

References
[1] C. H. Chapman. Fundamentals of Seismic Wave Propagation, Cambridge University Press, 2004.
[2] J. Virieux. P-SV wave propagation in heterogeneous media: Velocity-stress finite-difference method, Geo-
physics, 889-901, 1986.
[3] Intel. ISA Extensions Intel AVX - Intelr Software: https://software.intel.com/en-us/isa-extensions/intel-avx,
Online, 2013.
[4] S. Shende and A. D. Malony. The TAU Parallel Performance System, International Journal of High Perfor-
mance Computing Applications, Volume 20 Number 2 Summer 2006. Pages 287-311.
[5] J. de la Puente. Website deploying a suite of geophysical tests for wave propagation problems on extreme scale
machines: http://hpc4e.eu, HPC4E Consortium Partners, 2015.
[6] C. Cerjan, D. Kosloff, R. Kosloff, and M. Reshef. A nonreflecting boundary condition for discrete acoustic
and elastic wave equations, Geophysics, 705-708, 1985

2 https://https://hpc4e.eu/downloads/software

49 4
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

I NFLUENCE OF PRECONDITIONING ON FINITE ELEMENT SIMULATION OF COL -


UMN COLLAPSE USING A WELL - POSED µ(I)- RHEOLOGY

Linda Gesenhues 1 , José J. Camata 3 , Adriano M. A. Côrtes 2 , Fernando A. Rochinha 1 , Alvaro L.G.A.
Coutinho 1
1 COPPE, Federal University of Rio de Janeiro, Brazil
2 NUMPEX-COMP, Campus Duque de Caxias, Federal University of Rio de Janeiro, Brazil
3 Computer Science Department, Federal University of Juiz de Fora, Brazil

Dense granular flows often occur in nature and industry, e.g., in avalanches, turbidity currents, grain
silos or material handling. Therefore, a high interest exists to gain further insights to predict such flows.
In the last decades, the µ(I)-rheology has been more and more established to compute dense granular
flows with a continuous model. Besides a dependency on the shear rate, this rheology model is based on
the Coulomb friction and assumes that the normal stress applied by the pressure are proportional by a
friction coefficient to the tangential stress [1]. This complex constitutive model results in an unusual vis-
coplastic shear-thinning behavior. The equations are highly non-linear and thus lead easily to numerical
instabilities, which may end in divergence, oscillation and extreme sensitivity to the input parameters.
Recently Barker et al [2] demonstrated that under some assumptions the governing equations are
ill-posed. Thus the solution, if not diverged before, might be grid-dependent and not unique. Barker
and Gray [3] suggest a regularization strategy to ensure well-posed constitutive equations. Besides the
regularization of Barker and Gray we have introduced other regularizations strategies based on thresholds
for the minimum and maximum viscosity values.
Earlier, we have managed to run 2D and 3D simulations of a column collapse using the well-posed
formulation of the µ(I)-rheology. In this experimental setup, a dense fluid column at rest is surrounded
by a lighter fluid. With time, the denser fluid collapses due to the gravity and the different densities [4].
Along the process, we noticed that different preconditioning strategies have a significant impact on the
solver’s convergence, and thus, have considerable importance for a successful simulation.
In this study, we compare the behavior of different parallel preconditioners (Block-Jacobi, ILU(n),
Algebraic Multigrid) for the column collapse using the µ(I)-rheology. This also includes a reordering of
the standard matrix (AIJ) to a blocked matrix (BAIJ). The simulations are governed by the Navier-Stokes
solver coupled with a transport solver employing the volume-of-fluid method which describes the inter-
face of the dense and light fluids. As finite element formulation, we use the residual-based variational
multiscale method. All implementations are based upon libMesh, a C++ library which interfaces PETSc
and other libraries. Furthermore, it supports adaptive mesh refinement and coarsening [5].

References
[1] Pierre Jop, Yoel Forterre, and Olivier Pouliquen. A constitutive law for dense granular flows. Nature,
441(7094):727–730, 2006.
[2] T. Barker, D. G. Schaeffer, P. Bohorquez, and J. M. N. T. Gray. Well-posed and ill-posed behaviour of the
µ-rheology for granular flow. Journal of Fluid Mechanics, 779:794–818, 2015.
[3] T. Barker and J. M.N.T. Gray. Partial regularisation of the incompressible µ(I)-rheology for granular flow.
Journal of Fluid Mechanics, 828(October):5–32, 2017.
[4] Linda Gesenhues, Jose J. Camata, and Alvaro L.G.A. Coutinho. Simulation of a column collapse for dense
granular flows. In CILAMCE 2017, 2017.

50 1
[5] Benjamin S. Kirk, John W. Peterson, Roy H. Stogner, and Graham F. Carey. libMesh : a C++ library for
parallel adaptive mesh refinement/coarsening simulations. Engineering with Computers, 22(3-4):237–254,
2006.

51 2
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

NUMERICAL SIMULATION OF WIND ACTION ON SUPERSTRUCTURE OF A BRIDGE AND


DYNAMIC RESPONSE RATE FOR DIFFERENT DECK GEOMETRIES

Anaximandro Souza¹, Leonardo Lisboa², Rodrigo Melo³


1
Universidade Estadual do Maranhão, anaxec1@gmail.com
2
Universidade Ceuma, lisboaleo77@gmail.com
3
Universidade Ceuma, rodrigo145889@hotmail.com
4
Unidade de Ensino Superior Dom Bosco, lsfneto@hotmail.com
5
Universidade de São Paulo, henriquecampelo78@hotmail.com

Abstract: The present work has as main study direction the analysis of the wind action in different types
of bridge decks geometries and their dynamic response, using classic bridge models already constructed
such as the North American bridge Tacoma Narrows and the Brazilian Rio Niteroi, such example, and
usual models (trapezoid, rectangular and double-T) for validation of results, with the objective of
obtaining study parameters that help in the analysis of these elements of vital importance for large
population centers and cities with regard to transport logistics, based on the assumption that different
from the own weight and overloads that end up generating deformations, the wind produces non-
conservative forces generating dynamic phenomena and aeroelastic effects, provoking instability
through the movement of its superstructure. In order to obtain the study parameters, we used methods
related to computational modeling and numerical simulation, aiming at obtaining results with maximum
accuracy and demonstrating the validation of the same as their practical use in the context of civil
construction, especially in bridges, considering their interrelations in the environment in which they are
inserted.
Keywords: Bridge decks, Dynamic response, Usual deck models, Aeroelastic effects

1.Introduction and Motivation

Bridges are essential elements for the development of a region, both in economic and
geographic criteria. They are made for the transition of people, vehicles and transport of objects
between two regions that are separated by rivers, seas, cliffs, etc. Therefore, studying their
interactions with the environment, in this case the wind and the conditions of the stresses that
they are subjected are of vital importance when considering that you want a bridge with good
performance and high longevity. Classically, simulations involving the action of internal or
external forces on medium and large structures were made from large-scale tests using wind
tunnels and destructive tests (in case of stress analysis), sometimes disregarding important
variables in the evaluation process. The CFD (Computational Fluid Dynamics) adds to the
evaluative process as a method that makes possible the study of the fluid-structure system, a
concept that analyzes the behavior of a structure not only in function of its stresses and demands
but as a global dynamic spectrum which surrounds the fluid in which it is immersed. In
agreement with the cfd, in the present work also were used mathematical methods of modeling,
attached to computational systems, that help in the obtaining of results with the maximum of

52
reliability and security. One of the most used features in computational modeling, and also used
in the formulation of the present article, is the finite element method, which consists of a
numerical procedure for the approximate solution of problems about boundary values of
differential equations, which are changed by a limited number of elements that present well
defined behavior where they are found in several forms such as triangular, quadrilateral, among
others depending on the type of problem studied. The finite elements are connected to each
other forming nodes or nodal points, when forming a set of nodes we get the name of mesh, a
concept used in the design and modeling of bridge decks for posterior simulation.

2. Main Objectives

The present research has as main objective to analyze different geometries of bridge decks and
how they are able to influence phenomena of aeroelastic instability, starting from the
assumption that the geometry of the deck can influence at generation of vortices that will
consequently induce phenomena in the all superstructure of the bridge. Using ANSYS
numerical modeling and simulation software, it was possible create usual bridge decks
geometries and subject them to simulations of wind actions under the superstructure of the
bridge, making 2D finite element analysis (FEA) meshes with a high degree of refinement and
analyzing their behavior from it. The results presented by the software refer to the frequency of
vortex shedding in low-pressure regions of the structure, also known as Strouhal number, as
also as drag (CD) and lift (CL) coefficients, comparing those coefficients with natural frequency
of the structure, it is possible to determine the performance of each deck and his tendency to
aeroelastic instability. These parameters are essential for the correct analysis of the bridge
behavior when considering the study of fluid-structure interaction (FSI). as also as drag (CD)
and lift (CL) coefficients, comparing those coefficients with natural frequency of the structure,
it is possible to determine the performance of each deck and his tendency to aeroelastic
instability. These parameters are essential for the correct analysis of the bridge behavior when
considering the study of fluid-structure interaction (FSI).

a) b)

Fig. 1: a and b, different perspectives of the Rio Nitéroi bridge evidencing its deck in double-t format

In Figure 1, the Brazilian bridge Rio Niterói, built in the 1970s with the objective of connecting
the cities of Rio de Janeiro and Niterói, presented in the 1990s vertical bending in the 50cm
range, according to records of the time in response to the action of the wind in the region. was
later corrected with the aid of shock absorbers and against weights installed in its structure

53
c) d)

Fig 2: c and d, the north american bridge Tacoma Narrows, before and after the collapse in 1940.

The Tacoma Narrows bridge, shown in figure 2, was a North American suspension bridge that
was located on the Tacoma Strait, in the United States. It collapsed in 1940, a few months after
its inauguration due to strong gusts of wind that induced dynamic effects related to resonance
and aeroelasticity in it.
Figures 1 and 2 show the importance of the study of the dynamic effects that the wind induces
in these types of structures.

3. Concluding remarks on current and future research

Wind action in static structures, such as buildings and bridges, is one of the main factors that
must be taken on consideration in the design and execution of a project, especially when
considering structures with robust shapes (ie non-aerodynamic), the shape of these types of
elements corroborates with the appearance of undesirable phenomena for the structure, such as
vortex detachment, hammer effects and excessive oscillation, Therefore, the analysis and
prediction, as well as the consideration of the natural frequencies and requesting loads to which
bridges will be subject is essential, this process involves from a good choice of materials to the
choices of people qualified to execute and study the project In this scientific work, besides
obtaining parameters of analysis, two conclusions were evidenced, the first concerns the use of
computational resources in the scope of civil construction, in view of its practical and
facilitating bias, in addition to providing more precise results in relation to studies of structural
nature and dynamic fluid (in the case of this work), assisting in the constructive process. The
second is about the isostatic nature of most bridges, a factor that in addition to what common
sense may suggest, is of the utmost importance, considering that these types of elements have
the need to move with a certain degree of freedom, allowing a structural integrity.

Referencies

BRAUN, A. L. 2007., Simulação numérica na engenharia do vento incluindo efeitos de interação


fluido estrutura. (Tese de Doutorado) Escola de Engenharia, Universidade Federal do Rio Grande do
Sul, Porto Alegre.

GOMES, H. C. 2013., Método dos Elementos Finitos com Fronteiras Imersas aplicado a Problemas de
Dinâmica dos Fluidos e Interação Fluido-Estrutura. (Tese de Doutorado), Universidade de São Paulo,
São Paulo.

54
BLESSMANN, J, 2005., Introdução ao Estudo da Ação Dinâmica do Vento. 2. ed. Editora da
Universidade. Universidade Federal do Rio Grande do Sul, Porto Alegre.

BISMARCK, M. N. 1940., Structural Dynamics in Aeronautical Engineering. AIAA educational


series, Reston.

HALLAK, P, H, 2002., Parâmetros Aeroelásticos para Pontes via Fluidodinâmica


Computacional. (Tese de Doutorado), Universidade Federal do Rio de Janeiro, Rio de Janeiro.

55
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

ANALYSIS AND STRUCTURAL CORRECTION A BELT CONVEYOR TR –


315K03 FROM VALE COMPANY IN BRAZIL USING COMPUTATIONAL
MECHANICS TECHNIQUES

Anaximandro Souza¹, Carlos Marinho², Luís Jorge³


1
Universidade Estadual do Maranhão, anaxec1@gmail.com
2
Vale, carlos.marinho@vale.com
3
Universidade Ceuma, luisjorgeed25@hotmail.com

Abstract: The present article has as main objectives the structural analysis of the belt conveyor TR-
315K-03 (2200) and in the verification of its mechanical components, such equipment implanted in
the company Vale. The research was carried out through computational models of a numerical
simulation software Procal 3D that uses finite elements method, where the obtained data adequately
show the quality of the materials used in analyzes from the force requests to which they are being
submitted and the importance of its study to ensure its efficiency.
Keys Words: Belt Conveyor, finite elements, structural analysis.

1. Introduction

Vale is considered one of the largest mining companies in the world, having facilities in
several countries such as Indonesia, Oman, Argentina, Brazil and Mozambique. Its main activity and
source of profit is designated as mining and transportation of the ore, and to do its work the railroad
systems are used to move between regions to accelerate the process in line with low emission
pollutants that the railway line can offer. Therefore the issue of logistics related to the handling of ore
is an essential factor for success in every process.
In this article, the belt conveyor used for all types of work that require the uniform and safe
movement of materials in the company was analyzed, with the objective of obtaining data through
structural and mechanical analysis. The interpretation of the results obtained through computational
modeling and numerical simulation is of vital importance for the useful life and durability of this
equipment that play an important role within the company.
Computational modeling consists of an area of knowledge that deals with the application
processes of mathematical models using computational techniques for the solution and understandings
of complex problems. Such problems with a high degree of complexity can result in an agglomeration
of physical variables for the control of the process, thus creating techniques for several models, aiming
at solving the problem.
In this way, one of the computational modeling methods is finite elements, which consists of a
numerical procedure to solve problems in the limits of differential equations, where they are modified
by a definite number of elements that have a well-defined behavior and that depending on the problem
can be found in various forms such as quadrangular, triangular and other forms that will depend on the
problem being studied.
The finite elements when connected to each other form nodes or in other words, nodal points,
consisting of a set of creating nodes, carrying a finite element mesh where their precision will depend

56
on the number formed of nodes and their formats, thus, how much smaller its size and more quantity
in a mesh, more accurate and correct will be the result. More and more modern simulation software is
used that, over time, is evolving more and more, improving the analysis, the type and the formation of
the mesh of elements, using the techniques proposed by the modeling, with more accurate results and
less number of failures.
Using the finite element methodology with the Procon 3D modeling and numerical simulation
software, it was possible to obtain the structural analysis data and the analyzes of its mechanical
components of the belt TK-315K-03 transporter where they are indispensable parameters for good
operation.

2. Main Objectives and Methodology

To evaluate the structural and mechanical aspects of the TR-315K-03, according to the
applicable standards and requirements, with the aid of the modeling and numerical simulation software
Procal 3D, where it was possible to analyze in detail the structural and mechanical components of the
equipment.

Figure 1: Finite element general model - TR-315K-03.

Figure 2: Finite element model of the stringer and stretching tower

3. Concluding remarks on current and future research

From the results it was observed that for the mechanical analysis, the design radius of the
concave curve is lower than the minimum required to avoid belt lifting and buckling of the edges that

57
can cause damage to the cables and delamination of the cover, belt lifting can cause material spillage
and misalignment. If this condition occurs, it is suggested to adjust it by installing an anti-lifting
system.
For the structural analysis of TR-315K-03, no major structural element was presented with
failure in the combinations used. The highest utilization rates found are located near the discharge
drum, considering the high-tension value in the belt.
For modal analysis, resonance possibilities were presented in the natural modes 2, 3 and 4
with the frequencies of the load and return rollers. During the analysis of the flexibility of the belt
changer platform, the deformation obtained from the natural frequencies of the conveyor did not have
the possibility of coupling with the excitation frequencies of the equipment, but it was found that this
displacement does not interfere with the operation of the conveyor

References

[1] ISO-5049-1, Mobile equipment for continuous handling of bulk materials, 1994.
[2] ABNT, NBR 6123, Forças Devido ao Vento em Edificações, 1988.
[3] ABNT, NBR 8800, Projeto de estruturas de aço e de estruturas mistas de aço e concreto de
edifícios, 2008.
[4] NBR 6678, Transportadores de correia - Roletes – Dimensões, 2010.
[5] FEM, Rules for the design of mobile equipment for continuous handling of bulk materials, France,
1997.
[6] ISO 5048, Continuous Mechanical Handling Equipment – Belt Conveyors with carrying idlers,
Calculation of operating power and tensile forces, 1989.
[7] CEMA 6th Edition, Belt Conveyor for Bulk Materials, 2007.
[8] DIN 22101, Continuous conveyors – Belt conveyors for loose bulk materials – Belt for calculation
and dimensioning, 2011.
[9] AISC 2005 - Specification for Structural Steel Buildings, Allowable Stress Design and Plastic
Design, July 2006, American Institute of Steel Construction.
[10] ASCE Standard, Minimum design loads for buildings and other structures, 2005.
[11] AS 4324.1 – Mobile equipment for continuous handling of bulk materials, Part 1: General
requirements for the design of steel structures, Australian Standard, 1995.
[12] AS 3774 – Loads on bulk solid containers, Australian Standard, 1996;
[13] DNV-RP-C203, Fatigue design of offshore steel structures, 2010;
[14] HENFEL, Caixas para rolamentos, catálogo de caixas de mancais;
[15] VALE, EG-M-401, “Especificação geral para fornecimento de equipamentos mecânicos”, 2009.
[16] VALE, ET-M-406, "Especificação técnica para transportadores de correia", 2012.
[17] VALE, CP-S-501, “Critérios de projeto para estruturas metálicas”, 2013.
[18] Lemmon, R., Local Stresses in Belt Turnovers in Conveyor Belts, Bulk Material Handling by
Conveyor Belt IV, 2002.
[19] Desenhos técnicos, listados na Tabela 12.1, no ANEXO II – Relação dos desenhos consultados;
[20] DALLY J. W., Experimental stress analysis, McGRAW-HILL, 3rd ed., 1991.
[21] VISHAY PRECISION GROUP, Instruction Bulletin B-129-8 – Surface Preparation for Strain
Gage Bonding, Micro-Measurements, Rev 19, 2011.
[22] VISHAY PRECISION GROUP, Instruction Bulletin B-127-14 – Strain Gage Installations with
M-Bond 200 Adhesive, Micro-Measurements, Rev 14, 2011.

58
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

PROBABILISTIC PROGRESSIVE COLLAPSE OF A TRUSS:


A BENCHMARK EXAMPLE OF SYSTEM RBDO AND RISK-BASED OPTIMIZATION
André T. Beck1, Rodolfo K. Tessari2, Henrique M. Kroetz3
1
University of São Paulo, atbeck@sc.usp.br
2
University of São Paulo, tessari.rodolfo@usp.br
3
University of São Paulo, henrique.kroetz@usp.br

System Reliability-Based Design Optimization (RBDO) methods and examples in the published
literature are not many. Specifically, benchmark examples with general system behavior and
progressive failure are almost non-existent, but very important to test system RBDO methods. In this
paper, a benchmark example involving progressive failure of a hyper-static truss is addressed. Two
formulations of the problem are addressed: the classical RBDO formulation, with reliability
constraints; and a risk-based formulation, which takes into account consequences of failure. It is
demonstrated that a typical system RBDO formulation always leads to isostatic structures, as the
formulation does not offer any incentive for the permanence of hyper-static members. In the risk-based
formulation, failure costs are differentiated w.r.t. primary member failure, in hyper-static structures
(existence of warning before eventual collapse), and failure of isostatic members (no warning). In this
formulation, optimal designs also include hyper-static structures, but this depends on cost of failure
scenarios. Results presented herein are relevant in the modern context of robust design considering
progressive collapse.

Keywords: structural optimization; RBDO; System RBDO; risk optimization; progressive collapse.

Acknowledgements: The authors acknowledge funding of this research project by Brazilian agencies
CAPES (Brazilian Higher Education Council), CNPq (Brazilian National Council for Research, grant
n. 306373/2016-5) and FAPESP (São Paulo State Foundation for Research, grant n. 2017/01243-5).

References

[1] Aoues Y, Chateauneuf A, 2008: Reliability-based optimization of structural systems by adaptive


target safety – Application to RC frames, Structural Safety 30, 144–161.
[2] Aoues Y, Chateauneuf A, 2010: Benchmark study of numerical methods for reliability-based
design optimization. Structural and Multidisciplinary Optimization, 41:277–294.
[3] Beck AT, Gomes WJS, 2012: A comparison of deterministic, reliability-based and risk-based
structural optimization under uncertainty. Probabilistic Engineering Mechanics 28, 18-29.
[4] Beck AT, Gomes WJS, Bazán FAV, 2012: On the Robustness of Structural Risk Optimization
with Respect to Epistemic Uncertainties. Int. J. for Uncertainty Quantification 2, 1 - 20.
[5] Beck AT, Gomes WJS, Lopez RH, Miguel LFF, 2015: A comparison between robust and risk-
based optimization under uncertainty, Struct. Multidisc. Optim. 52, 479-492.
[6] Beck AT, Kougioumtzoglou IA, Santos KR., 2014: Optimal Performance-Based Design of Non-
Linear Stochastic Dynamical Systems. Engineering Structures 78, 145-153.

59
[7] Beck AT, Verzenhassi CC, 2008: Risk Optimization of a Steel Frame Communications Tower
Subject to Tornado Winds, Latin American Journal of Solids and Structures 5, 187-203.
[8] Beyer HG, Sendhoff B, 2007: Robust optimization - A comprehensive survey, Computer Methods
in Applied Mechanics and Engineering 196, 3190-3218.
[9] Enevoldsen I, Sørensen JD, 1993: Reliability-based optimization of series system of parallel
systems. J. Struct. Eng. (ASCE) 119(14), 1069–84.
[10] Gomes WJS, Beck AT, 2013. Global structural optimization considering expected consequences
of failure and using ANN surrogates. Computers & Structures 126, 56–68.
[11] Gomes WJS, Beck AT, 2014a: Optimal Inspection and Design of Onshore Pipelines Under
External Corrosion Process, Structural Safety 47, 48-58.
[12] Gomes WJS, Beck AT, 2014b: Optimal Inspection Planning and Repair Under Random Crack
Propagation, Engineering Structures 69, 285-296.
[13] Gomes WJS, Beck AT, Haukaas T, 2013: Optimal Inspection Planning for Onshore Pipelines
Subject to External Corrosion. Reliability Engineering & Systems Safety 118, 18-27.
[14] Liang J, Mourelatos ZP, Nikolaidis E, 2007: A single-loop approach for system reliability-based
design optimization. J Mech Des 129(12):1215–1224.
[15] Lopez RH, Beck AT, 2012: Reliability-based design optimization strategies based on FORM: a
review. Journal of the Brazilian Society of Mech. Sciences and Eng. 34, 506-514.
[16] McDonald M, Mahadevan S, 2008: Design Optimization with System-Level Reliability, ASME
Journal of Mechanical Design 130, 021403.
[17] Nguyen TH, Song J, Paulino GH, 2010: Single-Loop System Reliability-Based Design
Optimization Using Matrix-Based System Reliability Method: Theory and Applications, ASME
Journal of Mechanical Design 132, 011005-1-11.
[18] Okasha NM, 2016: An improved weighted average simulation approach for solving reliability-
based analysis and design optimization problems, Structural Safety 60, 47–55.
[19] Schuëller GI, Jensen HA, 2009: Computational methods in optimization considering uncertainties
– an overview, Computer Methods in Applied Mechanics and Engineering 198, 2-13.

60
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

UNCERTAINTY QUANTIFICATION & RISK ANALYSIS IN ENGINEERING, WITH


APPLICATIONS TO OIL & GAS EXPLORATION
André T. Beck1
1
University of São Paulo, atbeck@sc.usp.br

Current global climate trends, perceived trough higher ocean levels and higher temperatures, are
leading to more intense and more frequent extreme weather events. This, combined with the increase
in urbanization and in coastal development, has the potential to drastically increase consequences of
extreme weather and other natural hazards. A major hurricane hit the coast of Santa Catarina in 2004,
for the first time on record, with winds of up to 180 km/h. Five hurricanes that hit the Gulf of Mexico,
between 2004 and 2008, where stronger than anything previously recorded, in over 50 years of
offshore exploration in the Gulf area. As a consequence, design loads for fixed offshore structures
increased by nearly 50%. As we approach times of higher uncertainties, regulations on engineering
facilities are becoming stricter, with mandatory risk analysis for all offshore and most onshore Oil &
Gas facilities. The profession is learning that absolute safety is an utopia, hence consequences of
failure have to be explicitly addressed in engineering design. These recent challenges have also led to
the development of new engineering design concepts, such as infra-structure robustness and resilience,
and performance-based design. Addressing the challenges above involves uncertainty quantification,
reliability and risk analysis. These issues should drive future research efforts in the Oil & Gas and
other industries. Addressing these problems will require a change in the deterministic way that the
engineering profession is perceived, which should also drive changes in engineering curricula.

Keywords: extreme weather, risk analysis, uncertainty quantification, reliability, Oil & Gas.

Acknowledgements: The authors acknowledge funding of this research project by Brazilian agencies
CAPES (Brazilian Higher Education Council), CNPq (Brazilian National Council for Research, grant
n. 306373/2016-5) and FAPESP (São Paulo State Foundation for Research, grant n. 2017/01243-5).

References

[1] Gilbert RB et al., 2012: Achieving Reliability in the Face of Extreme Uncertainty, Proc. 5th
Asian-Pacific Symposium on Structural Reliability and its Applications.
[2] Haldar A, Mehrabian A, 2008: Structural engineering in the new millennium: opportunities and
challenges, Structural Survey 26, 279-301.
[3] Petroski H, 1992: To Engineer Is Human: The Role of Failure in Successful Design, 1st Edition,
Vintage Books.
[4] Petroski H, 2006: Success through Failure: the paradox of design, Princeton University Press,
UK.

61
[5] Petroski H, 2012: To Forgive Design: Understanding Failure, Harvard University Press.
[6] Taleb NN, 2010: The Black Swan - The Impact of the Highly Improbable, Random House, NY.
[7] Vogel RM, 1999: Stochastic and deterministic world views, Editorial: Journal of Water Resources
Planning and Management 125, 311-313.

62
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

DRYING OF CERAMIC BLOCKS IN AN INDUSTRIAL TUNNEL DRYER: MODELING AND


SIMULATION

Anderson Melchiades Vasconcelos da Silva1, Mariana Julie do Nascimento Santos2, Antonio


Gilson Barbosa de Lima 3
1
Federal University of Campina Grande, Department of Mechanical Engineering, Av: Aprígio Veloso,
882, Bodocongó, Campina Grande-PB, Zip Code 58429-900, Brazil, andersonmelch@gmail.com
2
Federal University of Campina Grande, Department of Mechanical Engineering, Av: Aprígio Veloso,
882, Bodocongó, Campina Grande-PB, Zip Code 58429-900, Brazil, marianajulie@outlook.com
3
Federal University of Campina Grande, Department of Mechanical Engineering, Av: Aprígio Veloso,
882, Bodocongó, Campina Grande-PB, Zip Code 58429-900, Brazil, antonio.gilson@ufcg.edu.br

Clay materials manufacturing has been reported in the literature based on the material chemical
composition and commercial type products, such as brick, roof tiles, blocks, etc. These products are
very different in size, shape, detail, complexity, material formulation, structure, and cost. The clay
product, during the manufacturing process, goes through the steps of product molding, drying and
firing. In the molding process, water is added to the clay to increase plasticity, and to facilitate the
forming of a specified product. The drying is a stage of the process that precedes firing, and needs a
large amount of thermal energy, to remove the water, which was added during the molding process.
However, drying can cause different troubles in the materials. If this moisture content is not removed
adequately, severe stresses occurs inside the material causing cracking and fissures reducing quality of
product post-drying process. Then the drying air conditions affect the moisture removal and heating of
the product. In this sense, the published [1-3] literature shows that if special care is taken during the
drying, energy costs can be reduced and a better quality product, with a smaller process time, can be
obtained.
Drying plays an important role in different industrial applications, thus, mathematical models to
describe the drying process have been reported by many researchers, especially for ceramic material
[4,5]. In this sense, this work focuses on the drying of industrial clay blocks in a cross-flow tunnel
dryer by using numerical simulation. The proposed mathematical model is based on an industrial
tunnel dryer and energy and mass transfer between the product and hot air in the drying process.
The proposed problem is time dependent and was solved using the finite volume numerical method
[7-9]. The integration of the partial differential equations, in the volume and time, resulted in a set of
linear equations in discretized form. To obtain the numerical results, a computational code in language
of the Mathematica® program, for the solutions of the equations used, was developed. Details about
the process parameters and numerical procedure can be found in the literature [4-6,10,11].
Predicted results of the moisture content and the surface product temperature (vertices) as a
function of the drying time, are presented and compared with experimental results (drying of 1 block
in oven) [6]. This comparison is possible because the analysis considers only the first layer of blocks
in the bed (y = 0).
In this sense, in accordance with the predicted results of the hollow blocks drying in a tunnel dryer
of cross-flow type, it can be concluded, that the mathematical modeling and numerical method

63
proposed in this work can be used to simulate the drying process of blocks and other clay materials in
a cross flow dryer, because a good approximation between the numerical and experimental results of
the moisture content and surface temperature of the block were obtained. Here, was confirmed that the
blocks of the first layer get dried and heated a little more than those in the last layer. This is due to the
fact that when the drying air arrives in the first layer of bricks it has a low absolute humidity and as it
goes on penetrating in the next layers of the product it receives water vapor of the blocks, increasing
the absolute humidity and reducing its temperature. During the drying process, low gradients of
temperature and moisture content, within the bed, were obtained, providing at the end of the process, a
more uniform drying and obtaining good quality products.
The results is in agreement with the concept that a non-uniform drying and high thermo-hydro-
mechanical stresses in the ceramic blocks may cause cracking, fissure and deformation in them,
affecting its quality at the end of the drying process. Since the drying process studied in the present
work has been more controllable, so products with good quality can be obtained.

References

[1] W.D. Callister, Jr. Materials science and engineering an introduction, 7td ed., John Wiley & Sons,
Inc. USA, 2007.
[2] D. A. Brosnan, G. C. Robinson. Introduction to drying of ceramics. Ohio: The American Ceramic
Society, 2003.
[3] J.S. Reed, Principles of ceramics processing. John Wiley & Sons, Inc., New York, 1995.
[4] G.S. Almeida. Simulation and experimentation of red ceramics drying in industrial thermal
systems, Ph.D Thesis (Doctorate in Process Engineering), Federal University of Campina Grande,
2008. 88f (In portuguese).
[5] G. S. Almeida, J.B. Silva, C. Joaquina e Silva, R. Swarnakar, G. A. Neves, A. G. B Lima, Heat
and mass transport in an industrial tunnel dryer: modeling and simulation applied to hollow
bricks. Applied Thermal Engineering, v. 55, p. 78-86, 2013.
[6] A.M.V Silva. Drying of industrial ceramics blocks: Modeling and simulation, Ph.D Thesis
(Doctorate in Process Engineering), Federal University of Campina Grande, 2018. (In portuguese).
[7] C.R. Maliska. Computational heat transfer and fluid mechanics. Rio de Janeiro: LTC-Livros
Técnicos e Científicos Editora S.A, p453, 2004. (In Portuguese).
[8] S.V. Patankar. Numerical heat transfer and fluid flow, New York: Hemisphere Publishing
Coorporation, p197, 1980.
[9] H. K. Versteeg, W. Malalasekera. An introduction to computational fluid dynamics: the finite
volume method, Prentice Hall, London, 1995.
[10] F. V. S. Tavares, S. R. Farias Neto, E. S. Barbosa, A. G. B. de Lima, C. Joaquina e Silva. Drying
of ceramic hollow bricks in an industrial tunnel dryer: A finite volume analysis. The International
Journal of Multiphysics, v. 8, p. 297-312, 2014.
[11] G.S. Almeida, F.V.S. Tavares, W.M.P.B. Lima, A.G. Barbosa de Lima, Energetic and exergetic
analysis of the clay bricks drying in an industrial tunnel dryer. Defect and Diffusion Forum. v.369,
p.104 - 109, 2016.

64
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

NON-ISOTHERMAL ABSORPTION OF WATER IN VEGETABLE FIBER REINFORCED


POLYMER COMPOSITES: AN APPLICATION TO THE LANGMUIR-TYPE MASS
TRANSPORT MODEL AND THE FINITE-VOLUME METHOD

Rafaela Quinto da Costa Melo1, José Vieira da Silva2, Antonio Gilson Barbosa de Lima3
1
Federal University of Campina Grande, Department of Materials Engineering, Av: Aprígio Veloso,
882, Bodocongó, Campina Grande-PB, Zip Code 58429-900, Brazil, rafaelaquinto@live.com
2
Federal University of Campina Grande, Department of Mechanical Engineering, Av: Aprígio Veloso,
882, Bodocongó, Campina Grande-PB, Zip Code 58429-900, Brazil, jvieira7@gmail.com
3
Federal University of Campina Grande, Department of Mechanical Engineering, Av: Aprígio Veloso,
882, Bodocongó, Campina Grande-PB, Zip Code 58429-900, Brazil, antonio.gilson@ufcg.edu.br

Vegetal fibers have presented great potential for technological applications due to its potential as
substitute of synthetic fibers. Their use as reinforcement in polymer composite materials present
several advantages such as, low density, low cost, being obtained from renewable sources, are widely
available and environmentally superior to synthetic fiber [1-6]. The main disadvantage of these
materials is their susceptibility to the influence of environmental agents, especially humidity and
temperature [2,6-8]. Thus, the objective of this work was to develop a transient and one-dimensional
mathematical modeling and its numerical solution to predict heat and mass transfer during water
absorption in polymeric composite reinforced with vegetable fiber based on the Langmuir model.
The Langmuir diffusion model (Non-Fickian diffusion) was initially proposed by Carter and Kilber
[9] and it is based on the hypothesis of the existence of probabilities describing two "states" of
absorbed water [5,6,8,10]. In the first state, the free water molecules diffuse inside the porous solid
and are absorbed with a probability λ per unit time. In the second one, the water molecules leave the
trapped state and become free with a probability μ per unit of time. For the mathematical model was
considered a plate with thickness 2a, immersed in a fluid solute (water), placed in a container of
thickness (2L+2a). The solid is homogeneous, isotropic and totally dry at the beginning of the process.
Further, the solid is symmetrical with respect to their center. The total moisture content inside the
material in a specific position x and instant t is found from the sum of the amounts of free and
entrapped solute, and the average moisture content of the solid at any time is given by integration of
the moisture content in the volume.
For the numerical solution of the governing equation, the discretization process (finite-volume
method) is done by applying the integral in all terms of the governing equation in volume and time
[11]. To solve the equation system originated by the discretization process, a computer code was
developed on the Wolfram Mathematica® platform. The equations were solved iteratively using the
Gauss-Seidel method. Details about the modeling and numerical procedure can be found in the
literature [5,6,8,9].
The formulation was applied to water absorption process in polymeric composites reinforced with
Caroá vegetable fibers. The composite studied have very large width and length compared to the
thickness, so, the pure water penetrates only in the direction of the thickness. For the validation of the
model, numerical results of the average moisture content was compared with experimental data

65
reported by Silva et al. [12] for polymer composite materials reinforced by Caroá fiber. From the
analysis it was verified a good agreement between the predict and experimental results, which confirm
that the mathematical model adequately predict the water diffusion process inside the material.
From the analysis of the predicted results, it was concluded that the water absorption is fast in the
initial stages and tends to decay for long exposure times to water until reaching the hygroscopic
saturation point. I was observed that, in regions near the surface, water absorption is faster. The water
penetrates the interior of the material generating a moderate concentration gradient along the thickness
and decreasing with the progress of the absorption process. Thus, at any point inside the solid, the
moisture content increases with immersion time until it reaches its hygroscopic saturation condition.
Here, is noticed that with a longer immersion time there is an increase in the amount of molecules
trapped within the material while decreasing the amount of free molecules to diffuse until the
equilibrium condition (μS=λC). Thus, the higher the concentration of the free solute, the higher the
concentration of immobilized solute found inside the polymer composite.

References

[1] S. V. Joshi, L. T. Drzal, A. K. Mohanty, S. Arora. Are natural fiber composites environmentally
superior to glass fiber reinforced composites?. Composites Part A: Applied science and
manufacturing, v. 35, n. 3, p. 371-376, 2004.
[2] L. Mohammed, M. N. Ansari, G. Pua, M. Jawaid, M. S. Islam. A review on natural fiber reinforced
polymer composite and its applications. International Journal of Polymer Science, v. 2015, 2015.
[3] E. Omrani, P. L. Menezes, P. K. Rohatgi. State of the art on tribological behavior of polymer
matrix composites reinforced with natural fibers in the green materials world. Engineering Science
and Technology, v. 47, p.777-780, 2015.
[4] H. Patel, A. Parkhe, P. K. Shrama. Mechanical behaviors of banana and sisal hybrid composites
reinforced with epoxy resin. International Journal of Research–Granthaalayah, v. 4, n.1, p.206-216,
2016.
[5] W. R. G. Santos, R. Q. C. Melo, A. G. B. Lima. Water absorption in polymer composites
reinforced with vegetable fiber using Langmuir-type model: An exact mathematical treatment, Defect
and Diffusion Forum, v. 371, p. 102-110, 2016.
[6] R. Q. C. Melo, A. G. B. Lima, Vegetable fiber-reinforced polymer composites: fundamentals,
mechanical properties and applications. Diffusion Foundations, v.14, p.1 - 20, 2017.
[7] L. H. Carvalho, E. L. Canedo, S. R. Farias Neto, A. G. B. Lima, Moisture transport process in
vegetable fiber composites: Theory and analysis for technological applications. In: Industrial and
Technological Applications of Transport in Porous Materials. Springer Heidelberg, p. 37-62, 2013.
[8] R. Q. C. Melo, W. R. G. Santos, A. G. B. Lima, W. M. P. B. Lima, J. V. Silva, R. P. Farias. Water
absorption process in polymer composites: Theory analysis and applications. In: Transport
phenomena in multiphase systems. Series: Advanced Structured Materials.1 ed. Heidelberg
(Germany): Springer International Publishing, 2018, v. 93, p. 219-249.
[9] H. G. Carter, K. G. Kibler. Langmuir-type model for anomalous moisture diffusion in composite
resins. Journal of Composite Materials, v. 12, n. 2, p. 118-131, 1978.
[10] T. I. Glaskova, R. M. Guedes, J. J. Morais, A. N. Aniskevich, A comparative analysis of moisture
transport models as applied to an epoxy binder. Mechanics of Composite Materials, v. 43, n. 4, p.
377-388, 2007.
[11] C. R. Maliska. Computational heat transfer and fluid mechanics. Rio de Janeiro: LTC-Livros
Técnicos e Científicos Editora S.A, 2004. (In Portuguese).
[12] C. J. Silva, A. G. B. Lima, E. G. Silva, T. H. F. Andrade, R. Q. C. Melo, Water absorption in
caroá-fiber reinforced polymer composite at different temperatures: A theoretical investigation.
Diffusion Foundations, v.10, p.16 - 27, 2017.

66
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A N EFFICIENT DOMAIN DECOMPOSITION METHOD WITH CROSS - POINT TREAT-


MENT FOR H ELMHOLTZ PROBLEMS

Axel Modave 1 , Xavier Antoine 2 , Christophe Geuzaine 3


1 POEMS, CNRS-ENSTA-Inria, Palaiseau (France) axel.modave@ensta-paristech.fr
2 Université de Lorraine, CNRS, Inria, IECL (France) xavier.antoine@univ-lorraine.fr
3 Université de Liège (Belgium) christophe.geuzaine@uliege.be

Solving high-frequency time-harmonic scattering problems using finite element techniques is chal-
lenging, as such problems lead to very large, complex and indefinite linear systems. Optimized Schwarz
domain decomposition methods (DDMs) are currently a very promising approach, where subproblems
of smaller sizes are solved in parallel using direct solvers, and are combined in an iterative procedure.
It is well-known that the convergence rate of these methods strongly depends on the transmission
condition enforced on the interfaces between the subdomains. Local transmission conditions based on
high-order absorbing boundary conditions (HABCs) have proved well suited [1, 2]. They represent a good
compromise between basic impedance conditions (which lead to suboptimal convergence) and the exact
Dirichlet-to-Neumann (DtN) map related to the complementary of the subdomain (which is expensive
to compute). However, a direct application of this approach for domain decomposition configurations
with cross-points, where more than two subdomains meet, does not provide satisfactory results.
We present an improved DDM that efficiently addresses configurations with cross points. Noting
that these points actually are corners for the subdomains, our strategy consists in incorporating a corner
treatment developed for HABCs into the DDM procedure. After a presentation of the key aspects of the
methods, the effectiveness of our approach is discussed with two-dimensional finite element results.

1. Helmholtz problem with HABC and corner treatment


In order to solve scattering problems set on infinite or very large domains by finite element methods,
a common strategy consists in computing the numerical solution only on a truncated computational
domain, and using a non-reflecting treatment at the artificial boundary, such as a HABC or a perfectly
matched layer (PML). To describe our approach, we consider a two-dimensional Helmholtz problem
defined on a rectangular computational domain Ω:

∆u + k2 u = s, in Ω,
(
(1)
∂n f u + B u = 0, on Γ f , ( f = 1 . . . 4)

where k(x) is the wavenumber, s(x) is a source term, Γ f is an edge of the rectangular domain, ∂n f is
the exterior normal derivative on Γ f ⊂ ∂Ω, and B is a non-reflecting boundary operator. Following [3],
we use a Padé-type HABC, which is obtained by approximating to exact DtN operator for the half-space
problem, with constant k and s = 0 outside. It corresponds to using, for each edge Γ f ,

on Γ f ,
2ıαk N
B u = −ıkαu − ∑ ci (u f ,i + u),
M i=1

and introducing N auxiliary fields {u f ,i }Ni=1 governed by the auxiliary equations

∂2ττ u f ,i + k2 (α2 ci + 1)u f ,i + α2 (ci + 1)u = 0, on Γ f , (i = 1 . . . N) (2)




67 1
where ∂2ττ is the second-order tangent derivative, α = eıφ/2 , ci = tan2 (iπ/M) and M = 2N + 1. The
accuracy of the numerical solution at the boundary depends on the number N and the angle φ [3].
Because of the spatial derivative in equation (2), additional boundary conditions must be prescribed
on the auxiliary fields at the boundary of the edges (i.e. at the corner of the domain) to close the sys-
tem. Following [4], we introduce new relations that ensure the compatibility of the system without any
supplementary approximation. With these relations, the auxiliary fields defined on adjacent edges are
coupled at the common corner. For the fields {u f ,i }Ni=1 defined on Γ f , having an adjacent edge Γ f 0 , the
boundary conditions at the corner Pf ,c = Γ f ∩ Γ f 0 can be written as
∂n f ,c u f ,i + D u f ,i = 0, on Pf ,c , (i = 1 . . . N) (3)
with

on Pf ,c ,
N
2ıαk (α2 ci + 1 − α2 )u f ,i − α2 (ci + 1)u f 0 , j
D u f ,i = −ıkαu f ,i −
M ∑ cj α2 ci + α2 c j + 1
= 0, (i = 1 . . . N)
j=1

where ∂n f ,c is the exterior normal derivative at Pf ,c ⊂ ∂Γ f , and {u f 0 , j }Nj=1 are the auxiliary fields defined
on Γ f 0 . Finally, the problem consists in solving the main field u(x) on the domain with boundary con-
ditions on the edges (equation (1)) and N auxiliary fields on each edge with boundary conditions at the
corners (equations (2)-(3)). See [4] for a three-dimensional version of this strategy.

2. A non-overlapping DDM with cross-point treatment


The Helmholtz problem defined on Ω is decomposed into subproblems defined on non-overlapping
subdomains ΩI (I = 1 . . . N dom ), with Ω = I ΩI and ΩI ΩJ = 0/ if I 6= J. We consider here a structured
S T

decomposition of a rectangular domain Ω into an array of rectangular subdomains (see figure 1). The
edges of each subdomain ΩI are denoted ΓI, f ( f = 1 . . . 4). Each edge could be an interface edge (if there
is a neighboring subdomain beyond the edge) or a boundary edge (if it belongs to the boundary of Ω).
At each iteration of the DDM algorithm, the subproblems are
solved in parallel, and data are exchanged at the interfaces between
the subdomains to synchronize the solutions. The additive Schwarz
∗ • ∗
DDM can be described as follows, at iteration ` + 1:
Ω4 Ω3

• For all subdomain ΩI , compute u`+1 solution to


g4,f,d,i
I
(
∆u`+1 + k2 u`+1 = s, in ΩI ,
• •◦ •

= g`I, f , on ΓI, f ,
I I g1,f,c,i
∂nI, f u`+1
I + B u`+1
I ( f = 1 . . . 4) g1,f g2,g Γ

where g`I, f is a transmission variable if ΓI, f is an interface Ω1 Ω2


edge, or it is set to zero if ΓI, f is a boundary edge.
∗ • ∗

• For all interface between neighboring subdomains ΩI and ΩJ ,


⇌ Transmission variables
for interface edges

update the transmission variables gI,`+1


f and gJ,g according to
`+1 ⇌ Transmission variables
for boundary/cross points

on ΓJ,g ,
I, f = −gJ,g + 2B uJ , Figure 1: Example of domain decom-
g`+1 ` `+1

J,g = −gI, f + 2B uI ,
g`+1 ` `+1
on ΓI, f , position (2 × 2 configuration) with
the transmission variables.
where ΓI, f = ΓJ,g is the common interface edge.
The convergence of this algorithm is accelerated by using an iterative Krylov method (GMRES) on the
top of the procedure for updating the transmission variables. See e.g. [1] for more details.
Because the HABC is used as transmission operator (on the interface edges) and as boundary op-
erator (on the boundary edges), the description of the DDM algorithm is incomplete. It should include
the auxiliary fields defined on the edges and, since these fields are governed by equation (2), additional
conditions should be prescribed at the boundary of the edges. The boundary points of each edge ΓI, f
are denoted PI, f ,c (c = 1, 2). Each point could be a cross point (at the cross of two interface edges), a
corner point (at the cross of two boundary edges) or a boundary point (at the cross of one interface edge
and one boundary edge). They are represented with the symbols ◦, • and ∗ on figure 1.

68 2
At iteration ` + 1 of the DDM algorithm, we have the additional operations:
• For all boundary/interface edge ΓI, f , compute the auxiliary field u`+1 I, f ,i solution to

on ΓI, f ,
( 2 `+1 
∂ττ uI, f ,i + k2 (α2 ci + 1)u`+1 2
I, f ,i + α (ci + 1)uI
`+1
= 0,

I, f ,i + D uI, f ,i = gI, f ,c,i , at PI, f ,c ,


∂nI, f ,c u`+1 `+1 `
(c = 1, 2)

where g`I, f ,c,i is a transmission variable if the adjacent edge of ΓI, f at the boundary point PI, f ,c is
an interface edge, or it is set to zero if the adjacent edge is a boundary edge.
• For all boundary/cross point shared by the edges ΓI, f and ΓJ, f of neighboring subdomains ΩI and
ΩJ , update the transmission variables g`I, f ,c,i and g`J, f ,d,i according to:

I, f ,c,i = −gJ, f ,d,i + 2D uJ, f ,i ,


g`+1 ` `+1
on PJ, f ,d ,

J, f ,d,i = −gI, f ,c,i + 2D uI, f ,i ,


g`+1 ` `+1
on PI, f ,c ,

where PI, f ,c = PJ, f ,d is the common point.


The auxiliary fields of two adjacent edges of one subdomain are coupled by the operator D at the com-
mon corner. All these operations are rather naturally included in the DDM algorithm. A GMRES is used
for updating all the transmission variables, which are now associated to shared edges and shared points.

3. Preliminary finite element results


In order to verify and to analyze the efficiency of the proposed DDM, we present finite element results
obtained with two 2D benchmarks. The numerical scheme is based on a Galerkin method adapted from
[1], with meshes made of triangles, nodal finite elements, and second-degree basis functions. The sim-
ulations are made with the GetDP and GetDDM environments [5].
In the first benchmark, we consider the scattering of an incident plane wave by a disk. The scattered
field is computed on a rectangular domain Ω, which is partitioned into six subdomains (figure 2a). A
Neumann condition is prescribed on the boundary of the disk, and the HABC with the corner treatment
is used on the exterior boundary. In the DDM algorithm, transmission operators based on an optimized
impedance condition [2] and the HABC are tested. In the former case, a specific treatment ensures the
compatibility between the impedance condition (on the interface edges) and the HABC (on the boundary
edges) at the boundary points. The effect of the boundary/cross-point treatments is analyzed by keeping
or removing the corresponding terms in the finite element scheme.

(a) Solution (b) Relative residual (c) Relative L2 −error


100 100

10−2 10−2
Relative L2−error
Relative residual

10−4 10−4

10−6 10−6

Impedance + no point treat Impedance + no point treat


10−8 Impedance + boundary-point treat 10−8 Impedance + boundary-point treat
HABC + no point treat HABC + no point treat
HABC + boundary-point treat HABC + boundary-point treat
HABC + boundary/cross-point treat HABC + boundary/cross-point treat
10−10 10−10
0 20 40 60 0 20 40 60
GMRES iteration GMRES iteration

Figure 2: Scattering benchmark: Configuration and solution (a). History of residual (b) and L2 −error (b) for trans-
mission operators based on the basic impedance condition (red lines) or the HABC (black lines) with/without point
treatments. Parameters: N = 4 and φ = π/3

69 3
(a) Velocity model

(e) Number of iterations for different configurations

1 line + (N dom ) columns

(b) Domain 1 × 45
250 3 lines + (N dom /3) columns
5 lines + (N dom /5) columns

Number of GMRES iterations


200

150

(c) Domain 3 × 15
100

50

(d) Domain 5 × 9 0
0 20 40 60 80 100
Number of subdomains N dom

Figure 3: Marmousi benchmark: Velocity model (a) and solution with three kinds of domain partition (b)-(d). The
HABC is used both as boundary condition and transmission operator. The point treatments are enabled. The
number of iterations to reach the relative residual 10−6 is plotted as a function of the number of subdomains for
the three kinds of domain partition (e).

For both kinds of transmission operator, the boundary-point treatment is required for converging
towards the good solution. Indeed, without this treatment, the residual decreases with the iteration
number, but the error stagnates (figures 2b and 2c, respectively). By contrast, the error decreases cor-
rectly when the boundary-point treatment is enabled. In that case, the convergence is faster with the
transmission operator based on the HABC, and it is even faster with the cross-point treatment.
We finally address a more challenging benchmark with a heterogeneous medium: the Marmousi
model, which represents a realistic geological structure (figure 3a). Although the HABC was initially de-
rived by assuming a constant wavenumber, it provides good accuracy for problems with heterogeneous
media [4], and we have observed that it accelerates the convergences of the DDM for the Marmousi
benchmark (results not shown here). On figure 3, the convergence of the method is compared for dif-
ferent domain partitions. For a larger number of subdomains (which is required to solve large prob-
lems), the DDM converges significantly faster with a multi-dimensional partition than with the mono-
dimensional partition, which confirms the efficiency and the interest of our approach.

References

[1] Y Boubendir, X Antoine, and C Geuzaine. A quasi-optimal non-overlapping domain decomposition algorithm
for the Helmholtz equation. Journal of Computational Physics, 231(2):262–280, 2012.
[2] M Gander, F Magoules, and F Nataf. Optimized Schwarz methods without overlap for the Helmholtz equation.
SIAM Journal on Scientific Computing, 24(1):38–60, 2002.
[3] R Kechroud, X Antoine, and A Soulaïmani. Numerical accuracy of a Padé-type non-reflecting boundary con-
dition for the finite element solution of acoustic scattering problems at high-frequency. International Journal
for Numerical Methods in Engineering, 64(10):1275–1302, 2005.
[4] A Modave, A Atle, J Chan, and T Warburton. A GPU-accelerated nodal discontinuous Galerkin method with
high-order absorbing boundary conditions and corner/edge compatibility. International Journal for Numerical
Methods in Engineering, 112(11):1659–1686, 2017.
[5] B Thierry, A Vion, S Tournier, M El Bouajaji, D Colignon, N Marsic, X Antoine, and C Geuzaine. GetDDM: an
open framework for testing optimized Schwarz methods for time-harmonic wave problems. Computer Physics
Communications, 203:309–330, 2016.

70 4
Stock Price Change Prediction Using News Text Mining
Marcelo Beckmann1, Nelson F.F. Ebecken1a, and Beatriz S. L. Pires de Lima1b
1
Department of Civil Engineering, Federal University of Rio de Janeiro, Centro de Tecnologia - Bloco B, Sala
101 - Ilha do Fundão, PoBox 68506, 21941-909, Rio de Janeiro - RJ - Brazil

Abstract. Along with the advent of the Internet as a way of propagating news in a digital format, came the need
to understand and transform this data into information. The advances of data mining and text mining techniques for
this purpose brought new opportunities of research applied to financial market.

This work presents a computational framework that aims to predict the changes of stock prices along the day, given
the occurrence of news articles related to the companies listed in the Down Jones Index. For this task, an automated
process that gather, clean, label, classify, and simulate investments was developed. This process integrates the existing
data mining and text algorithms, with the proposal of new techniques of alignment between news articles and stock
prices, pre-processing, and classifier ensemble. The result of experiments in terms of classification measures like
Accuracy, Precision, Recall, AUC, G-mean, F-measure, and the Cumulative Return obtained through investment
simulation outperformed the other results published in the reviewed literature.

This work also argues the techniques of accuracy measure and cross validation applied to this field of research
has too few to contribute in terms of investment recommendation for financial market, and the successful use of
text mining for this purpose must follow some precepts.

Altogether, the developed methodology and results contribute with the state of art in this emerging research field,
demonstrating that the correct use of text mining techniques is an applicable alternative to predict price
movements in financial market.
Keywords: Financial Market, Financial Economics, Stock Market Prediction, Predictive Analytics, Natural
Language Processing, Text Mining, Sentiment Analysis, Data Mining

1. Introduction

The advances in data mining and text mining, allied with the velocity and the way the news articles
are published, created opportunities to use text mining and sentiment analysis applied to financial
market prediction (TMFP). This work aims to provide scientific evidences that data mining and text
mining can be used to automatically interpret news articles and learn patterns to predict the market
movements, providing this way investment recommendations to be used by traders and automated
trading systems to achieve returns. To accomplish this objective, a complete process of data mining
and text mining was developed to predict the price movements in the stock market for the 30
companies listed in the Down Jones Industrial Index (DJIA) along the day (intraday). Because the
complex and unstable nature of financial markets, the traditional data mining algorithms were not
enough to make correct predictions, then a new data preparation technique to deal with imbalanced

Marcelo Beckmann, Ph.D., E-mail: beckmann.marcelo@gmail.com


a
Nelson Francisco Fávilla Ebecken, Ph.D., E-mail: nelson@ntt.ufrj.br
b
Beatriz S. L. Pires de Lima, Ph.D., E-mail: bia@coc.ufrj.br

71
class problem, and a classifier ensemble technique to remove class overlapping was proposed in this
work.

(Beckmann, Ebecken, & De Lima, 2017) discovered after an extensive survey related to TMFP that
around 50% of reviewed works use only Accuracy as a classification measure which is not
recommended for imbalance problems (Weiss & Provost, 2001), (Ling, Huang, & Zhang, 2003),
(Weis, 2004), (He & Garcia, Learning from Imbalanced Data, 2009), (He & Ma, Imbalanced
Learning: Foundations, Algorithms, and Applications, 2013), (Ali, Shamsuddin, & Ralescu, 2013),
and others studies use cross validation which also is not recommended for time series (Hastie,
Tibshirani, & Friedman, 2003), (Arlot & Celisse, 2010), (Bergmeir, Hyndman, & Koo, 2015).

The experiment results for the developed framework are demonstrated in terms of classification
measures like Accuracy (Sokolova & Lapalme, 2009), Precision, Recall, AUC (Fawcett, 2004), G-
mean (Barandela, Sánchez, García, & Rangel, 2003), and F-measure (Van Rijsbergen, 1979), and an
investment simulator was developed to validate the predictions generated by the classifier. The
classifier measures and the Cumulative Return obtained with the investment simulation
outperformed the other results published in the reviewed literature.

2. Methodology

The main processing flow of the proposed methodology can be seen in Fig 1, and this process is
repeated for each company listed in the DJIA, with each company owing a predictive model. Only
data mining and text mining techniques will be used, and no econometric techniques were applied
during this process. All the TMFP process was developed with RapidMiner platform and its
respective extensions (Mierswa, Wurst, Klinkenberg, Scholz, & Euler, 2006), and the innovation
proposed in this work was developed in a new extension called TradeMiner.

2.1 Data Gathering

The process depicted in Fig 1 starts with the gathering of news articles (A), from internet using a
web crawler in full operation from January to September of 2013. The source of news came from
Yahoo Finance, and Google Finance. Each news article record is composed of the news content in
English, the stock symbol, and the published date and time. The stock prices (also known as market
data) associated with the DJIA companies were collected minute by minute, using a free web service
(B). The market data were labelled (D) as SURGE, for prices with rise >= 75% of the maximum
ascent observed during the day, and PLUGE for prices with fall >= 75% of the maximum descent
observed during the day. The news articles need to be labelled (E) according the changes in the stock
prices τ minutes after its publication, by associating a label r(C) to a new article published at time t,
given a set of stock prices labels C={c(t-1), c(t), c(t+1), ..., c(t+τ)}, as explained in the equation (1).

𝑞𝑠 > 𝑞𝑝 𝑎𝑛𝑑 ∆𝐶 > 0, 𝑆𝑈𝑅𝐺𝐸 (1)


𝑟(𝐶) = {𝑞𝑠 < 𝑞𝑝 𝑎𝑛𝑑 ∆𝐶 < 0, 𝑃𝐿𝑈𝑁𝐺𝐸
𝑁𝑂𝑇 𝑅𝐸𝐶𝑂𝑀𝐸𝑁𝐷𝐸𝐷

In equation (1), qs is the number of occurrences of SURGE and qp, the occurrences of PLUNGE,

72
and the price delta before and after C, represented as C = c(t+τ+1)-c(t-1). The rationale for this
alignment proposal is: only a strong turnaround in the stock prices, and the continuous change of
prices before and after the time offsets, will make possible to identify the proper characteristics in
the news articles for a profitable trading recommendation.

Fig. 1 The text mining modelling process applied to stock price change prediction.

2.2 Data Splitting

To maximize the classifier efficacy, the predictive model is trained every week, in a technique known
in the literature as sliding window (Dietterich, 2002). The training dataset incorporates 6 months of
news articles kept in its chronological order, and the test dataset contains 1 week of new records to
evaluate the model. As the processing advances to a new week, the training dataset incorporates the
week tested previously, and discards the first week 6 months ago (F).

2.3 Training

The news articles are converted into a Bag of Words (BOW) matrix (Harris, 1954). The stop words,
terms with size less than two characters, and terms with frequency lower than 2% and greater than
95% were removed (Miner, et al., 2014), (Zhai & Massung, 2016). The use of n-grams (Sidorova &
al., 2014), which consists in a series of consecutive words of size n, with the maximum n=3 in this
work, helped to reduce the dimensionality and carries the existing semantic from the original text.
The discovered n-grams and individual words were represented as an Inverse Document Frequency
(TF-IDF) measurement (Robertson, 2004) (G). The most representative BOW features were selected
(H) by using a Pearson's Chi-Square statistic (Pearson, 1900), (Forman, 2003). This work aims to
predict only the SURGES, then to simplify the decision surface, the PLUNGE examples were
merged with the NOT RECOMMENDED (the majority class). To balance the dataset (I), the noisy
examples from majority class where removed using a novel technique called KNN Undersampling
(Beckmann, Ebecken, & De Lima, 2015). The prepared data is then submitted to a Support Vector
Machine (SVM) (Cortes & Vapnik, 1995) Machine Learning algorithm, with the LIBSVM

73
implementation (Chang & Lin, 2011), and used the Radial Basis Function (RBF) as kernel, with C
and Gamma parameters adjusted via grid search (Hsu, Chang, & Lin, 2003).

2.4 Test

The test phase consists in to apply the models generated in the training phase into the test dataset.
The word list is used to generate a new BOW with TF-IDF measurements (G). The Chi-Square word
weights are used to select the most relevant words (H), and the model generated by the SVM
algorithm is used to predict the SURGES and NOT RECOMMENDED in the test dataset. This
prediction is applied to each new article, but several news articles can be published in the same time
offset τ, however, only one recommendation needs to be given for each period of time, then a novel
classification ensemble algorithm that uses a Genetic Algorithm (Holland, 1975), (Goldberg, 1989),
(Whitley & Sutton, 2012) was developed to mitigate the class overlapping problem, by adjusting the
voting threshold for a best recommendation, given the predictive outcome of each news article at
that period of time (K).

2.3 Evaluation

The output of test phase are the classification results of 30 predictive models, corresponding to 30
stocks listed on DJIA index. These models are evaluated given the G-Mean classification measure.
If at least 10 models have a minimal value of G-Mean >= 55.00, then these models are considered
stable enough to perform an investment simulation, otherwise the entire TMFP process needs to be
adjusted (L). The investment simulation (M) consists in to use the output of (K) to buy stocks using
a short-term investment strategy. The simulation is evaluated given its Cumulative Return (CR).
Positive CRs indicates these models could be used in a real investment scenario.

Further details about this methodology and experiments can be found in (Beckmann, Ebecken, &
De Lima, 2017).

3. Experiments

To observe how the stock prices are affected by news articles, a set of experiments using the
proposed methodology demonstrated the classifier performances when predicting a SURGE or NOT
RECOMMENDED movement in the stock prices, in a period of τ=1, 2, 3, and 5 minutes after the
news article be published. The maximum averaged results after 10 runs, in terms of classification
measures such as Accuracy (99.77), Precision (99.88), Recall (92.74), AUC (67.87), G-mean (92.66),
F-Measure (76.00), and CR (21.47) are all associated with τ=1. These values outperformed the other
results found after an extensive literature review in (Beckmann, Ebecken, & De Lima, 2017). The
source and data for this methodology are open, and the instructions to reproduce these experiments
and make new developments are available in (Beckmann, Ebecken, & De Lima, 2018).

4. Conclusions

This work presented a computational framework using data mining and text mining to find patterns
between the news articles published, and the respective movements in the stock prices, creating a
predictive model to forecast the stock prices changes along the day (intraday), for the 30 companies

74
listed in the DJIA, as published in (Beckmann, Ebecken, & De Lima, 2017). The best experiment
used a time offset of 1 minute after the news article was published, and the maximum results in
terms of classification measures and Cumulative Return obtained after three months of investment
simulation outperformed the other results published in the reviewed literature. These positive results
can be accredited to the precise workflow developed, the proper use of F-measure and G-mean as
classification measures and process adjusting, and the new algorithms proposed in this work.

These results offer evidences that the stock prices movement can be predicted using text mining,
and indicates the stock prices started to be affected for the news articles in the few minutes after they
are published. Nevertheless, when the news articles are accumulated in a wider period, occurs a loss
of signal, probably because there is no mechanism developed to distinguish which news articles are
affecting the stock price, being more difficult to obtain a stable model under these conditions.
Despite the good results presented in the experiments, the association between news articles and
stock prices accumulated in a wider period deserves more attention in a future work.

References

Ali, A., Shamsuddin, S., & Ralescu, A. (2013). Classification with class imbalance problem: a review.
Int. J. Advance Soft Compu. Appl, Vol. 5, No. 3.
Arlot, S., & Celisse, A. (2010). A survey of cross-validation procedures for model selection.
Statistics Surveys, 4, 40-79.
Barandela, R., Sánchez, J., García, V., & Rangel, E. (2003). Strategies for learning in class imbalance
problems. Pattern Recognition, 36(3), pp. 849-851.
Beckmann, M., Ebecken, N., & De Lima, B. (2015). A KNN Undersampling Approach for Data
Balancing. JILSA - Journal of Intelligent Learning Systems and Applications, 7, 104-116.
Beckmann, M., Ebecken, N., & De Lima, B. (2017). Thesis: Stock Price Change Prediction Using
News Text Mining. Rio de Janeiro, Brazil: Civil Engineering Program/COPPE, Federal
University of Rio de Janeiro.
Beckmann, M., Ebecken, N., & De Lima, B. (2018). A Practical Guide for Stock Price Prediction
Using Text Mining. Rio de Janeiro, Brazil: Civil Engineering Program/COPPE, Federal
University of Rio de Janeiro.
Bergmeir, C., Hyndman, R., & Koo, B. (2015). A Note on the Validity of Cross-Validation for
Evaluating Time Series Prediction. Monash University, Departmnet of Econometrics and
Business Statistics.
Camerer, C., & Loewenstein, G. (2004). Advances in Behavioral Economics. Princeton University
Press.
Chang, C., & Lin, C. (2011). LIBSVM : a library for support vector machines. ACM Transactions
on Intelligent Systems and Technology, pp. 2:27:1-27:27.
Cortes, C., & Vapnik, V. (1995). Support-Vector Networks, Machine Learning. Machine Learning,
vol. 20, no. 3, pp. 273-297.
Dietterich, T. (2002). Machine Learning for Sequential Data: A Review. Proceedings of the Joint
IAPR International Workshop on Structural, Syntactic, and Statistical Pattern Recognition
(pp. 15-30). London: Springer-Verlag.
Fawcett, T. (2004). ROC Graphs: Notes and Practical Considerations for Researchers. HP
Laboratories.

75
Forman, G. (2003). An extensive empirical study of feature selection metrics for text classification.
Journal of Machine Learning Research 3, 1289-1305.
Goldberg, D. E. (1989). Genetic Algorithms in Search, Optimization and Machine Learning. Boston,
MA: Addison-Wesley.
Harris, Z. (1954). Distributional Structure. Word, 10, 146–162.
Hastie, T., Tibshirani, R., & Friedman, J. (2003). Model Assessment and Selection. In The Elements
of Statistical Learning, Data Mining, Inference and Prediction (pp. 245-247). New York:
Springer Series in Statistics.
He, H., & Garcia, E. (2009). Learning from Imbalanced Data. IEEE Transactions on Knowledge
and Data Engineering, Volume 21 Issue 9, 1263-1284.
He, H., & Ma, Y. (2013). Imbalanced Learning: Foundations, Algorithms, and Applications (1st
Edition ed.). Wiley-IEEE Press. doi:http://dx.doi.org/10.1002/9781118646106
Holland, J. (1975). Adaptation in Natural and Artificial Systems. Cambridge: MIT Press.
Hsu, C. W., Chang, C. C., & Lin, C. J. (2003). A practical guide to support vector classification.
Taipei, Taiwan: National Taiwan University.
Ling, C. X., Huang, J., & Zhang, H. (2003). AUC: A Better Measure than Accuracy in Comparing
Learning Algorithms. In B. C.-d. Yang Xiang (Ed.), Lecture Notes in Computer Science (Vol.
2671). Halifax, Canada.
Mierswa, I., Wurst, M., Klinkenberg, R., Scholz, M., & Euler, T. (2006). YALE: Rapid Prototyping
for Complex Data Mining Tasks. Proceedings of the 12th ACM SIGKDD International
Conference on Knowledge Discovery and Data Mining.
Miner, G., Delen, D., Elder, J., Fast, A., Hill, T., & Nisbet, R. (2014). Pratical Text Mining and
Statistical Analysis for Non-structured Text Data Applications. Elsevier.
Pearson, K. (1900). On the criterion that a given system of deviations from the probable in the case
of a correlated system of variables is such that it can be reasonably supposed to have arisen
from random sampling. Philosophical Magazine Series, 5.
Robertson, S. (2004). Understanding inverse document frequency: On theoretical arguments for IDF.
Journal of Documentation. 60 (5), 503–520. doi:10.1108/00220410410560582
Sidorova, G., & al., E. (2014). Syntactic N-grams as machine learning features for natural language
processing. Expert Systems with Applications, Vol. 41, Issue 3, 853-860.
doi:10.1016/j.eswa.2013.08.015.
Sokolova, M., & Lapalme, G. (2009). A systematic analysis of performance measures for
classification tasks. Information Processing and Management, 45, 427-437.
Van Rijsbergen, C. (1979). Information Retrieval (2nd ed.). Massachusets: Butterworths.
Weis, G. (2004). Mining with rarity: a unifying framework. ACM SIGKDD Explorations Newsletter
- Special issue on learning from imbalanced datasets Homepage archive Volume 6 Issue 1,
7-19.
Weiss, G., & Provost, F. (2001). The Effect of Class Distribution on Classifier Learning: An
Empirical Study. Technical Report MLTR-43, Dept. of Computer Science, Rutgers
University.
Whitley, D., & Sutton, A. (2012). Genetic Algorithms — A Survey of Models and Methods. In G.
Rozenberg, T. Bäck, & J. Kok, Handbook of Natural Computing (pp. 637-671). Springer.
Zhai, C., & Massung, S. (2016). Text Data Management and Analysis: A Practical Introduction to
Information Retrieval and Text Mining. ACM. doi:10.1145/2915031

76
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

CONCRETE BEAM WITH FIXED ENDS ON ELASTIC FOUNDATION


SUBJECTED TO THE UNIFORMLY DISTRIBUTED LOADING: NUMERICAL
ANALYSIS WITH FTOOL® AND SAP2000®

Breenda Lorrana Vieira Lima1, Michell Macedo Alves2


1
Instituto Federal de Educação, Ciência e Tecnologia Goiano, breendalorranavieiralima@gmail.com
2
Instituto Federal de Educação, Ciência e Tecnologia Goiano, michell.macedo@ifgoiano.edu.com

1. Introduction
Entre os vários problemas de engenharia a serem estudados, um de grande interesse para a engenharia
de fundação, é o problema de contato entre vigas e as bases elásticas nas quais elas se apoiam. Esse
conceito foi introduzido por [1], baseando-se na hipótese de [2], em que o modelo da base elástica é
caracterizado fisicamente por um sistema de molas com rigidez equivalente à do solo. Nesse trabalho,
especificamente, será abordado o estudo de vigas de concreto com extremidades engastadas sobre base
elástica submetidas a carregamento uniformemente distribuído. O objetivo principal é apresentar
metodologias de modelagem da base elástica por meio das ferramentas fornecidas pelos softwares
Ftool® [3] e SAP2000® [4]. Dessa forma, buscou-se apresentar formas mais adequadas de disposição do
conjunto de molas sob a viga e o valor da constante de mola, a fim de se obter valores de deformações
e esforços solicitantes com boa aproximação, comparados com aqueles determinados pelas formulações
analíticas de [1]. Essa é uma alternativa mais acessível e simples de análise do problema, em comparação
aos altos custos computacionais gerados pelos softwares de Método dos Elementos Finitos.

2. Analytical Formulation

Conforme [1], considere uma viga sobre base elástica sujeita às cargas verticais. Devido a ação dessas,
a viga fletirá e surgirão forças de reações distribuídas ao logo do contato viga-solo, no sentido oposto
ao deslocamento vertical da viga, denominadas de pressões de contato 𝑝 (Fig.1 (a)). Para [2] a
intensidade da pressão de contato em um ponto da viga, é diretamente proporcional a deflexão da viga
𝑦 nesse mesmo ponto. Significa que o comportamento da base é elástico-linear - “Lei de Hook”, e
caracterizada pela constante de mola 𝑘. Em vigas cuja a base elástica são camadas de solo, a constante
de mola, depende do produto da largura 𝑏 da viga em contato com o solo e o coeficiente de recalque do
solo 𝑘0 (Eq.1). [1] definiu o parâmetro 𝜆 que engloba a rigidez da viga à flexão e a elasticidade da base
(Eq.2). E, para [5], pode-se classificar as vigas sobre base elástica em três grupos conforme a rigidez
relativa: curtas (𝜆 𝑙 < 𝜋/4), médias (𝜋/4 < 𝜆 𝑙 < 𝜋) e longas (𝜆 𝑙 > 𝜋).
Pela condição de equilíbrio em um elemento infinitesimal 𝑑𝑥 de uma viga sobre base elástica sujeita a
um carregamento uniformemente distribuído 𝑞. Na qual atuam os esforços solicitantes e seus
acréscimos: momento fletor 𝑀 e esforço cortante 𝑉 (Fig.1 (b)). E, utilizando-se as expressões
fundamentais da Resistência dos Materiais de [6], [7]. [1] determinou a equação diferencial geral
fundamental da linha elástica de viga sobre base elástica (Eq.3). Cuja solução 𝑦(𝑥) para o problema
abordado é (Eq.4).

77
(a) (b)

Figura 1: Viga sobre base elástica: (a) pressões de contato; (b) elemento infinitesimal

𝑘 = 𝑏 𝑘0 (1) 𝜆 = ∜(𝑘/(4 𝐸𝐼)) (2) 𝐸𝐼 𝑑4 𝑦/𝑑𝑥 4 = − 𝑘 𝑦 + 𝑞 (3)

𝑞 𝑠𝑒𝑛ℎ(𝜆𝑥) 𝑐𝑜𝑠(𝜆(𝑙 − 𝑥)) + 𝑠𝑒𝑛(𝜆𝑥) 𝑐𝑜𝑠ℎ(𝜆(𝑙 − 𝑥)) + 𝑠𝑒𝑛ℎ(𝜆(𝑙 − 𝑥)) 𝑐𝑜𝑠(𝜆𝑥) + 𝑠𝑒𝑛(𝜆(𝑙 − 𝑥)) 𝑐𝑜𝑠ℎ(𝜆𝑥)
𝑦(𝑥) = [1 − ]
𝑘 𝑠𝑒𝑛ℎ(𝜆𝑙) + 𝑠𝑒𝑛(𝜆𝑙)
(4)
3. Modeling Methodologies
Os procedimentos para a modelagem da viga sobre base elástica na interface dos softwares consistem
em: 1º) Criar a geometria (nós e barras); 2º) Aplicar os apoios às extremidades da viga (apoios de 3º
gênero/engastes); 3º) Criar e atribuir o material (concreto) e as dimensões e formatos das seções
transversais (retangulares); 4º) Atribuir a base elástica ( conjunto de molas independentes caracterizadas
pela constante de mola e rigorosamente espaçadas sob os pontos da viga) e; 5º) Aplicar o carregamento
ao modelo. Para a modelagem específica da base elástica, apresenta-se três metodologias:
I) Ftool® v. 3.01 - Apoios Elásticos: Para cada segmento de 1 metro de viga é necessária 1 mola sob a
metade desse segmento. A quantidade de molas equivale ao valor do comprimento da viga. É necessário
informar: direção (eixo Y) e valor da constante de mola por metro de viga na direção Y (Fig.3).
II) SAP2000® v. 19 - “Joint Springs”: Semelhante a metodologia utilizada no Ftool®: quantidade e
distribuição das molas. É necessário informar: tipo (simples); direção (impede parcialmente a translação
na direção da coordenada Z do eixo global) e valor da constante de mola por metro de viga (Fig.4 (a)).
III) SAP2000® v. 19 - “Line Springs”: Atribui aos pontos associados aos tramos da viga um conjunto de
molas distribuídas linearmente. É necessário informar: tipo (simples); valor da constante de mola por
metro de viga; comportamento da mola aos esforços (somente compressão) e vetor-direção da linha de
molas (componente Z do eixo global) (Fig.4 (b)).
Logo, a quantidade e disposição dessas molas sob a viga respaldam-se no valor da constante de mola
que representa a rigidez do solo (mola) por um metro de viga.

Figura 2: Viga sobre base elástica - Ftool® v. 3.01

(a)

(b)

Figura 3: Viga sobre base elástica - SAP2000® v. 19: (a) Joint Springs; (b) Line Springs

78
4. Numerical Examples

Segue os modelos de viga sobre base elástica, ilustrados na (Fig.4) e (Tab.1).

Figura 4: Viga biengastada de concreto sobre base elástica - carregamento uniformemente distribuído

Tabela 1: Propriedade das vigas sobre base elástica – exemplos numéricos


Geometria Módulo de Coeficiente de Solicitaçã
Modelo Viga
𝑙 (20x40) Elasticidade 𝐸 Recalque 𝑘0 o𝑞
V05AF média 5m Areia Seca Fofa
V10AF longa 10 m 12 753 kN/m³
Concreto
V05AC longa 5m Areia Seca 10 kN/m
25 GPa
Compacta
V10AC longa 10 m
166 770 kN/m³

A modelagem das vigas de 10 metros sobre base elástica consistiu na criação de uma barra contendo
100 pontos espaçados de 10 cm, ou 100 tramos de 10 cm, e atribuição de 10 molas sob os pontos
estratégicos. Para as vigas de 5 metros, 50 pontos espaçados de 10 cm e aplicação de 5 molas. Enquanto,
para as vigas de 1 metro, composta de 100 tramos de 1 cm e 1 mola. O objetivo é mapear os resultados
para melhor verificação.

5. Results
Observou-se que as configurações dos diagramas são semelhantes para as vigas sobre base elástica
modeladas tanto na interface Ftool® como SAP2000® - “Joint Springs”. Fato que se justifica pelo
princípio de ambas modelagens serem a mesma.
Notou-se que as vigas sobre areia seca fofa, para as três metodologias, os valores do deslocamento
vertical e rotação de tangente à elástica foram iguais dos analíticos. Enquanto, o diagrama de momento
fletor presenciou pontos angulosos no meio do vão da viga, em especial, as vigas de 10 m , para a
modelagem com o Ftool® e SAP2000® - “Joint Springs” (Fig.5(a)). O mesmo não aconteceu na
modelagem com o SAP2000® - “Line Springs”. O diagrama de esforço cortante, nos pontos de molas
apresentou-se descontinuidades nos valores, significativamente nas vigas de 10 m. Contudo, fazendo a
média entre os valores nas seções logo antes e depois dos pontos de mola, observou que o resultado
aproximou do analítico (Fig.5(b)). Esse comportamento não aconteceu com o recurso “Line Springs”.
Quando a base elástica se torna mais rígida, vigas sobre areia seca compacta, os efeitos mencionados
anteriormente são mais significativos, tanto na viga de 5 e 10 metros. Mesmo assim, o comando “Line
Springs” continua sendo um recurso viável (Fig.6).

79
(a) (b)

Figura 5: Vigas de concreto sobre areia seca fofa: (a) Momento fletor; (b) Esforço cortante

Figura 6: Esforços solicitantes em vigas de concreto sobre areia seca compacta

6. Conclusions

Os exemplos numéricos possibilitaram compreender que a modelagem utilizando o recurso de molas do


Ftool® é viável para a determinação das deformações da viga (deslocamentos verticais e rotação da
tangente à elástica). Em contrapartida, nos resultados dos esforços solicitantes, apresentaram-se pontos
angulosos no diagrama de momento fletor e descontinuidades nos diagramas de esforço cortante. Esses
pontos angulosos e descontinuidades tendem a aumentar quanto mais longas forem as vigas e mais rígida
for a base elástica, ou seja, quanto maior for a constante de mola. Referente às análises de vigas feitas
com o software SAP2000®, especificamente utilizando o comando “Joint Springs”, verificou-se que os
resultados obtidos são semelhantes aos do Ftool®. Por outro lado, através do comando “Line Springs”
do software SAP2000® foi possível obter resultados dos esforços solicitantes de maneira mais precisa,
sem pontos angulosos ou descontinuidades nos diagramas de momento fletor e esforço cortante,
respectivamente. Em síntese, constatou-se que quanto mais longa for a viga e mais rígido o meio
elástico, maior serão as discrepâncias dos valores dos esforços solicitantes obtidos mediantes a utilização
do software Ftool® e SAP2000® (comando “Joint Springs”) comparados com a solução analítica.

80
Acknowledgments

Ao Instituto Federal de Educação, Ciência e Tecnologia Goiano Campus Rio Verde (IFGoiano).
À empresa Multiplus®, representante no Brasil do SAP2000® v. 19 e ao Luiz Fernando Martha, autor do
Ftool® v. 3.01, os quais autorizaram a publicação dos resultados mediante a utilização dos softwares.

References

[1] Hetényi, M; (1946). Beams on Elastic Foundation. Scientific Series, vol. XVI. Ann Arbor: The
University of Michigan Press, University of Michigan Studies.
[2] Winkler, E; (1867). Die Lehre von der Elastizitat und Festigkeit. Dominicus, Prague.
[3] FTOOL® - Two – Dimensional Frame Analysis Tool. Educational Version 3.01; 2015.
[4] SAP2000® - “Static and Dynamic Finite Element Analysis of Structure.Structural Analysis
Program’’. Advanced 19, Computers and Structures, Inc; 2016.
[5] Süssekind, J. C; (1980). Curso de Análise Estrutural 2: Deformações em Estruturas e Métodos das
Forças. Vol. 2. Ed. 4. Porto Alegre: Editora Globo.
[6] Hibbeler, R. C; (2010). Resistência dos materiais. Ed. 7. São Paulo: Pearson Prentice Hall.
[7] Beer, F. B.; Johnston, E. R; (2015). Mecânica dos Materiais. Ed. 7. São Paulo: McGraw Hill.

81
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

TOWARDS A METHODOLOGY FOR THE OPTIMAL DESIGN OF A GROUP OF RISERS IN


LAZY-WAVE CONFIGURATION

Bruno da Fonseca Monteiro1, Bruno Martins Jacovazzo2, Carl Horst Albrecht3, Breno
Pinheiro Jacob4
1
Polytechnic School of Federal University of Rio de Janeiro, bruno.monteiro@poli.ufrj.br
2
Polytechnic School of Federal University of Rio de Janeiro, brunojacovazzo@poli.ufrj.br
3
Polytechnic School of Federal University of Rio de Janeiro, carl@poli.ufrj.br
4
Laboratory of Computer Methods and Offshore Systems, LAMCSO - PEC/COPPE/UFRJ,
breno@lamcso.coppe.ufrj.br

This paper describes the first steps towards of an innovative computational tool, based on evolutionary
optimization methods, for the synthesis and optimization of a group of risers in lazy-wave
configuration that connect a floating platform at the sea surface to the wellheads at the sea bottom. In
previous works [1, 2, 3], some authors have presented such an optimization tool based on evolutionary
algorithms for the synthesis and optimization of lazy-wave riser; however, considering only one
individual riser. The objective in that case was reduce costs, only. The main advance of this work
consists in the optimization of a set of close risers among themselves, where in addition to minimizing
the cost, it is sought to avoid the clash between them. Moreover, for optimization assessment,
geometric parameters are taken as the design variables and the design constraints consider for
structural integrity. Extreme environmental conditions of a Brazilian offshore field are used in the case
study. The results open many roads for future developments, indicating that evolutionary algorithms
(EAs) can provide an efficient riser system with multiple lines.

References

[1] Pina AA, Albrecht CH, de Lima BSLP, Jacob BP. Tailoring the particle swarm optimization
algorithm for the design of offshore oil production risers, Optimization and Engineering, 12:215–235,
2011. DOI:10.1007/s11081-009-9103-5.
[2] Vieira IN, Jacob BP, Lima BSLP, Bio-inspired algorithms for the optimization of offshore oil
production systems, International Journal for Numerical Methods in Engineering (Print) , v. 1, p.
n/a-n/a, 2012. http://onlinelibrary.wiley.com/doi/10.1002/nme.4301/pdf
[3] Monteiro BF, Pina AA, Baioco JS, Albrecht CH, Lima BSLP, Jacob BP, Toward a methodology
for the optimal design of mooring systems for floating offshore platforms using evolutionary
algorithms, Marine Systems & Ocean Technology, v. 1, p. 1, 2016.
https://link.springer.com/article/10.1007%2Fs40868-016-0017-8

82
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

A COMPARATIVE STUDY OF NEIGHBOURING PARTICLE SEARCH METHODS


IN COMPUTATIONAL FLUID MECHANICS: DIRECT SEARCH AND LINKED
LIST TECHNIQUES EVALUATION

Carlos Alberto Dutra Fraga Filho1, Lucas Lustosa Schuina1, Brenda Silva Porto1

1
Development, Implementation and Application of Computational Tools for Problem
Solving in Engineering Research Group, Federal Institute of Espírito Santo,
Mechanical Engineering Coordination, Av. Vitória, 1729
29040-780 – Jucutuquara, Vitória, ES, Brazil
email: cadff1@gmail.com

Abstract. Particle methods are being increasingly employed in solving problems in


Engineering, Applied Mathematics and Physics. Regardless of the problem studied, the search
for neighbouring particles must be done at each numerical iteration. The aim of this paper is to
present results found in the particle search process using the direct method and the linked list,
allowing a discussion about the efficiency of both methods and the variables that influence it.
Two classical hydrodynamic problems (dam breaking and lid-driven cavity) have been
simulated. The particle numbers and the interpolation functions (kernels employed in the SPH
particle method) varied in simulations. Results for different combinations of kernel/ particle
numbers were found. The linked list algorithm implemented presented similar performance to
direct search. The evaluation of proposals of maintaining the neighbours list without
modifications during some numerical iterations, aiming to improve computational efficiency,
needs to be performed.

Keywords: neighbour particle search; direct search; linked list; computational efficiency

1. Introduction
The literature on Lagrangian particle models presents several techniques for the search of
neighbouring particles. Two of these techniques are the direct search and linked list. In the
first and more simple technique, a reference particle is fixed and the distances to all other
particles are found. If a distance is smaller than the cut-off radius (kh), a neighbouring particle
is found and stored in a matrix with a number of lines equal to the number of particles in the
domain and a number of columns equal to the expected number of neighbouring particles for
each fixed particle.
Figure 1 presents the flowchart of the direct search algorithm and the neighbourhood of a
fixed particle (region within the circumference of radius equal to kh). Figure 2 presents the
algorithm of the second technique implemented in this work (the linked list algorithm).

83
Figure 1. The direct search algorithm and the graphical representation of a fixed particle and its
neighbourhood.

Figure 2. The linked list algorithm implemented.

84
The second technique implemented and tested was the linked list technique [1-3]. In this
method, the domain is divided into a grid containing a certain number of cells. Each cell
contains a number of particles that can vary during the numerical simulation. Figure 3 shows
the grid and cells defined in this technique. The reduced search region is shaded in light blue.

Figure 3. The region in which the search is performed in the linked list method.
1. Objective
The main objective of this work is to implement two searching techniques: direct search and
linked listed in order to evaluated their computational efficiencies.
2. Findings
Two hydrodynamic cases, with the solution of the equations of motion being provided by the
SPH method, were evaluated in a 2-D domain: dam breaking and lid-driven cavity flow.
3.1. Dam breaking
The 2-D geometry simulated was a tank with a height a length of 0.420 m and a height of
0.440 m. The damned water had a height of 0.114 m height and and a width of 0.114m. The
simulation finished when the wave achieved the left wall of the reservoir. The physical-
mathematical modelling of this dynamic problem can be found in [4]. The interpolation
function/ particle number combinations and the results found are shown in Table I.
Table I. Processing time for differents kernel/ number of particles in dam break simulation

Number of particles Interpolation CPU time


function Direct search Linked List
1296 Cubic spline 2h36m25s 3h29m20s
1296 Lucy’s Quartic 1h40m13s 3h23m7s
1296 Quintic spline 2h33m51s 4h30m28s
Considering the surprising results found in the first simulations, with lower performance
of the linked list, more studies were carried out to optimize the first algorithm implemented.
Liu and Liu (2003) [5] presented a form of storage of neighbouring particles in pairs, which
has the following characteristics:
 along the process of searching for neighbours of a fixed particle i, after the comparison
between the interparticular distances and the cutting radius, a particle j is identified as a
neighbour of i. The pair of particles, i and j, is stored.
 in the search for neighbours of the particle j, we take advantage of the information that the
particle i is a neighbour of the particle j (it is not necessary to redo the computational operations
to locate this pair of neighbours).

85
Thus, fewer computational operations will be performed and it is expected that there will be
a decrease in CPU time.
2.2. Lid-driven cavity flow
The modified routines have been employed in simulations of a square lid-driven cavity,
isothermal flow. The sides of the square cavity were 1.0 × 10−3 m in length. The modelling of
this problem is available in [6]. The simulation finished when the steady-state has been
reached. The results found are shown in Table II.
Table II. Processing time for different kernel/ number of particles combinations
Interpolation Number of particles CPU time (s)
function (per side of cavity) Direct search Linked List
31x31 294 330
Cubic spline 41x41 580 548
51x51 1049 1053
31x31 297 337
Quintic spline 41x41 655 657
51x51 1147 1149
The FORTRAN Programming Language was employed in the serial computational codes
used in simulations, which were performed in a computer with 4GB of RAM memory, Intel
Core i5-3210M 3rd generation processor, and Linux operational system.
3. Concluding remarks and future research
The analysis of the results provided by the implemented search algorithms allowed to conclude
that the simple use of the linked list technique is not sufficient for the best efficiency in the
process of neighbouring particle searching. The best results were found when larger numbers
of particles were employed in the discretization of the domain. The storage of neighbouring
particles in pairs promoted a significant improvement in computational efficiency, which is in
accordance with the conclusions presented in [5].
There are proposals of maintaining the neighbours list (without modifications) during some
numerical iterations [1-3]. They need to be tested in order to verify the correct neighbours
identification (accuracy maintenance) and improvement of the computational efficiency.
References
[1] Viccione G., Bovolin V., Carratelli E. P. Defining and optimizing algorithms for
neighbouring particle identification in SPH fluid simulations. Int. J. Numer. Meth. Fluids
58:625–638, 2008.
[2] Dominguez J. M., Crespo A. J. C., Gómez-Gesteira M., Marongiu J. C. Neighbour lists in
smoothed particle hydrodynamics. Int. J. Numer. Meth. Fluids 67(12): 2026-2042, 2011.
[3] Winchenbach R., Hochstetter H., Kolb A. Constrained Neighbor Lists for SPH-based Fluid
Simulations. Proceedings of Eurographics/ ACM SIGGRAPH Symposium on Computer
Animation, Zurich, Switzerland, 2016.
[4] Fraga Filho C.A.D., An algorithmic implementation of physical reflective boundary
conditions in particle methods: Collision detection and response. Physics of Fluids 29, 113602,
2017.
[5] Liu G.R., Liu M.B. Smoothed Particle Hydrodynamics: a Meshfree Particle Method.
World Scientific, Singapore, 2003.
[6] Fraga Filho C.A.D., Chacaltana J.T.A., Pinto W.J.N. Meshless Lagrangian SPH method
applied to isothermal lid-drive cavity flow at low-Re numbers. Comp. Part. Mech., 2018.
https://doi.org/10.1007/s40571-018-0183-x

86
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

CONNECTION BETWEEN BERNOULLI-EULER RODS AND KIRCHHOFF-LOVE SHELLS


Cátia da Costa e Silva1, Sascha Maassen2, Nils Viebahn3, Paulo de Mattos Pimenta4, Jörg
Schröder5
1
Polytechnic School at University of São Paulo, Brazil, catiacosta@ifsp.edu.br
2
Universität Duisburg-Essen, Germany, nils.viebahn@uni-due.de
3
Universität Duisburg-Essen, Germany, sascha.maassen@uni-due.de
4
Polytechnic School at University of São Paulo, Brazil, ppimenta@usp.br
5
Universität Duisburg-Essen, Germany, j.schroeder@uni-due.de

ABSTRACT
This work presents a geometrically exact Bernoulli-Euler rod [5] and a Kirchhoff-Love Shell [4]
connection model. For the rods transversal shear deformation is not accounted for and the shell is a shear
rigid model. The connection between the two models are made by having de displacements on both
elements be the same and making the rotation compatible as well. The parameterization of the rotation
field is done by the rotation tensor with the Rodrigues formula that makes the updating of the rotational
variables very simple. This parametrization can be seen in [5]. Cubic Hermitian interpolation for the
displacements together with a quadratic Lagrange interpolation for the torsion angle of the rod were
employed within the usual Finite Element Method, leading to adequate C1 continuity. For the shell a
consistent plane stress condition is incorporated at the constitutive level of the model. A triangular finite
element, with a quadratic interpolation for the displacements are applied for the shell model. The
connection between the two models is done by a Lagrange multiplier. This model can be used in any
case of rod and shell working together like with reinforced shells.
1. INTRODUCTION
The objective of this work is to present the compatibility of the geometrically exact Bernoulli-Euler rod
formulation and Kirchhoff-Love shell formulation. The basic kinematical assumption that select the
class of admissible motions is obtained by imposing that rod cross sections that are initially orthogonal
to the chosen axis remain orthogonal to it after deformation for the rod. The basic kinematical
assumption for the shell is based on a deformation gradient written in terms of the first- and second-
order derivatives of the displacements. The resulting finite element formulation provides C1-continuity
using a penalty approach, which penalizes the kinking at the edges of neighboring elements. This
approach enables the application of well-known C0-continuous interpolations for the displacements,
which leads to a simple finite element formulation, where the only unknowns are the nodal
displacements. Our theory is called geometrically exact because no approximation is employed after the
basic kinematical assumption is made. Displacements and rotations can be unlimited large.
2. ROTATION RODRIGUES PARAMETERS
As seen in  denote the vector of Rodrigues parameters. The Rodrigues parameterization furnishes
4  
ˆ     
Q    1 2  , where   Skew    and 2     . (2.1)
4 2 2 

Q  ei  eir .
Vector  can be obtained from with the aid of

87
4
 axial  SkewQ  (2.2)
1  trQ

The curvature  is
  axial Q QT   Ξ  and r  QT   axial QTQ    ΞT  , (2.3)

and  is its back-rotated counterpart with


r

4  
Ξ   I  1   . (2.4)
4      2 
2.1. INCREMENTAL DESCRIPTION

Let
i and
 i 1
denote a quantity
  at instants ti
and ti 1
, respectively. And let


be an
incremental quantity. Thus, one gets for the rotation tensor the following relations
Qi 1  QQi , where
ˆ  i 1  , Q  Q
ˆ    ˆ  i  . (2.5)
Qi 1  Q and Qi  Q

We recall the following result by Rodrigues, which is probably the most relevant result by him,
4  
i 1       1     (2.6)
4  i    i
2 i 


In the incremental description we have


4  
Ξ   I  1    . (2.7)
4   2  2 

From [5] for the rod we have

e3i  e3i 1   
2 1
  e3m  e3m e3m , (2.8)
where  is the incremental torsion angle.

With aid of (2.2), (2.9) is used for the shell

2
  e ij 1  e ij (2.9)
1  eki 1  eki

3.GEOMETRICALLY EXACT BERNOULLI-EULER ROD KINEMATICS

Let
e r , e r , e r 
1 2 3
be an orthogonal system placed at the reference configuration of the rod. The vectors
e ,   1, 2
r
r
, are placed on the cross section of the rod, shown in [5]. Thus, e3 is orthogonal to this plane.
The position of the rod material points in the reference configuration is  and r the director
r

    r r ,   e3r , r r  er . (3.1)

    er      0,  
We introduce the coordinate 3 , , where  is the rod length at reference
configuration. In the current configuration, the position of the material points is given by
1
x  z  r , r  Qr r , with e3  z  z  , and z   e3r  u  and z   u  . (3.2)

88
Introducing a linear elastic material law, a simple potential for the beam can be formulated based on
 r 
r   r 


generalized cross-sectional strains  
, incorporating the axis stretch and curvature. The strain
energy per unit reference length is then given by
1 r
  r     Dr . (3.3)
2
A constitutive tensor D is used to model linear elasticity in this case. The potential will be minimized
using the standard finite element procedure.
4. GEOMETRICALLY EXACT KIRCHHOFF-LOVE SHELL KINEMATICS
The middle surface of the shell is plane in the reference. Let
e r , e r , e r 
1 2
be an orthogonal system with
3

r r
the vectors e placed on the shell reference mid-plane and e3 normal to this plane. The position of the
rod material points in the reference configuration is  and r is the director just like in the rod.
r

    r r ,   er , r r  3e3r . (4.1)


In the current configuration, the position of the material points is given by
x  z  r , r  Qr r , z    u , (4.2)

  , 

With

z, = er  u, and z,  u, with   ,  (4.3)


1
e1  z,1 z,1 ,
1
e3  z,1  z,2 (z,1  z,2 ), (4.4)
e2  e3  e1 .

5. CONTINUITY CONSTRAINS

Figure 1: general case the finite elements threshold showing field continuities
continuity constraint: S =R
constraint functional: con = (S  R )  0 (5.1)
with S , R indicating sides of each element threshold, shell and rod respectively
The procedure to induce satisfaction of this constraint states that by this constraint the rotation field
appears as a continuous field.

89
Here the Lagrange multiplier approach is introduced that ensures the constrain in an exact manner.
6.EXAMPLE
To show the applicability of the formulation a simple example shown in Figure 2 was computed with
an elasticity module of 21000 and Poisson of 0.3125. Fist the beam was clamped on extremity A (case
A) then, to evaluate just the torsion, the displacements degrees of freedom were set to zero also on
extremity B (case B). As seen on table 1 the example shows a similar response to the linear problem,
then one can conclude that this connection is working.

Figure 2: shell + rod example

Non-Linear Non-Linear
Linear
Case A Case B
ux 0 2,6E-05 0
uy 0 9,5E-05 0
uz 0,05362 0,05389 0
uy 0,00135 0,00135 0
 0,00354 0,00367 0,00367
Table 1 :Values at point B

7. CONCLUSIONS
This work is relevant to structural analysis because there are many practical cases that need shells
coupled with rods. And the connection between two analogous theory was made and granted good
results. A very simple example was computed to show that to connection through the rotation degree of
freedom works the results were compared with a linear analysis and it showed similar response. These
rod and shell theories are used for very flexible structures, like thin shells and slender rods.

REFERENCES
[1] Pimenta P. M. and Yojo T., “Geometrically-exact analysis of spatial frames”, Applied Mechanics
Reviews, ASME, New York, v.46, 11, 118-128, 1993.
[2] Pimenta P. M., “Geometrically-Exact Analysis of Initially Curved Rods”, in: Advances in
Computational Techniques for Structural Engineering, Edinburgh, U.K., v.1, 99-108, Civil-Comp
Press, Edinburgh,1996.
[3] Pimenta P. M., Campello E. M. B., “Shell curvature as an initial deformation: geometrically exact
finite element approach”, Int. J. Num. Meth. Engrg., 78, 1094-1112, 2009.
[4] Viebahn, N., Pimenta, P.M. & Schroeder, J., “A simple triangular finite element for nonlinear thin
shells - Statics, Dynamics and anisotropy”, Computational Mechanics, online, 2016.
[5] Silva, C.C., Maassen, S., Pimenta, P.M. & Schröder, J. “Geometrically exact analysis of Bernoulli-
Euler rods” submitted to Computer Methods In Applied Mechanics and Engineering, 2018

90
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

NUMERICAL MODELING OF THE BRAZE WELDED ASSEMBLY OF A COPPER COATED


WITH A NICKEL-BASED ALLOY AND STEEL

T. Chawki1, 2, 3, C. Bertoni2, E. Feulvarch1, H. Klocker3, J.M. Bergheau1


1
National School of Engineers of Saint-Etienne, Chawki.Tahri@enise.fr, eric.feulvarch@enise.fr,
jean-michel.bergheau@enise.fr
2
Federal Mogul Powertrain Ignition Products, Christophe.Bertoni@federalmogul.com
3
National School of Mines of Saint-Etienne, klocker@emse.fr

Abstract
Ever since the invention of the internal combustion engine, an essential material has always been present
in spark plug: Nickel-based alloys. Over the years, these materials gained great importance. They are
widely used in automotive and aerospace, as they can withstand high homogeneous temperature,
aggressive corrosion, different stress levels, etc. Manufacturing of more and more complex structures
makes the selection of materials the most important factor. Independently from the level of stress,
temperature and corrosive environment, components must be reliable during service. Alloys that can
meet the above mentioned requirements are often high-priced materials which require advanced
technologies during the production of components [1]. Resistance welding is one of the micro-joining
processes, in which a weld is performed between two workpieces; shell (steel) and copper coated with
a nickel-based alloy (Fig. 1.a), whose chemical compositions are given in table 1.

Table 1: Chemical compositions of the samples.


Ni Cr Fe Al C Mn S Si Bi Cu

Gradient

Nickel alloy 58% 21% balance 1% 0.1% 1% 0.015% 0.5% - 1%

Copper - - - - - - - - 0.0005 99.95%

The copper coated with a nickel-based alloy is welded, through an electric resistance machine, to the
shell using a high frequency direct current (Fig. 1.a). The welding frequency plays a major role in the
welded zone quality (Fig. 1). Optimization of the welding sequence and process is one way to reach
this goal. However, experimental optimization requires prototyping and measurements which are
extremely expensive and time consuming and finally, very few solutions can be used. For the numerical
simulation of high resistance welding using high frequency, the literature is not very abundant. The
presence of eddy currents changes indeed the distribution of power due to joule effect. Thereby changing
the joule effect along the contact interface and consequently strongly influencing the quality of the weld
achieved. Finite-element simulations can be used in that aim but the difficulty is that welding processes
involve complex geometries and physical phenomena.

91
2D Model
Nickel-based
alloy

Copper Gripper

Gripper
Current density J
h (

H
Shell Shell Y
(a) X (b)

Copper coated with a


nickel-based alloy Welded zone

Shell

(c)
Figure 1 : (a) Scheme of the welding strategy, (b) 2D modeling of welding process and (c) 3D modeling of welding process.

The local models are insufficient to predict residual stresses and distortions, only global 3D models can
correctly evaluate it [2]. These types of simulations require substantial computing times and
performance. The idea is to compare the different results coming from an electrokinetic model using a
3D and a 2D (with different thicknesses (e) (Fig. 1.b, c)) geometry. The results obtained using a
simplified two-dimensional model are discussed and compared to results from a complete 3D
simulation. Once the appropriate thickness (e) is determined (Fig. 1.a, b), electrodynamic-thermal
simulations are performed using a 2D model. This new procedure is suitable to investigate the influences
of the welding machine parameters, such as the frequency (50 and 5000 Hz), the gripper’s altitude (H),
and the copper’s altitude (h) (Fig. 1.a) as well as other factors that contribute to weld quality. A
comprehensive analysis procedure has been developed to perform the incrementally coupled thermal-
electrical-mechanical analysis to simulate the resistance welding of nickel-based alloys. These
simulations are accompanied with experiments.

Keywords: Resistance welding. Physical couplings. Electrodynamic-thermal. FEM.

92
References

[1] Pollock T. M., Tin S., Nickel-based superalloys for advanced turbine engines: chemistry,
microstructure, and properties. Journal of Propulsion And Power, 22, 2 (2006) 361-374.
[2] Feulvarch, E, V Robin, and J M Bergheau. “Thermometallurgical and Mechanical Modelling of
Welding – Application to Multipass Dissimilar Metal Girth Welds.” Science and Technology of
Welding and Joining 16, no. 3 (April 2011): 221–31.

93
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

O N THE NUMERICAL SIMULATION OF DISCRETE FRACTURE NETWORKS


Philippe R B Devloo 1 , Chensong Zhang 2
1 FEC/UNICAMP, phil@fec.unicamp.br
2 Chinese Academy of Science, zhangcs@lsec.cc.ac.cn

Different formulations and numerical approximations are presented for the simulation of discrete
fracture networks in two dimensions. In all cases the fracture flow is represented by a one dimen-
sional flow coupled with a two dimensional flow through porous media. The different formulations are
compared in terms of computation of total flux, conservation properties and size of global system of
equations.
Abstract We compare H 1 approximation, H(div) approximation, H(div) approximation with MHM
(Multiscale Hybrid Mixed) multiscale reduction and H(div) with SBFem enhanced resolution at the
fracture tips.
The different formulations are implemented using the NeoPZ programming environment that is freely
available on github [https://github.com/labmec/neopz.git||NeoPZ-github].

94 1
Some Results About Existence and Uniqueness
Solution for Stochastic Bending Beam
∗1 *2
Cláudio R. A. da Silva Júnior , Pedro D. Danizete
1
Graduate Program in Mechanical Engineering and Materials (PPGEM), Federal Technological
University of Parana (UTFPR), Curitiba, Paraná, Brazil.
2
Graduate Program in Mathematics, Federal University of Parana, Curitiba, Paraná, Brazil.

(Received keep as blank , Revisedkeep as blank, Accepted keep as blank )

Abstract. This paper present a theorem of existence and uniqueness for solution by stochastic beam
bending. The used stochastic version of the Lax-Milgram, Babuska 2004, for demonstration this
theorem. Present results of the existence and uniqueness for theoretical solutions for theory of beam
Euler-Bernoulli and Timoshenko.
Keywords: Stochastic Bending Beam; Lax-Milgram lemma; Euler-Bernoulli's beam; Timoshenko's
beam

1. Introduction

The field of stochastic mechanics has been subject of extensive research and significant
developments in recent years. Stochastic mechanics incorporates the modeling of randomness or
uncertainty in the mathematical formulation of mechanics problems. This is in contrast to the more
established field of structural reliability, where uncertainty and randomness are also addressed, but
where problem solutions are obtained mainly based on deterministic mechanics models.
The analysis of stochastic engineering systems has received new impulse with use of finite
element methods to obtain response statistics. Initially, finite element solutions where combined
with the Monte Carlo method, and samples of random system response where obtained.
Perturbation and Galerkin methods where used in this context, Araújo and Awruch (1994). Such
methods allowed representation of uncertainty in system parameters or in loads by means of
stochastic processes. At the end of the 80’s, Spanos and Ghanem (1989) used the Galerkin finite
element method to solve a stochastic beam bending problem, where Young’s modulus was
modeled as a Gaussian stochastic process. The space of approximate solutions was built using the
finite element method and chaos polynomials. These polynomials form a complete system in
L 2 ( Ω,F , P )
L2 ( Ω , F ,P ) = Ψ , where Ψ = span {ψ i }i =0  is the space generated by the chaos

 

∗Corresponding author, Professor Claudio Avila, E-mail: avila@utfpr.edu.br

95
polynomials and ( Ω, F , P ) a probability space. The ideas presented in this study where
innovative and represented a new method to solve stochastic problems.
Babuska al. (2005) presented a stochastic version of the Lax-Milgram lemma. The paper
presents a hypothesis which represents limitations to the modeling of uncertainty via Gaussian
processes. For certain problems of mechanics, use of Gaussian processes can lead to loss of
coercivity of the bi-linear form associated to the stochastic problem. This difficulty was indeed
encountered in the study of Ávila (2004), and resulted in non-convergence of the solution for the
bending of plates with random parameters. This non-convergence was due to the choice of a
Gaussian process to represent the uncertainty in some parameters of the system. This failure to
converge also affects solutions based on perturbation or simulation methods.
This paper presented results of the existence and uniqueness of theoretical solutions for
stochastic bending for Euler-Bernoulli and Timoshenko beam theory. These results were obtained
with the use of the Lax-Milgram lemma using appropriate hypotheses about the problem data

2. Stochastic Bending of the Euler-Bernoulli Beams

In this section, the strong and weak formulations of the problem of stochastic bending of Euler-
Bernoulli beams are presented. At the end of this section, the Lax-Milgram lemma is used to
present a proof of existence and uniqueness of the solution. The strong form of the stochastic beam,
 2
 dx 2( dx
2
)
 d E.I ( x,ω ) . d u2 = f , ∀ ( x,ω ) ∈ ( 0 ,l ) × Ω;
 (1)
u ( 0 ) = u ( l ) = 0;
 2
d u d 2u
 dx 2
= = 0;
 x =0
dx 2 x =l
where E is the Young’s modulus, I is the moment of inertia, Ω is a sample space and f is a load
term. For the existence and uniqueness of the response, the following hypotheses are considered:
(
H1 : ∃ c , C ∈ ℝ + : P ω∈ Ω : E .I ( x , ω ) ∈  c , C  , ∀ x ∈  0, l  = 1 )
(2)
( )
H2 : f ∈ L 2 Ω , F , P ; L 2 ( 0, l ) .
Hypothesis H1 ensures that the beam stiffness modulus is positive-defined and uniformly limited
in probability e a hypothesis H2 ensures that the stochastic loading process is limited.
In order to study existence and uniqueness, the abstract variational problem associated to the
strong form (Eq. 1) needs to be defined. The abstract variational problem associated to the beam
bending problem defined in Eq. (1) is defined in V = L2 ( Ω, F , P; Q ) , with
 d 2u d 2u 
Q = u ∈ H 2 ( 0, l ) u ( 0 ) = u ( l ) = 0 ∧ 2 = = 0 ,
 dx x =0
dx 2 x =l 

 
L 2 ( Ω , F , P; Q ) = u : ( 0, l ) × Ω → ℝ u é mensurável e ∫ u ( ω ) dP ( ω ) < +∞  . (3)
2
H 2 ( 0,l )
 Ω 

96
Expression (2) means that an element u ∈ L 2 ( Ω , F , P; Q ) for ω ∈Ω , fixed, u ( ⋅, ω ) ∈ Q . On
the other hand, for x ∈ ( 0, L ) , fixed, u ( x , ⋅ ) ∈ L 2 ( Ω , F , P ) . Defining the tensorial product
between v ∈ L 2 ( Ω , F , P ) and w ∈ Q as u = v.w , one should note that for fixed ω∈Ω ,
u ( ⋅, ω ) = v ( ⋅ ) .w ( ω ) ∈ Q , (4)
whereas for a fixed x ∈ ( 0, L ) , u ( x, ⋅ ) = v ( x ) .w ( ⋅) ∈ L 2 Ω , F , P . Hence, one has ( )
L ( Ω, F , P; Q ) ≃ L ( Ω, F , P ) ⊗ Q ⇒ V ≃ L ( Ω, F , P; Q ) ⊗ Q . It is also necessary to
2 2 2

redefine the differential operator for the space obtained via tensorial product. The operator
Dωα : V → L 2 ( Ω , F , P ) ⊗ L 2 ( 0, l ) , Matthies & Keese (2005), acts over an element u ∈ V the
following way,

( ) ( x ).w ( ω) ,
α
Dωα u : d αv (5)
dx
where α ∈ ℕ and α ≤ 2. V is a Hilbert space, with internal product defined as
(u,v )V = ∫ ( Dω2u ( ω) ,Dω2v ( ω) )L (0,l ) dP ( ω) .
2
(6)

The bilinear form a :V ×V → ℝ is defined as,
a ( u,v ) = ∫ ( E.I .Dωu ( ω) ,Dωv ( ω) )L (0,l ) dP ( ω) .
2 2
2
(7)

The abstract variational problem associated to the strong form, Eq. (1), is defined as follows:
 Find u ∈V such that
 (8)
 a ( u,v ) = ℓ ( v ) , ∀v ∈V .
From the hypotheses of limited probability one can show that the bilinear form has the following
properties:
a. continuity
a ( u,v ) ≤ C ∫ Dω2u ( ω) 2 Dω2v ( ω) 2 dP ( ω) ≤ C u V v V ;
L ( 0 ,l ) L ( 0 ,l )

b. coercivity
(
a ( u,u ) ≥ c ∫ Dω2u ( ω) ,Dω2u ( ω) ) dP ( ω) ≥ c ∫ Dω2u ( ω) dP ( ω) = c u
2
2
L2 ( 0,l )
V
L ( 0,l )
2
Ω Ω
From hypothesis H1 and H2, which state the continuity and coervicity of the bilinear form, and
from the Lax-Milgram lemma, it is guaranteed that the problem defined in Eq. (1) has an unique
solution, and continuous dependency on the data.

3. Stochastic Bending of the Timoshenko Beam

In this section, the strong and weak formulations of the problem of stochastic bending of
Timoshenko beams are presented. The Lax-Milgram lemma, again, is used to present a proof of
existence and uniqueness of the solution. The strong form of the stochastic Timoshenko beam
bending problem is given as,

97
( )
 d α d φ + β ( dw − φ ) = 0;
 dx dx dx
d

(
 dx β ( dw )
dx − φ ) = − f , ∀ ( x,ω) ∈ ( 0,l ) × Ω;
(9)
 w ( 0,ω) = w ( l,ω) = 0;

φ ( 0,ω) = φ ( l,ω) = 0, ∀ω∈Ω;
where α= E.I and β = G.A are the bending and shear stiffness, respectively, Ω is the sample
space, w is the transversal beam displacement field, φ is the angular displacement field and f is a
load term. In order to guarantee existence and uniqueness of the solution, the following hypotheses
are necessary:

(
∃α, α ∈ ℝ + \ {0} : P {ω∈Ω : α ( x, ω) ∈[ α,α] , ∀x ∈[ 0, l ]} = 1;
 )
H1: 
({
∃β, β ∈ ℝ + \ {0} : P ω∈Ω : β ( x, ω) ∈ β, β , ∀x ∈[ 0, l ] = 1;

}) (10)

(
H2: f ∈ L2 Ω, F , P; L2 ( 0, l ) . )
Hypothesis H1 ensures that the beam stiffness modulus is positive-definite and uniformly limited
in probability. Hypothesis H2 ensures that the stochastic load process has finite variance. These
hypotheses are necessary for the application of the Lax-Milgram lemma, Babuska al (2005), which
is used in the sequence, to demonstrate the existence and uniqueness of the solution. In this section,
a brief theoretical study of existence and uniqueness of the solution of stochastic Timoshenko
beam bending problems is presented. For operators with derivatives of order greater than two, no
such study has been found in the literature. The study requires definition of stochastic Sobolev
spaces, tensorial product and density between distribution spaces and Lp spaces. The study also
requires definition of the abstract variational problem associated to the strong form, Eq. (9). The
stochastic Sobolev space where the solution to the stochastic beam bending problem is constructed

( (
is V = L2 Ω, F , P; H 01 ( 0, l ) ) ) , such that,
2

 w and φ are measurable and 


 
V = ( w, φ ) : ( 0, l ) × Ω → ℝ 2
. (11)
w ( ω ) H 1 0,l dP ( ω ) , ∫ φ ( ω ) ( )
2 2

 ∫ ( ) H 1( 0,l )
dP ω < +∞

 Ω Ω 
Eq. (11) means that, for ω ∈Ω fixed, ( w (⋅, ω) , φ (⋅, ω) ) ∈ H ( 0, l ) × H ( 0, l ) , whereas for
1
0
1
0

x ∈ ( 0, l ) fixed, w ( x, ⋅) , φ ( x, ⋅) ∈ L2 ( Ω, F , P ) . Defining the tensorial product between


v ∈ L 2 ( Ω , F , P ) and ( θ, ϑ ) ∈ H 01 ( 0, l ) × H 01 ( 0, l ) as ( w, φ ) = ( v.θ, v.ϑ ) , one has, for fixed
ω∈Ω ,
 w ( ⋅, ω) = v ( ⋅) .θ ( ω) ∈ H 01 ( 0, l ) ;
 (12)
φ ( ⋅, ω) = v ( ⋅) .ϑ ( ω) ∈ H 0 ( 0, l ) ;
1

98
whereas for fixed x ∈ ( 0, l ) ,
 w ( x, ⋅) = v ( x ) .θ ( ⋅) ∈ L2 ( Ω, F , P ) ;
 (13)
φ ( x, ⋅) = v ( x ) .ϑ ( ⋅) ∈ L ( Ω, F , P ) .
2

Hence, one has

( 2
)
L2 Ω, F , P; ( H 01 ( 0 ,l ) ) ≃ L2 ( Ω, F , P ) ⊗ ( H 01 ( 0,l ) ) ⇒ V ≃ L2 ( Ω, F , P ) ⊗ ( H 01 ( 0 ,l ) ) .
2 2

It is also necessary to redefine the differential operator for the space obtained via tensorial product.
( )
The operator Dων : V → L2 ( Ω, F , P ) ⊗ L2 ( 0, l ) , acts over an element w ∈ V the following
2

way,
ν
( )
Dων w : d vν ( x ) .θ ( ω) ,
dx
(14)

where ν ∈ ℕ and ν ≤ 2. Let u, v ∈V , with u = ( w, φ) and v = ( h, υ ) , define inner product


(⋅, ⋅) V
:V ×V → ℝ ,
l

( u,v)V = ∫∫ ( ( Dωφ.Dωυ) + ( Dωw −φ) .( Dωh − υ) ) ( x,ω) dxdP ( ω) . (15)


Ω0

= ( u , v )V . The bilinear form


1
2
The inner product defined in Eq. (15) induces the V norm, u V

a : V ×V → ℝ , is defined as
l
a ( u, v ) = ∫∫ ( ( α.Dωφ.Dωυ) +β.( Dωw − φ) .( Dωh − υ) ) ( x, ω) dxdP ( ω) . (16)
Ω0

The abstract variational problem associated to the strong form, Eq. (9), is defined as follows:
 Find u ∈ V such that
 (17)
 a ( u,v ) = ℓ ( v ) , ∀v ∈V ;
where ℓ :V → ℝ is a linear functional given by,
l
ℓ ( v ) = ∫ ∫ ( f .v ) ( x,ω ) dxdP ( ω) . (18)
Ω 0
From the hypotheses H1, Eq. (10), one can show that the bilinear form has the following
properties:

a. continuity

99
l
a ( u,v ) ≤ ∫ ∫ α ( Dωφ.Dωυ )( x,ω) + β . ( ( Dω w − φ ) .( Dωh − υ ) ) ( x,ω)  dxdP ( ω)
 
Ω0

  l
1 1
 l 
2 2

≤ C. ∫  ∫ ( Dωφ )( x,ω) dx  .  ∫ ( Dωυ )( x,ω) dx  dP ( ω)


2 2

Ω  0  0 

1 1
l  l 
2 2

+ ∫  ∫ ( Dω w − φ )( x,ω) dx  .  ∫ ( Dωh − υ )( x,ω) dx  dP ( ω) 


2 2

Ω 0  0  
≤ C.  Dωφ . Dωυ + Dω w − φ
 L 2 ( Ω ,F , P )⊗ L 2 ( 0 ,l ) L 2 ( Ω ,F , P )⊗ L 2 ( 0 ,l ) L 2 ( Ω ,F , P )⊗ L 2 ( 0 ,l )

. Dωh − υ  ≤ C. u . v ,
L 2 ( Ω ,F , P )⊗ L 2 ( 0 ,l )  V V

{ }
where C = max α, β .
b. coercivity
l
a ( u,u ) ≥ ∫∫ ( α.Dωφ.Dωφ) +β.( Dωw −φ) .( Dωw −φ)  ( x,ω) dxdP ( ω)
Ω0
l
≥ c ∫∫ ( Dωφ.Dωφ) + ( Dωw −φ) .( Dωw −φ)  ( x,ω) dxdP ( ω) ≥ c. u V2 ,
Ω0

where c = min {α ,β} .


From hypothesis H1, which state the continuity and coervicity of the bilinear form, and from the
Lax-Milgram lemma, Babuska al (2005), it is guaranteed that the problem defined in Eq. (17) has
unique solution and continuous dependency on the data ( α, β, f ) .

4.Conclusions

In this work, results were presented on the existence and uniqueness of weak solutions for the
stochastic bending problems of Euler-Bernoulli and Timoshenko beams. For that, hypotheses
related to the limitation and regularity of the elastic coefficients and loading term were necessary.
From this, the Lax-Milgram lemma was used to obtain the result of existence and uniqueness of
weak solutions. In addition, we obtained the continuity of the weak solutions with respect to the
data.

Acknowledgments

This article is part of a research project of process number 420615/2016-4 developed in the
PPGEM/UTFPR and supported by the National Research Council, CNPq.

100
References

Araújo, J.M. and Awruch, A.M. (1994), “On stochastic finite elements for structural analysis”, Computers
and Structures, 52(3), 461-469.
Ávila da S. Jr, C. R.,, (2004), "Application of the Galerkin method to stochastic bending of Kirchhoff plates",
Doctoral thesis. Florianópolis (SC, Brazil): Department of Mechanical Engineering, Federal University
of Santa Catarina, [in Portuguese].
Babuska, I.; Tempone, R.; Zouraris, G. E., (2005), “Solving elliptic boundary value problems with uncertain
coefficients by the finite element method: the stochastic formulation”, Computer Methods in Applied
Mechanics and Engineering, 194(12), 1251-1294.
Matthies, H. G. and Keese, A., (2005), “Galerkin methods for linear and nonlinear elliptic stochastic partial
differential equations”, Computer Methods in Applied Mechanics and Engineering, 194(12), 1295-1331.
Spanos, P.D. and Ghanem, R., (1989), “Stochastic finite element expansion for media random”, Journal
Engineering Mechanics, 125(1), 26-40.

101
Bubnov Galerkin Method Applied Euler Bernoulli
Beam Stochastic Bending
Cláudio R. A. da Silva Júnior1, Pedro D. Danizete*2
1
Graduate Program in Mechanical Engineering and Materials (PPGEM), Federal Technological
University of Parana (UTFPR), Curitiba, Paraná, Brazil.
2
Graduate Program in Mathematics, Federal University of Parana, Curitiba, Paraná, Brazil.

(Received keep as blank , Revisedkeep as blank, Accepted keep as blank )

Abstract. This research is presented a study about stochastic bending Euler Bernoulli beam. The
uncertain is modeled by random variables. From this, the Bubnov-Galerkin method is applied to obtain
approximate numerical solutions for the stochastic transverse displacement process. The approximation
spaces are generated by the generalized polynomials of chaos and dense global functions in the spaces
of the theoretical solutions. The Monte Carlo method is used for evaluation perfomance's Bubnov
Galerkin.
Keywords: Bubnov Galerkin method; generalized polynomials of chaos; stochastic bending Euler's
beam.

1. Introduction

The last few decades have witnessed tremendous developments in the modeling of the
structural behavior of mechanical systems, due to advances in computational mechanics.
Numerical, and computational methods such as finite element, finite difference, boundary element,
and so on, have reached wide acceptability and ample coverage of applications. New
developments address the solution of complex, non-linear problems. Multi-physics analysis allows
the investigation of new, unforeseen interaction effects between structures and soils or fluids. But
despite the tremendous robustness and accuracy of modern computational mechanics tools, one
significant aspect of most engineering problems has been neglected by so-called deterministic
computational mechanics: uncertainties, Spanos and Ghanem (1989).
In the present paper, the Bubnov-Galerkin method and chaos polynomials are used to obtain
approximate solutions for the stochastic displacement response of stochastic bending of Euler-
Bernoulli beams. Uncertainties in bending stiffness are modeled as parameterized stochastic
process, indexed in uniform random variables, in agreement with the conditions for existence and
uniqueness of the solution. The space of approximate solutions is constructed using generalized
chaos polynomials. A stochastic version of the Lax–Milgram lemma, Babuska al (2005), is used
for a brief theoretical study about the existence and uniqueness of the solution. The Bubnov-

Corresponding author, Professor Claudio Avila, E-mail: avila@utfpr.edu.br

102
Galerkin method is applied to obtain numerical solution for a example stochastic bending
problems. The performance of numerical solutions is evaluated by comparing first and second
order moments of the approximated displacement responses with the same statistics obtained via
Monte Carlo simulation.

2. Stochastic Bending of the Euler-Bernoulli Beams

In this section, the strong and weak formulations of the problem of stochastic bending of Euler-
Bernoulli beams are presented. At the end of this section, the Lax-Milgram lemma is used to
present a proof of existence and uniqueness of the solution. The strong form of the stochastic
beam,
ì 2
ï dx 2 ( dx
2
)
ï d E.I ( x,w) . d u2 = f , " ( x,w) Î ( 0 ,l ) ´ W;
ï
íu ( 0 ) = u ( l ) = 0;
ï 2
ïd u =
d 2u
= 0;
ï dx 2 dx 2 x=l
î x =0 (1)
where E is the Young’s modulus, I is the moment of inertia, Ω is a sample space and f is a load
term. For the existence and uniqueness of the response, the following hypotheses are considered:
(
H1 : $c, C Î ¡ ++ : P wÎW : E.I ( x, w) Î éëc, C ùû , "x Î éë0, l ùû = 1.)
H2 : f Î L2 ( W, F , P; L2 ( 0, l ) ) .
(2)
Hypothesis H1 ensures that the beam stiffness modulus is positive-defined and uniformly limited
in probability e a hypothesis H2 ensures that the stochastic loading process is limited.
In order to study existence and uniqueness, the abstract variational problem associated to the
strong form (Eq. 1) needs to be defined. The abstract variational problem associated to the beam
V = L 2 ( W, F , P ; Q )
bending problem defined in Eq. (1) is defined in , with
ì
ï d 2u d 2u ü
ï
Q = íu Î H 2 ( 0, l ) u ( 0 ) = u ( l ) = 0 Ù 2 = 2 = 0ý
ï
î dx x=0 dx x =l ï,
þ
ì
ï ü
ï
L2 ( W, F , P; Q ) = íu : ( 0, l ) ´ W ® ¡ u é mensurável e ò u ( w) dP ( w) < +¥ ý.
2
H 2 ( 0,l )
ï
î W ï (3)
þ
u Î L ( W, F , P; Q )
2
u ( ×, w) Î Q
Expression (2) means that an element for w ÎW , fixed, . On
x Î ( 0, L ) u ( x, ×) Î L2 ( W, F , P )
the other hand, for , fixed, . Defining the tensorial product
v Î L2 ( W, F , P )
between and wÎ Q as u = v.w , one should note that for fixed wÎW ,
u ( ×, w ) = v ( ×) .w ( w ) Î Q
, (4)

103
x Î ( 0, L ) u ( x, ×) = v ( x ) .w ( ×) Î L2 ( W, F , P )
whereas for a fixed , . Hence, one has
L ( W, F , P; Q ) ; L ( W, F , P ) Ä Q Þ V ; L ( W, F , P; Q ) Ä Q
2 2 2
. It is also necessary to
redefine the differential operator for the space obtained via tensorial product. The operator
Dwa : V ® L2 ( W, F , P ) Ä L2 ( 0, l )
, Matthies and Keese (2005), acts over an element u Î V
the following way,

( ) ( x ).w ( w) ,
a
Dwa u : d av
dx (5)
where a Î ¥ and a £ 2. V is a Hilbert space, with internal product defined as
(
(u,v )V = ò Dw2u ( w) ,Dw2v ( w) L (0,l ) dP ( w)
W
) 2

. (6)
The bilinear form a :V ´V ® ¡ is defined as,

W
(
a ( u,v ) = ò E.I .Dw2u ( w) ,Dw2v ( w) ) L2 ( 0 ,l )
dP ( w)
. (7)
The abstract variational problem associated to the strong form, Eq. (1), is defined as follows:
ìïFind u ÎV such that
í
ïîa ( u,v ) = l ( v ) , "v ÎV . (8)
From the hypotheses of limited probability one can show that the bilinear form has the following
properties:
a. continuity
a ( u,v ) £ C ò Dw2u ( w) 2 Dw2v ( w) 2 dP ( w) £ C u V v V
L ( 0 ,l ) L ( 0 ,l )
W ;
b. coercivity
(
a ( u,u ) ³ c ò Dw2u ( w) ,Dw2u ( w) ) dP ( w) ³ c ò Dw2u ( w) dP ( w) = c u
2
2
L2 ( 0,l )
V
L ( 0,l )
2
W W
From hypothesis H1 and H2, which state the continuity and coervicity of the bilinear form, and
from the Lax-Milgram lemma, it is guaranteed that the problem defined in Eq. (1) has an unique
solution, and continuous dependency on the data.

3. Uncertainty Representation

In most engineering problems, complete statistical information about uncertainties is not


available. Sometimes, the first and second order moments are the only information available. The
probability distribution function is defined based on experience or heuristically. In order to apply
Galerkin’s method, an explicit representation of the uncertainty is necessary. Given the incomplete
information about the probability distribution of a given parameter, an hypothesis of finite
dimensional noise is assumed. This implies that the uncertainty over a given input parameter
+
J : ( 0,l ) ´ W ® ¡
will be represented in terms of a finite set of random variables:

104
J ( x,w) = J ( x, x ( w) ) = J ( x,x1 ( w) , K , x N ( w) ) .
(9)
From this hypothesis, the uncertainty in beam and foundation stiffness coefficients are modeled
via parameterized stochastic processes. These are defined from a linear combination of
deterministic functions and random variables, Grigoriu (1995),
N
J ( x, w ) = m J ( x ) + å ji ( x ) x i ( w) = m J ( x ) + F t ( x ) .x ( w )
i =1 , (10)
m J ( ×) J ( ×, × ) F : ( 0,l ) ® ¡ N
where is the expected value of random process , is a vector-valued
( 0, l ) , "i Î {1,K , N} . x ( w) = {xi ( w)}i=1
N
ji Î C0 ( 0, l ) I C
2

function with terms is a vector of


independent random variables, such that:
ìïE [ xi ] = 0, "i Î {1, K , N} ,
í
ïî P ( wÎ W : xi ( w) Î G i ) = 1, "i Î {1,K , N} , . (11)
E [×] G x
where is the expected value operator. In Eq. (11), i is the image of random variable i ,
Gi = xi ( W ) Gi = [ ai , bi ] Ì ¡ Gi = bi - ai < ¥, "i Î {1, K , N}
that is, , with , , limited. In this
{Gi }i=1 , is given by
N

form, the image of random vector x : W ® G , with G Ì ¡


N
, and in terms of
N
G = Õ Gi
i=1 . Since the random variables are independent, the joint probability density is given by:
N
r ( x ( w ) ) = Õ r i ( xi )
i =1 , (12)
ri ( ×) xi . Hence, the
where is the marginal probability density function of random variable
dP ( ×)
probability measure is defined as:
N
dP ( x ( w) ) = Õ ri ( xi )d xi
. i =1 (13)
From the measure and integration theory one knows that the probability measure defined in Eq.
(13) is the measure obtained from the product between probability measure spaces associated to
x ( w) = {xi ( w)}i =1
N
x :W ®G
the random variables , with i i.
From the Doob-Dynkin lemma, Rao and Swift (2010), the transversal displacement random
x ( w) = {xi ( w)}i =1
N

process will be a function of random variables , hence:


u ( x, w) = u ( x, x ( w) ) = u ( x, x1 ( w) , K , x N ( w) )
. (14)
In this paper, polynomials of the Askey-Wiener scheme are used construct the problems solution
space.

105
4. Bubnov-Galerkin Method

The Galerkin method is used in this paper to solve the stochastic beam bending problem with
uncertainty in the beam and foundation stiffness coefficients. In order to develop numerical
solutions which are compatible with the conditions for existence and uniqueness of the theoretical
solution, existence and uniqueness results are used. An element from a space isomorph to the
V ; L2 ( W, F , P ) Ä Q
space obtained via tensor product, , can be represented from elements of
L ( W, F , P )
2
separable spaces, dense in spaces and Q . Hence, the strategy to construct
approximated, numerical solutions and to use Bubnov-Galerkin method is to use bases of finite
L 2 ( W, F , P )
dimensions, but dense in and Q . Due to the simplicity of the spatial domain for the
present problem, no spatial discretizations are employed. Hence, functions employed in
construction of the responses are defined in the whole problem domain. For more complex spatial
domains, special techniques like finite element, boundary element or finite differences would have
to be employed.
It is proposed that approximated solutions to the stochastic displacement response of the beam
have the following form:
¥
u ( x, x ( w) ) = å uidi ( x, x ( w) )
i =1 , (15)
where i
u Î ¡ , "i Î ¥ are coefficients to be determined and i are the test functions.
d ÎV
Numerical solutions to the variational problem defined in Eq. (11) will be obtained. Hence, it
becomes necessary to define spaces less abstract than those defined earlier, but without
compromising the existence and uniqueness of the solution. From the theorem of Cameron and
L2 ( W ,F ,P )
S = L2 ( W,F ,P )
Martin (1947), have . Consider two complete orthogonal systems
ì d 2f d 2f ü
F = span ïífi Î C ( éë0, l ùû , ¡ ) fi ( 0 ) = fi ( l ) = 0 Ù 2i ( 0 ) = 2i ( l ) = 0 , "i Î ¥ ïý
2

ïî dx dx ïþ
and
L2 ( W,F , P )
= span {yi }i =1 = L ( W, F , P )
¥
Y= Y Q
= 2
, such that F = Q , and define the tensor product
between F and S as:
(fÄy )i ( x, x ( w) ) = f j ( x ) .yk (x (w)) , with j, k Î¥ . (16)
d = ( fÄy )
To simplify the notation, we will use i i . Since approximated numerical solutions
are derived in this paper, the solution space has finite dimensions. This implies truncation of the
= span {yi }i =1
Yn =
n

complete orthogonal systems F and S . Hence one has F m and , which


VM = F m Ä Y n
results in Y n ( n = dim Y n =)
. The dimension of , , depends on the dimension

106
x ( w)
of the random variable vector and on the order of chaos polynomials. Let "s" be the
x ( w)
dimension of random vector and “p” the order of chaos polynomials , then the dimension of
Y n is given by:
( s + p )!
n= .
s!.p! (17)
Since
dim F n =< ¥ and dim Y n =< ¥ , one has that the dimension of the approximation space,
VM
, is given by
M = dim (VM ) = dim ( F m Ä Y n ) = dim ( F m ) .dim ( Y n ) = m.n
(18)
With the above definitions and results, it is proposed that numerical solutions are obtained from
truncation of the series expressed in Eq. (15) at the M th term:
M
uM ( x, x ( w) ) = å ui di ( x, x ( w) )
i =1 . (19)
Substituting Eq. (19) in Eq. (8), one arrives at the approximated variational problem consists in
finding the coefficients of the linear combination expressed in Eq. (19). Using a vector-matrix
representation, the system of linear algebraic equations can be written as
KU = F , (20)
K Î MM (¡ ) U = {ui }i =1 F = { f i }i=1
M M

where is the stiffness matrix, is the displacement vector and


is the loading vector. Elements of the stiffness matrix are defined as,
l
K = éë kij ùû
M ´M
(
, kij = ò ò EI .Dw2di .Dw2d j ) ( x, w)dxdP ( w).
W0 (21)
The load vector is given by
l
F = { f j} f j = ò ò ( q.d j ) ( x, w)dxdP ( w)
M
,
j =1
W0 . (22)

5. Numerical Examples

In this section, two numerical examples are presented, for bending of random Euler-Bernoulli.
In the example, uncertainty on the beam stiffness is considered. The uncertainties in stiffness are
represented using parameterized, weakly stationary, random processes. The beam is simply

supported at both ends, the span equals one meter


(l = 1 m ) , the cross-section has width
b = 1100 m and the beam is subject to an uniform distributed load of q ( x ) = 1 kPa.m, "x Î [ 0, l ] .
In this example the uncertainty is assumed on the Young´s modulus of beam material,
E : [ 0, l ] ´ W ® ¡ +
, which is modeled as a parameterized random process:

107
( n.lx ) + x ( w) sin ( n.lx )ùûú
N
E ( x, x ( w) ) = m E + 3.s E å éx2.n -1 ( w) cos
ê
n =1 ë
2.n
, (49)
where
m E is the expected value and s E is the standard deviation of Young´s modulus and
x ( w) = {xn ( w)}n=1
2.N

, is a vector of four uniform, independent random variables, hence, N = 2 ,


( 2.N = dim x ( w) ). Numerical solutions are obtained for
s E = ( 101 ) .m E
. Results obtained via
Monte Carlo simulation, and used as reference. The Fig. 1 presents results for the expected value
of mid-spam displacements, obtained trough Monte Carlo simulation, and by means of the
p Î {1, 2,3}
Galerkin solutions of order . The difference between the curves is imperceptible,
showing that even a Galerkin solution of order p=1 is already accurate.

Figure 1: Expected value of displacements.

The Fig. 2 show the variances graphics obtained via Monte Carlo simulation and via Galerkin
p Î {1, 2 ,3}
method for , respectively. Other than what was observed for the expected value,
approximations for the variance are only accurate for p=2.

108
Figure 2: Variance of value of displacements.

4.Conclusions

In this paper, theoretical and practical results for bending of stochastic Euler-Bernoulli beams
have been presented. The uncertainty in beam stiffness was modeled as parameterized random
processes. The Lax-Milgram lemma was use to establish a proof for existence and uniqueness of
the theoretical solutions. This study guided construction of the approximate solution space. The
Bubnov-Galerkin method was employed in the construction of approximate, numerical solutions to
Euler-Bernoulli beam bending. Tensor product of finite-dimensional spaces was used to construct
the approximate solutions. In order to represent the random displacement response, a family of
Legendre polynomials, originated from the Askey-Wiener scheme, were employed. In general, it
was observed that even low-order numerical solutions (p=1) provide accurate approximations of
the expected value of the response. For the variance of the displacement responses, higher order
polynomials were needed, but accurate results were already obtained for p≥3, for all cases studied.
The developed numerical solution developed herein, using chaos polynomials of the Askey-Wiener
scheme, was shown to be an accurate and efficient solution for problems of stochastic bending of
Euler-Bernoulli beams. If a proper family of chaos-polynomials is selected, and if the conditions
for existence and uniqueness of the solution are respected, excellent convergence rates are
obtained in the approximation of random displacement responses.

Acknowledgments

This article is part of a research project of process number 420615/2016-4 developed in the
PPGEM/UTFPR and supported by the National Research Council, CNPq.

References

Babuska, I.; Tempone, R. and Zouraris, G. E., (2005), “Solving elliptic boundary value problems with
uncertain coefficients by the finite element method: the stochastic formulation”, Computer Methods in
Applied Mechanics and Engineering, 194(12), 1251-1294.
Cameron, R.H. and Martin, W.T., (1947), “The orthogonal development of nonlinear functionals in series of
FourierHermite functionals”, Annals Mathematics, 48, 385-392.
Grigoriu, M., 1995, “Applied Non-Gaussian Processes: Examples, Theory, Simulation, Linear Random
Vibration, and Matlab Solutions”, Prentice Hall.
Matthies, H. G. and Keese, A., (2005), “Galerkin methods for linear and nonlinear elliptic stochastic partial
differential equations”, Computer Methods in Applied Mechanics and Engineering, 194(12), 1295-1331.
Rao, M. M. and Swift J. R., 2010, “Probability Theory with Applications”, Springer; 2nd ed.

109
Spanos, P.D. and Ghanem, R., (1989), “Stochastic finite element expansion for media random”, Journal
Engineering Mechanics, 125(1), 2640.

110
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A FULLY- CONSERVATIVE FINITE VOLUME FORMULATION FOR COUPLED PORO -


ELASTIC PROBLEMS

Hermínio T. Honório 1 , Felipe Giacomelli 2 , Lucas G. T. da Silva 3 , Clovis R. Maliska 4


1 Federal University of Santa Catarina, herminio@sinmec.ufsc.br
2 Federal University of Santa Catarina, felipe.g@sinmec.ufsc.br
3 Federal University of Santa Catarina, guesser@sinmec.ufsc.br
4 Federal University of Santa Catarina, maliska@sinmec.ufsc.br

Solid mechanics is a research field that deals with the mechanical behavior of a wide variety of materials
undergoing external loads. Among the various types of solids, porous materials, for instance, can be
found in applications such as soil and rock mechanics, biomechanics, ceramics, etc. These applications
are studied in the field of poromechanics, which is a specific branch of the solid mechanics that considers
all types of porous materials. An important characteristic of such materials is that they contain a network
of interconnected pore channels saturated with a fluid. In most situations the mechanical behavior of the
porous matrix and the fluid flow through the pore channels are two tightly coupled phenomena interfering
with each other. When the fluid moves from one region to another in the porous matrix it changes the
pressure field inside the pore channels, which is perceived by the porous matrix as a force imbalance. As a
consequence, the porous matrix tends to deform in order to find a new configuration of stress equilibrium.
When the porous matrix deforms, the pore channels are also modified, which directly affects the fluid
flow and the pore pressure field. It is clear then that a fluid flow model and a structural model must be
considered in order to solve this coupled phenomenon.
The basis of the theory that describes coupled poromechanics has been established by Terzaghi
(1923) [1], where the effective stress principle has been presented. According to this principle, the
effective stresses σ 0 acting on the solid porous matrix is balanced by the pore pressure p and the total
stress tensor σ externally applied to the system, that is:

σ 0 = σ + αpII (1)

where α is the Biot’s coefficient and I is the second-order identity tensor. Almost 20 years later, Biot
(1941) [2] generalized this theory to three-dimensions and it has become known as Biot’s consolidation
theory. In this theory, the governing equations are the mass conservation equation for deformed porous
media and the stress equilibrium equations for a porous matrix. Considering only small strains and linear
behavior for the porous matrix, the stress equilibrium equations can be written as:

∇s · (C∇s u − αpi) + ρg = 0 (2)

with C being the fourth-order tensor written in Voigt notation, ∇s being the symmetric nabla operator
and i being the Voigt represetation of I. Finally, the mass conservation equations in this case reads,
1 ∂p 
+ ∇ · v f + αvs = q (3)
M ∂t
with the Biot’s modulus given by a combination of the solid and fluid compressibilities (cs and c f ,
respectively), porosity φ and α, that is, M = [φc f + (α − φ)cs ]−1 . Additionally, the seepage velocity
(Darcy velocity) and the solid velocity are respectively given by:
k ∂u
v = − · (∇p − ρ f g) and vs = (4)
µ ∂t

111 1
Figure 1: (a) Corner-point and (b) unstructured grids.

in which k, µ, ρ f , g and u stand for the absolute permeability tensor, fluid viscosity and density, gravita-
tional acceleration vector and the displacement vector, respectively.
For real applications, the system of coupled partial differential equations composed of Equations
(3) and (2) must be solved by numerical techniques. In the groundwater community, the most common
approach is to apply the Galerkin Finite Element Method (FEM) for discretizing both fluid flow and
structural models. The use of FEM for solving coupled geomechanics is probably because of historical
reasons, as problems involving solid mechanics have always been solved by FEM. Although this method
presents the advantage of being applied to unstructured grids (see Figure 1b), thus providing great ge-
ometrical flexibility, it does not ensure local mass conservation, which is an important characteristic
specially for multiphse flows. In reservoir simulators, for instance, where multiphase flows are the main
mathematical models considered, the Finite Volume Method (FVM) is the most common choice. The
reason for employing the FVM in reservoir simulators is because its basic premisse is to ensure local
conservation in every control volume of the grid. In this context, another common approach for solving
coupled geomechanics is to solve the fluid flow model in a resevoir simulator with the FVM, and then
solve the geomechanical model in a separate FEM software. There are a number of drawbacks in this
approach that deserves further discussion. First, reservoir simulators are usually applied to corner-point
grids, as the one depicted in Figure 1a, where the variables are stored at the grid block centroids. In the
FEM software, unstructured grids are usually employed, with the variables stored at the grid nodes. This
situation requires the interpolation of variables between two different grids, which represents an extra
source of numerical errors and additional computational cost. Moreover, synchronizing two different
softwares and managing the traffic of information between the two of them can be a cumbersome task.
In order to avoid these drawbacks, a number of researchers have been proposing unified methodologies
for solving both geomechanical and fluid flow models. In the FEM community, a number of works in
this direction can be mentioned ([3, 4, 5] and many others). On the other hand, a few important attempts
have being proposed for solving coupled geomechanics in a unified finite volume formulation. For ins-
tance, Shaw & Stone (2005) [6] solved linear poroelasticity in unstructured cell-centered grids, although
emphazis has been placed on corner-point grids. Later on, dal Pizzol and Maliska (2013) [7] presented
a finite volume formulation for coupled geomechanics in Cartesian staggered grids for two-dimensional
problems. Important advances on cell-centered finite volume formulations were also developed in [8, 9]
for two-dimensional unstructured grids.
The present work proposes the solution of coupled geomechanics by employing the Element-based
Finite Volume Method (EbFVM) for discretizing the partial differential equations of both fluid flow and
geomechanical models. As a finite volume method, the EbFVM ensures mass and momentum (force)
conservation for each control volume of the grid, which is an important feature specially for fluid flows.
Moreover, the EbFVM provides great geometrical flexibility as it is naturally applied to unstructured
grids. In this work, three-dimensional unstructured grids composed of tetrahedra, hexahedra, prisms and
pyramids are employed. These types of grids are of particular interest for building radial grids around
wells in order to better capture the radial flow pattens in this region (see Figure 1b). In the EbFVM, the
control volumes are built around the nodes of the grid, therefore the variables of the problem (p and u, in
this case) are stored at the grid nodes, characterizing a cell-vertex method. As shown in Figure 2a, each
element of the grid is subdivided into sub-elements, or sub-control volumes, associated to each element
vertex. The control volume is then built by the union of all sub-elements sharing a common node. Figure
2b shows a control volume built around a node of a three-dimensional grid. Each control volume Ωi is
bounded by a control surface Γi composed of faces identified by one integration point, ip, on its centroid

112 2
Figure 2: Geometrical entities for (a) two-dimensional grid and (b) three-dimensional grid.

and an area vector, s, point outwards the control volume.


The discretized mass and stress equilibrium equations are obtained by integrating Equations (3) and
(2) in each control volume and applying the divergence theorem. The resulting surface integrals over
Γi are then evaluated at the integration points of the control volume. This means that mass fluxes and
forces are computed at the control volume surfaces, which is precisely what ensures mass and momentum
(force) conservation. When the algebraic representation of Equations (3) and (2) are grouped together,
the following linear system is obtained:
    
−K L u bu
Q A   =   (5)
−H p bp
∆t ∆t
where the block matrices K and L accounts for the effective stresses and the pore pressures acting on
control volumes’ surfaces. The second block-line of Equation (5) contains the mass conservation equa-
tions, where matrices A, H and Q represent the accumulation terms, the mass fluxes due to the seepage
velocity and the mass fluxes due to the solid movement. The linear system of Equation (5) is solved in a
monolithic way by an LU decomposition.
The proposed methodology is first validated against the well known Mandel’s problem, where a
rock slab is compressed in vertical direction and the lateral boundaries are fully permeable, as depicted
in Figure 3. In this problem, the poroelastic equations cannot be decoupled as in the one-dimensional
poroelastic column of Terzaghi, which makes it a suitable test case for assessing the proposed formula-
tion. The problem has been solved with grids composed of four types of elements: hexahedra, tetrahedra,
prisms and pyramids. As it can be seen, good agreement with the analytical solution is obtained for all
types of grids.
The final problem intends to reproduce a water withdrawal from a 12 meters aquifer composed of
sand. The aquifer is trapped between two layers of silty clay with low permeability. The whole struc-
ture consists of a cylinder with 250 meters radius and 50 meters height. A vertical well with prescribed

Figure 3: Mandel’s problem: Pressure and displacement profiles for different time steps.

113 3
constant pressure is placed at the center of the structure. Due to symmetry, only a quarter of the geom-
etry is considered, as depicted in Figure 4. The left side of this figure shows the pressure and vertical
displacement fields. The graphic in the middle shows the pressure profile along the vertical center line
of the structure. The well is placed between z = 27, 5 and z = 40 meters, where pressure is constant. It
is interesting to notice the positive values of pressure that establishes in the adjacent aquitards. This is
known as the Noordbergum effect and it’s an evidence of the coupling between fluid flow and geome-
chanics. The rightmost graphic of Figure 4 shows the vertical displacement in the radial direction. After
250 days, the maximum subsidence observed is of 40 mm right above the well.

Figure 4: Groundwater withdrawal.

In this work the EbFVM has been used for solving both physical models involved in geoemechanics:
the fluid flow and geomechanical model. This a promising alternative for solving coupled geomechanics
for two main reasons. Since it is a fully conservative method, it is able to accurately solve multiphase
flows in porous media. Moreover, the momentum equation is also satisfied for each control volume of
the grid. The second reason is because it is able to handle unstructured grids composed of different types
of elements. This allows for the use of radial grids in the near-well region in order to better capture the
flow patterns in the vicinity of the well. To the knowledge of the authors, there is no other numerical
scheme that present all this features together.

References
[1] K. Terzaghi. Die berechnung der durchlassigkeitsziffer des tones aus dem verlauf der hydrodynamis-
chen spannungsercheinungen, Sitzung berichte. Akademie der Wissenschaften, Wien Mathematiesch-
Naturwissenschaftliche Klasse, 1923.
[2] M.A. Biot. General theory of three-dimensional consolidation. Journal of Applied Physics 12 (1941), 155–
164.
[3] J. A. White, R. I. Borja Stabilized low-order finite elements for coupled soliddeformation/fluid-diffusion and
their application to fault zone transients. Computer Methods in Applied Mechanics and Engineering 197
(2008), 4353–4366.
[4] M. Ferronato, N. Castelletto, G. Gambolati A fully coupled 3-d mixed finite element model of Biot consolida-
tion. Journal of Computational Physics 229 (2010), 4813–4830.
[5] J. Choo, R. I. Borja A stabilized mixed finite elements for deformable porous media with double porosity.
Computer Methods in Applied Mechanics and Engineering 293 (2015), 131–154.
[6] G. Shaw, T. Stone Finite volume methods for coupled stress/fluid flow in commercial reservoir simulators.
SPE Reservoir Simulation Symposium (2005), Houston, Texas U.S.A.
[7] A. dal Pizzol, C. R. Maliska A finite volume method for the solution of fluid flows coupled with the mechanical
behavior of compacting porous media. Porous Media and its Applications in Science, Engineering and Industry
1453 (2012), 205–210.
[8] J.M. Nordbotten Cell-centered finite volume discretizations for deformable porous media. Int. J. Numer. Meth.
Engng. v100, issue 6, (2014), 399–418.
[9] J.M. Nordbotten Convergence of a cell-centered finite volume discretization for linear elasticity. SIAM J.
Numer. Anal. v53, issue 6, (2012), 2605–2625.

114 4
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

A LAGRANGIAN TRANSPORT MODEL TO SIMULATE THE TRAJECTORY OF OCEANIC


FLOATS

Daiane G. Faller1, Clare Eayrs1, Jhon Mojica1, Nelson F. F. Ebecken3, David M. Holland1,2
1
Center for global Sea-Level Change, New York University Abu Dhabi, Abu Dhabi, United
Arab Emirates, dgf3@nyu.edu
1
Center for global Sea-Level Change, New York University Abu Dhabi, Abu Dhabi, United
Arab Emirates, clare.eayrs@nyu.edu
1
Center for global Sea-Level Change, New York University Abu Dhabi, Abu Dhabi, United
Arab Emirates, jfm11@nyu.edu
2
Courant Institute of Mathematical Sciences, New York University, New York, 10012,
United States of America, david.holland@nyu.edu
3
Laboratory of Technology Transference Nucleus-COPPE, Federal University of Rio de
Janeiro, Brazil, nelson@ntt.ufrj.br

In this study, we investigate the ability of a Lagrangian transport model to predict float
trajectories in the Southern Ocean (Weddell Sea - Antarctica, 20°E-15°W, 62°S-67°S).
Between 2014 and 2018, the project SOCCOM (Southern Ocean Carbon and Climate
Observations and Modeling) deployed 4 floats within the selected region with the endeavor to
measure different oceanographic variables such as temperature, nutrients, oxygen, pressure,
among others. The float is designed to drift between the surface and 2000 meters collecting
data profiles, transmitting the data via satellite each time it reaches the surface. However, if
the surface is covered by ice, to avoid being trapped, the float avoids the surface, stores the
data and descends for another round of profiling, waiting until the ice retreats to safely
transmit the data. In such cases, the data is transmitted without tracking (position) information
as the float relies on GPS to obtain its location.
Aiming to simulate the ‘virtual’ trajectory of the floats we used the offline particle-tracking
ARIANE and velocity fields from HYCOM (HYbrid Coordinate Ocean Model) ocean model
outputs with 1/12° horizontal resolution and 32 vertical levels. HYCOM is a primitive
equation general circulation model with high horizontal resolution and adaptive vertical
coordinate system, using terrain-following coordinate at shallow water and z-level coordinates
in the mixed layers and/or unstratified water. When combined with NCODA (Navy Coupled
Ocean Data Assimilation), HYCOM is able to assimilate data from in situ profiles and

115
altimeter data, which allows the model physics to be adjusted by the real observational data,
resulting in more realistic results. ARIANE is a Fortran toolkit with a staggered C-grid for the
Lagrangian interpretation of the circulation calculated by numerical ocean models. ARIANE
performs an analytical calculation, where each trajectory segment respects locally the
continuity equation.
Before the simulation, the float temperature profiles with known locations were used to
validate the reliability of the HYCOM output. The model and float profiles showed good
correlation and small misfit (observational data minus model data) from the surface down to
1000 meters, so the velocity fields from the first 14 layers of the HYCOM output
(representing the first 1000 m) were used with the particle-tracking model. During the particle
simulation virtual particles are introduced into the flow field near the initial float position,
with the central particle located at the known float position. The method used create a grid
with 4 by 4 degrees, with the same resolution as the HYCOM grid (1/12°), which represents
144 particles per grid cell (totaling 2304 particles). The particles are allowed to move freely
via hydrodynamic forcing in the horizontal domain of each layer, however, vertical
movements are not considered, i.e., particles cannot move between layers. At the end of the
simulation, the position of the particles along each float pathway is extracted as well the
profiles from the HYCOM fields (temperature, salinity and velocity). The robustness of the
results is checked by comparing the modelled profiles with those obtained by the floats.
Regional and global models that use data assimilation are only able to assimilate data with
known position. There is limited information on the real ocean state during winter at the polar
regions. Using this method to retrieve valuable data sets that previously had no positional
information vastly improves the winter oceanographic database, particularly in the poorly
sampled Southern Ocean. Our results provide a larger data set from which to examine and
interpret meso- and small- scale processes and identify the main ocean features of the region.

116
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

FINITE ELEMENT STRATEGIES FOR MODELING THE SIZE EFFECT IN


NANO-REINFORCED MATERIALS
Dang Phong Bach, Delphine Brancherie, Ludovic Cauvin

Sorbonne Universités, Université de Technologie de Compiègne


Laboratoire Roberval FRE UTC-CNRS 2012, Compiègne, France

1. Introduction
Popularity of nanocomposites is still encreasing due to their remarkable physical properties (mechanical,
electrical,...). Thus, understanding of their behavior has become a crucial issue on the experimental point
of view but also on the analytical and numerical aspects. With the use of nano-reinforcement, size effects
usually not observed with “standard” (in terms of size) reinforcement appear: the mechanical properties
are not only dependent on the volume fraction of reinforcement but also on the size of the reinforcements.
This size effect, due to local phenomena at the atomic scale, could be attributed to the increase of the
ratio (interface matrix-inclusions) / (volume fraction of inclusions). In order to model this size effect,
a coherent interface at the matrix-inclusion interface is usually considered. The introduced interface is
therefore characterized by a surface stress-strain relation leading to a surface elastic behavior. In terms
of modelisation, most of the works consider analytical developments and tend to estimate the overall
mechanical properties of nano-composites [1–6]. The main limitations of such analytical works is that
the shape of the inclusions is limited to cylindrical and spherical ones. Moreover, such techniques do
not allow to easily consider material non linear behavior. The development of dedicated computational
approaches is therefore necessary to circumvent those limitations. If we omit the high cost works based
on molecular dynamics [7, 8] few numerical approaches are proposed in the literature to describe the
behavior of nano-reinforced materials. The first one consists in introducing interface-type elements in a
standard Finite Element (FEM) framework to account for the surface elasticity [9]. The second approach
investigated in the literature by Yvonnet and co-authors [10] consists in taking into account the surface
effect in the framework of the eXtended Finite Element Method (XFEM) . The last one discussed in
this work is the Embedded Discontinuity Finite Element Method (EFEM) which take into account the
interface at the local level of the elements. We propose first to compare the two former approaches (FEM
and XFEM). After validation of the implemented strategies, we investigate, the influence of the homog-
enization hypothesis in terms of boundary conditions on the computed effective mechanical properties.
In particular, a statistical analysis is performed on random RVEs ( Representative Volume Elements).
Further results are presented in the context of elastoplastic behaviors. A comparison of the two previous
numerical approaches with EFEM will be discussed during the presentation.

2. Problem definition and associated numerical formulations


We consider here a continuum body which is described by a bounded domain Ω ⊂ Rd (d = 2 or 3) , with
boundary ∂Ω. This domain consists of two phases Ω(1) and Ω(2) ( in the remainder of this paper Ω(1)
and Ω(2) denote inclusion and matrix, respectively). These two phases are divided by an interface Γ (see
figure 1a).

117 1
(a) Geometry (b) FEM conformal mesh (c) XFEM mesh (d) EFEM mesh

Figure 1: Cylindrical inclusion with imperfect interface treated in plane strain models

The equilibrium equations in the bulk can be written:

divσ(i) + b = 0 in Ω(i) (i = 1, 2) (1)

The generalized Young-Laplace equation [11] is then used to deal with the coherent interface equilib-
rium:
[[σ]] · n + divs σs = 0 (2)
where [[σ]] denotes the jump of stress across the interface Γ, divs denotes the surfacic divergence and σs
is the surfacic stress tensor.
By appealing to the extended divergence theorem, we can build the weak form of the previous problem
as: Z Z
δu · (div σ + b)dΩ + δū¯ · (divs σs + JσK · n)dΓ = 0 ∀(δu, δū¯ ) (3)
Ω Γ
For XFEM enrichment strategy, the interface is defined as the zero level-set of the function φ(x) (see
figure 1c). The displacement is interpolated as:

n n n

u (x) = ∑ Ni (x)ui + ∑ N j (x)ψ(x)a j with ψ(x) = ∑ |φi |Ni (x) − ∑ φi Ni (x)
h
(4)
i=1 j∈VΓ i=1
i=1

where Ni (x) is the shape function associated to node i and VΓ denotes the set of nodes associated to
elements whose support is cut by the interface.
The use of the previous interpolation in the weak form of equilibrium equations (3) leads to the following
discrete system to be solved:
(K + Ks) d = f (5)
For the FEM with interface-type elements approach (see figure 1b), taking into account coherent interface
is based on an explicit discretization of the interface through the use of interface elements with surface
elasticity. In that case, we have to consider the classical interpolations for the 2D elements in the bulk
and the 1D elements on the interface Γ:
n m
uh (x) = ∑ Ni (x)ui and uhs (x) = ∑ Ñi(x)ũi (6)
i=1 j=1

where Ñi is the shape function of 1D interface elements associated to node i and ũi corresponds to
the displacements of node i along the interface direction obtained by projecting the components of the
displacement in the global frame onto the local frame:

ũi = ui · tΓs (7)

where tΓs is the unit tangent vector of the interface.


Using those interpolation in the weak form of the equilibrium system leads to the discretised system to
be solved:
(K + Ks) u = f (8)

118 2
3. Several numerical results
In this section, few selected results of this work are presented. Additional results will be discussed during
the presentation. Figure 2a presents the comparison in terms of rate of convergence and efficiency of two
implemented numerical approaches (XFEM and Interface Element). Figure 2b presents the effective
(homogenized) normalized bulk modulus obtained for different sets of interface elastic properties (set
A: λs = 6.842 N/m, µs = −0.375 N/m; set B: λs = 3.48912 N/m, µs = −6.2178 N/m; set C: λs = 0
N/m, µs = 0 N/m) in terms of the radius of nanopore for a fixed volume fraction f=0.2. For the effective
properties both approaches give very similar results but the efficiency and the rate of convergence is
better for Interface element (see figure 2a).

-0.4 0.58
setA XFEM/level set XFEM/level set
-0.6 setB XFEM/level set 0.56 Set A Interface element

Normalized effective plane strain


setC XFEM/level set GSCM [Le Quang et He 2009]
setA Interface element
-0.8 0.54
setB Interface element
setC Interface element

bulk modulus
-1 0.52
log 10 (e)

Set C
-1.2 r = 0.6 0.5

-1.4 r = 0.73 0.48

-1.6 0.46
1
r = 0.95
-1.8 0.44
r=1 Set B
-2 0.42
-1.8 -1.6 -1.4 -1.2 -1 -0.8 -0.6 0 10 20 30 40 50
log 10 (h) Ro in nanometers, f=0.2

(a) Convergence analysis with different (b) Size-dependent effective bulk modulus for
interface parameters periodic boundary condition

Figure 2: Cylindrical nanopore problem with the coherent interface

Influence of the homogenization hypothesis on the computed effective mechanical properties has been
studied in terms of boundary conditions (Kinematic Uniform (KUBC), Static Uniform (SUBC) and Pe-
riodic). A comparison with an analytical homogenization technique [3] is also carried out. Periodic
boundary condition show a good agreement with the analytical model.

1.1
KUBC
1 SUBC
Normalized effective plane strain

PERIODIC
Set A GSCM
0.9
bulk modulus

0.8
Set B
0.7

0.6
Set C
0.5

0.4
0 0.05 0.1 0.15 0.2
Volume fraction

Figure 3: Size-dependent effective bulk modulus for different type of boundary condition

Finally, in order to get closer to real microstructures, different RVEs consisting of randomly distributed
nanopores are analyzed to compute the effective properties in terms of the radius of the pores:

119 3
0.44

Normalized effective plane strain


0.43

0.42

bulk modulus
0.41

0.4

0.39

0.38
0 10 20 30 40 50
R0 in nanometers, f=0.3

Figure 4: Size-dependent effective bulk modulus with random distribution

References
[1] P Sharma and S Ganti. Size-dependent eshelby’s tensor for embedded nano-inclusions incorporating sur-
face/interface energies. Transactions-American society of mechanical engineers journal of applied mechan-
ics, 71(5):663–671, 2004.
[2] P Sharma, S Ganti, and N Bhate. Effect of surfaces on the size-dependent elastic state of nano-
inhomogeneities. Applied Physics Letters, 82(4):535–537, 2003.
[3] H. Le Quang and Q. C. He. Estimation of the effective thermoelastic moduli of fibrous nanocomposites with
cylindrically anisotropic phases. Archive of Applied Mechanics, 79(3):225–248, Mar 2009.
[4] HL Duan, Jian-xiang Wang, ZP Huang, and Bhushan Lal Karihaloo. Size-dependent effective elastic con-
stants of solids containing nano-inhomogeneities with interface stress. Journal of the Mechanics and Physics
of Solids, 53(7):1574–1596, 2005.
[5] Sébastien Brisard, Luc Dormieux, and D Kondo. Hashin–shtrikman bounds on the bulk modulus of a
nanocomposite with spherical inclusions and interface effects. Computational Materials Science, 48(3):589–
596, 2010.
[6] Sébastien Brisard, Luc Dormieux, and Djimedo Kondo. Hashin–shtrikman bounds on the shear modulus of a
nanocomposite with spherical inclusions and interface effects. Computational Materials Science, 50(2):403–
410, 2010.
[7] D. Brown, V. Marcadon, P. Mélé, and N. D. Albérola. Effect of filler particle size on the properties of model
nanocomposites. Macromolecules, 41(4):1499 – 1511, 2008.
[8] S. Güryel, M. Walker, P. Geerlings, F. De Profta, and M. R. Wilsonb. Molecular dynamics simulations of the
structure and the morphology of graphene/polymer nanocomposites. Physical Chemistry Chemical Physics,
19:12959 – 12969, 2017.
[9] Gao Wei, YU Shouwen, and Huang Ganyun. Finite element characterization of the size-dependent mechani-
cal behaviour in nanosystems. Nanotechnology, 17(4):1118, 2006.
[10] Julien Yvonnet, H Le Quang, and Q-C He. An xfem/level set approach to modelling surface/interface effects
and to computing the size-dependent effective properties of nanocomposites. Computational Mechanics,
42(1):119–131, 2008.
[11] Yu Z Povstenko. Theoretical investigation of phenomena caused by heterogeneous surface tension in solids.
Journal of the Mechanics and Physics of Solids, 41(9):1499–1514, 1993.

———————————————————————

120 4
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

CONCRETE QUANTITY MINIMIZATION FOR A SHALLOW FOUNDATION INTEGRATING


PARTICLE SWARM OPTIMIZATION AND A SOIL STRUCTURE-INTERACTION ANALYSIS.

David Sebastián Cotes Prieto 1, César Andrés Méndez Poveda2, William Giovanny Alfonso León3, Oscar
Javier Begambre Carrillo4

1
Industrial University of Santander, david.cotes@correo.uis.edu.co
2
Industrial University of Santander, cesar.mendez@correo.uis.edu.co
3
Industrial University of Santander, william.alfonso@correo.uis.edu.co
4 Industrial University of Santander, ojbegam@uis.edu.co

1. Abstract
In current engineering practice in Colombia, footings linked by beams is one of the more employed shallow
foundation systems when the ratio of transferred (from the super-structure to the foundation) flexural moment
and axial load is high. Its broad implementation is due to its capacity to control excessive differential
settlements, soil pressure and uplifting. There are a lot of simplified “traditional” methodologies used for the
analysis of these foundation systems, however more accurate results can be obtained when taking into account
the effect of soil-structure interaction (SSI). Often a discouraging “side-effect” of the use of SSI is that wider
dimensions for the elements are usually needed to accomplish minimum structural codes requirements. This
work presents preliminary results of a research project that pretends to encourage the use of SSI in analysis and
design of these kind of foundations by taking advantage of the particle swarm optimization (PSO) ,which has
proven being efficient for solving other structural engineering problems, to minimize the quantity of concrete
for the footings and beams, required to achieve proper differential settlements and soil pressure, and to avoid
uplifting (according to Colombian structural and construction code “NSR-10”). Structural analysis of the
system was done using finite element analysis integrating: (i) elastic springs with stiffness according to the
ballast coefficient to simulate the soil, (ii) the Kirchhoff plates theory for footings modeling and (iii) the Euler-
Bernoulli beams theory for link beams modeling. All the analysis and optimization were done using a self-
made code in Matlab R2017a. The efficiency of the proposed analysis and optimization methodology was
proved using a real-life problem of an already existing foundation. Results showed that the proposed
methodology is a good alternative to reduce the quantity of concrete required in this kind of foundations.
Keywords: Particle swarm optimization, shallow foundation, soil-structure interaction, finite element method.
2. Methodology
2.1. Structural case
The foundation system proposed for analysis and design is form by two footings linked by a beam. One of the
footings is commonly known as a center footing while the other is an eccentric footing. The solicitations are
modeled in the geometrical center of the footings and consists in an axial load and a flexural moment (that
causes flexure in the beam) for each, these forces are directly transferred from the super-structure to the
foundation system. Additional flexural moment is calculated for the eccentric footing based on the eccentricity
of the transferred force, this condition gives a high ratio of the total flexural moment and the axial force.
Described above is presented in Figure 1.

2.2. Structural modeling


Kirchhoff plates theory to model footings, and Euler-Bernoulli frames theory to model beams [1], were
employed using the finite element method (FEM) implementing rectangular four-node plate elements for the
footings and linear two-node bar element for the beam [1]. To model soil-structure interaction, linear springs
were used as supports in the nodes [2]. Springs stiffness corresponds to the geotechnical parameter known as
ballast coefficient [3] [4]. The structural model was programed using Matlab R2017a [5]. Results obtained
from the model, needed for the optimization method restrictions, are: (i) maximum soil pressure (Qmax) (related
to springs force), (ii) differential settlement (ds) (between footings), and (iii) maximum displacement (dmax);
this last result is needed to verify if there is or not uplifting in the foundation [6]. Figure 2 illustrates deformed
shape for a foundation system obtained in the Matlab 121 R2017a program.
Figure 1. Foundation system conventions and modeling.

Figure 2. Structural model results

2.3. Particle swarm optimization


To minimize the quantity of concrete required for the foundation system, particle swarm optimization (PSO)
was implemented [7]. This heuristic method was selected due to its successful use on the optimization of other
structural engineering problems [8]–[10]. The objective function to minimize is the volume of the system as
described in Equation 1 [11].

ffitness = (bx1 ∗ by1 ∗ bz1 ) + (bx2 ∗ by2 ∗ bz2 ) + (hv ∗ bv ∗ Lv ) (1)

The function has eight independent variables which describe the footings and beam dimensions and one
dependent variable which is volume. Equation 2 presents the restrictions adopted (described in 2.2) for the
optimization, which are based on structural and geotechnical engineering criteria proposed on the Colombian
structural and construction code “NSR-10” [12]. Minimum and maximum values allowed for the independent
variables were chosen based on NSR-10 recommendations and constructive criteria. To control de velocity of
the method a hybrid between the constriction factor [13] and the inertial weight [14] was used. Figure 3
presents the flow diagram that describes the integration of the structural modeling and the PSO
programed by the authors.

Lcolx
differencial settlement ≤
Restrictions = { 160 (2)
maximum soil pressure ≤ admisible soil pressure
maximum displacement ≤ 0

122
Start

Maximum and minimum values of independent variables, restrictions,


material parameters, loads, iterations number and particle numbers.

Randomly determine initial positions and velocities of the swarm particles.

Calculate differential settlements, maximum soil pressure and maximum displacement using
FEM.

NoEnd
Are restrictions
satisfied?

Yes

it=1

Calculate stochastic factors (rand, χ) and inertial weight (w).

Calculate new velocity and position for the swarm particles.

Calculate differential settlements, maximum soil pressure and maximum displacement using
FEM.

Are restrictions
satisfied?

New position is accepted (particle New position is not accepted (particle


moves). does not move).

Calculate objective function.

Selection of the best positions (for each particle and for all
the swarm).

it=it+1

it<=maxit

Best value of objective function.

Best position of the swarm particles.

End

Figure 3. Flow diagram that describes the integration of the structural modeling and the PSO.

3. Partial results
Results presented in Table 1 are based on a real case of study, the algorithm was executed three times for three
different values of maximum iterations (maxit=5, 10, 15) maintaining a constant value of swarm particles
(N=10), for a total of nine numerical experiments performed. The movement of the particles can be seen in
Figure 4. It can be noticed the all experiments of maxit=5 resulted in values of the standard deviation high
when compared to maxit=10 and 15, so a higher number of iterations is required to obtain better results,
however as it can be seen in Figure 5 in all cases (for the minimum standard deviation of each set of
experiments) it was possible to obtain lower values for the objective function (volume) compared to the real
case of study results, so minimization process was123achieved.
Table 1. PSO results.

PSO Results (10 particles)


Real maxit=5 maxit=10 maxit=15
1 2 3 1 2 3 1 2 3
bx1 [m] 2.000 2.753 1.284 2.904 1.128 1.263 1.193 3.450 1.364 3.200
by1 [m] 2.000 1.681 3.096 1.986 3.205 1.022 1.105 1.445 3.652 1.364
bz1 [m] 0.400 0.474 0.520 0.473 0.449 0.586 0.650 0.503 0.421 0.560
bx2 [m] 1.800 0.763 0.501 1.652 2.908 0.817 1.759 1.369 1.150 1.581
by2 [m] 1.800 1.404 2.352 0.747 0.473 2.713 1.825 0.521 0.589 1.785
bz2 [m] 0.400 0.673 0.687 0.440 0.625 0.582 0.425 0.498 0.532 0.509
hv [m] 0.800 0.363 0.376 0.668 0.365 0.677 0.688 0.432 0.248 0.455
bv [m] 0.500 0.390 0.493 0.676 0.519 0.273 0.265 0.514 0.652 0.346
ds [mm] 2.562 4.006 3.787 2.521 4.446 4.654 3.217 3.437 3.481 3.512
Qmax [kPa] 181.069 183.660 212.117 143.944 238.810 313.712 227.533 173.530 173.224 205.523
dmax [mm] -1.965 -0.586 -1.516 -1.078 -1.524 -3.188 -2.470 -0.902 -0.850 -1.626
ffitness [m]3 3.560 3.432 3.850 4.651 3.286 2.992 3.086 3.455 3.241 3.275
st dev [m] - 0.529 1.151 0.821 0.533 0.046 0.175 0.120 0.798 0.007

3.7

3.6

3.5 maxit=5 (st dev=0.529)


maxit=10 (st dev=0.046)
Best function [m]3

3.4 maxit=15 (st dev=0.007)


Real
3.3

3.2

3.1

2.9
1 3 5 7 9 11 13 15
Iteration
4.
Figure 4. Swarm displacement behavior (maxit=10, std Figure 5. Best function results for minimum standard
dev=0.046 [m]). deviations obtained.

References

[1] E. Onate, Structural Analysis with the Finite Element Method - Linear Statics, vol. 1. 1989.
[2] T. R. S. Mullapudi and A. Ayoub, “Soil Structure Interaction through Two Parameter Foundation,” in Proceedings of
the ASME 2010 29th International Conference on Ocean, Offshore and Arctic Engineering, 2016, no. 2002, pp. 1–6.
[3] A. D. Kerr and N. E. Soicher, “A Peculiar Set of Problems in Linear Structural Mechanics,” Int. J. Solids Struct., vol.
33, no. 6, pp. 899–911, 1996.
[4] M. Mesa Lavista, J. Álvarez Pérez, E. Tejeda Piusseaut, and C. A. Recarey Morfa, “Determination of the domain
dimensions in embankment numerical modeling,” Dyna, vol. 83, no. 198, p. 44, 2016.
[5] The Mathworks Inc, “Matlab R2017a.” The Mathworks Inc, Massachusetts, USA, 2017.
[6] G. Gazetas, “4th Ishihara Lecture. Soil-foundation-structure systems beyond conventional seismic failure thresholds,”
Soil Dyn. Earthq. Eng., pp. 23–29, 2014.
[7] J. Kennedy and R. Eberhart, “Particle swarm optimization,” Neural Networks, 1995. Proceedings., IEEE Int. Conf., vol.
4, pp. 1942–1948 vol.4, 1995.
[8] S. Talatahari, E. Khalili, and S. M. Alavizadeh, “Accelerated Particle Swarm for Optimum Design of Frame Structures,”
Math. Probl. Eng., vol. 2013, 2013.
[9] P. K. Jena and D. R. Parhi, “A Modified Particle Swarm Optimization Technique for Crack Detection in Cantilever
Beams,” Arab. J. Sci. Eng., vol. 40, no. 11, pp. 3263–3272, 2015.
[10] M. Khajehzadeh, M. R. Taha, A. El-Shafie, and M. Eslami, “Modified particle swarm optimization for optimum design
of spread footing and retaining wall,” J. Zhejiang Univ. A, vol. 12, no. 6, pp. 415–427, 2011.
[11] A. Petrowski, J. Dréo, P. Siarry, and E. Taillard, Metaheuristics for Hard Optimization. Berlin-Germany: Springer,
2006.
[12] Comisión asesora permanente para el régimen de construcciones sismo resistentes, Reglamento colombiano de
construcción sismo resistente. NSR-10. Bogotá D.C, 2010.
[13] R. C. Eberhart and Y. Shi, “Comparing inertia weights and constriction factors in particle swarm optimization,” Inst.
Electr. Electron. Eng., vol. 1, no. 7, pp. 84–88 vol.1, 2000.
[14] R. Correa, O. Begambre, and J. Carrillo, “Validación de un algoritmo híbrido del PSO con el método simplex y de
topología de evolución paramétrica,” DYNA, pp. 255–265, 2011.
124
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

ANALYTICAL-NUMERICAL STUDY OF VIBRATIONS IN CYLINDRICAL SHELL


UNCOUPLED AND COUPLED WITH FLUID

Davidson de Oliveira França Júnior1, Lineu José Pedroso2


1
University of Brasilia, davidson.francajunior@gmail.com
2
University of Brasilia, lineu@unb.br

1. Introduction
The cylindrical shells are characterized by a curved surface object with a small thickness compared to
their other dimensions, being generally made of solid material. Despite extensive theoretical studies on
vibrations in cylindrical shells, most works are limited to classical boundary conditions: pinned-
pinned, clamped-clamped, and clamped-free. However, it is known that in engineering applications,
the walls of cylindrical shells are generally connected to other structural components (cover, bottom
plate, etc) through various types of connections, in which the idealization of different conditions (Fig.
1) represents a fundamental role in the analysis of these situations in the mathematical and/or
numerical model.

Figure 1: A Diagram of the tank: (a) the cross-section; (b) the transversal section.

Several studies have theoretical analytical, numerical and experimental approaches to the study of
vibrations in cylindrical shells for different boundary conditions, such as [1-7]. In this work, the free
vibrations of an empty, fully filled cylindrical reservoir with an inviscid and incompressible liquid are
studied for various boundary conditions. The reservoir is modeled by a cylindrical shell and the fluid
represented by an acoustic cavity. The analytical solution is based on the variational calculation
through the energy theory implemented in MAPLE software. The fluid is modeled by the wave
equation and the fluid-structure interaction performed by incorporating an additional mass of fluid into
the dynamic equations of motion of the shell. The numerical discretization is done with the finite
element method (MEF) using the software ANSYS. The forced vibration comes from a harmonic
point load in the shell, in which the frequency domain response spectra were investigated for the
decoupled shell with different boundary conditions.

125
2. Analytical Formulation
The cylindrical shell (Fig. 1) is composed of thin walls of length L, median radius R, radial angle φ
and thickness h. The shell material is considered to be elastic with Young's modulus E, Poisson
coefficient v and specific mass ρ. The coordinates of the displacement vector on the surface of the
shell in the axial, circumferential and radial direction are respectively u, v and w. The deformations
and the changes in the curvature of the shell assumed based on the linear theory of Flügge [8]. The
dynamic motion equations can be obtained through the Rayleigh-Ritz procedure through the
Lagrangian function (Γ). The Lagrangian function, the internal energy of deformation and the kinetic
energy of the cylindrical shell, according to [9], is given by:
h
2 2 L
EP =
1
2    ( z  z )
+     +  z  * z dv
h 0 0

 = Ec máx − E P máx 2
(1)
2 L  2 2 2
Rh  u   v   w 
Ec =
2    t  +  +

 t

 t 
 dzd

0 0

where ECMÁX is the maximum kinetic energy of the shell and EPMÁX is the internal energy of the
maximum deformation of the cylindrical shell. Part of the beam modes presented by [10]. When
applying the Rayleigh-Ritz procedure (  A = 0 ;  B = 0 ;  C = 0 ) in the Lagrangian function, we have
a system of three equations of motion. These equations can be expressed in matrix form, by symmetric
terms, in which for nontrivial solutions, the determinant of the characteristic matrix is defined as zero.
This yields the characteristic polynomial in which the roots characterize the natural frequencies.
Through the analyzed analytical procedure, for a cylindrical shell with boundary condition pinned-
pinned, the coupling is performed by the additional mass of fluid imposed in the shell equation in
terms compounded by the mass of the structure. The expression of the additional mass was developed
by [11] and imposing such a parameter has a hydrodynamic pressure of the transformed fluid in an
additional virtual mass in the shell structure.

3. Numerical Simulation
In the modeling the following properties of the materials were adopted: the steel shell considered to be
made with Young's modulus E=200 GPa, Poisson Coefficient υ=0,29 and specific mass ρe=7760
Kg/m3; the liquid is water with specific mass ρf=1000 Kg/m3 and speed of sound propagation in water
c=1500 m/s. For the geometric properties, were adopted: the cylindrical shell with radius R=0,175 m,
length L = 0.664 m and thickness h = 0.00102 m; the acoustic cavity with radius R = 0,175 m and
length L = 0,664 m. The formulation used was U-P, displacement for structure and pressure for the
fluid.

Figure 2: Numerical models: (a) Uncoupled shell; (b) Uncoupled fluid and (c) Fluid-Structure
interaction.

126
4. Results
The analytical results were compared with the numerical results and are presented in Figure 3. The
study was limited to a low-frequency range (m = 1) and different numbers of circumferential waves.

(a) (b)
Figure 3: Natural frequencies of the mode shape m = 1 for different boundary conditions for (a)
uncoupled shell and (b) coupled with fluid.

Based on Figure 3, it is possible to observe that the analytical and numerical results practically
coincide, validating the analytical technique and numerical modeling in ANSYS®. It is observed that
the lower natural frequencies of the cylindrical shell do not necessarily occur for small values of n,
both for the uncoupled shell and for the shell coupled with fluid, and also, as n increases, natural
frequencies tend to be common values. Such a phenomenon is justified by the fact that when the shell
vibrates in small circumferential modes the energy of membrane extensional deformation
predominates in the total deformation energy of the system. However, as the number of modes n
increases, the flexural deformation energy becomes predominant in the system.
The presence of the fluid reduced the natural frequencies, showing that the mode of additional mass in
which the structure predominates over the fluid is predominant during vibration. From the numerical
modeling validated in the analysis in free vibrations, the harmonic point load was imposed on the
cylindrical shell, P(z,ϕ,t) = 100 sen (ϖt), in the radial direction, at the center of the cylindrical shell
(z=L/2) and for its angle ϕ=00. The excitation frequency (ϖ) of the load was varied from 0 to 2500
rad/s and the damping ratio adopted was equal to two percent (ξ = 2%). Thus, it was possible to obtain
the displacement amplitude curves as a function of this applied excitation frequency.

(a) (b)

127
(c) (d)

(e)
Figure 4: Frequency spectrum for a cylindrical shell for z = L / 2 and ϕ=180º: (a) pinned-pinned, (b)
pinned-free, (c) clamped-clamped, (d) clamped-free e (e) clamped-pinned.

Based on the frequency spectra shown, it can be concluded that the peaks of the curves with the
displacement amplitudes occur at applied frequencies approximately equal to the first natural
frequencies at which the shell is excited. This is observed for all boundary conditions, where the
difference between the applied and the natural frequency did not exceed 3%. As can be analyzed, the
largest amplitudes were identified both in the radial (w) direction of the shell and in the
circumferential direction, showing that for forced vibrations the circumferential displacement becomes
important in the analysis.

5. Conclusions
With the numerical modeling, it was possible to analyze the effect of the fluid-structure coupling
through the identification of the dominant modes of the system. This fact was fundamental to observe
that even having influence in the dynamic behavior, the different bindings produce typical
characteristic modes of the structure. In general, both for the decoupled and coupled fluid-structure
problem it can be concluded that the boundary conditions with one or both free edges cause the lower
frequencies, i.e. it is clear that a reduction of system rigidity occurs and consequently a decrease in
natural frequencies. By imposing the harmonic point load on the decoupled cylindrical shell, it was
possible to show the way of obtaining the operational modes, since the analytical treatment in these
situations is more complex.

Acknowledgments

The authors thank the University of Brasília (UNB), the Coordination for the Improvement of Higher
Education Personnel (CAPES) and the Foundation for Research Support of the Federal District (FAP-
DF) for the resources received.

128
References

[1] Kurylov Y. & Amabili M. (2010). Polynomial versus trigonometric expansions for non linear
vibrations of circular cylindrical shells with diferente boundary conditions. In: Journal Sound and
Vibration; 329 (9): 1435–49.
[2] Qu Y.; Hua H. & Meng G.; (2013). A domain decomposition approach for vibration analysis of
isotropic and composite cylindrical shells with arbitrary boundaries. In: Journal Composite
Structures. 95 (2013) pp. 307-321.
[3] Mendes, N. B., Pedroso, L. J. & Ribeiro, P. M. V., (2014). Um estudo de vibrações livres
acopladas em cascas cilíndricas com anéis enrijecedores e contendo fluido. CILAMCE 2014 –
XXXV Iberian Latin-American Congress on Computational Methods in Engineering, Fortaleza, CE,
Brasil, 19 p.
[4] Ma X., Jin G., Xiong Y. & Liu Z. (2014). Free and forced vibration analysis of coupled conical-
cylindrical shells with arbitrary boundary conditions. International Journal Mechanical Sciences;
88:122–37.
[5] Tang D., Yao X., Wu G. & Peng Y. (2017). Free and forced vibration analysis of multi-stepped
circular cylindrical shells with arbitrary boundary conditions by the method of reverberation-ray
matrix. In: Journal Thin-Walled Structures. 116 (2017) 154-168.
[6] França Jr, D. O., Pedroso, L. J. & Mendes, N. B., (2017). Estudo de Vibrações Livres
Desacopladas e Acopladas Fluido-Estrutura em Cascas Cilíndricas para Diferentes Condições de
Contorno. CILAMCE 2018 – XXXVIII Iberian Latin American Congress on Computational Methods
in Engineering, Florianópolis, SC, Brasil, 20 p.
[7] Li, H; Luo, H.; Wei S. & Wen B.; (2018). The Influence of Elastic Boundary on Modal Parameters
of Thin Cylindrical Shell. International Journal of Acoustics and Vibration, Vol. 23, pp. 93–105.
[8] Flugge, W., (1934). Statik und Dynamic der Schalen. Berlin, Julius Springer.
[9] Brush, D. O.; Almroth, B. O.; (1975). Buckling of Bars, Plates, and Shells. New York: McGraw-
Hill. 379 p.
[10] Blevins, R. D., (1979). Formulas for Natural Frequency and Mode Shape. First Edition, Van
Nostrand Renhoid Company, New York, United States.
[11] Lindholm, U. S., Kana, D. D. & Abramson, H. N. (1962). Breathing vibrations of a circular
cylindrical shell with an internal liquid. In: Journal Aerospace Science. 29, 1052-1059.

129
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

STUDY OF FLUID-STRUCTURE INTERACTION THROUGH ADDITIONAL MASS IN


CYLINDRICAL TANKS UNDER FREE VIBRATIONS

Davidson de Oliveira França Júnior1, Lineu José Pedroso2


1
University of Brasilia, davidson.francajunior@gmail.com
2
University of Brasilia, lineu@unb.br

1. Introduction
The dynamic structural analysis of cylindrical shells is considerably complex due to the fact that they
have flexural and extensional vibrations together. In this sense, a shell has three deformed modalities:
the deformed axial (u), circumferential (v) and radial (w). According to [1], because they have higher
displacement amplitudes, the radial deformations produce the lower natural frequencies for a cylindrical
shell. This fact directly influences the coupled analysis, since to perform the fluid-structure interaction
the modal deformed of the tank with fluid are assumed the same as in the vacuum.
The study of vibrations in cylindrical shells over the years, and still does today, following the classic
works of [2-5], among other various theories that have been improved over time. From this, several
works bring theoretical analytical and numerical approaches to the fluid-structure coupling, in which we
can mention some works that couple the fluid as a virtual mass in the shell equations only in the radial
modal deformed direction, such as: [6-10]. In this work, the free vibrations of a cylindrical tank totally
filled with an incompressible and incompressible liquid are studied, for different forms of mass coupling
between fluid and structure. The tank is modeled by a cylindrical shell and fluid-structure interaction is
accomplished by incorporating a virtual additional mass into the shell equations in seven different ways:
coupled mass in the axial, circumferential and radial deformed as well as the combinations between
them. The analyzed analytical methodology is based on the energetic formulation implemented in
MAPLE software. For the validation of the procedure, the results were compared with the values
obtained in the experiment performed by [9].

2. Analytical Formulation
2.1 Cylindrical Shell

The cylindrical shell (Fig. 1) is composed of thin walls of length L, median radius R, radial angle ϕ and
thickness h. The shell material is considered to have Young's modulus E, coefficient of Poisson v and
specific mass ρ. The coordinates of the displacement vector on the surface of the shell in the axial,
circumferential and radial direction are respectively u, v and w. The deformations and the changes in the
curvature of the shell assumed on the basis of the linear theory of [4].

130
Figure 1: Schematic diagram of tank: (a) Coordinate system and geometry of cylindrical shell (b) the
cross-section; (c) fluid-structure interaction by additional mass of fluid.

The dynamic equations of motion can be obtained through the Ritz procedure through the Lagrangian
function, given by:
 = Ec máx − E P máx (1)

where ECMÁX is the maximum kinetic energy of the shell and EPMÁX is the internal energy of the maximum
deformation of the cylindrical shell. The internal energy of deformation and the kinetic energy of the
cylindrical shell, according to [11], is given by:
h
2 2 L
EP =
1
2    ( z  z +     +  z  * z dv ) (2)
h 0 0

2

2 L  2 2 2
Rh  u   v   w 
Ec =
2    t  +  t  +  t   dzd

(3)
0 0

Since A, B and C are constants that expose the amplitude of the axial, circumferential and radial
displacements, respectively. The number of longitudinal half-waves is m and the number of
circumferential waves is n. The plot ϕm(z)represents the longitudinal displacements of the shell. In the
proposed method, a beam with pinned-pinned boundary condition presented by [12]. Applying the Ritz
procedure in the Lagrangian function (  A = 0 ;  B = 0 ;  C = 0 ), we have a system of three equations
of motion. These equations can be expressed in matrix form, expressed by symmetric terms, and are
given by:

 C11 C12 C13   A


C  
 21 C22 C23   B  = 0 (4)
C31 C32 C33  
C 

For non-trivial solutions, the determinant of the characteristic matrix is defined as zero. This yields the
characteristic polynomial in which the roots characterize the natural frequencies. Through the
implementation on the software MAPLE, it is possible to obtain the eigenvalues and eigenvectors of the
uncoupled shell.

2.2 Fluid-Structure Interaction

Through the analyzed analytical procedure, for a cylindrical shell with pinned-pinned boundary
condition, the coupling is performed by the additional mass of fluid imposed on the shell equation in
terms compounded by the mass of the structure. The expression of the additional mass was developed
by [13] and is given by:

131
  2mH  
sen   
f  R   H  L   J n (i ) 
 = 
 
   −  
 h   L (5)
  2m  R J n (i )
'
  = 

  r R
 

Since ζ is the virtual additional mass expression, ρf is the density of the acoustic fluid and J'n is the
derivative of the Bessel functions. By imposing such a parameter there is a hydrodynamic pressure of
the transformed fluid in an additional virtual mass in the shell structure. Mass couplings were studied in
the following way:

𝐶11 + 𝜁𝜔2 𝐶12 𝐶13 𝐶11 𝐶12 𝐶13 𝐶11 𝐶12 𝐶13
[ 𝐶21 𝐶22 𝐶23 ] [𝐶21 𝐶22 + 𝜁𝜔2 𝐶23 ] 𝐶
[ 21 𝐶22 𝐶23 ]
⏟ 𝐶31 𝐶32 𝐶33 ⏟𝐶31 𝐶32 𝐶33 ⏟𝐶31 𝐶32 𝐶33 + 𝜁𝜔2
CASE A CASE B CASE C

𝐶11 + 𝜁𝜔2 𝐶12 𝐶13 𝐶11 + 𝜁𝜔2 𝐶12 𝐶13 𝐶11 𝐶12 𝐶13
[ 𝐶21 𝐶22 + 𝜁𝜔2 𝐶23 ] [ 𝐶21 𝐶22 𝐶23 ] [𝐶21 𝐶22 + 𝜁𝜔2 𝐶23 ]
𝐶33 + 𝜁𝜔 2 𝐶33 + 𝜁𝜔2 (6)
⏟ 𝐶31 𝐶32 𝐶33 ⏟ 𝐶31 𝐶32 ⏟𝐶31 𝐶32
CASE D CASE E CASE F

𝐶11 + 𝜁𝜔2 𝐶12 𝐶13


[ 𝐶21 𝐶22 + 𝜁𝜔2 𝐶23 ]
⏟ 𝐶31 𝐶32 𝐶33 + 𝜁𝜔2
CASE G

In the modeling, the following properties of the materials were adopted: the steel shell with Young's
modulus E=200 GPa, Poisson ratio υ=0,29 and specific mass ρe=7760 Kg/m3; the liquid is water with
specific mass ρf=1000 Kg/m3 and velocity of sound propagation in water c=1500 m/s. Regarding the
geometric properties, the following was adopted: the cylindrical shell with radius R=0,175 m, length
L=0,664 m and thickness h=0,00102 m; the acoustic cavity with radius R=0,175 m and length L=0,664
m.
The formulation implemented on the software MAPLE is analogous to the uncoupled problem, where
the characteristic matrix determinant is defined as zero. With this, the roots of the characteristic
polynomial, in this case, represent the coupled natural frequencies.

4. Results
The analytical results obtained were compared with the experimental results and are presented in Table
1. The N index is the order of the frequencies, m is the longitudinal mode shape number and n is the
number of circumferential waves.

Table 1 – Comparisons of natural frequencies (Hz) the cylindrical shell coupled with fluid.
Diff. Diff. Diff. Diff. Diff. Diff. Diff.
N m n Ref [9] Case A Case B Case C Case D Case E Case F Case G
(%) (%) (%) (%) (%) (%) (%)
1 1 4 92,00 220,37 139,53 199,58 116,93 103,28 12,26 198,94 116,24 103,19 12,16 100,80 9,57 100,71 9,47
2 1 5 104,00 230,13 121,28 217,84 109,46 116,12 11,65 217,56 109,19 116,08 11,62 114,40 10,00 114,36 9,96
3 1 3 119,00 311,53 161,79 256,05 115,17 134,03 12,63 253,74 113,23 133,69 12,34 128,15 7,69 127,86 7,45
4 1 6 147,00 297,80 102,59 288,13 96,01 159,74 8,67 287,96 95,89 159,71 8,65 158,17 7,60 158,14 7,58
5 1 7 206,00 396,05 92,26 387,71 88,21 223,51 8,50 387,60 88,16 223,49 8,49 221,96 7,75 221,94 7,74

Based on Table 1 it is possible to observe that the additional mass can increase the inertial load of the
shell in the axial, circumferential and radial direction, and among their combinations. The shell
displacements are larger in the radial direction, but incorporating the additional mass of the axial,

132
circumferential and radial direction (Case G) brings more realistic results, since, even if small, the
viscous tension of the fluid in these directions also influences the phenomenon physical vibration. It
should be noted that cases C, E, and F also produced good results, but cases A, B and D do not simulate
adequately and cannot be used. In this way, Case G can be recommended to perform the fluid-structure
coupling.

5. Conclusions
This work addressed the aspects of fluid-structure interaction in problems applied in the dynamic
analysis in vertical cylindrical tanks. The influence of the way of incorporating the additional mass of
fluid in the shell has become clear and must be taken with care since it has been found that the error
between the analytical and experimental results differ based on this parameter. The procedure of
coupling the fluid in the three deformed shells (u, v and w) can be used to analyze the tank in different
modes of vibration by producing results very close to the experimental result.

Acknowledgments

The authors thank the University of Brasília (UNB), the Coordination for the Improvement of Higher
Education Personnel (CAPES) and the Foundation for Research Support of the Federal District (FAP-
DF) for the resources received.

References

[1] Soedel, W., (2005). Vibrations of Shells and Plates. Third Edition, Marcel Dekkler, New York,
United States.
[2] Love, A. E. H., (1888). The small free vibrations of a thin elastic shell. International Phil. Trans.
Roy. Soc., pp. 491–549.
[3] Donnell, L. H., (1933). Stability oh thin walled tubes under torsion. Inc. NACA Report, No 479.
[4] Flugge, W., (1934). Statik und Dynamic der Schalen. Berlin, Julius Springer.
[5] Mushtari, K. M., (1938). On the stability of cylindrical shells subject to torsion. International
Trudy Kaz. Avais, Russian.
[6] Lakis, A. A. & Paidoussis, M. P., (1971). Free vibration of cylindrical shells partially filled with
liquid. International Journal of Sound and Vibration, pp. 1–15.
[7] Gonçalves, P. B. & Batista, R. C., (1986). Frequency Response of Cylindrical Shells Partially
Submerge dor Filled with Liquid. In: Journal of Sound and Vibration. DOI: 0022-460X/87/040059+12.
[8] Fernholz, C. M, Robinson, J. H., (1990). Fully-Coupled Fluid/Structure Vibration Analysis Using
MSC/NASTRAN. NASA technical memorandum 102857.
[9] Amabili, M. & G. Dalpiaz; (1995). Breathing Vibrations of a horizontal circular cylindrical tank
shells, partially filled with liquid. International Journal of Vibration and Acoustics, pp. 117–187.
[10] França Jr, D. O., Pedroso, L. J. & Mendes, N. B., (2017). Estudo de Vibrações Livres Desacopladas
e Acopladas Fluido-Estrutura em Cascas Cilíndricas para Diferentes Condições de Contorno.
CILAMCE 2018 – XXXVIII Iberian Latin American Congress on Computational Methods in
Engineering, Florianópolis, SC, Brasil, 20 p.
[11] Brush, D. O.; Almroth, B. O.; (1975). Buckling of Bars, Plates, and Shells. New York: McGraw-
Hill. 379 p.
[12] Blevins, R. D., (1979). Formulas for Natural Frequency and Mode Shape. First Edition, Van
Nostrand Renhoid Company, New York, United States.
[13] Lindholm, U. S., Kana, D. D. & Abramson, H. N. (1962). Breathing vibrations of a circular
cylindrical shell with an internal liquid. In: Journal Aerospace Science. 29, 1052-1059.

133
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

T HE STRONG DISCONTINUITY APPROACH FOR DUCTILE FAILURE WITH DAM -


AGE

Jérémie Bude 1 , Delphine Brancherie 1


1 Laboratoire Roberval, FRE UTC-CNRS 2012, delphine.brancherie@utc.fr

This work deals with a modeling tool aiming at describing, without mesh dependency, ductile failure
phenomena. The non linear behavior of ductile materials can be decomposed into two phases precursor
to failure: the pre-peak phase in which plasticity and damage are evolving in the bulk, and the softening
post-peak phase in which strain localization occurs and leads to a macro-crack growth responsible for
the failure of the structure. This second phase is responsible for the mesh-dependency observed numer-
ically when dealing with such softening materials. Many authors have proposed localization limiters::
non-local techniques [11, 8], rate-dependency regularization [9, 10] or mesh size depending softening
parameters [4, 1], to cite only a few of them.
In this work, we present the extension of the strong discontinuity approach as a localization lim-
iter to the regularization of a coupled damage-plasticity version of Lemaitre’s phenomenological model
[7]. The proposed extension of the strong discontinuity approach (also called embedded discontinuity
approach) takes into account the two phases of the nonlinear behavior of the considered material. The
pre-peak phase, characterized by the evolution of plasticity and damage in the bulk, is described through
the use of classical continuum models. The second phase of the behavior, characterized by the coales-
cence of micro-cracks and development of macro-cracks, is described by the introduction of a surface
of discontinuity of the displacement field. A traction-displacement jump relation is considered on the
interface to represent the localized dissipation occuring due the development of macro-cracks. The reg-
ularization properties of the strong discontinuity approach has to be attributed to the introduction of the
interface and the related surfacic dissipation.
The Finite Element implementation of the method is based on the use of the incompatible modes
method [6]. It has proven its regularizing effect in the case of brittle materials [2]. In this work, we
intend to attest its regularizing effects in the ductile fracture framework by considering both plane strain
and axisymmetric test problems. We investigate, in particular, the influence of the criterion for the
introduction of the displacement discontinuity.
Moreover, the key points of the extension of the strong discontinuity approach to the framework of
finite strain will be presented and discussed.
Several numerical examples will be presented to assess the regularizing capabilities of the approach
as well as its efficiency in reproducing the crack path observed experimentally.

References
[1] Barenblatt, G.I., The mathematical theory of equilibrium crack in the brittle failure. Advances in Applied
Mechanics, Vol.7, 55–125, 1962.
[2] Brancherie, D., Ibrahimbegovic, A.. Novel anisotropic continuum-discrete damage model capable of repre-
senting localized failure of massive structures. Part I : theoretical formulation and numerical implementation.
Journal of Engineering Computations, Vol. 26, 100–127, 2009.
[3] de Souza Neto, E.A.. A fast, one-equation integration algorithm for the Lemaitre ductile damage model.
Communications in Numerical Methods in Engineering, Vol. 18, 541–554,2002.
[4] Dugdale, D.S., Yielding of steel sheets containing slits. Journal of the Mechanics and Physics and Solids,
Vol. 8, 100–104, 1960.

134 1
[5] Dujc, J., Brank, B., Ibrahimbegovic, A.. Quadrilateral finite element with embedded strong discontinuity for
failure analysis of solids. Computer Modeling in Engineering and Sciences, Vol. 69, 223–258, 2011.
[6] Ibrahimbegovic, A., Wilson, E.L.. A modified method of incompatible modes. Communications in Applied
Numerical Methods, Vol. 7, 187–194,1991.
[7] Lemaitre, J., Chaboche, J.L., Mechanics of Solid Materials. University Press: Cambridge, 1990.
[8] Mulhaus, H.B., Aïfantis, E.C., A variational principle for gradient plasticity. International Journal of Solids
and Structures, Vol. 28, 845-857, 1991.
[9] Needleman, A., Material rate dependence and mesh sensitivity in localization problems. Computer Methods
in Applied Mechanics and Engineering, Vol. 63, 69–85, 1988.
[10] Niazi, M.S., Wisselink, H.H., Meinders, T., Viscoplastic regularization of local damage models: revisited.
Computational Mechanics, Vol. 51, 203–216, 2013.
[11] Pijaudier-Cabot, G., Bažant, Z.P., Non-local damage theory. ASCE Journal of Engineering Mechanics, Vol.
113, 1512-1533, 1987.

135 2
CILAMCE 2018 C ONGRESS
12-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

U LTRASONIC COMPUTED TOMOGRAPHY USING FULL - WAVEFORM INVERSION


FOR QUANTITATIVE IMAGING OF BONES IN THE VISCOELASTIC CASE

Dimitri Komatitsch1 , Simon Bernard1 , Vadim Monteiller1 , Philippe Lasaygues1 and Régine Guillermin1
1 Aix Marseille Univ., CNRS, Centrale Marseille, LMA, Marseille, France, komatitsch@lma.cnrs-mrs.fr

We introduce an ultrasonic quantitative imaging method for long bones based on full-waveform in-
version. The cost function is defined as the difference in the L2 -norm sense between observed data and
synthetic results at a given iteration of the iterative inversion process. We use a full-wave viscoelastic
implementation, but in order to reduce the computational cost we resort to a two-dimensional approxi-
mation. The inverse problem is solved iteratively based on a quasi-Newton technique called the Limited-
memory Broyden-Fletcher-Goldfarb-Shanno method. We show how the technique can be made to work
well for benchmark models consisting of a single cylinder, and then five cylinders, the latter case includ-
ing significant multiple diffraction effects. We then show pictures obtained for a tibia-fibula bone pair
model. Convergence is fast, typically in 15 to 30 iterations in practice in each frequency band used. We
discuss the so-called ‘cycle skipping’ effect that can occur in such full waveform inversion techniques
and make them remain trapped in a local minimum of the cost function. We illustrate strategies that can
be used in practice to avoid this, as also demonstrated in [1] in the simpler acoustic case only. Future
work should include real data instead of synthetic ones only.
The main idea behind our work from a medical tomography point of view is that bone strength is re-
lated to bone mass, geometry, architecture, and composition [2]. In particular, recent studies have shown
that the thickness and porosity of the cortical bone layer plays a major role in bone resistance to fracture
at various sites such as the hip, radius and vertebrae [3, 4, 5, 6]. Bone diseases such as osteoporosis
affect bone remodeling, which results in increased cortical bone porosity and reduced cortical thickness,
and therefore in an increased risk of fracture. Osteoporotic fractures cause substantial suffering to the
patient, have a high mortality rate, and are an increasing source of burden in aging societies [7].
Bone mineral density (BMD) assessed from dual-energy-X-ray absorptiometry (DXA) is the most
commonly-used predictor of bone fracture risk [7]. DXA-measured BMD correlates to fracture risk, but
has a limited predictive power [8]. This comes from the fact that DXA measures bone mineral quan-
tity but ignores other relevant parameters related to bone geometry and mechanical properties. Higher-
resolution methods such as high-resolution peripheral quantitative computed tomography (HR-pQCT)
and high-resolution magnetic resonance imaging (HR-MRI) can provide detailed images of bone ge-
ometry and architecture [9] but are of limited clinical use because of high costs, irradiation, or long
acquisition time.
Quantitative Ultrasound (QUS) techniques have been developed as an alternative to X-ray methods
for bone quality assessment [10, 11]. Ultrasounds can indeed provide information on bone mechanical
parameters through the measurement of the speed of sound and/or of attenuation, as well as geometrical
information. They are non-ionizing and low cost, two considerable advantages over X rays. While QUS
methods have originally focused on trabecular bone assessment, they have recently been extended to
cortical bone measurements [12]. Most notably, model-based approaches relying on guided ultrasound
wave propagation in the cortical shell of long bones have yielded promising results for instance for
thickness measurements of the radius bone [13].
Ultrasonic Computed Tomography (USCT) has been proposed as a tool to provide quantitative im-
ages of the speed of sound in the cross-section of long bones with millimetric resolution [14], allowing
for accurate assessment of cortical thickness. The main difficulty of bone USCT is the large impedance

136 1
Figure 1: Result of FWI for wave velocity (a) and mass density (b) for a realistic tibia-fibula bone pair
model. The insert shows the true geometry. The actual velocity values are 2800 m.s−1 for bone and 1500
m.s−1 for the surrounding media, while the density values are 1800 kg.m−3 for bone and 1000 kg.m−3
for the surrounding media. It can be observed that, while the velocity map is correctly estimated, the
estimation of density is difficult.

contrast between hard bones and the surrounding soft tissues. In that context, ray-based and first-order
Born or Rytov approximation methods commonly used in USCT of soft tissues [15, 16] do not provide
quantitative images, except in the case of specific but time-consuming algorithms such as Compound
USCT [17]. Iterative approaches based on high-order approximations have been proposed to address the
nonlinear inverse problem for high-contrast targets [18, 19]. Lasaygues et al. [14] applied one of such
iterative methods, called distorted Born diffraction tomography (DBDT), to bone-mimicking phantoms
and obtained fairly accurate estimates of their geometry, with a relative error on their size of about 5%,
as well as of their wave speed, with a relative error of about 10%. Recently, a Born-based inversion
method was proposed to image the internal structure of long bones from reflection data acquired in an
axial configuration [20], but no quantitative assessment of the wave speed was obtained.

References
[1] Simon Bernard, Vadim Monteiller, Dimitri Komatitsch, and Philippe Lasaygues. Ultrasonic computed to-
mography based on full-waveform inversion for bone quantitative imaging. Physics in Medicine and Biology,
62(17):7011–7035, 2017.
[2] C. J. Hernandez and T. M. Keaveny. A biomechanical perspective on bone quality. Bone, 39(6):1173–1181,
2006.
[3] Gerold Holzer, Gobert von Skrbensky, Lukas A. Holzer, and Wolfgang Pichl. Hip fractures and the contri-
bution of cortical versus trabecular bone to femoral neck strength. Journal of Bone and Mineral Research,
24(3):468–474, 2009.
[4] Roger M. D. Zebaze, Ali Ghasem-Zadeh, Ann Bohte, Sandra Iuliano-Burns, Michiko Mirams, Roger Ian
Price, Eleanor J. Mackie, and Ego Seeman. Intracortical remodelling and porosity in the distal radius and
post-mortem femurs of women: a cross-sectional study. Lancet, 375(9727):1729–1736, 2010.
[5] Jean-Paul Roux, Julien Wegrzyn, Monique E. Arlot, Olivier Guyen, Pierre D. Delmas, Roland Chapurlat, and
Mary L. Bouxsein. Contribution of trabecular and cortical components to biomechanical behavior of human
vertebrae: an ex vivo study. Journal of Bone and Mineral Research, 25(2):356–361, 2010.
[6] Yohann Bala, Roger Zebaze, Ali Ghasem-Zadeh, Elizabeth J Atkinson, Sandra Iuliano, James M Peterson,
Shreyasee Amin, Åshild Bjørnerem, L Joseph Melton, Helena Johansson, John A Kanis, Sundeep Khosla,

137 2
and Ego Seeman. Cortical porosity identifies women with osteopenia at increased risk for forearm fractures.
Journal of Bone and Mineral Research, 29(6):1356–1362, 2014.
[7] O. Ström, F. Borgström, John A. Kanis, Juliet Compston, Cyrus Cooper, Eugene V. McCloskey, and Bengt
Jönsson. Osteoporosis: burden, health care provision and opportunities in the EU. Arch. Osteoporos., 6:59–
155, 2011.
[8] S. C. E Schuit, M van der Klift, A. E. A. M Weel, C. E. D. H de Laet, H. Burger, E. Seeman, A. Hofman,
A. G Uitterlinden, J. P. T. M van Leeuwen, and H. A. P Pols. Fracture incidence and association with bone
mineral density in elderly men and women: the Rotterdam study. Bone, 1(34):195–202, 2004.
[9] Eve Donnelly. Methods for assessing bone quality: a review. Clin. Orthop., 469(8):2128–2138, 2011.
[10] Pascal Laugier. Instrumentation for in vivo ultrasonic characterization of bone strength. IEEE Transactions
on Ultrasonics, Ferroelectrics, and Frequency Control, 55(6):1179–1196, 2008.
[11] Pascal Laugier and Guillaume Haïat. Bone Quantitative Ultrasound. Springer Netherlands, 2011.
[12] Kay Raum, Quentin Grimal, Peter Varga, Reinhard Barkmann, Claus C. Glüer, and Pascal Laugier. Ultra-
sound to assess bone quality. Curr. Osteoporos. Rep., 12(2):154–162, 2014.
[13] Quentin Vallet, Nicolas Bochud, Christine Chappard, Pascal Laugier, and Jean-Gabriel Minonzio. In vivo
characterization of cortical bone using guided waves measured by axial transmission. IEEE Transactions on
Ultrasonics, Ferroelectrics, and Frequency Control, 63(9):1361–1371, 2016.
[14] Philippe Lasaygues, Régine Guillermin, and Jean-Pierre Lefebvre. Ultrasonic computed tomography. In Pas-
cal Laugier and Guillaume Haïat, editors, Bone Quantitative Ultrasound, pages 441–459. Springer Nether-
lands, 2010.
[15] Cuiping Li, Nebojsa Duric, Peter Littrup, and Lianjie Huang. In vivo breast sound-speed imaging with
ultrasound tomography. Ultrasound in Medicine and Biology, 10(10):1615–1628, 2009.
[16] Roberto J. Lavarello and Andrew J. Hesford. Methods for forward and inverse scattering in ultrasound
tomography. In Jonathan Mamou and Michael L. Oelze, editors, Quantitative Ultrasound in Soft Tissues,
pages 345–394. Springer Netherlands, 2013.
[17] Philippe Lasaygues, Edgard Ouedraogo, Jean-Pierre Lefebvre, Marcel Gindre, Marilyne Talmant, and Pas-
cal Laugier. Progress towards in vitro quantitative imaging of human femur using compound quantitative
ultrasonic tomography. Physics in Medicine and Biology, 50(11):2633, 2005.
[18] C. Lu, J. Lin, W. Chew, and G. Otto. Image reconstruction with acoustic measurement using distorted Born
iteration method. Ultrason. Imaging, 18(2):140–156, 1996.
[19] Régine Guillermin, Philippe Lasaygues, Guy Rabau, and Jean-Pierre Lefebvre. Quantitative non-linear ul-
trasonic imaging of targets with significant acoustic impedance contrast - An experimental study. J. Acoust.
Soc. Am., 134(2):1001–1010, 2013.
[20] Rui Zheng, Lawrence H. Le, Mauricio D. Sacchi, and Edmond Lou. Imaging internal structure of long bones
using wave scattering theory. Ultrasound in Medicine and Biology, 41(11):2955–2965, 2015.

138 3
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

COMPARATIVO DE DIMENSIONAMNETO DOS SISTEMAS ESTRUTURAIS SLIM FLOR E


STEEL DECK

Djemerson Mateus de Andrade1,


Geraldo Donizetti de Paula2,
Rovadávia Aline de Jesus Ribas3
1
Universidade Federal de Ouro Preto, djemersonmateus@yahoo.com.br
2
Universidade Federal de Ouro Preto, gdepaula9@gmail.com
3
Universidade Federal de Ouro Preto, roviaaline@gmail.com

RESUMO

O sistema misto de laje com forma incorporada steel deck é muito utilizado no Brasil. No entanto em
outros países aplicam-se outros sistemas de construção. Neste artigo faz-se um estudo comparativo do
sistema Slim Floor com o sistema misto convencional. O sistema Slim Floor é caracterizado por
proporcionar ganho na espessura final do pavimento, pois as vigas se encontram incorporadas a laje.
Assim, este trabalho visa contribuir para o conhecimento dessa outra técnica no país. Ao aplicar a
metodologia do sistema Slim Floor, em comparação com o steel deck, conseguiu-se uma redução de
45% na espessuara do pavimento tomado como referência.

Palavras-Chave: Estruturas Mistas; Laje Mista; Pavimento de Pequena Espessura.

ABSTRACT:

The mixed slab system with built-in steel deck is widely used in Brazil. However in other countries
other construction systems are applied. In this article we make a comparative study of the Slim Floor
system with the conventional mixed system. The Slim Floor system is characterized by a gain in the
final thickness of the floor, since the beams are incorporated in slab. Thus, this work aims to
contribute to the knowledge of this other technique in the country. When applying the methodology of
the Slim Floor system, in comparison with the steel deck, a reduction of 45% in pavement thickness
was obtained as reference.

Key Words: Slim Flor; Composite Structures; Slab of small thickness

139
1- INTRODUÇÃO

O sistema misto de laje com forma incorporada steel deck vem sendo muito aplicado no Brasil. Em
outros países aplicam-se ainda outros sistemas de construção, dentre eles o sistema conhecido como
Slim Floor, no qual a mesa inferior é mais larga que a superior, podendo a laje se apoiar nessa mesa
(Figura 1). Esse perfil é obtido ao se cortar um perfil I ou H longitudinalmente e a posterior soldagem
de uma chapa de aço, mais larga que a mesa original. No entanto, a atual norma brasileira de estruturas
de aço, a NBR 8800 [1], não contempla esse tipo de dimensionamento, que é pouco conhecido por
profissionais da área [2]. Neste artigo é focada a aplicação das vigas mistas incorporadas tipo Slim
Floor ou Pavimentos Mistos de Pequena Espessura, fazendo-se um estudo comparativo do sistema
sistema Slim Floor com o sistema misto convencional em steel deck. Para o dimensionamento do
sistema Slim Floor foi utilizado o software CoSFB® [3].

Figura 1: Viga mista Slim Floor [3]

2- DESCRIÇÃO DO PROBLEMA

Para o estudo, utilizou-se uma estrutura de uma edificação localizada na cidade de Ouro Branco-MG,
que será um parte de um Instituto Federal. Na Figura 2, tem-se um detalhe do projeto que mostra a
viga em estudo. Essa viga é formada por um perfil W530x66, resultando em um sistema laje mais viga
com uma altura total de 675 mm (Figura 3).
A sobrecarga para o dimensionamento do sistema steel deck é de 3 kN/m², para salas de aula,
conforme a NBR 6120 [4]. Há também o peso do revestimento em granilite, que representa uma carga
permanente de 0,2 kN/m², considerando-se uma espessura de 10 mm. As cargas referentes ao peso
próprio do sistema estrutural utilizado não foram discriminadas, pois é apresentado apenas o resultado
do cálculo em steel deck, ou seja, a espessura final do pavimento, que consiste da viga mais a laje.

Figura 2: Viga mista [5] Figura 3: Viga mista [5]

140
3- DIMENSIONAMENTO DO PAVIMENTO SLIM FLOOR

Para o dimensionamento, em atendimento às normas NBR 6118 [6] e NBR 8800 [1], do pavimento
Slim Floor, foram utilizadas as mesmas cargas citadas anteriormente, ou seja, carga permanente
devido ao granilite de 0,2 kN/m² e a sobrecarga de utilização referente a uma sala de aula de 3 kN/m².
Nesse processo é considerado um perfil tipo IFB tipo A, em que a carga permanente referente ao peso
do perfil é somente a metade de seu peso próprio, pois ele é utilizado cortado ao meio. O peso da
chapa soldada ao perfil e o peso do concreto da vigota pré-moldada e da capa de concreto são
calculados pelo próprio programa CoSFB® [3].
Ao efetuar o dimensionamento, têm que ser consideradas duas etapas importantes: a primeira, em que
o concreto está fresco, nesse caso só a viga atua na resistência do pavimento, e a segunda, em que o
concreto já está curado, atuando todo o sistema em conjunto [7, 8].

4- RESULTADOS OBTIDOS NO DIMENSIONAMENTO

No primeiro caso, no qual é feita a verificação no estado limite último, em que só a viga resiste aos
esforços solicitantes, o Eurocode 4 [9] usa para essa verificação, o fator de segurança de 1,35 para a
carga permanente do sistema estrutural. A sobrecarga não é considerada nessa etapa, pois não há
nenhuma sobrecarga atuando na laje nesse momento. A carga que atua, então, é a soma do peso
próprio da viga e do peso próprio da laje de concreto (pré-laje mais a capa), resultando em uma
espessura de 375 mm, no caso em estudo. O peso total do sistema Slim Floor, considerando a laje
mais a viga escolhida – que foi um perfil HE 550 A – e a placa soldada na parte inferior, representa
uma carga permanente de 9,84 kN/m². Como as vigas estão espaçadas de 3 m, isso resulta em uma
carga distribuída de 29,52 kN/m. Levando em consideração que o sistema estrutural é composto por
vigas bi apoiadas e um fator de segurança de 1,35, chega-se a um momento solicitante no meio do vão
de 435,54 kNm. Na Tabela 1 mostram-se esses valores obtidos no software CoSFB® [3].

Tabela 1 – Carga distribuída


Carga na laje 9,84 kN/m²
Carga Distribuída Posição inicial (x=0) Posição Final (9,35)
na viga 29,52 kN/m 29,52 kN/m

Nessa etapa, a viga resiste sozinha a um momento de 837,24 kNm, o que resulta em segurança
estrutural em relação ao momento fletor atuante, e na Tabela 2 mostram-se os valores desses
momentos obtidos no software CoSFB® [3].

Tabela 2 – Momentos solicitante e resistente


Posição X(m) Med (KNm) Mrd (KNm)
4,67 435,54 837,27

O esforço cortante atuante na viga é de 186,33 kN, enquanto ela resiste a 833,77 kN, resultando,
também, em segurança estrutural ao sistema nessa fase inicial de construção. Na Tabela 3 tem-se o
valor do esforço cortante resistente.

Tabela 3 – Esforço cortante resistente


(kN) 833,77

Foi feita a análise para verificar se a quantidade de conectores stud bolts utilizada seria o suficiente
para haver interação entre a laje e a viga, tendo-se aplicadas duas linhas de conectores espaçados de
170 mm. Para que haja interação parcial entre os dois elementos, teria que se ter uma interação mínima

141
de 0,60. A interação existente foi de 0,83, na qual o valor do esforço horizontal de cálculo foi de
3477,03 kN e a força resistente do conector de 2897,67 kN, conforme mostra-se na Tabela 4.

Tabela 4 – Interação laje e viga do sistema Slim Floor


x(m) Ncd(kN) Nstuds(kN) η ηmín OK
4,68 3477,03 2897,67 0,83 0,6

A seguir, foi feita a análise do sistema trabalhando em conjunto, em que a laje e a viga resistem
conjuntamente aos esforços solicitantes. Além do peso próprio do sistema, tem-se a sobrecarga
atuando, de 3 kN/m², referente a uma sala de aula [4]. O Eurocode 4 [9] considera, para o estado
limite último, a combinação de cargas com os fatores de segurança de 1,35 para o peso próprio e 1,5
para a sobrecarga. Com esses coeficientes, tem-se uma carga distribuída de 53,35 kN/m. Isso resulta
em um momento solicitante de 582,83 kNm e um esforço cortante de 249,44 kN, enquanto os esforços
resistidos pela interação entre a viga e a laje são de 1347,09 kNm e 833,77 kN, respectivamente,
conforme mostra-se na Tabela 5. O sistema, do ponto de vista estrutural, considerando o ELU, mostra-
se bastante seguro, mas deve-se verificar o ELS.

Tabela 5 – Momento e esforço cortante resistentes e solicitantes com interação laje e viga
Med= 582,83 kNm 1347,09 kNm (Mpl, red.) OK
Ved= 249,44kN 833,77 kN (Vpl,rd) OK

Segundo Barros [7], os parâmetros para verificação da flecha em estruturas do tipo Slim Floor, são
diferentes das estruturas convencionais, sendo permitida uma flecha de L/200. A norma NBR 8800 [1]
prevê uma flecha admissível de L/350 para pisos. Como o trabalho segue normas europeias, foi
admitida a flecha de L/200, que resultou em uma flecha admissível de 4,68 cm. Na verificação nesse
estudo, foi encontrada uma flecha inicial de 4,53 cm e uma final de 4,75 cm (valores muito próximos
ao limite de 4,68 cm - L/200).

5- CONCLUSÕES

Ao se utilizar o sistema de dimensionamento Slim Floor, chegou-se auma espessura do pavimento de


375 mm, uma redução de 300 mm, que representa 45% da altura do pavimento convencional.
Devido à flecha excessiva, teve-se que utilizar um perfil de dimensões maiores, o que resultou em uma
folga considerável no tocante ao momento fletor e esforço cortante. Para o momento fletor, o momento
final resistido pelo sistema Slim Floor foi de 1347,09 kNm, enquanto o momento solicitante foi de
582,83 kNm, ou seja, 2,3 vezes maior que o necessário.
Para o esforço cortante ocorreu a mesma condição, na qual o esforço que o sistema resiste é de
833,77 kN, e o atuante foi de 249,44 kN, ou seja, o sistema Slim Floor resiste a um esforço cortante
3,3 vezes maior que o necessário. Então, para se atender a limitação da flecha, foi necessário aumentar
a espessura do pavimento, fazendo com que a viga ficasse com uma segurança exagerada em relação
ao momento fletor e ao esforço cortante, o que gera aumento de custos em uma construção.

REFERÊNCIAS

[1] ASSOCIAÇÃO BRASILEIRA DE NORMAS TÉCNICAS. NBR 8800. Projeto de estruturas de


aço e de estruturas mistas de aço e concreto de edifícios. Rio de Janeiro: ABNT, 2008.

142
[2] Andrade, D.M. Comparativo de Dimensionamento de Pavimento Misto de Steel Deck com o
Sistema Slim Floor Utilizando o Software CoSFB. 2017. 68p. Dissertação (Mestrado em Engenharia
Civil). Escola de Minas, Universidade Federal de Ouro Preto, Ouro Preto, 2017.
[3] CONSTRUCTALIA. Slim Floor. An innovative concept for floors. Manual da Arcelor Mittal,
[1996]. Disponível em: < http://constructalia.arcelormittal.com/files/SlimFloor_EN--
9b4c958d6dca408cc87f6ffff15a8f4a.pdf>. Acesso em: 15 mai. 2017.
[4] ASSOCIAÇÃO BRASILEIRA DE NORMAS TÉCNICAS. NBR 6120. Cargas para o cálculo de
estruturas de edificações. Rio de Janeiro: ABNT, 1980.
[5] MD BRASIL ARQUITETOS. Projeto do Instituto Federal de Minas Gerais de Ouro Branco.
Fortaleza, 2013.
[6] ASSOCIAÇÃO BRASILEIRA DE NORMAS TÉCNICAS. NBR 6118. Projeto de estruturas de
concreto - Procedimento. Rio de Janeiro: ABNT, 2014.
[7] Barros, M. O. Análise e dimensionamento de pavimentos mistos slim floor. 2011. 132 p.
Dissertação (Mestrado em Engenharia Civil). Faculdade de Ciências e Tecnologia, Universidade Nova
de Lisboa, Lisboa, 2011.
[8] SANTOS, V. J. M.; LIMA, D. M.; SAKIYAMA, F. I. H.; Slim Floor: método de dimensionamento
e estudo paramétrico. Revista Eletrônica de Engenharia Civil, Universidade Federal de Goiás, v.13, n°
1, 2017.
[9] EUROPEAN COMMITTEE FOR STANDARDIZATION. Eurocode 4 – Design of composite
steeland concrete structures – part 1-1: General rules and rules for buildings. Bruxelas, 2004.
.

143
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

NUMERICAL STUDY OF BALLAST-FLYING PROCESS CAUSED BY ACCRETING SNOW/ICE ON


HIGH-SPEED TRAINS BY USING THE DISCONTINUOUS DEFORMATION ANALYSIS (DDA)

D.Ding1, A.Ouahsine2, P.Du3


1
UT Compiegne-Sorbonne Univ., Lab. Roberval, 60203 Compiègne Cedex, France, dong.ding@utc.fr
2
UT Compiegne-Sorbonne Univ., Lab. Roberval, 60203 Compiègne Cedex, France, ouahsine@utc.fr
3
UT Compiegne-Sorbonne Univ., Lab. Roberval, 60203 Compiègne Cedex, France, , peng.du@utc.fr

Abstract: High speed trains that run through snowfall blow up the snow which sticks to the underfloor
equipment and freezes rapidly into ice. When trains enter a warmer region, the frozen snow and ice
drop at high speed from train causing the ballast to fly up and seriously damage the car body and the
environment along the track. To clarify the dynamic behavior of ballast stones at the time of collision
with a lump of snow/ice, discontinuous deformation analysis (DDA) method has been was carried out.
The DDA simulation illustrates the relationship between the number of ballast stones and the mass,
shape, speed and angle of the ice /snow hitting the track.

Keywords: Discontinuous Deformation Analysis (DDA), high speed train; ballast flying;

1. Introduction and Motivation

During winter, train services in snowy countries are usually struck by constantly recurring
problems related to snow, ice and coldness. The occurrences of dry snow and temperature changes are
two different scenarios that have been pointed out as particularly unfavorable. The dry and light snow
can whirl around and cling to the train while running, as seen in Figure 1(a). This phenomenon causes
snow packing as shown in Figure 1(b). The trains carrying snow enter a warmer region and the
generated heat and vibration cause the snow to melt. The frozen ice/snow then drops at high speed
from the train, causing the ballast to fly up and seriously damage the car body and the environment
along the track [1-3]. The discontinuous deformation analysis (DDA) method was used to investigate
dynamic behavior of ballast stones at the time of collision with a lump of snow/ice. It also used to
analysis influence factors of ballast flying, thus providing theoretical basis for optimizing ballast bed
profile and preventing ballast flying.

(a) (b)
Figure 1. Snow fly and stick to the underfloor equipment

1
144
2. Method and Model
2.1 Discontinuous Deformation Analysis
In this study a numerical model based on the method of Discontinuous Deformation Analysis
(DDA) [5,6] is presented to simulate the dynamic behavior of ballast stones colliding with the accreted
snow/ice. The ballast stones are considered as a stack of rigid blocks moving against each other [4]. The
contact and cohesion between stones as well as their geometric shape are included in this analysis.
In two dimensions, the first order approximation of the displacement (u, v) at any point (x, y) of a
block i is interpolated as:
u   1 0 ( y  y0 ) ( x  x0 ) 0 ( y  y0 ) 
   d   (T ) d  (1)
v   0 1 ( x  x0 ) 0 ( y  y0 ) ( x  x0 ) 
where d is the vector of variables associated with an individual block comprising the rigid body
translations and rotation at the centroid of the block, 0 ,  0 and  0 and the normal and shear strains,
 x ,  y and  xy , respectively:
d T  0  0  0  x  y  xy 
T
(2)
The equilibrium formulation in the DDA method is provided by the principle of potential energy
minimization. For a system of n blocks, the total potential energy  is given by
 =d Tp k pq dq  d Tp f p p, q  1,....6n (3)
where  includes contributions from block strain energy and energy terms emanating from point
and gravity loads, block to block contacts, initial stresses and boundary constraints. For more
information regarding the derivation of these contributions, the reader is referred to Shi (1992) and Shi
and Goodman(1989)[7,8]. Minimization of the above energy potential yields

 k pq d q  f q  0 (4)
d q
Consequently, Equation (4) takes the partitioned form:
 k11 k12 k12 ... k1n  d1   f1 
    
 k21 k22 k23 ... k2 n  d 2   f 2 
 k31 k32 k33 ... k3n  d3    f3  (5)
    
 ... ... ... ... ...  ...   ... 
k kn 3 ... knn    
 n1 kn 2  d n   f n 
where fi is a 6  1 vector of forces acting on block i and d i contains the variables associated with
block i . The off-diagonal submatrices kij , ( i  j ) contain the stiffness components associated with
the contact between block i and j and kij refers also to the components of the material stiffness of
block i .
2.2 Track bed model
According the European standard (13450, 2002) [9-11], half of the ballasted track bed model was
built as shown in Figure 2. The distance between the bottom of the vehicle and the top of the ballast
is set to 600 mm, which is equal to the sum of the heights of half of bogie and rail. The properties of
the materials (ice and ballast) such as unit mass, elastic modulus and Poisson ratio are shown in
Table1.

2
145
Figure 2. Track bed model
Table1. Properties of ice and ballast
unit mass elastic modulus Poisson ratio velocity
ice 800 kg /m3 5600 Pa 0.06 250(300/350) km /h
ballast 2600 kg /m3 500000 Pa 0.25 0

3. Validation and Conclusion


The validation of the numerical results is done by using Japanese experimental data [2]. In these
experiments, an air cannon was used to eject ice/snow with different masses and angles as shown in
Figure3(a). The geometries of the ballast bed and ice/snow are created wherein the ballast bed
corresponds to the European gradation standards. The relationship between the number of ballast
stones and the mass, shape, speed and angle of the ice/snow will be investigated and compared with
test data by DDA simulation.

(a) (b)
Figure 3. Japanese snow/ice flying experiment and results

References

[1] Kloow L. High-Speed Train Operation in Winter Climate. Foreign Rolling Stock, 2015.
[2] Kawashima I, Endo F. Experimental studies on ballast-flying phenomenon caused by dropping
of accreted snow/ice from high-speed trains, RTRI Report, 17 (8), 2003.
[3] G.Q.Jing, D.Ding. Snow Flight Characteristic and Prevention Research of High Speed Railway
in Cold Region. Journal of Railway Engineering Society,2017, (9):29-34.
[4] Hadji S, Ouahsine A, Naceur H, et al. Modelling of transport and collisions between rigid
bodies to simulate the jam formation in urban flows. International Journal of Multiphysics,
2(2):247-266, 2008.
[5] S. Kaidi, M. Rouainia, A. Ouahsine (2012).. (2012), ’Stability of breakwaters under
hydrodynamic loading using a coupled DDA/FEM approach. J. Ocean Engineering, Vol.55,
2012, 62–70.
[6] S. Kaidi, A. Ouahsine, P.Sergent, M. Rouainia (2012). Discontinuous Deformation Analysis to
assess the stability of rockfill dams under seismic loading’. C.R.Acad. Sci. Paris, Série IIb, vol.
340 (10), 2012. 731-738

3
146
[7] Shi, Gen-Hua. "Discontinuous deformation analysis: a new numerical model for the statics and
dynamics of deformable block structures." Engineering computations 9.2 (1992): 157-168.
[8] Shi, Gen‐ hua, and Richard E. Goodman. "Generalization of two‐ dimensional discontinuous
deformation analysis for forward modelling." International Journal for Numerical and
Analytical Methods in Geomechanics 13.4 (1989): 359-380.
[9] Bonnett, Clifford F. Practical railway engineering. Imperial College Press, 2005.
[10] Selig, Ernest Theodore, and John M. Waters. Track geotechnology and substructure
management. Thomas Telford, 1994.
[11] Alemu, A. Y. "Survey of Railway Ballast Selection and Aspects of Modelling
Techniques." Division of Highway and Railway Engineering, Department of Transport Science,
School of Architecture and the Built Environment, Royal Institute of Technology ,2011.

4
147
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

O N THE DYNAMIC BEHAVIOR OF RODS AND BEAMS WITH NONLINEAR BOUND -


ARY STIFFNESS

D. R. Santo1 , J.-M. Mencik2 , B. Tang3 , P. J. P. Gonçalves 4 , M. J. Brennan5


1 São Paulo State University, Department of Mechanical Engineering, Bauru, Brazil
e-mail: douglas.roca@unesp.br
2 National Institute for Applied Sciences Central Val de Loire, Blois, France

e-mail: jean-mathieu.mencik@insa-cv.fr
3 Dalian University of Technology, Institute of Internal Combustion Engine, Dalian, China

e-mail: btang@dlut.edu.cn
4 São Paulo State University, Department of Mechanical Engineering, Bauru, Brazil

e-mail: paulo.jpg@feb.unesp.br
5 São Paulo State University, Department of Mechanical Engineering, Ilha Solteira, Brazil

e-mail: mj.brennan@unesp.br

1. Introduction
Periodic structures, also called metamaterials, are a new class of composite materials [1] capable of
exhibit novel properties never seen before in materials found in nature, offering new possibilities to
enhance the performance of structures using these materials [2]. Metamaterials have important char-
acteristics to deal with electromagnetic, acoustic and elastic waves and, at first, they were proposed to
deal as electromagnetic wave absorbers [3, 4]. The analogy between electromagnetic waves and acoustic
waves led the development of metamaterials to deal with acoustic waves [5]. Metamaterials, when re-
lated to structural vibration, are composed tailoring identical periodic substructures, designed to decrease
wave propagation through the structure, enhancing determined phenomena. One of those, the band gap
or stop-band phenomena, describe a certain frequency region which the wave propagation on periodic
structures doesn’t exist. This phenomena can be used as a form of passive control of vibration and noise
in engineer structures, specially in strategic areas as naval, civil construction, automotive and aerospace
industries. Metamaterials offers the exceptional capacity to create desired properties through structural
arrangements of its inner elements. Therefore, take advantage of the capacity to control the parameters
that effective act on the metamaterial, changing its properties and, consequently, its dynamics, have at-
tracted considering interest over the last decade. One way to design tunable metamaterials occurs with
the introduction of nonlinear elements [6].
Motivated by the challenge in understand the bad gaps of wave propagation in periodic structures at
frequencies often encountered in engineering applications, this work aims to study the band gap behavior
of a periodic structures as rods and beams to which grounded stiffness elements are attached at set
distances apart. According to [7], Bloch’s theorem states that for any periodic structure with identical
unit periodically arranged in the structure, the change the wave amplitude across the unit cell, due to
propagating wave without attenuation, does not depend upon the location of the unit cell within the
structure. Therefore, it is possible to understand the wave propagation through the entire structure just
by considering the wave motion in a single unit cell. The first order Harmonic Balance Method (FHBM)
is used to derive the dynamic stiffness terms of the nonlinear spring and nonlinear stiffness coefficient of
the spring is considered both positive and negative. Next, analytical and numerical simulation studies of
the nonlinear behavior of the adopted model are performed.

148 1
2. Problem statement
The problem considered in this work is illustrated in Figure (1), where a rod and a beam of length L is
coupled at x = 0 to a nonlinear spring, with linear and nonlinear coefficientes represented by s1 and s3 ,
respectively. At the other end, x = L, there is a harmonic force with magnitude FL . Rod displacements at
x = 0 and x = L are U0 and UL . Where the linear and angular displacements at x = 0 and x = L are W0
and wL , respectively. According to [8], the dynamic stiffness matrix relating forces and displacements at

(a) (b)

Figure 1: (a) Rod and (b) beam coupled to nonlinear spring at x = 0 and force at x = L.

both ends of a rod and beam elements can be written as Eqs., respectively.
    
F0 EAkl cot (kl L) + Dr0 −EAkl / sin (kl L) U0
= (1)
FL −EAkl / sin (kl L) EAkl cot (kl L) UL
    
F0 −K11 + Db0 −P K12 V W0
 M0  EIkb3  −P Q11 −V Q 12
  Θ0 
     (2)
 FL  = N  K12 −V −K11 P   WL 
ML V Q12 P Q12 ΘL
where,

K11 = EIkb3 (cos (kb L) sinh (kb L) + sin (kb L) cosh (kb L)) /N, K12 = EIkb3 (sin (kb L) + sinh (kb L)) /N

P = EIkb2 (sin (kb L) sinh (kb L)) /N, V = EIkb2 (cos (kb L) − cosh (kb L)) /N
Q11 = EIkb (cos (kb L) sinh (kb L) − sin (kb L) cosh (kb L)) /N, Q12 = EIkb (sin (kb L) − sinh (kb L)) /N
N = cos (kb L) cosh (kb L) − 1
Rod amplitudes of displacement at end x = 0 and x = L are U0 and UL , respectively. The quasi-
longitudinal wavenumber is represented by kl . Beam amplitudes of linear and angular displacements
at ends x = 0 and x = L, and the bending (flexural) wavenumber are represented by W0 , Θ0 , WL , ΘL
and kb , respectively. Forces and moments applied to the ends of rod and beam at x = 0 and x = L
are F0 , FL , M0 and ML . Using the first order Harmonic Balance Method (FHBM), the dynamic stiff-
ness terms of the nonlinear stiffness attached to the rod and beam are given as Dr0 = s1 + 3s3U02 /4 and
Db0 = s1 + 3s3W02 /4, respectively. For convenience, the nonlinear spring coefficients for the rod and
beam study cases are written in a nondimensional form as α1 = s1 L/ES, α3 = s3 L/ES, γ1 = s1 L3 /EI,
γ3 = s3 L3 /EI, where E, S and I are, respecively, the Young’s Modulus, rod cross-section area and beam
second moment of area.

3. Analysis and Results


Figures (2) and (3) shows the propagation constants versus the nondimensional wavenumber, kl L, of a
rod grounded to a nonlinear stiffnes spring for different values of α1 and α3 . In Fig (2a) it is possible to
observe the propagation constant for the linear spring case, α3 = 0, and shows that the spring has more
influence over the system dynamics in the low frequency. When the nonlinear coefficient α3 is different
from zero, Figs. (2b) and (3), it is possible to observe the increase of the bandwidth of attenuation
in low frequencies and, also, that the system shows great attenuation for higher frequencies. Figures
(4) and (5) shows the propagation constants versus the nondimensional wavenumber, kb L, of a beam

149 2
2 6

P hase ← | → Atten. 1 4

P hase ← | → Atten.
0
2
-1
0
-2
-2
-3

0 2 4 6 8 10 12 0 2 4 6 8 10 12
kl L kl L
(a) (b)

Figure 2: Propagation constants for the rod grounded to a nonlinear springs at x = 0, force at x = L and
U0 = 0.1. (a) (blue) α1 = 0.1, α3 = 0; (red) α1 = 0.5, α3 = 0; (black) α1 = 1, α3 = 0. (b) (blue) α1 = 1,
α3 = 100; (red) α1 = 1, α3 = 1000; (black) α1 = 1, α3 = 10000.

4
P hase ← | → Atten.

-2

0 2 4 6 8 10 12
kl L

Figure 3: Propagation constants for the rod grounded to a nonlinear springs at x = 0, force at x = L and
U0 = 0.1. (a) (blue) α1 = 1, α3 = −100; (red) α1 = 1, α3 = −1000; (black) α1 = 1, α3 = −10000.

grounded to a nonlinear stiffnes spring for different values of γ1 and γ3 . Figure (4a) shows the propagation
constants curves for γ3 = 0. As expected, the linear spring attached to the beam has great influence at
low frequencies. In Fig. Fig. (4b), the attenuation constants increases for both low and high frequency as
the positive value of γ3 increases. For negative γ3 , Fig. (4), the attenuation constants decreases for both
low and high frequency as γ3 value increase.

4. Conclusions
This work presented a study of the band gap behavior of a periodic rod and beam which grounded nonlin-
ear stiffness elements are attached in one of the ends and harmonic forced on the other one. The nonlinear
stiffness coefficients considered are both positive and negative. Such a system can be considered as a
metamaterial whose properties can be tailored to achieve a certain dynamic behavior. A particular chal-
lenge is to arrange for the band gap in wave propagation along the beam to occur at low frequencies,
so that it can be used to control vibration transmission at frequencies often encountered in engineering
applications. The work involved analytical and numerical investigation into the band gap behavior of rod
and beam with such elements attached. It is expected that the results showed in this work improves the
comprehension of wave propagation and band gaps in periodic structures like rods and beams.

150 3
2 2

P hase ← | → Atten.

P hase ← | → Atten.
1 1

0 0

-1 -1

-2 -2

-3 -3

0 2 4 6 8 10 12 0 2 4 6 8 10 12
kb L kb L
(a) (b)

Figure 4: Propagation constants for the beam grounded to a nonlinear springs at x = 0, force at x = L
and W0 = 0.1. (a) (blue) γ1 = 0.1, γ3 = 0; (red) γ1 = 0.5, γ3 = 0; (black) γ1 = 1, γ3 = 0. (b) (blue) γ1 = 1,
γ3 = 100; (red) γ1 = 1, γ3 = 1000; (black) γ1 = 1, γ3 = 10000

2
P hase ← | → Atten.

-1

-2

-3

0 2 4 6 8 10 12
kb L

Figure 5: Propagation constants for the beam grounded to a nonlinear springs at x = 0, force at x = L and
W0 = 0.1. (a) (blue) γ1 = 1, γ3 = −100; (red) γ1 = 1, γ3 = −1000; (black) γ1 = 1, γ3 = −10000.

References
[1] J. Li, C.T. Chan Double-negative acoustic metamaterial, Physical Review E, Vol. 70, No. 5 (2004), pp. 055602.
[2] A. Sihvola, Metamaterials in electromagnetics, Metamaterials, Vol. 1, No. 1 (2007), pp. 2-11.
[3] J.B. Pendry, Negative refraction makes a perfect lens, Physical Review Letters, Vol. 85, No. 18 (2000), pp.
3966.
[4] T. Tanaka, Plasmonic metamaterials produced by two-photon-induced photoreduction technique, J. of Laser
Micro/Nanoengineering, Vol. 3, No. 3 (2008), pp. 152-156.
[5] G.W. Milton, J.R. Willis, On modifications of Newton’s second law and linear continuum elastodynamics,
Proceedings of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, Vol. 363,
No. 2079 (2007), pp. 855-880.
[6] R.C. McPhedran, I.V. Shadrivov, B,T. Kuhlmey, Y.S. Kivshar, Metamaterials and metaoptics, NPG Asia Ma-
terials, Vol. 3, No. 11 (2011), pp. 100.
[7] A.S. Phani, J. Woodhouse, N.A. Fleck, Wave propagation in two-dimensional periodic lattices, The Journal of
the Acoustical Society of America, Vol. 119, No. 4 (2006), pp. 1995-2005.
[8] F. Fahy, J. Walker, Advanced applications in acoustics, noise and vibration, CRC Press, 2004.

151 4
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

EFFECT OF PARTICLE SPIN ON THE SPATIO-THERMAL FIRE HAZARD DISTRIBUTION OF


INCANDESCENT FRAGMENTS

Eduardo M. B. Campello 1
1
Department of Structural and Geotechnical Engineering, University of São Paulo, Brazil
campello@usp.br

The release of incandescent fragments from bulk materials (usually metallic or ceramic) is a common
side effect in many human activities and engineering applications. This is the case of (but is not
restricted to) manufacturing processes that involve cutting, grinding, sanding and welding, as well as
metal-to-metal contact observed in construction sites, worn-out brakes, ballistic impacts, explosions,
pyrotechnics, etc. Besides being potentially hazardous if in direct contact to humans, incandescent
fragments may be the starting point of unwanted fires, especially if in dry or semi-dry environments.
Depending on (i) their thermal state, (ii) their kinetic energy when released and (iii) the type of
material on which they land, incandescent fragments can be a dangerous source of fire ignition and
human injury. These aspects depend on a number of factors, among which the most relevant are the
thermal properties and the trajectories of the fragments. In a recent work by Zohdi [1], a simple
computational model was proposed to track the time evolution of hot fragments with respect to both
position and temperature. Therein, however, the fragments were allowed to have only translational
motion. The present work is an extension of Zohdi´s model in the sense that particle spin is now
incorporated into the problem´s dynamics, affecting both the particles´ trajectories and their cooling
rates. Our approach is based on discrete particle dynamics following the works of Campello [2,3],
combined with standard differential equations from heat transfer to capture the temperature evolution
of the fragments. We use the released energy from an initial blast pulse to provide the starting kinetic
energy of a system of particles, and then numerically compute the trajectory and thermal state of the
particles over time. Gravitational settling, drag forces, drag-induced heating, as well as convective and
radiative cooling, are considered. Additionally, since the particles are allowed to spin, and thereby
have a more complex interaction with the surrounding air, the influence of Magnus forces is further
investigated. It is shown that particle spin may have a quite significant effect on the spatio-thermal
footprint of the fragmentation, especially when small to mid-sized fragments are present (of the order
of fractions of millimeters to a few millimeters). We believe that simple particle models of the type as
shown here may be a useful tool to estimate fire hazards distributions as well as to guide fire safety
rules in industrial and construction workplaces.

References

[1] T. I. Zohdi. Modeling the spatio-thermal fire hazard distribution of incandescent material ejecta in
manufacturing, Comput. Mech., 2018 (online first) (https://doi.org/10.1007/s00466-018-1617-2).
[2] E. M. B. Campello. A computational model for the simulation of dry granular materials, Int. J.
Nonlinear Mech., v. 106, 89–107, 2018 (https://doi.org/10.1016/j.ijnonlinmec.2018.08.010).

152
[3] E. M. B. Campello. A description of rotations for DEM models of particle systems, Comp. Part.
Mech., v. 2, 109–125, 2015.

153
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

RELIABILITY ANALYSIS OF A CONCRETE FRAME.

Eduardo A. K. Naccache1, Valério S. Almeida2,


1
University of São Paulo, eduardo.naccache@usp.br.
2
University of São Paulo, valerio.almeida@usp.br

Abstract. As the concrete used in each floor of a building or in each group of elements comes from
different mixtures, its statistical parameters can be assumed constant only inside the same mix. In this
work a reliability analysis of the horizontal displacement of a concrete frame is undertaken
considering different statistical parameters for the concrete of different groups of elements. The
horizontal load is an equivalent static wind with statistical properties. The concrete frame is linear
elastic and the reliability methods applied are the Monte Carlo with simple sampling and the First
Order Reliability Method (FORM).

Keywords: Reliability; Finite element method; Service limit state; Equivalent static wind.

1. Introduction
Reliability has found a large application field in structural engineering. Although very far from a pure
probabilistic approach, the current reliability methods allow the performance assessment of different
structural requirements. A performance-based wind engineering (PBWE) is proposed in Tessari,
Kroetz and Beck (2017). The performance of a steel latticed tower subjected to wind loads, either
equivalent static wind as dynamic wind, are assessed by means of reliability methods. The potential of
the PBWE with probabilistic approach for the design of steel towers was made evident as it is a way to
make a balance between expected failure costs for different performance measures and to optimize the
structure performance before different load scenarios.
In earthquake engineering the performance-based design has found even more acceptance. A four-
storey concrete building under seismic action has its horizontal displacement performance verified in
Abdelouafi, Benaissa and Abdellatif (2015). The seismic analysis was a static equivalent lateral load
called pushover analysis. The results of some reliability methods such as the FORM, Importance
sampling and Response surface were compared and found good agreement but with FORM being a
much less time consuming as it is expected when the problem involves a low number of variables.
In Corelhano, Corrêa and Beck (2012), the reliability of concrete frames regarding the top horizontal
displacement was studied too. The most important feature is the model error used in the limit state
function. The horizontal displacement of forty-two frames with different reinforcement ratios; number
of floors; number of bays; columns sections; beams sections were calculated one time with simplified
models and one time with rigorous models to represent cracking of the concrete. The horizontal

154
displacement ratio between these two situations is the variable which represents de model error and its
parameters could be estimated from these forty-two observations.
In this work the reliability analysis of concrete frames considering different concrete parameters for
different groups of elements is aimed. The non-linearities are not considered so the structure model is
relatively simple, but on the other hand, the reliability analysis regards hundreds of variables as each
element dimension is regarded as such.

2. Main objectives
This work has some important objectives. First one is applying reliabilities methods in a problem that
involves a great number of variables and understand the pros and cons of each one of them. The
second one is to develop a design procedure in the moulds of the Performance based engineering
(PBE) that would lead a probable designer to an optimal and functional solution. The third and the last
one is to make way for more complex structural analysis to be associated with the performance-based
design, also for not letting the simpler ones be forgotten as not deserving a modern treatment.

Figure 1 – Two-dimensional frame without and with soil structure interaction.

3. Concluding remarks
With this work the authors expect to get along a promisor area of the engineering research, what very
probably is, for the evolution of the design process seems to pass through a performance measure,
slowly leaving behind the traditional design philosophies.
Beside the evolution of the design process goes the search for increasing realism of the structural
analysis. In this matter, the authors expect to extend the reliability analysis for structures with
increasingly complexity degree such as three-dimensional frames and considering the soil structure
interaction.

155
What becomes clear is the fact that as the complexity level of the structural problems increases, the
performance-based design comes out as the more reasonable approach. It is not by chance that in the
earthquake engineering this new philosophy has found its first application. Then, without mistake, one
can state that as the engineering community abandon the very much simplified methods and
considerations such as the equivalent static wind and the rigid interaction between the soil and the
foundations, more necessary the performance-based design will be.

Acknowledgments
The authors would like to thank the academic support given by the PPGEC-EPUSP (Programa de Pós-
Graduação em Engenharia Civil da Escola Politécnica da Universidade de São Paulo) and the financial
support of CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior) and Cnpq
(Consenlho Nacional de Desenvolvimento Científico e Tecnológico).

References

[1] Al-Ansari, M. S. Reliability index of tall buildings in earthquake zones. Open journal of
Earthquake Research, 2, 39-46, 2013.
[2] Abdelouafi, E. G., Benaissa, K. Abdellatif, K. Reliability Analysis of reinforced concrete
buildings: comparison between FORM and ISM. Procedia Engineering, 114, 650-657, 2015.
[3] Augusti, G., Ciampoli, M. Performance-based design in risk assessment and reduction.
Probabilistic Engineering Mechanics, 23(4):496-508, 2008.
[4] Beck, A. T., Carrêa, M. R. S. New Design Chart for Basic Wind Speeds in Brazil. Latin American
Journal of Solids and Structures, v. 10, pp 707-723, 2013.
[5] Beck, A. T., Souza Jr, A. C. A first attempt towards reliability-based calibration of brazilian
structural design codes. Journal of the brazilian society of mechanical sciences and engineering, Vol.
32, No.2, p 119-127, 2010.
[6] Ciampoli, M., Petrini, F., Augusti, G. Performance-based wind engineering: towards a general
procedure. Structural Safety, 33(6), 367-378, 2011.
[7] Ciampoli, M., Petrini, F. Performance-based aeolian risk assessment and reduction for tall
buildings. Probabilistic Engineering Mechanics, 28, 75-84, 2011.
[8] Corelhano, A. G. B., Corrêa, M. R. S., Beck, A. T. Reliability of buildings in service limit state for
maximum horizontal displacements. Ibracon structures and materials journal, v.5, pp. 84-103, 2012.
[9] Davenport, A. G. The relationship of reliability to wind loading. Journal of wind engineering and
industrial aerodynamics, 13, 3-27, 1983.
[10] Ellingwood, B. R. Probability-based codified design: past accomplishments and future
challenges. Structural Safety, 13(3):159-76, 1994.
[11] Ellingwood, B. R., Tekie, P. B. Wind load statistics for probability-based structural design.
Structural Engineering ASCE, 125(5):453-63, 1999.
[12] Epaarachchi, D. C., Stewart, M. G., Rosowsky, D. V. Structural reliability od multistory
buildings during construction. Journal of structural engineering, vol. 128, No. 2, 205-213, 2002.
[13] Hendawi, S., Frangopol, D. M. System reliability and redundancy in structural design and
evaluation. Structural Safety, 16(1-2):47-71, 1994.
[14] JCSS. Probabilistic model code. Joint Committee on structural safety. 2001.
[15] Kliukas, R., Kudzys, A. Reliability index design in reinforced concrete structures of annular
cross sections. 10th International Conference on Modern buildings materials, structures and technique,
2010.
[16] Martins, R. G., Santos, S. H. C. Análise de confiabilidade de fundações de máquinas pelo método
de monte carlo. XI Congresso brasileiro de pontes e estruturas, 2016.

156
[17] Naccache, E. A. K., Almeida V. S. S. Analysis of a framed structure considering the SSI and
subjected to dynamic wind load. Proceedings e, Journal, page1-pageN, year.
[18] Naccache, E. A. K., 2017. Confiabilidade de estacas. Novas edições acadêmicas. Saarbrücken,
Deutschland, 2017.
[19] NBR6123:1988: Wind Loads in Buildings. ABNT – Brazilian Association of Technical Codes,
Rio de Janeiro.
[20] Martins, R. G., Santos, S. H. Análise de confiabilidade de fundações de máquinas pelo método de
Monte Carlo. IX Congresso brasileiro de pontes e estruturas, 2016.
[21] Paulotto, C., Ciampoli, M., Augusti, G. Some proposals for a first step towards a performance-
based wind engineering. First International forum in engineering decision making, 5-9, 2004.
[22] Rackwitz, R. Optimization – the basis for code-making and reliability verification. Structural
Safety, vol. 22, 27-60, 2000.
[23] Ranganathan, R., Deshpande, C. Reliability analysis of reinforced concrete frames. Journal of
Structural Engineering, vol. 113, No. 6, 1315-1328, 1987.
[24] Sorensen, J. D., Kroon, I. B., Faber, M. H. Optimal reliability-based code calibration. Structural
Safety, vol. 15, 197-208, 1994.
[25] Tessari, K. R., Kroetz, M. H., Beck, A. T. Performance-based design of steel towers subject to
wind action. Engineering Structures, 143, 549-557, 2017.
[26] Wilson, E. L. Three-Dimensional Static and Dynamic Analysis of Structures. Computers and
Structures Inc. Berkeley, California, 2002.

157
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

PARAMETERS E STIMATION FOR KINETIC MODELING OF 18 F-FDG IN P OSITRON


E MISSION T OMOGRAPHY (PET)
Eliete Biasotto Hauser 1 , Evandro Manica 2 , Gianina T. Venturin 3 , Samuel Greggio 4 ,Eduardo Rigon
Zimmer 5 , Wyllians Vendramini Borelli 6 , Jaderson Costa Da Costa 7
1 Brain Institute of Rio Grande do Sul (BraIns)/School of Sciences, PUCRS, eliete@pucrs.br
2 Institute of Mathematics, UFRGS, evandro.manica@ufrgs.br
3 Brain Institute of Rio Grande do Sul (BraIns), PUCRS, gianina.venturin@pucrs.br

4 Brain Institute of Rio Grande do Sul (BraIns), PUCRS, samuel.greggio@pucrs.br


5 Department of Biochemestry, UFRGS, eduardo.zimmer@ufrgs.br
6 School of Medicine, PUCRS, wyllians.vb@gmail.com
7 Brain Institute of Rio Grande do Sul (BraIns)/School of Medicine, PUCRS, jcc@pucrs.br

The main purpose of this study is to estimate the unknown transport rates in a irreversible two-tissue
compartment model, solved by Laplace transform method, for kinetic modeling of [18F]2-fluor-2deoxy-
D-glucose(FDG) , in order to quantify amyloid in Positron Emission Tomography(PET) image. The
image-derived arterial input function (IDAIF) is obtained, noninvasively, from a first order differential
equation that describe the dynamics of the radiotracer on the from the discrete time activity curve in
a carotid volume of interest(VOI) (known to be amyloid free). The transport constants are determined
applying the initial conditions, the effective dose injeceted, the half-life of the tracer, the discrete activity
curves (TAC) and, iteratively, minimizing the sum of the quadratic residuals(Levenberg-Marquardt).
After calculating a convolution integral, the analytical solution is completely described. We present
numerical results generated by this methods applied in data obtained from the projects at the Brain
Institute (InsCer/BraIns) at Pontifical Catholic University of Rio Grande do Sul (PUCRS) and comparing
our results with those published in the literature.
The mathematical model for the problem is expressed by the system of two differential equations:


 d
 dt C1 (t) = K1 Ca (t) − ( k2 + k3 )C1 (t)

(1)

 d

 C2 (t) = k3 C1 (t)
dt
where, Ca (t) is the IDAIF, C1 (t) and C2 (t) are, respectively, the concentration in the non-displaceable
and displaceable compartments, K1 and k2 , k3 , k4 are kinetic rate constants which have to be determined.
In order to determine the IDAIF Ca (t) , we use the dynamics of the radiotracer on the reference
region(carotids), governed by the differential equation
dCr 0 0
= K1Ca (t) − k2Cr (t) (2)
dt
where Cr (t) is the concentration of the radiotracer in the reference region, K1 0 > 0 and k2 0 > 0 are pro-
portionality rates describing, respectively, the tracer influx into and the tracer outflow from the reference
tissue(Carotids VOI). Cr (t) will be approximated by means of nonlinear regression of the data obtained
from a TAC curve on a Positron Emission Tomography(PET) image (Figure1 and Figure2). Data used
to determine the TAC and Ca (t) come from PET image data obtained from a experiment performed at
Brain Institute(BraIns) at Pontifical Catholic University of Rio Grande do Sul (PUCRS).
The equations (1) and (2) are integrated with respect to time together with their partial derivatives
with respect to the parameters ki .

158 1
Figure 1: Carotids Volume of interest(VOI) on the positron emission tomographic(PET).

Figure 2: Discrete time activity curves(TAC)- Left and right carotids.

 We apply  the Laplace transform with respect to t in Eq.(1), denoting £ {Ci (t)} = Ci (s) and
d Ck (t)
£ = sCi (s) −Ci (0) .
dt
With C1 (0) = 0 and C2 (0) = 0, obtain the an algebric system of 2 equations:

 ( s + k2 + k3 )C1 (s) = K1Ca (s)
(3)

− k3C1 (s) + s C2 (s) = 0

Then, using the inverse Laplace transform, Ci (t) = £−1 Ci (s) , and we consider ∗ to denote the
convolution operation, we obtain

159 2
   
−1 K1Ca (s) −1 1 
C1 (t) = £ = K1 £ ∗ £−1 Ca (s)
( s + k2 + k3 ) ( s + k2 + k3 )
(4)
 
k3C1 (s) 
C2 (t) = £−1 = k3 ∗ £−1 C1 (s) .
s
The analtical solution is
Z t
−( k2 +k3 )t
C1 (t) = K1 e ∗ Ca (t) = K1 e−( k2 +k3 ) (t−u) Ca (u) du
0
Z t
(5)
C2 (t) = k3 ∗ C1 (t) = k3 C1 (u) du .
0

Keywords: Alzheimer’s Disease (AD), Image-Derived Arterial Input Function (IDAIF), Kinetic Mo-
delling, Parameter Estimation, Laplace Transform, Positron Emission Tomography(PET).

References
[1] A. Arnold, D. Calvetti, A. Gjedde, P. Iversen, E. Somersalo. Astrocytic tracer dynamics estimated from
[1 −11 C]-acetate PET measurements , Mathematical Medicine and Biology, page367-page382, 2015.
[2] D. M. Bates, D. G.Watts Nonlinear Regression and Its Applications, Wiley, 1988.
[3] W. V. Borelli, L.Schilling, G. Radaelli, J.C.DaCosta. Neurobiological findings associated with high cogni-
tiveperformance in older adults: a systematic review, International Psychogeriatrics, page1-page13, 2018.
[4] V. J. Cunningham,T. Jones. Spectral analysis of dynamic PET studies, J. Cereb Blood Flow Metabolism,
page15-page23, 1993.
[5] P. Z. Fregonara, E. M. Fadaili, R. Maroy, C. Comtat, A. Souloumiac, S. Jan, M. Ribeiro, V. Gaura, A. B.Hen ,
R. Trebossen. Comparison of eight methods for the estimation of the image-derived input function in dynamic
[18F]-FDG PET human brain studies, Journal of Cerebral Blood Flow Metabolism, page1825âpage1835,
2009.
[6] E. B. Hauser,G. T. Venturin, s. Greggio, W. Borelli, J. C. da Costa, Image-Derived Carotid Arterial In-
put Function as an Inverse Problem in Kinetic Modeling of [18F]2-Fluoro-2 Deoxy-D-Glucose(FDG) in
Azheimer’s Disease, 15th International Symposium on Computer Methods in Biomechanics and Biomedical
Engineering, page1-page7, 2018.
[7] E. B. Hauser, G. T. Venturin, S. Greggio, E. Manica, E. R. Zimmer, J.C.C. Costa. Laplace Transform Method
for 11C-PIB Two-Tissue Reversible Compartment Model with Image-Derived Arterial Input Function, Ibero-
Latin American Congress on Computational Methods in Engineering, ABMEC, page1-page10, 2017.
[8] E. B. Hauser, G. T. Venturin, S. Greggio, J.C.Costa. Mathematical modeling to quantify the pharmacoki-
netic process of [18F]2-fluor-2deoxy-D-glucose (FDG). Constanda,Kirsch. Integral Methods in Science and
Engineering, Birkhäuser, 2015.
[9] E. B. Hauser. Laplace Transform in Tracer Kinetic Modeling, Meeting on Nuclear Applications, ABEN,p
age1-page7, 2013.
[10] R. B. Innis, V. J. Cunningham, J. Delforge, M. Fujita, A. Gjedde, R. N. Gunn, J. Holden, S. Houle, S. Huang,
M. Ichise, H. Iida, H. Ito, Y. Kimura, R. A. Koeppe, G. M. Knudsen, J. Knuuti, A. A. Lammertsma, M.
Laruelle, J. Logan, R. P. Maguire, M. A. Mintun, E. D. Morris, R. Parsey, J. C. Price, M. Slifstein, V. Sossi,
T. Suhara, J. R. Votaw, D. F. Wong, R. E. Carson. Consensus nomenclature for in vivo imaging of reversibly
binding radioligands, Journal of Cerebral Blood Flow Metabolism, page1533âpage1539, 2007.
[11] F. Kiessling, B. J. Pichler. Small Animal Imaging, Springer, 2011.
[12] M. M. Khalil, Basic Sciences of Nuclear Medicine, Springer, 2011.
[13] ] R. A. Koeppe, V. A Holthoff. K.A. Frey, M.R. Kilbourn and D.E. Kuhl. Compartmental Analysis of
[llC]Flumazenil Kinetics for the Estimation of Ligand Transport Rate and Receptor Distribution Using
Positron Emission Tomography, Journal of Cerebral Blood Flow and Metabolism, page735-page744, 1991.

160 3
[14] R. Laforest, T. L. Sharp, J. A. Engelbach, N.M. Fetting, P. Herrero, J. Kim, J. S. Lewis, D. J. Rowland, Y. C.
Tai, M. J. Welch. Measurement of input functions in rodents: challenges and solutions. Nuclear Medicine and
Biology. page679-page685, 2005.
[15] N. A. Lassen. Neuroreceptor Quantitation In Vivo by the Steady-State Principle Using Constant Infusion or
Bolus Injection of Radioactive Tracers, Journal of Cerebral Blood Flow and Metabolism, page709-page716,
1992.
[16] C. C. Lee, A. Nayak, A. Sethuraman, G. Belfort, G. J. McRae. A Thrre-Stage Kinetic Model of Amyloid
Fibrillation, Biophysical Journal, page3448-page3458, 2007.
[17] B. J. Lopresti, W.E. Klunk, C.A. Mathis, J.A. Hoge, S.K. Ziolko, X. Lu, C.C. Meltzer, K. Schimmel, N.D.
Tsopelas, S. T. DeKosky, J.C. Price. Simplified quantification of Pittsburgh Compound B amyloid imaging
PET studies: a comparative analysis.. J. Nucl Med. page1959-page72, 2005.
[18] D. W. Marquardt. An Algorithm for Least-Squares Estamation of Nonlinear Parameters,J. S. Industr. Appl.
Math,page431-page441, 1963.
[19] N.Nelissen, J.Warwick, P. Dupont. Kinetic Modelling in Human Brain Imaging, Chia-Hung Hsieh, 2012.
[20] A. B. Rodell, G. O’Keefe, C.C. Rowe, V.L. Villemagne, A. Gjedde. Cerebral Blood Flow and AB-Amyloid
Estimates by WARM Analysis of [11C] PiB Uptake Distinguish among and between Neurodegenerative Dis-
orders and Aging, Frontiers in Aging Neuroscience, page1-page11, 2017.
[21] Y. Su, T. M. Blazey, A. Z. Snyder, M.E. Raichle, R. C. Hornbeck,P. Aldea, J. C. Morris, T. L. S. Benzinger.
Quantitative Amiloid Imaging Using Image-Derived Arterial Input Function, PLoS One. page1-page16, 2015.
[22] D. Vriens, L-F. de Geus-Oei, W.J.G. Oyen, E.P. Visser. A Curve-Fitting Approach to Estimate the Arterial
Plasma Input Function for the Assessment Of Glucose Metabolic Rate and Response to Treatment, The Journal
of Nuclear Medicine, page1933-page1939, 2009.
[23] S. Basu, H. Zaidi. S. Holm, A. Alavi. Quantitative Techniques in PET-CT Imaging, Current Medical Imaging
Reviews. page216-page233, 2011.
[24] H. Zaidi. Quantitative Analysis in Nuclear Medicine Imaging, Springer, 2006.
[25] . Zhou, K. Chen, E.M. Reiman, D.Li, B. Shanet. A method for generating image-derived input function in
quantitative 18F-FDG PET study based on the monotonicity of the input and output function curve, Nuclear
Medicine Communications, page362âpage370, 2012.

161 4
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A FIBER OPTIMIZATION METHOD BASED ON NORMAL DISTRIBUTION FUNC -


TION FOR HYPERELASTIC MATERIALS

Andre Luis Ferreira da Silva1 , Ruben Andres Salas1 , Emilio Carlos Nelli Silva 1
1 Department of Mechatronics System Engineering, University of Sao Paulo, SP, Brazil

1. Introduction
Fiber-reinforced composite materials have been used more frequently and their properties can be im-
proved by optimum fiber orientation. Models which are based in discrete candidate angle as the models
DMO, SFP, BCP and NDFO are efficient to avoid multiple local minima and problems related to initial
fiber configuration. Among the methods of optimization of fiber angles the NDFO method stands out.
Such method uses only one variable per element, regardless of the number of chosen angles. However,
the NDFO method has been applied so far in materials where the stress-strain relationship can be con-
sidered linear. This work proposes the application of an adaptation of NFDO using the Neo-Hookian
model of Ciarlet-Simo, which can be used for materials subjected to large deformations. To consider
the effect of the fiber on the element, a transversely isotropic part is added to the isotropic portion of
the strain energy function. The optimization problem solved in this work is the well-known minimum
compliance design. The unwanted effects of fiber discontinuity are treated using the filter proposed by
Lazarov which consider as a filter a solution of a modified Helmholtz equation. A numerical example is
presented to demonstrate the result of the proposed method.

2. Theoretical Formulation
A hyperelastic material is one for which there is a strain energy function per unit undeformed volume Ψ
depend upon the right Cauchy-Green tensor C as [1] [2] [3]

Ψ = Ψ(C) (1)
Since 12 Ċ is work conjugate to the second Piola-Kirchhoff tensor S, its possible to determine the deriva-
tive of Ψ in respect to time as [3]

∂Ψ 1
Ψ̇ = : Ċ = S : Ċ (2)
∂Ċ 2
and therefore S is defined as

∂Ψ
S=2 (3)
∂C
For isotropic materials, the strain energy function can be written as a function of the invariants of C as

Ψ(C) = Ψ(I1 , I2 , I3 ) (4)

162 1
where the invariants I1 , I2 and I3 are defined as [4]

I1 =tr(C) (5a)
I2 =C : C (5b)
2
I3 =det(C) = J (5c)

therefore the second Piola-Kirchhoff tensor is determined by

∂Ψ ∂Ψ ∂Ψ −1
S=2 I+4 C + 2I3 C (6)
∂I1 ∂I2 ∂I3
Due to Ψ can be an arbitrary function of the invariants, Eq. 6 becomes complex because of a large number
of possible constitutive parameters [4]. The Neo-Hookean hyperelastic equation can be used for large
strain range and has only two constitutive parameters. The strain energy and the second Piola-Kirchhoff
tensor for Neo-Hookean hyperelasticity are defined as [2]

1 1
Ψnh = µ(I1 − 3) − µ ln J + λ(J − 1)2 (7)
2 2

Snh = µ(I − C−1 ) + λJ(J − 1)C−1 (8)


where µ and λ are Lamé coefficients, J is the Jacobian and I is the identity matrix.
A transversely isotropic material is characterized by an isotropic plane and physical properties symmet-
rical with respect to this plane. Fig. 2. represents a transversely isotropic in an undeformed configuration
Ωr and in a deformed configuration Ω. The isotropic plane is obtained from the directions 1 − 2. In
direction 3 there is a vector a which represents the main direction of orthotropy.

Figure 1: Transversely isotropic solid in undeformed and deformed configurations.

The portion of strain energy Ψtrn which represents the transversely isotropic behavior is a function of
two pseudo invariants of C given by

I4 =ar · C ar (9a)
2 r
I5 =ar · C a (9b)

The transversely isotropic portion Ψtrn can then be defined as

1
Ψtrn = [α + β lnJ + γ(I4 − 1)] − α(I5 − 1) (10)
2
and the corresponding second Piola-Kirchhoff tensor is [2]

Strn = 2β(I4 − 1)C−1 + 2[α + 2β lnJ + 2γ(I4 − 1)]ar ⊗ ar − α(Car ⊗ ar + ar ⊗ Car ) (11)
where α, β and γ are constants that depend on Young’s modulus and the Poisson’s ratio.

163 2
3. Topology Optimization Formulation
The optimization of fiber angles based in discrete methods follows the basic concept of choosing a set of
candidate angles and using a weight function to determine an effective elastic tensor as a weighted sum
of the elastic tensors which corresponds to each candidate. However, owing to the fact that in this work
it is used a hyperelastic model, it is necessary to utilize a different approach. Since the collaboration
of the fiber for the stiffness of the material is considered in the transversally isotropic part of second
Piola-Kirchhoff tensor and considering the fact that when some load is applied parallel to the isotropic
plane, only the matrix of the composite material is loaded, thus the second Piola-Kirchhoff tensor can be
defined as
nc
S = Snh + ∑ wi Stri (12)
i=1
where wi are the weighting functions and nc the number of candidate angles.
All methods which use a selection of discrete angles are intended to drive only one weighting function
wi to 1, whereas all the others are driven to 0. What differentiates fiber optimization methods based on
the selection of candidate angles is the way the wi function does the parameterization. In this work wi is
given by [5]

ŵi
wi = nc (13)
∑k=1 ŵk
where nc is the number of candidate angles and ŵi is a normal distribution function which can be written
as [6]
(θ−θic )2

ŵi = e 2p2
θ
(14)

where i represents the number of candidate angles, θ is the design variable, θic represents the candidate
angles and pθ is a penalization constant.
The optimization problem solved in this work consisting of the well-known compliance minimization
which is defined as
Z Z
min b · u dΩ + t · u dS
Ω ΓN
Z Z Z
s.t. P : ∇v dΩ = b · v dΩ t · v dS (15)
Ω Ω ΓN
min
θc ≤ θ ≤ θmax
c
where b are the body forces, t are the surface forces, v is a test function, u are the displacements and P
is the first Piola-Kirchhoff stress tensor.
The relation between the first and the second Piola-Kirchhoff tensors is given by

P = S · FT (16)
where S is the second Piola-Kirchhoff tensor and F is the deformation gradient. S is calculated as a sum
of the isotropic part with a weighting sum of the transversally isotropic part, such as
nc
S = Snh + ∑ wi Strn,i (17)
i=1

4. Numerical Example
The numerical example presented here is a box domain with dimensions represented by Fig 4. Table 1
shows the material properties used in this example. The candidates angles are

θc = [−78.75, −67.5, −56.25, −45, −33.75, −22.5, −11.25, 0, ...symmetric..., 90] (18)

164 3
and the domain has 120x40x6 tetrahedral elements linearly interpolated.

(a) Domain (b) Solution

Figure 2: Fully clamped beam with in-plane distributed load

EA [N/mm2 ] E[N/mm2 ] GA [N/mm2 ] ν F[N] θ0 [◦ ] pθ,0


38.3 · 103 8.27 · 103 4.14 · 103 0.26 [0, −100, 0]T 45 10

Table 1: Material properties and initial conditions

5. Conclusion
An optimization method for fiber orientation based on NDFO method considering hyperelastic mate-
rial has been proposed. The proposed method uses weighting functions to select only one transversely
isotropic second Piola-Kirchhoff among a set of candidates.
Continuity is achieved by using a modified Helmholtz equation as a filter method. By increasing the r
value in the filter, it is obtained higher values of the objective function.
The result shows in the example demonstrates it is possible to optimize the fibers orientation by using
a formulation where the linear elastic behavior approximation is not considered. With this formulation,
the number of possible modeled materials has been increased and there is no limitation imposed by the
small deformations.

6. Acknowledgements
ALF Silva thanks CAPES ( Coordination of Superior Level Staff Improvement) and INCT/CEMTEC
(Nacional Institute on Advanced Eco-Efficient Cement-Based Technologies) for the financial during his
Master’s Degree under Grant 88887.165790/2018-00. Authors also acknowledge the support of the
RCGI (Research Centre for Gas Innovation), hosted by the University of São Paulo (USP) and spon-
sored by FAPESP (2014/50279-4) and Shell Brazil.

References
[1] Ogden, Raymond W Non-linear elastic deformations, Courier Corporation, 1997.
[2] Bonet, J. and Burton, AJ. A simple orthotropic, transversely isotropic hyperelastic constitutive equation for
large strain computations, Computer methods in applied mechanics and engineering, 151-164, 1998.
[3] Bonet, Javier and Wood, Richard D. Nonlinear continuum mechanics for finite element analysis, Cambridge
university press, 1997.

165 4
[4] Wriggers, Pete Nonlinear finite element methods, Springer Science & Business Media, 2008.
[5] J. Stegmann, E. Lund. Discrete material optimzation of general composite shell structures. Int J Number
Methods Eng 2008, 62(14) 2009-27
[6] Kiyono, C. Y., Silva, E. C. N., & Reddy, J. N. (2017). A novel fiber optimization method based on normal
distribution function with continuously varying fiber path., Composite Structures, 160, 503-515.

166 5
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

N UMERICAL INVESTIGATION OF STRONG ADDED MASS EFFECT FOR FLUID -


STRUCTURE CALCULATIONS APPLIED TO MOVING HYDROFOILS

Emmanuel LEFRANÇOIS

Roberval Laboratory CNRS FRE 2022 - Sorbonne Universités - Université de Technologie de Compiègne
CS 60319, 60203 Compiègne, France - e-mail: emmanuel.lefrancois@utc.fr

1. Context and interest of this study


The research presented in this article focuses on the development of a numerical tool for investigating
fluid-structure interactions (FSI) between a fluid flow that is not confined (infinite) and a current turbine
with blades. Many similarities may be observed from its aerial version (wind turbine) but a major point
of concern results from a fluid density 800 times higher than in the air. The constant search for an
optimal solution (by increasing size and reducing mass) inevitably leads to flexible behavior resulting
from hydrodynamic loads, and this flexible behavior may have serious impacts on the efficiency of the
device. The FSI model may here without loss of generality be restricted to a 2D airfoil placed in a flow
and having two degrees of freedom (dof), namely plunging and pitching motions. The Panel Method
approach [2, 3] is of particular interest for fluid flow calculations around lifting device due to the fact it
has been originaly designed for. This potential approach is restricted to incompressible and irrotational
flows, however if completed by a Kutta condition, it can be extended to lifting flows. The FSI approach
is here based on a partitioned coupling with a dedicated solver for each of the two physics (namely fluid
flow and structure dynamics). Exchanges take place regularly between the two solvers via a coupling
scheme [4, 5] that is based on successive solutions produced by the fluid and structure solvers. The
coupling is said to be loosely coupled partitioned if only one shot (that is to say a single computation)
per time step is required for each field, and strongly coupled partitioned if an iterative procedure is used
to ensure convergence of the coupled solution. The major drawback of the standard partitioned FSI
coupling scheme is that where higher density fluids are involved (meaning strong effects of added mass),
convergence is no longer guaranteed, and divergence will generally be observed, regardless of the chosen
time step for incompressible flows [6]. The objective in this paper is to show that in order to take into
account heavy fluid flows such as in sea currents, the coupling scheme must be corrected, as described
for example in [1], in order to counteract the penalizing impact of the added mass effect on the classical
FSI coupling scheme. This correction is based on estimating an added mass matrix [Madd e ] that may

considerably improve and/or ensure the iterative phase of a strongly coupled partitioned approach.

c
θ(t) 25%c
~n P
Kθ ~eθ ~r
w(t) P F
V∞, ρ α G F cpi ~t
chord c N −1
z,~k Kz z N
N +1
y,~ Position at rest x 4 cp3
3 cp2
2 cp1 1

0 x,~ı O
(a) Dynamics of a 2D hydrofoil (b) Panel Method discretization
Figure 1: Fluid-stucture interaction for a 2D hydrofoil with plunging and pitching motions

167 1
2. Fluid-structure coupling for a 2D movable hydrofoil
2.1. Dynamics for a hydrofoil with two degrees of freedom
Here we consider a 2D airfoil with chord length c, of mass m and flexibly attached to a fixed point,
as illustrated in Figure 1(a). Two dof are here considered, namely plunging w(t) and a pitching θ(t)
motions. From Lagrange’s equation and the kinetic and potential energies of the airfoil we obtain the set
of two equations:
       
m −mPG ẅ Kz 0 w Rz
+ = (1)
−mPG IP θ̈ 0 Kθ θ M P + PG mg
with Kz denoting the axial rigidity along the z−axis, Kθ the torsional rigidity with respect to the y−axis
and I |G the mass moment of inertia about G. The terms Rz and M P denote respectively the vertical
component of the generalized force obtained from the pressure integration, and the resulting pitching
moment at P. Finally, this may be condensed to the following:

[M]{Ü } + [K]{U (t)} = {Fp (t)}, with {U (0)} = {U 0 } and {U̇ (0)} = {0}, (2)

where [M] and [K] denote respectively the mass and the rigidity matrices corresponding to the attachment
of the airfoil, and {U } denotes the two dof. The term {Fp } denotes for the sollicitation vector resulting
from aerodynamic loads.

2.2. Panel Method (PM) for lifting potential flows


Panel methods are particularly suitable for calculating the flow field over an airfoil that undergoes un-
steady time-dependent motion in a fluid that may be assumed inviscid and incompressible. The main
idea in the Panel Method is not to solve this Laplacian equation in the classical way for the entire fluid
domain, but to cast the same analysis in a boundary integral equation form, where the Hess & Smith
Panel Method (HSPM) was introduced, is considered to be the reference paper. In this approach, with


2D non-stationary flows being restricted as set out in [9], the velocity V at any point ~x = (x, z) of the
fluid domain is decomposed according to:


− →
− σ(s)~r τ(s) V2 ∂ϕ
˛ ˛
V (~x) = V ∞ +~v with ~v = ds + ~eθ ds and p + ρgz + ρ +ρ = f (t). (3)
2πr 2 2πr 2 ∂t
S S



where V ∞ defines the velocity of the uniform flow at infinity. The vector ~v denotes the disturbance
field due to the airfoil and results from two contributions, since the airfoil may be represented by two
elementary flows (also called singularities) corresponding to source flow (σ(s)) and vortex flow [9] (τ(s)).
For points belonging to the interface between the flow and the airfoil, the boundary condition that
stipulates no flow through surface enables us to define a given set of equations to be solved for σ(s).
Completed by the Kutta condition that stipulates that the flow must leave the trailing edge smoothly, the
set to be solved for σ(s) and τ is now complete. In order to couple the fluid flow with the structure,
we need to know the pressure p. This may be calculated at any point using the non-stationary form of
Bernoulli’s equation.

2.3. Fluid-structure coupling scheme corrected from added mass effect


The coupling process is required to perform FSI calculations in order to regularly update the variables
common to both the fluid and the structure solvers. In order to better counteract the added mass effect that
results from heavy fluid flow such as in a liquid (sea currents) and that may lead to divergence, here we
propose correcting the classical FSI coupling scheme in relation to the added mass effect. The main idea
(in the case of conservative systems only) is that if the real added mass matrix [Madd, f ] could be calculated
exactly, the force term appearing in equation (2) would be exactly replaced by {Fp }i ≡ −[Madd, f ]{Ü }i .
For most cases, the real added mass matrix [Madd, f ] is out of reach. The classical partitioned coupling

168 2
scheme (denoted by CLAS) is then modified in accordance with [1], and equation (2) is now related to
the corrected scheme (denoted by CORR):

([M] + [Madd,e ]) {Ü }i+1 + [K]{U }i+1 = {Fp }i + [Madd,e ]{Ü }i , (4)

where i and i + 1 are indexes for the iterative process, and [Madd,e ] is the matrix corresponding to the
estimated added mass effect resulting from the pressure load all around the structure. Each component
Madd,e (i, j) is related to the force on the body in the i-axis resulting from a unit acceleration along the
j-axis. At convergence, the two additional terms cancel out and we get back to the original form of the
coupling equation (2). Adding extra terms on both parts of the original equation, helps to reduce the
penalizing effect of {Fp } and to increase the beneficial effect of [M]{Ü } on the convergence process.

3. Results
In this example an hydrofoil (NACA 2412, 105 panels) is flexibly attached to a fixed point, as illustrated
in Figure 1(a). Immersed in a uniform flow (V∞ , ρ), the airfoil is initially removed from its position
at rest (wo = 0.2 m and θo = 8 deg.) until a stationary fluid state is reached. It is then relaxed to allow
the fluid-structure coupling process to take place freely: it will be remarked that the mechanical energy
decreases with time because of a transfer of energy to the fluid tracked in the form of a vortex wake. The
point of this example is to look at the influence of the volumetric mass ρ on the convergence property
of the FSI scheme, to show the severe limit observed for the classical coupling scheme (that we term
CLAS), and finally to show the beneficial effect of the scheme corrected from the added mass effect (that
we term CORR). Fluid flow conditions and structure characteristics are summarized in Table 1.

ρ V∞ m I |G KZ Kθ Madd,e (1, 1) Madd,e (1, 2) Madd,e (2, 2)


[kg/m3 ] [m/s] [kg] [kg/m2 ] [N/m] [Nm/rad] [kg] - -
[1 − 2000] 5 10 100 104 104 0.71ρ −0.15ρ 0.05ρ
Table 1: Fluid flow conditions and structure properties
The estimated added mass matrix [Madd,e ] (symmetric) is calculated according to [10]. The airfoil
is considered fixed over a given number of time steps nFIX , then its flexibility is restored in order to start
the free fluid-structure interaction. The time step ∆t = 10−3 s is related to the two natural frequencies
( f1 = 1.59 Hz, f2 = 5.03 Hz) extracted from an eigenvalue analysis. The same analysis, in accordance
with equation (4), is conducted over nstep time steps, for both the CLAS scheme (with [Madd,e ] = [0])
and the CORR scheme, for a range of volumetric mass given in Table 1. For each time step, the number
of iterations to convergence is extracted. The case (ρ = 8 hg/m3 ) corresponds to the limit observed
for the CLAS scheme to converge. Any density above this value causes the coupling to diverge, as
reported in Table 2. For the considered case, the CLAS scheme excludes densities higher than 20 kg/m3 .

ρ [kg/m3 ] 1 10 20 25 100 1000 2000


# iterations (CLAS) 10 8 8 ∞ ∞ ∞ ∞
# iterations (CORR) 10 8 7 6 6 6 6
Table 2: Effect of density on the averaged total number of iterations for convergence
Above this value the coupling scheme systematically diverges. The CORR scheme, on the other hand,
systematically converges, whatever the density value, which confirms its capacity to support heavy fluids.
But the number of iterations is not the only determinant of whether a scheme is capable of converging.
The results below are solely obtained by using the CORR scheme. Figure 2(a) and (b) shows that
increasing the fluid density gives rise to a predictable higher level of energy dissipation in the fluid,
since a wake and vortex are generated that are simply convected downstream of the flow with their own
energy (which the airfoil is consequently deprived of). Energy signals are normalized according to the
initial energy Eo resulting from initial perturbations wo and θo . The two signals Em and W have opposite
behaviors, in agreement with the principle of energy conservation. In other words, an irreversible transfer
of the mechanical energy Em is observed between the main flow and the airfoil, due to the produced work
W . The higher the density, the higher the observed dissipative effect.

169 3
0.5
0.5

0 0

z [m] −0.5 −0.5

−1 −1
0 1 2 3 4 5 6 7 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5
x [m] x [m]
1 1

0.8
Em 0.8

Ep
E /Eo

0.6 0.6

0.4 0.4
Ek
0.2 0.2

0
W 0
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
time [s] time [s]

(a) ρ = 1 kg/m3 (b) ρ = 10 kg/m3


Figure 2: Dissipative effect due to airfoil energy transfer to the wake
4. Conclusions and prospects
This paper presents a corrected version of a strongly iterative partitioned FSI scheme for studying the
dynamics of an airfoil flexibly attached and immersed in a heavy fluid. The intended application of our
work mainly concerns the fluid-structure coupling that may operate between a moving lifting component
(such as a marine current turbine) and a surrounding heavy fluid such as water. The mathematical model
for the fluid is based on the potential Panel Method that offers the dual benefit of being restricted to a
boundary element analysis and of ensuring the lifting capability of the component. The mathematical
model for the structure, on the other hand, is based on the dynamics of a 2D airfoil that encounters
plunging and pitching motions. Correcting the FSI scheme to counteract the penalizing effect of the
added mass allows convergence to be ensured, whatever the value of the added mass. The application
case is intended to show how the classical FSI scheme is only applicable to a narrow range of fluids
(ρ ≤ 8 kg/m3 ), whereas taking into account the added mass effect on the coupling scheme can ensure the
convergence required by coupling considerations. We are currently looking at the possibility of extending
this approach to 3D applications to cover more realistic rotor geometries (wind mills, marine turbines),
in order to establish the full requirements of FSI calculations for such devices.
References
[1] M. Song, E. Lefrançois and M. Rachik. Development of a partitioned algorithm for fluid-structure coupling
with no fluid density dependency. Computer & Fluids, DOI : 10.1016/j.compfluid.2013.05.022, 2013
[2] Joseph Katz and Allen Plotkin, Low-Speed Aerodynamics, Cambridge Aerospace Series, 2nd Edition, 2001
[3] John J. Bertin and Russell M. Cummings, Aerodynamics for Engineers, Pearson Edition, 6th Edition, 2014
[4] Piperno S, Farhat C, Larrouturou B. Partitioned procedures for the transient solution of coupled aeroelastic
problems. part I: Model problem, theory and two-dimensional application. Comput. Methods Appl. Mech.
Engrg. 124, pages 79-112, 1995.
[5] Felippa CA, Park KC, Farhat C. Partitioned analysis of coupled mechanical systems. Computer Methods in
Applied Mechanics and Engineering, Vol. 190, Issues 24-25, 3247-3270, 2001.
[6] van Brummelen E.H., Added mass effects of compressible and incompressible flows in fluid-structure interac-
tion, J. Appl. Mech. 76 (2009), 021206-7.
[7] Fernández M. A., Gerbeau J.-F. and Grandmont C., A projection semi-implicit scheme for the coupling of an
elastic structure with an incompressible fluid. Int. J. Numer. Meth. Engng., 69 (4), pp.794-821, (2005)
[8] Küttler U., and Wall W., Fixed-point fluid-structure interaction solvers with dynamic relaxation, Computa-
tional Mechanics 43 (2008), no. 1, 61-72.
[9] T.Cebeci, M.Platzer, H.Chen, K.-C. Chang and J.P.Shao, Analysis of Low-Speed Unsteady Airfoil Flows,
Springer, Horizons Publishing, 2005
[10] C. E. Brennen, A review of added mass and fluid inertial forces, Technical Report, Report Number CR 82.010.
Contract Number N62583-81-MR-554, 1982.

170 4
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

N OVEL METHOD FOR ACOUSTIC FLUID - STRUCTURE INTERACTION IN APPLI -


CATION TO OVERALL SAFETY OF STRUCTURES IN QUASI - STATIC SETTING

E. Hadzalic 1,2 , A. Ibrahimbegovic 1 , S. Dolarevic 2


1 Université de Technologie de Compiègne, France
emina.hadzalic@utc.fr, adnan.ibrahimbegovic@utc.fr
2 Faculty of Civil Engineering, University of Sarajevo, Bosnia and Herzegovina

emina.hadzalic@gf.unsa.ba, samir.dolarevic@gf.unsa.ba

I NTRODUCTION
In fluid-structure interaction problems, the external fluid acts both as the source of the loading on the
structure and as source of the saturation of the material. The both aspects of the external fluid influence
have to be taken into account in order to provide a sound design of the structure in terms of structural
integrity and durability. In order to do so, the structure in the interaction problems of practical inter-
est such as dam-reservoir systems, storage tanks or water containers, has to be represented in terms of
saturated porous media. For numerical representation of the structure response, we propose the use of
coupled discrete beam lattice model based on Voronoi cell representation of the domain with nonlinear
Timoshenko beam finite elements acting as cohesive links, capable of modeling the response and lo-
calized failure in pore-saturated structures. Numerical implementation of the internal solid phase-pore
fluid coupling, which is governed by Biot’s porous media theory, results with an additional pressure-type
degree of freedom placed at each node of Timoshenko beam finite element. In fluid-structure interaction
problems of our interest, we can assume a small motion of the external fluid. This hypothesis allows us to
choose mixed displacement/pressure based finite element formulation for the external fluid motion, with
the governing equations derived from the acoustic wave theory. The chosen formulation of the external
fluid motion features both displacement and pressure degrees of freedom. This allows for the structure
and the external fluid finite elements to be connected directly at the common nodes, without any special
numerical treatment of the fluid-structure interface.

D ISCRETE MODEL OF STRUCTURE BUILT OF SATURATED POROUS MEDIA


For numerical representation of the structure response, we choose coupled discrete beam lattice
model based on Voronoi cell representation of the domain with cohesive links as nonlinear Timoshenko
beam finite elements enhanced with additional kinematics [1, 2, 3]. These enhanced kinematics in terms
of embedded strong discontinuities in axial and transverse direction allow for modeling crack formation
and propagation both in mode I and mode II. Mode I relates to crack opening, and mode II relates to
crack sliding. The coupling between the solid phase and the internal pore fluid is introduced in the model
through Biot’s porous media theory and Darcy’s law governing the pore fluid flow [4, 5]. The pore fluid
flow is spread across the mesh of CST (Constant Strain Triangle) finite elements, which coincide with
Delaunay triangles. The duality property between the Voronoi cell and Delaunay triangle representations
in combination with Hammer quadrature rule for numerical integration on CST elements, results with an
additional pore pressure degree of freedom placed at each node of Timoshenko beam finite element.
Biot’s formulation of coupled problem combines equilibrium and continuity equations. The strong
form of equilibrium equations for Timoshenko beam finite element is written as

dN dV dM
+ n(x) = 0; + q(x) = 0; +V (x) + m(x) = 0 (1)
dx dx dx

171 1
where n, q and m are the distributed external loads, N = N 0 − bpAe , V = V 0 and M = M 0 are the total
stress resultants (the axial force, the shear force and the bending moment, respectively), p is the pore
pressure and b is the Biot coefficient. We assume that p is positive in compression. Following Terzaghi’s
principle of effective stresses, the superscript (0 ) denotes effective stress resultants.
The strong form of continuity equation exploits Darcy’s law for pore fluid flow, and is written as
1 k 
ṗ + b∇ · u̇ − ∇ · ∇p = 0 (2)
M γf

where M is Biot modulus, u is the displacement of the solid phase, k is the coefficient of permeability of
isotropic porous media, and γ f is the specific weight of the pore fluid.

E XTERNAL FLUID MODEL


For numerical representation of the external fluid motion we choose mixed displacement/pressure
finite element formulation. We assume that the motion of the external fluid remains small and irrotational,
with inertial effects being negligible due to quasi-static simplification. The strong form of governing
equations is written as [6, 7]
p Λ
∇p + ∇ × Λ − fb = 0; ∇ · u + = 0; ∇ × u − = 0 (3)
β ϑ
where p is the pressure, u is the displacement vector, Λ is the ’vorticity moment’ and ϑ is the penalty
parameter. The parameter β is the bulk modulus of the external fluid, and fb is the external load vector.
For numerical discretization of the external fluid domain, we choose Q4-P4-Λ4 finite element. The
’vorticity moment’ degree of freedom can be statically condensed on the element level, so the only
unknown variables remaining on the global level are displacements and pressures.

N UMERICAL RESULTS
In this section, we simulate the response of a nonlinear dam-reservoir system with the aim of quanti-
fying the overall safety of the dam structure against localized failure. The geometry of the dam-reservoir
system and the loading program are shown in Figures 1a and 1b.
1.0 m

pervious
A
p=0
Specific weight [MPa]
ious

Imposed
perv

displacements
6.0 m
5.1 m

γs

y γw

x
0 7 14 21 28
Time [days]
3.0 m 8.0 m

(a) Geometry of dam-reservoir system (b) Loading program

Figure 1: Dam-reservoir system

We compute the maximum overload the dam structure can withstand, on top of already acting self-
weight and hydrostatic loading. The computation of maximum overloads for both horizontal and vertical
direction is carried out by imposing corresponding displacements at the top base of the dam. The specific
weight of the water is γw = 0.01 MPa, and the specific weight of the dam material is γs = 0.02 MPa.
The mechanical properties of Timoshenko beam finite element are: Young’s modulus E = 10000 MPa,

172 2
the yield and fracture limits in tension, compression and shear σy,t = 0.015 MPa; σy,c = 0.20 MPa;
σy,s = 0.015 MPa; σ f ,t = 0.02 MPa; σ f ,c = 0.30 MPa; σ f ,s = 0.02 MPa. The coefficient of permeability
of dam material is k = 10−7 m/s, and the bulk modulus of external fluid is β = 1000 MPa. All numerical
computations are performed with a research version of the computer code FEAP, developed by R. L.
Taylor [8].
Pressure [kPa]
0.00
0.000e+00 12.75
0.0127 25.50
0.0255 38.25
0.0382 51.00
5.100e-02

1
Numerical model
0.8 Analytical solution

y/Hreservoir
0.6

0.4

0.2

0
0 10 20 30 40 50 60
Hydrostatic pressure kPa]

(a) Pressure and pore pressure distribution (b) Hydrostatic pressure distribution

Figure 2: Computed pressure fields due to self-weight and hydrostatic loading

We first assume linear elastic response of dam structure, and perform the computation for the first two
loading phases indicated in Figure 1b. The computed pressure field in the reservoir and in the body of the
dam due to the self-weight and hydrostatic loading are shown in Figure 2a. The computed pressure field
in the reservoir is practically identical with the analytical solution for the hydrostatic pressure distribution
(Figure 2b).

9
Horizontal overload [kN]

0
-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0
Horizontal displacement [mm]
(b) Deformed (c) Cracks formed and (d) Cracks formed and
(a) Horizontal overload configuration propagating in mode I propagating in mode II

Figure 3: Horizontal overload: computed results

140
Vertical overload [kN]

120
100
80
60
40
20
0
-1.4 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0
Vertical displacement [mm]
(b) Deformed (c) Cracks formed and (d) Cracks formed and
(a) Vertical overload configuration propagating in mode I propagating in mode II

Figure 4: Vertical overload: computed results

173 3
Next, we assume nonlinear behavior of dam material, and perform computations following the load-
ing program shown in Figure 1b, with the aim to determine the maximum overload dam can withstand
until ultimate failure. The computed admissible horizontal overload is 8.3810 kN (Figure 3a). The to-
tal horizontal reaction resulting from the self-weight and hydrostatic loading is 130.05 kN. Thus, we
can conclude that the computed factor of safety against failure is 1.06. The failure mode for horizontal
overload is shown in Figures 3b-3d. The computed admissible vertical overload is 139.2103 kN (Figure
4a). The total vertical reaction resulting from the self-weight and hydrostatic loading is 252.7516 kN.
Thus, we can state that the computed factor of safety against failure is 1.55. The failure mode for vertical
overload is shown in Figures 4b-4d.

C ONCLUDING REMARKS
In this paper, we have presented an efficient approach for dealing with acoustic fluid-structure in-
teraction problems in quasi-static setting. The structure is considered as saturated porous media whose
response is obtained with nonlinear coupled discrete beam lattice model. The external fluid motion is
described with mixed displacement/pressure based formulation. The chosen finite element representa-
tions of the structure response and the external fluid motion feature the same displacement and pressure
degrees of freedom. This allows for the structure and the external fluid finite elements to be connected
directly at the common nodes ensuring the exchange of both pressure and motion at the fluid-structure
interface, without any special numerical treatment of the common boundary. The results of numerical
simulations show that with the proposed model we are able to capture localized failure in the pore-
saturated structure for a particular loading program, and to quantify overall safety of the pore-saturated
structure against failure with respect to fundamental load cases, self-weight and hydrostatic loading. The
proposed numerical model of interaction shows excellent performance in terms of fast convergence rates
and efficient computational time.

ACKNOWLEDGMENTS
This work was supported by the French Ministry of Foreign Affairs, and French Embassy in Bosnia
and Herzegovina. Professor Adnan Ibrahimbegovic was supported by the funding for Chaire de Mé-
canique Picardie (120-2015 RDISTRUCT-000010 and RDISTRUCT-000010), EU funding (FEDER) and
IUF-Institut Universitaire de France (Membre Senior). These grants and financial supports are gratefully
acknowledged.

References
[1] A.Ibrahimbegovic. Nonlinear solid mechanics: theoretical formulations and finite element solution methods,
Springer Science & Business Media, 2009.
[2] M. Nikolic, A.Ibrahimbegovic, P. Miscevic. Brittle and ductile failure of rocks: embedded discontinuity ap-
proach for representing mode I and mode II failure mechanisms, International Journal for Numerical Methods
in Engineering, 102(8):1507–1526, 2015.
[3] E.Hadzalic, A.Ibrahimbegovic, S.Dolarevic. Failure mechanisms in coupled soil-foundation systems, Coupled
Systems Mechanics, 7(1):27–42, 2018.
[4] M. Nikolic, A.Ibrahimbegovic, P. Miscevic. Discrete element model for the analysis of fluid-saturated frac-
tured poro-plastic medium based on sharp crack representation with embedded strong discontinuities, Com-
puter Methods in Applied Mechanics and Engineering, 298:407–427, 2016.
[5] E.Hadzalic, A.Ibrahimbegovic, M.Nikolic. Failure mechanisms in coupled poroplastic medium, Coupled Sys-
tems Mechanics, 7(1):43–59, 2018.
[6] KJ.Bathe, C.Nitikitpaiboon, X.Wang. A mixed displacement-based finite element formulation for acoustic
fluid-structure interaction, Computers & Structures, 56(2-3):225–237, 1995.
[7] X.Wang, KJ.Bathe. Displacement/pressure based mixed finite element formulations for acoustic fluid-structure
interaction problems, International journal for numerical methods in engineering, 40(11):2001–2017, 1997.
[8] O.C. Zienkiewicz, R.L. Taylor. The Finite Element Method, vols. I, II, II, Elsevier, 2005.

174 4
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

M ULTI - SURFACE PLASTICITY MODEL WITH SOFTENING FOR SOLIDS WITH MARKED
DIFFERENCE OF FAILURE

Emir Karavelić 1,2 , Adnan Ibrahimbegovic 1


1 University of Sarajevo, Faculty of Civil Engineering, 71000 Sarajevo, Bosnia and Herzegovina,
emir.karavelic@utc.fr
2 Université de Technologie de Compiègne/Sorbonne Universités, Laboratoire Roberval de Mécanique,

Centre de Recherche Royallieu, 60200 Compiègne, France, adnan.ibrahimbegovic@utc.fr

1. Introduction
In this work we develop plasticity model much inspired by bahaviour of concrete. We present the con-
crete model built in the view of failure models for massive structures, where the elastic behavior is
followed by the creation of the fracture process zone with a large number of micro-cracks and subse-
quent final failure mode in terms of the macro-cracks. The fracture process zone is represented by the
non-associated Drucker-Prager model since the number of micro-cracks is considered sufficiently large
and their orientation random. The macro-crack is represented with a surface of displacement discontinu-
ity on which are concentrated all localized dissipative mechanisms due to the apparition and development
of localization zones [1].

2. Multi-surface model
The constitutive model governed by Hooke’s law, in the presence of plastic deformation, is no longer
pertaind to the total, but the elastic deformation. Namely, by assuming the independence of the elastic
response on plastic flow, the total deformation can be split additively into elastic εe and plastic part ε p ,

ε = ε e +εε p (1)

By further assuming that the elastic response remains linear, reducing to Hooke’s law in the absence of
plastic deformation, we can construct the free energy potential as a quadratic form in terms of deforma-
tions
ψ(εε,εε p ,ζζ) := 21 (εε −εε p ) · C(εε −εε p )) + 12 ζKζ (2)
Besides the stress tensor σ we define the stress-like internal variable q, which is dual to the strain-
like internal state variable ζ. We assume that this dual variable is used to define the yield criterion which
corresponds to the classical Drucker-Prager model:
r
1 2 (3)
σ, q) :=k dev[σ
φy (σ σ] k + tan(ϕ)tr[σσ]) − (σy − q)
3 3
where
σ] = σ − 13 (tr[σ
dev[σ σ])1 (4)
is the deviatoric part of the stress tensor, tan(ϕ) is material parameter which can characterize the internal

friction and σy is uni-axial yield stress. In (10) the Frobenius norm is used with k σ k:= σ : σ. Instead
to the yield function, plastic flow develops along the normal to the plastic potential function resulting

175 1
with nonassociated Drucker-Prager model. Here, we will use plastic potential function similary to yield
function
φ p (σ σ] k + 13 tan(ψ)tr[σ
σ, q) :=k dev[σ σ] (5)
where tan(ψ) is material parameter describing the angle of dilatancy.
Given three fundamental equations in (1), (2) and (3) with (5), are sufficient to completely define
the stress tensor computation as well as the internal variables evolution corresponding to the plasticity
model, simply by using the the second principle of thermodynamics and the principle of maximum plas-
tic dissipation [2]. It can be shown that the corresponding value of Lagrange multiplier γ̇ for associative
plasticity model is obtained from the consistency condition, which assures that in a plastic loading pro-
cess, subsequent stress and deformaton states remains on subsequent yield surface. Consequences of the
principle of maximum plastic dissipation that characterize such associative plasticity model are the con-
vexity of the yield surface in stress space and normality of plastic flow with respect to the yield surface.
Drucker’s stability postulate is also in agreement with this principle thus we can conclude that proposed
model with Drucker-Prager nonassociative flow rule is not stable in sense of Drucker. However, stability
postulate is sufficient but not a necessary criterion. Since the uniqueness of stress and strain trajectories
for a given loading exists, the material can be regarded as locally stable, thus the condition of uniqueness
rather than the stability postulate may be regarded as a basic for establishing stress-strain relationship.
For non-associative flow rule, the plastic multiplier γ̇ is obtained by using the plastic consistency
condition φ˙y = 0, and can be expressed as

∂φy
γ̇ = h−1 · Cε̇ε (6)
σ
∂σ
where we used relation
∂φy ∂φp ∂φy ∂φy
h−1 = ·C + ·K (7)
σ
∂σ σ
∂σ ∂q ∂q
We can simplify corresponding results in (6) for proposed model to obtain:

2Gνν : ε̇ε + 3Kα1 tr(ε̇ε)


γ̇ = (8)
2G + 9Kα1 α2 + 32 Kh,iso

Once the plastic multiplier γ̇ is determined the corresponding stress increment can be obtained and
the elastic-plastic constitutive

Cep = C + C p (9)

where C p is plastic tangent stiffness tensor and represents the degradation of the stiffness of material
due to plastic flow. It can be noted from (33) that tensor C p lacks symmetry, and so does Cep if a
nonassociative flow rule is used which we stated for Drucker-Prager model:

(2Gννn+1 + 3Kα1 1) ⊗ (2Gννn+1 + 3Kα2 1)


Cep = C − (10)
2G + 9Kα1 α2 + 23 Kh,iso

The stress update and formation of the elastoplastic tangent modulus requires the derivative of the
yield and of the plastic potential function. Thus, even for linear criteria this is a cumbersome task when
carried out in the general six-dimensional stress space. It will be shown in the following that computation
in principal stress space simplifies the presented procedure above, remarkably. Firstly, the dimension of
the problem reduces from six to three, and secondly, in the three-dimensional stress space the stress
states can be visualised graphically, making it possible to apply geometric arguments. The approach is
applicable for general isotropic yield criteria, but in the following only Drucker-Prager criteria will be
considered. As only isotropic material models are considered the manipulations can be carried out with
respect to any set of coordinate axes. Therefore the predictor stress is transformed into principal stress
space and returned to the yield surface. Considering the fact that the stress return preserves the principal
directions, the updated stress can then be transformed back into the original coordinate system. The

176 2
constitutive matrices are also formed in principal stress space and then subsequently transformed. All
transformations rely on standard coordinate transformation.
Drucker-Prager model in principal stress space is right-circular cone with its axis equally inclined
with the respect to each of the coordinate axes and its apex in the tension octant. It can be shown that
plastic deformation must be accompanied by an increase in volume if ϕ 6= 0. This property known as
dilatancy is consequences of the dependency of yield surface for associative or plastic potential surface
for nonassociative flow rule, on hydrostatic pressure. Figure 1 shows the Drucker-Prager yield surface
open in direction of the negative hydrostatic axis along with plastic potential function. From flow rule the
plastic strain increment dεipj is perpendicular to plastic potential surface at the actual yield point M. The
vector dεipj can be decomposed into vertical and horizontal component where horizontal components
dεipvj represents the plastic volume change, which is always positive for ψ > 0 so and for associative
plasticity. When using a negative dilatancy angle, we find that stress-strain curve gradually approaches
a line with a negative slope. In other words, hardening is followed by softening and during this unstable
behaviour the material resistance vanishes completely.
J1/2
2

d pv
ij
Current loading surface
p
d ij
Plastic potential surface

Subsequent loading surface

-I1

Figure 1: The loading and plastic potential surfaces for the Drucker-Prager material with a nonassociated
flow rule

Invoking three main ingredients sufficent for the constitutive model of plasticity we can provide the
stress tensor computation as well as internal variables evolution. The principal stresses and deformations
as well as their directions are found by solving the well-known eigenvalue problem. The three eigen-
vectors of deformation tensor form a coordinate transformation tensor, thus forming the transformation
matrix T . All the remaining ingredients of the plasticity model can be obtained from the standard ther-
modynamics considerations. After the computation of plastic multiplier and updating stress tensor we
also have to obtain the elastoplastic tangent modulus which constist of a material and of a geometric part
[3].

σ 3 3
∂σ ∂σi ∂
C= =∑ ni ⊗ ni + ∑ σi (ni ⊗ ni )
∂εε i=1 ∂εε ε
i=1 ∂ε (11)
| {z } | {z }
Cmat Cgeo

In order to take into account two types of dissipative mechanism, for compression and tension one, we
build multisurface model in order to better reproduce the behavior of massive structures: a bulk dissipa-
tion characterized by the development of micro-cracks, which is taken into account by introduction of
the Drucker-Prager model and a surface dissipation taking place at the level of the localization zones in
terms of the macro-cracks triggered with plasticity criterion in strain space defined by three surfaces [4].

φ1 (εεe ) = εe1 − (εy − q) ≤ 0


φ2 (εεe ) = εe2 − (εy − q) ≤ 0 (12)
e
φ3 (εε ) = εe3 − (εy − q) ≤0

Figure 2a shows proposed criterion in terms of principal stresses. One can note that in tension region
elastic stage is followed by softening, while in compression the fracture process zone is defined, followed
by softening.
Figure 2b cleary reveals that the only loading case which is not limited is the three-axial compression
along the three axes such as produced by hydrostatic pressure. Figures 3 and 4 shows stress-strain curve
for performed numerical tests.

177 3
s
2
1/2
J2
s 3

s 1

Drucker-Prager

-I1
hydrostatic
axis

(a) (b)
Figure 2: Multisurface criterion (a) in principal stress space (b) in meridian plane

Isochoric-simple shear test


6

4
Shear stress [MPa]

=0
1
=20
=-10

(a) (b)
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Shear strain [%]

Figure 3: Computed responses of concrete in isochoric shear tests


5 0
Linear hardening
4.5 -5 Saturation hardening

4
-10

3.5
-15
Stress [MPa]

Stress [MPa]

3
-20
2.5
-25
2
-30
1.5

-35
1

0.5 -40

(a) (b)
0 -45
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 -0.4 -0.35 -0.3 -0.25 -0.2 -0.15 -0.1 -0.05 0
Axial Strain [%] Axial Strain [%]

Figure 4: Macroscopic response: (a) uniaxial tension test; (b) uniaxial compression test

3. Conclusion
It has been shown that proposed model combining the inelastic hardening and inelastic softening is able
to explain the failure of a massive structure. The class of problem of main interest for this kind of model
pertains to the failure of massive structures where final failure mechanism is preceeded by significant
development of plastic zone and where the contribution of the so-called fracture process zone remains of
equal importance for total plastic dissipation as the actual failure mechanism itself.
Acknowledgements: This work was supported by a scholarship from the French Ministry of For-
eign Affairs and French Embassy in Bosnia and Herzegovina.This support is gratefully acknowledged.
Adnan Ibrahimbegovic was supported by the funding for Chaire-de-Mecanique and IUF. This support is
gratefully acknowledged.

References
[1] Karavelic E, Nikolic M, Ibrahimbegovic A (2017) Concrete meso-scale model with full set of 3D failure modes
with random distribution of aggregate and cement phase. Part I: Formulation and numerical implementation.
Comput. Methods Appl. Mech. Engrg., doi:10.1016/j.cma.2017.09.013
[2] Ibrahimbegovic A (2009) Nonlinear Solid Mechanics: Theoretical Formulations and Finite Element Solution
Methods. Springer, London
[3] Ibrahimbegovic A, Gharzeddine F, (1999) Finite deformation plasticity in principal axes: from a manifold to
the euclidean setting. Comput. Methods Appl. Mech. Engrg. 171,341-369
[4] Ibrahimbegovic A, Brancherie D, (2003) Combined hardening and softening constitutive model of plastic-
ity,precursor to shear slip line failure. Computational Mechanics,31,88-100, doi:10.1007/s00466-002-0396-x

178 4
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

COLLAPSE OF PARABOLICALLY TAPERED CHS STEEL COLUMNS:


FINITE ELEMENT MODEL VALIDATION

Zampaolo, T. C.1, Massaroppi Jr., E.2, Abambres, M.3, Ribeiro, T. P.4


1
Department of Mechanical Engineering, University of São Paulo, Brazil, tzampaolo@gmail.com
2
Department of Mechanical Engineering, University of São Paulo, Brazil, massarop@sc.usp.br
3
Abambres’ Lab, Portugal, abambres@netcabo.pt
4
Tal Projecto, Portugal, tpribeiro@gmail.com

Abstract: Tapered steel beams and columns are increasingly being used as primary load carrying
structures. To design and determine the load carrying capacity of these structures numerical modeling
methods, such as the finite element method, have been widely employed in the past few years. This
paper presents a systematic study on the influence of the finite element model on the ultimate bearing
capacity and postbuckling behavior of tapered steel columns with circular hollow section. The aim is
to investigate the influence of model simplifications, mesh density and quality, element types and
boundary conditions.

Keywords: Finite element model validation; mesh quality; nonlinear buckling; tapered column.

1. Introduction and Motivation

The considerable stiffness, strength and ductility of metals, as well as recent advances in
manufacturing technology, have led to the construction of highly efficient structural metal systems
(large strength-to-weight ratios) – in particular, those composed by thin-walled members have shown
to be a quite economic solution featuring reduced transportation and construction times (Loughlan
2004). Non-prismatic (tapered) members are widely used in modern metal construction in Civil,
Mechanical and Aeronautical industries, mostly due to their (i) structural efficiency, (ii) functionality,
and (iii) low fabrication costs (Zhang and Tong 2008) - Fig. 1 shows some typical applications of long
tapered steel columns in airport structures.

(a) (b)
Fig. 1. Tapered steel columns in (a) Barajas (Madrid) and (b) Beijing international airport.

In order to take advantage of those benefits, accurate, simple and efficient design methods
must be available. Nevertheless, it is well-known (Marques et al. 2012) that safety verifications in

179
steel standards (e.g., CEN 2005, 2006, AISC2010, SA 2016), most mostly
ly adapted from prismatic member
rules, might be unsafe (up to 300% in some cases - Bedynek et al. 2013), difficult to perform, and/or
quite conservative (not taking advantage of the economy of non non-prismatic
prismatic members). A commonly
adopted alternative to thosee methods, as recommended by design codes, is the use of advanced
(physically and geometrically nonlinear) FEA, which is obviously unfeasible in current design practice
due to their time and know-how how requirements (besides involving expensive FEA software). Although
the large amount of research performed in the last few decades, either concerning (i)
numerical/analytical
/analytical formulations (e.g., Ghadban et al. 2017, Kim and Jang 2017, Balduzzi et al. 2017,
Lee and Lee 2018), or (ii) design methods (e.g., Marques et al. 2012, Papp 2016), it is still imperative
the development of groundbreaking (i.e., simultaneously accurate, easy-to-use,
easy use, versatile, efficient and
affordable) design rules/tools for tapered steel members. Within this context, the work presented
herein is part of an ongoing investigation that aims to propose (until the end of the year) an Artificial
Neural Network (ANN)-based based design scheme for simply supported circular hollow section (CHS)
tapered steel columns, exhibiting a parabolic taper and typicall
typically used (i) in high-rise
rise building (high-
(high
ceiling) lobbies, and (ii) as support of large span roof structures (see Fig. 2). For that purpose, the first
step consists in the performance of an extensive parametric finite element analysis (FEA) for the
computationon of (i) elastic buckling and (ii) ultimate bearing capacity loads.

Fig. 2.. Generic tapered CHS column analyzed in this work.

The parametric analysis (PA) involves nine independent (input) variables, as defined in Tab. 1.
All combinations of input variable values were taken for the PA, resulting in a total of 58320 distinct
columns to be simulated.

Tab. 1.. Variables/values involved in the FE


FE-based parametric analysis.

180
2. Main Objectives

The accuracy of the finite element simulations is highly dependent on the model adopted (i.e.
the simplifications adopted to the geometry to be analyzed), boundary conditions and the mesh quality
and density. Thus, care has to be taken during this initial stage of the FE simulations in order to
achieve accurate and reliable results. This paper addresses details and important conclusions about the
modelling, simplifications and mesh validation procedures carried out before the onset of the PA, all
performed using the FE package ANSYS Mechanical APDL (Ansys Inc 2014).

3. Future research

In Civil Engineering, ANNs have provided a convenient and often highly accurate solution to
problems within all branches, appearing from publications statistics to be one of the great successes of
computing (Flood 2008). The first journal article on civil engineering applications of neural networks
was published by Adeli and Yeh (1989). The authors used a simple one-neuron model to the design of
steel beams. Since then, many other applications of ANNs within all fields of Civil Engineering arose
with increased complexity and sophistication (Adeli 2001). Areas like (i) buckling load prediction
(e.g., Sharifi and Tohidi 2014), (ii) bearing capacity prediction (e.g., Gandomi et al. 2013), (iii)
constitutive modeling (e.g., Oeser and Freitag 2016), (iv) structural reliability and/or optimization
(e.g., Papadrakakis and Lagaros 2016), or (v) structural health monitoring (e.g., Min et al. 2012), have
received special focus until today. Many successful ANN-based models have been proposed to assess
the behavior of metals and structures, when composed by prismatic members (e.g., Guzelbey et al.
2006, Efstathiadesa et al. 2007, Lu et al. 2009, Sheidaii and Bahraminejad 2012, Xu et al. 2013,
Tohidi and Sharifi 2015, 2016, Nazari et al. 2015, Banu and Rani 2016). Several works have revealed
a huge decrease in computing time when comparing the proposed ANN model with the FEA
counterpart, and without compromising accuracy – e.g, when estimating the temperature of a tubular
truss under fire, Xu et al. (2013) concluded that the ANN computes the desired output 1800 times
faster than FEA. Surprisingly, unlike for prismatic members, virtually no effort has been done to
develop analysis and design methods for tapered metal members based on ANNs.

References

Adeli H (2001). Neural networks in civil engineering: 1989–2000, Computer‐Aided Civil and
Infrastructure Engineering, 16(2), 126–142.
Adeli H, Yeh C (1989). Perceptron Learning in Engineering Design, Computer-Aided Civil and
Infrastructure Engineering, 4(4), 247-256.
AISC – American Institute of Steel Construction (2010). Specification for Structural Steel Buildings
(ANSI/AISC 360-10), AISC, Chicago, USA
Ansys Inc. (2014). ANSYS APDL (Mechanical), release 15.0, Canonsburg, PA.
Balduzzi G, Morganti S, Auricchio F, Reali A (2017). Non-prismatic Timoshenko-like beam model:
Numerical solution via isogeometric collocation, In Computers & Mathematics with Applications,
74(7), pp. 1531-1541.
Banu PSN, Rani SD (2016). Knowledge-based artificial neural network model to predict the properties
of alpha+ beta titanium alloys, Journal of Mechanical Science and Technology 30 (8), 3625-3631.
Bedynek A, Real E, Mirambell E (2013). Tapered plate girders under shear: Tests and numerical
research, 46(January), pp. 350-58
CEN - Comité Européen de Normalisation (2005). Eurocode 3: Design of steel structures. Part 1-1:
General rules and rules for buildings (EN 1993-1-1), Brussels
CEN - Comité Européen de Normalisation (2006). Eurocode 3. Design of steel structures. Part 1–5:
Plated structural elements (EN 1993-1-5), Brussels

181
Efstathiadesa C, Baniotopoulosa CC, Nazarkob P, Ziemianskib L, Stavroulakisc GE (2007).
Application of neural networks for the structural health monitoring in curtain-wall systems,
Engineering Structures, 29(12), 3475-84.
Flood I (2008). Towards the next generation of artificial neural networks for civil engineering,
Advanced Engineering Informatics, 22(1), 4–14.
Gandomi AH, Yun GJ, Alavi AH (2013). An evolutionary approach for modeling of shear strength of
RC deep beams, Materials and Structures, 46(12), 2109–2119.
Ghadban AA, Al-Rahmani AH, Rasheed HA, Albahttiti MT (2017). Buckling of Nonprismatic
Column on Varying Elastic Foundation with Arbitrary Boundary Conditions, Mathematical problems
in engineering, Volume 2017, 1-14.
Guzelbey IH, Cevikb A, Gogus MT (2006). Prediction of rotation capacity of wide flange beams using
neural networks, Journal of Constructional Steel Research, 62(10), pp. 950–61
Kim H, Jang G-W (2017). Higher-order thin-walled beam analysis for axially varying generally
shaped cross sections with straight cross-section edges, Computers & Structures, 189(September), 83-
100.
Lee J, Lee B (2018). Elastica and buckling loads of nonlinear elastic tapered cantilever columns,
Engineering Solid Mechanics, 6(1), 39-50.
Loughlan J (Ed.) (2004), Thin-Walled Structures − Advances in Research, Design and Manufacturing
Technology, Institute of Physics Publishing, Bristol.
Lu Y, Ye L, Su Z, Zhou L, Cheng L (2009). Artificial Neural Network (ANN)-based Crack
Identification in Aluminum Plates with Lamb Wave Signals, Journal of Intelligent Material Systems
and Structures, 20(1), 39-49.
Marques L, Taras A, Silva LS, Greiner R, Rebelo C (2012). Development of a consistent buckling
design procedure for tapered columns, Journal of Constructional Steel Research, 72(May), 61-74.
Min J, Park S, Yun C-B, Lee C-G (2012). Impedance-based structural health monitoring incorporating
neural network technique for identification of damage type and severity, Engineering Structures,
39(June), 210–220.
Nazari A, Rajeev P, Sanjayan JG (2015), Offshore pipeline performance evaluation by different
artificial neural networks approaches, Measurement, 76 (December), pp. 117-28
Oeser M, Freitag S (2016). Fractional derivatives and recurrent neural networks in rheological
modelling – part I: Theory, International Journal of Pavement Engineering, 17(2), 87–102.
Papadrakakis M, Lagaros ND (2016) Reliability-based structural optimization using neural networks
and Monte Carlo simulation, Computer Methods in Applied Mechanics and Engineering, 191(32),
3491–3507.
Papp F (2016). Buckling assessment of steel members through overall imperfection method,
Engineering Structures, 106 (January), pp. 124-36
SA - Standards Australia (2016). Australian standard: Steel Structures, AS 4100—1998 (R2016),
Sydney.
Sharifi Y, Tohidi S (2014). Lateral-torsional buckling capacity assessment of web opening steel
girders by artificial neural networks — elastic investigation, Frontiers of Structural and Civil
Engineering, 8(2), 167–177.
Sheidaii MR, Bahraminejad R (2012). Evaluation of compression member buckling and post-buckling
behaviour using artificial neural network, Journal of Constructional Steel Research, 70(March), 71–77.
Tohidi S, Sharifi Y (2015). Neural networks for inelastic distortional buckling capacity assessment of
steel I-beams, Thin-Walled Structures, 94 (September), pp. 359-371
Tohidi S, Sharifi Y (2016). Load-carrying capacity of locally corroded steel plate girder ends using
artificial neural network, Thin-Walled Structures, 100 (March), pp. 48-61
Xu J, Zhao J, Wang W, Liu M (2013). Prediction of temperature of tubular truss under fire using
artificial neural networks, Fire Safety Journal, 56 (February), pp. 74-80
Zhang L, Tong GS (2008). Lateral buckling of web-tapered I-beams: A new theory, Journal of
Constructional Steel Research, 64(12), 1379-93.

182
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

Repeated Richardson extrapolation with verification of the order of


accuracy of numerical schemes and the forms of applying
boundary conditions of the finite volume method on the Laplace
equation

Fabiana de Fátima Giacomini1, Ana Paula da Silveira Vargas2


1
Technological Federal University of Parana, fgiacomini@utfpr.edu.br
2
Technological Federal University of Parana, anavargas@utfpr.edu.br

Reliable accuracy and predictions of transport phenomena are the main purpose in Computational Fluid
Dynamics (CFD). The final intent of scientific interest is the validation of a mathematical model and for
this numerical verification is necessary. Two verification steps are: the computational code verification
and the numerical solution verification. Therefore, the objective of this work is to verify the effect on
the discretization error and its order caused by the form to apply the boundary conditions in problems
solved with the finite volume method. For this, they are considered: one and two-dimensional Laplace
equation with the verification of the order of accuracy a priori and a posteriori; uniform grids; up to
seven primary and secondary interest variables with numerical approximations of first and second orders
of accuracy; Dirichlet boundary conditions; tri and pentadiagonal solvers; grids with up to millions of
nodes; quadruple precision; and number of iterations suficiente to achieve the machine rounding error.
The form of apply the boundary conditions considered are four: with and without ghost volume; half
volume; and zero thickness volume. The a priori estimate of the order of accuracy is performed by
means of the deduction of the formal order based on the Taylor series. This approach is used to
effectively verify the order of accuracy of the numerical solutions calculated a posteriori by the use of
hibrid techniques, with numerical approximations of different orders, or not. The a posteriori estimate
is made based on the estimation of the discretization error obtained through Repeated Richardson
Extrapolation (RRE) [1]. As results obtained it is expected that: i) if the mixing fator exist, it does not
change the effectiveness of RRE [2]; ii) the accuracy order obtained a posteriori by means of RRE
corroborates the formal order obtained a priori; iii) RRE provides subsidies for cases in which it is not
possible to estimate a priori or a posteriori the orders of the numerical error; iv) the form to apply
boundary conditions with half volume results in the smallest numerical error [3]. Additional topics
include deduction of true orders from the numerical scheme, apparent order analysis, numerical
parameter effect on hibrid scheme, if any, and parabolic equation with or without source term.

References

[1] C. H. Marchi, L. A. Novak, C. D. Santiago, A. P. da S. Vargas. Highly accurate numerical solutions


with repetead Richardson extrapolation for 2D laplace equation, Applied Mathematical Modelling 37,
7386-7397, 2013.

183
[2] A. P. da S. Vargas. Repeated Richardson extrapolation and schemes of 1st and 2nd order, mixed and
Crank-Nicolson on the advection-difusion and Fourier 2D equations, Doctoral Thesis, PGMEC-UFPR,
2013.
[3] F. F. Giacomini. Verification of the form to apply boundary conditions in one-dimensional problems
with the finite volume method, Master’s Dissertation, PGMEC-UFPR, 2009.

184
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

GLOBAL CONTACT ANALYSES BETWEEN PIPES THROUGH


DYNAMIC SIMULATIONS USING FRAME FINITE-ELEMENT MODELS
Tiago Sten Freitas1, Fabrício Nogueira Corrêa2, Breno Pinheiro Jacob3
1
Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia - COPPE/UFRJ,
tiagosten@lamcso.coppe.ufrj.br
2
Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia - COPPE/UFRJ,
fabricio@lamcso.coppe.ufrj.br
3
Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia - COPPE/UFRJ,
breno@lamcso.coppe.ufrj.br

ABSTRACT

O termo problemas de contato é empregado para as situações gerais onde ocorrem interações entre
um ou mais corpos. Os problemas de contato são frequentes em mecânica dos sólidos, visto que
as ações de contato são o mecanismo mais comum para as transferências de forças entre estruturas
ou sólidos em geral.
Problemas de contato são notadamente complexos quando não são conhecidos os pontos ou
regiões de contato, principalmente em problemas dinâmicos definidos no espaço tridimensional.
Para tal, deve-se conhecer a trajetória dos corpos no espaço e, por sua vez, a distância entre eles.
Assim que o contato é identificado, deve-se solucionar a equação de movimento dos corpos
considerando suas equações constitutivas, as condições de contorno e as condições iniciais que
definem o contato.
Este tipo de problema é notadamente não-linear por além de envolver não-linearidades
geométricas, envolvem não-linearidades físicas devido a vários fatores, tais como: grandes
deformações da seção dos corpos em contato, efeitos de atrito, etc.
Tradicionalmente, o problema de contato em modelos numéricos é baseado no uso de escalares
generalizados ligados por dois nós, um de cada corpo, definidos por molas não lineares com
patamar nulo para representar que os sólidos estão afastados e não nulos para representar forças
elásticas durante o contato.
O objetivo deste trabalho é apresentar um algoritmo mais aprimorado para identificação de
contato, baseado na interpenetrabilidade volumétrica, em análises globais de estruturas esbeltas
de seção circular, modeladas através de elementos finitos reticulados, de pórtico ou treliça
espacial. As forças elásticas de ação-reação são calculadas de forma expedita pela aplicação de
uma rigidez de contato, enquanto a dissipação de energia do contato é definida pelo
amortecimento estrutural de Rayleigh.
Este método é adequado para tratar o contato externo ou mesmo de uma ou mais estruturas
internas a outra (pipe-in-pipe).
Neste contexto, cabe recordar que as análises globais tem como objetivo avaliar, por exemplo, o
comportamento dinâmico das estruturas esbeltas, os modos de vibração, os esforços e curvaturas;
não sendo objeto da avaliação os efeitos mais localizados e complexos que envolveriam, por
exemplo, o uso de elementos finitos tridimensionais e equações constitutivas mais rigorosas para
estimativa das tensões e deformação da seção devido ao contato/impacto.
Deste modo, análises globais são aplicáveis em projetos de cabos de torres teleféricas, linhas de
ancoragem e risers de estruturas de exploração de petróleo offshore - que são dutos de escoamento

185
responsáveis pela ligação entre o poço e a unidade flutuante, aplicados nas operações de
perfuração, produção, exportação, injeção e completação. Em qualquer uma destes problemas, o
contato entre estruturas adjacentes pode ocorrer e ser mitigado globalmente com o método
apresentado neste trabalho.
Em sistemas de risers, por exemplo, onde grupos de risers são colocados lado a lado, o contato
externo entre eles (clashing) deve ser evitado e depende do afastamento dos slots, da força da
corrente marítima sobre cada estrutura, da diferença de peso e demais características físicas e
geométricas entre risers adjacentes, etc. A Figura 1 ilustra de forma esquemática a necessidade
de verificação de contato entre dois ou mais risers.

Figura 1 - Contato entre diferentes riser.

Ainda no cenário offshore, o contato externo entre estruturas pode ocorrer, por exemplo, em uma
situação de blackout de uma sonda de perfuração posicionada dinamicamente por propulsores. A
unidade de perfuração em blackout desconecta seu riser do poço, ficando em deriva livre pela
perda de funcionamento dos propulsores com o riser de perfuração pendurado. Nesta situação, ela
pode se aproximar de uma unidade de produção ancorada na mesma localidade, havendo risco da
colisão entre a coluna de perfuração e as linhas de ancoragem (Figura 2). Se isto acontecer, a linha
de ancoragem pode exceder seu limite de carga e romper, o riser de perfuração pode ser danificado
e a coluna de perfuração pode ainda se prender na linha de ancoragem, guiando a unidade de
perfuração em deriva para cima da unidade de produção.
Além disso, existem situações mais específicas onde o contato pode existir quando uma estrutura
está dentro de outra tubular (pipe-in-pipe). Na área de exploração de petróleo, operações de
descida de colunas de revestimento pelo interior da coluna de riser em operações de perfuração
de poços é um bom exemplo deste tipo de contato.
Em todos os casos citados, as análises globais envolvendo a modelagem das linhas por elementos
finitos reticulados devem ser capazes de simular o efeito do contato, contando com um algoritmo
de identificação dos pontos de contato e de aplicação das forças de ação-reação no modelo de
elementos finitos solucionado no domínio do tempo.

Figura 2 - Colisão entre riser de perfuração e uma linha de ancoragem.

186
Neste contexto, este trabalho mostra que o algoritmo desenvolvido para identificação de contato
é adequado para análises dinâmicas globais no domínio do tempo de estruturas esbeltas modelas
por elementos finitos reticulados de seção circular. A abordagem envolve a identificação continua
de contato através de um método analítico, evitando que usuário pré-defina possíveis regiões de
contato, escalares generalizados, ou se preocupe com o tamanho dos elementos finitos nestas
regiões. O desempenho computacional é garantido através do uso de bounding box para busca de
possíveis regiões de contato.
Dois modelos genéricos serão estudados apenas para apresentar a funcionalidade do método: um
simulando o contato externo de linhas; e outro, o contato pipe-in-pipe. A eficiência computacional
também será abordada.

Keywords: Global Analysis, Contact Algorithm, Finite Elements.

REFERENCES

[1] NETO, F, Q. Modelagem de problemas de contato-impacto empregando formulações


penalizadas do método dos elementos finitos. 2002. P. 156 Tese (Doutor em Ciências em
Engenharia Civil). Programa de Pós-Graduação e Pesquisa de Engenharia (COPPE) da
Universidade Federal do Rio de Janeiro – UFRJ, Rio de Janeiro, 2002.

[2] SILVA, D, M, L. Ferramentas computacionais para análise e projeto de Instalação de dutos


submarinos. 2009. P. 156 Tese (Doutor em Ciências em Engenharia Civil). Programa de Pós-
Graduação e Pesquisa de Engenharia (COPPE) da Universidade Federal do Rio de Janeiro –
UFRJ, Rio de Janeiro, 2009.

[3] AZEVEDO, C, T.; ROCHA, L, A, S. Projetos de Poços de Petróleo: geopressões e


assentamentos de colunas de revestimentos. 2. ed. Rio de Janeiro: Interciência, 2009.

[4] BELYTSCHKO, T.; FISH, J. Um Primeiro Curso em Elementos Finitos. Rio de Janeiro, RH,
Editora LTC, 2009. 256 p.

[5] FERNÁNDEZ, E. F.; JUNIOR, O. A. P.; PINHO, A. C. Dicionário do Petróleo em Língua


Portuguesa: exploração e produção de petróleo e gás uma colaboração Brasil, Portugal e Angola.
Rio de Janeiro: Lexikon: PUC-Rio 2009.

[6] “ISO 11960-2:2004 Petroleum and natural gas industries - Steel pipes for use as casing or
tubing for wells”. Geneva, Suíça, ISO – International Organization for Standardization, 2006.

[7] CORRÊA, F.N., Ferramentas Computacionais para Análise Acoplada de Sistemas Offshore,
Tese de D.Sc., COPPE/UFRJ, Rio de Janeiro, RJ, Brasil, 2008.

[8] CHAKRABARTI, S. K., Handbook of Offshore Engineering – 2vols., 1st Ed., Elsevier, 2008.

[9] WRIGGERS, P., Computational Contact Mechanics. Chichester, John Wiley & Sons, 2002.

[10] LAURSEN, T.A., Computational Contact and Impact Mechanics. Berlim, Springer, 2002.

[11] HUGHES, T.J.R., TAYLOR, R.L., SACKMAN J.L., CURNIER A., KANOKNUKULCHAI
W., “A Finite Element Method for a Class of Contact-Impact Problems”, Comput. Methods Appl.
Mech. Engrg., vol.8, pp. 249–276, 1976.

[12] GOLDSMITH, W., Impact. Mineola, Dover Publications, 2001.

[13] JONES, N., Structural Impact. Cambridge University Press, 1997.

187
[14] BELYTSCHKO, Ted, YEH, I.S., “The Splitting Pinball Method for Contact-Impact
Problems”, Comput. Methods Appl. Mech. Engrg., vol. 105, pp. 375–393, 1993.

[15] BELYTSCHKO, Ted, LIN, J.I., “A Three-Dimensional Impact-Penetration Algorithm with


Erosion”, Computer and Sructures, vol. 25, pp. 95–104, 1987.

[16] QUARANTA, F., Modelagem de Problemas de Contato-Impacto Empregando Formulações


Penalizadas do Método dos Elementos Finitos, Tese de D.Sc., COPPE/UFRJ, Rio de Janeiro, RJ,
Brasil, 2002.

[17] HUNĔK, I., “On Penalty Formulation for Contact-Impact Problems”, Computers &
Structures, vol.48, pp. 193–203, 1993.

[18] HALLQUIST, J.O., GOUDREAU, G.L., BENSON, D.J., “Sliding Interfaces with Contact-
Impact in Large-Scale Lagrangian Computations”, Comput. Methods Appl. Mech. Engrg., vol.
51, pp. 107–137, 1985.

[19] BATHE, K.-J., Finite Element Procedures, New Jersey, Prentice-Hall, 1996.

[20] ZIENKIEWICS, O. C., TAYLOR, R. L., The Finite Element Method, 6 ed, Singapore,
McGraw-Hill, 2005.

[21] OÑATE, E, ROJEK, J., “Combination of Discrete Element and Finite Element Methods for
Dynamic Analysis of Geomechanics Problems”, Comput. Methods Appl. Mech. Engrg., vol.193,
pp. 3087–3128, 2004.

[22] HUGHES, T. J. R., The Finite Element Method – Linear Static and Dynamic Finite Element
Analysis, New York, Dover Publications, 2000.
[23] COOK, R. D., MALKUS, D. S., PLESHA, M. E, WITT, R. J., Concepts and Applications
of Finite Element Analysis, 4 ed, United States, Jonh Wiley & Sons, 2002.

[24] O’ROURKE, J., Computational Geometry in C, Massachusetts, Cambridge University Press,


1997.

[25] BERG, M., KREVELD, M.V., OVERMARS, M., SCHWARZKOPF, O., Computational
Geometry – Algorithms and Applications, Springer, 2000.

[26] ZHONG, Z.-H., NILSSON, L., “Automatic Contact Searching Algorithm for Dynamic Finite
Element Analysis”, Computers & Structures, vol.52, pp. 187–197, 1994.

[27] BENSON, D.J., HALLQUIST, J.O., “A Single Surface Contact Algorithm for the Post-
Buckling Analysis of Shell Structures”, Comput. Methods Appl. Mech. Engrg., vol.78, pp. 141–
163, 1990.

[28] WRIGGERS, P., ZAVARISE, G., “On Contact Between Three-Dimensional Beams
Undergoing Large Deflections”, Comm. Numer. Methods Engrg., vol.13, pp.

[29] SILVA, D.M.L., CORRÊA, F.N., JACOB, B.P., “A Generalized Contact Model for the
Simulation of Complex Offshore Operations”, XXVII CILAMCE – Iberian Latin-American
Congress on Computational Methods in Engineering, September 3-6, Belém, Pará, Brazil, 2006.

[30] SILVA, D.M.L., CORRÊA, F.N., JACOB, B.P., “A Generalized Contact Model for Dynamic
Analysis of Floating Offshore Systems”, 25st OMAE – International Conference on Offshore
Mechanics and Arctic Engineering, June 4-9, Hamburg, Germany, 2006.

188
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

PARAMETRIC STUDY TO OPTIMIZE THE HYDRODYNAMIC EFFICIENCY OF A SUBMERGED


PRESSURE DIFFERENTIAL WAVE ENERGY CONVERTER

Raí Mariano Quintas1, Fabrício Nogueira Corrêa2, Carl Horst Albrecht3


1
Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia – COPPE/UFRJ,
raiquintas@lamcso.coppe.ufrj.br
2
Instituto Alberto Luiz Coimbra de Pós-Graduação e Pesquisa de Engenharia – COPPE/UFRJ,
fabricio@lamcso.coppe.ufrj.br
3
Escola Politécnica – Universidade Federal do Rio de Janeiro, carl@lamcso.coppe.ufrj.br

Faced with the threat posed by global warming and climate change to the future of the planet,
society has intensified the pressure to reduce greenhouse gas emissions. As a result, countries have
been mobilizing to reduce the use of fossil fuels, which is a major contributor to anthropogenic
greenhouse gas emissions.

At the same time, as the world population continues to grow, demand for energy will increase.
Thus, to ensure the supply of this energy, without counting on fossil fuels, it will be necessary to
invest in alternative energy sources.

Because of that, renewable energy has gained increasing importance in the energy production
industry. Among the sources of renewable energy, one that has a great potential for generation, but is
still not much explored, is the wave energy.

Wave energy has some advantages over other sources of renewable energy, such as a higher
energy density and its generation consistency, but the efficiency of the devices developed to convert
this energy into useful electrical energy is generally very low, which makes the production of energy
by waves still unpractical [1].

Therefore, this work aims to study and propose strategies to theoretically increase the
hydrodynamic efficiency for a wave energy conversion device.

For this purpose, it will be used a submerged pressure differential device with cylindrical
geometry. This device will be simulated numerically for the cases of regular and irregular sea to find
the value of the parameters that optimize the device’s efficiency.

References

[1] B. Drew, A. R. Plummer, M. N. Sahinkaya. A review of wave energy converter technology, Proc.
IMechE Vol. 223 Part A: J. Power and Energy, 887-902, 2009.

189
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

3D DUCTILE CRACK SIMULATION BASED ON H-ADAPTIVE METHODOLOGY

FT. YANG1, A. RASSINEUX, C. LABERGERE2, K. SAANOUNI


1
Laboratoire Roberval UMR CNRS/FRE 2012, Sorbonne Universités, Université de Technologie de
Compiègne, fangtao.yang@utc.fr
2
ICD/LASMIS, UMR STMR 6279, Université de Technologie de Troyes, 12 rue Marie Curie,
BP2060, F10010 Troyes, France

1. Introduction
The metal forming process is of great importance in the industry. The damage of ductile material as well
as the fracture occurring during metal forming process is a great concern during metal forming
processes. The 2D simulation of ductile damage [1] [2] is widely studied in the literature, however 3D
simulation of ductile damage, especially the representation of the cracks based on h-adaptive
methodology is still an opening problem to be discussed.

2. Damage model
The elasto-plastic model proposed by Hooputra et al. [3] is used to describe the inelastic behavior of the
ductile material. The equivalent plastic strain is given by Eq.1.
 T sinh  k0        T sinh  k0     
 Dpl  ,  pl   (1)
sinh  k0      

Where  T and  T correspond to the equivalent plastic strain at ductile damage initiation for equibiaxial
tensile and equibiaxial compressive deformation, respectively. For isotropic materials the stress
triaxiality in equibiaxial tensile deformation state,   , is 2/3 , and in equibiaxial compressive
deformation state,   , is -2/3. The stress triaxiality  is defined as    p / q where p is the pressure
stress, q is the Mises equivalent stress and  pl is the equivalent plastic strain rate. The damage variable
D is defined in Eq. 2 which increases monotonically with plastic deformation.
d pl
D   pl (2)
 D  ,  pl 

3. 3D h-adaptive methodology
3.1 3D mesh refinement strategy
During the metal forming processes, with the plastic strain accumulating, the damage increases which
results in a highly localized physical field. These two physical quantities are chosen as indicators to
make a referenced size map. The mesh is then refined by bi-section technique [4]. This technique given
by a minimum size ratio 2 between old coarse mesh and new fine mesh. This requires field transfer
operator to capture the evolution of the physical field in a reliable way.

3.2 Element deletion and volume conservation

190
The cracks are represented by deleting totally damaged elements. The damage value ranges from 0 to 1.
We define a critical damage value as Dc  0.99 [5]. If all integration points in an element having the
damage value bigger than this critical damage value, this element is defined as totally damaged elements.
As discussed in section 3.1, the totally damaged elements located within very narrow damage bands and
the size of totally damaged elements are limited to be equal or smaller than minimum mesh size.
Therefore, when these elements are removed from the mesh, the volume reduction is very small. Based
on this fact, the compensation of the volume can be performed by relocating the nodes on the crack
surface. This process is similar to a smoothing process, however the displacements of the nodes are
small. As a result, the topology of the mesh around the crack surface can be kept.

3.3 Field transfer operator


3.3.1 Nodal field transfer --- shape function interpolation
The mesh is refined according to cumulative strain and damage field. Then the nodal fields and
integration fields should be transferred from old (coarse) mesh to the new (fine) mesh. Nodal fields are
transferred by shape function interpolation as expressed in Eq. 3:
nGauss

j ( x, y , z ) 
S new N
i 1
old
i ( x, y , z ) Siold (3)

where N iold is the shape function corresponding to node of the containing element in old mesh which
contains the node j in the new mesh. The nodal value Siold is at ith node of the containing element. The
nodal value S new
j is at jth node in the new mesh. Sometimes, after refinement and relocation process, a
node in the new mesh may be located outside the old mesh. In order to interpolate around all boundary
of the old mesh, the shape function here can be negative outside the element. In this case, we choose the
element in the old mesh which is nearest to the considered node in the new mesh as the containing
element. Because this node is very near to the “containing” element, the continuity of shape function is
not disturbed.

3.3.2 Integration field transfer --- hybrid interpolation


Some authors [7] in the literature transfer the integration field by a meshless interpolation called diffuse
approximation [6]. This method is aimed to reconstruct the field by an interpolation function using a
group of neighboring points based on moving least square. A brief introduction is expressed as
following, in Eq. 4:
 a0 
a 
S ( X  X 0 )  P ( X  X 0 )   X 0   1 x  x0 y  y0 z  z0   1 
T

 a2 
 
(4)  a3 
in which X 0 is the evaluation point and S is the field we want to reconstruct in the vicinity of the
evaluation point. P is the basis of a polynomial function and  is the coefficient vector to be solved.
The coefficient vector is obtained by satisfying the following criterion in Eq. 5:
n n
Min : J  ( X  X 0 )   wi ( S ( X i  X 0 )  Si ) 2   wi ( P ( X i  X 0 )   Si ) 2 (5)
T

i 1 i 1

in which Si is the field value on the neighboring point X i . wi is the interpolation weight corresponding
to each neighboring point.
However, most of them treat with a 2D problem or a nearly non-localized problem (no crack
propagation). In our case, the integration field is 3D and highly localized. According to our experience,
the only diffuse approximation is not enough to achieve a robust transfer. The solution is to introduce
also the shape function interpolation process. The whole meshless-FEM based hybrid transfer process
is illustrated in the following Fig. 1. The integration fields of the old mesh are transferred to the nodal
points of the new mesh by diffuse interpolation (meshless) and then again transferred to the integration

191
points of the new mesh by finite element shape function interpolation. The advantages of this hybrid
transfer process is that it preserves the extrema value of the field, especially when the field is localized
as well as minimum the numerical diffusion of the transfer process.

Figure 1. The hybrid transfer process

4. Numerical results
This adaptive methodology is used to simulate crack initiation and propagation on aluminum board
under tensile test. The aluminum alloy EN AW-7108 T6 is used to run the simulation. The numerical
process is carried out by ABAQUS Explicit® in a quasi-static frame. The parameters in equivalent
plastic strain are listed in Tab. 1.

Table 1 Experimentally determined ductile failure parameters [3]


Parameters  T  T k0
Values 0.26 193.0 5.277

Three different shape of specimens with a uniform thickness 0.5mm are used. For each of the specimen,
the initiation and propagation of the crack are displayed. The adaptive process does not influence on the
evolution of the plasticity and damage field. The cracks propagate within a narrow bands. The damage
field on the specimens are colored with blue equals to 0 and red equals to 1.

Figure 2. The crack initiation and propagation on the double arc specimen

Figure 3. The crack initiation and propagation on the double notched specimen

192
Figure 4. The crack initiation and propagation on the specimen with double holes

We can see in the Fig. 2, the double arc specimen is fixed at the left extreme and a load is added at the
right extreme. The crack initiates in the middle of the specimen and propagates to the two arc boundaries.
In Fig. 3, the double notched specimen is also fixed at its left extreme and a load is given at the right
extreme. The crack initiates around two notches and propagates to the middle of the specimen. The
specimen with two holes in Fig. 4, is fixed at the bottom and a load is given at the top. Three cracks
appear on the specimen. We can see that in all numerical results the initial meshes are very coarse
meshes and the meshes are refined around the cracks because of the accumulation of the plasticity and
damage in these zones. The advantages of our method are that saving lots of the computational cost and
no need to pre-refined mesh along the potential crack path.

References

[1] Hooputra, H., Gese, H., Dell, H., & Werner, H. (2004). A comprehensive failure model for
crashworthiness simulation of aluminium extrusions. International Journal of Crashworthiness, 9(5),
449-464.
[2] C. Labergere, A. Rassineux, K. Saanouni, Numerical simulation of continuous damage and fracture
in metalforming processes with 2d mesh adaptive methodology, Finite Elements in Analysis and
Design 82 (2014) 46–61.
[3] Bouchard, P. O., Bay, F., Chastel, Y., & Tovena, I. (2000). Crack propagation modelling using an
advanced remeshing technique. Computer methods in applied mechanics and engineering, 189(3),
723-742.
[4] A. Rassineux, 3d mesh adaptation. Optimization of tetrahedral meshes by advancing front technique,
Computer Methods in Applied Mechanics and Engineering 141 (3) (1997) 335–354.
[5] K. Saanouni, Damage mechanics in metal forming: Advanced modeling and numerical simulation,
John Wiley & Sons, 2013.
[6] P. Breitkopf, G. Touzot, P. Villon, Consistency approach and diffuse derivation in element free
methods based 22on moving least squares approximation, Computer Assisted Mechanics and
Engineering Sciences 5 (4) (1998) 479–501.
[7] D. Brancherie, P. Villon, and A. Ibrahimbegovic. On a consistent field transfer in non-linear inelastic
analysis and ultimate load computation. Computational Mechanics, 42(2):213-226, 2008.

193
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

FLOW OVER A CYLINDER FIXED BY SPRING AND DAMPER SOLVED BY FOURIER


PSEUDOSPECTRAL AND IMMERSED BOUNDARY METHODS

Felipe Pamplona Mariano1, Andreia Aoyagui Nascimento2, Aristeu da Silveira Neto3


1
Universidade Federal de Goiás, fpmariano@ufg.br
2
Universidade Federal de Goiás, aanascimento@ufg.br
3
Universidade Federal de Uberlândia, aristeus@mecanica.ufu.br

Phenomena involving aeroacustic, transition to turbulence and combustion are problems that
engineers aim to understand using techniques of Computational Fluids Dynamics (CFD) [1]. In these
problems are used methods of high order accuracy to obtain results to analyze the real physical
phenomena mentioned. High order methods provide an excellent accuracy, for example, high order
finite differences method and compact schemes [2]. On the other hand, they have disadvantaged of
computational expensive cost in comparison to conventional methods. The mathematical formulation
of spectral methods become possible to join high accuracy with low computational cost [3,4]. This low
computational cost is given by the Fast Fourier Transform (FFT) [5], since the order (O) of the
computational cost of a fluid flow problem solution with finite differences method is O(N2), where N
is the number of the grid points, the cost of the FFT is of O(Nlog2N). Furthermore, it was also
developed the projection method [4], which uncouple the pressure field in the spectral space. Using
the projection tensor is not necessary to calculate the Poisson equation, as has been done by
conventional methodologies to simulate incompressible fluid flows. Normally, to solve this equation is
the most expensive part of a CFD numerical codes [1]. The disadvantage of the spectral methodology
is the difficulty to work with complex geometries and boundary conditions [5]. One of the most
practical methodologies to work with complex and moving geometries is the Immersed Boundary
(IBM) [6]. It is characterized by the imposition of a term source, which has the role of a body force
added to the Navier-Stokes equation to represent virtually a body immersed in the flow [7].
A methodology used, in the present paper, works with Fourier pseudospectral method connected
with Immersed Boundary Method [8]. It is proposed to simulate flows with non-periodic boundary
conditions making use of the term source of immersed boundary. On the other hand, the accuracy of
immersed boundary is improved, at least to second order for smooth solution problems.
The present work shows the simulations of flows over cylinders stationary and flows over cylinders
fixed by springs and dampers. These are some classical problems to validate the procedure of Fluid
Structure Interaction (FSI), i.e., the cylinder can move into the flow and interacts of the fluid flow
edges. These kinds of problems have a great industrial interest and a challenge for researchers in CFD
[9, 10, 11].
It is generate an inlet uniform profile flow, the fluid flow cross the section of a circular cylinder
and we verify the drag and lift coefficients, this variables determine the forces that act on bodies
immersed in flow, the drag coefficient (Cd) determines the resistance force of the fluid on the body
immersed, while the lift coefficient (Cl) determines the force that exists in the direction perpendicular
to incoming flow, an interesting problem in aeronautical engineering is the optimization of airfoils,
that consist in maximize the lift and minimized the drag of the airfoil profiles. Other parameter

194
analyzed is the Strouhal number (St) which determines the non-dimensional vortex shedding, it is
important to solve problems of fluid-structure interaction, for example, pillars of bridges, or wings of
aircraft, submitted to a flow, if the frequency of vortex shedding is close to the natural frequency is
extremely damaging to these structures.
The Fig. 1 shows vorticity isocontours (-1,0<w<1,0) of fluid flow over a cylinder, at Reynolds
number equal 100. In the Fig. 1 it is possible to see the periodical boundary conditions and the
Immersed Boundary method acting over the flow.

Figure 1. Isocontours of vorticity (-1,0<w<1,0) at Re=100. - negative vortice; -- positive vorticity.

The Tab. 1, shows the comparison between Cd, Cl and St for different Reynolds numbers and with
different authors.

Table 1. Comparison of drag coefficient and Strouhal number.


[12] [13] [14] [15] Present work
Re Cd Cl St Cd Cl St Cd Cl St Cd Cl St Cd Cl St
100 1,39 0,20 0,160 1,44 0,33 0,165 1,42 0,34 0,171 1,39 0,34 0,160 1,45 0,35 0,175
150 1,37 0,25 0,175 1,47 0,58 0,184 1,37 0,49 0,200
200 1,42 0,66 0,202 1,38 0,68 0,192 1,27 0,47 0,213
300 1,22 0,27 0,190 1,08 0,39 0,221

To study the effects of fluid structure interaction (FSI) we realize simulations of flows over
cylinder fixed by two spring-damper systems in agreement of the works of [10] and [11]. Furthermore,
we perform the simulations of two different time advanced methods, the first is the classical explicit
Euler and the second one is the forth order Runge-Kutta with six steps and low dissipation and low
dispersion method (RK46) purposed by [16]. In the Fig. 2 is shown the vorticity field of both methods
at the same time and at the same Reynolds number.

Figure 2. Vorticity field of flows over cylinder fixed by spring. In the left figure the simulation is
realized using the explicit Euler and the right using the forth order Runge-Kutta method.

195
In the Fig. 2 it is noteworthy the change of the flow pattern of the wake vortices, in the simulation
of RK46 the pattern of flow is the 2-C and in the explicit Euler the pattern is 2-S. The Fig. 3 presents
the maximum displacement of center of the cylinder using the RK46 in comparison with the [10]. The
solution is very similar to work of the reference, but it is a little overestimated. It is important
understand that the results of [10] are obtained using a Finite Volume Method.

Figure 3. Maximum displacement of cylinder center.

The Fourier pseudospectral method allows solver the incompressible Navier-Stokes equations with
high order accuracy and with low computational cost when compared another high order method,
because the pressure disentail and the use FFT algorithm. In the simulations of the flow over a circular
cylinder it is possible observe the drag and lift coefficients and Sthrouhal number quite like other
authors. In the simulations of fluid structure interaction, it is important highlight the time advanced
method in both, fluid and rigid body movement, differential equations. It is required the use the high
order time-advanced method, in the present paper we adopted the Runge-Kutta with low dissipation
and low dispersion purposed by [16]. This way allows we obtain the correct flow pattern and
displacement of the cylinder center.

References

[1] S.V. Patankar. Numerical heat transfer and fluid flow. Hemisphere, 1980.
[2] G.E. Karniadakis, S.A. Orszag. Nodes, modes and flow codes. Physics Today, 34-42, 1993.
[3] S.A. Orszag. Spectral Methods for Problems in Complex Geometries, Journal of Computational
Physics, 70-92, 1970.
[4] C. Canuto, M.Y. Hussaini, A. Quarteroni, T.A. Zang. Spectral methods: fundamentals in single
domains, Springer-Verlag, 2006.
[5] W.L. Briggs, V.E. Henson. The DFT, SIAM, 1995.
[6] C.S. Peskin. The immersed boundary method, Acta Numerica, 479–517, 2002.
[7] Z. Wang, J. Fan, K. Luo. Combined multi-direct forcing and immersed boundary method for
simulating flows with moving particles, International Journal of Multiphase Flow, 283-302, 2008.
[8] F.P. Mariano, L.Q. Moreira, A. Silveira Neto, C.B. da Silva, J.C.F. Pereira. A new incompressible
Navier-Stokes solver combining Fourier pseudo-spectral and immersed boundary methods, Computer
Modeling in Engineering Science, pp. 181-216, 2010.

196
[9] S. Enriquez-Remigio, A. Silveira Neto. A new modeling of fluid-structure interaction problems
through immersed boundary method/virtual physical model (IBM/VPM), Proceedings of the 19th
Brazilian Congress of Mechanical Engineering, 1-10, 2007.
[10] M.J. Chern, Y.H. Kuan, G. Nugroho, G. TingLu, T.L. Horng. Direct-forcing immersed boundary
modeling of vortex-induced vibration of a circular cylinder, Journal of Wind Engineering and
Industrial Aerodynamics, 109–121, 2014.
[11] S.P. Singh, S. Mittal. Vortex-induced oscillations at low Reynolds numbers:
Hysteresis and vortex-shedding modes, Journal of Fluids and Structures, 1085–1104, 2005.
[12] A.L.F. Lima e Silva, A. Silveira Neto, J. Damasceno. Numerical simulation of two dimensional
flows over a circular cylinder using the immersed boundary method, Journal of Computational
Physics, 351-370, 2003.
[13] M.C. Lai, C.S. Peskin. An immersed boundary method with formal second order accuracy and
reduced numerical viscosity, Journal of Computational Physics, 705–719, 2000.
[14] S. Xu and Z.J. Wang, An immersed interface method for simulating the interaction of a fluid with
moving boundaries, Journal of Computational Physics. 454-493, 2006.
[15] D.V. Le, B.C. Khoo, K.M. Lim, An implicit-forcing immersed boundary method for simulating
viscous flows in irregular domains, Computer Methods in Applied Mechanics and Engineering, 2119-
2130, 2008.
[16] V. Allampalli, R. Hixon, M. Nallasamy, S. Sawyer. High-accuracy large-step explicit Runge-
Kutta (Hale-RK) schemes for computational aeroacoustics, Journal of Computational Physics, 3837-
3850, 2009.

197
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A NUMERICAL STUDY OF FLOW AROUND CIRCULAR AND RECTANGULAR CYLIN -


DERS

Elder Gualberto Alves1 , Felipe Pamplona Mariano2 , Andreia Aoyagui Nascimento3


1 Universidade Federal de Goiás, eldergualberto55@gmail.com
2 Universidade Federal de Goiás, fpmariano@ufg.br
3 Universidade Federal de Goiás, aanascimento@ufg.br

The study of flow around slender structures (circular and rectangular cylinders) are of practical inter-
est to many fields of engineering because many civil structure and industrial structure can be assimilate to
this shape. Many publications about circular cylinder such as Sarpkaya (1979, 2004), Sumer and Freds0e /
(1997), Bearman (1984), Williamson and Govardhan (2004, 2008), are motivation to any researchers sty-
ding this physical problem. In contrast, flow around rectangular cylinders has not been investigated to
the same extent, although this has many technical applications such as building aerodynamics; Franke et
al. (1990), Klekar and Patankar (1992), Davis and Moore (1982), Davis et al. (1984), Okajima (1982,
1990), Mukhopadhyay et al. (1992) and Suzuki et al. (1993). In this paper, addresses the application
Fourier Pseudoespectral Method (FPM) coupled with Immersed Boundary Method (IBM) for incom-
pressible and isothermic flow, two-dimensional domain, through cylinders of circular and rectangular
cross-sections. The Navier-Stokes equations are solving for FPM, and the geometries are represented by
immersed boundary. Computer simulations were solving for low Reynolds numbers and four rectangu-
lar models having width-to-height ratios of 1.0, 2.0, 4.0 and 16.0 were considered. Results about vortex
sheddings and Strouhal are discussed.

References
[1] Franke, R., Rodi, W., Schönung, B., Numerical calculation of laminar vortex shedding past cylinders. J. Wind
Eng. Ind. Aerodyn. 35, 237-257, 1990.
[2] Klekar, K.M., Patankar, S.V., Numerical prediction of vortex shedding behind square cylinders. Int. J. Numer.
Meth. Fluids 14, 327-341, 1992.
[3] Davis, R.W., Moore, E.F., A numerical study of vortex shedding from rectangles. J. Fluid Mech. 116, 475-506,
1982.
[4] Davis, R.W., Moore, E.F., Purtell, L.P.,A numerical-experimental study of confined flow around rectangular
cylinders. Phys. Fluids 27, 46-59, 1984.
[5] Okajima, A., Strouhal numbers of rectangular cylinders. J. Fluid Mech. 123, 379-398, 1982.
[6] Okajima, A., Numerical simulation of flow around rectangular cylinders. J. Wind Eng. Ind. Aerodyn. 33,
171-180, 1990.
[7] Mukhopadhyay, A., Biswas, G., Sundararajan, T., Numerical investigation of confined wakes behind a square
cylinder in a channel. Int. J. Numer. Meth. Fluids 14, 1473-1484, 1992.
[8] Suzuki, H., Inoue, Y., Nishimura, T., Fukutani, F., Suzuki, K., Unsteady flow in a channel obstructed by a
square rod (crisscross motion of vortex). Int. J. Heat Fluid Flow 14 (1), 2-9, 1993.
[9] Sarpkaya, T.,Vortex-induced oscillations: a selective review. J. Appl. Mech. 46, 241-258, 1979.
[10] Sarpkaya, T.,A critical review of the intrinsic nature of vortex-induced vibrations. J. Fluids Struct. 19, 389-
447, 2004.

198 1
/ J., Hydrodynamics Around Cylindrical Structures. World Scientific, Singapore, 1997.
[11] Sumer, B.M., Freds0e,
[12] Williamson, C.H.K., Govardhan, R., Vortex-induced vibrations. Annu. Rev.Fluid Mech. 36, 413-455, 2004.
[13] Williamson, C.H.K., Govardhan, R., A brief review of recent results in vortex-induced vibration, J. Wind Eng.
Ind. Aerodyn. 96, 713-735,2008.
[14] Bearman, P.W., Vortex shedding from oscillating bluff bodies. Annu. Rev. Fluid Mech. 16, 195-222, 1984.

199 2
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A SSESSMENT OF NON - CONVENTIONAL CONSTRAINTS IMPOSED TO A HY-


PERELASTIC MODEL

Felipe Tempel Stumpf 1 , Rogerio Jose Marczak 1


1 Departamento de Engenharia Mecanica, Universidade Federal do Rio Grande do Sul.

In order to satisfy requirements such as unicity of solutions and physically plausible mechanical
behavior, it is mandatory that the strain energy function W of hyperelasticity fulfills some mathematical
conditions, such as policonvexity and the Baker-Ericksen inequalities. Policonvexity is assured by the
positive definiteness of the Hessian of W, implying uniqueness of solutions and numerical stability in the
framework of computer simulations with Newton type methods. The Baker-Ericksen inequalities, on the
other hand, are sufficient to guarantee that the material behaves in a physically plausible way, although
they are rarely taken into account during the procedure of curve fitting. A hyperelastic model must,
therefore, be developed in a way that its equation for W allows the simultaneous satisfaction of both of
these requirements. Here we propose such an isotropic constitutive model based on a modification of a
previously developed equation in order to enable the fulfillment of the aforementioned conditions as well
as incorporating anisotropic capability based on the Holzapfel-Gasser-Ogden model.
——————————————-

200 1
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

S OFTWARE EROSION ON A STRUCTURAL ANALYSIS SOFTWARE - A CASE


STUDY

Flavio Torres da Fonseca 1 , Roque Luiz da Silva Pitangueira 2 , Samuel Silva Penna 3
1 Universidade Federal de Minas Gerais, flaviotf@dees.ufmg.br
2 Universidade Federal de Minas Gerais, roque@dees.ufmg.br
3 Universidade Federal de Minas Gerais, spenna@dees.ufmg.br

Abstract
The INSANE (Interactive Structural Analysis Software) is being developed by the Structural Engineering
Department of Federal University of Minas Gerais (DEES/UFMG) since 2002. Through these years,
the system has evolved from a simple 2D frame element analysis software to a very complex system,
with many different methods, elements, constitutive models and solution types. However, the system
also suffered from a well known phenomena called software erosion: a slow deterioration of software
performance over time. In this paper, the INSANE system will be briefly described and a diagnosis
of its actual state will be presented. Software erosion causes, consequences and possible solutions or
preventions will be discussed.

1. A brief history of the INSANE project


The INSANE (Interactive Structural Analysis Software) software started to be developed in 2002, by
the Structural Engineering Department of Federal University of Minas Gerais (DEES/UFMG), as an
undergraduate research program. Its initial objective was to create a simple 2D frame element analysis
software to be used as an academic resource in Structural Analysis courses [1, 2]. However, since the
beginning, its software architecture and class organization were planned to be as general as possible, in
order to enable its expansion.
Since then, many undergraduate and graduate researchers have worked on it, resulting in 29 Master
Thesis and 5 Doctoral Dissertations [3].
Today, the INSANE is a complex system with many methods (Finite Element Method - FEM, Gener-
alized Element Method - GFEM, Boundary Element Method - BEM and Mesh Free Methods), linear and
non linear (physically and geometrically) analysis, static and dynamic analysis, as well as many different
constitutive models [4, 5, 6, 7, 8, 9].
Despite many efforts have been put in creating a friendly and complete user interface [10], its devel-
opment did not follow the advances of the analysis modulus, so the last stable version of INSANE was
released in 2004 and presents only the frame elements interface.

2. Software erosion on the INSANE source code


Software erosion can be described as a slow deterioration of software performance over time. It is a
phenomena that has been studied by computer scientists for decades and has been detected not only
in academic softwares, but also in well known open sources and proprietary projects, as Mozilla web
browser, VIM text editor, ANT and version 2.4 of Linux kernel [11, 12, 13].

201 1
The biggest problem with software erosion is that its effects accumulate over time, making the source
code harder to maintain and to understand.
One of the main causes of this phenomena in the INSANE comes from the project’s own nature: an
academic project, in which each developer has an individual objective and a rigid time schedule. The
rotativity of contributors is very high, so the understanding of the system as a whole and the formation of
a developing culture is not achieved. Another point is the lack of sense of community. Some people tend
to think only in its own work and do not worry about the effects of the modifications on other developers’
work.
In the INSANE source code the most commons signs of software erosion are cyclic dependencies,
dead code (unused code, large number of commented lines) and code clones (identical or near-identical
code fragments).
A sign of erosion which appeared after the adoption of Git as the version control system is the huge
number of branches that were created and that are never updated with the master branch modifications
or nerver merged into master. There are also some orphan branches which were created, modified and
abandoned.
Another problem in INSANE code, but that is not direct related to software erosion but indirectly
contributes for it, is the small quantity of implemented unit tests. To maintain a stable code and enable
refactoring, it is imperative to have tests, to ensure that the software works properly.

3. Recommendations to avoid and reverse software erosion


The first step to avoid the software erosion process is to put into the mind of all developers that the
INSANE system is bigger than their individual work and that other people will need to understand and
use their source code. By knowing this, it is expected that all contributors will create good ‘legacy’ code
for the upcoming developers.
Another step is to introduce code review process, which can be done within the adopted version
control system (Git and GitLab). This will force the code to be well written and dead code to be removed.
To reverse the software erosion, the existing source code must be refactored. This refactorization
comprises dead code removal, proper documentation of source code, numerical methods optimization
and software architecture modification. For this last change, one of the possibilities is to adopt a plug-in
based architecture, which will modularize and encapsulate even more the source code, protecting the
core classes and methods.
In order to guarantee that the code is correct and functional, unit tests must be implemented.

4. Conclusion
The INSANE source code has suffered a heavy process of software erosion and a lot of work must be
done in order to make it useful, understandable and expandable. A complete refactoring and modification
of its software architecture, adopting a plug-in based architecture, is the best option and will be done in
the next few years.

5. Acknowledgments
The authors would like to acknowledge CNPq (National Council of Scientific and Technological De-
velopment), CAPES (Coordination of Improvement of Higher Education Personnel), FAPEMIG (Minas
Gerais State Research Foundation) and PROPEEs/UFMG (Structural Engineering Graduate Program of
the Federal University of Minas Gerais) for financial supports.

References
[1] F.T. Fonseca, R.L.S. Pitangueira. Um programa gráfico interativo para modelos estruturais de barras, XXV
CILAMCE - Iberian Latin American Congress on Computational Methods in Engineering, 2004.

202 2
[2] R.L.S. Pitangueira, K. Caldas. Projeto de software livre para modelos do Método dos Elementos Finitos, XXVI
CILAMCE - Iberian Latin American Congress on Computational Methods in Engineering, 2005.
[3] INSANE project website. http://www.insane.dees.ufmg.br [Accessed: 31/05/2018].
[4] F.T. Fonseca, R.L.S. Pitangueira. Insane: uma plataforma para computação científica, X Encontro de Mode-
lagem Computacional, 2007.
[5] P.D. Alves, F.B. Barros, R.L.S. Pitangueira. An object-oriented approach to the Generalized Finite Element
Method, Advances in Engineering Software, 1-18, 2013.
[6] A.B. Monteiro, A.R.V. Wolenski, F.B. Barros, R.L.S. Pitangueira, S.S. Penna. A computational framework for
G/XFEM material nonlinear analysis, Advances in Engineering Software, 380-393, 2017.
[7] R.G. Peixoto, F.E.S. Anacleto, G.O. Ribeiro, R.L.S. Pitangueira, S.S. Penna. A solution strategy for non-
linear implicit BEM formulation using a unified constitutive modelling framework, Engineering Analysis with
Boundary Elements, v. 64, p. 295-310, 2016.
[8] Silva, R.P. Análise Não-Linear de Estruturas de Concreto por meio do Método Element Free Galerkin, Doc-
toral dissertation, Federal University of de Minas Gerais, Belo Horizonte, Brazil, 2012.
[9] L. Gori, S.S. Penna, R.L.S. Pitangueira. A computational framework for constitutive modelling, Computers &
Structures, 1-23, 2017.
[10] S.S. Penna, R.L.S. Pitangueira. Projeto orientado a objetos de um pós-processador para modelos do Método
dos Elementos Finitos, XXVII CILAMCE - Iberian Latin American Congress on Computational Methods in
Engineering, 2006.
[11] M. Dalgarno. When good architecture goes bad, Methods & Tools, Editor, 27-34 , 2009.
[12] L. Silva, D. Balasubramaniam. Controlling software architecture erosion: A survey, The Journal of Systems
and Software, 132-151, 2012.
[13] R. Terra, M.T. Valente, K. Czarnecki, R.S. Bigonha. Recommending refactorings to reverse software ar-
chitecture erosion, 16th European Conference on Software Maintenance and Reengineering, IEEE, 335-340,
2012.

203 3
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

E NTROPIC CONSIDERATIONS ON THE LBGK MODEL FOR ADVECTION - DIFFUSION


Florian De Vuyst1 , Thomas Douillet-Grellier 2

1 LMAC, EA 2222, Sorbonne Universités, Université de Technologie de Compiègne


2 CMLA, ENS Cachan, CNRS, Université Paris-Saclay, 94235 Cachan, France

1. Introduction
The kinetic origin of the LB method yields the question of the compatibility of the LB method
with the H-theorem. This topic has gained attention [1, 2] because it is strongly connected to
the stability properties of the method. In particular, in order to palliate the fact that classic
LB schemes do not verify the H-theorem, new LB formulations have been introduced based
on the construction of specific equilibria equipped with an H-theorem [3] or entropy-controlled
schemes [4, 5].
In this work, we study the entropy properties of the D1Q3 LBGK scheme applied to the 1D
advection-diffusion equation. First, we introduce the considered model and its relevant proper-
ties. Then, we show how it satisfies a discrete H-theorem for a particular entropy functional.
In this process, we reinterpret the BGK operator as a gradient of a Lagrangian functional while
exhibiting the the entropy dissipation source terms and the entropy fluxes. Our observations
are validated through numerical experiments. Moreover, equilibrium distributions are obtained
from minimization principles.

2. LB method for the 1D advection-diffusion equation


Let’s consider the 1D advection-diffusion equation with a constant advection speed u ∈ R:
2
∂t ρ + ∂x (ρu) − ν ∂xx ρ = 0, x ∈ R, t > 0 (1)

with ρ = ρ(x, t) the density and ν the viscosity. The initial condition is ∀x ∈ R, ρ(x, t = 0) = ρ0 (x)
where ρ0 is the initial density distribution. Additionally, we assume ν > 0 and we only consider
positive solutions of equation (1) such that ρ(x, t) ≥ 0.
Let us consider the D1Q3 lattice with its discrete velocities v = (v− , v0 , v+ )T = (−c, 0, c)T and
its discrete distribution vector f = (f− , f0 , f+ )T . The LB equations (LBE) are given hereafter
in vector form

∂t f + Λ ∂x f = q(f ), (2)
where the source terms q is the collision term and with the diagonal and constant advection ma-
trix Λ defined by Λ = diag(−c, 0, −c) We connect the LBE (2) with the macroscopic equation (1)
by imposing the follwoing constraints

hf , ei = ρ, hq(f ), ei = 0 (3)

where h., .i denotes the standard scalar product, and e = (1, 1, 1)T , q(f ) = (qi (f ))i .
The BGK collision model is defined as a relaxation term towards a discrete equilibrium
distribution f eq , i.e.
f eq − f
q(f ) = (4)
τ0

204 1
with τ 0 > 0 a relaxation time.
It is also expected that f eq has the 0th -order moment property but also the 1st -order and
2 -order moment consistency properties. For a certain constant γ > 0, We can write
nd

hf eq , ei = ρ, hf eq , vi = ρu, hf eq , v 2 i = γρc2 . (5)

3. Equilibrium distributions
In this section, we will obtain the equilibrium distributions associated with the entropy functional
defined in equation (6) from minimization principles.

3.1. Entropy functional


As shown in the previous section, consistency is obtained under the hypothesis of small flow
velocities. Thus, it is always assumed that |u|  c. In particular, the inequality 1 − 3|u|
2c < 1
holds. Therefore, the follwoing functional H(f ) is clearly strictly convex
X
H(f ) = hi (fi ), (6)
i∈{−,0,+}

fi2 1 1
with hi defined as hi (fi ) = 2 wi , i ∈ {−, 0, +} and w− =3u , w0 = 1, w+ = .
1 − 2c 1 + 3u
2c
We call H(f ) an entropy for f . The entropy functional H of equation (6) can be rewritten
1 1
H(f ) = hDf , f i = kf k2D . (7)
2 2
where D = diag( 1−13u , 1, 1+13u )
2c 2c

3.2. Entropy minimization problem


To begin, let us consider the constrained minimization problem
min H(f ), (8)
f =(f− ,f0 ,f+ )T

subject to 
f ≥ 0,
(9)
hf , ei = ρ
The functional H is coercive, strictly convex and the admissible set is non-empty, closed and
convex. Therefore, there is a unique solution f eq to the problem (8-9). To this minimization
problem, we associate a Lagrange multiplier λ ∈ R and the Lagrangian functional L : R3 ×R → R
expressed as
L (f , λ) = H(f ) + λ ϕ(f ). (10)
The necessary 1st -order optimality conditions are given by the Euler-Lagrange equations:
∇L (f eq , λ? ) = ∇H(f eq ) + λ? e = 0, (11)
It leads to the following solution
ρ ρu ρ ρ ρu
(f− )eq = − , (f0 )eq = , (f+ )eq = + .
3 2c 3 3 2c

3.3. Reinterpretation of the BGK collision term


We now focus on the collision term of the LBE. Let us choose the collision term as
1
q(f ) = − 0 D−1 ∇f L (f , λ? ), (12)
τ
with τ 0 > 0 a characteristic time of collision and L defined in equation (10). We can write
1 −1 f eq − f
q(f ) = − D (∇H(f ) + λ?
e) = .
τ0 τ0
We retrieve the standard BGK collision term [6].

205 2
4. Discrete H-Theorem
The time partial derivative is now discretized by the Euler explicit scheme. The lattice relaxation
τ0
time is chosen as τ = .
∆t

4.1. Collision step


We first study the entropy property of the collision step. To simplify notations, we omit the
dependency in x position, and denote f at discrete time tn = n∆t by f n . The collision step on
a time step ∆t reads
1
fˆn = f n + (f eq,n − f n ) . (13)
τ
After some algebraic manipulations, we can obtain
1 1
H(fˆn ) − H(f n ) = − (τ − ) kf eq,n − f n k2D . (14)
τ 2 2
We then get a local entropy dissipation proportional to kf eq,n −f n k2 , as soon as τ > 12 . We notice
that the numerical stability limit, τ > 1/2, is naturally recovered through entropic considerations.

4.2. Transport step


Let us now consider the transport step. At a lattice node x and at instant tn+1 , this reads

fin+1 (x) = fˆin (x − vi ∆t). (15)

Let us now consider the transport of entropy quantities. From (15), we get the conservation
scheme
∆t n
H(f n+1 (x)) = H(fˆn (x)) − [φ (x + h/2) − φn (x − h/2)] . (16)
h
where the numerical entropy flux φn (x + h/2) = φn (fˆ(x), fˆ(x + h)) is given by

φn (x + h/2) = c h+ (fˆ+ (x)) − c h− (fˆ− (x + h)). (17)

4.3. Whole discrete entropy inequality


By combining entropy balances for both collision (14) and transport (16) steps, we get a discrete
entropy balance for the LBGK scheme:
∆t n 1 1
H(f n+1 (x)) − H(f n (x)) + [φ (x + h/2) − φn (x − h/2)] = − 2 (τ − ) kf eq,n − f n k2D ≤ 0. (18)
h τ 2

5. Numerical experiments
As a numerical illustration of the behavior of equation (18), a couple of numerical experiments
have been undergone. The spatial domain is Ω = [0, 1] (in meters for example) and is closed with
periodic boundary conditions. Ω is discretized with 4000 lattice points, so that the space step
h = 2.5×10−4 m. Regarding the physical parameters, we have set the advection speed u = 0.1m/s
and the speed of sound c = 2m/s. Consequently, the time step is dt = h/c == 2.5 × 10−4 s. The
initial density profile to be advected is chosen as piecewise constant ρ0 (x) = 1x∈[1/3,2/3] (x).
Figure 1 shows the density profile at t = 10s and the corresponding entropy dissipation for two
different values of τ .
It is observed on both figures that the the entropy dissipation is strictly negative up to
machine precision as predicted by equation (18). Note that we have tested to compute the
entropy dissipation as H(f n+1 (x)) − H(f n (x)) + ∆t 1
h [φ (x + h/2) − φ (x − h/2)] or as − τ 2 (τ −
n n
1 eq,n − f n k2 and that it gives exaclty the same results. The entropy dissipation is a good
2 ) kf D
indicator of density changes and could eventually been used as a tool to track discontinuities in
a multi-fluid context.

206 3
2 2

Entropy Dissipation (.10−7 )

Entropy Dissipation (.10−7 )


Profile at t = 10s Profile at t = 10s
Profile at t = 0s 0 Profile at t = 0s 0

1 1
-1 -1
ρ

ρ
0 -2 0 -2

-3 -3

-1 -1
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

x(m) x(m) x(m) x(m)

(a) (b) (c) (d)

Figure 1: Advection of a rectangular profile for 10s and its associated entropy dissipation for
τ = 0.51 (a,b) and τ = 0.501 (c,d)

6. Conclusion
In this paper, we have shown that by considering a particular entropy functional, it is possible
to derive a discrete H-theorem for the LBGK scheme applied to the 1D advection-diffusion
equation in a non-standard way. A fine evaluation of the entropy dissipation is presented where
the dissipation source terms due to the collision step and the entropy fluxes associated with the
transport step are exhibited. The quantification of entropy dissipation is verified by numerical
experiments that confirm the theoretical results.
Through the exploration of the intrinsic entropic properties of this LBGK scheme, we have
obtained the equilibrium distribution by solving an entropic minimization problem and pro-
posed a reinterpretation of the BGK collision term as the gradient of the Lagrangian functional
associated with this minimization problem.
Future work will be focused on extending our approach to more general entropy functionals,
to 2D, to different collision operators and to non-periodic domains. We aim to improve the
understanding of the numerical stability conditions of the LBGK scheme using these entropic
considerations.

References
[1] Iliya V Karlin, Alexander N Gorban, Sauro Succi, and V Boffi. Maximum entropy principle for lattice
kinetic equations. Physical Review Letters, 81(1):6, 1998.
[2] Li-Shi Luo. Some recent results on discrete velocity models and ramifications for lattice boltzmann
equation. Computer Physics Communications, 129(1):63 – 74, 2000.
[3] Santosh Ansumali and Iliya V Karlin. Entropy function approach to the lattice boltzmann method.
Journal of Statistical Physics, 107(1-2):291–308, 2002.
[4] Santosh Ansumali and Iliya V. Karlin. Stabilization of the lattice boltzmann method by the H-
theorem: A numerical test. Phys. Rev. E, 62:7999–8003, Dec 2000.
[5] Bruce M Boghosian, Jeffrey Yepez, Peter V Coveney, and Alexander Wager. Entropic lattice boltz-
mann methods. Proceedings of the Royal Society of London A: Mathematical, Physical and Engineer-
ing Sciences, 457(2007):717–766, 2001.
[6] P. L. Bhatnagar, E. P. Gross, and M. Krook. A model for collision processes in gases. i. small
amplitude processes in charged and neutral one-component systems. Phys. Rev., 94:511–525, May
1954.

207 4
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

C OMPACT THIRD ORDER EXPANSION OF LATTICE B OLTZMANN SCHEMES


François Dubois 12
1 LMSSC, Conservatoire National des Arts et Métiers, Paris
2 Department of Mathematics, University Paris Sud, Orsay
francois.dubois@u-psud.fr
30 May 2018

The lattice Boltzmann scheme in his actual form has been developed with the contributions of Lallemand,
Succi, d’Humières, Luo [1, 2, 3, 4] and many others. In order to derive the equivalent partial differential
equations, a classical of the Chapman Enskog expansion is popular in the lattive Boltzmann community
(see e.g. [4]). A main drawback of this approach is the fact that multiscale expansions are used without a
clear mathematical signification of the various variables and functions. Independently of this framework,
we have proposed in [5, 6] the Taylor expansion method to obtain formally equivalent partial differential
equations. The infinitesimal variable is simply the time step (proportional to the space step with the
acoustic scaling). This approach has been experimentaly validated in various contributions [7, 8]. A
third order extension for fluid flow has been proposed in [9] and an efficient implementation up to fourth
order accuracy is presented in [10].
In this contribution, we consider a regular lattice L composed by vertices x separated by distances that
are simple expressions of the space step ∆x. A discrete time t is supposed to be an integer multiple of a
time step ∆t > 0. A very general lattice Boltzmann scheme with q discrete velocities of the form

f j (x, t + ∆t) = f j∗ (x − v j ∆t, t) , 0 ≤ j < q .

The distribution f ∗ after relaxation is defined with moments m such that

mk = ∑ Mk` f j .
j

The d’Humières matrix [3] M is invertible and we decompose the moments in the following way:
 
W
m≡ .
Y

The conserved variables W are not modified after relaxation: W ∗ = W . The microscopic variables Y
are changed in a nonlinear way by the relaxationprocess:

Y ∗ = Y + S (Φ(W ) −Y ) .

The matrix S is invertible, and ofter chosen as diagonal. It is supposed to be fixed in the asymptotic
process presented hereafter. The equilibrium values Y eq = Φ(W ) are given smooth functions of the
conserved variables. When ∗∗ is evaluated, we have simply

f ∗ = M −1 m∗ .

With this general framework, we determine in this contribution an asymptotic system of partial differen-
tial equations for the conserved variables:

∂t W + Γ1 (W ) + ∆t Γ2 (W ) + ∆t 2 Γ3 (W ) = O(∆t 3 )

208 1
and an asymptotic expansion for the microscopic variables as a differential nonlinear function of the
conserved variables:
Y = Φ(W ) + ∆t Φ1 (W ) + ∆t 2 Φ2 (W ) + O(∆t 3 ) .
The coefficients Γ1 , Φ1 , Γ2 , Φ2 and Γ3 of this expansion are recursively determined as a function of the
data v j , M, Φ(W ) and S. We compare our new result with the particular third order expansion proposed
in [9] and the linear approach presented in [10].

References
[1] FJ. Higuera, S. Succi, R. Benzi. Lattice gas dynamics with enhanced collisions, Europhysics Letters, volume 9,
issue 4, p. 345-349, 1991.
[2] YH Qian, D d’Humières, P. Lallemand. Lattice BGK models for Navier-Stokes equation, Europhysics Letters,
volume 17, issue 6, p. 479-484, 1992.
[3] D. d’Humières. Generalized Lattice-Boltzmann Equations, in Rarefied Gas Dynamics: Theory and Simula-
tions, volume 159 of AIAA Progress in Astronautics and Astronautics, p. 450-458, 1992.
[4] P. Lallemand and L.-S. Luo. Theory of the Lattice Boltzmann Method: Dispersion, Dissipation, Isotropy,
Galilean Invariance, and Stability, Physical Review E, volume 61, 6546, 2000.
c
[5] F. Dubois. Une introduction au schéma de Boltzmann sur rà seau, ESAIM Proceedings, volume 18, p. 181-
215, 2007.
[6] F. Dubois. Equivalent partial differential equations of a lattice Boltzmann scheme, Computers and Mathemat-
ics with Applications, volume 55, p. 1441-1449, 2008.
[7] F. Dubois, P. Lallemand. Towards higher order lattice Boltzmann schemes, Journal of Statistical Mechanics:
Theory and Experiment, P06006, 2009.
[8] F. Dubois, P. Lallemand. Quartic Parameters for Acoustic Applications of Lattice Boltzmann Scheme, Com-
puters and Mathematics with Applications, volume 61, issue 12, p. 3404-3416, 2011.
[9] F. Dubois. Third order equivalent equation of lattice Boltzmann scheme, Discrete and Continuous Dynamical
Systems-Series A, volume 23, issue 1/2, p. 221-248, 2009.
[10] A. Augier, F. Dubois, B. Graille et P. Lallemand. On rotational invariance of Lattice Boltzmann schemes,
Computers and Mathematics with Applications, volume 67, issue 2, p. 239-255, 2014.

209 2
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

O N THE OPTIMIZATION OF POWER GENERATED BY COMBINED CYCLE HEAVY-


DUTY GAS TURBINE

Paulo Guilherme Inca1 , Gabriel Maidl 2 , Eduardo Massashi Yamao 3 , Renato de Arruda Penteado Neto
4 , Leandro dos Santos Coelho 5 , Cintia de Carvalho Toledo 6

1 Lactec and Federal University of Parana, paulo.inca@lactec.org.br


2 Lactec and Pontifical Catholic University of Parana, gabriel.maidl@lactec.org.br
3 Lactec, eduardo.yamao@lactec.org.br
4 Lactec, renato@lactec.org.br
5 Pontifical Catholic University of Parana and Federal University of Parana, leandro.coelho@pucpr.br
6 U. E. G. Araucaria, cintia@uega.com.br

1. Introduction
The usage of natural gas to generate power has been increasing over the years. This is explained by
the fact that natural gas fuel emits less pollutant when compared with coal, for instance [1]. However,
awareness with climate change has also been raising over the last decades [2]. This scenario makes
companies that produces energy seek for solutions that reduces the amount of pollutant emitted by their
power plants, while trying not to reduce the power generated.
Some researches aiming to reduce the emissions of SO2 , NOx , CO, and CO2 include the installation
of pollutant cleaning, switching to low-emission fuels, replacement of aged parts of the power plant, and
emissions dispatch [3]. This environment awareness and the legislation that regulates pollutant emissions
by power plants, enforce companies to be under a certain limit. The Brazilian legislation delimits the
level of pollutant gases emission in the environment coming from energy production in power plants. It
also applies fines and criminal penalties to companies emitting over the limit [4].
The energy consumption is also rising all over the world. If it keep the trend, energy demand is
supposed to increase 65% from 2004 to 2030. Nevertheless, most of the energy consumed worldwide
is not renewable [5]. Considering these reasons, developing new methodologies to reduce the emissions
of harmful gases that pollute the atmosphere or make the combustion of the fuel more efficient is likely
to scale down the environmental impact caused by the energy production from thermal power plants.
Therefore, this paper proposes a optimization of power generated by a combined cycle heavy-duty gas
turbine using the Self-adaptive Differential Evolution (SaDE) algorithm.

2. Combustion Chamber
The combustion chamber studied in this article is part of a heavy-duty gas turbine used for power genera-
tion in a gas power plant. The gas turbine operates using natural gas in a combined cycle, where the hot
gases generated by the combustion in the gas turbine are used to move a steam turbine. The combustion
system is composed by sixteen combustion chambers, assembled in a circular arrangement.
Each combustion chamber has a basket, where the combustion occurs, a transition piece, which
directs the hot gases produced in the combustion to the turbine, and a set of injectors, which has a
pilot injector surrounded by eight secondary injectors. These injectors conduct the gas fuel into the
combustion chamber through calibrated holes that promote a strong turbulence, favoring the mixture
between the compressed air that comes from the compressor, and the fuel.The secondary injectors are

210 1
geometrically equal and they are called stages A and B, alternately arranged around the pilot injector.
The pilot injector differs from the others by its size and presence of a complementary part used for air
injection, which characterizes the low NOx emission system.

3. Models Identification
The modeling of both power and NOx models uses the gas demand in the stages A, B, C, and pilot, and
the ambient temperature in Celsius. The NOx model also take into consideration the power generated
by the turbine in Megawatts (MW). The process of identification of these models uses neural networks,
with a variation of the standard network which is called Radial Basis Function (RBF) [6]. An RBF
neural network uses a Gaussian function as the activation function and it is used in many applications for
function approximation.

4. Optimization Procedure
Different techniques can be used to solve optimization problems, such as convex programming [7] and
metaheuristics [8]. This research is focused on metaheuristic optimization which uses stochastic methods
to search for the global minimum and maximum of the objective function, aiming to avoid local soluti-
ons. Although these methods may not always find the optimal solution, the result is often a satisfactory
solution for the problem [9], and it is restricted in the feasible region, meaning that the answer found
respects all the constraints of the problem. Furthermore, metaheuristics are a class of algorithms that can
solve the problems in a reasonable amount of time.

4.1. Differential Evolution


Differential Evolution (DE) is population-based method, stochastic function optimizer which was firstly
introduced by Storn and Price. This meta-heuristic draw attention due to its performance on the First
International Contest on Evolutionary Computation managing to finish at the third place. After that, a
detailed description of the algorithm was presented in [10]. Following the success of this algorithm many
variants derived from the original method [11], and it was applied in a variety engineering problems [12].

4.2. Self-adaptive Differential Evolution


In the classical Differential Evolution there are three control parameters that have to be set by the user,
they are the mutation factor, the crossover rate (CR), and the population size. It is well known that there
are some specific values of CR that make the algorithm converge much faster [13]. Empirical studies
have suggested values to be used in this parameters [10] and [14].
To tackle this time consuming fine tune of the control parameters Qin, Huang, and Suganthan in-
troduced the Self-adaptive Differential Evolution (SaDE) [15]. This algorithm adapts the value of the
mutation factor and the crossover rate. The only parameter left to user is the population size. Moreover,
it has a poll of strategies in the mutation operator that is also adapted to choose the strategy that is more
likely to generate a trial chromosome that will survive to the next generation.

4.3. Objective function


Using the models of power and NOx to simulate the process of power generation and NOx emissions
by the thermal power plant, the objective is to maximize the power, this means adjust the valves of the
combustion chamber to decide which gas demand produces more energy. The restriction of this problem
is the NOx emissions, that must be under a certain limit.
In this research only a specific range of generated power is optimized. In this interval, the gas
demand of the stage B is the value as the gas demand of the stage A. Hence, the optimization procedure
only manipulates the stages A, C, and pilot.

211 2
5. Results
This chapter brins the results of the power generation optimization. To run the SaDE optimization,
the parameters setup were selected empirically and the best results were with a population size of 10
chromosomes with a maximum of 1000 generations.
For the models simulation it has been defined a variation of temperature and different values to the
restriction limit of NOx emission. Each simulation setup was executed 10 times and the best fitness was
considered. The search space for the combustion chamber stages were set from 0 to 4.5 kilograms per
second of natural gas for the stages A and B, from 0 to 2.5 kilograms per second for stage C, and from
0 to 2.0 kilograms per second for the pilot stage. This is the physical limitation of gas demand in the
stages.

5.1. Obtained Results


The best result of 10 executions for each scenario is shown in the from Table 2 to Table 3, where A, B, C,
and Pilot are the turbine stages, GD is the gas demand which is the sum of gas in the four stages and AT
is the ambient temperature of the simulation parameter, and OP is the output power in MW optimized by
the optimization procedure. In Table 3 NOx is also specified since no restriction has been set.

Tabela 1: Optimal stages configuration for 20 ppm%02 NOx restriction


A B C Pilot GD AT OP
3.41 3.41 0.35 0.62 7.80 18 116.81
3.31 3.31 0.34 0.65 7.61 23 113.43

Tabela 2: Optimal stages configuration for 25 ppm%02 NOx restriction


A B C Pilot GD AT OP
4.36 4.36 0.19 0.45 9.36 18 152.50
4.08 4.08 0.29 0.46 8.91 23 140.09

Tabela 3: Optimal stages configuration for no NOx restriction


A B C Pilot GD AT OP NOx
4.36 4.36 0 0.61 9.33 18 158.51 30.17
4.36 4.36 0 0.61 9.33 23 158.51 33.19

From the results it is possible to notice that if the NOx emissions is more restricted, meaning that
less pollutant gases should be emitted, the gas turbine will produce less power. Without any emission
restriction is where the turbine can produce more energy, as shown in Table 3, although it exceeds the
legal maximum proposed by Brazilian authorities. For a NOx restriction of 20 PPM%O2 the efficiency of
the turbine will be reduced on average 21%, Table 1. Finally, the environment temperature affects directly
the NOx emissions, thus an increase in the ambient temperature incurs a decrease on the maximum power
generated by the turbine in due to emissions constraint, as shown in 1 and 2.

6. Conclusion
The present research has provided an experimental study on the power generation of a heavy duty gas
turbine taking into account environmental restrictions. The optimization method applied in this expe-
riments has shown to be appropriate, hence the convergence to the optimal value has occurred with a
small population and few generations. Moreover, the usage of this optimization method has reduced the
amount of computer processing due to the good parameters setup. All results found for different environ-
ment temperature, desired output power, and NOx emissions limits are within the legislation restrictions

212 3
and are the optimal adjustment for the gas demand of each stage of the combustion chamber. It was
also presented an hypothetical results with NOx emissions restriction slightly below the current Brazilian
regulation to demonstrate that the simulations can cope well with different scenarios of the emissions
limits regulations.

Referências
[1] M. T. Tolmasquim. Energia Termeletrica: Gas natural, Biomassa, Carvao, Nuclear. Rio de Ja-
neiro: EPE - Empresa de Pesquisa Energetica, 2016.
[2] Nnamdi I. Nwulu and Xiaohua Xia. Multi-objective dynamic economic emission dispatch of elec-
tric power generation integrated with game theory based demand response programs. Energy Con-
version and Management, 89(Supplement C):963 – 974, 2015.
[3] J. H. Talaq, F. El-Hawary, and M. E. El-Hawary. A summary of environmental/economic dispatch
algorithms. IEEE Transactions on Power Systems, 9(3):1508–1516, Aug 1994.
[4] Brasil Conama. Dispoe sobre padroes de qualidade do ar, previstos no pronar. Diario Oficial da
Uniao, 22/08/1990, 1990.
[5] Manuchehr Irandoust. The renewable energy-growth nexus with carbon emissions and technologi-
cal innovation: Evidence from the nordic countries. Ecological Indicators, 69(Supplement C):118
– 125, 2016.
[6] Chien-Cheng Lee, Pau-Choo Chung, Jea-Rong Tsai, and Chein-I Chang. Robust radial basis func-
tion neural networks. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics),
29(6):674–685, Dec 1999.
[7] Sébastien Bubeck. Convex optimization: Algorithms and complexity. Found. Trends Mach. Learn.,
8(3-4):231–357, November 2015.
[8] Ilhem Boussaïd, Julien Lepagnot, and Patrick Siarry. A survey on optimization metaheuristics.
Information Sciences, 237(Supplement C):82 – 117, 2013. Prediction, Control and Diagnosis using
Advanced Neural Computations.
[9] Ibrahim H. Osman and Gilbert Laporte. Metaheuristics: A bibliography. Annals of Operations
Research, 63(5):511–623, Oct 1996.
[10] Rainer Storn and Kenneth Price. Differential evolution – a simple and efficient heuristic for global
optimization over continuous spaces. Journal of Global Optimization, 11(4):341–359, Dec 1997.
[11] S. Das and P. N. Suganthan. Differential evolution: A survey of the state-of-the-art. Trans. Evol.
Comp, 15(1):4–31, February 2011.
[12] Kit Po Wong and Zhao Yang Dong. Differential evolution, an alternative approach to evolutionary
algorithm. In Proceedings of the 13th International Conference on, Intelligent Systems Application
to Power Systems, pages 73–83, Nov 2005.
[13] F. Penunuri, C. Cab, O. Carvente, M.A. Zambrano-Arjona, and J.A. Tapia. A study of the classical
differential evolution control parameters. Swarm and Evolutionary Computation, 26(Supplement
C):86 – 96, 2016.
[14] J. Ronkkonen, S. Kukkonen, and K. V. Price. Real-parameter optimization with differential evolu-
tion. In 2005 IEEE Congress on Evolutionary Computation, volume 1, pages 506–513 Vol.1, Sept
2005.
[15] A. K. Qin, V. L. Huang, and P. N. Suganthan. Differential evolution algorithm with strategy
adaptation for global numerical optimization. IEEE Transactions on Evolutionary Computation,
13(2):398–417, April 2009.

213 4
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

THREE-DIMENSIONAL FINITE ELEMENT MODEL TO PREDICT FLEXURAL BEHAVIOR OF


REINFORCED CONCRETE BEAMS

Gabriela Bandeira de Melo Lins de Albuquerque1, Valdir Pignatta e Silva2, João Paulo
Correia Rodrigues3
1
University of São Paulo – Brazil and University of Coimbra – Portugal, gabriela.lins@usp.br.
2
University of São Paulo – Brazil, valpigss@usp.br.
3
University of Coimbra – Portugal, jpaulocr@dec.uc.pt.

This paper presents the basis for the conception of a three-dimensional finite element model capable of
simulating the flexural behavior of simply supported reinforced concrete beams. The modeling was
performed from the software DIANA [1], considering the characteristics of the full scale beams that
were submitted to four-point bending tests by the authors of this research in the Laboratory of Testing
Materials and Structures of University of Coimbra, Portugal. Besides the concrete beam itself, the
geometry of the model has the steel plates used to distribute the applied loads, the blocks of the roller
and pinned supports and the longitudinal and transverse reinforcements (Figs 1 and 2). Crack
formation and propagation was incorporated from a rotating crack model [2], that is based on smeared
cracking approach, initially proposed by [3]. The crack bandwidth was automatically determined by
the software with the use of Rots’ element based method [4]. The effects of material softening, as well
as the concrete stiffening between cracks, were other phenomena included to represent its nonlinear
behavior. Both the exponential [5] and the parabolic models [6], adopted to mimic the response of
concrete to tensile and compressive stresses, respectively, were derived from the principles of fracture
mechanics. The compressive strength and the density of the concrete were defined by experimental
tests, while its other mechanical properties were calculated from [7, 8]. The steel of the loads
distribution plates and of the support blocks was characterized from an isotropic linear-elastic
constitutive model with mechanical properties calculated from [9]. An elasto-plastic stress-strain
diagram with hardening was chosen to simulate the behavior of the reinforcements, for which some
mechanical properties were again measured by experimental tests and the others defined according to
the standard [10]. The 3D model was discretized with a mesh of a twenty-node isoparametric solid
brick elements with three degrees of freedom that are based on quadratic interpolation and Gauss
integration, applying a 3 point integration scheme to solve the stiffness matrix of the element. Those
elements are the most recommended for the analysis of concrete structures as they can describe a
greater number of deformation modes [1, 11]. The reinforcements were embedded in the mesh of the
"mother elements" and considered as uniaxial bar elements. Therefore, they didn't have degrees of
freedom of their own and their mechanical properties were calculated only in the longitudinal
direction, in a 2 point integration scheme. Perfect bond between the reinforcement and the surrounding
material was considered. An element size of 40 x 40 x 40 mm was defined after a mesh sensitivity
analysis. It was verified that this FE mesh presented a good equilibrium between accuracy and
efficiency. The behavior of the reinforced concrete beam was evaluated from a nonlinear structural
analysis, in which the equations related to the problem were iteratively solved by the Newton-Raphson
Regular method with the tangent stiffness matrix being derived for every iteration. The load increment

214
applied to the beam was performed by force control, with relatively small load steps (1 kN). Due to the
high degree of nonlinear phenomena involved and the feasible difficulty to converge some steps, the
arc length control was activated for some critical stages of the analysis, such as the beginning of the
concrete cracking. The convergence criterion was controlled by the displacement norm, assuming a
tolerance equal to 5 x 10-2. In order to increase the convergence rate and to maintain stable solutions,
the line search algorithm was also applied [12, 13]. The model was validated by comparing predicted
to experimental data. Good agreement was verified for the curves of applied loads vs deflections of the
beam (Fig 3). The crack fields, including the crack widths, also managed to be quite representative of
those measured during the experimental tests (Fig 4). The flexural failure modes were achieved by
means of the numerical model too, as can be seen in Fig 5. Thus, the consistency of the hypotheses
adopted for the numerical modeling presented in this research was demonstrated, since it was able to
represent in a satisfactory and accurate way the flexural behavior of simply supported reinforced
concrete beams.

Fig 1 Geometry of the beam, of the steel plates used to Fig 2 Detail of the longitudinal and transverse
distribute the applied loads and of the supports blocks reinforcements embedded in the geometry of the beam

45
40
35
30
P [kN]

25
20
15 A_ka0_kr0_1
10 A_ka0_kr0_2
5 A_ka0_kr0_3
DIANA
0
0 10 20 30 40 50 60 70 80 90 100 110 120 130
dv,s2 [mm]
Fig 3 Comparison between experimental and numerical curves of applied loads vs beam deflection

215
Fig 4 Crack field obtained by DIANA, with the values of crack width Ecw1, compared to the one measured during
experimental tests for 25 kN of total applied load

Fig 5 Comparison between beam deformed shapes for the maximum deflection measured during tests

References

[1] DIANA FEA. User’s manual: release 10.1. Delft: DIANA FEA BV, 2016a. 13 v.
[2] Cope R. J. et al. Modelling of reinforced concrete behaviour for finite element analysis of bridge
slabs. In: Numerical methods for nonlinear problems 1. Swansea: Pineridge Press, 1980. p. 457-470.
[3] Rashid Y. R. Ultimate strength analysis prestressed concrete pressure vessels. Nuclear
Engineering and Design, v. 7, n. 4, p. 334-344, 1968.
[4] Rots J. G. Computational modeling of concrete fracture. 1988. 132 p. PhD Thesis, Delft University
of Technology, Delft, 1988.
[5] Vecchio F. J., Collins M. P. The modified compression-field theory for reinforced concrete
elements subjected to shear. ACI Journal, v. 83, n. 22, p. 219-231, 1986.

216
[6] Feenstra P. H. Computational aspects of biaxial stress in plain and reinforced concrete. 1993. 151
p. PhD Thesis, Delft University of Technology, Delft, 1993.
[7] European Committee for Standardization. EN 1992-1-1: Eurocode 2: design of concrete structures
- part 1.1: general rules and rules for buildings. Brussels: CEN, 2004. 225 p.
[8] Comité Euro-International du Béton. CEB-FIP MC 2010: fib Model Code for concrete structures
2010. 2011. 653 p.
[9] European Committee for Standardization. EN 1993-1-1: Eurocode 3: design of steel structures -
part 1.1: general rules and rules for buildings. Brussels: CEN, 2005. 91 p.
[10] European Committee for Standardization. EN 1992-1-2: Eurocode 2: design of concrete
structures - part 1.2: general rules - structural fire design. Brussels: CEN, 2004. 97 p.
[11] Belletti B. et al. Guidelines for nonlinear finite analysis of concrete structures. Utrecht:
Rijkswaterstaat Centre for Infrastructure, 2016. 66 p. (Rijkswaterstaat Technical Document RTD:
1016-1:2016).
[12] Crisfield M. A. Accelerated solution techniques and concrete cracking. Computer Methods in
Applied Mechanics and Engineering, v. 33, n. 1-3, p. 585-607, 1982.
[13] Schweizerhof K. Consistent concept for line search algorithms in combination with arclength
constraints. International Journal for Numerical Methods in Biomedical Engineering, v. 9, n. 9, p.
773-784, 1993.

217
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

MULTISCALE ANALYSIS OF MULTIPLE CRACKS IN AIRCRAFT FUSELAGE

T. A. A. Oliveira1, G. Gomes2, F. Evangelista Junior3, A. M. Delgado Neto4

University of Brasilia
1
eng.thiagoarnaud@gmail.com
2
ggomes@unb.br
3
fejr.unb@gmail.com
4
alvaro_fausto@hotmail.com

1. Introduction

This paper shows an analysis of multiple cracks in an aircraft fuselage through the automation in
software specialized to evaluate the cracks propagation - BemCracker2D [1, 2]. The first analysis
refers to the macro element in which is performed a rivet modeling at the software BemLab2D [3] and
analyzed via Dual Boundary Element Method the SIF at BemCracker2D. With the SIF, the stresses in
a micro element are then calculated. This micro element is composed by two pre-established initial
cracks and a circular hole. As a result, there is relationship between fatigue life (number of load
cycles) and compliance of the edges of this micro element at each crack increment. The main objective
is evaluating the compliance of micro elements under several levels of loads, cracks, and holes. And as
specific objectives: obtaining edge displacements at each crack propagation and the number of loading
cycles for each increment.

2. Material and Methods


The methodology of this paper consists of the analysis via Dual Boundary Elements Method of fatigue
life of aircraft fuselage with two cracks pre-established and a circular hole. For this, an aircraft
fuselage panel model, representing the macro analysis, was automated in order to obtain elastic
stresses in the regions of these voids.

2.1. Macro model analysis


BemLab2D was used to create the boundary element mesh to represent the numerical model (see
Figure 1), and the BemCracker2D to compute the SIF.

Figure 1: Boundary element mesh of the macro model with BemLab2D.

218
SIF values are listed in Table 1.
Analytical Numerical
Tip 1 Tip 2 Tip 3 Tip 4 Tip 5 Tip 6
0.07092803 0.07155231 0.07155169 0.07156367 0.07156367 0.07155169 0.07155231
Table 1: Analytical and numerical SIF results.
These results therefore present the purpose of corroborating the model and the subsequent analysis to
be performed.
2.2. Micro model analysis
With the SIF computed, the stress field at the crack tip is calculated through the Eqs. (1), (2) and (3)
according to [4]. To the tip 1 in Figure 1 (KI=0.07155231).
𝐾𝐾𝐼𝐼 𝜃𝜃 𝜃𝜃 3𝜃𝜃
𝜎𝜎𝑥𝑥 = cos �1 − sin sin � (1)
√2𝜋𝜋𝜋𝜋 2 2 2
𝐾𝐾𝐼𝐼 𝜃𝜃 𝜃𝜃 3𝜃𝜃
𝜎𝜎𝑦𝑦 = cos �1 + sin sin � (2)
√2𝜋𝜋𝜋𝜋 2 2 2
𝐾𝐾𝐼𝐼 𝜃𝜃 𝜃𝜃 3𝜃𝜃
𝜏𝜏𝑥𝑥𝑥𝑥 = sin �cos cos � (3)
√2𝜋𝜋𝜋𝜋 2 2 2
where KI is the SIF, r the distance to the tip and θ the angle.
Figure 2 shows the stress field near the crack tip for three different angles used to the analysis in the
micro element.

(a) (b) (c)


Figure 2: Stress field for different angles θ
a) θ=0° b) θ=45° c) θ=90°

As near the crack in the case of the LEFM the stresses tend to infinity near the crack tip, in order to
calculate the maximum ones in the elastic stage it was used r as the distance of the Irwin’s plastic zone
(2𝑟𝑟𝑝𝑝 ).
The stresses field values are listed in Table 2.

Angle σx σy τ
θ=0° 239.00 239.00 0.00
θ=45° 142.74 298.88 32.34
θ=90° 84.50 253.50 -84.50
Table 2: Stresses Fields

219
The micro model is highlighted in Figure 3, which shows two pre-established cracks and a central
circular hole. For the analysis, a BemLab2D model of a 1 cm square side element was adopted with
the stresses calculated in Eqs. (2) to (4) and displacement constraint on the left and lower sides, with a
central hole of radius 0.1 cm and two 0.1 cm cracks inclined 45°.

Figure 3 - Model of the Micro Element.


3. Results

According to each increment there are the points for the construction of the fatigue life curve (N) x
Compliance in (m/N) of both the right and upper edge, shown in Figures 4, 5 and 6 for θ=0°, 45° and
90°, respectively. It can be seen that as the crack propagates (number of cycles) compliance increases.
This is due to the fact that the edge displacement rates increase gradually due to the loss of rigidity of
the plate with the crack increments, so that when the number of cycles reaches about 106, the edge
displacements are already quite amplified, resulting in high compliance. It is also seen Figure 4 that
due to the symmetry of the plate layout, the graph of Crack 1 Tip 1 is equal to that of Crack 2 Tip 2
and that of Crack 1 Tip 2 is equal to that of Crack 2 Tip 1.

Figure 4: Load cycles x Compliance for θ=0°


4. Conclusions

Regarding the use of software (BemLab2D and BemCracker2D), the results found for SIF validation
were very close to the analytical ones with identical values up to the second decimal place. Therefore,
it can be concluded that the use of these were satisfactory resulting in output data as expected.

220
Figure 5: Load cycles x Compliance for θ=45°

Figure 6: Load cycles x Compliance for θ=90°


5. Acknowledgement

The authors are grateful to the Brazilian National Research Council (CNPq), to the Brazilian
Coordination for the Improvement of Higher Education (CAPES) for the supporting funds for this
research and to Federal District Research Support Foundation (FAP-DF).
6. References

[1] Gomes, G. "Data Structure for Representing Bidimensional Models of Boundary Elements". MSc.
Thesis, Brazilia University, Department of Civil and Environmental Engineering, Brazilia, 2000.
[2] Gomes, G. "Aplication of BEM and DRM in in Object Oriented 2D Plasticity Problems". DSc.
Thesis, Brazilia Univeristy. Department of Civil and Environmental Engineering, Brazilia, 2006.
[3] Delgado Neto, A. "BEMLBA2D: Graphical Interface for Modeling, Visualization and Analysis
with Boundary Elements - An Application in Elastomeric Problems". MSc. Thesis, Brazilia University,
Department of Civil and Environmental Engineering, Brazilia, 2017.
[4] Anderson, T. L., 2005. Fracture Mechanics: Fundamentals and Applications. CRC Press. Boca.

221
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

IDENTIFICATION OF PRINCIPAL RESEARCH TOPICS STUDIED IN ENGINEERING IN BRAZIL

Jether Gomes1, Thiago M. R. Dias¹, Gray F. Moita1


1
CEFET-MG, jetherog@gmail.com
1
CEFET-MG, thiago@div.cefetmg.br
1
CEFET-MG, gray@dppg.cefetmg.br

¹CEFET-MG - Federal Center for Technological Education of Minas Gerais Av. Amazonas, 7675,
Nova Gameleira, 30510-000 Belo Horizonte, MG, Brazil

Abstract
The publication of scientific papers has increased significantly in recent years. Consequently, several
studies on scientific production data have been carried out in order to analyze what has been
researched. Since the understanding of how research has evolved can, for example, serve as a basis for
the construction of scientific policies. Thus, the objective of this work is to identify and analyze the
main topics of research published between 1962 and 2016 in congress annals and journals by Brazilian
PhD researchers who work in engineering.

Keywords: Keywords. Research topics. Bibliometrics. Lattes Platform.

Introduction
The internet and its services have been instrumental in the impressive growth in the number of
publications of scientific articles in recent decades. According to Dias [9], the great ease of accessing
the articles made available on the Internet contributes directly to the expansion of knowledge. On the
other hand, according to Yi and Choi [8], understanding about the evolution of scientific production
can promote new advances in science. In Brito et. al. [1], the authors point out that works of this
nature are considered urgent in Brazil and can portray what is developed and published in science,
technology and innovation, making it possible to generate parameters to guide efforts and investments
in order to boost research results.
As a consequence of this, there is a global growth in the number of researchers from all areas to
extract knowledge about what has been developed in science ([9]; [6]; [5]; [10]; [11]). At this point, it
is important to emphasize that the work can be carried out under different perspectives, such as:
productivity indicators, scientific collaboration and analysis of research topics. In the latter case, a
topic can be understood as a term (or descriptor) that represents one of the subjects associated with a
particular document.
At the national level, however, very few studies are doing research topics, and when they do, they
usually use a restricted amount of data that comes from specific international repositories, and
therefore can not represent what is produced in the Brazil. Thus, analyzing data from repositories that
encompasses several types of publications, especially in national vehicles and in several areas of
knowledge, becomes a relevant task for understanding the evolution of Brazilian science. In this
context, Ferraz et. al. [7] points out that the large volume of data in the curriculum repository of the
Lattes Platform represents the most important instrument for studies on Brazilian scientific production.

222
One of the most interesting approaches for studies of research topics is the analysis of the
keywords of a set of scientific publications, since they are carefully inserted by their respective authors
and provide a possibility to describe the main subjects that permeate the work ([8]; [3]). Thus, in the
current work, keywords of scientific articles are also referred to as research topics.
In addition, for Pritchard [2], bibliometrics stands out as one of the main metric sciences of
content analysis and can be applied to scientific data repositories in order to obtain quantitative
information about publications. While in Dias [9], the author emphasizes that with the use of
bibliometrics it is possible to identify the trends and growth of scientific knowledge in several areas,
observe the dispersion of scientific knowledge, support investment policies and understand how
scientific developments occur.
Therefore, this study aims to analyze all keywords of scientific articles referring to large area
engineering, published in congress annals and journals between 1962 and 2016 by the Brazilian PhD
researchers who have curricula registered in the Lattes Platform. The study contemplates bibliometric
analyzes of the keywords in order to identify and map the main research topics in engineering by
Brazilian researchers.

Development
The motivation of choosing the Lattes Platform as a source of information is basically related to
four factors: (1) data are available on the internet and not widely analyzed ([6]); (2) to deal with the
integration of scientific production data of all the areas of S & T that exist in Brazilian science along
the whole trajectory; (3) for not neglecting articles published in national journals that are often not
indexed, as well as articles in congress annals ([9]); and, (4) as a powerful source for providing high
quality data to measure and measure national academic performance.
The acquisition of the PhD's curricula in the XML version was done through the use of
LattesDataXplorer, developed by Dias [9] to collect the scientific data contained in the curricula
registered in the Lattes Platform. Subsequently, a framework of components was developed
responsible for processing the data of the curricula in order to support the desired analyzes (Figure 1).
In it, the components "data filtering" and "data treatment" are responsible for the entire process of
selecting, treating and modeling information from the curricula that actually need to be processed to
achieve the proposed objectives, and, at the same time, computational processing.

Figure 1. A set of components developed for filtering and processing data.

223
The "filtering" component performs the mining step in the curricula to extract information from
the articles, storing them separately in a scientific publications archive, thereby defining the core data
set to be studied. Information from the articles include: curriculum identifier; large area of publication;
year of publication; type of publication; title and keywords. For this work, it was considered all the
publications in congress annals and in journals that have as great area, the engineering. On the other
hand, the "treatment" component processes data from the archive of scientific publications to treat and
characterize them and builds from them a set of standard results files to facilitate analysis. This
component basically performs three steps, namely: cleaning and grouping the data, normalizing the
data and building the files.
The cleaning and grouping the data step is for processing the keywords in such a way as to
exclude terms that do not represent study topics. It also groups distinct keywords that have the same
semantic value. For this, initially, the developed method obtains the keywords of each article analyzed.
Then, each of the keywords are associated with the language registered for the article, serving as
reference in the process of radicalization. The lowercase process converts all words to lowercase in
order to standardize the set. In the process of stopWords are removed the terms that do not have
semantic values. Subsequently, in the normalization process all accents and scores are taken from the
keywords. Finally, the process of radicalization is responsible for reducing the keyword to its radical.
However, in the case of composite keywords, this process is executed on each term individually, and,
concatenated in a single word.
At the normalizing the data stage, all the articles in the scientific publications archive are
processed to identify and process the collaborative works. This procedure is performed for data from
the same scientific article not being considered more than once during the analyzes. In order to do so,
it was necessary to adapt the ISCooll method proposed by Dias and Moita [11] to identify high-
volume scientific collaborations with linear cost data. Thus, while the original ISCooll method uses a
dictionary to link articles (keys of a dictionary) to their authors (curriculum identifiers), the adapted
ISCooll method adopts a dictionary to link the titles of articles to the union set of their keywords.
Finally, the building the files is intended to facilitate the analyzes and contribute to the reduction
of the data to be processed. For more comprehensive analyzes, the standard results files were divided
into: files that consider works done in co-authoring (processed through cleaning and normalization
steps) and those containing the entire set of articles (processed only by the cleaning step). Therefore,
three files of the following formats were constructed: (1) Individuals file by keywords (Curriculum
identifier, keyword1, ... keywordN); (2) Keyword archive by year and large area (Keyword, year,
frequency and large area) and (3) Keyword frequency file by year and publication quantity (Year,
keyword frequency, amount of publication and large area).

Results and Discussions


Data were collected in April 2017, totaling 265,170 PhD's curricula, where 23.437 of these
curricula have engineering as their main area of expertise. For the analyzes were considered all the
unique articles (collaboration) published by the doctors in congress annals and in journals between
1962 and 2016 referring to the large area of engineering, totaling 906.775 articles and 2.626.707
keywords. In addition, it should be noted that most of the curricula were recently updated, with 47%
having a date last updated in 2017 and 72% updated in the last two years. In spite of this, it is worth
mentioning that, according to Dias [9], the non-updating of the rest of the curricula may have varied
and difficult to recognize reasons.
In continuation, to know the topics that are the main research topics in the last 55 years of
engineering research in Brazil, all the keywords processed were ranked according to the importance of
a topic based on frequency. Table 1 presents the top 15 research topics studied in engineering in
Brazil.

224
Table 1: Main topics of study used by individuals who work in Engineering.
Rank 1962-2016 Frequency
1 Composites 6.088
2 Optimization 5.535
3 Concrete 5.300
4 Simulation 4.994
5 Corrosion 4.937
6 Recycling 4.388
7 Biodiesel 4.360
8 Modeling 4.155
9 Ergonomics 3.977
10 Biomaterial 3.958
11 Ceramics 3.852
12 Mechanical Property 3.755
13 Finite elements 3.715
14 Solid Waste 3.611
15 Adsorption 3.579

Finally, it is emphasized that this type of analysis is characterized as an important mechanism,


since it makes it possible to identify the most impacting topics within a research community in each
epoch. When analyzing the keywords of the curricula registered in the Lattes Platform it is possible to
consider publications made in congresses annals, which is not feasible when verifying international
data repository or in non-indexed journals. Therefore, it is possible to obtain a more precise view of
the topics most investigated over time by the Brazilian researchers in the areas of engineering and,
from this, to perform analyzes about phenomena and tendencies to extract knowledge to be used in the
most varied types of decision in the scientific field. However, despite the validity of the results, it is
noticed that the analysis of the keywords using only measure of popularity importance based on
frequency is influenced quantitatively, which, in the case of the Lattes Platform that has articles
published in the most varied conferences and periodicals with different levels of qualities, only
quantitative analyzes can lead to misperceptions. Thus, as future studies, it is expected that analyzes
will be performed considering temporal factors to determine the relevance of a topic in each epoch,
analyzes based on social network techniques to determine the most central keywords, and qualitative
characteristics appropriate to the context of the Lattes Platform.

Referências
[1] A. G. C. de Brito, L. Quoniam, and J. P. Mena-Chalco, “Exploração da plataforma lattes por assunto:
proposta de metodologia,” Transinformação-ISSN 2318-0889, vol. 28, no. 1, 2016.
[2] A. Pritchard, “Statistical bibliography or bibliometrics,” Journal of documentation, vol. 25, pp. 348, 1969.
[3] G. F. Khan and J. Wood, “Information technology management domain: Emerging themes and keyword
analysis,” Scientometrics, vol. 105, no. 2, pp. 959–972, 2015.
[4] J. Choi, S. Yi, and K. C. Lee, “Analysis of keyword networks in mis research and implications for predicting
knowledge evolution,” Information & Management, vol. 48, no. 8, pp. 371–381, 2011.
[5] J. P. Mena-Chalco, L. A. Digiampietri, F. M. Lopes, and R. M. Cesar, “Brazilian bibliometric coauthorship
networks,” Journal of the Association for Information Science and Technology, vol. 65, no. 7, 2014.
[6] L. Digiampietri, “Análise da rede social brasileira,” Ph.D. dissertation, School of Arts, Sciences and
Humanity, University of São Paulo (USP), 2015.
[7] R. R. N. Ferraz, L. M. Quoniam, and E. A. Maccari, “The use of scriptlattes tool for extraction and on-line
availability of academic production from a department in management,” in CONTECSI, vol. 17, 2014.
[8] S. Yi and J. Choi, “The organization of scientific knowledge: the structural characteristics of keyword
networks,” Scientometrics, vol. 90, no. 3, pp. 1015–1026, 2012.
[9] T. M. R. Dias, “Um estudo da produção científica brasileira a partir de dados da plataforma lattes,” Ph.D.
dissertation, CEFET-MG, 2016.
[10] T. M. R. Dias, G. F. Moita, P. M. Dias, and T. H. J. Moreira, “Caracterização de redes científicas de dados
curriculares,” iSys-Revista Brasileira de Sistemas de Informação, vol. 7, no. 3, pp. 5–18, 2014.
[11] T. M. R. Dias and G. F. Moita, P. M. Dias, “A method for the identification of collaboration in large
scientific databases,” Em Questão, vol. 21, no. 2, pp. 140–161, 2015.

225
CILAMCE 2018 CONGRESS
11 – 14 NOVEMBER 2018, PARIS/COMPIÈGNE, FRANCE

The Fuzzy Logic Technique for Emulation of the Decision-making


Process during Human Movement Simulation

H.C. Braga1, G.F. Moita2, Paulo E.M. Almeida3


1
CEFET-MG, henriquecbragaseg@gmail.com
2
CEFET-MG, gray@dppg.cefetmg.br
3
CEFET-MG, pema@lsi.cefetmg.br

Abstract. Not rarely the movement of people is linked to the flow behaviour of particles, or even to a
behaviour similar to that found in fluid dynamics. Thus, the modelling of the movement can benefit
from the knowledge already established for the flow of particles and fluids. However, in addition to the
dynamic aspects already established, there are, concurrently in moving people, human factors, such as
cognitive and social phenomena. Thus, for most adequate models, is necessary consider these human
aspects. In previous work was used the model Fuga to simulate the evacuation process of some
environments. This model uses the Fuzzy Logic for emulation of decision-make human process. With
the fuzzy logic, it is possible work with qualitative and quantitative variables within the same process,
which is very useful to work with ergonomics aspects. However there are no commercial software
using fuzzy logic for simulate human movement, and even in academic contest there are very few.
Why? In this work will be detailed described how the Logic Fuzzy can be used in the context of human
movement simulation. The mainly advantages are specified, and examples will be showed. However
the disadvantages will be commented too. The mainly is probably the required time for a complete
fuzzy processing. In the simulation, each agent will pass for a complete fuzzy process several times
each second. The fuzzy logic is a process relatively sophisticated, and considering that in simulations
can participate hundreds, or until thousands, of agents, the processing time is an important factor to be
considered. However was developed a technique capable of improve the processing time for fuzzy logic
named Fuzzy Matrix. An example will be showed where the use of Fuzzy Matrix became the system
about 140 times faster. Thus the fuzzy logic appears a very promiser instrument to be used in pedestrian
dynamic.
Keywords: Fuzzy Logic; Human movement; Decision-make process; Multi-agent simulation.

1. Introduction
The movement of people sometimes is linked to the flow behaviour of particles, or even to a behaviour
similar to that found in fluid dynamics. Thus, the modelling of the movement can benefit from the
knowledge already established for the flow of particles and fluids. However, in addition to the dynamic
aspects already established, there are, concurrently in moving people, human factors, such as cognitive
and social phenomena. Thus, for most adequate models, is necessary consider these human aspects.
In previous works the model Fuga, developed to simulate the human movement in some environments,
both in situation of evacuation during an emergency [1, 2] and in bidirectional flows [3], was used. This
model uses Fuzzy Logic for the emulation of human decision-making process. With fuzzy logic, it is
possible to work with qualitative and quantitative variables within the same process, which is very
useful to deal with ergonomics aspects. However, there are no commercial software using fuzzy logic to
simulate human movement, and even in the academic context there are very few available. Thus, this
work describes how fuzzy logic can be efficiently used in the context of human movement simulation.
2. Fuzzy Decision-Making Process

226
One person located at any position of a given environment needs to take a decision in order to choose
the best route to a target (usually the nearest exit). Several variables are taken into account in this
process and different route options are available. For each route option, each person will calculate its
“quality”, named route quality (RQ). The higher the RQ value for a certain route, the more appropriate
is the choice of that route. Hence:
RQk = ƒFuzzy (inputsk) (1)
RQke = max (RQk) (2)
ke  k RQke RQk (3)

Where:
inputs = input variables;
ƒFuzzy = fuzzy system function response;
k = movement options;
ke = the option of movement effectively chosen;
RQk = each of the RQs for each of the k move options calculated by the fuzzy system; and
RQke = RQ of the effectively chosen route.

In Fuga v. 1.0, the input variables were AD, WE, PR, SL, IE and AE. Figure 1 shows the representation
of this adopted fuzzy system.

Figure 1: Representation of a fuzzy system for emulating human decision-making process [2].
Hence:
- AD (Apparent Distance): AD is related to the distance of the agent to the exit or target obtained by
comparing the value of the distance map [4] of the location of the agent with the value of the distance
map of the neighbouring cell under analysis. The AD values will then be:
AD = 0, if in the comparison the apparent distance increases;
AD = 0.5, if it remains the same apparent distance; and
AD = 1, if the direction checked indicates a cell with less apparent distance than the current occupied.
- WE (Wall Effect): it is related to the search for the comfort zone [5]. Associated with the model,
there is an algorithm that verifies if there is a wall around the cell occupied by the agent. Thus, the WE
values will be:
WE = 0, if the neighbouring cell of the agent in question is also immediate neighbour to a wall (great
repulsive effect);
WE = 0.5, if the distance from any wall is of two matrix elements (each matrix element have 9 cm by
9 cm); and
WE = 1, if position is three or more matrix elements away from any wall (no repulsive effect).
- PR (Preferential Route): PR is obtained by comparing the direction of movement indicated by the
value of the escape route matrix of the cell of agent, with the effective direction of movement that the
agent would do to this neighbouring cell in verification. The PR values will then be:
PR = 0, if direction of movement of the agent to neighbouring cell in analysis is opposite to the
direction indicated by the cell of the matrix route of escape occupied by the agent;
PR = 0.5, if direction indicated is orthogonal; and
PR = 1, if the direction checked coincides with the indicated direction.

227
- SL (Stress Level): before each simulation (input parameter) the level of stress of the each agent (low,
medium, high or very high) was determine. Thus, based on the U-inverted hypothesis [6] and
depending on this previously defined stress level, the SL values will be:
SL = 0, for medium stress situations;
SL = AD, for situations of low or high stress; and
SL = rand (1), for situations of very high stress.
Additionally, SL also, regardless of its initial value, will have a random component. Thus, at each
iteration, the SL will have a probability (usually 5%) of having a random value.
- IE (Inertia Effect): IE is obtained by comparing the direction of movement that the agent would make
if it moved to this neighbouring cell with the last direction of movement. The IE values will then be:
IE = 0, if the direction is opposite to last performed;
IE = 0.5, if it is orthogonal to the last performed; and
IE = 1, if movement maintains the same direction of the last performed.
- Automata Effect (AE): there is an algorithm that verifies if there is another agent around the cell
occupied by the automaton associated with the model. Thus, the AE values will be:
AE = 0, if there is a neighbouring agent up to a distance of six matrix elements in the considered
direction; and
AE = 1, if the direction is not obstructed by agents.
For example, Figure 2 shows a sequence of images of an evacuation simulation with 400 persons in a
square environment of 200 m2 with two exits using this fuzzy system to choice the best route until the
nearest exit.

Figure 2: Sequence of images of an evacuation with a fuzzy decion-make process [2].

3. Enhancing The Fuzzy Performance


As showed, the fuzzy system can be used in order to emulate the decision-make process in a human
movement simulation. However, one of the factors that may be contributing to the difficulty of
applying fuzzy logic is its processing time. A fuzzy system encompasses a series of operations, such as
scalar to fuzzy conversion, all inference machine processing and its association to the rule bank, as
well as the final fuzzy conversion to scalar.
This whole process has some sophistication, therefore a high computational cost. Considering an
application in a process control, for example, the computational time spent for a complete cycle of a
fuzzy decision making technique may, in practice, be considered negligible, so it is not a limitation to
its use. However, multi-agent modelling can involve thousands of agents - or even more, where each
agent, individually, may need to perform several complete fuzzy processing cycles every second. In
this case, the processing time becomes a negative point to the use of this tool out of the academic
context.
A possibility would be the direct use of fuzzy data, for example, in the form of a table (FT). Ideally this
FT should stay in the computer's dynamic memory, for the reduction in the search time, which will be a
great advantage. One issue is that input quantities may have continuous variations, that is, some criteria
should be adopted for the generation of FT. This TF would have as many columns as the input and
output quantities. An alternative to simplify this table, and also greatly facilitate the research on, is to
use the Fuzzy Matrix (FM) concept.

228
The FM is an array that has only the values of the output. Each of the input quantities is treated as one
different dimension. Knowing the values in each dimension (input quantities), the output value is
directly obtained. It will be necessary to convert the values of each input quantity of the FT into its
matrix equivalent index.
Table 1. Example of PR conversion from FT to FM.
Fuzzy Table Fuzzy Matrix
Input variable Input value Respective dimension Dimension index
0 1
PR 0.5 PRd 2
1 3

Thus, the RQ for each possible movement option can now be simply obtained reading the desired value
on the FM:
RQk = FM [PRd; ADd; WEd; SLd; IEd; AEd]k (4)
6
To verify the reduction of the processing time, a set of 10 complete fuzzy system by Fuga (random
values of the inputs) were carried out on an ASUS Notebook computer, Intel Core i5 2.53 GHz, 64 bits,
Windows 7, using the program Matlab 7.9.0 (R2009b) in its own programming environment and its
Fuzzy Toolbox package (FIS Editor).
As a result, the effective processing time of obtaining the output fuzzy value was reduced by about 140
times, which is very expressive and no doubt facilitates the expansion of the field of application of the
fuzzy logic. Regarding the quality of the result, the only expected difference is associated to the
rounding of the SL values to a multiple number of 0.1; since the other input quantities did not change.

4. Conclusions
The fuzzy logic technique can be considered a promising alternative tool to be used in human
movement simulation software in order to emulate the decision-make process. To enhance its results,
the use of fuzzy matrix concept can reduced very expressively the processing time without loss in the
quality of the information accurate. This is a valuable finding that contributes for a more efficient use of
the fuzzy logic in multi-agent simulations of complex systems.
Acknowledgements
The authors would like to thanks CEFET-MG, CNPq and CAPES for the financial support received
during this work.
References
[1] H.C. Braga, G.F. Moita, and P.E.M Almeida. Simulation of People Flow by a New Fuzzy Discrete
Automata Model and an Ergonomic Approach, Traffic and Granular Flow 2015, pp. 137-144,
2016.
[2] H.C. Braga, and G.F. Moita. Modeling with Fuzzy Logic the dynamic of people flow during the
evacuation of constructed environments dimensioned according to the Brazilian legislation. In:
11th World Congress on Computational Mechanics, 2014, Barcelona. Proceedings ..., Barcelona:
CIMNE. v. V., p. 6458-6467, 2014.
[3] H.C. Braga, G.F. Moita, and P.E.M Almeida. Simulation with Fuzzy Logic of Human Movement
in Bidirectional Flows with the formation of Lane Formation and Jamming, Abakós, v. 6, n. 1, pp.
19-34, 2017.
[4] P.A. Thompson, and E.W. Marchant. Testing and application of the computer model SIMULEX,
Fire Safety Journal, n. 24, pp. 149-166, 1995.
[5] X. Pan, et al. A multi-agent based framework for the simulation of human and social behaviors
during emergency evacuations, AI and Society, n. 2, v. 22, pp. 113-132, 2007.
[6] M.A. Staal. Stress, cognition and human performance: a literature review and conceptual
framework, NASA, 171 p. TM 212824, 2004.

229
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

FE2 MULTI - SCALE METHOD APPLIED TO SIMULATE COMPOSITE MATERIALS


STRUCTURES IN SUPERCOMPUTERS .

G.Giuntoli 1 , S. Oller 2 , M. Vázquez 3 , G. Houzeaux 4


1 Barcelona Supercomputing Center, guido.giuntoli@bsc.es
2 Universitat Politècnica de Catalunya, sergio.oller@upc.edu
3 Barcelona Supercomputing Center, mariano.vazquez@bsc.es
4 Barcelona Supercomputing Center, guillaume.houzeaux@bsc.es

There is no doubt that composite materials are widely used nowadays in different industries. One of
the main areas of applicability is in the aircraft industry, a clear evidence is the A350 from Airbus which
has more than 50% of its weight in composite materials. These materials allow to build lighter aircrafts
that save fuel and reduce emissions while at the same time has enough strength.
One way of designing composite material structures can be done by simulating the behavior of a
prototype under different working loads, by this way several computational methods can be used (see
Ref. [1]). In this case we focus particularly on the FE2 multi-scale method (FE2) 1 . The main idea
is to create a coarse grid at the macroscopic scale that is solved by FEM and the constitutive average
properties of this macroscopic problem are calculated with a micro-model by FEM calculations too. In
Fig. 1 we represent a basic idea of the FE2 algorithm where the structure is composed of two different
micro-structures and at the time it needs a property (the stress σ) it asks to the micro-models to calculate
it using the strain ε. This last process is called localization of the strain and it is reached through the
imposition of specific boundary conditions at the micro-structure.

ε
h h h

h
h h

h
h h

σ
h h h h h h h

h h
h

h h
h

h h h

ε h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

h
h

σ
h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

Figure 1: Scheme of the FE2 multi-scale method process. In this 2D representation the structure is
composed of two different micro-structures: layers of matrix and circular fibers with different orientation
angles.

This work is divided in two main parts, in the first one we analyze a non-linear composite material
problem using the FE2 method and we compare the solution with other one but applying FEM directly
as a single micro-structure, we consider this last one an exact solution. Once we show that both solutions
are coherent we measure the computational cost and we compare them, we expect the cost of the FE2
to be significantly less than the other. The particular problem that we solve consists on a beam like the
one shown on Fig. 2. For simulating the non-linear behavior, the micro-structure is made of two layers
of different materials, one of them is linear and the other has a plastic behavior.
The main challenges on this problem is that for each integration point at the micro-structure a certain
amount of internal variables should be stored in order to compute the plastic response or non-linear
1 The name FE2 is because the finite element method is applied at two scales: microscopic and macroscopic.

230 1
response. For the case of the linear materials the solution is much more easy to obtain because only one
calculation should be done at the microscopic scale and then the result is reused in all the integration
points of the macroscopic scale. We compare also this last linear case in order to see the accuracy and
the efficiency.

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

h h h h h h h h h h h h h h h

Macro-problem Micro-problem
F

Figure 2: Problem that we are going to solve to test the accuracy and efficiency of the FE2 multi-scale
method.

The code that we prepared for applying the FE2 method consist in a coupling between two codes. For
solving the macroscopic problem we use Alya, a parallel multi-physics code capable of being executed
in distributed architectures which has a solid mechanics module. For the microscopic problem we use
MicroPP, which is able to localize the strains from the macroscopic code and to homogenized stress and
return them to the macro-scale.
The second part of this work is devoted to explain the main design details of the codes and to measure
the performance and trying to figure out the maximum computational problem size that we can solve
using this strategy.
The main bottleneck of the algorithm resides on the microscopic calculations because all the integra-
tions points in the macro-structure should get the average properties to assemble the matrix and the RHS
of the macro-problem. By this way the macroscopic problem cannot calculate until all the microscopic
problems would have being solved and that is why we designed the microscopic code from scratch in
order to fulfill some requirements and reach good efficiency.
The main characteristic of MicroPP that we discuss in detail are:

• Structured grid design

• ELL matrix format

• Efficient memory management.

Solid mechanic module FE2 Structured grids

Parallel MPI code ELL matrix format

Alya MicroPP Memory management

Figure 3: Coupling scheme of codes to solve the FE2 problem. Alya is used to calculate over the
macroscopic scale and MicroPP over the microscopic one.

References
[1] S. Oller. Numerical Simulation of Mechanical Behavior of Composite Materials, Springer, 2014.

231 2
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

A L AGRANGE M ULTIPLIER F ORMULATION FOR C ONSIDERING M EMBER D E -


FORMATION C ONSTRAINTS INTO M ATRIX S TRUCTURAL A NALYSIS

Guilherme Barros 1 , Luiz Fernando Martha 2 , Evandro Parente Junior 3


1 Department of Civil Engineering, Pontifical Catholic University of Rio de Janeiro, gbarros@tecgraf.puc-rio.br
2 Department of Civil Engineering, Pontifical Catholic University of Rio de Janeiro, lfm@tecgraf.puc-rio.br
3 Department of Structural Engineering and Civil Construction, Federal University of Ceara, evandro@ufc.br

This paper is concerned with the analysis of framed structures with inextensible and rigid members, i.e.
members without axial and bending strains. Rigid and inextensible members may be useful in educa-
tional software because they capture the essence of the structural behavior with a reduced the number of
variables. In addition, they allow a comparison with results obtained by hand calculation using classical
structural analysis methods. There are three main approaches to constraint handling: transformation,
penalty function and Lagrange multiplier methods [1,2,3]. The transformation methods, also known as
master-slave elimination, use each constraint to eliminate one equilibrium equation, reducing the num-
ber of degrees of freedom. However, this approach does not allow the determination of axial forces in
inextensible member or bending moments in rigid members. The implementation of the penalty func-
tion method is trivial, since physically it corresponds to assign a very large number to the axial stiffness
(EA) of inextensible members and a very large number to the bending stiffness (EI) of rigid members.
However, as the penalty factor increases, the stiffness matrix becomes increasingly ill-conditioned, lead-
ing to large solutions errors. This paper presents a methodology for considering structural member
deformation constraints using Lagrange multipliers. It consists of adding strain constraints into the total
potential energy minimization, leading to a quadratic programming problem. In addition, this approach
is very suitable for computational implementation because it does not affect the generic characteristic of
a matrix structural analysis. The solution gives rise to one Lagrange multiplier per constraint, which is
essential for computing member internal forces. However, there are situations in which inextensible and
rigid member constrains may be redundant, which prevents the determination of dependent Lagrange
multipliers. In these cases, it is not possible to determine internal forces in the members with redundant
constraints. Although not implemented, a special treatment is indicated for determining member internal
forces in these situations.

References
[1] O.C. Zienkiewicz, R.L. Taylor, J.Z. Zhu. The Finite Element Method: Its Basis and Fundamentals, 2013.
[2] R.D. Cook, D.S. Malkus, M.E. Plesha, R.J.W. Witt. Concept and Applications of Finite Element Analysis,
2002.
[3] C.A. Felippa, Introduction to Finite Element Methods, 2017.

232 1
CILAMCE 2018 C ONGRESS
11-14 N OVEMBER 2018, PARIS /C OMPIÈGNE , F RANCE

O BJECT ORIENTED FRAMEWORK FOR MULTIPHYSICS ANALYSIS


G. G. Botelho 1 , H. A. S. Monteiro 2 , R. G. Peixoto 3 , R. L. S. Pitangueira 4
1 Universidade Federal de Minas Gerais - UFMG, guilhermebotelho@ufmg.br
2 Universidade Federal de Minas Gerais - UFMG, hmonteiro@ufmg.br
3 Universidade Federal de Minas Gerais - UFMG, rodrigo.peixoto@dees.ufmg.br
4 Universidade Federal de Minas Gerais - UFMG, roque@dees.ufmg.br