FACULDADE DE ENGENHARIA

UNIVERSIDADE DO PORTO

Probabilistic Seismic Assessment of Bridges

Ricardo Nuno Carvalho Monteiro

A thesis submitted to the University of Porto in accordance with the

requirements of the degree of Doctor of Philosophy in Civil Engineering,
in the Faculty of Engineering

Porto, 2011
“We build too many walls and not enough bridges”

Sir Isaac Newton (1643 – 1727)

Acknowledgments

Seismic assessment of bridge structural systems is an extremely interesting field of

research in which I had the opportunity to work since the last year of my Civil Engineering
degree, back in 2004. Such early discovery and all the work carried out within this
challenging step of my professional career became definitely more attractive and
motivating due to the contribution of a number of people to whom I would like to express
my most sincere gratefulness.

To Professor Raimundo Delgado, my advisor, I would like to thank for his crucial role in
this work, not only academically, with his wisdom and intelligence, but personally as well,
for his permanent encouragement and motivation, making everything definitely become
easier. To him I believe to owe a fair amount of my scientific knowledge and interest in
structural and seismic analysis.

To Professor Aníbal Costa, my co-advisor, I would like to thank for his experienced
guidance and practicability, as well as his ability to introduce me to the seismic analysis
topic from the beginning of my research years.

To Professor Rui Pinho and Doctors Chiara Casarotti and Helen Crowley, I’m thankful for
their help and guidance throughout the enriching training periods I spent, in 2007 and
2008, at the EUCENTRE, Pavia. I would like to specially thank Professor Rui Pinho, who
became a friend, for his faith in me and my work, for his incentive and leadership skills,
which definitely inspired me.

To Professors António Arêde, Nelson Vila Pouca, Miguel Castro, Rui Calçada, Ana Maria
Sarmento, Paula Milheiro, Maria do Carmo Coimbra and Isabel Silva, I wish to thank for
their interest in my work, for their helpful scientific contribution and supporting words.

To my PhD colleagues, friends and academic partners ever since undergraduate times,
Mário, Cristina and Nuno, I would like to thank for their companionship, help in a non

v
measurable number of aspects at all levels, daily friendship and support. A particular
mention goes to Mário for his brotherhood.

To the other colleagues and friends at FEUP or at the EUCENTRE, Sara, Sérgio,
Fernando, João, Ricardo, Pedro, Patrício, Carlos, Maria Serena, Romain, Claudia and
Tuba, amongst others, I’m grateful for their team spirit, help provided with no hesitation
and thoughtfulness towards me and my work.

To Luís Sousa, Rui Ribeiro and Ângelo Carvalho, former MSc students I had the pleasure
to work with, I wish to thank for the contribution with provided data and other elements
from their individual studies.

To the Structural Department and Laboratory staff, especially Marta for her esteem, I
would like to thank for their availability and support in all the logistical tasks.

To my dear friends Lisa, Maria Inês, Pedro, Rita, Andreia, Nuno and Luís Pedro I specially
thank for their patience and affection, for providing me with numerous wonderful moments
and for helping me to go through the toughest ones.

Finally, but firstly in my heart, I would like to deeply thank my family, who always took
care of me, embracing me in every moment, allowing me, so many times, to focus on
nothing else but my work, spending less time with them than I would wish to. To my
parents, António and Maria Rosa, I’m thankful for their endless love and unconditional
support, whichever my choices are. To my sisters, Eugénia and Susana, I’m grateful for
their friendship and altruism, showed by their constant dedication to my well-being and
happiness.

This research work would not have been possible without the financial support of FCT
(Portuguese Foundation for Science and Technology) in the form of the PhD grant
SFRH/BD/22959/2005.

vi
Abstract

The main goal of this thesis is to outline the entire process of assessing the seismic safety
of single existing reinforced concrete bridges, within a probabilistic context. Given that
several possible methodologies are currently available for seismic assessment with the
inclusion of nonlinear structural behaviour, this work intends to contribute with a global
overview on the essential aspects surrounding any safety assessment procedure, with
particular focus on bridge structures. A specific methodology using statistical
characterization of the assessment problem, conducting to the failure probability
computation was employed and optimized.

Firstly, a characterization process of the seismic action has been carried out, with the
purpose of portraying this sort of loading, which is well known for its complexity and
variability, coming from the occurrence of highly unpredictable natural seismic hazards.
Furthermore, the optimum way for seismic loading to be set up is looked for, in terms of
reduced dispersion and number of earthquake records, with a view to the use in subsequent
repeated probabilistic safety calculations.

The modelling issue is afterwards analysed, with a brief mention to the sort of structural
models currently employed for bridges and ways of accounting for material and
geometrical nonlinearity. Special emphasis is put on the latter – different modelling
alternatives for the material plasticity representation are scrutinized, a subject seen as quite
pertinent.

The estimation of the structural response for the duly modelled structure, subjected to
properly characterized seismic loading, which constitutes the next logical task, has been
addressed by means of an extensive comparative parametric study of the two most
commonly accepted approaching tools: nonlinear dynamic or nonlinear static analysis,
featuring different sorts of seismic action input.

vii
Finally, the intended safety verification is carried out probabilistically for a case study mad
up of a set of different bridge configurations, in length, regularity and structural type,
together with an equally wide set of earthquake records, both already used for the previous
calibration studies conducted for the different components of the safety assessment
problem. Both capacity and seismic demand are characterized statistically, according to
premises settled in the modelling and structural response estimation sections, respectively.
Statistical characterization is managed through proper random simulation techniques, such
as the employed Latin Hypercube Sampling scheme. Different alternatives to reach the
failure probability, involving distinct uncertainty levels, are presented and compared, with
the aim to conclude over applicability and accuracy of the methodologies, as well as the
relative seismic safety of the tested configurations.

viii
Resumo

O trabalho apresentado nesta tese teve como principal objectivo percorrer as várias etapas
do processo de verificação de segurança sísmica individual, num contexto probabilístico,
de pontes existentes em betão armado. Presentemente, várias alternativas se encontram
disponíveis para a abordagem ao problema de segurança sísmica e, por conseguinte, este
trabalho pretende dar um contributo com a elaboração de uma perspectiva global do tema,
sintetizando os aspectos mais relevante, com especial ênfase na análise de pontes. É
proposta e aplicada uma metodologia de avaliação de segurança por cálculo da
probabilidade de ruína da estrutura, com caracterização estatística das diferentes variáveis.

O trabalho conduzido começa por focar a caracterização da acção sísmica,

reconhecidamente complexa e com elevada variabilidade, consequência natural da
imprevisibilidade associada à ocorrência de fenómenos sísmicos. Desse modo, procura-se
estudar a melhor maneira de definir tal acção, investindo na redução da dispersão e no
número de acelerogramas reais a utilizar, com vista à sua utilização em aplicações praticas
de verificação de segurança, envolvendo repetição de cálculos probabilísticos.

A temática da modelação é abordada fazendo referencia aos modelos estruturais

correntemente utilizados para análise não linear de pontes bem como aos diferentes modos
de ter em consideração a não linearidade material e geométrica. É dada especial atenção ao
primeiro tipo de não linearidade enumerado, estudando a viabilidade de emprego de
diferentes estratégias de modelação da plasticidade material, uma questão tida como
particularmente pertinente na análise sísmica de estruturas.

A estimativa da resposta estrutural das pontes devidamente modeladas, quando submetidas

a uma acção sísmica convenientemente caracterizada, é igualmente alvo de estudo com
relativo detalhe por intermédio de um estudo paramétrico extenso de comparação das duas
vias de análise tipicamente usadas: análise dinâmica não linear ou análise estática não
linear, a que correspondem diferentes formas de representação da acção sísmica.

ix
Finalmente, a verificação de segurança propriamente dita é efectuada para um caso de
estudo constituído por diferentes configurações, em extensão, regularidade ou tipo de
aparelhos estruturais, usando com um conjunto igualmente vasto de acelerogramas,
igualmente utilizados na calibração das diversas componentes do processo da avaliação de
segurança. A capacidade resistente e a exigência estrutural são caracterizadas
estatisticamente de acordo com os resultados dos estudos sobre a acção sísmica, modelação
e previsão da resposta. A caracterização estatística é conseguida por intermédio de técnicas
de amostragem convenientemente aplicadas. São propostos e comparados diferentes
procedimentos conducentes ao cálculo da probabilidade de ruína, com diferentes níveis de
incerteza, com o objectivo de concluir quanto à sua aplicabilidade bem como da segurança
sísmica das diversas pontes testadas.

x
Résumé

L'objectif principal de cette thèse est de décrire l'ensemble du processus d'évaluation de la

sécurité sismique de ponts existants en béton armé, dans un contexte probabiliste. Lorsque
plusieurs méthodes possibles sont actuellement disponibles pour l'évaluation sismique avec
l'inclusion du comportement des structures non linéaires, ce travail veut contribuer à une
vision globale sur les aspects essentiels entourant toute la procédure d'évaluation de la
sécurité, avec un accent particulier sur les structures de pont. Une méthodologie spécifique
en utilisant la caractérisation statistique du problème d'évaluation, tenue à la probabilité de
défaillance de calcul a été utilisé et optimisé.

Tout d'abord, un processus de caractérisation de l'action sismique a été réalisée, dans le but
de représenter ce type de chargement, qui est bien connue pour sa complexité et la
variabilité, provenant de la présence très imprévisible des risques naturels sismiques. En
outre, la meilleure façon pour le chargement sismique à mettre en place est recherché, en
termes de dispersion réduite et le nombre de dossiers tremblement de terre, en vue de
l'utilisation répétée dans les prochains calculs probabilistes de sûreté.

La question de la modélisation est ensuite analysée, avec une brève mention de ce genre de
modèles structurels actuellement employé pour les ponts et les moyens de la comptabilité
pour le matériel et la non-linéarité géométrique. Un accent particulier est mis sur ce
dernier – la modélisation de/sur différentes alternatives pour la représentation plasticité du
matériau sont examinées, un sujet considéré comme tout à fait assez pertinente.

L'estimation de la réponse structurale de la structure modélisée dûment, soumis à des

séismes de chargement bien caractérisé, qui constitue la prochaine étape logique, a été
abordée par le biais d'une étude comparative approfondie des paramètres des deux outils
les plus communément acceptés approche: dynamique non linéaire ou non linéaire analyse
statique, avec différentes sortes d'entrées de action sismique.

xi
Enfin, la vérification de sécurité destinés probabiliste est effectué pour une étude de cas
folle d'un ensemble de configurations différentes pont, dans la longueur, la régularité et le
type de construction, avec une large série de documents tout aussi tremblement de terre,
toutes deux déjà utilisées pour les études d'étalonnage précédente réalisée pour les
différentes composantes du problème d’évaluation de la sécurité. Les deux capacités de la
demande sismique sont caractérisés statistiquement, selon locaux installés dans les sections
et structurelles réponse estimation de modélisation, respectivement. Caractérisation
statistique est géré par des techniques appropriées de simulation aléatoires, telles que
l'échantillonnage par hypercubes latins régime des travailleurs salariés. Différentes
alternatives pour atteindre la probabilité de défaillance, impliquant les niveaux
d'incertitude distincts, sont présentés et comparés, dans le but de conclure sur l'applicabilité
et la précision des méthodes, ainsi que la sismique de sécurité relative des configurations
testées.

xii
Contents

Acknowledgments ................................................................................................................v

Abstract...............................................................................................................................vii

Resumo.................................................................................................................................ix

Résumé.................................................................................................................................xi

Contents .............................................................................................................................xiii

1. Introduction...................................................................................................................1.1

1.1 General......................................................................................................................1.1

1.2 Motivation and goals ................................................................................................1.3

1.3 Thesis outline............................................................................................................1.4

2. Seismic Assessment of Bridges.....................................................................................2.1

2.1 Introduction...............................................................................................................2.1

2.2 Recent earthquake events – major observations and lessons learned .......................2.3
2.2.1 Loma Prieta earthquake .....................................................................................2.4
2.2.2 Northridge earthquake .......................................................................................2.7
2.2.3 Kobe earthquake ................................................................................................2.9
2.2.4 L’Aquila earthquake ........................................................................................2.12
2.2.5 Haiti earthquake...............................................................................................2.13
2.2.6 Christchurch earthquake and Tōhoku earthquake and tsunami .......................2.15
2.2.7 Summary..........................................................................................................2.17

2.3 The seismic action ..................................................................................................2.19

2.3.1 Selection and scaling of real records ...............................................................2.20
2.3.2 Number of records ...........................................................................................2.24

xiii
 2.4 Nonlinear seismic analysis – methods for structural response prediction ..............2.26
2.4.1 Nonlinear static analysis – pushover................................................................2.28
2.4.2 Nonlinear dynamic analysis .............................................................................2.31
2.4.3 Innovative displacement based approaches .....................................................2.33
2.4.4 Nonlinear response prediction of bridges ........................................................2.35

2.5 Nonlinear modelling overview................................................................................2.37

2.5.1 Material nonlinearity........................................................................................2.38
2.5.2 Geometrical nonlinearity..................................................................................2.41

2.6 Safety assessment methods .....................................................................................2.42

2.6.1 Deterministic assessment .................................................................................2.42
2.6.2 Probabilistic assessment...................................................................................2.43

2.7 Summary .................................................................................................................2.44

3. Seismic Input .................................................................................................................3.1

3.1 Accelerograms ..........................................................................................................3.2

3.2 Real accelerograms – selection of records ................................................................3.3

3.2.1 Initial considered set ..........................................................................................3.3
3.2.2 A homogeneous base of records – Clustering analysis......................................3.5
3.2.3 Application to the set of selected records ..........................................................3.6

3.3 Real accelerograms – scaling....................................................................................3.8

3.3.1 Record quantities based intensity measures.......................................................3.9
3.3.2 Spectrum quantities based intensity measures .................................................3.10
3.3.3 Comparative analysis of typical intensity measures ........................................3.12
3.3.4 Response based intensity measures..................................................................3.14
3.3.4.1 Preliminary Approach on a Displacement-Based Intensity Measure.......3.15
3.3.4.2 Procedure..................................................................................................3.16
3.3.4.3 Structural application ...............................................................................3.18

3.4 Artificial accelerograms..........................................................................................3.26

3.4.1 Algorithm of generation...................................................................................3.27
3.4.2 Reference response spectrum for generation of records ..................................3.28
3.4.3 Artificial ground motion records generation parameters .................................3.29

xiv
 3.5 Comparison between real and artificial records – structural application................3.29
3.5.1 Structural description.......................................................................................3.30
3.5.2 Real records – original and homogeneous bases .............................................3.31
3.5.3 Minimum necessary number of accelerograms ...............................................3.32
3.5.4 Artificial records ..............................................................................................3.34
3.5.5 Comparison of results ......................................................................................3.35

3.6 Conclusions.............................................................................................................3.37

4. Nonlinear Modelling .....................................................................................................4.1

4.1 Material models ........................................................................................................4.2

4.1.1 Concrete models ................................................................................................4.2
4.1.1.1 Modified Kent-Park model..........................................................................4.4
4.1.1.2 Mander-Priestley-Park model ....................................................................4.6
4.1.2 Steel models.......................................................................................................4.7
4.1.2.1 Giuffrè-Menegotto-Pinto model..................................................................4.9

4.2 Material nonlinearity approaches............................................................................4.10

4.2.1 Lumped plasticity models (plastic hinges) ......................................................4.10
4.2.1.1 Plastic hinge constitutive law – trilinear curve ........................................4.11
4.2.1.2 Plastic hinge length and location .............................................................4.12
4.2.2 Distributed plasticity models (fibres) ..............................................................4.18

4.3 Case study ...............................................................................................................4.19

4.3.1 Selected bridges ...............................................................................................4.19
4.3.1.1 Piers ..........................................................................................................4.22
4.3.1.2 Deck ..........................................................................................................4.24
4.3.1.3 Abutments..................................................................................................4.25
4.3.2 Seismic action ..................................................................................................4.26

4.4 Comparative study ..................................................................................................4.27

4.4.1 Elastic modal analysis comparison ..................................................................4.28
4.4.2 Moment-curvature relationships ......................................................................4.32
4.4.3 Pushover capacity curves.................................................................................4.35
4.4.4 Structural response parameters ........................................................................4.38

4.5 Conclusions.............................................................................................................4.42

xv
5. Seismic Demand ............................................................................................................5.1

5.1 Nonlinear Static Analysis..........................................................................................5.2

5.2 Nonlinear Static Procedures......................................................................................5.4

5.2.1 Capacity Spectrum Method (CSM)....................................................................5.5
5.2.2 N2 Method .........................................................................................................5.8
5.2.3 Displacement Coefficient Method (DCM) ......................................................5.10
5.2.4 Modal Pushover Analysis (MPA) ....................................................................5.11
5.2.5 Adaptive Capacity Spectrum Method (ACSM) ...............................................5.13
5.2.5.1 Damping-based spectral reduction...........................................................5.17
5.2.5.2 Ductility-based spectral reduction............................................................5.27
5.2.6 Adaptive Modal Combination Procedure (AMCP) .........................................5.28
5.2.7 Method comparison..........................................................................................5.30

5.3 Parametric Study .....................................................................................................5.31

5.4 Results.....................................................................................................................5.33
5.4.1 Results representation ......................................................................................5.33
5.4.2 Preliminary evaluation .....................................................................................5.35
5.4.2.1 Capacity Spectrum Method.......................................................................5.35
5.4.2.2 N2 Method.................................................................................................5.38
5.4.2.3 Modal Pushover Analysis..........................................................................5.43
5.4.2.4 Adaptive Capacity Spectrum Method........................................................5.46
5.4.3 Comparative study results................................................................................5.53
5.4.3.1 Global results............................................................................................5.53
5.4.3.2 Intensity level results.................................................................................5.55
5.4.3.3 Bridge configuration results .....................................................................5.57

5.5 Conclusions.............................................................................................................5.62

6. Safety Assessment .........................................................................................................6.1

6.1 Failure probability computation................................................................................6.2

6.1.1 Seismic action – intensity level probability density function ............................6.5
6.1.2 Capacity distribution ..........................................................................................6.8
6.1.3 Latin Hypercube Sampling ................................................................................6.9
6.1.4 Seismic structural effects .................................................................................6.12

xvi
 6.1.5 Failure probability with local uncertainty........................................................6.13
6.1.6 Failure probability for a given intensity level..................................................6.17
6.1.7 Failure probability with global uncertainty .....................................................6.19
6.1.8 Reliability index...............................................................................................6.21

6.2 Application to a set of bridges ................................................................................6.22

6.2.1 Latin Hypercube sampling in capacity definition............................................6.23
6.2.2 Seismic action probability density function.....................................................6.29
6.2.3 Failure probability with traditional NSA approach .........................................6.31
6.2.3.1 Vulnerability function from nonlinear dynamic analysis .........................6.31
6.2.3.2 Vulnerability function from nonlinear static analysis ..............................6.37
6.2.3.3 Structural effects and failure probability..................................................6.46
6.2.4 Failure probability with global uncertainty LHS approach .............................6.56
6.2.4.1 Simulation process calibration .................................................................6.56
6.2.4.2 Structural effects and failure probability..................................................6.63
6.2.5 Comparison of methodologies .........................................................................6.67

6.3 Conclusions.............................................................................................................6.71

7. Conclusions....................................................................................................................7.1

7.1 Concluding remarks ..................................................................................................7.1

7.2 Future developments.................................................................................................7.9

References.............................................................................................................................1

xvii
1. Introduction

1.1 General

The occurrence of a major earthquake event is still seen as a catastrophe, with devastating
consequences over human lives, buildings and transportation networks, causing a large
scale impact in any society and multiple costs of different natures. Indeed, even if
becoming particularly devastating to poor, underdeveloped countries, it equally greatly
affects populations and several infrastructures in modern, industrialized countries, where,
despite the advanced status of seismically designed structures, it is still possible to point
out weakness points, such as historical or late XX century constructions, in need for
retrofitting. From a more global picture, the occurrence of earthquakes can have an
outcome which is reflected in affected country policies, during several years, given the
deep changes in building philosophy or large investments in seismic design strategies.

A number of relatively recent earthquakes are typically referred to, due not only to the their
destructive impact but also to the lessons that were learnt, leading, sometimes, to profound
changes or the establishment of turning points in seismic design philosophies. The San
Fernando (1971), Loma Prieta (1989) or Northridge (1994) earthquakes in the USA; Kobe
earthquake in Japan (1995); the 921 earthquake (1999), in Taiwan, are some of the
examples of such important recorded events in the past few decades, together with other
earthquakes that occurred in Turkey, Greece or China. The particular case of recent

1.1
Introduction

earthquake events that stroke Italy in 2009 (L’Aquila earthquake); Haiti, in 2010; New
Zealand and Japan, in 2011, denote essentially two different scenarios, both very pertinent.
The first scenario, corresponding to the earthquakes that have occurred in so-called
developed countries, with improved codes for seismic design, reveals how an earthquake
can still cause severe damage, found in old, non retrofit constructions and in some new,
supposedly seismic resisting, structures. On the other hand, the Haiti earthquake revealed
the chaotic scenario that can be verified when the shaking takes place in poor, under
development countries. Thousands of human losses occurred and several basic facilities
became inoperable, mobilizing substantial international aid. The recognition of the
importance of the role that earthquakes play in modern societies is, therefore, largely
evident, in a much wider range of issues than sometimes thought.

There are two essential key intervening aspects, which any society needs to consider when
drawing a response strategy to seismic hazard. The first is to develop efficiently enhanced
building codes and guidelines, which become able, in the end, to assure that new
constructions feature proper ductility characteristic, as well as energy dissipation
mechanisms, allowing the structure to accommodate a much larger deformation demand,
resisting and performing quite better, without collapsing, when facing earthquakes. The
other intervention field is at the existing constructions assessment level. The number of
existing structures that were built in the past, from hundreds to a few decades ago, with no
regard for any seismic provisions is enormous. Such considerable portion of the total
construction is extremely exposed to forthcoming seismic events, presents probably high
vulnerability and needs, therefore, to be seismically assessed and, eventually, retrofit.

It is the author’s belief, however, that effort has been more concentrated in the design of
new structures side, rather than in the assessment of existing structures one. On that matter,
considerable improvement has been made in the past few years, with the development of
further refined methodologies, which include adequate consideration of relatively complex
nonlinear structural models as well as highly refined probabilistic safety computation
methods. Such approaches, seen as more accurate are, though, yet to become common
procedure among practitioners. This is, in part, due to the several different methodologies
that are available to reach a failure probability of an existing structure, as well as to the
complexity of issues surrounding any possible safety assessment procedure. Moreover,
detailed nonlinear analysis procedures continue to be complex, time-consuming and, for

1.2
 Introduction

that reason, common practice still majorly consists of elastically analysing the structures,
with subsequent application of behaviour reduction factors. A summarizing work of the
different approaches, stressing the relevant criteria, discarding unnecessary alternatives, in
order to unify procedures and reduce the actual dispersion in methodologies is justified.

All of the drawn considerations become even more relevant on bridge structures, given
that, traditionally, the majority of studies and frameworks have been focused on buildings,
leaning space for a gap of knowledge that has, nonetheless, become less evident in the
recent past. Even if typically highly resistant structures, many times protected by the
deck’s elastic behaviour, bridges are key elements in any region stricken by an earthquake
and require therefore, special attention, due to their vital importance as connecting and
transportation facilities. Indeed, in the immediate aftermath of an earthquake, closure of a
bridge can impair emergency response operations.

The current challenge within the field of assessing existing structures, and therefore
bridges, is to reach a failure probability, which takes into account the uncertainty of all the
variables at stake. Furthermore, the applicability of such procedures in a current practice
context is fundamental. To accomplish so, intermediate methodologies, less complex, but
still improving with respect to the currently employed ones, by inclusion of simpler
nonlinear features, can be considered within a transition process.

1.2 Motivation and goals

The main goal of this work has been to probabilistically assess the seismic vulnerability of
single existing reinforced concrete bridges to seismic action. Even if inevitably
characterized by a certain level of ambiguity, the safety degree has been evaluated in a
probabilistic fashion, i.e., through the computation of failure probability of single bridges,
whose capacity features duly characterized uncertainty, when facing a specific seismic
scenario, defined statistically as well. A deeper knowledge on different approaches for the
collapse probability computation is sought and the influence of the different elements, or
steps, of the safety assessment process is also studied in detail.

Indeed, the path to reach a final conclusion about the safety, or lack of, offered by a
structural system is everything but straightforward and there is still no consensus among

1.3
Introduction

the worldwide research community on a large number of aspects. As a result, this work
aimed to look into as much detail as possible the different topics that need to be covered
for the vulnerability evaluation, focusing on the particular case of bridge structures, trying
to contribute to the exposure and clarification of the possible alternatives within each issue,
providing as well, if possible, recommendations and guidelines for practitioners and future
research. Such different issues include, from a macroscopic perspective, the definition of
seismic input, which is naturally extremely variable, highly unpredictable, and able to be
represented by different input modes; the structural modelling, featuring geometrical and
material nonlinearity; the structural response estimation, which can be static or dynamic-
based; and the final probabilistic computation itself, intersecting demand and capacity.

The study of each of the mentioned aspects has been systematically carried out by means
of parametric studies, in which results were object of statistical treatment, in order to
enable the covering of a high number of different structural configurations and earthquake
records as well as to reach, as much as possible, generalized findings. An extensive battery
of regular and irregular bridges, together with a relatively wide seismic input, is used to
allow the fulfilment of the generalization purposes.

1.3 Thesis outline

In order to fulfil the devised work objectives, each chapter of this thesis will go through
every aspect considered to be relevant within a seismic assessment of bridges context,
trying, as much as possible, to make it coincide with well defined steps of the natural path
to follow within a full probabilistic vulnerability evaluation procedure. Most of the topics
are strictly correlated and contribute to each others reciprocally, given that, structurally,
everything is connected in some way. However, it would be impossible to present the work
without sorting the topics in a predefined order, even if every choice would be arguable.
The reader is hence asked to keep this in mind when going through the different chapters.

Firstly, an introduction to the seismic phenomenon is made in Chapter 2, featuring a brief

description of the major earthquake events of the past decades, with the intent of stressing
the importance and growing relevancy of the seismic assessment of existing structures,
with special mention to bridges, as well as seismic design of new ones. Not only the
devastating character of the ground shaking is highlighted but reference is continuously

1.4
 Introduction

made to specific structural damage occurring on bridges due to this sort of natural hazard.
Afterwards, an initial approach to all the subjects involved in the process of assessing the
seismic safety of bridges is presented, with the aim to summarize the state of knowledge
and the main available paths for each of the tasks that need to be carried out towards the
safety verification. The most significant advances in each theme are reviewed and topics
still in need for further development are object of particular attention, hence, introducing
the subsequent chapters.

The first of the aforementioned safety assessment process stages can be seen as the
preparation of the seismic action to which the structure is expected to be subjected to.
Accordingly, Chapter 3 deals with the seismic action characterization, establishing a
permanent correlation with the seismic analysis methods and its corresponding needs in
terms of seismic input (Chapters 5 and 6). As a consequence, given the superior accuracy
of nonlinear dynamic analysis, which is becoming more popular and disseminated within
seismic assessment research, the main focus is on accelerograms, the most adequate input
form of earthquake events. Aspects such as type of records, real or artificial, as well as
selection and scaling techniques according to distinct intensity measures, when choice falls
on the primer, are discussed and investigated, with a view to optimization. Based on a
European integrated project, LESSLOSS, a brief case study, made up of a long, irregular
viaduct, is parametrically analysed. An innovative intensity measure based on simplified
pushover analysis is additionally proposed and preliminarily validated.

Chapter 4 steps into the modelling topic, in which one of the most important issues is the
inclusion of nonlinearity features. Indeed, nonlinear models are conspicuously needed for
the employment of up to date methodologies, whether static or dynamic (reviewed in
Chapter 5). An overview on geometrical and material nonlinearity is carried out, with
emphasis on the latter. On such matter, the nonlinear material models, used for the
characterization of reinforced concrete structures, are described, under monotonic and
cyclic loading conditions. Furthermore, different ways of accounting for the material
plasticity, namely through plastic hinges or fibres, are thoroughly evaluated by means of a
detailed parametric study, which will enable the consideration of the most suitable type of
model in extensive analyses in Chapter 6. The description of a case study of fourteen
different bridge structures, as well as the seismic input, both used for the nonlinear
modelling studies and within the subsequent subjects, is carried out.

1.5
Introduction

After the seismic action and the nonlinear structural model are calibrated, the estimation of
the structural response to such input, essential to the seismic demand definition, is the main
scope of Chapter 5. Again, among the wide range of structural performance prediction
possibilities, the analysis is centred in the comparison of the currently most considered as
eligible tools: nonlinear static analysis or nonlinear dynamic analysis. Four commonly
employed nonlinear static procedures are selected, individually calibrated and further
statistically compared to nonlinear dynamic analysis, in terms of piers, deck and abutments
different response parameters. Comparison is carried out for the case study and seismic
input, both described in Chapter 4. The validity of using static approaches, when compared
to dynamic counterparts, is investigated and relative performance of the different selected
procedures is looked at as well.

The seismic safety assessment itself is at last dealt with in Chapter 6, making use of the
previous chapters’ conclusions. Featuring a deeply probabilistic nature, different
methodologies are presented for the safety assessment, which foresee the failure
probability computation. In order to accomplish so, as well as to duly incorporate the
uncertainty associated to all the safety problem variables, both capacity and demand are
statistically characterized by means of assumed and/or adjusted distributions, with the
contribution of the findings and guidelines drawn in Chapters 3 and 4. Such statistical
definition is carried out through the obtaining, when necessary, of random samples making
use of the Latin Hypercube sampling technique, which is also thoroughly tested and
calibrated. Furthermore, different ways of dealing with the uncertainty of the different
variables, together with the use of different response prediction techniques, studied in
Chapter 5, will yield distinct failure probability quantities. Distinction is essentially
established between accounting for uncertainty local or globally and the use of static or
dynamic vulnerability/fragility curves. Again, the different approaches are tested for the
presented case study and observations are made towards the recognition of an optimum
procedure, identifying patterns as a function of the bridge structural configuration.

Finally, in Chapter 7, a summary of concluding remarks, featuring a discussion of the

most important findings of each subject, is presented. The level of accomplishment of the
initial objectives of the work is appraised and future developments are suggested.

1.6
2. Seismic Assessment of Bridges

2.1 Introduction

Within the possible loading types to which a structure can be subjected to, seismic action is
a highly unpredictable one, a characteristic that assumes an extreme importance and is
widely recognized within the structural engineering field. For that reason, in situ post
earthquake damage observation, experimental testing and probability based analysis
methods are essential for an accurate seismic assessment procedure.

Unfortunately, the devastating effects of earthquakes that have been still affecting several
populations in different countries for the past few years continue to be one of the disaster
events that most uncover unexpected failure mechanisms and causes in structures subjected
to such type of demand. Important advances in earthquake engineering of bridges have
definitely been associated to the occurrence of intense seismic disasters, given that the
observation of major seismic events allows the recognition of the structural damage extent,
stimulating the development of more accurate and refined assessment methodologies as
well as the definition of retrofitting solutions for the resisting elements, such as shear
strengthening of the piers or improvements on the solutions for the foundations, pier-deck
connection, amongst others. The initially observed damages were mainly related to the
failure of the ground, whereas, as the bridges started to be built up of reinforced and

2.1
Seismic Assessment of Bridges

prestressed concrete, several different failure mechanisms, mostly the pier rupture kind of
ones, appeared.

Additionally to the particular unpredictable character of the seismic action, bridge

structures behave in a considerable different manner from building structures, something
that has to do with the naturally irregular nature of the former sort of structures, highly
vulnerable in the transverse direction, with respect to the development of the deck, as well
as highly unknown, in terms of seismic behaviour, until the recent past.

Together with the experience that comes from observing the post-earthquake status of the
structures, the thorough study of the seismic action itself is another fundamental issue,
which has undergone intense development, mainly in the pursuance of an optimum way of
selecting ground motion records and using them, so as to minimize the typically high
dispersion that is found. Indeed, a major extent of the scatter that the results of any sort of
nonlinear analysis present is frequently related to the irregularity characterizing the loading
input.

At the same time, the techniques that are used to model the structure can equally lead to
extensive debate, as a consequence of the considerable number of available alternatives,
which go from geometrical to material modelling possibilities, including different degrees
of elements discretization or hysteretic energy dissipation models as well. On such matter,
the evolution along the past few years has definitely been outstanding, caused by a
significant progress in the computational capacity, allowing more refined models to be
considered and a greater amount of results data to be extracted from the analysis.

Any safety assessment process is carried out following a specific methodology that
combines all the elements defining the capacity and the demand and, herein, the variety is
even more pronounced, with a wide set of methods that can be used, from simplified,
rough procedures to highly refined techniques. Several studies have been conducted,
improving at the same time, methodologies with different detail levels, further and further,
with the purpose of including them in current assessment, design and codes or guidelines.
One of the greatest challenges has been the development of successively updated simple
methodologies, yet credible and nonlinear, as the seismic action requires, for the use of
practitioners and to be implemented in design codes. At the same time, more refined

2.2
 Seismic Assessment of Bridges

alternatives have still continued to be calibrated until they become more user-friendly and
the engineering community gets more familiarized with the sometimes complex concepts
that are involved.

From a state-of-art point of view, within this chapter, a short review on the thematic of this
work is intended, denoting the aspects of major relevance that are related to the seismic
behaviour and safety assessment of bridges. Firstly, a few major earthquake events that
have occurred in the recent past, which considerably affected reinforced concrete bridges,
are mentioned emphasising the learned lessons and the critical issues that have lead to
structural malfunction. Furthermore, with the recognition of the need for means to assess
and identify vulnerable configurations, whether existing or new ones, a perspective on the
different elements that are part of such endeavour and on what has been done in the past is
carried out. The seismic action, a fairly complex one, is the first analysed element, in terms
of types of representation, scaling techniques, amongst others. Indeed, the better
characterized the seismic action is, the clearer and easier will the assessment process
become. Structural modelling techniques, concerning nonlinear material models or energy
dissipation, are another fundamental aspect, this one from the structure side. Finally, an
overview on the typical ways of carrying out the safety assessment itself is presented,
which will constitute no more than a contribute, given the wideness characterizing such
matter.

Many of the concepts and notions that are herein presented are, in a certain way, universal,
able to be implemented within the study of any sort of structures. It is well know, in
addition, that the general knowledge in seismic behaviour of buildings, for instance, is
quite superior than the one for bridges, which sometimes leads to the attempt of adapting
some methodologies, developed bearing building frames in mind, to that sort of structures.
According to such premises, the exposure that follows will consistently try to make the
connection to the specific bridges case.

2.2 Recent earthquake events – major observations and lessons learned

Bridges are a set of different elements working together in sometimes unexpected ways
during earthquakes, rendering highly important the study of its behaviour during
earthquakes. It has been relatively frequent that significant earthquakes have occurred and

2.3
Seismic Assessment of Bridges

strongly shaken populations all over the globe, affecting well developed, industrialized
countries and less developed ones as well. The consequences are always significant and,
sometimes, devastating, involving human lives and structures. In the past fifty years,
approximately, the number of bridges, viaducts and special structures has increased hugely
and so the effects of earthquakes have become more visible, calling the attention of the
international community, even though not always to the desirable extent.

In previous studies on seismic behaviour of bridges (e.g., (Vaz, 1992; Delgado, 2009))
other authors pointed out the most relevant earthquake events in the twentieth century as
well as more recent ones, in the past decade, recalling how undeniably topical seismic
action is. It is not intended, therefore, herein, to be repetitive or exhaustive in describing
the History around ground shaking but to provide the reader a context that enables better
understanding of the importance of such matter and how the observation contributes to
post-earthquake work guidance. The twentieth century has witnessed hundreds of
earthquakes of magnitude 6 and above all over the globe. From a structural engineering
point of view, the latter and the ones occurring in high building density regions, are more
interesting, given that they have put to test a higher amount of bridges and viaducts, as well
as more enhanced, up to date, seismic design criteria. On such basis, a set of the most
representative earthquakes, having occurred in North America and Asia in the past thirty
years, will be mentioned, mainly for their important contribute to bridges earthquake
engineering. Furthermore, two major recent ground shaking events that took place in
Europe and Central America will be briefly referred to, owing to their contemporaneity.

2.2.1 Loma Prieta earthquake

On such basis, the Loma Prieta earthquake, which shook the San Francisco Bay area of
California in 1989, is a classical example of a natural disaster that largely affected a highly
population density area, endowed with a vast highway transportation network. Causing a
total of 63 casualties and near 4000 injuries, a particular aspect of this quake was that 42 of
the casualties were due to the collapse of the two levels Cypress Street Viaduct of
Interstate 880 in West Oakland, as illustrated in Figure 2.1. Nevertheless, less than five
percent of the bridges exposed to ground shaking were damaged.

2.4
 Seismic Assessment of Bridges

Figure 2.1 – Cypress viaduct collapse: left – aerial view; right – piers’ failure detailed view.

Two major factors, apart from soon after determined resonance effects, induced the
collapse of the late 1950s-built viaduct: geotechnical issues and deficient reinforced
concrete design. The upper and lower levels composing the structure were connected by
two-column bents in a combination of cast concrete and four pin (shear key) connections.
The upper deck in some sections was not securely fastened to the lower deck and, as the
bridge vibrated during the earthquake, the pins connecting the levels also began to vibrate,
causing the concrete surrounding the pins to crumble and break away. The lack of
transverse reinforcement in the nodes connecting piers and deck was evident – see Figure
2.2 for detail. Indeed the connections of elements are often subjected to higher demand
than the resisting elements themselves. The prediction of ductile connections was
definitely not taken into consideration by the time the viaduct was built or strengthened.

As for geotechnical reasons, the viaduct was built on soft mud, weak soil, highly
susceptible to liquefaction during an earthquake and exhibiting larger ground motion.
Without the presence of concrete under the piers, those elements slid sideways under the
weight of the upper deck and allowed a large portion of the upper deck to collapse
(Yashinsky, 1998a). Several aspects of the design and construction of the structure have
hence been suggested as contributing to its failure: inadequate transverse reinforcement in
the piers, ineffective bent cap and pin connection design (Moehle, 1999) and improper
compensation for the weak soil conditions (Yashinsky, 1998a).

2.5
Seismic Assessment of Bridges

Figure 2.2 – Cypress viaduct: lack of proper reinforcement in pier-deck connections.

The famous Oakland Bay Bridge, on the other hand, ended up suffering minor damage, as
a section of the upper deck of the eastern truss portion of the bridge collapsed onto the
deck below, indirectly causing one death. The satisfactory behaviour of such bridge has
largely to do with the majority of the structure being made up steel, a material less
vulnerable to seismic events than concrete, due to its inherently available ductility. The
need for considerable attention to be paid to reinforced concrete bridges was, hence,
confirmed.

Figure 2.3 – Oakland Bay Bridge: collapsed portion of the upper deck.

2.6
 Seismic Assessment of Bridges

2.2.2 Northridge earthquake

The Northridge earthquake, named after a neighbourhood in the city of Los Angeles,
California, occurred in 1994, lasting for about 45 seconds. With one of the highest ever
measured ground accelerations in urban North America, in the range of 1.0g, it became one
of the costliest natural disasters in the United States history, causing seventy two deaths
and over 9000 injured.

After the Loma Prieta earthquake, in 1989, a retrofitting program was begun over
Californian bridges and was yet to conclude when the Northridge earthquake took place.
There were about 2000 bridges in the epicentral region; six of these bridges experienced
failure and four others were so badly damaged they had to be replaced. The shaking was in
the origin of considerable damage in the vast freeway network, with particular emphasis to
the Santa Monica Freeway, which serves millions of commuters everyday, and the
Antelope Valley Freeway, illustrated in Figure 2.4.

Figure 2.4 – Collapsed sections in Santa Monica freeway (left) and Antelope Valley Freeway
(right).

The failure of those bridges was primarily due to the failure of the supporting columns that
had been designed and constructed before 1971, a critical timing, given that after the San
Fernando earthquake the standards for earthquake design began to be “toughened”
considerably. The lack of proper concrete-core confinement, resulting in “birdcadging”
effect of steel reinforcement, or poor behaviour of flared pier tops are some typical
examples of column failure – see illustration in Figure 2.5.

2.7
Seismic Assessment of Bridges

The retrofitted bridges did not suffer serious damage, which consisted of, according to
Yashinsky (1998b), minor cracks to the slope paving and settlement of the approach. In
contrast, damage to unretrofit bridges at the same site was extensive.

Figure 2.5 – Failure of columns: insufficient confinement (left); shear failure of flared column with
transverse reinforcement slipping (right).

Figure 2.6 illustrates the contrast between pier failure, due to lack of efficient
reinforcement, and satisfactory behaving pier, provided with retrofitting (Cooper et al.,
1994).

Figure 2.6 – Failure of column due to failure of circular confinement steel (left); good behaviour of
column retrofitted with steel jackets (right).

Indeed, bridges designed and built before 1971 performed worse than those designed
according to most recent standards and piers were the most damaged components (Basoz
and Kiremidjian, 1998). Still, the damage caused by the earthquake revealed that some
structural specifications did not perform as well as expected, such as the case of two

2.8
 Seismic Assessment of Bridges

bridges, both constructed shortly after the 1971 earthquake, on the Simi Valley-San
Fernando Valley Freeway. These bridges presented severe column distress that resulted in
bridge failure.

Other damage to bridges included spalling and cracking of concrete abutments, spalling of
column-cover concrete, settlement of bridge approaches, and tipping or displacement of
both steel- and neoprene-type bearings. Moreover, bridges with non-monolithic abutment
types, discontinuous spans and single column bents performed poorly.

2.2.3 Kobe earthquake

The Great Hanshin earthquake, often referred to as Kobe earthquake, due to the city that
was majorly stricken, occurred in January of 1995. It killed more than six thousand people,
4600 from the city of Kobe only. Even though experiencing ground shaking on a regular
basis, this was Japan’s worst earthquake since 1923. Comparison to the Northridge
earthquake, which had occurred only one year before, was inevitable and less fortunate to
the Kobe city side, given that damage was much greater than the one caused by the
American quake. Such difference comes mainly from the type of ground beneath Kobe and
the light, unreinforced construction type in masonry and wood.

The damage in highway bridges was one of the main post-earthquake images, coming from
several different scenarios: substructure failure, originating from simple shear failure in
reinforced concrete columns; premature shear failure at terminations of longitudinal bars
with insufficient development lengths; extensive failure of steel columns, the first in the
world; soil liquefaction, leading to settlements and tilting of foundations and substructures
and lateral spreading of ground associated with soil liquefaction, causing movements of
foundations.

One of the most impressive structural disasters was the collapsed eighteen spans in Fukae
viaduct, inserted in Route 3, within the elevated Hanshin Expressway, illustrated in Figure
2.7.

2.9
Seismic Assessment of Bridges

Figure 2.7 – Collapsed Fukae Viaduct (left) and Premature Shear Failure of Reinforced Concrete
Column, Fukae Viaduct (right) (Kawashima, 2007).

The viaduct was designed in accordance with the 1964 Design Specifications and was
completed in 1969. The insufficient code provisions led to important problems in the
design of the viaduct: the overestimated allowable shear stress, the insufficient
development length of longitudinal bars terminated at mid-height and the insufficient
amount of tie bars. The combination of these aspects originated the aforementioned
extensive premature shear failure. Comparing the performance of this viaduct to the one of
the parallel Route 5 viaduct of the same mentioned Expressway, completed in the early to
mid-1990s, enhances the importance of the new seismic design stipulations, given that the
latter behaved far better than the former, despite the potentially worse soil conditions.

Figure 2.8, taken from Kawashima’s lecture notes (Kawashima, 2007), sketches the
probable failure mechanism of the viaduct, where is extremely visible the importance of
the proper reinforcement of the piers, especially at the base level, in order to assure a
sufficient ductility level. To such extent, the adequate behaviour under high seismic
demand will require good actual anchorage conditions.

2.10
 Seismic Assessment of Bridges

Figure 2.8 – Failure mechanism of Fukae Viaduct (Kawashima, 2007).

Soil failure was equally frequent as many of the bridges were founded on sand–gravel
terraces (alluvial deposits) overlying gravel–sand–mud deposits at depths of less than ten
meters, a condition which is believed to have led to site amplification of the bedrock
motions. Several situations of liquefaction and lateral spreading, resulting in permanent
substructure deformations and loss of superstructure support also occurred (Moehle and
Eberhard, 1999). The collapse of the Nishinomiya-ko Bridge approach span, Figure 2.9, is
good example of how the site conditions largely increased the vulnerability.

Figure 2.9 – Nishinomiya-ko Bridge approach span collapse (Kobe Collection, EERC Library,
University of California, Berkeley).

2.11
Seismic Assessment of Bridges

Additional collapse of bridges resulted from damage of unseating prevention devices and
from forces transfer through unseating prevention devices.

2.2.4 L’Aquila earthquake

In April, 2009, the L'Aquila earthquake occurred in the region of Abruzzo, in central Italy,
featuring a magnitude of 6.3 on the Richter scale. Its epicentre was near L'Aquila, the
capital of Abruzzo, which together with surrounding villages suffered most damage. The
earthquake was felt throughout central Italy and 308 people are known to have died
making it one of the deadliest earthquakes to hit Italy since the 1980. In spite of having
occurred in a well developed country, with relatively advanced seismic regulation, the
stroke effects were significant mostly because of the high number of medieval, historical
constructions and unreinforced masonry buildings, known for being quite susceptible to the
seismic action.

Not only medieval structures in L’Aquila suffered damage. Many relatively modern
buildings, such as a local university dormitory, with nonductile concrete, soft-storey
irregularities or new precast, suffered substantial damage as well. Damage extent even
reached some buildings that were believed to be earthquake resistant. L'Aquila Hospital's
new wing, which opened in 2000 and was thought capable of safely facing a strong
earthquake, suffered extensive damage and had to be closed. This was, at some level,
surprising, given that the event was moderate and the country modern. Failure was
admitted as primarily due to the lack of consistent seismic design (Miyamoto et al., 2009).

Nevertheless, transportation facilities, namely bridges, behaved reasonably, suffering

minor damage. Indeed, major damage to such sort of structures would be expected for a
higher magnitude earthquake or longer duration of strong ground shaking. The worst case
corresponded to a short, 35 meters long, three-span, continuous reinforced concrete bridge,
not far from the epicentre, which collapsed onto the riverbed, as visible in Figure 2.10.

2.12
 Seismic Assessment of Bridges

Figure 2.10 – Bridge collapsed on the Aterno river, near Fossa (Grimaz and Maiolo, 2010).

The four reinforced concrete columns with hexagonal sections broke at the pier-slab
connections, sliding sideways and penetrating the deck slab. Furthermore, it is believed
that vertical downward motion was relevant to the failure of the bridge.

Other occurrences consisted of damage at the frame piers top of another three span
continuous bridge, with additional damage of the superstructure by tensile cracking due to
the movement of the piers towards the centre of the river, caused by movements of the
embankment on both sides. A masonry arch bridge, which had collapsed before and was
repaired by filling crashed limestone, collapsed again during the earthquake, due to
probable movement of abutments, resulting in the loss of the arching effect. Finally, some
of the viaducts within the A24 expressway near L’Aquila were affected by the earthquake,
although the expressway itself did not collapse anywhere (Aydan et al., 2009). Again, a
major viewpoint is that, given the moderate nature of the ground motion, well seismic
designed bridges should not have collapsed, which indicates a need for code review or a
thorough existing structures safety assessment.

2.2.5 Haiti earthquake

The Haiti earthquake, one of the latest ground shaking events and probably one of the most
catastrophic ones, took place in February, 2010, with massive human losses, taking the
lives of more than 230 000 people. The consequences were devastating mainly due to the
fact that there was a generalized lack of attention to earthquake-resistant design and
construction practices and the poor quality of much of the construction. Indeed, the

2.13
Seismic Assessment of Bridges

historical pattern of earthquakes in Haiti indicates that an earthquake of high magnitude

could strike southern Haiti near Port-au-Prince at any time.

According to field reports, it was not recognised any bridge failure due to the earthquake.
Within Port-au-Prince, the majorly affected city, most of the crossings over streams were
accommodated by box culverts, Figure 2.11, which did not appear to be damaged. In any
case, these crossings can be hazardous in the future, given that large amount of garbage
accumulated upstream such culverts may combine with silt and debris to prevent water
from passing through the culverts (Eberhard et al., 2010).

Figure 2.11 – Box coulvert in Port-au-Prince (left); damage to shear key at intermediate support
of bridge (right) (Eberhard et al., 2010).

There were, anyway, river crossings on Nationale No. 2, most of them spanned by bridges
with precast girders resting on cast-in-place reinforced concrete bents and supporting a
cast-in-place deck. In two of such bridges damage was observed. One of them, the bridge
over the Momance River presented minor pounding damage at the shear key at one of the
intermediate supports that probably had not been adequately and/or properly reinforced. A
similar bridge, in the Carrefour section of Port-au-Prince, suffered damage in the external
shear keys at both intermediate supports, apparently caused by the lack of hook anchorage
at the end of the top beam reinforcement.

One of the major findings of the scrutiny of Haiti’s earthquake was that, even with the
absence of seismographic stations during the main earthquake or its largest aftershocks,
which could estimate accurately the intensity of ground motions, the indirect suggestion is
that the earthquake did not produce ground motion enough to severely damage

2.14
 Seismic Assessment of Bridges

well-designed structures. Several buildings and bearing-wall structures survived the

earthquake with no signs of large deformation demands. Similarly, bridges near the
epicentre suffered only minor damage and were able to function immediately after the
earthquake. Such picture confirms, unfortunately, the lack of seismic-resistant conditions
that still persist throughout the globe, with large incidence in poor countries.

2.2.6 Christchurch earthquake and Tōhoku earthquake and tsunami

The year of 2011 started early to feature severe earthquake disasters. Both events are
herein presented together, not only for having taken place the same year but also because
both occurred in so-called developed, wealthy countries, New Zealand and Japan.

The 2011 Christchurch earthquake was a 6.3-magnitude earthquake that struck the
Canterbury region in New Zealand's South Island on 22 February 2011, causing
widespread damage and multiple fatalities although no bridge collapses have been
reported. It followed nearly six months after the 7.1 magnitude 2010 Canterbury
earthquake that caused significant damage to the region but no direct fatalities. Analysts
estimated that the earthquake could cost insurers US$12 billion. Of the 3000 buildings
inspected within the main avenues of the central city, around 45% have been given red or
yellow stickers to restrict access because of safety problems and one thousand were
expected to be demolished (around 25% of the total number of buildings). Many heritage
buildings were also given red stickers after inspections but, in general, not many buildings
have collapsed. Some examples include two six-storey buildings: the Canterbury
Television building and the PGC Building, illustrated in Figure 2.12, a reinforced concrete
building that had been constructed in 1963-1964, which drew significant amount of
attention of the high vulnerability of pre-1970s mid- and high-rise buildings. The poor
seismic behaviour of these buildings includes column shear failure, beam-column joint
failure, onset of soft-storey failure, shear wall failure, etc. The 26-storey Grand Chancellor
building, Christchurch's tallest hotel, was reported to be on the verge of collapse and was
indicated for demolition pointed to be demolished over the following six months. While
damage occurred to many older buildings, particularly those with unreinforced masonry
and those built before stringent earthquakes codes were introduced, high rises built within
the past twenty to thirty years performed well (Kam, 2011).

2.15
Seismic Assessment of Bridges

Figure 2.12 – The Pyne Gould Corporation (PGC) Building following the 2011 Christchurch
earthquake.

The 2011 Tōhoku earthquake, officially named the Great East Japan Earthquake, was a
9.0-magnitude 9.0 undersea megathrust earthquake off the coast of Japan that occurred on
March 11th. It was considered the most powerful known earthquake to have hit Japan, and
one of the five most powerful earthquakes in the world overall since modern record-
keeping began in 1900. The earthquake triggered extremely destructive tsunami waves,
causing numerous casualties (approximately 14 616 deaths, 5278 injured and 11 111
people missing), destruction of infrastructures and a number of nuclear accidents. The
overall cost could exceed $300 billion. Structurally, over 125 000 buildings were damaged
or destroyed, as well as roads, railways and bridges, and a dam collapsed.

Apart from the obviously numerous structural failure occurrences, such as the bridge
failure illustrated in Figure 2.13, the 2011 Japan earthquake caused a serious nuclear
accident, affecting, in the present and future years, the planet natural resources, as well as
worldwide population.

2.16
 Seismic Assessment of Bridges

Figure 2.13 – Bridge collapse in 2011 Japan Earthquake and Tsunami.

The devastating way how ground shaking recently stroke New Zealand and Japan
constitutes a sad, yet important, reminder of how vulnerable modern societies, even in
relatively well-prepared countries, still are to seismic events.

2.2.7 Summary

The purpose of this section was to identify and briefly describe types of
earthquake-induced damage to bridges and, where possible, to identify the causes of the
damage. This is a task of recognized importance because of the obviously dramatic and
severe consequences associated to bridge failure as well as the need for taking lesson from
those disasters. It is not, however, a simple endeavour, given that damage process is
complex and comes from the interaction of several contributing variables. Moreover, when
damage is intense to the point of concealing detail and damage deformation itself, some
speculative judgment may be needed for each particular collapse, being difficult to
generalize the causes of bridge damage.

2.17
Seismic Assessment of Bridges

In the past, bridges responded to earthquakes in a very particular manner, each case
depending on the characteristics of the ground motion at the particular site and the
construction details of the specific bridge. Despite such individuality, it was still possible
to learn from past earthquake damage, once that many types of damage occur repeatedly.
With such observation, over the years, bridge seismic design practices evolved, largely
reflecting lessons learned from performance in past earthquake events.

Within the North American situation, the rapid change in bridge construction practice
following the 1971 San Fernando earthquake was evident (Hall, 1995). Prior to that,
California design and construction practice was based on significantly lower design forces
and less stringent detailing requirements compared with current requirements. With the
1989 Loma Prieta and the 1994 Northridge earthquakes, insightful study on bridge
structural failure has gained strength, with the recognition for a needed review and update
of seismic design, code and construction criteria. Indeed, major collapse situations due to
the insufficient reinforcement of the piers, poorly designed connections between piers and
deck, subjected to high demand, were observed, denoting a turning point towards more
refined seismic specifications.

Moreover, although typically associated to older bridge rupture situations, soil failure still
kept on being the cause of several collapses, influenced by proximity of the bridge to the
fault and site conditions. Site conditions were clearly responsible for the bridge response in
the 1989 Loma Prieta earthquake leading to the conclusion that local site conditions have
significant impact on amplifying strong ground motion, and the subsequent increased
vulnerability of bridges on soft soil sites. During the 1995 Kobe earthquake, significant
damage and collapse likewise occurred in elevated roadways and bridges founded on
alluvial deposits, which is believed to have led to site amplification of the bedrock motion.

In spite of the considerable effort that has been made in the past two or three decades in the
improvement of seismic reinforcement and regulation, there are still disastrous events on a
regular basis, compromising numerous existing structures as well as the lives of those who
daily make use of them. Two of the most recent earthquakes that have been mentioned,
L’Aquila and Haiti, prove how, in the past couple of years, it is still possible, not to say
likely, for high structural and human damage to occur, even with modern construction and
regulations. In such cases, failure took place because of the lack of attention in design and

2.18
 Seismic Assessment of Bridges

construction to the possibility of earthquakes or poor construction practices. The Haiti case
foresees probable future scenarios of huge metropolis being hit by earthquakes, suffering
from massive destruction and human lives taking. The equally recent New Zealand and
Japan earthquakes have again proved their destructive capacity, even if in presence of
seismically prepared countries. The certainty that there is still a long path to run, with
respect to seismic design and performance, is probably an equally important lesson to
constantly keep in mind.

2.3 The seismic action

As one of the key issues in any vulnerability analysis, the load definition requires
considerable attention and sometimes, such as the case of the seismic action, thorough
calibration. Moreover, reduction of dispersion is often needed given the high variability
and uncertainty that characterizes earthquakes. The most logical and insightful way of
representing the seismic action is by means of accelerograms, the time-history registration
of the ground motion accelerations (eventually displacements or velocities) that are
measured throughout the duration of the shaking. Depending on the sort of analysis that is
being employed, other simplified representation possibilities, such as response spectra, that
are usually implemented in code provisions or equally simplified methods, such as
nonlinear static procedures, may be used. On the other hand, if a nonlinear dynamic
analysis is intended, the use of accelerograms is fundamental and should be handled with
care, given the disparity that often occurs within the several available types of records.

Accelerograms play an extremely important role when performing dynamic analyses,

especially if nonlinear behaviour is considered. As an element which characterizes seismic
action, it is possible to adopt accelerograms associated to different intensity levels and
occurrence probability. At the same time, shape and energetic content characteristics of
each record strongly influence any structural response. Thus, taking the pointed reasons in
account, the importance of carefully evaluating the set of accelerograms to use is easily
justified. A lot has been done in the recent past within this domain, particularly regarding
selecting and scaling techniques, which is related to type of records and intensity level,
respectively.

2.19
Seismic Assessment of Bridges

A very relevant aspect that can be deepened is the option between using real or artificial
ground motion records, putting synthetic ones aside. This is actually the main issue: what
sort of ground motion records is best to use? The choice for any of these types naturally
involves each one’s advantages and potentialities. At the same time, the interest remains on
a practical and systematic analysis, a feature that may be associated to artificial records,
but on the other hand one should avoid seismic records that may conduce to excessive
structural effects, with loss of reality and overestimating design. Additionally, the sort of
structure that is being object of analysis can be conditioning as well.

2.3.1 Selection and scaling of real records

The use of real records is naturally advantageous, given that the analysis will definitely
become more genuine. Moreover, online databases of real ground motion records are
becoming more and more available, for every typical spot of intense earthquake activity. It
seems, hence, that an interesting combination of such a realistic as possible analysis, which
is adequate, given the specific features of the seismic action, together with what is intended
in a seismic analysis, arises with the use of real records, avoiding the need for generating
artificial ones. Nevertheless, it is immediate to realise that the record-to-record variation
associated not only to the real earthquake records parameters, such as duration, magnitude,
epicentral distance, peak ground acceleration but to the subsequent structural engineering
demand parameters median values as well, namely response spectra or ductility demand,
will be higher in a great amount. Furthermore, not all the magnitude-distance-soil
combinations are covered and spectra are generally not smoothed.

Although it is still not current practice, Eurocode 8 itself already considers the possibility
of using real earthquake records. In fact, it recommends them, in Part 2 – Bridges (CEN,
2005b), comparing to the use of artificial ones, when performing nonlinear analyses,
whereas for buildings, Part 1 (CEN, 2005a), no recommendation is given. With the choice
for real records type in nonlinear dynamic analyses, a great number of other aspects arises,
which has been the study target of quite wide research activity, with some considerable
progress in the past few years.

In a logical proceeding order, the first step is to perform an initial selection of the
accelerograms to use. Should it be randomly or based on geotechnical parameters or any

2.20
 Seismic Assessment of Bridges

alternative techniques? Studies from Shome et al. (1998) or Bommer and Acevedo (2004)
have addressed this issue whether selecting and cataloguing records based on magnitude
and occurrence distance, in the case of the former, whether proposing a selection based on
a code response spectrum matching procedure, minimizing the residual distance between
the spectra.

After the initial selection of records, there is the need for a scaling technique to put them at
a same level. The effectiveness of a certain technique may depend or not of ground motion
parameters and should be verified on different structural systems. The majority of the past
research studies were developed focusing intensity-based methods to scale ground
motions, which keep the original non-stationary content and only modify its amplitude.
Instead, spectral matching techniques that modify the frequency content or phasing of the
record to match its response spectrum to the target spectrum can be considered.

Generally, two categories for intensity-based scaling techniques can be defined, according
to the nature of parameters in which they are based. These can derive form records’
inherent characteristics or from the corresponding response spectra. The first significant
attempt to establish a comparison between several different intensity-based techniques was
performed by Nau and Hall (1984). Such work has proved that scaling ground motions to
match a target value of peak ground acceleration, which was the earliest approach, yielded
inaccurate estimates with large dispersion. The study analysed six scaling techniques based
on ground motion data and two based on response quantities used to normalize earthquake
response spectra. The results seemed to show that traditional techniques, such as the ones
considering peak ground reference values, tend to lead to high scatter, while spectrum
intensities or Fourier amplitudes were considered as promising alternative scaling
parameters. That study was, although, carried out considering only response spectra
scaling, thus, effects on single-degree-of-freedom (SDOF) systems. The twelve selected
records followed no particular criteria, such as the conventional seismologic properties,
magnitude or epicentral distance, for instance. The purpose of covering a relatively wide
range on those variables was sought, though. Similar findings can be found in work from
Miranda (1993), Vidic et al. (1994) or Shome and Cornell (1998).

Later, Shome et al. (1998) looked at the use of real accelerograms from a different
perspective, trying to address other issues, like the initial selection of records or the

2.21
Seismic Assessment of Bridges

appropriate number to use. The initial selection and cataloguing was based on the
earthquakes magnitude and distance. Nevertheless, based on the nonlinear response of a
five-DOF steel structure, the study ended up by concluding that those ground motions
parameters do not have great influence. Different scaling techniques were once more
tested. On this matter, conclusions pointed towards the consideration of spectral
acceleration at the fundamental frequency of the structure as a better performing technique,
recommending the use of peak ground acceleration to be disregarded. It is also noted by
the authors that the conclusions were based on the analysis of the single
multi-degree-of-freedom (MDOF) referred structure, highlighting the fact that, namely on
scaling parameters, little work has been done on that sort of structures.

More recently, works conducted by Kappos and Kyriakakis (2000) and Kurama and
Farrow (2003) revisited the natural records issue. The former presented some
developments, mainly focusing on comparative analysis and new scaling parameters, and
different seismotectonic environments on elastic or inelastic spectra, hence, SDOF
systems. Also MDOF structures were analysed, particularly multi-storey frames.
Conclusions on elastic spectral scaling indicated that generally, for intermediate and long
period range, velocity-related parameters performed well, whilst, in inelastic conditions,
spectrum intensity scaling produces better effects. Concerning effects on MDOF systems,
spectrum intensity scaling continues to be the choice, either based on elastic spectra or
inelastic pseudo-velocity spectrum. Similarly, Kurama and Farrow (2003), in their study,
went back into ground motion scaling methods, analysing seven different techniques,
insisting on different site conditions, such as soil profile and epicentral distance. SDOF and
MDOF systems were tested with twenty records per soil profile. The study proposes
another parameter to scale the records, which generally performs better across the
considered site soil characteristics, for different structural types.

All the other scalar intensity measures, in general, have hence been found to be inaccurate
and inefficient as well. Moreover, within most of the studies, with the exception of the
work from Shome et al. (1998), none of the tested parameters considered any property
of the structure to be analyzed. By including vibration data of the structure when scaling
records to a target value of the elastic spectral acceleration, for instance, from a code-based
design spectrum or PSHA-based uniform hazard spectrum at the fundamental vibration
period of the structure, results became actually quite improved, nevertheless, for structures

2.22
 Seismic Assessment of Bridges

whose response is dominated by their first-mode. When structures respond significantly in

their higher vibration modes or far into the inelastic range, accuracy and efficiency
decrease, as shown by the works of Kurama and Farrow (2003), Mehanny and Deierlein
(2000), Alavi and Krawinkler (2004). There have been, though, approaches to scalar
intensity measure parameters that consider higher modes response, combining spectral
accelerations at the first two periods, developed by Bazzurro (1998) and Shome and
Cornell (1999). In spite of improving accuracy, such measure remained inefficient for
near-fault records with a dominant velocity pulse, as observed by Baker and Cornell
(2006).

Alternative approaches, such as scaling earthquake records to minimize the difference

between elastic response spectrum and target spectrum have also been followed, in some
cases, implementing genetic algorithms (Alavi and Krawinkler, 2004; Kennedy et al.,
1984; Malhotra, 2003; Naeim et al., 2004).

Intensity based measures can still be distinguished depending on whether they are based on
the elastic response of the structure, relying on the structural period only, or they account
for the inelastic features, taking structural strength into consideration. When the inelastic
spectral deformation is significantly larger than corresponding elastic one, the elastic
response based parameters, such as the ones that are tested in the studies previously
mentioned, become less appropriate. The recognition of such situation led to the recent
proposal of scaling parameters that are based on inelastic deformation spectrum, with
improved estimates of median values and dispersion of control parameters (Bazzurro and
Luco, 2006; Luco and Cornell, 2007). Studies carried out by means of dynamic analyses of
generic frames subjected to different intensity levels, revealed promising results obtained
when records were scaled with parameters defined as the inelastic deformation of the first
mode equivalent single degree of freedom system, especially when compared to elastic
response based ones (Tothong and Luco, 2007; Tothong and Cornell, 2008).

In terms of existing regulations, available American guidelines concerning building codes

have addressed this issue prescribing methodologies to the scaling of ground motion
records for a site-specific hazard. International Building Code (IBC) (ICC, 2006) and
California Building Code (CBC) (CBSC, 2007) require earthquake records to be scaled
according to the ASCE 7-05 provisions (ASCE, 2005). Provisions are established for 2D

2.23
Seismic Assessment of Bridges

analysis of regular structures, establishing that ground motions are scaled such that the
average value of the 5%-damped elastic response spectra for a set of scaled motions is not
less than the design response spectrum over a specific period range. Specific conditions are
defined for structures having plan irregularities or without independent orthogonal lateral
load resisting systems where 3D analyses need to be carried out.

Following the structural response based scaling parameters trend, a recent advanced
method has been proposed by Kalkan and Chopra (2010), consisting of a modal-pushover-
based scaling (MPS) technique to scale ground motions for use in nonlinear dynamic
analysis of buildings and bridges. Ground motions are scaled to match (to a specified
tolerance) a target value of the inelastic deformation of the first-mode inelastic single-
degree-of-freedom system whose properties are determined by first-mode pushover
analysis. The authors consider it appropriate for first-mode dominated structures, extending
it, though, for structures with significant contributions of higher modes by considering
elastic deformation of higher-mode SDOF systems in selecting a subset of the scaled
ground motions. Within the application of the methodology, two bridges and six actual
buildings, covering low-, mid-, and high-rise building types in California, were tested,
confirming the accuracy and efficiency of the MPS procedure as well as its superiority
over the ASCE 7-05 (ASCE, 2005) scaling procedure.

2.3.2 Number of records

Following the initial selection and scaling of records, in order to carry out the seismic
analysis, the number of real records to use assumes critical importance. Basically, such
issue has the underlying idea that one expects a certain number of different ground motion
records to be able to estimate, trustfully, mean or median results, avoiding a cumbersome
procedure, made of too many runs. Shome et al. (1998) dealt with the number of ground
motion records to consider, in their aforementioned study, in a statistical way, through the
definition of a confidence interval necessary to guarantee a dispersion level. That
dispersion turned out as acceptable if seven records were used. Also the work from
Bommer and Acevedo (2004) has addressed this issue, mainly from a qualitative point of
view, analysing current design codes guidance and recent research on that matter. Their
conclusions, in accordance with the existing variability in recommendations, tried to leave

2.24
 Seismic Assessment of Bridges

the subject as flexible as possible, in a sense that each case should be considered
individually, according to variables of multiple natures.

A recent work from Bradley (2011), estimated the seismic demand from seismic response
analyses, from a probabilistic point of view, making use of the 84th percentile of the
distribution of the sample mean seismic demand as the design seismic demand. That study
took into account the number of ground motions considered, how the ground motions were
selected and scaled and the differing variability in estimating different types of seismic
response parameters, within a proposed simple three-step procedure suitable for routine
design implementation.

From what has been mentioned, seems clear that using real earthquake records is no simple
task, involving several aspects that conspicuously need to be looked at with some care. The
advantages of using such genuine sort of records, and corresponding response spectra, to
represent the seismic action has motivated intense research in the recent past, with the
purpose of reducing the natural variation that is found. Some issues, such as scaling
techniques, have been focused quite more thoroughly than others previously pointed out.
Apart from that, the conducted studies until the moment are characterized by large
heterogeneity, as far as topics and variables in the analyses or considered structural types
are concerned. Moreover, the tendency has been to approach the problem by the inelastic
structural effects side, rather than by the records’ characteristics one, which will certainly
be the basis of future code provisions.

As stated before, some studies considered geotechnical parameters while others denied
their importance when selecting real accelerograms; some based their conclusions on the
study of SDOF systems while others tried to call out results from MDOF systems; some
follow certain criteria to initially select the records or try to reasonably define how many to
use, others simply use an apparently sufficient number, in order to cover a considerable
range of possibilities. Moreover, and moving forward within the use of accelerograms in
seismic analyses, artificial records still represent a viable alternative and should therefore
be eventually confronted to real records, something that seems not to have still been
properly addressed. The option between one of the types, or even the possible indifferent
choice for one of them, relating such option with the structural type that is being looked
thorough, bridges in particular, is therefore a topic of clear significance.

2.25
Seismic Assessment of Bridges

2.4 Nonlinear seismic analysis – methods for structural response prediction

Following the proper characterization of the seismic action, its multiple level uncertainties
and dispersion, the seismic analysis itself, including the explicit consideration of the
nonlinear effects, is an equally complex task, involving several aspects that need thorough
concern. The main distinction between basic possibilities of approaching the problem has
been, however, quite simple and immediate; the choice has laid on inelastic static analysis,
typically recurring to pushover based algorithms, or inelastic dynamic analysis.

As noted by Elnashai (2002), inelastic static analysis is widely implemented in design

office environment and is by far more familiar to practitioners than the dynamic
alternative. Reasons such as complexity of time-integration algorithms or difficulties in
defining damping, as well as the variation induced in acceleration and force related
quantities by the combination of both issues, are pointed out by the mention state-of-art
review to justify the still actual tendency. Other motives can be added, such as the
significant amount of output information that nonlinear dynamic analysis yields, or the
extensive number of parameters that need to be calibrated for the analysis to work out
properly. Appropriate adjustment of such parameters may become a harsh task if the user
has no experience or knowledge in advanced nonlinear analysis, which is probably the case
of most of the design office structural engineers. Furthermore, the application domains of
both sorts of analysis have not necessarily been crossing each other, something that has
enabled a certain development in parallel. Static analysis has proven to represent dynamic
response with a satisfactory accuracy level and has seen considerable progress, starting
with quite simple attempts (Freeman et al., 1975; Saiidi and Sozen, 1981; Kunnath et al.,
1992; Lawson et al., 1994) but recently evolving to proposals of advanced pushover-based
methodologies. Nevertheless, nonlinear static analysis is mostly useful and intuitive for the
estimation of capacity. Structural effects, or demand, need to be determined later, through
the inclusion of response spectra, not in a direct fashion, hence. Contrarily, dynamic
analysis has the advantage of allowing structural members, with corresponding capacity
duly modelled, to be submitted to the seismic action, knowing demand time-step by time-
step, which again reinforces the idea that nonlinear dynamic analysis, despite possible
complexity issues, has its place.

2.26
 Seismic Assessment of Bridges

It is still not clear how the nonlinear analysis domain is to be shared between static or
dynamic analysis. It seems clear that both domains tend to coexist and intersect each other.
When static procedures are too simplistic for structures that demand higher accuracy,
dynamic analysis will work as a complementing alternative, and vice-versa. There is no
clear evolution trend when it comes to analysis types: in spite of being considered the most
true and complete methodology, inelastic dynamic analysis is still not implemented and
nonlinear static procedures, as well as other recent simplified alternatives, are nevertheless
a great improvement over presently employed elastic evaluation procedures.

In the beginning of the past decade, Elnashai (2002) presented an interesting diagram,
reproduced in Figure 2.14, illustrating the static and dynamic interaction domains, as
function of strong-motion peculiarity and structural irregularity.

Figure 2.14 – Effect of development of advanced static analysis on model-input motion

application domain (Elnashai, 2002).

Although it has been argued that static domain tends to expand as a consequence of higher
refinement in nonlinear static procedures that have undergone, it is to the author’s belief
that such tendency merely reflects the actual analysis methodologies distribution within
design office environment and that a future scenario will consist of increasing parity
between such two major domains.

In what follows a brief discussion on the evolution and conditions of implementation of the
different analysis alternatives is carried out.

2.27
Seismic Assessment of Bridges

2.4.1 Nonlinear static analysis – pushover

The basic concept behind the technique of a pushover analysis has actually no rigorous
theoretical basis and is to assume that, as referred in (Elnashai, 2002), “if a set of actions or
deformations can be found such that a particular response mode, or a combination of
modes, is represented statically, then the response of the structure under a monotonically
increasing vector of actions or deformations may replace results from dynamic analysis”.
The major assumption is simultaneously the main limitation of any pushover analysis, used
to assess the capacity of the structure. In order to reach the structural effects caused by
certain seismic action, the capacity curve must be intersected with the demand spectrum,
which will define the performance point, often referred to by the corresponding target
displacement. Herein, different possibilities have been suggested by different authors:
Freeman (1998) proposed a method making use of elastic overdamped spectrum, whereas
Fajfar (1999) was pioneering in implementing inelastic spectra in nonlinear static pushover
procedures.

Other disadvantages of the pushover-based methods for estimating demand are the needed
transformations of the MDOF structures under analysis into equivalent SDOF systems,
which, understandably, carry a considerable degree of simplification. Once more, the
formulation of such SODF system is not unique, each author following a slightly different
path, based on the common principle that the deformed shape of the structure is not
excessively altered during the dynamic loading. Consequently, the inclusion of more than
the first mode contribution to the analysis, which can play an extremely important part, has
become one of the key challenges within recent proposals of pushover techniques. If bridge
structures, of well known irregular and higher mode dependent behaviour, are kept in
mind, the need for such endeavour becomes even more apparent.

Krawinkler and Seneviratna (1998) addressed the use of pushover analysis in seismic
performance evaluation, summing up advantages and pitfalls, recognizing it as a valuable
tool in today’s limited states of knowledge and practice. A pushover analysis used for
demand prediction is certainly not highly accurate but neither the seismic action nor
capacity estimates are. According to such review, the accuracy is essentially affected by
the aspects around the estimation of the target displacement and the selection of load
patterns that supposedly deform the structure in a similar way to expected when an

2.28
 Seismic Assessment of Bridges

earthquake occurs. The authors emphasize difficulties and ambiguities around

modifications needed to be applied to elastic demand spectrum, fundamental for target
displacement location, to account for yield strength, stiffness degradation or pinching
effect, strength deterioration, P-delta effects, viscous damping, among others. The load
pattern is considered to be even more critical and is pointed out by the authors as probably
the weakest point of the pushover analysis procedures, mainly if invariant patterns are used
in long period structures with localized yielding mechanisms. Indeed, regardless the load
pattern that is chosen, it will trigger certain deformation modes and miss others that are
propelled by dynamic response and inelastic response characteristics. Major conclusions
point towards the acknowledgment of the potential use of inelastic pushover analysis in
demand prediction in many cases, able to provide information more relevant than static or
dynamic analysis, mostly for structures that vibrate primarily in the fundamental mode.

As an attempt to overcome the mentioned drawbacks, namely the lack of higher modes
contribution or the invariant nature of the load pattern, more advanced algorithms of
pushover analysis have been proposed in the recent past, such as a Multi-Modal Pushover
Procedure, by Paret et al. (1996), later improved by Moghadam and Tso (2002). A modal
pushover analysis method, consisting of the repetition of single pushover analysis,
corresponding to each relevant mode, with quadratic combination of results in the end, has
been proposed by different authors Chopra in the beginning of the past decade (Chopra and
Goel, 2001, 2002). This is probably the most intuitive way of considering higher mode
effects, although not necessarily the most advantageous. Several improvements and
application studies to probe the suitability of the method to different structural types have
been presented thereafter (Chintanapakdee and Chopra, 2003; Chopra, 2005; Chopra and
Chintanapakdee, 2004; Chopra and Goel, 2004; Goel, 2005; Goel and Chopra, 2004,
2005a, 2005b).

From a different perspective, other alternative procedures have been recently proposed,
working on the type of pushover analysis that is employed. With the purpose of including
higher mode effects, as well as better accounting for degradation characteristics with
increasing loading and including the characteristics of the input ground motion, several
authors, Bracci et al. (1997), Gupta and Kunnath (2000), Elnashai (2001), Antoniou and
Pinho (2004) or Antoniou et al. (2002), proposed adaptive or fully adaptive pushover
analysis techniques. Frequently applied together with spectrum scaling, it consisted of the

2.29
Seismic Assessment of Bridges

application of displacements, or forces, in an adaptive fashion, that is, with the possibility
of updating the loading pattern according to the structural properties of the model at each
step of the analysis, as in Figure 2.15, left. In such way, the structural stiffness at different
deformation levels is considered in the evaluation of the new forces, the system
degradation and period elongation can be accounted for and the alteration of the inertia
loads during dynamic analysis for different deformation levels may be successfully
modelled. The advantage in such procedure is to avoid the repetition of independent
pushover analysis, as many as the relevant vibration modes, using a unique pushover
analysis that, hopefully, includes all the relevant features neglected by the conventional
analysis. The innovative algorithm proved to be numerically stable, even in the highly
inelastic region, whereas the additional modelling and computational effort, with respect to
conventional pushover procedures, is negligible. At most, it can be argued that additional
complexity may come in terms of access to an efficient eigenvalue solver, scaling forces
by spectral ordinates, updating applied forces or displacements vectors. Such possible
complications are, however, on the programmer, rather than the user side, hence, no
complexity is effectively added.

In Figure 2.15, right, results taken from work of the adaptive pushover authors show the
proximity of response, using the adaptive technique for a building frame, to dynamic
analysis, when compared to conventional pushover with two different load patterns.

Figure 2.15 – Adaptive pushover: shape of loading vector updated at each analysis step (left)
(Pinho and Antoniou, 2005); Adaptive vs. conventional pushover procedures (right) (Antoniou et
al., 2002).

2.30
 Seismic Assessment of Bridges

Conventional pushover, regardless the load shape, seems to provide a certain envelope to
dynamic analysis results whilst adaptive pushover is the one that gets closer.

Additional recent attempts to improve the results obtained by the nonlinear static
procedures have been done by carrying out adaptive pushover analyses for each significant
vibration mode, together with energy based modal capacity curves, such as the proposal of
Kalkan and Kunnath (2006). Notwithstanding the value and benefits coming from further
and further advanced procedures, it is important not to fall into a complexity level similar
to the dynamic analysis one, losing the original purpose of simplicity of pushover based
methods.

2.4.2 Nonlinear dynamic analysis

Nonlinear dynamic analysis has always been seen as the most general and natural approach
to predict the dynamic structural response, however, due to the large computational
demand, its implementation within seismic analysis and design is yet to become usual
practice. The requirements around the dynamic analysis, summarized by Elnashai (2002),
in comparison to static pushover analysis, are substantially higher in number and
complexity level, as reviewed in Table 2.1.

Table 2.1 – Comparison of requirements for static and dynamic analysis (Elnashai, 2002).

Static Analysis Dynamic Analysis

Detailed models needed Detailed models needed
Stiffness and strength represented Stiffness and strength represented
No mass representation required* Mass representation required
No damping representation required Damping representation required
No additional operators required Time integration operators required
No input motion required Input motion required
Target displacement required Target displacement is an output
Action distribution fixed* Actions vary in time
Usually faster than dynamic analysis Usually slower than static analysis
*This may not be the case for advanced adaptive pushover

Essentially, dynamic analysis is widely recognized as the excellence tool for estimating
seismic response, but is at the same time demanding, computationally and in terms of
technical knowledge. It is still not common that practitioners experience covers all the
needed parameters. The integration scheme is a crucial aspect and has been found to
profoundly affect the results, as well as to compromise the analysis in several other

2.31
Seismic Assessment of Bridges

aspects, such as numerical stability or time step. Another critical parameter is the damping
definition, elastic viscous or hysteretic, which is believed to cause variations of 50% or
more in force response prediction (Elnashai, 2002; Priestley and Grant, 2005; Hall, 2006),
as well as to contribute to numerical (in)stability. The input force itself, the ground motion,
expressed in the form of one or more accelerograms, is a decisive and problematic issue
itself, in terms of type of records, intensity or standardization, among others (see 2.3 for
further detail).

The importance of the mentioned factors, is generally recognized, as summarized by

Spacone et al. (2008), describing the unresolved issues within the use of nonlinear
dynamic analysis as: the selection of an appropriate structural model, the selection of
ground motions, the selection of a correct damping matrix, the heavy computational effort
and the large volume of output quantities to be analysed.

Some indications are given as for the selection and nature of ground motion records within
regulation codes, such as EC8 (CEN, 2005a). Moreover, it is commonly believed that
recorded accelerograms are less demanding on the structures and should be used for actual
design of structures. On the other hand, generated accelerograms present the opposite
tendency and, therefore, are claimed by some to be used in research work to refine
nonlinear methods of analysis. Given the extensive nature of the topic of nonlinear
analysis, there are still no thorough studies and discussion is still going on in countries
where dynamic analysis is starting to be applied broadly.

The damping issue is as important as controversial in nonlinear dynamic analysis.

Hysteretic damping is usually responsible for the majority of the energy dissipation during
the earthquake action. A remaining smaller amount of non-hysteretic or elastic viscous
damping, the designation lacks consensus, resulting from physical phenomena involving
structural and non-structural elements is also activated. Such portion has been traditionally
considered by means of an equivalent viscous damping parameter, quantified by the
Rayleigh damping model, as advocated by Clough and Penzien (1993) or Chopra (1995),
believed to range from 1% to 8%, depending on several factors (Wakabayashi, 1986), or
around 2% to 3%, always under 5% for an inelastic structure, as referred in (Spacone et al.,
2008). Generally, the more accurate the nonlinear structural model is, the lower the
damping ratios should be (Panagiotou and Restrepo, 2007). The approach of using

2.32
 Seismic Assessment of Bridges

equivalent viscous damping concept to represent energy dissipation sources that are not
explicitly included in the model are admonished by authors such as Wilson (2002), but still
supported by others, although with the recommendation for abandoning the Rayleigh
damping model, which is proportional to mass and stiffness. A stiffness proportional only
model is argued by many (Abbasi et al., 2004; Hall, 2006; Pegon, 1996; Wilson, 2002) to
avoid the spurious energy dissipation generated by mass proportional part. Particularly,
tangent stiffness proportional damping has been seen as the most promising choice by
Priestley and Grant (2005). The damping issue is rather complex and even difficult to
solve, given that small changes in damping, mainly for very low damping, imply large
variations in the results.

2.4.3 Innovative displacement based approaches

Recent developments in seismic behaviour assessment of bridges have focused the

transition from force design philosophies to displacement based ones. In the present time,
with a variety of new design approaches that require increased emphasis on displacement,
rather than on strength, the most common approach has been to attempt to modify force-
based design procedures, rather than to completely revise seismic design procedure in a
more rational manner, as stated in (Calvi, 2004).

The evolution from forces (or accelerations) evaluation to displacements has been based in
the recognition of a series of points, according to (Priestley, 1993, 2003): it is generally
accepted that damage can be related to material strains, and that material strains can be
related to maximum response displacements, but not to response accelerations; equal
displacement approximation is known to be non-conservative for short-period structures.
Priestley (2003) showed that all formulations are correct at some part of the period range
of structural response, and all are wrong at other periods. New, displacement-based (some
times referred to as performance-based), methods for design have therefore been developed
recently, with emphasis to one of the approaches (Priestley and Calvi, 2003b), the Direct
Displacement Based Design (DDBD), in which the fundamental difference from force-
based design is that the structure to be designed is characterized by a SDOF representation
of performance at peak displacement response, rather than by its initial elastic
characteristics, based on the Substitute Structure approach (Gulkan and Sozen, 1974;
Shibata and Sozen, 1976). Shortly, within the DDBD methodology, a structure is

2.33
Seismic Assessment of Bridges

designed/assessed based on its behaviour at maximum response, by determining the

inelastic displacement pattern, characterizing the equivalent SDOF system, applying the
displacement based design concept to the SDOF structure and determining the strength
required to the structural elements. All the procedure, as well as the computed quantities, is
defined for the structure to reach a predetermined displacement when subjected to an
earthquake consistent with the design level event.

For the particular case of bridges, carrying out the first step, estimating the inelastic
displacement pattern of the deck, compatible with the displacement at the top of the piers,
will allow, in the end, the definition of the piers capacity, if design levels corresponding to
increasing seismic intensity are considered. The critical pier can therefore be readily
identified and, in most cases, will be the shortest one ruling the selection of the
displacement pattern. Generally, the lateral displacements for bridge piers are based on a
set of limiting longitudinal strains consistent with the desired damage level.

It is straightforward to compute the design displacement from strain limits, considering the
strain profile at maximum deflection of a simple bridge pier under transverse response,
defined by the maximum concrete compression strains, εc, and the maximum reinforcement
tensile strain, εs, for the considered performance state. Accordingly, there are two possible
limit state curvatures, based on the concrete compression and the reinforcement tension
respectively. Plastic and yielding curvatures may be easily computed, enabling curvature
ductility to be known. If ductility in displacements at the top of the pier intended,
expressions including the pier height contribution ought to be used.

The use of such approach will definitely constitute a recently developed and greatly
simplified alternative for the seismic analysis of bridges. It is certainly faster and less
demanding than any nonlinear pushover based approach, not to mention nonlinear dynamic
analysis. Some questioning may rise around the need for a SDOF substitute structure,
which is, in any case, already considered in nonlinear static procedures or the task of
associating a seismic intensity level to a specific performance level. Even though originally
developed bearing design in mind, it can be used for assessment of existing structures
purposes (Priestley et al., 2007).

2.34
 Seismic Assessment of Bridges

2.4.4 Nonlinear response prediction of bridges

When it comes to bridges, it is commonly recognized that any of the aforementioned issues
has been rather less scrutinized than for the case of building structures. Bridges have been
designed by reference to acceleration response spectra, according to (Casarotti et al.,
2005), for the past 40 years at least, apparently for historical reasons, given that common
practice has always dealt with other load types that not the seismic action: self-weight,
traffic or wind. When structural design of bridges started to become routinely
implemented, the first rough approaches consisted in procedures similar to the adopted for
the case of wind loading, assuring that the structure would remain elastic for a portion of
the vertical weight, applied as a uniform lateral force. There was no inelastic response
being studied herein, the behaviour was fully elastic and, consequently, underestimation of
deformation or deflections, together with overestimation of force, leading to the absence of
significant strength degradation or insufficient reinforcement length (Priestley et al., 1996;
Kawashima, 2000), was widely verified during the occurrence of earthquakes (see 2.2).

Particularities regarding this sort of structures are often invoked to justify such different
state of progress. To mention some, it can be pointed out that the superstructure, the deck,
is designed to remain elastic and, hence, as long as piers enter their inelasticity range, the
deformed shape of the structure will be essentially governed by the elastic behaving deck;
effects of superior mode shapes, already highly pointed out, are more important for
irregular bridges; complex torsional and distorsional effects are expected, due to the
contrast between deck and piers element types.

In accordance with what has been just stated, standard and recently developed pushover
procedures, as well, have been thoroughly tested for buildings, but not for less investigated
typologies, such as bridges. The instant tendency would to extrapolate the procedures
assessed with buildings, a practice that must be handled with care, due to the inherent
mentioned differences between the two structural systems. Fischinger et al. (2004) even
question the validity of traditional pushover procedures to bridges. Similarly, Spacone et
al. (2008) argue that pushover procedures are still not readily applicable to irregular
structures, where bridges fit in.

2.35
Seismic Assessment of Bridges

Nevertheless, and despite the unfavourable characteristics of bridge structural systems,

there have been a few recent research studies indicating that nonlinear static procedures are
possible to use within such configurations (Casarotti and Pinho, 2007; Isakovic and
Fischinger, 2006; Lupoi et al., 2007; Paraskeva et al., 2006; Pinho et al., 2007; Pinho et
al., 2009). Conclusions of the majority of such studies are drawn, though, towards the
recommendation for using advanced, higher mode effects including nonlinear static
analysis techniques.

With respect to nonlinear dynamic analysis, the application to bridges is frequently less
demanding than for other sort of structures, such as buildings. Models are definitely less
complex, with less structural elements, which make the dynamic analysis considerably
more feasible. Typical discouraging issues, computational demand, analysis output or
integration scheme, assume herein slightly lightened significance. Indeed, seismic
assessment of important bridges is increasingly performed using dynamic analysis in the
time domain. Nevertheless, some issues are traditionally pointed out as key simplifying
assumptions when it comes to dynamic analysis of bridges response to appropriately
selected and scaled time histories: the seismic motion that is transmitted to the structure
through its supports is synchronous and identical for all piers and abutments; the local site
conditions are accounted for in terms of site categorization and the superstructure is fully
fixed at the pier base points (Sextos et al., 2003). Such assumptions are important to avoid
incorporating more complex models, which leads to often uneconomic and sometimes
numerically sensitive analyses.

Damping stands, to what bridges are concerned, as an essential issue, assuming values that
are typically lower, given that not only there are less non-structural elements, from where
the so-called elastic viscous damping can appear, but also the structural hysteretic energy
dissipation locations correspond to the base of the piers, only. The fluctuation of a bridge
model response as a function of the damping can be extensive and calibration is frequently
needed, as noted in (Carvalho, 2009).

The use and validation of Direct Displacement Based Design philosophy within bridges
seismic analysis has been object of quite a few recent research studies (Alvarez, 2004;
Dwairi and Kowalsky, 2006; Kowalsky, 2002; Priestley and Calvi, 2003a). The transverse
response of bridges is especially more complex than the longitudinal response, requiring

2.36
 Seismic Assessment of Bridges

careful consideration. Some aspects needing deeper attention have to do with: the
definition of the displacement profile, which demands some consideration regarding the
relative stiffness of the piers, when compared to the deck or the abutments; to this matter,
Kowalsky (2002) and Dwairi and Kowalsky (2006) proposed the concept of effective
mode shapes, whereas Priestley and Calvi (2003b) adopt more pragmatic approaches.
Alvarez (2004) or Alfawakhiri and Bruneau (2000) use first mode based displacement
shapes. System damping is another important issue although, given the simplified nature of
the methodology, less complex of taking into account, as prescribed by the works of
Kowalsky et al. (1995) and (1994) or Priestley and Calvi (2003a), both estimating damping
in the individual elements, combining them later on according to the work or shear force,
respectively, carried out by each member.

2.5 Nonlinear modelling overview

The seismic action and nonlinear analysis methods have been seen to contemplate different
characterizing possibilities, comprising several critical issues, to which sometimes
corresponds significant uncertainty or dispersion. The modelling task, including material
and geometric features, is equally not immune to different alternatives and need for choices
and assumptions. Indeed, structural modelling for seismic evaluation becomes particularly
noteworthy as a natural reflex of the complexity that is typically associated to such action.

The different structural modelling possibilities that are commonly seen as available can be
classified according to the purpose of their use, as summarized by Spacone (2001). Global
Models, or Lumped Parameters Models, feature the nonlinear response of a structure at
specific degrees of freedom. Discrete Finite Element Models, also known as Member
Models, Structural Elements Models or Frame Models, characterize a structure by
connecting frame elements with duly modelled inelasticity. Finally, Microscopic Finite
Element Models, using the Finite Element (FE) general method, approximate the solution
of a problem in continuum mechanics by the analysis of an assemblage of two or three-
dimensional FEs connected at a finite number of nodal points. The choice for each of the
mentioned alternatives, reflecting the refinement level of the model, has plenty to do with
the desired accuracy as well as with the computational effort that is implicated.
Furthermore, the sort of analysis procedure that is being considered will definitely make a

2.37
Seismic Assessment of Bridges

difference, with the objective to avoid superposition of factors aggravating complexity of

the whole process. It is, however, generally reasonable to assume that sophisticated FE
models are more suitable for the study of structural details, which require high accuracy
with bearable computational demand. At the same time, it is commonly accepted that
frame models represent the best compromise between accuracy, provided by nonlinear FE
modelled members, and simplicity, offering a sound enough view over the structural
elements response.

Within a bridge structural system, frame models are typically used, with the nonlinearity
engaged to the piers (the bearing structure), since deck and abutments are usually protected
against collapse or severe damage, for reasons of cost and life safety (Casarotti and Pinho,
2006). The piers become therefore the fundamental elements (as recurrently stated) that
need more detailed modelling of their nonlinear behaviour.

There are two major sources of nonlinearity: material and geometric. The former source
has been definitely paid more attention over the years probably due to the fact that
geometric nonlinearities become more important in a later phase of the structural response
to high intensity, closer to the ultimate limit state.

2.5.1 Material nonlinearity

Material nonlinearity, regarding inelastic response of structures to seismic action, is

typically accounted for by means of lumped or spread plasticity models.

Lumped plasticity models correspond to the use of linear elastic behaving elements, with
the exception for certain, well defined, zones of the element, where the plasticity is
admitted to be concentrated under the form of a plastic hinge, according to the observed
typical concentration of inelasticity of RC frames at the extremities of the elements. As a
consequence, the first approaches to model this type of behaviour considered nonlinear
springs at the member ends (Clough and Johnston, 1966; Giberson, 1967; Takizawa,
1976), as referred in (Spacone, 2001). Plastic hinges have no standard implemented
definition and its location, and mostly length, can be derived from several different
approaches. Their definition, i.e., their constitutive law, can be specified using ad hoc
plastic laws or more advanced fibre based cross section models (Spacone et al., 2008).
Many of them try to reproduce the effect of phenomena such as stiffness degradation in

2.38
 Seismic Assessment of Bridges

flexure and shear (Clough and Benuska, 1967; Takeda et al., 1970; Brancaleoni et al.,
1983), pinching under load reversal (Banon et al., 1981; Brancaleoni et al., 1983) or bar
pull-out effects (Otani, 1974; Filippou et al., 1983a). Modified versions of the previous
have been developed as well, so as to include the different described issues, such as the one
proposed by Costa and Costa (1987), later refined by Varum (1996), which considers
stiffness and strength degradation and pinching all together. More contentious is the length
along which the plastic the behaviour develops, LP, needed to transform plastic rotations
into plastic curvatures, and vice-versa. Again, there is no established formula or procedure
to define such parameter, approaches have been refined over the years and several
proposals have been made since the first approach by empirical expressions suggested by
Baker and Corley, where the plastic hinge length is proportional to the distance from the
critical section to the point of contraflexure, as mentioned by Park and Paulay (1975).
Different proposals and refinements arose in the following years (Kappos, 1991; Paulay
and Priestley, 1992; Priestley and Park, 1984), complemented by calibration and/or
comparative studies by other authors, such as (Vaz, 1992; Guedes, 1997). The general
conclusion has been to notice that no specific proposal stands out as outstandingly better or
more accurate, given that a concentrated plasticity model itself is inevitably approximate.
More simplified and expedite expressions do not, therefore, necessarily lead to worse
estimates. This is actually the major drawback commonly associated to lumped plasticity
models, the fact that structures are assumed almost totally elastic, although, at the same
time, they are typically more employed within current commercial structural analysis
computer programs (e.g., (Computers&Structures, 2006)). It is most likely an easier to
grasp concept for inexperienced analysts (Spacone et al., 2008). Other authors, though,
counter this trend and call out the attention to the need for being careful when using a
concentrated plasticity approach by users that are inexperienced in the calibration of the
characterization of the constitutive laws (Casarotti and Pinho, 2006). Studies from Charney
and Bertero (1982) or Bertero et al. (1984), amongst others, focused the limitations and
pitfalls of employing lumped plasticity.

Spread plasticity models, commonly referred to as fibre models, consider the material
nonlinearity in a totally distributed way, at each integration point. Considered to be more
accurate when describing the continuous structural characteristics of RC members, with no
inelastic regions definition, require simple geometric and material properties. The cross

2.39
Seismic Assessment of Bridges

section response is then estimated by classical plasticity theory in terms of stress and strain
resultants or by explicitly discretizing the elements in fibres with uniaxial behaviour, with
material inelasticity spread along the member longitudinal axis, which assures accurate
estimation of damage even in the highly inelastic range. First approaches to distributed
nonlinearity resulted in proposals by Aktan et al. (1974), Hellesland and Scordelis (1981)
or Marí and Scordelis (1984), the latter two making use of the classical stiffness method
with cubic hermitian polynomials. Improved modelling algorithms, including axial force-
bending moment interaction (Menegotto and Pinto, 1973), shear effects (Bazoant and Bhat,
1977) or alternative flexibility based formulations (Mahasuverachai and Powell, 1982;
Kaba and Mahin, 1984; Zeris and Mahin, 1988, 1991) have followed. Implementation in
commercial software, available for research or office design analysis and assessment is,
however, relatively recent, with focus on OpenSees (McKenna, 1997; McKenna and
Fenves, 2006), MIDAS (2006), SeismoStruct (SeismoSoft, 2008), among others, which
allow the use of displacement-based or force-based elements.

In the end, both sorts of models use the same cross section constitutive laws, at an
integration point level, for spread plasticity models, or at a plastic hinge level, for lumped
plasticity ones. The concrete and steel material models are, hence, the basic input source,
which admit, as well, different refinement and subsequent complexity levels. Amongst
numerous available steel models, the bilinear, the Menegotto and Pinto (1973) and the
Monti and Nutti (1992) proposals are frequently used, whereas for concrete there are
current tri-linear (more simplified) or nonlinear with constant or variable confinement,
such as the models from Kent and Park (1971), Scott et al. (1982) or Mander et al.
(1988b), just to mention some.

FE fibre models are, nevertheless, the most developed and up to date, given the ability to
directly take into account phenomena such as interaction between axial and bending forces
or shear effects, through the entire length of the element with no need for distinction
between elastic and inelastic sub-elements, as in plastic hinges based models. Furthermore,
due to its exact integration nature, fibre models are commonly seen as more accurate and
precise but, to some extent, more delicate to use, according to Spacone et al. (2008),
particularly when sections of structural elements feature softening; in such cases,
deformations will occur concentrated in the extreme sections with solution objectivity loss,
which somehow recalls a plastic hinge scenario. In fact, contributions from the two

2.40
 Seismic Assessment of Bridges

modelling possibilities may be reconciled, using plastic hinge typical lengths for
discretization of elements within fibre based models (Casarotti and Pinho, 2006; Calabrese
et al., 2010).

As far as bridges are concerned, the relative positioning of the two modelling types is at
least interesting. On the one hand nonlinear behaviour of bridges is deemed to occur at
well defined structural elements, the piers, which, in turn, witness quite concentrated,
delimited, plasticity effects at the bottom or, at most, at the top, depending on the
connection to the deck. This set of aspects will easily induce the suitability of lumped
plasticity models. On the other hand, given the relative simplicity that characterizes bridge
structural systems, and keeping in mind that nonlinear modelling will be restricted to the
main bearing substructure (the piers), the argument of high computational demand and
analysis complexity loses relevance and finite fibre-based elements modelling becomes
more appealing.

2.5.2 Geometrical nonlinearity

There is not much experience in modelling geometric nonlinearity in frame analysis, in

spite of being generally recognized their importance during earthquake-induced ground
motion, as the structures may experience considerable lateral drifting. Different sources,
local and global, are usually admitted for geometric nonlinear behaviour, corresponding to
beam-column effects and large displacements or rotations, respectively. Quoting
Spacone et al. (2008), when displacement-based elements are used, contemplation of P-∆
effects or large displacements and moderate rotations is a “well established field” whilst,
for force-based elements, general co-rotational formulations have been posed (Crisfield,
1990; Sandhu et al., 1990; Neuenhofer and Filippou, 1998; Izzuddin, 2001; Felippa and
Park, 2002; Scott and Filippou, 2007). Recent work from Correia and Virtuoso (2006)
allows generalized accounting for large displacements/rotations and large independent
deformations relative to the frame element's chord (the P-∆ effects), through the
employment of a total co-rotational formulation.

The importance given to this matter within design codes is relative, such as the example of
EC8, which does not include specific guidelines on geometric nonlinearity, indicating only
when and how to consider the phenomenon, in an approximated way. Second order effects

2.41
Seismic Assessment of Bridges

will be taken or not into account, depending on an approximate formula for interstorey
drift.

According to Priestley et al. (1996), concerning bridge response, lateral displacements

during an earthquake event are typically small, with regards to the piers height and cross
section dimension, which generally allows nonlinear geometric effects to be neglected. For
the case of highly flexible bridges or slender or very tall piers, geometrical nonlinearity
may become significant, especially when combined with considerable vertical load, and
must therefore be included. Nevertheless, second-order (P-∆) effects consideration, instead
of full, large-deformation nonlinear geometric behaviour, will be enough.

2.6 Safety assessment methods

Several paths can be followed with the purpose of evaluating the seismic safety of bridges,
when properly modelled, subjected to duly characterized seismic loading. The main
distinction that can be made is between deterministic and probabilistic methods. Both of
the approaches within this basic division can recur to the same tools for characterizing the
seismic input (response spectra, accelerograms), estimating structural capacity or
predicting structural effects (fibre or plastic hinges modelling approaches, nonlinear static
or dynamic analysis) but the main difference will be in the way uncertainty (of the seismic
input, capacity models and analysis procedures) is taken into account and how it is
reflected in the final assessment output.

2.6.1 Deterministic assessment

The deterministic assessment is currently the most employed tool, mainly if one considers
the design office environment. It is basically given by the crude comparison of the
capacity, member-by-member, with the corresponding demand, when the structure is
submitted to seismic forces. The capacity, as prescribed by many codes (e.g., EC8) is often
affected by empirical reduction coefficients, applied to the mean or characteristic values,
so as to account for uncertainty. For the case of the seismic input, a minimum number of
records is also frequently prescribed in order to guarantee a good average or maximum
demand. The safety verification will then be typically carried out according to predefined
limit states, currently associated to specific return periods, which will govern the way of

2.42
 Seismic Assessment of Bridges

considering the seismic input and computing the capacity and demand for further
comparison.

2.6.2 Probabilistic assessment

The employment of probabilistic methods is, on the other hand, far from large
dissemination among the professional engineering community, even though they have been
rather well-established in the recent past (few decades). Their main practical application
has been actually the calibration of the deterministic approaches used in codes, based on
the use of partial safety factors. The reason that is mostly pointed out for such state of the
art is the abstractness, mathematical complexity and much greater consciousness on the
actual physical terms of the problem, unlike the traditional design procedures, which offer
clear-cut guidance, as stated by Pinto et al. (2007). A new wave of probabilistic methods,
developed over basic probabilistic concepts has come up in the past decade, especially in
the US. Their “simplicity” with respect to theoretical background makes them more easily
approachable by engineers and, thus, more appealing to the community. Their popularity
has reached the state of use as complement to code-based design, assessment of existing
structures or even design of new ones. Even if “simple” any probabilistic procedures
requires a substantial amount of different information, in comparison to the deterministic
ones, to characterize the uncertainty in the structural safety problem. Such data includes
probabilistic characterization of seismic action, as well as description of the capacity and
structural demand, accompanied by a measure of the corresponding variability.

According to (Pinto, 2001) probabilistic assessment of structures is usually

simulation-based, FORM-based (First Order Reliability Methods) or response-surface
based. Several different proposals for implementation are available in literature within each
category but it generally recognized that the seismic risk subject is still in considerable
development and definition, due to the growing awareness of the international community
of the need for including probabilistic measures in seismic assessment practice.

2.43
Seismic Assessment of Bridges

2.7 Summary

This chapter featured the main goal of providing a background to this work, regarding the
current status of the different steps within probabilistic seismic assessment of bridges,
calling the attention out to the vital importance of considering the seismic events in bridge
engineering as well as summarizing the different topics in discussion and in need for
improvement. Based on such considerations, Chapters 3 to 6 will focus the different
identified subjects, which will allow, in the end, the obtaining of a consistent probabilistic
safety measure.

2.44
3. Seismic Input

The definition of the seismic input is probably the primary step within any seismic
assessment analysis and is sometimes assumed as a straightforward task, which can be
highly misleading. Effectively, seismic action has been proved as complex, unpredictable,
allowing several different representation modes, each with considerable dispersion coming
from natural variation. The way how seismic loading is accounted is deeply related to the
employed analysis method: accelerograms are used for nonlinear dynamic analysis
whereas response spectra are typically utilized for nonlinear static procedures.

The use of accelerograms is definitely more realistic, given that it enables the application
time-step by time-step of the dynamic loading, becoming more suitable than response
spectra for multi-degree-of-freedom (MDOF) structures. It is, however, understandable
that such advantageous use of accelerograms carries a corresponding fair amount of
complexity, usually requiring preliminary treatment. Response spectra, on the other way,
are usually more consensual and information is generally represented in a more smoothed
way, especially if within code or guidelines applications. The immediate disadvantage is
that, by definition, response spectra refer to single-degree-of-freedom (SDOF) structures
which will require MDOF systems to be converted to equivalent SDOF ones, if analysis
through such sort of seismic action representation is intended.

3.1
Seismic Input

The substantial differences concerning the complexity surrounding the mentioned

approaches indicate that, clearly, a higher number of calibrating issues are associated to
accelerograms. Moreover, nonlinear dynamic analysis is becoming more and more
popular, given the advances in current computational capability. Consequently, the main
focus of this chapter will be on the use of accelerograms for seismic analysis.

3.1 Accelerograms

Several aspects are worth considering within the process of setting up real or artificial
records for the use in nonlinear dynamic analysis. A brief overview is herein intended, as a
contribution to this matter, especially regarding the application to bridge structures. Much
of the work within this topic has been carried out under the European Project LESSLOSS
(2004a), which considered the application of seismic risk assessment methodologies of
several research teams of different institutions to the analysis of two distinct structures: a
building frame, named ICONS and a viaduct, named LORDO.

Starting with the scaling and selection issue, a set of twenty records was initially
considered, serving as the original base of records, within a certain range of magnitude,
epicentral distance and soil profile types. Several different scaling techniques were
compared using elastic acceleration response spectra, therefore, results of SDOF systems.
Simultaneously, a double selection is performed, one related to the scaling parameters that
markedly perform better and the other based on the disregarding of accelerograms which,
due to their inflated effect, increase greatly the scatter among the records, forming a
reduced base. This selection is performed by means of a statistical grouping technique,
which is based on clusters – groups of elements with similar characteristics or effects,
formed by likeness measuring. The records are afterwards used, scaled according to the
best performing previously found procedures, to the nonlinear seismic analysis of the two
mentioned case study MDOF structures: the building frame and the bridge. Furthermore,
analyses with artificial accelerograms, matched to the response spectra of the real records,
are also carried out. Results, namely ductility in displacements, are used to: i) check the
stability effect of the clustering selection; ii) search for a minimum number of records
necessary and sufficient to reach the same mean/median result that the global set would

3.2
 Seismic Input

provide; iii) compare scaling techniques, previously filtered based on response spectra; iv)
confront performance of real and artificial accelerograms.

3.2 Real accelerograms – selection of records

3.2.1 Initial considered set

The study on real records was carried out over a set of twenty accelerograms selected from
the European Strong Motion Database (Ambraseys et al., 2002). The records have been
selected from firm soil sites, free-field, including large and distant, large and close,
moderate and close as well as intermediate earthquake records, in an attempt to cover a
sufficiently representative range of distances and the expected range of magnitude in
Europe. Figure 3.1 illustrates the range of magnitudes and epicentral distances across the
initially considered base of records, where it can be observed how the chosen
accelerograms are spread out within the four quadrants, although scenarios of
simultaneously high or low magnitude and distance are prominent. Figure 3.2, in turn,
represents the cumulative distribution of the measured magnitudes and epicentral
distances. The cumulative distribution shape is a good indicator of how homogeneously the
parameter varies within its range: a slope close to 45 degrees would be the ideal situation.

80

60
Distance (km)

40

20

0
5 5.5 6 6.5 7
Magnitude (MW)

Figure 3.1 – Magnitude-Distance used pairs of values.

3.3
Seismic Input

1 1

0.8 0.8

0.6 0.6
F(M)

F(d)
0.4 0.4

0.2 0.2

0 0
5 5.5 6 6.5 7 0 20 40 60 80
Magnitude (MW) Distance (km)

Figure 3.2 – Magnitude and distance cumulative distributions.

Table 3.1 briefly presents the main seismotectonic characteristics of the twenty initially
selected records.

Table 3.1 – Selected real records – main features

Fault Distance (km)
Epicentral Distance

Acceleration (m/s²)
Earthquake Name

Magnitude (MW)

Peak Horizontal
Duration (s)
(km)
Date

1 Montenegro 1979 6.9 56 46 15.94 0.572

2 Campano Lucano 1980 6.9 58 41 13.04 0.532
3 Erzincan 1992 6.6 65 58 18.56 0.297
4 South Iceland 2000 6.5 52 50 21.58 0.692
5 Montenegro 1979 6.9 25 12 48.23 4.453
6 Montenegro 1979 6.9 24 9 48.22 2.88
7 Campano Lucano 1980 6.9 16 13 45.54 1.725
8 Friuli (aftershock) 1976 5.3 21 15 7.19 1.701
9 Friuli (aftershock) 1976 5.5 15 17 23.82 2.273
10 Friuli (aftershock) 1976 5.5 15 17 86.06 0.898
11 Friuli (aftershock) 1977 5.4 7 6 48.98 2.365
12 Dursunbey 1979 5.3 6 6 18.31 2.824
13 Preveza 1981 5.4 28 7 14.93 1.402
14 Etolia 1988 5.3 20 12 21.88 0.584
15 Etolia 1988 5.4 21 6 30.50 0.383
16 Umbria Marche 1997 5.7 25 25 46.33 0.519
17 Valnerina 1979 5.8 39 37 38.41 0.386
18 Lazio Abruzzo 1984 5.9 69 64 23.21 0.171
19 Potenza 1990 5.8 28 29 20.38 0.944
20 South Aegean 1994 6.1 64 – 35.00 0.565

3.4
 Seismic Input

Earthquake duration is quite variable, ranging from 7.19 to 86.06 seconds, as well as peak
ground acceleration values, which go from 0.01g to 0.46g. A scaling parameter based on
earthquake duration has been considered, and will be presented further on, so as to provide
some information on the importance of the variability associated to that parameter.

3.2.2 A homogeneous base of records – Clustering analysis

Starting from the initial set, a selection procedure is thought to be advantageous in the
sense that extremely severe or particularly peculiar earthquakes may be segregated with
benefits to the consistency of the analysis. Consequently, a homogeneous base of records is
intended to represent a smaller set of earthquake records, based on the original ones, made
up by removing those having very distinct characteristics, without loss of identity of the
initial base. This representativity is assured by grouping elements with largest similarities,
which has been measured, at a first stage, using response spectra. Afterwards, also
displacement ductility demand quantities have been object of clustering procedure, applied,
therefore, to structural effects. The methodology used to aggregate the original real ground
motion records into groups, based on similarity measures, is called Clustering Analysis.

Clustering analysis can be seen as a collection of statistical techniques that can be used to
classify objects, resulting in a size reduction of the available data. Given that it is possible
to measure similarity and dissimilarity in a number of ways, data classification will depend
upon the method being used, such as distance measuring, e.g., (Everitt et al., 2001).
Elements that are significantly different from others will be grouped into clusters of a
single element, which reflects the fact that the association degree is strong between
members of a same cluster and weak between members of different clusters. Clustering
analysis classification is based on placing objects into more or less homogeneous groups,
in a way that the relationship degree between groups becomes more perceptible. The
application of this methodology is based on the two following steps: (i) measuring
distances or evaluating the similarity between objects and (ii) grouping objects based on
such distances or similarity measures.

There are two major clustering techniques, corresponding to hierarchical and

non-hierarchical algorithms. The hierarchal procedure involves putting up a hierarchy tree
defining the relationship among individuals, whilst, when a non-hierarchical algorithm is

3.5
Seismic Input

used, the position in the measurement is taken as the central place and distance is measured
from that central point. Currently, the non-hierarchical method has been less considered,
due to the difficulties in choosing an ideal reference central point. To what hierarchical
clustering is concerned, the technique can be divisive or agglomerative. A divisive method
begins with all cases in one cluster, which is gradually split up into smaller clusters. On the
other hand, the agglomerative technique starts with (usually) single member clusters that
are gradually merged until one large cluster is formed.

When carrying out a clustering analysis, the first step is to establish distance measurements
(e.g., Euclidean, squared Euclidean or Chebychev distances) or similarity measurements
(e.g., Pearson’s correlation, Russell, Rao or Jaccard coefficients) matrixes, in which the
linkage degree between each two units of analysis (response spectra or displacement
ductility for different intensity levels, for instance) will be evaluated, disposed in both
columns and rows and where the cell entries are the measure of similarity or distance for
every pair of cases. Distance measures indicate how distinct two observations are from
each other, in a way that cases which are considered to be alike will correspond to low
distance amounts.

The following task is the definition of a clustering method that will determine the way how
clusters will be combined at each step (e.g., nearest neighbour, further neighbour or
centroid method). Once several objects have been put together these methods define the
distances between those clusters and the remaining elements, creating eventual new
clusters (Everitt et al., 2001). Output results can be either graphical representations (e.g.,
dendograms or icicle plots) or tables (e.g., cluster membership table or agglomeration
schedule). It should be noted that clustering analyses can be performed using easily
available commercial statistical software.

3.2.3 Application to the set of selected records

In order to assess the similarity of the 20 initially selected ground motion records,
characterized in 3.2.1, elastic response spectrum ordinates were used as the variable herein
submitted to agglomerative hierarchical clustering analysis, a rather simple procedure.
Within clustering analysis, all the previously referred distance and similarity measures
differ on the distance definition itself as well as on the way of evaluating it. For this

3.6
 Seismic Input

specific case, a likeness measure, the Pearson’s correlation, was chosen, given that it
allows comparing the linkage between each two variables (different response spectra). The
chosen clustering combination method was the “nearest neighbour”, which groups objects
based on the minimum aforementioned distance between them. Table 3.2 indicates, for
specific numbers of considered clusters, 3 to 8, which are the similar variables, i.e., the
ones that are grouped in a same cluster.

Table 3.2 – Cluster definition and homogeneous base

Case/Record 8 Clusters 7 Clusters 6 Clusters 5 Clusters 4 Clusters 3 Clusters

1 1 1 1 1 1 1
2 1 1 1 1 1 1
3 1 1 1 1 1 1
4 2 2 2 2 2 2
5 3 2 2 2 2 2
6 4 3 3 3 3 3
7 5 4 4 4 1 1
8 6 5 5 4 1 1
9 1 1 1 1 1 1
10 7 6 2 2 2 2
11 3 2 2 2 2 2
12 2 2 2 2 2 2
13 2 2 2 2 2 2
14 3 2 2 2 2 2
15 3 2 2 2 2 2
16 3 2 2 2 2 2
17 1 1 1 1 1 1
18 1 1 1 1 1 1
19 1 1 1 1 1 1
20 8 7 6 5 4 2
Homogeneous
1 2 3 4 5 9 11 12 13 14 15 16 17 18 19
base

For instance, if 8 clusters are considered, i.e., if one wants to divide the 20 records into
eight groups by means of the predefined similarity measures, according to Table 3.2,
records 1, 2, 3, 9, 17, 18 and 19 would belong to Cluster No. 1; records 4, 12 and 13 to
Cluster No. 2; records 5, 11, 14, 15 and 16 to Cluster No. 3; whereas records 6, 7, 8, 10 and
20 would constitute the single-element clusters No. 4, 5, 6, 7 and 8, respectively. As an
example, a schematic illustration of the records division, when defining 8 Clusters, is
provided in Figure 3.3. Records 6, 7, 8, 10 and 20 are defined as the most “different” from
all the rest, according to the adopted criteria, given that the clustering analysis determined
that they should be in single-element clusters.

3.7
Seismic Input

2
12 20
4 8
1 9 13 5
3 7
2 17 6
8
1 18 19 4
15 6
5 14
11 10
3 16 7

Figure 3.3 – Records distribution when defining 8 clusters.

Analogously, the same interpretation can be made for the case of 7 to 3 clusters,
identifying the records that are considered individually (shadowed cells in Table 3.2).
Clearly, records 6 and 20 are consistently individuated, whereas records 7, 8 and 10, are
highlighted if the lower boundary of 6 or 7 clusters is considered. Based on such results,
the selection was made to group variables into 7 clusters, for which members 6, 7, 8, 10
and 20 are not alike, hence defining five clusters of one member only. By segregating the
corresponding five response spectra from the original set, a homogeneous reduced set of
ground motion records, with the remaining 15 elements, was defined.

3.3 Real accelerograms – scaling

Considering accelerograms exactly as they were recorded, significant differences come up,
a natural feature on the use of real records. Subsequently, a seismic analysis involving
different real records requires the accelerograms to be put at a ‘same’, or at least
comparable, intensity level. In practice, what has been typically considered is the
imposition that different records share a specific characteristic, which can be an exceeding
probability for a predefined hazard level or a ground motion parameter, such as peak
ground or spectral acceleration in order to enable the average or median representation of
results. The immediate consequence is that accelerograms will necessarily be modified
through the application of an appropriate scaling factor.

The main goal, when selecting an appropriate scaling technique, related to a specific
standardizing parameter, is to reduce, ideally minimizing, the associated scatter. Recalling
what has been said in 2.3, several scaling methodologies have been studied in previously
published work and have been traditionally divided in two major groups: scaling factors

3.8
 Seismic Input

obtained from the characteristics of the accelerogram or from corresponding response

spectra.

The scaling parameter to choose may be closely related to eventual structural safety
verification, if the work is being developed within a probabilistic framework, depending on
the safety assessment selected method. If a failure probability is intended, recurring to the
statistical characterization of the capacity and the demand, the exceeding probability
density function for the selected scaling parameter is needed, and its availability has to be
taken into account when that choice is made. For such reason, some parameters have been
tested and emphasis has been given to those presenting a well-known or easily obtainable
exceeding probability density function.

3.3.1 Record quantities based intensity measures

Based on the accelerogram itself, simple quantities can be computed and considered as
scaling parameters. Among several possible ones, the following were selected for testing.

Peak ground acceleration (pga)

Each accelerogram is scaled in order to have its maximum acceleration equal to a
predefined level.

Square Ground Motion (s)

Integral of the square ground acceleration.

Root-square ground motion (rs)

Square root of the square ground motion.

Mean-square ground motion (ms)

Mean square value of the ground acceleration history for the time interval
corresponding to the attainment of 5% and 95% of the record’s energy.

Root-mean-square ground motion (rms)

Square root of the mean-square ground motion.

3.9
Seismic Input

3.3.2 Spectrum quantities based intensity measures

Including effects on SDOF systems, response spectrum, either in acceleration or pseudo-

velocity, of each record can be the base of certain scaling procedures. Additionally, power
spectrum may be used with the same purpose, bringing out a total of three other possible
scaling parameters.

Spectrum intensity (si)

Integral of the pseudo-velocity between 0.1 and 2.5 seconds.

Spectral acceleration (sa)

Accelerograms are scaled to have the same spectral acceleration at the fundamental
period of the structure.

Power spectrum area (psa)

Area under the power spectrum

For each scaling technique, the adopted normalization procedure was to establish a record
as a reference (the first one in Table 3.1 was herein considered) and scale all the others to
match the reference value of the considered scaling parameter for that reference record.
Figure 3.4 shows the set of scaled response spectra according to one of the scaling
parameters, with the purpose of illustrating the existing dispersion within the initial set of
real earthquake records.

10

8
Acceleration (m/s )
2

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Period (s)

Figure 3.4 – Scaled response spectra according to peak ground acceleration

3.10
 Seismic Input

The considerable heterogeneity level among the initial set can be relevant to check if any
of the alternative scaling parameters is actually particularly advantageous, given that a
highly homogeneous set of records may not put in evidence the real differences on the
performance of the different parameters.

Figure 3.5, in turn, stands for the average response spectrum for each scaling technique.
The ms technique yields a median response spectrum outstandingly different from all the
others, which denotes inappropriateness regarding the use of this parameter. For that
reason, a new plot of the average response spectrum for each scaling technique, featuring
not the ms technique, is presented in Figure 3.6.

35
pga
s
30
rs
ms
25 rms
si
Acceleration (m/s2)

sa (0.59s)
20 psa

15

10

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Period (s)

Figure 3.5 – Mean response spectrum to each of the scaling techniques

9
pga
8 s
rs
7 rms
si
Acceleration (m/s )

6 sa (0.59s)
2

psa
5

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Period (s)

Figure 3.6 – Mean response spectrum to each of the scaling techniques (without ms)

3.11
Seismic Input

In spite of a similar general shape on the spectra, differences are significantly higher for
periods under 1s, the range of interest, mainly at the first half of the interval. Furthermore,
no significantly distinct trend between record quantities based and response spectrum
based scaling techniques has been encountered, which denotes a clear need for a more
thorough analysis in terms of dispersion among the selected ground motion records (by
means of coefficients of variation).

3.3.3 Comparative analysis of typical intensity measures

Comparison of techniques has been carried out by computing a simple statistical measure,
the coefficient of variation (COV), referring to the ordinates of the twenty response spectra
normalised to each of the reference scaling parameters. The coefficient of variation, along
with the period or frequency, is presented in Figure 3.7 and Figure 3.8, respectively, for the
eight scaling procedures. Spectral acceleration technique is employed for a matching
period of 0.59 seconds, which corresponds to the case study of Lordo viaduct, considered
within the research project referred in Section 3.1.

In agreement with the markedly different behaving mean spectrum that has been observed,
the ms technique performs generally worse than the others. It features the largest
coefficient of variation across almost the whole range of periods. Similarly, other
parameters based on the square of the acceleration tend to introduce no noticeable
reduction in the dispersion levels: the s technique performs rather poorly, comparatively to
ms; with rs and rms better results are achieved, mainly for low periods, whereas at the
opposite side, such improvement does not stand. Velocity-based parameter, si, assumes the
best behaviour for periods over 0.5 seconds (or frequencies under 2.0 Hz).

Regarding the other parameters, as expected, scatter is small around the considered
structural fundamental frequency, when using the spectral acceleration matching
technique, sa, but average to poor results are obtained elsewhere. If one focus on the
frequency-dependent plot, scaling procedure based on peak ground acceleration, pga, the
typically most employed parameter, performs fairly well, when compared to the other
parameters, except for low frequencies (high periods), where its performance is surpassed
by other techniques.

3.12
 Seismic Input

150
pga s rs ms rms si sa (0.59s) psa

125

100
COV (%)

75

50

25

0
0 0.5 1 1.5 2 2.5 3 3.5 4
Period (s)

Figure 3.7 – Response spectra ordinates coefficient of variation for each scaling technique (period
dependent)

150
pga s rs ms rms si psa sa(0.59s)

125

100
COV (%)

75

50

25

0
0 5 10 15 20 25
Frequency (Hz)

Figure 3.8 – Response spectra ordinates coefficient of variation for each scaling technique
(frequency dependent)

The standardization by means of the area under the power spectrum, psa, a different
parameter regarding its nature, has a constantly median performance, therefore, not worth
the use. In general, it may be said that, considering the entire range of frequencies, pga and
rms techniques are the best, though for low frequencies (<1.0 Hz) si and rs techniques

3.13
Seismic Input

yield somewhat better results. Furthermore, if the structure under analysis is particularly
considered, the spectral acceleration matching procedure assumes great relevance in the
region of the fundamental period. According to the exposed, pga and sa techniques were
assumed for the rest of the seismic action characterization study. Such choice has been
based on two important aspects: first, if a full-range efficient technique is intended, pga
performs well, whilst the others have shown a less steady performance. Furthermore, the
use of a different parameter, instead of the traditional, well known peak ground
acceleration, seems, therefore, not worthy. Secondly, if a procedure that is calibrated for
the specific structure being assessed is sufficient, spectral acceleration technique behaves
extremely well, as previously demonstrated.

3.3.4 Response based intensity measures

Within probabilistic seismic demand analysis, the choice for the intensity measure (IM) to
be adopted relies not only on its ability to harmonize the seismic input, reducing the scatter
levels around response of SDOF or MDOF systems but on its characteristics of
practicality, sufficiency (i.e., the structural response does not depend on other ground
motion features given the IM), efficiency (i.e., the structural response is well correlated
with the IM) and hazard computability as well, as discussed in (Padgett et al., 2008). Up to
a relatively recent past the focus on intensity measures has been on the ground motion
parameters or the response of SDOF structures side, through the use of response spectra, as
detailed in the previous sections. Typically, peak ground acceleration (PGA) or spectral
acceleration (Sa) at a specific period or the fundamental period of the structure have been
used in most of the bridge engineering applications (Shinozuka et al., 2000; Mackie and
Stojadinovic, 2004; Nielson and DesRoches, 2007). The task of selecting appropriate or
optimal IMs to condition the demand and serve as basis to structural fragility curves has
been object of improvement and new approaches have been recently developed, given that
the choice of an optimal IM is still a subject in need for further addressing, as far as classes
of bridges are concerned. Such new approaches include denser formulations, such as the
inclusion of simplified static analysis of the structure to take into account accurate
estimates of its period and stiffness characteristics. It is actually recognized in the work by
Padgett et al. (2008), how structurally based IMs are challenging to implement in a
regional risk assessment context because often there is no sufficient available information
to estimate structural parameters, such as fundamental period.

3.14
 Seismic Input

An alternative approach has been studied so as to eventually replace the need for running
nonlinear dynamic analysis with pre-scaled accelerograms, through the establishment of a
new displacement based intensity measure, developed from simplified displacement based
principles, applied to bridges.

3.3.4.1 Preliminary Approach on a Displacement-Based Intensity Measure

A nonlinear static analysis (pushover) based procedure has been adopted, using the results
afterwards in a displacement based fashion, characterizing an equivalent SDOF system and
determining the bridge performance through the consideration of specified limit states. The
main goal was to seek for the relationship between the responses coming from a simplified
procedure and the result coming from nonlinear dynamic analysis, represented by a
specific number of accelerograms.

The procedure starts by carrying out a nonlinear static analysis of the structure, which
enables the characterization of the structural system. Three different limit states, which are
identified and located in the pushover curve, have been considered: the yielding limit state
(LSy) and two post-yielding ones (LS2 and LS3). The post-yielding limit states have been
defined in accordance with the displacements at which the first element reaches pre-
defined strains in the steel or concrete. The strains in the steel and concrete have been
taken as the ones admitted by Bal et al. (2008) as 0.35% for the concrete and 2% for the
steel for the second limit state (LS2) and 0.75% in the concrete and 3.5% in the steel for
the third limit state (LS3). Secant yield and post-yield periods have been defined
accordingly. Equivalent SDOF properties are defined and the displacement for each limit
state is computed from all the locations of the structure (for a bridge structure, all the deck
nodes at piers location will be considered), hence, in a Displacement Based fashion,
according to Equation (3.1), depending, therefore, on the actual load factor corresponding
deformed shape of the deck, di.

d SDOF =
∑m ⋅d
i i
2

(3.1)
∑m ⋅d
i i

3.15
Seismic Input

3.3.4.2 Procedure

The procedure used to estimate the intensity measure for a given structure and the
calculation of the engineering demand parameter from nonlinear dynamic analyses, for
calibration, is summarised in the following steps and should be carried out for each of the
considered accelerograms.

1. Calculate the displacement response spectra (5% damping);

2. Run the nonlinear dynamic analyses for the bridge using tangent stiffness
proportional damping;

3. From a static nonlinear analysis calculate the displacement capacity for each limit
state as described in 3.3.4.1 and using Equation (3.1);

4. Calculate the effective periods of vibration and the equivalent viscous damping for
each limit state. Several different proposals can be assumed in this step;

5. Obtain the spectral displacement demand for each limit state, based on the effective
period and equivalent viscous damping, and compare it with the displacement
capacities to estimate which limit state the bridge is predicted to exceed; the
intensity measure (IM) can thus be estimated. If the bridge does not exceed the first
limit state, the first limit state demand is used as the intensity measure; if the bridge
falls between the first and second limit state the second limit state demand is used
as the intensity measure; if the bridge falls between the second and third limit state
the third limit state demand is used as the intensity measure and if the bridge is
assumed to collapse then no intensity measure is assigned and the results are
removed;

6. According to the deformed shape associated to the Limit State that the structure is
not exceeding, a representative reference location, xref, is defined employing
Equation (3.2), where xi is the distance measured along the deck, measured from
one of the abutments. The engineering demand parameter (EDP) herein considered
is the maximum displacement at the reference location of the bridge (Ddyn), which
is obtained from the nonlinear dynamic analysis for the specific record;

3.16
 Seismic Input

xref =
∑x ⋅m ⋅d
i i i
(3.2)
∑m ⋅d i i

7. Plot IM against EDP.

A flowchart representing the procedure is presented in Figure 3.9.

Define bridge displacement capacity for each Limit State

dLSy, dLS2 and dLS3
()

Compute effective period, equivalent viscous damping and

spectral reduction factor per Limit State (TLSy, TLS2 and TLS3)

Define displacement demand for each effective period

SD(TLSy), SD(TLS2) and SD(TLS3)
(bridge displacement capacity for each Limit State)

SD(TLSy) ≤ dLSy yes IM = SD(TLSy)

no

SD(TLS2) ≤ dLS2 yes IM = SD(TLS2)

For each record

no

SD(TLS3) ≤ dLS3 yes IM = SD(TLS3)

no

Collapse – no IM is defined

Determine x ref in correspondence with deformed shape at the

Limit State previously assigned to the bridge and define EDP as
max |Ddyn(x ref)|

Plot IM against EDP

Figure 3.9 – Bridge displacement-based intensity measure procedure.

3.17
Seismic Input

3.3.4.3 Structural application

A case study consisting of fourteen well-known bridges, used for application in other
studies (Casarotti and Pinho, 2007), has been adopted, comprising two bridge lengths (with
regular, irregular and semi-regular layout) and two types of abutments. The complete
description and detailed information on the selected bridge configurations is carried out in
Section 4.3. Three different types of pushover analysis for each bridge have been carried
out: (i) adaptive displacement-based pushover; (ii) conventional pushover, with 1st mode
proportional load shape; (iii) conventional pushover, with uniform load shape. Further
details on nonlinear static analysis can be found in Chapter 5. Regarding the definition of
the limit states, within the reported framework, the yield limit state of the structure has
been considered to occur as soon as the steel on the first pier yields. Figure 3.11, referring
to one of the analysed structural configurations (Figure 3.10), illustrates the limit states
definition for the adaptive pushover curve computation case. The first pier to yield is P2,
the central one, in which Limit States 2 and 3 occur before the yielding of the third pier.

Figure 3.10 – A123 configuration.

4
x 10
3

A123
2.5

2
Base Shear (kN)

LS3
LS2 P3
1.5
P1
P2
1

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6
Equivalent SDOF system displacement (m)

Figure 3.11 – Adaptive pushover curve and limit states for A123 bridge.

3.18
 Seismic Input

With respect to the computation of the equivalent damping, effective period and spectral
reduction factor, two proposals, Priestley et al. (2007) and FEMA-440 (ATC, 2005) will
both be considered. According to Priestley et al., the effective period can be obtained using
the expression in Equation (3.3), the equivalent viscous damping is given by Equation (3.4)
and the correction factor applied to the displacement spectrum is given by Equation (3.5),
where Ty is the yield period, µLS is the ductility at a given limit state, ξeq is the equivalent
viscous damping and η is the correction factor.

TLS = Ty µ LS (3.3)

ξ eq = 0.565
(µ LS − 1) + 0.05
(3.4)
µ LS ⋅ π

7
η= (3.5)
2 + ξ eq

Effective period, equivalent viscous damping and correction factor proposed in FEMA-440
guidelines are as presented from Equations (3.6) to (3.8), respectively.

{
TLS = Ty 0.2( µ LS − 1) 2 − 0.038( µ LS − 1) 3 + 1 } (3.6)

ξ eq (%) = 4.9( µ LS − 1) 2 − 1.1( µ LS − 1) 3 + 5 (3.7)

5.6 − ln(ξ eq )
η= (3.8)
4

3.19
Seismic Input

The nonlinear dynamic analyses were carried out using ten earthquake real records from
the SAC Project (SAC, 1997), scaled to match the 10% probability of exceedance in 50
years (475 years return period) uniform hazard spectrum for Los Angeles. Detailed
description of the seismic records ensemble can be found in Section 4.3.

Applying the procedure described in Figure 3.9 for the 14 bridges and 10 earthquake
records, the preliminary plots with the relation IM–EDP, shown in Figure 3.12, are
obtained. The effective periods have been calculated using Equations (3.3) and (3.6), the
equivalent viscous damping using Equations (3.4) and (3.7) and the correction factor
applied to the displacement spectrum using Equations (3.5) and (3.8).

The black line in the plots in Figure 3.12 shows the linear regression of the data (y=ax). It
should be noted that the slope of the linear regression is a measure of the bias in the data
and σ is a measure of its efficiency (a low level of dispersion denotes an efficient intensity
measure). On average, the intensity measure underpredicts the maximum displacement
response at the reference location of the bridge.

The data dispersion, σ, calculated as the standard deviation of EDP/IM has been found to
be roughly from 0.4 to 0.6. The use of FEMA-440 approach improves the efficiency, given
that dispersion is usually 0.1 less. However the same does not happen with the bias, which
is between 1.2 and 1.3 for the closest to one situation, the adaptive pushover one. The
improvement in the scatter does not seem very surprising considering that in FEMA-440
equivalent linear parameters (i.e., effective period and damping) were determined through
a statistical analysis that minimizes, in a rigorous manner, the extreme occurrences of the
difference (i.e., error) between the maximum response of an actual inelastic system and its
equivalent linear counterpart. The scatter around the IM is larger for higher intensity
levels.

3.20
 Seismic Input

Adaptive Pushover - Priestley Damping Adaptive Pushover - FEMA Damping

25 25

EDP=1.2371 IM, σ=0.49849 EDP=1.2912 IM, σ=0.46938

20 20
EDP, ∆dyn(xref) (cm)

EDP, ∆dyn(xref) (cm)

15 15

10 10

5 5

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Intensity Measure, Sd(TLS) (cm) Intensity Measure, Sd(TLS) (cm)

st st
Conventional 1 Pushover - Priestley Damping Conventional 1 Pushover - FEMA Damping
25 25

EDP=1.2831 IM, σ=0.36933 EDP=1.3618 IM, σ=0.30631

20 20
EDP, ∆dyn(xref) (cm)

EDP, ∆dyn(xref) (cm)

15 15

10 10

5 5

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Intensity Measure, Sd(TLS) (cm) Intensity Measure, Sd(TLS) (cm)

Conventional uni Pushover - Priestley Damping Conventional uni Pushover - FEMA Damping
25 25

EDP=1.4124 IM, σ=0.44651 EDP=1.4539 IM, σ=0.44601

20 20
EDP, ∆dyn(xref) (cm)

EDP, ∆dyn(xref) (cm)

15 15

10 10

5 5

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Intensity Measure, Sd(TLS) (cm) Intensity Measure, Sd(TLS) (cm)

Figure 3.12 – Intensity measure vs. EDP.

In order to look at whether a procedure based on the increase in the period of vibration
with limit state displacement is meaningful, the response period of vibration from each

3.21
Seismic Input

nonlinear dynamic analysis has been calculated through the Fourier Amplitude Spectrum
of the response history at the reference location of the bridge. The period of vibration
obtained in this way for each dynamic analysis is plotted against the EDP, the dynamic
analysis results for the reference location, in Figure 3.13.

25

1.899
EDP=12.4 Teff

20
EDP, ∆dyn(xref) (cm)

15

10

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Effective Period, Teff(s)

Figure 3.13 – Response period of vibration versus EDP.

There seems indeed possible to establish some kind of trend between the effective period
and the respective EDP measure. Furthermore, Figure 3.14 shows how these periods
compare with the effective periods proposed in FEMA-440, which have been calculated
using Equation (3.6) where the ductility is based on the results of the pushover analysis
(Figure 3.11). It appears that these limit effective periods are, for a significant number of
cases, upper bounds to the actual periods of vibration recorded.

In order to investigate whether a closer relationship between the intensity measure and the
displacement at the reference location can be obtained using these periods of vibration, the
intensity measure has been recalculated using the effective periods shown in Figure 3.13,
with the ductility being estimated as EDP/∆LSy, where ∆LSy is obtained as described in
Sections 3.3.4.1 and 3.3.4.2. The updated IM vs. EDP plots are presented in Figure 3.15.

3.22
 Seismic Input

2
Adaptive Pushover - FEMA Damping

1.5
Period (s)

LS3
0.5 LS2

LSy

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration

2
st
Conventional 1 Pushover - FEMA Damping

1.5
Period (s)

LS3

0.5 LS2

LSy

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration
2

Conventional uni Pushover - FEMA Damping

1.5
Period (s)

LS3

0.5 LS2

LSy

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration

Figure 3.14 – Response periods of vibration and corresponding effective periods of vibration from
Equation (3.6) for different bridge configurations and types of pushover analysis.

It can be seen that the underprediction given by the intensity measure has increased. The
use of the effective measured periods does not seem indeed to improve the procedure.
Indeed, even though the clouds of points do seem to have become slimmer, the number of
outliers has apparently increased as well. Nevertheless, adaptive pushover based procedure
is the less underpredicting one. FEMA-440 approach for the equivalent viscous damping
and spectral reductions factor keeps on leading to closer to EDP estimates and lower
dispersion.

3.23
Seismic Input

Adaptive Pushover - Priestley Damping Adaptive Pushover - FEMA Damping

25 25

EDP=1.6504 IM, σ=0.64679 EDP=1.4615 IM, σ=0.54932

20 20
EDP, ∆dyn(xref) (cm)

EDP, ∆dyn(xref) (cm)

15 15

10 10

5 5

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Intensity Measure, Sd(TLS) (cm) Intensity Measure, Sd(TLS) (cm)

st st
Conventional 1 Pushover - Priestley Damping Conventional 1 Pushover - FEMA Damping
25 25

EDP=1.743 IM, σ=0.57699 EDP=1.538 IM, σ=0.5094

20 20
EDP, ∆dyn(xref) (cm)

EDP, ∆dyn(xref) (cm)

15 15

10 10

5 5

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Intensity Measure, Sd(TLS) (cm) Intensity Measure, Sd(TLS) (cm)

Conventional uni Pushover - Priestley Damping Conventional uni Pushover - FEMA Damping
25 25

EDP=1.7012 IM, σ=0.65244 EDP=1.4824 IM, σ=0.59644

20 20
EDP, ∆dyn(xref) (cm)

EDP, ∆dyn(xref) (cm)

15 15

10 10

5 5

0 0
0 5 10 15 20 25 0 5 10 15 20 25
Intensity Measure, Sd(TLS) (cm) Intensity Measure, Sd(TLS) (cm)

Figure 3.15 – Intensity measure vs. EDP using dynamic analysis-measured Teff and ductility.

Some of the assumptions that have been made to streamline the procedure can be discussed
and revisited:

3.24
 Seismic Input

The Limit State strains considered for concrete and steel may be reviewed, or
adjusted, since this is fundamental data for the displacement capacities of the
structure, within the intensity measure procedure; the fact that yielding and
subsequent Limit States are assumed in the structure as soon as they occur in one of
the piers, may as well influence the procedure in a relevant manner. A possibility
would be to try an approach similar to what is typically followed for buildings,
where the yielding displacement is taken as the one corresponding to 75% of the
maximum base shear. However, this may not be easy, since it can be seen that the
first pier reaches LS2 and LS3 shortly after yielding. Additionally, the definition of
the maximum base shear it s not clear for all the bridge configurations;

The equivalent SDOF displacement is another important issue, which has been
defined according to Equation (3.1). A possible alternative is to use the
displacement corresponding to the reference location, xref, shown in Equation (3.2).
For the buildings case, the effective height is used, which seems to be a simpler
approach. An approach simply using the deck central node as reference, as
proposed in EC8 and other codes and guidelines within nonlinear static procedures,
seems therefore worth a try;

The relationship of the effective measured periods with actually observed Limit
State in the dynamic analysis, as well as with the effective LS periods coming from
Priestley or FEMA approaches, is still missing. This may be important to justify the
use of such effective periods instead of the empirical proposed ones;

The use of a reference node which is dependent on the displacement shape renders
also somewhat difficult the evaluation of dynamic analysis results. The
displacement shape at the Limit State that the bridge was considered not to exceed
was used, in agreement with the determined intensity measure. For that shape the
reference location was computed and response of that node was submitted to
Fourier analysis to get the effective period. Moreover, EDP results from the first
run had to be used to estimate the measured ductility EDP/LSy used on the second
run, given that EDP results depended on the Intensity Measure which was still
being computed. This seems to suggest an iterative procedure. Anyhow, EDP used

3.25
Seismic Input

to estimate ductility and EDP obtained in the new run proved not to be very
different.

3.4 Artificial accelerograms

It has been discussed, up to now, the use of real accelerograms as well as additional needed
procedures, such as selection and scaling, which come from the mentioned and observed
dispersion level among this sort of records. Consequently, at this point, the consideration
of artificial accelerograms starts to gain relevance. The use of artificial accelerograms is no
innovative, yet relevant, tool, especially if the difficulty in assuring a representative real
ground motion records base is considered. From a design code point of view the use of
artificial records is taken into account, namely on Eurocode 8, which considers their use as
much as recorded accelerograms. The use of simulated accelerograms, as they are referred
to for the analysis of buildings, shall be adequately qualified with regard to the
seismogenetic features of the sources and to the soil conditions appropriate to the site. Also
for bridges, appropriate simulated accelerograms may be used in case the required number
of recorded ground motions is not available. The choice for this type of records is
surrounded by several issues, as previously introduced, being that the main disadvantage
lies exactly on their nature, and the possibly associated lack of authenticity, whereas its
major argument is the overcoming of the complexity allied to the use of real records.

The term artificial is herein employed with a wide meaning and one can argue that the
general designation non-real or synthetic records should be used, instead. Briefly, these can
refer to spectrum compatible artificial earthquake records, which are based on random
vibration theory and wavelets and make use of spectral density function and random
phases (e.g., SIMQKE (Gasparini and Vanmarcke, 1976)) or non-stationary stochastic
vector processes (e.g., TARSCTHS (Papageorgiou et al., 2001)) or to synthetically
simulated records, which are generated through complex numerical simulation of source
and travel path mechanism of a seismic event. Engineering applications of these models
are still few at the present, whereas much progress is being achieved on the models and on
their calibration (Hwang and Huo, 1994; Lam et al., 2000), which contextualizes the
consideration of artificial records in this study. Therefore, the goal of this section consists
mainly on scrutinizing the advantages that can be found when using artificial

3.26
 Seismic Input

accelerograms compatible with the set of selected real records, by direct comparison of the
structural response.

3.4.1 Algorithm of generation

The frequency is the varying parameter along the duration of the time history. The chosen
process of coming up with artificial accelerograms consisted on generating earthquake
time-histories compatible with an elastic velocity response spectrum, based on the
methodology proposed by Gasparini and Vanmarcke (1976), later improved by Barbat and
Canet (1994). The procedure is based on the definition of the ground motion records, ä(t),
by a superposition of series of sinusoidal waves, as in Equation (3.9). Frequency,
amplitude and phase angle, are therefore the main varying parameters, according to the
same equation.

n
a&&(t ) = I (t ) ⋅ ∑ Ai sin (ωi t + Φ i ) (3.9)
i =1

Ai represents the amplitude, ϖi each angular frequency and Φi is the phase angle.
Essentially, by generating different arrays of phase angles, different ground motion records
can be generated. I(t) is an intensity envelope function. The algorithm starts by setting n
frequencies (ϖi) equally spaced and n phase angles randomly generated (between 0 and
2π). A group of amplitudes (Ai) is obtained by using the simplified relation, Equation
(3.10), with the power spectral density function (Sẍ), which can be determined according to
Gasparini and Vanmarcke (1976), as in Equation (3.11).

Ai ≈ 2 ⋅ S &x& (ωi ) ⋅ ∆ωi (3.10)

3.27
Seismic Input

ϖ 2i ⋅ S v2 (ϖ i ) ωi 
S &x& (ϖ i ) ≈ S &x& (ϖ ).dϖ 
1
 π
⋅
  ζ p2,t e
− ∫
0  (3.11)
ϖi ⋅ − 1 
 4ν t 
 e 

In Equation (3.11), ζp,te represents the peak factor, defined as a function of the duration of
the generated motion, te, and of a probability level, p. Numerical research has shown that
for a value of p= 0.367 (corresponding to ln(p)= –1) rather good results are obtained, for
relatively short ground motion signals. Sv is the velocity response spectrum and νte is the
fictitious time-dependent damping factor, a fraction of the artificial critical damping factor.
Equation (3.11) has been employed by using the smallest natural frequency, ω1, with
which the integral in the right becomes zero. The adjustment between provided and the
generated spectrum is best for larger values of the duration, te.

3.4.2 Reference response spectrum for generation of records

Recalling one of the main goals of this section, to establish the comparison of the structural
response coming from the use of real and artificial ground motion records, in order to do
so, equivalent/comparable properties for both types need to be established. Hence, the
generation of artificial records has been based on equal probability response spectra and
knowledge of the corresponding Hazard distribution (Delgado et al., 2006). By considering
a specific exceeding probability level it is possible to define a reference response spectrum,
which can be in terms of peak ground acceleration (PGA) or spectral acceleration (Sa).

Defining the reference spectrum in terms of PGA consists in imposing the same peak
ground acceleration for the mean real response spectra, obtained by the peak ground
acceleration (pga) scaling technique (Section 3.3), and for the isoprobable response
spectrum. On the other hand, when defining the reference spectrum as a function of
spectral acceleration, the characteristics of the structure being analysed, namely the
fundamental vibration period, are taken into account. Taking the mean real response
spectrum, the corresponding exceedance probability level can be taken from the
isoprobable spectrum. That mean response spectrum is then scaled to match the spectral
acceleration of the isoprobable at the structure’s period.

3.28
 Seismic Input

Once the reference response spectrum is defined, the evaluation of the scatter levels
concerning the response of the analysed structures when subjected to real and artificial
records.

3.4.3 Artificial ground motion records generation parameters

The generation code established that the response spectra of the artificial records are
compatible with the reference one for a scatter (measured in terms of COV) lower than
20%. The main characteristics of the artificial ground motions records are related to the
duration, the intensity envelope function’s type, the time step and the peak ground
acceleration. As there are no criteria established in design codes for the duration, the
recommendation of a value not lower than 10 seconds was followed (CEN, 2005a). All the
signals were generated with 10 seconds and a time step of 0.01 seconds, assumed as
detailed and stable enough. A trapezoidal envelope function was adopted for the intensity,
for which the beginning and ending time of the constant function value had to be
prescribed.

3.5 Comparison between real and artificial records – structural application

It has been stated that scatter effects caused by the use of real records have different
meaning and proportion whether if SDOF or MDOF systems are to be considered. The
initial analysis on response spectra dispersion has shown scatter levels associated to
different scaling techniques. Subsequently, structural effects on MDOF structures, which
probably constitute the case of major interest on this kind of analyses, should be further
addressed. The role associated to nonlinear features on different structures makes the
nature of accelerograms and eventual scaling procedure much more relevant. Herein, a
MDOF structure (RC viaduct) has been considered as case study. The structure has been
subjected to the selected real records, scaled to peak ground or spectral accelerations, as
well as to comparable artificial accelerograms. Nonlinear analyses have been carried out
using PNL software, which is based on plastic hinges models. The algorithm, developed in
the eighties, has been extensively validated, supported on experimental testing, for the past
few years. More information may be found in (Costa and Costa, 1987; Costa, 1989;
Varum, 1996). Studied nonlinear response effects refer to ductility in curvature for critical

3.29
Seismic Input

sections. Statistical analysis has been applied, characterizing different results in terms of
average and standard deviation/coefficient of variation as dispersion measures.

3.5.1 Structural description

The structure that has been object of several dynamic nonlinear analyses consisted of the
Lordo viaduct, located on a mountainous area of high seismicity in Southern Italy, region
of Reggio Calabria, which has been studied within the European Integrated project
LESSLOSS (2004a). The structure is defined by two independent and parallel viaducts
corresponding to two motorways, with slight curvature in plan – Figure 3.16.

Figure 3.16 – Lordo viaduct (LESSLOSS, 2004a).

The analysis will focus the south viaduct, defined by piers 1 to 10 and abutments A and B,
hence, eleven spans with variable length between 40 and 110 meters, as shown in the
longitudinal profile in Figure 3.17. The deck is a continuous beam composed by a hollow
composite steel-concrete girder of rectangular cross-section, strengthened by diagonal
braces, struts, flanges and transverse stiffeners and diaphragms along span and at
abutments. The deck cross section has constant height over the major part of the spans.
Regarding the piers, they present polygonal hollow section with two different types. For
the piers 2 and 3 the cross-section was set for hydrodynamics reasons, therefore, these are
the only piers with hexagonal cross-section (6m deep and 3m wide); the remaining ones
feature rectangular hollow section with central sept (6m deep and 2m wide), as in Figure
3.17. Piers 2 to 5 are supported on piles, whereas the other piers have direct foundations.

3.30
 Seismic Input

Longitudinal profile - South Viaduct

A 1 2 3 4 5 6 7 8 9 10 B
60.00 75.00 110. 00 85.00 60. 00 60. 00 60.00 60.00 60.00 45.00 40.00

O
F OS S I NE
I Z O
DEVA
Deck cross-section Deck cross-section
over piers 2 and 3 over piers 1 and 4 -10
10.95 Cross-section Cross-section 10.95
. 0
0 6 0 .6 0
5
9 . 7 . 5
9 7

Piers 2 and 3 Piers 1 and 4 -10

0 .6 0 . 0
0 6

. 4 5
5 4
.5 5
2
0. 0

0 .2 0
1
0 . 2

0 .1 2
0 .0 3 .0 0 3
0 .3 5

0 .3 5
d . rt a sv .
Pe n n . rt a sv .
Pe d

0.7 5

0.30+2.50
4.50+0.30

0.0 5

0. 30

0 .30

2.00
3.00
1 .5 0

1 .40
0 .6 0

0 .30
0.3 0 2 .3 0 0 .30

0 .7 5
3.0 0 3.0 0

5.50
6.00

Figure 3.17 – Lordo Viaduct – South viaduct longitudinal profile and cross section of the piers
(LESSLOSS, 2004a).

The structure is schematically illustrated in Figure 3.18 and the critical sections in terms of
ductility demand, defined from preliminary analyses, are marked. For the subsequent
applications, due to similar ductility demand levels, only the P3-related results will be
presented.

Figure 3.18 – Lordo viaduct – critical sections: P3 and P4.

3.5.2 Real records – original and homogeneous bases

Effects of the original set of accelerograms on the structure are represented in Figure 3.19,
left. As expected, the original, unprocessed set of records yields quite scattered results,
across all intensity levels (each line represents a level). The graphical representation
enables the expedite conclusion that peaks are associated to accelerograms causing
instability, which advocates for the use an aggregating technique similar to the one
performed for response spectra in Section 3.2.2. As described in that section, a new cluster
analysis was therefore carried out. Linear dependence between variables was not evaluated
but Euclidean distances were measured instead. The cluster analysis highlights similar
elements, which allows disregarding records causing the major scatter effects, hence,
leading the original set of results to a higher homogeneity level, as illustrated in the same

3.31
Seismic Input

figure, on the right. The new reduced set of results corresponds to 15 of the initially
selected records.

80 80
pga initial pga selected
70 70

60 60
Ductility Demand

Ductility Demand
50 50

40 40

30 30

20 20

10 10

0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Record Record
45 45
sa initial sa selected
40 40

35 35
Ductility Demand
Ductility Demand

30 30

25 25

20 20

15 15

10 10

5 5

0 0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Record Record

Figure 3.19 – Ductility demand at LORDO viaduct for the original (left) and reduced (right) base
of real records, for the different records, scaled according to peak ground or spectral acceleration
(Pier P3).

3.5.3 Minimum necessary number of accelerograms

The initial set of records includes twenty accelerograms whereas the homogeneous base
has been set with fifteen accelerograms. When carrying out nonlinear dynamic analyses,
the use of whether twenty or fifteen records will definitely be more representative and
advantageous in the assessment of the scatter around the structural performance
predictions. However, twenty accelerograms imply an equal number of different, dense and
often time-consuming analyses, possibly leading to low feasibility levels. The number of
records to use for seismic analysis will not necessarily have to be that large, given that it is
expected that the engineering demand parameter at stake, such as ductility demand, will

3.32
 Seismic Input

stabilize, for a given tolerance, around a specific value. It can then be worthwhile to look
into the possibility of reducing the number of records to a minimum able to adequately
represent the whole set and produce the same mean/median structural response.

A former reduction procedure on the number of records has already been applied to the
original set of twenty accelerograms by removing the most diverging ones through cluster
analysis, as detailed in Section 3.2.3. It is thought to be possible to again reduce the fifteen
records from the homogeneous base and still achieve the same mean structural response
(ductility demand). The adopted procedure to study this intended reduction was based on
quite simple considerations, although other studied have gone through this matter on the
basis of confidence intervals (see Section 2.3.2). Samples of 3, 4 and so on, up to 10
accelerograms, were defined randomly. The n possible number of distinct samples of a
certain size p, is thus given by combinations of n, p to p. For each sample, the mean
ductility demand, Xi, and the coefficient of variation (COV) to the “real” mean value, X,
taken from all the 15 elements, is computed, according to Equation (3.12).

1 n
∑ ( X i − X )2 (3.12)
n i =1
COV =
X

Representing COV as a function of the sample dimension, one may expect to identify a
number of accelerograms that guarantees a satisfactory low distance to the “real” value,
which refers herein to the average result coming from the homogeneous non reduced set of
records. Figure 3.20 presents the curves obtained to the critical piers cross sections, for
several seismic intensity levels. The results for the remaining piers have been found to be
similar.

3.33
Seismic Input

25
0.5g
P3 0.75g
20 1.0
1.25g
1.5g
1.75g
15
COV (%)

COV (%)
2.0g

10

0
3 4 5 6 7 8 9 10 11 12
Sample Size

Figure 3.20 – Ductility demand COV according to sample size (Pier P3).

General scatter increases with the increase in nonlinearity, i.e., intensity level, which is an
expected feature, even though major differences come up for peak ground acceleration of
2g, only. Moreover, it can be observed how the decrease rate along with the sample size is
rather constant and, hence, it is difficult to define a number of records complying with the
defined goal. Notwithstanding the difficulties in interpreting the results, a deceleration in
the COV decrease is slightly visible from 6 records on and it is interesting to notice how
the dispersion level with respect to the real mean value can be seen as reasonably low –
mostly below 20%. If at least 6 records are considered, the COV lies around 10%. Using 7
records, instead, that amount is never reached.

The number of seven records is actually mentioned by the EC8 (CEN, 2005a) as the
minimum necessary for the average of the individual responses to be used, which is
reassuring, especially when the focus is on real records. In case less than seven nonlinear
dynamic analyses are carried out, the maximum response should be used.

3.5.4 Artificial records

Considering the conclusions just drawn on the suitable number of accelerograms to carry
out nonlinear dynamic analysis, the use of seven artificial ground motion records seems a
rational option, at least for a starting point. Most likely, and due to expected increased
homogeneity, less than seven artificial accelerograms will be enough to obtain a stable
median result. The ductility demand for each of the seven records, over the different

3.34
 Seismic Input

intensities, is represented in Figure 3.21, for the critical pier cross section. As expected, the
scatter characterizing results is much lower despite the fact that one of the records still falls
relatively out of the general panorama of the pga-based artificial accelerograms. The effect
of one outlier will be, however, easily absorbed by the mean or median value.

60 60
pga artificial sa artificial 0.5g
0.75g
50 50 1.0
1.25g
1.5g
Ductility Demand

Ductility Demand
40 40 1.75g
2.0g

30 30

20 20

10 10

0 0
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Record Record

Figure 3.21 – Ductility demand at pier P3 for the artificial records base, for different intensity
levels.

3.5.5 Comparison of results

The use of artificial records as an alternative to real accelerograms, or vice-versa, is still

matter of discussion, hence, the comparison of engineering demand parameters (ductility
demand) mean values as well as the associated scatter, when using both types of records is
pertinent. The initial set of real accelerograms (initial real) and also a new set involving
only the most similar 15 accelerograms (selected real) were used for analysis and
comparison. For the purpose of comparing the performance of real and artificial records,
the minimum number of 7 records has not been used given that, apart from the 10%
residual, the outcome is the same. Furthermore the comparative endeavour was developed
for the two scaling techniques previously selected, peak ground acceleration (pga) and
spectral acceleration (sa). Artificial records were generated in agreement, as well.

3.35
Seismic Input

140
pga initial real
pga selected real
120 sa initial real
sa selected real
pga artificial
100
sa artificial
COV (%)

80

60

40

20

0
0.5g 0.75g 1.0 1.25g 1.5g 1.75g 2.0g
Intensity

Figure 3.22 – Coefficient of variation for the ductility demand, according to the different scaling
techniques and types of accelerograms.

30
pga initial real
pga selected real
25 sa initial real
sa selected real
Mean Ductility Demand

pga artificial
sa artificial
20

15

10

0
0.5g 0.75g 1.0 1.25g 1.5g 1.75g 2.0g
Intensity

Figure 3.23 – Mean ductility demand for different scaling techniques and types of accelerograms.

According to Figure 3.22 real accelerograms scaled according to peak ground acceleration
(pga) yield the highest scatter levels, even if considering the selected base of records. The
same happens for the case of pga-scaled artificial records, when compared to sa-scaled
ones. Nevertheless, artificial accelerograms, as expected, feature the lowest dispersion.
When using real records scaled by means of spectral acceleration at the structural period
(sa) the coefficient of variation lies between the two mentioned categories. If the selected

3.36
 Seismic Input

real base is used, instead, the dispersion associated to pga scaled records moves closer to
the one obtained with sa scaling technique applied to the initial set of accelerograms.

With respect to the mean ductility demand, Figure 3.23, considerably higher mean values
are obtained when using pga scaled records and the initial base whilst rather lower ones
(around 50% of the maximum pga numbers), and the lowest at the same time, come from
the use of sa scaling technique, which yields similar results to both initial and selected base
of records. Employing artificial records, instead, again, intermediate or close to pga initial
real values for the ductility demand are obtained, which may denote an underestimating
trend for the sa results. Typically, all the alternative approaches tend to differ more among
each other, in terms of median values, as the intensity level increases, whereas, considering
COV, the difference tends to reduce, which can be advantageous from a collapse or near-
collapse limit states analysis point of view. The ductility demand estimates obtained when
carrying out the analysis with artificial records are generally in between the ones coming
from initial and reduced set of pga scaled accelerograms, which accounts for validity in the
use of artificial records, involving a lower number of analyses, as an alternative input to
nonlinear dynamic analysis.

3.6 Conclusions

The definition of the seismic action, together with all of the subsequent issues, is related to
the type of analysis. If nonlinear dynamic analysis is considered, seen as the most accurate
way of estimating the structural response, the seismic action is represented by
accelerograms; their selection, in number and type, scaling or homogenization has proven
to be no straightforward task. This chapter intended to provide a contribution, firstly, to the
matter of choosing between real ground motion records or artificial accelerograms.
Approaches to deal with the scatter, usually large, associated to the real records have been
addressed, as well as the topic of how many records to consider. A revisiting of traditional
scaling techniques, also seen as intensity measures, has been carried out focusing on
records or response spectrum base quantities. More advanced, yet preliminary,
displacement-based intensity measures have also been proposed. Finally, a comparison
between the use of artificial and real records, based on the structural response of a bridge

3.37
Seismic Input

MDOF system, has been presented. The main observations and conclusions are discussed
next.

A first approach to the consideration of real records issue has again emphasized the
considerable existing scatter, through the analysis of a database of real
accelerograms. A “filtering” technique, based on statistical clustering analysis, has
been applied to the initial set of records, in order to harmonize the database,
eliminating the individuals considered to be statistically out of the group. The
technique revealed itself easy to apply, making use of proper statistical tools and
the so called homogeneous base was then obtained.

Apart from selecting the most significant records for analysis, having eliminated
the most “distant” ones, the most efficient parameter to use when standardizing
them to a comparable level has been looked for, revisiting the typical used
measures, adding some less used ones. The comparative study considered eight
parameters, based on records or corresponding response spectrum quantities: peak
ground acceleration; square ground motion; root-square ground motion; mean-
square ground motion; root-mean-square ground motion; spectrum intensity;
spectral acceleration and power spectrum area. The dispersion around the spectral
ordinates, measured in coefficient of variation, was evaluated for the
correspondingly scaled accelerograms. In general, all the parameters behaved
similarly and it has been observed that, given a relatively heterogeneous database
of 20 ground motion records, the traditionally used peak ground and spectral
acceleration intensity measures performed superiorly, with respect to reducing the
scatter around spectral ordinates. Such observation features two positive aspects;
firstly, it goes along with most of past research work on this matter; furthermore,
despite the development of more refined intensity measures, recently proposed by
some authors (please refer to Section 2.3), the availability of Hazard data for such
measures is very limited, if not inexistent (this does not happen with peak ground
or spectral acceleration, which have been widely used). The latter becomes
particularly important within a probabilistic vulnerability assessment framework,
such as the one carried out in Chapter 6, which requires the seismic action to be
defined probabilistic. Indeed, typically, the probability density functions for the

3.38
 Seismic Input

seismic action are available for the peak ground acceleration (or spectral
acceleration, sometimes) intensity measure.

Moving forward in the intensity measures subject, an advanced formulation, yet

simplified in terms of analysis, has been studied. The proposed spectral
displacement-based intensity measure was based on the establishment of limit
states (defined through material behaviour) and corresponding secant periods,
determined by means of nonlinear static analysis. Different pushover analysis
types, as well as damping-based reduction factors, have been tested in an ensemble
of fourteen bridge structures and ten ground motion records. Interesting trends
between the results of nonlinear dynamic analysis and intensity measure have been
encountered, although further development will be needed, so as to improve its
efficiency. Indeed, some difficulties have been found in the definition of
displacement based theory simplified quantities for generalized bridges, such as
reference node location, which is somewhat expected, given that bridges are well-
known for its potentially high irregular behaviour, in height and plane.
Nevertheless, the experienced shortcomings have not prevented the proposed
measure from revealing itself as promising.

A structural application to investigate the viability of using artificial accelerograms,

when compared to real records, with respect to the structural effects (engineering
demand parameter considered as the ductility in curvature at the piers) in an
irregular viaduct has been carried out. Artificial records constitute a very attractive
alternative for two main reasons, which motivated the study. Firstly, the (complex)
process of selection and scaling of real records would be avoided. In addition,
within a safety assessment procedure based on nonlinear dynamic analysis, such as
the one in Chapter 5, the number of records to assure accurate results would be less,
with respect to real accelerograms, enabling therefore significant savings in
computational demand. The clustering technique was therein applied to the
ductility demand coming from the use of the initial base of accelerograms, leading
to the homogeneous base of records. Such procedure was actually something new,
given that a selection on the basis of a MDOF system response was employed.

3.39
Seismic Input

In order to avoid numerous nonlinear dynamic analyses, a simple parametric study

was carried out to determine the minimum number of real accelerograms that
would lead to the same results of the homogeneous base (within a specified
tolerance). It has been observed that 7 records would represent a good compromise
between feasibility and significance. To summarize, starting from a 20-record set, a
homogeneous base of 15 records has been derived from clustering analysis and, in
the end, a 7-record database could be used, assuring acceptable representativeness.

A simple algorithm of generation of artificial records was then selected and 7

accelerograms have been obtained, following the conclusions of the study on the
number of records. The comparison with real records (original and reduced set)
scaled in the two traditional intensity measures that performed best (peak ground
acceleration and spectral acceleration) was then carried out. The major conclusions
pointed the use of artificial records as valid, yet conservative, with logically much
lower scatter levels, standing between the median effect of the original set of
accelerograms, scaled by mean of peak ground acceleration and spectral
acceleration.

Once duly characterized, the seismic input will be applied to a proper nonlinear model of
the structure. As certainly not exempt from discussion, the structural nonlinear modelling
task will be addressed in the following chapter.

3.40
4. Nonlinear Modelling

The structural modelling topic is typically known for comprising a high number of
challenging issues, some necessarily calling out for assumptions to be made and some in
need for further development. Apart from the choice for the sort of model to use to
represent the bridges to study, the option between fibres or plastic hinges as the mode of
characterizing the material nonlinearity, the most important within bridge analysis (see
2.5), is another determinant aspect.

This chapter will consist, therefore, of detailing modelling aspects surrounding the work
that has been carried out. A brief mention to the type of models that have chosen to
represent a set of bridges is initially carried out, distinguished in accordance with the sort
of material nonlinearity that is respectively considered, the computation tools and analysis
procedures that are available. The used material models, concrete and steel, are afterwards
contextualized and described, highlighting the potentialities that lead to their choice.
Finally, and recognizing it as one of the current key issues, a parametric study is carried
out to compare the use of lumped (plastic hinges) or distributed (fibre models) plasticity
for the nonlinear material modelling.

The models that have been considered, in terms of spatial configuration, were selected
considering different criteria. Regarding the computation available tools, the lumped
plasticity models were idealized using software SAP2000 (Computers&Structures, 2006),

4.1
Nonlinear Modelling

defining the plastic hinge constitutive law by means of a fibre based cross section model in
BIAX (Vaz, 1992), whereas fibre models were built up using SeismoStruct (SeismoSoft,
2008). Both of the computer programs enable 3D frame models, which are, as stated in
(Casarotti and Pinho, 2006), currently the best compromise between simplicity and
accuracy, providing reasonable insight on the seismic response of both members and
global structure.

4.1 Material models

A brief discussion on the concrete and steel models that have been used to characterize the
bridge piers cross sections is in what follows. Whereas steel has been defined recurring to
the same base model, concrete behaviour, more complex and less standardized, has been
characterized through different approaches, depending on if plasticity is concentrated or
distributed.

4.1.1 Concrete models

Several concrete models are available in the literature, with different sophistication levels.
The importance of highly refined concrete models in the moment-curvature response of
reinforced concrete cross sections has been confirmed as not too relevant, in accordance
with analytical studies conducted by Aktan and Ersoy (1979). This is most likely due to the
fact that the expected ductile behaviour of RC sections will rely essentially on the steel
contribution, rather than the concrete one. This main feature will be of great importance
when carrying out, ahead, comparison parametric studies that make use of distinct concrete
models.

The typical concrete monotonic behaviour goes through relatively well defined different
damage stages, from initial cracking to rupture. The stress-train diagram has a first
approximately linear region which lasts until around half of the maximum compressive
strength is verified. From that point on, a nonlinear behaviour stage follows, classically
assumed as parabolic, caused by the considerable stiffness reduction due to cracking. The
end of the 2nd degree polynomial branch corresponds to the peak compressive strength,
which can be more or less evident, depending on whether the strength level is high or low.
After the stress peak, the resisting compressive stress diminishes significantly until the

4.2
 Nonlinear Modelling

ultimate strain is reached, corresponding generally to about 20% of the compressive peak
stress. That last descending portion is modelled as linear. Tensile concrete strength may be
assumed, typically between 10 and 20% of the compressive strength. As very low
contributing to the behaviour under bending forces, the tensile strength turns out to be
generally neglected.

Early proposals for concrete model constitutive stress-strain relationships came up in the
mid-late 1960s, by Shina et al. (1964) and Karsan and Jirsa (1969), both based on
experimental tests of a large number of concrete cylinders or cubes, subjected to uniaxial
compression. These were considerably preliminary models, which were, nevertheless,
useful to study and recognize cyclic loading and unloading path features of the concrete
material. Afterwards, from the decade of 1970 on, the number of proposals continued to
increase, together with several variants and modifications, as summarized in a work by
Yeh et al. (2002), referring to seismic performance of rectangular hollow bridge piers,
which considered nine different stress-strain rectangular reinforced concrete models and
found no significant dispersion in different model predictions, in particular for the
moment-curvature plots. The selection included well known and widely spread out model
from Kent and Park (1971), in its unconfined and confined versions; the modified Kent and
Park model by Park et al. (1982); a model proposed by Muguruma et al. (1978), similar to
the confined Kent and Park model with a different stress-strain curve shape and ultimate
confined concrete strain; the Sheikh and Uzumeri model (Sheikh and Uzumeri, 1980,
1982), similar to the modified Kent and Park model, except for a flat ultimate strength
within a strain range; the popular Mander et al. approach (Mander et al., 1988a; Mander et
al., 1988b), which makes use of a confining pressure to affect the shape of the stress-strain
curve; a proposal from Fujii et al. (1988), a model that is similar to the Muguruma et al.
(1978) model except for different control parameters; a modification of the modified Kent
and Park model with effective confining pressures of the rectangular cross section in both
directions, presented by Saatcioglu and Razvi (1992) and a model from Hoshikuma and
Nagaya (1997), which does not use effectively confined core area and suggests a residual
strength of half of the peak compressive strength at large strain.

The lumped plasticity models, carried out in SAP2000, using a RC cross section
constitutive trilinear law defined in BIAX (Vaz, 1992), make use of the modified
Kent-Park model, whilst the fibre-based finite element models idealized in SeismoStruct

4.3
Nonlinear Modelling

considered the Mander-Priestley-Park model. Both of them are described in higher detail,
as follows.

4.1.1.1 Modified Kent-Park model

The concrete nonlinear behaviour proposed by Park et al. (1982) is basically the
modification of the originally presented by Kent and Park (1971), approximately one
decade before. Such initial model takes into account the confinement effect in the concrete
ductility, provided by the transverse reinforcement steel but not the corresponding increase
in the compressive strength. The modification carried out by Park et al. consisted of
introducing a confinement factor, k, which accounts for the confinement phenomenon
when predicting the compressive strength, as well as the corresponding strain.
Additionally, it features a reduction in the slope of the descending branch in the tension-
strain diagram, reproducing the improvement in the concrete ductility provided by the
confinement. The stress-strain diagram is represented in Figure 4.1, where three distinct
branches, defined from points A to D, are distinguished as follows.

σc

B
kfc
tg(θm)=Zm k fc

fc θm

Confined Concrete

Unconfined Concrete
C D
0.2kfc

A
0.002 0.002k ε20c εc

Figure 4.1 – Modified Kent-Park model – confined concrete stress strain envelope diagram for
monotonic loading.

4.4
 Nonlinear Modelling

A–B (εc < 0.002k)

 2ε  2ε c  
2

σ c = k fc  c
−    (4.1)
 0.002k  0.002 k  

B–C (0.002k ≤ εc < ε20c)

σ c = k f c [1 − Z m (ε c − 0.002k )] (4.2)

C–D (εc ≥ ε20c)

σ c = 0 .2 k f c (4.3)

The confinement factor, k, is given by Equation (4.4); Zm, related do the slope, θm, of the
descending branch B–C, is defined in Equation (4.5); εc is the longitudinal concrete strain,
σc is the concrete compressive stress, (in MPa); fc is the unconfined concrete compressive
strength (in MPa); ε20c is the concrete strain corresponding to 20% of the maximum
compressive strength, in the branch B–C; fsyt is the yielding stress of the transverse
reinforcement steel (in MPa); ρν is the ratio between the transverse reinforcement steel
volume and the confined concrete volume; h is the confined concrete width and s is the
transverse reinforcement steel spacing.

ρν f syt
k = 1+ (4.4)
fc

4.5
Nonlinear Modelling

0 .5
Zm =
3 + 0.29 f c 3 h (4.5)
+ ρν − 0.002k
145 f c − 1000 4 s

The specific rules to model the concrete behaviour under repeated loading were originally
proposed by Park et al. (1972) but it has been foreseen that alternative rules from
Thompson and Park (CEB, 1983) are more realistic and were hence employed together
with the stress-strain model.

4.1.1.2 Mander-Priestley-Park model

The most interesting feature of the model proposed by Mander et al. (1988b) is probably
that it is fairly generalist. Indeed, it is possible to explicitly consider the sort and
arrangement of the transverse reinforcement steel, within a similar procedure to the one
adopted by Sheikh and Uzumeri (1979), when developing their model under monotonic
loading. The stress-strain constitutive relationship is illustrated in Figure 4.2 and is given
by the expression in Equation (4.6), in which the variables assume the following indicated
meanings.

σc

Confined Concrete
First hoop
' fracture
f cc

Unconfined
Concrete
f c'0
Assumed for
Ec cover concrete
Esec

εc0 2εc0 εcc εcu εc

Figure 4.2 – Mander et al. stress-strain model for confined concrete under monotonic loading.

4.6
 Nonlinear Modelling

f cc' x r
σc = (4.6)
r −1+ xr

f’cc is the confined concrete compressive strength; εc is the longitudinal concrete strain; f’c0
is the unconfined concrete compressive strength; εc0 is the concrete strain corresponding to
f’c0; x is the ratio between εc and εcc, given by Equation (4.7); r, in turn, is defined in
Equation (4.8).

  f cc' 
ε cc = ε c 0 1 + 5 '
− 1  (4.7)
  f c0 

Ec
r= ; Ec = 5000 f c'0 and Esec = f cc' ε cc (4.8)
Ec − Esec

Within the fibre modelling that has been carried out the cyclic rules proposed by
Martinez-Rueda and Elnashai (1997) have been used. The confinement effects provided by
the lateral transverse reinforcement were, in turn, incorporated through the rules proposed
by Mander et al. (1988b) whereby constant confining pressure is assumed throughout the
entire stress-strain range. The model considers, additionally, tensile strength, even if
evidently residual, which depends on the compression strain behaviour. It is assumed that
for εc > εcc, or as soon as cracking occurs, for the following loading cycles, the tensile
strength becomes null.

4.1.2 Steel models

Steel is, by nature, a less complex than concrete behaving material, which will be reflected
in the number and sophistication level of available models. Indeed, the steel monotonic
behaviour can be essentially divided in three stages: a first branch, linear, where stress is
constantly proportional to the strain; a second, where the stress is practically constant for
increasing strain, the yielding plateau; and a final, where the stress increases again until

4.7
Nonlinear Modelling

rupture occurs. Such typical behaviour refers to hot rolled steel, whereas cold rolled steel
does not exhibit the yielding plateau, with higher tensile strength and lower deformation
capacity. Some of the steel models are developed on the basis of material constitutive
theories; however, the majority of those are phenomenological models that characterize the
macroscopic response on the basis of experimental data. The fundamental characteristics of
the steel response are relatively simple, thus, the appropriate material model not only
predicts such response with reasonable level of accuracy but is also calibrated to fit
experimental data with relative ease.

A number of models characterizing the response of reinforcing steel subjected to reversal

cyclic loading on the basis of microscopic material response were identified and discussed
by Cofie (1984).

More representative models for the response of reinforcing steel subjected to reversing
cyclic loading can be achieved through the use of phenomenological models in which
nonlinear equations are calibrated on the basis of experimental data. Several models on
such basis have been proposed, since one the first approaches, in 1943, from
Ramberg and Osgood (1943). Bertero and Popov (1976) presented a model consisting of a
monotonic stress-strain law defined by seven points; Atkan-Karlson-Sozen (CEB, 1983)
proposed stress-strain relationships made up of four zones, between tow consecutive
loading reversals; Pinto and Giuffrè (1970) came up with a more complex behaviour
model, including asymptotic representation of hardening and Bauschinger effects. Shortly
after, Menegotto and Pinto (1973) applied the Giuffrè-Pinto model, systemizing a set of
nonlinear equations describing steel behaviour, which served as a base for a bunch of other
improving models: Stanton and McNiven (1979) improved computational efficiency;
Filippou et al. (1983b) worked better on the unloading response, with both reasonably
accurate prediction of response and relatively simple implementation and calibration;
sophisticated Chang and Mander (1994) model accounted for cyclic strain hardening,
providing quite accurate predictions of steel response.

Both lumped and distributed plasticity models have been defined taking the same
Giuffrè-Menegotto-Pinto steel model, eventually applied with further refinement for the
case of the SeismoStruct fibre models. A more detailed description of the main aspects of
such model is carried out next.

4.8
 Nonlinear Modelling

4.1.2.1 Giuffrè-Menegotto-Pinto model

The Giuffrè-Menegotto-Pinto model, used to describe the cyclic behaviour of the steel was
originally elaborated by Pinto and Giuffrè (1970) and later improved by Menegotto and
Pinto (1973). It manages to model the changes in stiffness and strength of the steel, due to
the inversion during cyclic loading, through curves developed within four asymptotes, such
as illustrated in Figure 4.3: two, parallel, with slope Es, based on the elastic branch of the
monotonic diagram, and two other, parallel as well, with slope Es1, corresponding to the
post-yield hardening stiffness.

σs ξ1

Es1 R0

Es

εs

R1(ξ1)
R2(ξ2)
Es1

ξ2

Figure 4.3 – Menegotto and Pinto stress-strain model for steel under cyclic loading.

Loading and unloading paths are completely involved by the monotonic loading bilinear
behaviour law. The general expression for the stress-strain relationship is given by
Equation (4.9), where effective strain and stress (σ*, ε*) are function of the unload/reload
interval; b is the ratio of the initial to post-yield tangent stiffness and R is a parameter that
defines the shape of the unloading curve, representing the Bauschinger effect.

ε*
σ * = (1 − b ) + bε *
(1 + ε )
*R
1R (4.9)

4.9
Nonlinear Modelling

Within the fibre modelling, computation implementation carried out by Monti et al.
(1996), initially performed by Yassin (1994), was used, together with the modification
introduced by Filippou et al. (1983b), to include isotropic strain hardening. This is actually
considered to be one of the most accurate and convenient models to use, combining
computational efficiency and very good agreement with experimental results.

4.2 Material nonlinearity approaches

Two main ways are currently followed when the purpose is to account for the nonlinear
material features: lumped and distributed plasticity models. Within this topic, in the section
that follows, each of the alternatives is looked into higher detail, with particular emphasis
on critical issues. A few calibration studies were carried out, prior to the comparison of
bridge response, obtained using each of the plasticity models.

4.2.1 Lumped plasticity models (plastic hinges)

When considering a concentrated plasticity model, the nonlinear behaviour of the bar
elements is located in a rotational spring in both extremities of the elastic behaviour part of
the element. Indeed, and regarding the particular application to bridges, studies carried out
in the recent past have shown that bridge piers have a clear tendency to assume a nonlinear
behaviour in well defined regions, which somewhat enables the plastic hinge approach
illustrated in Figure 4.4 (Monteiro et al., 2008b).

Figure 4.4 – Plastic hinge development at the bottom of a pier subjected to horizontal load.

4.10
 Nonlinear Modelling

Nevertheless, this kind of simplified model should be handled with care given that
accuracy of the results may be compromised when the user does not have reasonable
know-how on calibration of inelastic elements parameters. The assumption of the
concentrated plasticity zones for the structural elements, with corresponding plastic hinges
formation is currently used to estimate the real deformation capacity, taking duly account
of the material nonlinear behaviour. Using this kind of approach, the nonlinear analysis
process becomes greatly simplified, namely to what concerns the numerical data
processing. The deformation capacity of the element depends on the ultimate curvature and
plastic hinge length; different criteria used for the definition of these parameters may imply
a different deformation level. Within the current work, different possibilities for the
definition of the plastic hinge length were considered in a short complementary parametric
study. Moreover, the characterization of a plastic hinge requires a moment-curvature
diagram to be defined, or other “equivalent” one, which is obtained from the monotonic
loading of the cross section. The carried study has used, for the plastic hinge models
analyses, the bar finite element program SAP2000 (Computers&Structures, 2006) and
BIAX (Vaz, 1992), developed at University of Porto, for the moment-curvature
constitutive laws.

4.2.1.1 Plastic hinge constitutive law – trilinear curve

The plastic hinges constitutive law, commonly expressed in terms of moments-curvatures,

is probably the crucial contributing element within a lumped plasticity model analysis.
That law, representing the behaviour of reinforced concrete sections subjected to bending,
usually exhibits a trilinear approximate format, defined by cracking, yielding and rupture
(Park and Paulay, 1975). Its definition has been typically achieved by means of a fibre
discretization of the cross section of the element being considered, from which the
nonlinear behaviour characterization is intended. Such is the methodology employed by the
aforementioned fibre analysis algorithm BIAX, used and improved for hollow bridge pier
sections in (Delgado, 2000). The section is therefore discretized in fibres with longitudinal
uniaxial behaviour, which are analysed for biaxial bending, for a given axial compression
force, but neglecting shear effects. Fibre detailing distinguishes unconfined and confined
concrete and steel. Figure 4.5 illustrates the fibre modelling of the cross section of the
bridge piers under study.

4.11
Nonlinear Modelling

Concrete

Steel

Figure 4.5 – Pier cross section fibre discretization

The plane sections hypothesis is assumed, catering for the strain at each fibre to be easily
computed, as well as corresponding tension, making use of the employed material
stress-strain relationships for the monotonic loading.

Alternatively, the moment-curvature diagram under monotonic loading can be obtained

through a mechanistic procedure, in which the three characteristic point of the referred
curve, cracking, yielding and ultimate points, are computed using steel and concrete stress-
strain models, as defined, together with geometric characteristics of the section. Arêde and
Pinto (1996) carried out an efficient implementation of this procedure for a generic
T-shape reinforced concrete section, which can become rectangular, given that the
algorithm is general. The characteristic moments and curvatures are obtained by applying
imminent state conditions (known strains or stresses) along with elementary static
principles, provided the external axial force is known. This sort of monotonic loading
trilinear curve definition is particularly useful, within the context of such global section
models, particularly when dynamic calculations with several loading steps, are needed in a
large number, as has been done for probabilistic safety assessment – see Chapter 6.

4.2.1.2 Plastic hinge length and location

The plastic length is equally fundamental within a lumped plasticity approach. The
accuracy of the element behaviour, when admitting that nonlinearity is concentrated in its

4.12
 Nonlinear Modelling

extremity portions, where curvature are assumed to be constant, depends on a correct

estimation of the development extent of such behaviour. When duly characterized, the
limitation of the plastic effects to a delimited portion will enhance major computation
effort savings. The definition of the mentioned length is not, however, straightforward,
given the theoretical and physical complexity of the phenomenon.

The plasticity length to be considered is majorly defined according to the extent where the
longitudinal reinforcement bars have yielded. There are, nevertheless, other important
phenomena that occur, such as the yield penetration, illustrated in Figure 4.6,
corresponding to an additional rotation at the base support of the element, due to the
physical inability for the curvature to go from zero to its maximum value in an
infinitesimal length, leading to an extra rotation. Additionally, the plastic hinge length can
be “spread out” as a consequence of the shear forces cracking, when the plane sections
theory loses validity and steel deformation becomes larger than the computed one.

Shear effect
(Yield penetration)

Model distribution Real distribution

Figure 4.6 – Idealization of curvature distribution (left) and real curvature distribution, accounting
for yield penetration and shear effect (right).

Several empirical expressions have been proposed over the years for the plastic hinge
length, LP, since the first approaches from Baker, Corley or Sawyer, as documented by
Park and Paulay (1975), in which it was essentially assumed that the such length is
proportional to the distance of the critical point to the point of contraflexure. More
recently, Priestley and Park (1984), supported by a large number of experimental tests,
pointed out for the plastic hinge length to be reasonably estimated by half of the element
cross section height, h, according to Equation (4.10).

4.13
Nonlinear Modelling

L P = 0 .5 h (4.10)

A few years later, Kappos (1991), based on experimental tests from Canterbury University
as well, resented the expression in Equation (4.11), for the plastic hinge length, based on
the element length, l, and longitudinal reinforcement steel bars diameter, db.

LP = 0.08 l + 6d b (4.11)

Paulay and Priestley (1992) refined the proposal from Kappos and came up with
Equation (4.12), which includes the steel yielding strength, fsy.

LP = 0.08 l + 0.022d b f sy (4.12)

The more recent version of this approach ended up being presented by Priestley et al.
(2007), re-including the distance to the point of contraflexure, LC, as in the expression of
Equation (4.13), where fsu/fsy is the ratio of the ultimate tensile strength to yield strength of
the flexural reinforcement.

f 
LP = kLC + LSP = 0.2 su − 1 + 0.022 f sy d b (4.13)
 f 
 sy 

If the ratio is high, plastic deformations are expected to spread away from the critical
section as the reinforcement strain-hardens, whereas a low value will correspond to
concentrated plasticity close to the critical section, leading to lower hinge length.

Eurocode 8, in Part 2 – Bridges (CEN, 2005b), prescribes as well an expression for the
estimation of the length of a plastic hinge occurring at the top or the bottom junction of a

4.14
 Nonlinear Modelling

pier with the deck or the foundation body, with longitudinal reinforcement of yield stress
fsy and bar diameter db, as in Equation (4.14).

LP = 0.10 l + 0.015 f sy d b (4.14)

As demonstrated, the number of different proposals for the length quantification of the
plastic hinges is high and, therefore, a small parametric study has been carried out with the
purpose of selecting one of the approaches for both comparative study with fibre models
and subsequent nonlinear dynamic analysis (see Chapter 6). The study consisted of
performing a simple pushover analysis of a single pier supported at the bottom,
corresponding to one of the piers of the considered case study, described in Section 4.3.
The pier is 14 meters high, discretized in seven elements with rectangular hollow cross
section outer dimensions of 2.0x4.0m and constant width of 0.4m, as in Figure 4.7.

Figure 4.7 – Single pier used to calibrate the plastic hinge length (left) and corresponding cross
section (right).

From the presented alternatives for the plastic hinge length, three have been selected for
comparison: Kappos, Priestley and EC8, corresponding to Equations (4.11), (4.13) and
(4.14), which represent the most recent/improved versions of the different authorships. The
length obtained for the different tried versions (Lp1, Lp2 and Lp3) is presented in Table 4.1,

4.15
Nonlinear Modelling

whilst the corresponding base shear-top displacement curves, obtained with SAP2000, are
plotted in Figure 4.8.

Table 4.1 – Plastic hinge length for different versions.

Equation Plastic hinge length (m) Designation

(4.11) 1.24 Lp2
(4.13) 1.55 Lp3
(4.14) 0.77 Lp1

Vb
3000
(kN)

2500

2000

1500
lp1
1000 Lp1
Llp2
p2
500
Llp3
p3

0 d (m)
0 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40

Figure 4.8 – Capacity curves for different plastic hinge lengths.

The numbers in Table 4.1 feature considerable dispersion, with the proposal from Priestley
consisting of a plastic hinge length that is approximately the double of the EC8 one.
Nevertheless, as expected, the results in terms of capacity curves do not differ much,
especially for what concerns the ultimate base shear. On the other hand, the use of different
plasticity lengths passes on to the displacements evolution, when the structure enters the
inelastic field. The size of the difference between approaches increases with the inelasticity
level: the yielding plateau is reached much later when using Priestley’s approach, which
corresponds to the longest plastic hinge.

Another relevant parameter, from this concentrated plasticity perspective, is the plastic
hinge location in the element, which, according to the used software package, SAP2000,
can be at any position within the ending portion of the discretized bar. A study to on this
location parameter has been conducted, analogously to the plastic hinge length one. Three
location points have been tested, corresponding to the bottom (ph0), half (ph0.5) and top

4.16
 Nonlinear Modelling

(ph1) of the bottom element of the pier, which is 0.7m long. The plastic hinge length has
been estimated according to the formula proposed by Kappos, which provided intermediate
results. The capacity diagrams, illustrated in Figure 4.9, reveal that the plastic hinges-based
computation assumes the nonlinearity features completely concentrated in the specified
location, using the plastic hinge length to determine the rotation for that specific location.
Indeed, if the plastic hinge constitutive law is introduced in moments-rotations format, the
specification of its length is exempt.

Vb 3300.0
3300
(kN)
3000.0
3000
2700.0
2700
2400.0
2400
2100.0
2100
1800.0
1800
1500.0
1500
1200.0
1200 rot0
ph0
900
900.0
ph0.5
rot0.5
600.0
600
ph1
rot1
300
300.0

0.00 d (m)
0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20

Figure 4.9 – Capacity curves for different plastic hinge locations.

The plastic hinge is “triggered” as soon as the force that is being controlled (usually
bending moment) reaches the first inelastic moment defined in the plastic hinge
constitutive law. From that step on, deformation is controlled by that constitutive law.
Consequently, by shifting up the location of the plastic hinge along the bottom pier
element, supported at the base, according to the triangular moments’ diagram that is
installed, the base shear activating the plastic hinge behaviour is reached later. The closer
to the base the plastic hinge is located, the lower the global force able to trigger it will be.
Such finding justifies the differences found in Figure 4.9, which have been found, apart
from being negligible (variation in maximum base shear less than 5%), to be mainly
related to the maximum attainable base shear capacity rather than the curve shape itself.

4.17
Nonlinear Modelling

4.2.2 Distributed plasticity models (fibres)

A structural model that includes material nonlinearity in a distributed fashion, using finite
fibre elements, is able to characterize in higher detail the reinforced concrete elements and
thought to capture more accurately response effects on such elements. Geometrical and
material properties are the only required ones as input. According to Casarotti and Pinho
(2006) a fibre model manages to represents the propagation of the nonlinear effects over
the cross section of the element as well as along its extension. Consequently, higher
accuracy in the structural damage estimate is attained, even for the case of high inelasticity
levels. Fibre based analysis may have numerical solution using a stiffness or flexibility-
based formulation. Differences between the two possibilities have been studied in
Papaioannou et al. (2005) and the choice for the classic formulation based on the stiffness
matrix developed by Izzuddin (2001) has been made. For this kind of models analysis, the
fibre-based finite elements software package SeismoStruct (SeismoSoft, 2008), which
basically performs 3D finite element modelling, with behaviour prediction for high
displacement levels of structures subjected to static or dynamic loading, has been chosen.
Material nonlinearity and second order effects are taken into account. A stiffness based
cubic formulation is used to represent the development of the inelasticity along the
element, together with axial load and transverse deformation interaction. Numerical
integration makes use of two Gauss points per element, and the reinforced concrete cross
section is discretized in fibres, as shown in Figure 4.10.

One of the main advantages of the use of fibre models is precisely the fact that there is no
need for calibration regarding input variables, which are limited to the material and
geometrical properties of the elements. Indeed, within a uniaxial 2-node finite element
formulation, the variables that might be adjusted would be mainly the number of fibres or
the elements’ subdivision. The ideal number of section fibres, enough to assure an
adequate reproduction of the stress-strain distribution across the element's cross-section,
varies with the shape and material characteristics of the latter, depending also on the
degree of inelasticity to which the element will be forced to. A number of fibres between
100 and 200 is currently used; the latter, or more, is usually adopted for more complicated
sections, subjected to high levels of inelasticity.

4.18
 Nonlinear Modelling

Figure 4.10 – Reinforced concrete element discretization – fibre modelling approach (Casarotti and
Pinho, 2006).

Given the characteristics of the cross section of the piers, which is double-material, and of
the analysis taking the structures to high nonlinearity, 200 fibres have been used.
Regarding the number of elements, the extremities of the piers are the location for potential
plastic hinges, therefore, a possible approach is to adopt a discretization that roughly
follows such phenomenon. The adopted subdivision was, hence, to use five elements
corresponding to 15%, 20%, 30%, 20% and 15% of the structural length of the pier.

4.3 Case study

An extensive parametric study has been conducted over several topics, namely, material
plasticity modelling issues, nonlinear static procedures or probabilistic safety assessment,
using a set of bridges, differing in regularity level, in terms of piers’ heights and deck
length, and abutment types.

4.3.1 Selected bridges

The bridges selected for application were obtained by reformulation of the configurations
used in the former research project PREC8 – Bridge Research Programme (Pinto et al.,
1996; Guedes, 1997), a program launched to cover topics of the European standard design
code EC8 that needed to be clarified (in terms of structural regularity, evaluation of
behaviour factors improvement of methods of analysis and of capacity design procedures).

4.19
Nonlinear Modelling

The referred venture worked over two basis configurations, one regular and the other
irregular, designated Bridge 232 and Bridge 213, respectively. The label numbers 1, 2 and
3 stand for pier heights of 7, 14 and 21 meters, respectively, as represented in
200m (4x50m)

7.0m
7.0m
Short Pier (1)
7.0m
Median Pier (2)
Tall Pier (3)

0.4m
14.0m
3.0m
0.3m
DECK 0.3m
2.0m 3.2m PIER

6.5m

0.4m

0.4m 1.2m 0.4m

Figure 4.11, which illustrates the initially defined configuration 213, together with the
structural elements that are common to the rest of the bridges. Different versions based on
such geometrical configurations were considered by altering reinforcement features of the
piers as well as existence/distribution of isolating/dissipating devices.

The bridges have been designed for a peak ground acceleration of 0.35g, according to
(Calvi, 1994; Calvi and Pinto, 1995), in average soil conditions, following EC8 provisions.
A minimum reinforcement steel ratio equal to 0.5%, half of the 1% prescribed by the code,
has been adopted for the piers, in order to avoid highly forces to be attracted to the piers.
The deck has been assumed as elastic behaving for all the bridges.

4.20
 Nonlinear Modelling

200m (4x50m)

7.0m
7.0m
Short Pier (1)
7.0m
Median Pier (2)
Tall Pier (3)

0.4m
14.0m
3.0m
0.3m
DECK 0.3m
2.0m 3.2m PIER

6.5m

0.4m

0.4m 1.2m 0.4m

Figure 4.11 – Geometrical configuration of Bridge 213 and of the different structural elements,
constituting the studied bridges.

Using the aforementioned bridge 232 and 213 configurations, Casarotti et al. (2005),
extended the set to include deck lengths of 200 and 400 meters, with different
configurations as well, analysing the additional 123, 222, 2222222, 2331312 and 3332111
configurations. The parametric studies carried out within the work herein described were
based on the resulting final set of seven different bridge configurations.

Two bridge lengths have hence been considered (viaducts with four and eight 50m spans),
with regular, irregular and semi-regular layout of the piers’ height and with two types of
abutments; (i) continuous deck-abutment connections supported on piles, with bilinear
behaviour (type A bridges), and (ii) deck extremities supported on linear pot bearings
(type B bridges). The total number of bridges is thus fourteen, as implied by Figure 4.12.

4.21
Nonlinear Modelling

Label 222
Label 2222222
REGULAR
Label 232

Label 123 Label 3332111

SEMI
REGULAR /
IRREGULAR Label 213 Label 2331312

Figure 4.12 – Considered bridge configurations.

The fundamental period of vibration (see Table 4.2) ranges approximately from 0.3 to 0.5
seconds in short configurations and from 0.6 to 0.8 seconds in long ones.

Table 4.2 – Fundamental transverse periods of vibration (seconds)

Configuration 123 213 222 232 2222222 2331312 3332111

Type A abutments 0.46 0.40 0.41 0.51 0.59 0.65 0.71
Type B abutments 0.40 0.32 0.42 0.53 0.60 0.67 0.77

4.3.1.1 Piers

The piers are made up of hollow reinforced concrete cross sections, as illustrated in Figure
4.13, in which the reinforcement details are included as well. The concrete and steel
constitutive laws are considered as described in Section 4.1 and the corresponding defining
parameters are presented in Table 4.3 and Table 4.4.

4.22
 Nonlinear Modelling

Asw(k=1.2) [concrete cover = 20mm]

14Ø20 36Ø20

0.4m

1.2m

0.4m

0.4m 3.2m 0.4m

Figure 4.13 – Piers cross section

Table 4.3 – Concrete properties.

Parameter
Young’s Modulus (GPa) 30.5
Compressive stress (MPa) 42
Tension stress (MPa) residual
Strain at the peak compressive stress 0.002
Confinement factor 1.2
Specific weight (kN/m3) 24

Table 4.4 – Steel properties.

Parameter
Young’s Modulus (GPa) 200
Yielding stress (MPa) 500
Strain hardening parameter 0.005
Transition curve initial shape parameter 20
Transition curve shape calibrating coefficients (a1, a2) 18.5, 0.15
Isotropic hardening calibrating coefficients (a3, a4) 0, 1
Specific weight (kN/m3) 77

The modelling of the piers has featured particular attention to its bottom and top regions,
next to the ground and the deck, where the nonlinear behaviour will be concentrated.
According to Priestley et al. (1996), depending on the footing/connection conditions, the
plastic hinge regions should occur within an extent from 1/10 to 1/20 of the element’s
length. On the other hand, the length of the elements should not be very small, when
compared to the cross section depth, which is four meters. Bearing such considerations in

4.23
Nonlinear Modelling

mind, the piers have been divided in seven elements of length corresponding to 15%, 20%,
30%, 20% and 15% of the height, featuring, hence, higher refinement degree at the edges.

4.3.1.2 Deck

The deck has been assumed as elastic behaving, which is commonly accepted within
design codes, due to its considerable flexibility or even to the fact that it is typically
prestressed, not allowing plastic deformations. Furthermore, many times, isolating devices
are employed with the purpose of protecting the deck from damaging movement of the
soil. For the particular case of the bridges herein considered, the deck has been modelled
by means of an elastic beam 3D element accounting, nevertheless, for second order
geometrical nonlinear phenomena. The geometry of the deck is illustrated in Figure 4.14
and the corresponding mechanic characteristics are listed in Table 4.5, analogously to the
1:1 scale tested model at ISPRA (Guedes, 1997). The assumed Young’s and shear moduli
are 25GPa and 10GPa (ν=0.25). Each span of the deck is divided in eight elements, with
higher refinement in the regions over the piers and next to the abutments, only, given that
elastic elements will certainly not require large discretizing amount. Moreover, the bar
element representing the deck is shifted from the top of the piers of 1.508 meters, a height
corresponding to its cross section centre of mass (see Figure 4.14).

14.0m
3.0m
0.3m
0.3m G
2.0m
1.508m
0.25m

6.5m
Figure 4.14 – Deck geometry

Table 4.5 – Deck properties.

Parameter
EA (kN) 1.74 x108
EI2 (kN.m2) 1.34 x108
EI3 (kN.m2) 2.21 x109
GJ (kN.m2) 1.17 x108
Mass (ton/m) 17.4

4.24
 Nonlinear Modelling

The connection between piers and deck is assured by rigid link elements, of residual mass,
which guarantee that only shear and axial forces are passed to the piers, so as to simulate
the shear keys supporting the deck, as schematically represented in Figure 4.15, taken from
SeismoStruct model.

deck
rigid link

Figure 4.15 – Pier-deck connection through rigid link elements

It is worth noticing that the employed elastic element deck modelling does not take into
account shear contribution to the deformation.

4.3.1.3 Abutments

The abutments are essentially modelled through springs that intend to represent whether
the supporting on pot bearings or piles, whereas the piers are fully restrained at the bottom.
Bridge abutments design is usually carried out for serviceability limit states and afterwards
verified for the seismic demand. The use of elastic spring elements to model this sort of
structural element is common when analysing bridges, given the easiness in incorporating
the dynamic behaviour of the ground adjacent to the abutment, the different structural
features of the abutment and the soil-structure interaction (Casarotti et al., 2005). Further
detail and approaches on the modelling of abutments can be found in studies from Goel
and Chopra (1997), Megally et al. (2003), Bozorgzadeh et al. (2006) amongst others.

Adopting the modelling in the work carried out by Casarotti et al (2005), based on iterative
procedures or simplified approaches, the definition of the abutments consisted of
quantifying the stiffness of the different springs. For the selected configurations, two sorts
of abutment have been defined for testing: continuous deck-abutment connections
supported on piles (type A), with assumed bilinear behaviour, and deck extremities simply

4.25
Nonlinear Modelling

supported on pot bearing (type B), which feature linear response. Both cases are modelled
by means of four springs in parallel, which represent either the soil or the bearings,
connected through rigid arms, as shown in Figure 4.16.

abutment
abutment spring
abutment spring

abutment spring
spring

deck
rigid arms

Figure 4.16 – Abutments modelling scheme.

4.3.2 Seismic action

The employed set of seismic excitations is defined by an ensemble of ten records selected
from a suite of historical earthquakes (SAC, 1997) scaled to match the 10% exceeding
probability in 50 years (475 years return period) uniform hazard spectrum for Los Angeles,
which corresponds, in the current endeavour, to the intensity level 1.0. Figure 4.17
illustrates pseudo-acceleration and displacement spectra for the selected earthquake
records, with damping ratio of 5%, together with the median spectrum in thicker line.

Additional intensity levels, linearly proportional to the latter by a factor of 0.5, 0.75, 1.5,
2.0 and 2.5, have been also considered, with the purpose of allowing an overview on how
results evolve with increasing seismic intensity.

4.26
 Nonlinear Modelling

4.0 120

Spectral Displacement (cm)

Pseudo-Acceleration (g)
3.0 90

2.0 60

1.0 30

0.0 0
0.0 0.7 1.4 2.1 2.8 3.5 0.0 0.7 1.4 2.1 2.8 3.5
Natural Period (s) Natural Period (s)

Figure 4.17 – Elastic pseudo-acceleration and displacement spectra.

The ground motions were obtained from California earthquakes with a magnitude range of
6-7.3 recorded on firm ground at distances of 13-30 km; their significant duration
(Bommer and Martínez-Pereira, 1999) ranges from 5 to 25 seconds, whilst the PGA (for
intensity 1) varies from 0.23 to 0.99g, which effectively implies a minimum of 0.11g
(when intensity level is 0.5) and a maximum of 2.5g (when intensity level is 2.5). The
demand spectrum was defined as the median response spectrum of the ten records.

4.4 Comparative study

In the previous sections, both the modelling possibilities have been discussed and duly
calibrated, regarding the typically relevant parameters of each approach. The next step is to
look into the differences taking place when a nonlinear analysis is carried out, which,
ideally, would be insignificant or at least very low, regardless the type of model being used
for the analysis. In order to put away possible sources of divergence coming from the use
of dynamic nonlinear analysis, considered the most accurate analysis method, yet
definitely more complex, the employment of nonlinear static pushover analysis has been
decided. Furthermore, pushover-based procedures have been gaining ground as valid
methods for response estimation, when compared to dynamic analysis, mostly for
buildings, in the past few years. In addition, their suitability to the case of bridges has been
thoroughly investigated and confirmed in this work (Chapter 5). The comparative study
has focused the case study configurations described in the previous section. For the sake of
simplicity, and keeping in mind that material plasticity will be developed essentially at the
piers, only the first type of abutments, type A, has been considered. The comparison is,

4.27
Nonlinear Modelling

hence, based on a total of seven different configurations. A preliminary eigenvalue study

on the vibration modes of the different configurations has been carried out, with a view to
calibrate both analysis software programs models, in terms of stiffness and mass, assuring
the structural elastic validity of the subsequent comparisons.

The first comparison of material nonlinearity approaches that has been carried out was in
terms of the moment-curvature law obtained within the use of both of the computation
versions: the BIAX obtained one, used for the Plastic Hinge Model (PHM) in SAP2000,
and the one ‘internally’ considered by SeismoStruct, within a Fibre Model (FM). Such
initial comparison is due to the fact that for the capacity curve computation, the main
output of a pushover analysis, the need for an accurate consideration of the behaviour of
the piers’ cross sections is obvious, given that the piers will be the bridge elements with
nonlinear behaviour.

Two different cross sections, hollow and solid, submitted to three different axial loading
forces, have been tested and the moment-curvature behaviour given by both of the
approaches (plastic hinge or fibre based) have been compared. Pushover analyses were
subsequently carried out for the seven different considered bridge configurations,
presented in Section 4.3, followed by direct graphical comparison of pushover curves and
statistical parametric comparison of distinct response quantities: deck displacements, deck
bending moments, pier shear forces and abutment shear forces.

4.4.1 Elastic modal analysis comparison

The first three eigenmodes for the seven bridge configurations, computed for both
structural modelling versions, fibre based SeismoStruct (FM) and plastic hinges based
SAP2000 (PHM), are presented from Table 4.6 to Table 4.12.

4.28
 Nonlinear Modelling

Table 4.6 – Vibration modes for configuration 123 – Fibre vs. Plastic Hinge Modelling.

Period (s)
Bridge 123 transversal vibration modes Plastic Hinge
Fibre Model
Model

1st 0.4596 0.4597

2nd 0.2989 0.3006

3rd 0.2441 0.2450

Table 4.7 – Vibration modes for configuration 213 – Fibre vs. Plastic Hinge Modelling.

Period (s)
Bridge 213 transversal vibration modes Plastic Hinge
Fibre Model
Model

1st 0.4004 0.4012

2nd 0.3373 0.3377

3rd 0.2302 0.2321

4.29
Nonlinear Modelling

Table 4.8 – Vibration modes for configuration 222 – Fibre vs. Plastic Hinge Modelling.

Period (s)
Bridge 222 transversal vibration modes Plastic Hinge
Fibre Model
Model

1st 0.4147 0.4149

2nd 0.3312 0.3319

3rd 0.2729 0.2748

Table 4.9 – Vibration modes for configuration 232 – Fibre Model vs. Plastic Hinge Model.

Period (s)
Bridge 232 transversal vibration modes Plastic Hinge
Fibre Model
Model

1st 0.5053 0.5060

2nd 0.3312 0.3319

3rd 0.2842 0.2858

4.30
 Nonlinear Modelling

Table 4.10 – Vibration modes for configuration 2222222 – Fibre vs. Plastic Hinge Modelling.

Period (s)
Bridge 2222222 transversal vibration modes Plastic Hinge
Fibre Model
Model

1st 0.5886 0.5884

2nd 0.4254 0.4258

3rd 0.3699 0.3707

Table 4.11 – Vibration modes for configuration 2331312 – Fibre vs. Plastic Hinge Modelling.

Period (s)
Bridge 2331312 transversal vibration modes Plastic Hinge
Fibre Model
Model

1st 0.6522 0.6527

2nd 0.4741 0.4751

3rd 0.3604 0.3619

4.31
Nonlinear Modelling

Table 4.12 – Vibration modes for configuration 3332111 – Fibre vs. Plastic Hinge Modelling.

Period (s)
Bridge 3332111 transversal vibration modes Plastic
Fibre Model
Hinge Model

1st 0.7069 0.7071

2nd 0.4776 0.4787

3rd 0.3700 0.3712

The observation of the elastic modal analysis plots indicates a clear agreement between the
two models for each and every bridge configuration, given that whether the mode shape
(comparison omitted for the sake of simplicity) and the eigenvalues match quite perfectly.
The slight differences may be partially, at least, associated to the fact that the mass matrix
considered by SAP2000, as opposed to SeismoStruct, is diagonal and does not take into
consideration the mass contribution from the rotational degrees of freedom.

4.4.2 Moment-curvature relationships

Two reinforced concrete cross sections have been considered defining a 14 meters high
pier: solid rectangular, 0.5mx1.0m, and hollow rectangular, the same used for the plastic
hinge calibration (see Section 4.2.1.2), corresponding to the cross section of the piers of
case study bridges. Figure 4.18 illustrates the two tested elements. In spite of the fact that
the case study bridges feature hollow sections only, the presence of several solid section
piers in the existent bridges portfolio motivated the its inclusion in this calibration exercise.

4.32
 Nonlinear Modelling

Figure 4.18 – Solid (left) and hollow (right) reinforced concrete cross section piers.

Three axial loading levels have been applied, according to Table 4.13, in order to observe
the matching of approaches for different bearing conditions.

Table 4.13 – Tested axial load forces (N) and axial load ratio (υ) for each cross section.

Solid cross section Hollow cross section

Designation N (kN) υ N (kN) υ
N1 0 0 0 0
N2 500 0.024 3000 0.017
N3 1500 0.071 9000 0.052

Figure 4.19 and Figure 4.20 plot the moment curvature laws obtained for use within plastic
hinge or fibre models, using BIAX and SeismoStruct, respectively, for a solid and a hollow
cross section bridge pier. Whereas the BIAX computer module works with the cross
section characteristics as the only input, moment-curvature behaviour, using SeismoStruct,
has been determined through the application of an increasing loading force at the top of the
pier, recording data of interest at the closest to the base Gauss point.

4.33
Nonlinear Modelling
Solid cross section
2000

1800

1600

1400
Moment (kN.m)

1200

1000

800
Biax N1
600 SStruct N1
Biax N2
400
SStruct N2
200 Biax N3
SStruct N3
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Curvature

Figure 4.19 – Moment curvature constitutive laws, obtained for the solid cross section, with
different axial loading.

4
x 10

4
Moment (kN.m)

2
Biax N1
SStruct N1
Biax N2
1 SStruct N2
Biax N3
SStruct N3
0
0 0.002 0.004 0.006 0.008 0.01 0.012 0.014 0.016 0.018 0.02
Curvature

Figure 4.20 – Moment curvature constitutive laws, obtained for the solid cross section, with
different axial loading.

A first observation to make from the two sorts of cross sections is that both computation
approaches have given little relevance to tensile strength of the concrete, given that
constitutive laws are nearly bilinear when the axial load is inexistent. The first branch of
the trilinear behaviour for the rest of the situations will correspond, therefore, to the
‘elimination’ of the compressive axial load.

4.34
 Nonlinear Modelling

For the case of the simple, solid cross section, moment-curvature relationships are
extremely similar, with no differences noteworthy, besides the softer yielding showed by
SeismoStruct. With respect to the hollow cross section, for low axial load, there is again no
worth mentioning divergence, whilst, for a high axial loading scenario, there is some
difference until the yielding plateau is reached, with higher moments being recorded in the
BIAX analysis, and a higher stiffness initial branch. In any case, global similarity is
evident, which indicates how both computation tools are in agreement regarding cross
section constitutive laws.

4.4.3 Pushover capacity curves

The nonlinear static analysis of each of the bridges considered within this study has been
carried herein using two different deck loading patterns, uniform or 1st mode proportional,
as recommended by EC8. Regarding the node of control, two alternatives have been
admitted as well, for the case of the modal loading pattern: the centre of mass of the deck
and the maximum modal displacement node. The uniform loading pattern has been tested
with the centre of mass of the deck as reference node. Further detail on pushover-based
nonlinear static procedures can be found in Chapter 5. Within both the model analysis, the
control node of short bridges has been pushed until 0.75m, along 750 load steps, whereas,
for long bridges, the target displacement of the control node was 1.5m, divided in 1500
load increments.

The crude observation of the plots that follow, regardless the sort of bridge configuration,
bridge length, load shape or reference node, leads to the conclusion that there is a generally
good agreement between the capacity curves for both of the material plasticity models,
neither regarding the shape of the curve, nor the maximum base shear.

For short, irregular bridge configurations, Figure 4.21, the correspondence between the
Plastic Hinge Model (PHM) and Fibre Model (FM) based pushover curves, in the pre-
yield and post-yield regions, is notorious and independent from the loading pattern or the
reference node. A slightly higher difference in the transition of such two zones may be
found for the use of the first mode proportional load vector, which is, nevertheless,
probably a reflex of the same type of difference encountered in the moment-curvature
constitutive laws (Section 4.4.2).

4.35
Nonlinear Modelling

4 4
x 10 x 10

2.5

Base Shear (kN)

Base Shear (kN)

1.5
1.5

1
1

0.5 PHM modal central

0.5
PHM modal max
123 FM modal central PHM uniform central
123
FM modal max FM uniform central
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Displacement (m) Displacement (m)

4 4
x 10 x 10

2.5
2

2
1.5
Base Shear (kN)

Base Shear (kN)

1.5

1
1

0.5 PHM modal central

PHM modal max 0.5

213 FM modal central

213 PHM uniform central
FM modal max FM uniform central
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Displacement (m) Displacement (m)

Figure 4.21 – Capacity curves for short, irregular configurations 123 and 213.

4 4
x 10 x 10
2.5

2
2

1.5
Base Shear (kN)

Base Shear (kN)

1.5

1
1

0.5
PHM modal 0.5 PHM modal
PHM uniform PHM uniform
222 FM modal 232 FM modal
FM uniform FM uniform
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Displacement (m) Displacement (m)

Figure 4.22 – Capacity curves for short, regular configurations 222 and 232.

4.36
 Nonlinear Modelling

With respect to short, regular configurations, the main trend is fairly similar to the irregular
ones, as documented in Figure 4.22. The difference between the two approaches within the
first branch (pre-yield) is however slightly more pronounced.

Regarding long configurations, though, the general picture changes a little, according to
the plots in Figure 4.23. For irregular long bridges the difference between the curves, prior
to the yielding, tends to diminish, whereas the post-yield slope can differ to a larger extent,
in contrast to what happened for short configurations.

4 4
x 10 x 10
3.5
4.5

3 4

3.5
2.5
3
Base Shear (kN)

Base Shear (kN)

2
2.5

1.5 2

1.5
1
PHM modal central 1
0.5 PHM modal max
FM modal central 0.5 PHM uniform central
2331312 FM modal max 2331312 FM uniform central
0 0
0 0.5 1 1.5 0 0.5 1 1.5
Displacement (m) Displacement (m)

4 4
x 10 x 10

3 4

3.5
2.5

3
Base Shear (kN)
Base Shear (kN)

2
2.5

1.5 2

1.5
1

PHM modal central 1

0.5 PHM modal max

FM modal central 0.5 PHM uniform central
3332111 FM modal max 3332111 FM uniform central
0 0
0 0.5 1 1.5 0 0.5 1 1.5
Displacement (m) Displacement (m)

Figure 4.23 – Capacity curves for long, irregular configurations 2331312 and 3332111.

Finally, for the tested long regular configuration, Figure 4.24, the differences are equally
distributed along the entire domain of the capacity curves, with no relevant distinction
between the pre- and post-yield branches.

4.37
Nonlinear Modelling

4
x 10
4

3.5

Base Shear (kN)

2.5

1.5

1
PHM modal
PHM uniform
0.5 FM modal
2222222 FM uniform
0
0 0.5 1 1.5
Displacement (m)

Figure 4.24 – Capacity curves for long, regular configuration ‘2222222’.

In addition, it is worth to mention that plastic hinges models yielded consistently higher
shear capacity than the fibre models, for each end every studied bridge configuration, load
pattern or reference node. Such scenario was not always verified for the moment-curvature
comparison carried out for the two analysis computer programs, which indicates that,
indeed, the way of accounting for material plasticity plays an important part.

4.4.4 Structural response parameters

In addition to the comparison of the evaluation of the modelling possibilities when

assessing the structural capacity, herein carried out by means of the pushover curve, the
agreement level of the structural response, individuated in different parameters, for the
different approaches has been looked at as well. In correspondence with the
pushover-based capacity evaluation, the structural response comparison has made use of a
nonlinear static procedure, N2, prescribed in EC8 (CEN, 2005a, 2005b). Response data
corresponding to four parameters, measured in seven bridge configurations, across six
intensity levels, have been statistically processed. The four response parameters have been
deck displacements, deck bending moments, pier shear forces and abutment shear forces,
measured at each relevant location, which are, logically, the pier locations for the first
three parameters and the abutments for the latter.

A statistical parameter, Bridge Index (BI), has then been defined, corresponding to the
median of the ratio of Plastic Hinge Model (PHM) and Fibre Model (FM) responses,
across the relevant locations, as in Equation (4.15), written for a general response

4.38
 Nonlinear Modelling

parameter, ∆. The ratio is computed for the piers location, with the exception for the
abutment shear forces, which is measured at the two obvious locations. An ideal agreement
between the two sorts of plasticity modelling would, therefore, correspond to a unitary
Bridge Index.

∆ 
BI = median  PHM ,i 
 (4.15)
 ∆ FM ,i  i =1:npiers

Furthermore, if the median BI is computed across the different configurations, for each
intensity level, one will obtain plots with PHM–FM matching evolution trends as function
of the intensity level, as presented in Figure 4.25. On the other hand, if the median of the
Bridge Indexes is computed across the intensity levels, for each bridge configuration, the
evolutions of the results according to the sort of bridge will be obtained, illustrated in
Figure 4.26. For the sake of simplicity, only global median results will be presented,
instead of the detailed bridge-by-bridge median BI across different intensity levels.

The observation of results as a function of intensity level indicates, first of all, that,
generally, in median terms, the agreement between the two plasticity models’ predictions is
rather good. Deck displacements and bending moments’ predictions tend, however, to be
less steady than shear estimates, along the intensity increase and, for such situation, the use
of modal loading shape leads to better agreement between the modelling possibilities,
rather than uniform pattern. Displacements estimates are larger for the case of fibre
models, given that the ratios are mostly under unity, which does not happen, at least not so
evidently, for the rest of the parameters. Finally, it is noteworthy the fact that the modelling
approaches seem to get closer for higher inelasticity levels, a more visible scenario for the
case of shear forces and deck displacements.

In terms of bridge configuration, the results do not differ much. Deck bending moments are
particularly unstable for short, regular configurations, something that has already happened
for low intensity levels, although not to such extent in that representation mode. The
advantage in going for a 1st mode proportional loading pattern is not so explicit and
relative positions between the two loading shapes do tend to invert, along the different

4.39
Nonlinear Modelling

configurations. Again, agreement is definitely higher within shear predictions. No

significant correlation between similarity of nonlinear approaches and bridge configuration
has been detected.

Deck Displacements Pier Shear Forces

2 2
modal central modal central
1.8 modal max 1.8 modal max
uniform uniform
1.6 1.6

1.4 1.4

1.2 1.2
Median BI

Median BI
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Deck Moments Abutment Shear Forces

2 2
modal central modal central
1.8 modal max 1.8 modal max
uniform uniform
1.6 1.6

1.4 1.4

1.2 1.2
Median BI

Median BI

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 4.25 – Median Bridge Index (BI) for different response parameters, according to intensity
level.

A structural response comparison in terms of median indexes is interesting to assess the

matching extent of plastic hinges and fibre models predictions, across the entire bridges
(all locations) and intensity levels. In any case, from a design or safety assessment
perspective, the quantities of main interest will eventually be the maximum deformations
and/or forces. Based on such hypothesis, it would probably be of interest, or even more
accurate, within a safety assessment context, the analysis of the coherence of different
modelling strategies in terms of maximum demand. In order to do so, plots have been

4.40
 Nonlinear Modelling

redrawn, considering that, for each bridge configuration, the Bridge Index is not defined as
the median of the BIs at the piers locations, Equation (4.15), but as the ratio of the
maximum demand value for the structural parameter at stake, when using the plastic hinges
model and the corresponding value, when using the fibre model.

Deck Displacements Pier Shear Forces

2 2
modal central modal central
1.8 modal max 1.8 modal max
uniform uniform
1.6 1.6

1.4 1.4

1.2 1.2
Median BI

Median BI
1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
123 213 222 232 2222222 2331312 3332111 123 213 222 232 2222222 2331312 3332111
Bridge Configuration Bridge Configuration

Deck Moments Abutment Shear Forces

2 2
modal central modal central
1.8 modal max 1.8 modal max
uniform uniform
1.6 1.6

1.4 1.4

1.2 1.2
Median BI
Median BI

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
123 213 222 232 2222222 2331312 3332111 123 213 222 232 2222222 2331312 3332111
Bridge Configuration Bridge Configuration

Figure 4.26 – Median Bridge Index (BI) for different response parameters, according to bridge
configuration.

This alternative way of looking into results has, nevertheless, the limitation of eventually
being comparing maximum demand quantities coming from different locations, if the
response pattern differs greatly from lumped to distributed plasticity models. Results have
proven to be extremely similar to the ones obtained with the median indexes and are,
hence, not presented. Such fact contributes to the reassurance of the similarity of both
modelling techniques.

4.41
Nonlinear Modelling

4.5 Conclusions

The work that has just been presented had as main target the focus on the main issues
within the nonlinear modelling of reinforced concrete bridges in a seismic safety
assessment endeavour context. Structural modelling has been carried out by means of 3D
frame elements, using two available structural analysis software tools, able to include
geometrical and material nonlinear features.

Typically seen as less relevant when it comes to bridge structural systems, geometrical
nonlinearity requires, moreover, less calibration and admits less variation in the mode of
being accounted for. Material nonlinearity, on the other side, assumes greater importance,
representing one of the major points of energy dissipation during seismic response and
therefore has been looked into with higher care. Material models, for concrete and steel,
have been initially described, pointing out some of the several available alternatives and
detailing the ones that have been considered for the study. Furthermore, material
nonlinearity is typically modelled in a lumped way, through plastic hinges, or in a
distributed way, using fibre models. Recognizing this as a critical issue in the present
seismic analysis, a comparative study has been carried out with the purpose of identifying
the eventual convergence between the two possibilities. Comparisons have been based on
simple pushover-based nonlinear static analysis, a less complex structural analysis
methodology than dynamic analysis, for instance, which, hopefully, induced less
dispersion factors affecting the modelling comparison itself.

For the use of lumped plasticity models, a first calibration study has been carried out on the
characterization of the plastic hinge constitutive law. Different parameters have been
tested, such as length or location of the plastic hinge within the element, according to
different formulations, within a single pier pushover analysis. Three plastic hinge lengths
and locations have been tested and the results have proven that the influence of such
parameters is of limited significance, mainly on the maximum capacity and yielding
deformation. Major differences between the considered approaches for the plastic hinge
length occur in the elastic-cracked behaviour region only. It is therefore expected that no
considerable divergence occurs when the structure is under large deformation conditions.
Regarding plastic hinge location in the element, a specific feature of some computer

4.42
 Nonlinear Modelling

program algorithms, such as the used SAP2000, it has been seen that its relevance is
negligible.

The comparison between lumped and distributed plasticity models itself has been carried
out for a relatively large set of bridge configurations, regular, semi-regular and irregular,
for a wide range of intensity levels, in three different components. Firstly, the moment-
curvature constitutive laws used by both sorts of models have been compared for different
axial loading levels found in the tested bridge configurations. Subsequently, the structural
response, using both types of modelling, obtained through the employment of a nonlinear
static procedure, was compared, in terms of pushover capacity curves and response
parameters, namely deck displacements and bending moments and pier and abutment shear
forces. Main observations on such three aspects are as follows:

The moment-curvature constitutive laws, the essential input source for the
definition of the nonlinear behaviour of structural elements, have been found to be
rather similar, whether the obtained independently for plastic hinge characterization
or the ones intrinsically considered by the fibre model analysis software. Slight
differences have been found in the elastic pre- and post-cracking phases, which is
probably due to the formulation assumed within the used tools. For plastic hinges
definition, the employed BIAX algorithm makes use of equilibrium equations to
determine internal forces, whereas SeismoStruct works under interpolation between
Gauss points. The axial loading effect proved to be extremely important for the
maximum available bending moment, particularly when dealing with hollow cross
sections, a typical situation for bridges.

The capacity curve, as the main output result of a pushover analysis, has been
obtained for each of the modelling versions, according to different loading patterns
and reference nodes: 1st mode proportional load shape, tested with the centre of
mass of the deck or the maximum modal displacement as reference nodes, and
uniform load shape, carried out with the centre of mass of the deck as reference
node only. The pushover curves have turned out quite in agreement, regardless the
type of loading pattern, reference node or bridge configuration. Slight found
differences were probably due to the corresponding divergences previously
encountered in the reinforced concrete cross section constitutive laws.

4.43
Nonlinear Modelling

Finally, it has been studied the influence of the differences in the pushover curves
with different load shapes and reference nodes, representing the structural capacity,
in the structural demand, through four response parameters statistical comparison,
based on the definition of a median index parameter – the Bridge Index. The EC8-
recommended nonlinear static procedure has been used in order to obtain the
performance point at each of the six intensity levels, for each studied configuration.
The global observation of results points to the conclusion that, generally, both
lumped and distributed material plasticity models match in every of the analysed
parameters: displacements, bending moments and shear forces, at deck, piers and
abutments. The median Bridge Indexes were recurrently quite close to unity, across
the different intensity and regularity levels. Despite such similarity, it may be stated
that plastic hinge models tend to lead to higher shear force predictions, when
compared to fibre models, whereas the opposite scenario is verified for deck
displacements. Deck bending moments have presented a less constant trend.
Estimates obtained by means of pushover analysis with 1st mode proportional
loading pattern have been found generally more agreeing. For the majority of the
configurations and intensity levels, the maximum modal displacement node used as
reference led as well to more consistent estimates between the two approaches.
Agreement between the models has been found at its best when assessing shear
forces, even and mostly for high inelasticity levels. Disparity found for low/median
intensity levels may have to do with the employed nonlinear static procedure,
which carries out a bilinearization of the capacity curve, using equal areas criteria.
Such bilinear curve simplification can amplify dissimilarities in the pushover
curves of the different models for low to median seismic demand, where the
behaviour is deemed to be linear elastic. Under such conditions, displacements BIs
are above unity and shear forces BIs present the opposite trend, therefore,
overestimated by plastic hinge models, which indeed present higher initial elastic
stiffness, when compared to fibre models.

The adopted modelling technique and the included nonlinear features constitute a
particularly important step, given that it will largely influence capacity and structural
demand, the two fundamental variables within a seismic safety assessment procedure.
From the material models, under monotonic and cyclic loading, to the way of considering

4.44
 Nonlinear Modelling

plasticity characteristics of the structural members, different versions may be regarded as

eligible. Amongst the possible alternatives, with respect to the various issues that are
involved in the modelling task, special focus was given to the pertinent option between
concentrated or spread plasticity, given that, to the author’s opinion, the other aspects are
less controversial, at least for the time being. Is has been interestingly verified that the
option for a more refined fibre model over a more ‘simplified’, in the assumptions sense,
plastic hinges model does not necessary lead to more accurate evaluation of the seismic
behaviour of bridge structures. Such conclusions must be handled with relative care, given
that simplified nonlinear static analysis has been carried out, with no cyclic loading, such
as the one introduced by nonlinear dynamic analysis. Further studies, using more complete
analysis procedures, would provide a valuable contribute to this matter.

Accordingly, in the future, and in the rest of this work as well, the option for one or other
sort of model may be more influenced by other parameters such as computational effort or
model complexity, without expected limiting loss of accuracy. In particular, in Chapter 6,
the verified agreement between nonlinearity approaches will be used to carry out numerous
nonlinear dynamic analyses, within a simulation-based probabilistic context.

Firstly, though, in the following chapter, structural response prediction tools, the next step
when seismic input and structural model are characterized, will be discussed, from a bridge
application perspective.

4.45
5. Seismic Demand

Estimating the seismic effects in a structure subjected to earthquake loading is definitely

one of the key issues within the global structural safety assessment problem. Even if
naturally depending on the structure’s characteristics and the seismic action itself the main
differences, or not, arise from the nature of the approach, either static or dynamic, each
comprising different concepts, techniques and procedures. The distinction here will be
established between nonlinear dynamic analysis and nonlinear static analysis.

The main advantage in using nonlinear dynamic analysis is naturally the fact that the real
phenomenon is directly reproduced when the acceleration is applied to the ground
connections leading, therefore, to more accurate results. Nevertheless, several drawbacks
may be pointed out when considering this sort of analysis, starting with the seismic action
being particularly defined by an accelerogram. The ground motion records, the structural
modelling (including the reproduction of damping phenomena, the post-elastic behaviour
and corresponding energy dissipation, during loading and unloading periods) are all
important issues within nonlinear dynamic analysis and have been extensively debated in
Chapters 3 and 4. In addition, attention must also be paid to time/computational demand. It
is well known that dynamic analysis may require considerable amount of time to be carried
out, this being due to the analysis method itself, to a complex structural model, which will
typically not occur when assessing bridges, or to the long duration of the input ground
motion record.

5.1
Seismic Demand

The most commonly employed alternative to nonlinear dynamic analysis is the nonlinear
static analysis. Typically, this kind of approach is based in a pushover analysis carried out
on the structure, with the main goal being to somewhat envelope all, or at least a
considerable part, of the possible dynamic analysis results at each intensity level. This sort
of outcome is accomplished through the application of an increasing lateral force or
displacement vector to the structure. An example of a pushover curve, representing base
shear versus displacement of a reference node, is illustrated in Figure 5.1. The main
drawback to be found here is the assumed level of simplification, given that one expects
that the structural behaviour obtained from horizontal loading is able to replace the one
coming from the dynamic analysis.

This chapter looks into the way of estimating the seismic demand on bridges subjected to
such horizontal action. The main focus is the validation of static nonlinear simplified
procedures, which are directly compared to dynamic analysis. Several possibilities are
taken into account and a thorough parametric study is carried out on such validation.

Max. 120

Base Shear
(kN) 100

80

60

40

20

0 drift (%)
0 0.5 1 1.5 2 2.5 3

Figure 5.1 – Nonlinear static analysis: concept (Pinho, 2011).

5.1 Nonlinear Static Analysis

For the last few years, a considerable effort has been put in research of simplified, but
credible, methodologies to assess the seismic behaviour of structures. The so-called
Nonlinear Static Procedures (NSP) have therefore been developed and recognized for
application in some guidelines, such as the ATC-40 (ATC, 1996), FEMA-273 (ATC, 1997)
or the European Code (CEN, 2005a, 2005b). This type of procedure is extensively spread
within the earthquake engineering community, mainly due to its simplicity and potentially

5.2
 Seismic Demand

easy application when assessing a large number of structures. Although these methods
were initially thought having the application to buildings in mind, their extension to bridge
structures is full of interest and there is actually no reason for not doing it. Indeed, bridges
have the advantage of being, in general, structurally simple when compared to buildings.
Some very recent endeavours have dealt with the application to bridges of this kind of
procedures (e.g., (Isakovic and Fischinger, 2006; Paraskeva et al., 2006; Casarotti and
Pinho, 2007; Lupoi et al., 2007)). However, those studies were carried out for a specific
NSP, which was not scrutinised in several possible application ways, a gap that this chapter
tries to fill.

The aforementioned methods are based on the computation of the pushover curve, a
representation of the nonlinear force-deformation behaviour of the structure. With the use
of appropriate transformation relationships, the capacity curve of the single-degree-of-
freedom (SDOF) system equivalent to the original multi-degree-of-freedom (MDOF)
structure may be obtained. At this point the main differences between the proposed
methods arise, such as the contemplation or not and choice for a reference node, the way of
reducing the demand spectrum so as to account for the hysteretic energy dissipation, the
consideration of higher modes of vibration, on the way to determine the displacement
demand for a given ground motion.

There are quite a few possible procedures to carry out a nonlinear static analysis,
essentially depending on the sophistication level and time of appearance. From truly
simplified methods, based on correcting coefficients, to others, more complex, including
nonlinear and modal effects in a refined way, the choices are plenty.

Corresponding to the earliest attempts to propose a static method that would quickly lead
to accurate results, a first group of NSPs may be considered, including the pioneering
Capacity Spectrum Method (CSM), introduced by Freeman (1998) and implemented in
ATC-40 guideline, and N2 method, (Fajfar and Fischinger, 1988) included in the
recommended simplified procedures of European Code (CEN, 2005a). These first
proposals are appealing mainly for the simplicity of the method and usually consider either
the first mode proportional or uniform load distribution for the pushover curve
computation. Recently, an improved version of CSM method has been presented in
FEMA-440 guidelines (ATC, 2005), mainly consisting of the update of the prescribed

5.3
Seismic Demand

empirical relations to determine both the equivalent viscous damping and spectral
reduction factor.

Not much later than the ATC-40-adopted CSM, the Displacement Coefficient Method
(DCM) (ATC, 1997) has been proposed within the NEHRP Guidelines for the Seismic
Rehabilitation of Buildings. Whereas ATC-40 referred only to concrete structures, DCM is
applicable to a wide range of structural systems and computes the elastic response, which
is then tuned by a series of coefficients that account for the nonlinear behaviour. DCM
assumes, therefore, a certain empirical character, which distinguishes it from the previous
ones.

An additional group of procedures includes more recent approaches to nonlinear static

analysis, such as the Modal Pushover Analysis (MPA) (Chopra and Goel, 2002), the
Adaptive Modal combination Procedure (AMCP), by Kalkan and Kunnath (2006) or the
Adaptive Capacity Spectrum Method (ACSM), developed by Casarotti and Pinho (2007).
All of them come up with some improvements, mainly the inclusion of higher modes
contribution or, as in the ACSM, an alternative way of addressing the reference node issue.
The application of methods which try to implement new features, i.e., the ones belonging
to this group, may generate some controversy and it may be argued that effortlessness is
being left behind, towards to a complexity level increasingly similar to dynamic analyses.
Such perspective is relative. Indeed there may be so-called advanced NSPs, MPA or
AMCP, which foresee several repeated steps, whereas others, such as ACSM, carry no
additional effort to the analysis, since higher modes are included in the analysis all together
when working out the pushover curve. The importance of an increase in the computational
effort will depend, anyhow, on the corresponding difference in the performance.

5.2 Nonlinear Static Procedures

The six aforementioned procedures are briefly discussed in what follows, going through
the fundamental aspects of a nonlinear static analysis within the context of bridge
structures, such as loading pattern, equivalent SDOF system computation or eventual
spectral reduction technique, among others.

5.4
 Seismic Demand

Herein, each of the referred methods is studied individually, scrutinised in several

possibilities, according to its settlement variables. With this preliminary approach, an
optimisation of each procedure is looked for and a best performing way is chosen for
comparison. A confront of the four selected NSP is subsequently carried out, with the
purpose of emphasizing relative advantages and disadvantages or, eventually, coming up
with the choice for one of them as noticeably better. An extensive parametric study,
covering fourteen different bridge configurations, ten earthquake records scaled to six
intensity levels, and four control variables (deck displacements and flexural moments and
piers and abutments shear forces), was carried out in order to make a full comparison of
the four major available methods. The predictions accurateness evaluation is made by
means of confront with dynamic analysis results.

5.2.1 Capacity Spectrum Method (CSM)

The Capacity Spectrum Method was initially introduced by Freeman et al. (1975) in the
early seventies as a methodology to rapidly evaluate seismic vulnerability of buildings,
with the basis idea of getting a reasonable approximation of the elastic and inelastic limits
for the structure’s behaviour. Basically it consists in the intuitive, and rather rational,
graphical comparison of the capacity with the demand, represented by over-damped
response spectrum. Such graphical nature gives the engineer the opportunity to visualise
the relationship between demand and capacity. The capacity spectrum is given by the
displacement of a reference node and base shear, coming from the nonlinear pushover
curve, in the spectral displacement and acceleration (ADRS) format. The pushover curve is
obtained applying a lateral load distribution proportional to the fundamental vibration
mode shape of the structure. Linear elastic response spectra, scaled to account for the
energy loss due to hysteretic cyclic behaviour, are used to characterize the demand in an
inelastic response spectrum fashion. The reduction factors are based on effective viscous
damping levels that may be computed using Newmark-Hall relationships for damping and
ductility or according to idealized hysteretic parallelogram loops. The intersection of the
two curves is then considered to be the performance point of the structure.

The method has been included in ATC-40, a document that emphasizes the use of
nonlinear static procedures in general, such as Displacement Coefficient Method, but
focusing on CSM. The document includes step by step procedures to determine capacity,

5.5
Seismic Demand

demand and performance. Capacity is represented by a pushover curve, generally

constructed to represent the first mode response of the structure, which is based on the
assumption that the fundamental vibration mode is predominant. Demand is represented by
5% damped response spectrum, reduced in terms of effective damping, and performance
point is attained according to three methodologies, different in terms of their dependency
on analytical versus graphical techniques.

Two important aspects should be noted: classical CSM uses the fundamental mode shape
when computing the pushover and capacity curves and depends on a reference node,
usually considered the top floor, for buildings, or the deck centre of mass, in the case of
bridges. For the latter case, this a very particular issue because the accuracy of the deck
central node when representing the structure is quite dependent on the piers configuration.
In what concerns the demand curve, the response spectrum is reduced considering the
equivalent viscous damping, obtained summing the elastic viscous damping (recommended
by ATC-40 to be 0.05) with the hysteretic damping.

To summarize, according to what is prescribed in ATC-40, the Capacity Spectrum Method

can be applied taking the following steps:

8. Create a computer model of the structure and apply lateral forces to the structure in
proportion to the product of the mass and fundamental mode shape, including
gravity loads.

9. Calculate and record the base shear and reference node displacement, constituting
the pushover curve of the structure.

10. Convert the obtained capacity curve to the capacity spectrum, a representation of
the base shear-displacement curve in Acceleration-Displacement Response Spectra
(ADRS) format (spectral acceleration Sa versus spectral displacement Sd), using
first mode quantities to make the transformation.

11. Construct a bilinear representation of the capacity spectrum, for a first trial
displacement and acceleration, estimate effective damping, combining viscous
damping inherent to the structure and hysteretic damping, and appropriate
reduction of spectral demand. The hysteretic damping is considered to be related to
the area inside the loops of the base shear-structural displacement diagram during

5.6
 Seismic Demand

the earthquake occurrence. It is obtained using an approach based on the energy

dissipated by the structure in a single cycle of motion (Chopra, 1995). The spectral
reduction factors use, in turn, the relationships developed by Newmark and Hall
(1982), and are defined and limited according to the structural behaviour type (A, B
or C), which depends on the shaking duration and the quality of the primary
elements of the seismic resisting system.

12. Intersect capacity spectrum and reduced demand spectrum, build new bilinear
capacity spectrum for that point and update effective viscous damping and spectral
reduction factor;

13. If the updated damping acceptably matches the previously determined/assumed

one, the Performance Point (PP) of the structure is found, otherwise, iteration is
carried out until convergence is reached.

Recently, FEMA-440 guidelines (ATC, 2005), within the ATC-55 Project, performed an
evaluation of current Nonlinear Static Procedures, which included a review of the Capacity
Spectrum Method, prescribed in ATC-40. The document constitutes an effort to assess
current NSPs for the seismic analysis and evaluation of structures as well as to present
suggestions developed to improve such procedures for practical application in the analysis
of both existing and new structures. The improved recommendations rely on the original
procedure, with most of the process remaining intact. New expressions to determine the
effective damping and period are introduced as well as a new technique to modify the
resulting demand spectrum to coincide with the familiar CSM technique of using the
intersection of the modified demand with the capacity curve to obtain the performance
point of the structure. The suggested new expressions are set by type of hysteretic model –
bilinear or stiffness degrading – and post-yielding stiffness parameter α and its parameters
vary according to the ductility achieved by the system. Additionally, general expressions,
independent from the structural system type, are proposed.

According to the previously exposed, CSM was tested, further on this study, using its
original configuration as well as considering the FEMA-440 improving recommendations,
with expressions independent from structural system. Additionally, the reference node has
been tried either as the deck centre of mass or maximum modal displacement node,
yielding a total of 4 distinct variants.

5.7
Seismic Demand

5.2.2 N2 Method

A simple Nonlinear Analysis Method for Performance Based Seismic Design was formally
proposed in the late nineties by Fajfar (1999), combining pushover analyses of a MDOF
model with the response spectrum analysis of the equivalent SDOF system. The method is
named N2, where ‘N’ stands for its nonlinear analysis character, and ‘2’ for the use of two
separate mathematical models, the application of the response spectrum approach and
pushover analysis. With such methodology, a satisfactory balance between reliability and
applicability for everyday use is intended, along with a contribution to new trends in
seismic design, even if eventually originally restricted to the planar analysis of new or
existing building structures. Similarly to CSM, N2 method is formulated in the
acceleration-displacement format, providing a visual interpretation of the procedure. The
main difference is that N2 makes use of inelastic spectra rather then elastic spectra with
equivalent damping and period.

The nonlinear procedure has been considered and implemented in Eurocode 8 (CEN,
2005a) for application either to building or bridges. In order to come up with the MDOF
model pushover curve at least two load distributions are recommended: a uniform one, in
which the shape factor is unitary along the building’s height or deck and a first mode
proportional shape. The most essential assumption of the method becomes, therefore, the
time-independent lateral displacement shape, notwithstanding that the method’s results are
believed not to be excessively sensitive to changes in the assumed displacement shape. An
additional possibility, which has effectively been considered in this framework, is the
envelope of those two load shapes, a sort of a filtering technique that would try to capture
the best of each loading profile. The capacity curve is represented plotting the base shear
force as a function of the reference node displacement coming from the pushover analysis.

N2 is, conceptually, another procedure where higher modes will hardly be taken into
account properly, being therefore recommend for the application of building structures
oscillating predominantly in a single mode, even if irregular. If not, demand quantities are
expected to be underestimated. Solutions for this drawback, based on appropriate dynamic
magnification of selected quantities, are being sought, though. The method is equally
dependent on the choice of a control node, recommended to be the top floor of the building
or the centre of mass of the deformed deck. The design point is determined throughout the

5.8
 Seismic Demand

computation of the target displacement of the SDOF system, the performance point of the
structure, obtained through expressions that vary with the range where the equivalent
period falls in, short or medium-long. Those expressions are fundamentally based on the
ductility achieved by the system and on the spectral acceleration for the equivalent period.

The application of N2 procedure, according to EC8, consists on the following steps:

1. Create a computer model of the structure and apply lateral forces in proportion to
the product of the mass and a shape factor, which should be considered, for the
deck, as constant or proportional to the first mode shape, and for the piers,
proportional to the height above the foundation of the individual pier.

2. Build the capacity curve, given by the relation between base shear force and control
node displacement, which should be the centre of mass of the deformed deck.

3. Transform the MDOF structure to an equivalent SDOF system, computing the

equivalent mass, m*, force, F*, and displacement, d*.

4. Determine the idealized elasto-perfectly plastic force-displacement relationship,

defined for the plastic mechanism point in such a way that the areas under the
actual and the idealized force-deformation curves are equal. At this point, a
maximum displacement, dm*, the one corresponding to the formation of the plastic
mechanism, may be assumed, in order to proceed with the bilinearization.

5. Determine the period, T*, of the idealized equivalent SDOF system, based on the
idealized bilinear capacity curve.

6. Find the displacement target, dt*, depending on the comparison of the equivalent
period with the code spectral period, TC, i.e., depending on whether the structure is
in the short-period range or in the medium and long-period range.

7. If dt* is very different from dm*, assumed in the bilinear capacity curve, iteration
may be optionally carried out, re-bilinearizing the capacity curve for dt*.

Three different loading possibilities have been considered for the application herein carried
out of N2 method: uniform load distribution, first mode load distribution and their
envelope. Because of the method’s reference node dependency, each of the three
modalities was repeated changing the reference node to the maximum displacement one

5.9
Seismic Demand

instead of the centre of the mass of the deck. This way, N2 method was applied in 6
different versions.

5.2.3 Displacement Coefficient Method (DCM)

The Displacement Coefficient Method is another method that has been proposed in a
document prepared by the Applied Technology Council, within the ATC-33 Project,
funded by the Federal Emergency Management Agency, FEMA 273 publication. Such
document, dated of 1996, had as major goal the development of technically sound
guidelines for the seismic rehabilitation of buildings and proposed four distinct analytical
procedures, linear or nonlinear, static or dynamic. DCM is the document’s considered
Nonlinear Static Procedure and consists in pushing the structure to a target displacement,
expected to be equivalent to the experienced one during the earthquake event. The target
displacement corresponds to the displacement obtained using the equal displacements
approximation, then modified by various coefficients.

The methodology, mainly the tabled coefficients, is based on statistical analysis of

dynamic analysis of SDOF models of different types and provides a direct numerical
process for calculating the displacement demand, not requiring any conversion of the
capacity curve to spectral coordinates. It is recommended to be applied to all buildings,
new or existing, that are regular and do not have adverse torsional or multimode effects.

The structural model, directly incorporating inelastic material response is displaced by

monotonically increasing lateral forces until the target displacement, controlling
corresponding internal forces and deformations, or the building collapses. The
displacement target is, similarly to other procedures, referred to a control node, assumed by
FEMA-273 guidelines to be the centre of mass at the roof. The lateral load pattern should
be tested according to two possibilities, uniform or modal, proportional to total mass at
each floor level. Computing the effective fundamental period in the direction under
consideration and effective lateral stiffness, the displacement that the elastic-behaving
structure would have can be determined. Four coefficients are then applied to adjust that
displacement, considering the relation between spectral displacement and the building
roof’s displacement, the relation between expected maximum inelastic displacements to

5.10
 Seismic Demand

displacements calculated for linear elastic response, the effect of hysteresis shape on the
maximum displacement response and the dynamic P-delta effects.

The Displacement Coefficient Method is, due to the empirical expressions previously
developed, quite simple to apply, completing the following steps, as prescribed by
FEMA-273:

1. Create a computer model of the structure and apply lateral forces using at least two
distributions, uniform and modal.

2. Establish the relation between base shear force and lateral displacement of the
control node, the capacity curve.

3. Construct a bilinear representation of the capacity curve considering that the pre-
yielding branch passes through the point corresponding to a base shear of 0.6Vy,
where Vy is the yielding shear.

4. Calculate the effective fundamental period from the bilinear capacity curve.

5. Calculate the target displacement by the elastic behaving structure spectral

displacement, affected by a set of four coefficients, C0 to C3, which take into
account nonlinear response of the structure.

Displacement Coefficient Method was not formulated having application to bridges in

mind. Indeed, FEMA-273 guidelines state that the procedures have not been tested for each
and every structural type, particularly those that have generally been covered by their own
codes or standards, such as bridges. For that reason and due to its substantial empirical
nature, this nonlinear static procedure will not be included in the comparative study.

5.2.4 Modal Pushover Analysis (MPA)

The Modal Pushover Analysis was introduced by Chopra and Goel (2002) and consists in
the repeated application of a given nonlinear static analysis procedure for each of the
significant vibration modes of the structure, followed then by an adequate combination of
the results. Self claimed as a procedure based on structural dynamics theory, retaining the
conceptual simplicity and computational attractiveness of current procedures with invariant
force distributions, has been proposed and widely spread up in the past recent years. The
main reason for such acceptance was the awareness of the accuracy lack that pioneering

5.11
Seismic Demand

simplified methods, such as CSM in ATC-40, would many times provide. This is thought
to be mainly due to disregarding higher modes contribution to the response or the
redistribution of inertia forces because of structural yielding and associated changes in the
vibration properties, aspects deemed to assume a greater meaning when analysing bridges.

The whole procedure is actually quite similar to the one pursued by the other methods. It
starts with the computation of a number of MDOF pushover curves, each of which
obtained employing a load distribution that is proportional to the individual modes of
vibration being considered. The equivalent SDOF capacity curves are then determined by
transforming and bilinearizing the modal pushover-derived base shear-displacement
relations, making use of the nth modal quantities and the reference node displacement. The
main step, estimating the peak deformation, i.e., the performance point, can be done by
means of (i) response history analysis (RHA), (ii) inelastic design spectrum or (iii)
empirical equations for the ratio of deformations of inelastic and elastic systems. Response
history analysis consists in performing a nonlinear dynamic analysis of the equivalent
SDOF system for each significant mode, characterized by the bilinearized pushover curve.
Finally, a quadratic combination rule (e.g., through SRSS or CQC) is employed to combine
the responses obtained for each modal analysis.

Several improvements, with respect to its original formulation, have been carried out
within the Modal Pushover Analysis proposed procedure. The authors have focused on a
new way of computation of member forces, given that the initial quadratic combination of
results could lead to miscellaneous estimates, comparing them to the actual shear member
capacity. Additionally, other features have been included, such as inclusion of P-∆ effects
for all modes, different way of computing beam plastic rotations as an iterative procedure
to solve dependency on selected ground motion record for RHA.

More recently, a Modified Modal Pushover Analysis methodology, MMPA, has been
suggested as well, as a new, faster, MPA. Higher modes are considered with the structure
as elastic-behaving, which corresponds to a single nonlinear pushover analysis, the first
mode one, and, therefore, less computational effort.

The original procedure, together with the recent adjustments, widely found in available
literature (Chopra and Chintanapakdee, 2004; Chopra and Goel, 2004; Goel and Chopra,

5.12
 Seismic Demand

2004; Chopra, 2005; Goel, 2005; Goel and Chopra, 2005b, 2005a), can be summarized in
the following steps:

1. Compute the n natural frequencies and modes for the linearly elastic vibration of
the structure.

2. For the first mode, develop the base shear-roof displacement pushover curve for
force distribution proportional to the mass and mode shape.

3. Idealize the pushover curve as a bilinear curve and convert it, computing the first
mode inelastic SDOF system quantities.

4. Compute the peak deformation of the first mode inelastic SDOF system defined
previously using nonlinear response history analysis, inelastic design spectrum or
empirical equations for the ratio of deformations of inelastic and elastic systems.

5. Compute the dynamic response due to the first mode combining the effects of
lateral and gravity loads.

6. Compute the dynamic response due to higher modes under the assumption that the
system remains elastic, performing a classical modal analysis of a linear MDOF
system, skipping the need for additional pushover analysis.

7. Determine the total response combining the peak modal responses using SRSS rule.

As for the conventional NSPs, the MPA method relies on the choice of a given reference
node, hence two variants have once again been considered in the study that will follow,
one using with the central deck node as a reference and the other selecting the reference in
correspondence to the point of maximum deck deflection. It is also noted that the inelastic-
elastic response ratios approach was adopted in this work for the determination of the
performance point.

5.2.5 Adaptive Capacity Spectrum Method (ACSM)

Recent studies (Pinho et al., 2007; Monteiro et al., 2008a; Pinho et al., 2009) reported the
viability of employing Displacement-based Adaptive Pushover (Antoniou and Pinho,
2004) to estimate seismic response of bridges. Contextually, an equally adaptive NSP has
been proposed, and preliminarily verified, by Casarotti and Pinho (2007). The proposed
approach combines elements from the Direct Displacement-based design method (e.g.,

5.13
Seismic Demand

(Priestley and Calvi, 2003a)) and the Capacity Spectrum Method (ATC, 1996; Freeman,
1998), maintaining a spectrum-based approach which employs the substitute structure
methodology to model an inelastic system with equivalent elastic properties philosophy,
but elaborated and revised within an “adaptive” perspective, for which reason it can also be
viewed as an Adaptive Capacity Spectrum Method (ACSM). The procedure essentially
consists in deriving an adaptive SDOF capacity curve and plotting it versus the
Acceleration-Displacement Spectrum of the design earthquake, appropriately over-
damped, thus obtaining the design intersection. For that intersection, the Performance
Point, a model for the relationship between the hysteretic energy dissipation and equivalent
viscous damping is used, explicitly accounting for the ductility achieved by the system,
and iterations are carried out until convergence in damping is found.

The proposed method is therefore distinct from the original Capacity Spectrum Method,
making use of: (i) more reliable displacement-based adaptive pushover curves, (ii)
equivalent SDOF curve without reference either to any given elastic or inelastic mode
shapes, but calculated step by step based on the actual deformed pattern, either than
invariant elastic or inelastic modal shape, and not built on a modification of the capacity
curve referred to the displacement of a specific physical location. As a consequence, all the
‘equivalent SDOF quantities’ even though of same ‘format’ of the corresponding modal
quantities, are also calculated step-by-step based on the actual deformed pattern at each
analysis step, which, together with the fully adaptive pushover algorithm, stands for the
double adaptiveness of the procedure.

The procedure’s algorithm may be summed up in the following steps:

1. Perform a reliable pushover analysis on a nonlinear model of the MDOF structure

and derive, step-by-step, the equivalent SDOF adaptive capacity curve.

2. Apply the demand spectrum to the SDOF capacity curve, determining their
intersection, the performance point, for an assumed damping.

3. Bilinearize the capacity curve at the performance point and calculate corresponding
system damping.

5.14
 Seismic Demand

4. Determine if the actual damping matches the assumed one. If so, the performance
point is established, otherwise, update damping and repeat steps 2 to 4 until
convergence is found.

With respect to the demand spectrum reduction to account for the hysteretic energy
dissipation ability of the structures, several possible scenarios have been considered and
evaluated, nine damping-based and two ductility-based spectral reduction modalities,
seeking for an optimal way of applying ACSM, as described in (Casarotti et al., 2009).

For what seismic assessment of structures is concerned, the use of pushover-based

simplified procedures is considered as a useful alternative to the more rigorous nonlinear
dynamic analyses, if and when all relevant variables and effects are suitably taken into
account. Those variables include, namely within the Adaptive Capacity Spectrum Method,
amongst others, the hysteretic damping associated to the energy dissipation capacity that a
structure inherently presents during an earthquake. One of the main concerns, when
applying such procedure, is thus the definition of a demand spectrum that features
ordinates appropriately scaled-down to take due account of the aforementioned capacity of
structures for dissipating seismic energy through hysteresis.

Reduction of spectral ordinates may be carried out through the use of either over-damped
elastic or constant-ductility inelastic spectra. The former make use of equations that
estimate, as a function of ductility, values of the so-called equivalent viscous damping
which is then used as input into another set of expressions that provide the spectral scaling
factor. In alternative, the use of constant-ductility inelastic spectra, although perhaps less
commonly, has also been proposed as a means to estimate seismic demand within the
scope of nonlinear static assessment of structures (Chopra and Goel, 1999; Fajfar, 1999).

There are a relatively large number of past parametric studies dedicated to the derivation
and/or validation of different approaches to estimate spectral reduction factor values (e.g.,
(Miranda, 2000; Miranda and Ruiz García, 2002)), however such studies seem to have
focused mainly, if not exclusively, on SDOF systems. It seems, therefore, that verification
on full structural systems is conspicuously needed in order to verify the adequacy of using
existing SDOF-derived relationships in the assessment of MDOF systems. In the present
work, the case of bridges was considered together with, as previously mentioned, eleven

5.15
Seismic Demand

different approaches for taking into account, through spectral scaling, the energy
dissipation capacity of such systems.

As already mentioned, spectral Reduction Factors (RF) can be roughly divided in two
groups: damping-based and ductility-based. Roughly, the first family consists of all those
methods which, through the application of a reduction factor B based on the equivalent
viscous damping (elastic viscous plus hysteretic), reduce by the same amount both
displacement and acceleration spectral ordinates (see Equation (5.1) and Figure 5.2 left).
To the second category belong all those approaches which make use of a 5%-damping
elastic response spectrum and then reduce the spectral acceleration ordinates by a factor
defined as a function of ductility (Equation (5.2) and Figure 5.2 right). The spectral
reduction within ductility-based methods is not exactly vertical, given that displacements
are modified as well, however, for the range of periods considered in this work, R≈µ. A
few hybrid methodologies have also been proposed, as discussed subsequently.

S a,damp = B ⋅ S a,el −5%

S d,damp = B ⋅ S d,el −5%
(5.1)
S a,damp
S d,damp =
ω2

S a ,el −5%
S a ,duct =
R (5.2)
µ
S d ,duct = S a ,el −5% = C ⋅ S d ,el −5%
R

Sa Sa

Sd Sd

Figure 5.2 – Spectral reduction methods: damping-based (left) and ductility-based (right).

5.16
 Seismic Demand

5.2.5.1 Damping-based spectral reduction

With regard to damping-based spectrum scaling methods, a combination of different ways

of calculating the (i) equivalent viscous damping ξeq and (ii) corresponding spectral
reduction factor B are available. Some of the most commonly used ones are presented and
reviewed in what follows.

Equivalent viscous damping models

In the following equations, ξ0 stands for the so-called elastic viscous damping and µ for
ductility (which was a variable of the work, given that it varied with the intensity level and
the characteristics of the response of each bridge).

A) ATC-40, based on the modified Rosenblueth and Herrera model (herein termed
ATC40)

This proposal by Rosenblueth and Herrera (1964), and subsequently adopted by ATC-40
(ATC, 1996), was the first equivalent linear method to suggest the use of secant stiffness at
maximum deformation as the basis for considering inelastic response. In such approach, if
one considers a bilinear system with a post-yield stiffness ratio α, the viscous damping for
the equivalent linear elastic system is given by Equation (5.3), where κ is an empirical
parameter that takes account the degree to which the hysteresis response cycle resembles a
parallelogram or not; three possibilities are defined (A, B or C), depending on structural
system configuration and duration of ground shaking.

2  (1 − α )(µ − 1) 
ξ eq− ATC 40 = ξ 0 + κ (5.3)
π  µ − αµ + αµ 2 

B) Kowalsky, based on the Takeda hysteretic model with post-yield hardening (herein
termed TakKow)

Kowalsky (1994) derived an equation for equivalent viscous damping ratio that was based
on the Takeda hysteretic model. For a thin response mode (empirical parameter b=0) with

5.17
Seismic Demand

unloading stiffness factor of 0.5 and a post-yield stiffness ratio α, the equivalent damping
ratio is given by Equation (5.4).

1  (1 − α ) 
ξ eq −TakKow = ξ 0 + 1 − −α µ  (5.4)
π  µ 

C) Gulkan and Sozen, based on the Takeda model without hardening (herein termed
TakGS)

Gulkan and Sozen (1974) used the Takeda hysteretic model and experimental shaking table
results of small-scale reinforced concrete frames to develop empirical Equation (5.5) to
compute equivalent viscous damping ratio values.

 1 
ξ eq−TakGulSoz = ξ 0 + 0.21 −  (5.5)
 µ 

D) Iwan (herein termed Iwan)

Iwan (1980) derived empirically equations to estimate the equivalent viscous damping
ratio, Equation (5.6), using a hysteretic model derived from a combination of elastic and
Coulomb slip elements together with results from dynamic analyses using 12 earthquake
ground motion records.

ξ eq− Iwan = ξ 0 + 0.0587(µ − 1)0.371 (5.6)

E) Dwairi et al. (herein termed Dwairi or DwaKowNau)

Dwairi et al. (2007) recently developed new equivalent viscous damping relations for four
structural systems, defined as a function of ductility and effective period of vibration. With
the latter, the authors claim to having managed to significantly reduce the error in

5.18
 Seismic Demand

predicting inelastic displacements and minimize the scatter of results. Considering the
structural system that best fits the case of continuous deck bridges, the equivalent viscous
damping relation is that shown in Equation (5.7), where CST is a parameter dependent on
the effective period, ranging from a minimum of 0.3 to a maximum value of 0.65.

C ST  µ − 1 
ξ eq− DwaKowNau = ξ 0 +   (5.7)
π  µ 

F) Priestley et al. (herein termed Priestley)

The approach proposed by Priestley et al. (2007) can, in a somewhat simplified manner, be
represented by Equation (5.8). Actually, the equation should be applied to each individual
pier, and then a weighted average based on shear forces and response displacements would
be used to estimate the overall damping of SDOF system. Herein, for reasons of simplicity
and congruency with the employed NSP, the simplification of applying Equation (5.8)
directly to the full system is carried out.

 µ −1 
ξ eq−Pr iestley = ξ 0 + 0.444  (5.8)
 µπ 

Figure 5.3 plots the six previously listed approaches for equivalent viscous damping
estimation. It is noted that ATC-40 results are computed for a structural type B (the most
appropriate for the structures considered), no post-yield hardening was considered for
ATC40 and TakKow approaches, and Dwairi’s estimates are plotted for two representative
values of 0.4 and 0.5.

5.19
Seismic Demand

ATC40
TakKow
0.3 TakGS
Iwan
Dwairi (CST =0.4)
Damping ratio, ξ eq

0.25 Dwairi (CST =0.5)

Priestley
0.2

0.15

0.1

0.05
1 2 3 4 5 6 7 8 9 10
Ductility

Figure 5.3 – Considered damping models.

It is readily observed that whilst ATC40 and TakKow models distinguish themselves from
the rest by providing significantly higher equivalent viscous damping estimates, the
remaining four approaches (TakGS, Iwan, Dwairi and Priestley) yield results that are very
close. With the latter in mind, and considering its relative contemporariness with the work
of Gulkan and Sozen (1974), the expression proposed by Iwan (1980) will not be
considered on the subsequent parametric study, also because, in opposite to the other
relationships, it does not feature an upper bound limit. Such rationale could also have led
to the exclusion of one the two proposals from Dwairi et al. (2007) and Priestley et al.
(2007), however in this case both were kept in the parametric study given their diverse
nature in terms of application.

Damping-based reduction factors

As previously discussed, the computation of ξeq is then followed by the calculation of the
corresponding spectral reduction factor B. Different approaches may again be considered.

A) Newmark-Hall and ATC-40 (herein termed NH-ATC40)

In the well known method proposed by Newmark and Hall (1982), the damping reduction
factors BNH for median estimates of response (i.e. 50% probability of exceedance) are
given by Equation (5.9).

5.20
 Seismic Demand

 3.21 − 0.68 ln (100ξ eq )

 BNH ( acc ) = , Tb ≤ T < Tc
 2 . 21
 2.31 − 0.41ln (100ξ eq )
 BNH ( vel ) = , Tc ≤ T < Td (5.9)
 1.65
 1.82 − 0.27 ln (100ξ eq )
 BNH ( disp ) = , T ≥ Td
 1.39

The data of Newmark and Hall were limited to viscous damping ratios of 20% and are
obtained from a limited number of earthquakes prior to 1973. In addition they were derived
from the displacement response spectrum or pseudo-acceleration response spectrum. It is
noted that, for damping ratios higher than 5%, BNH (acc) < BNH (vel) < BNH (disp).

The method has been adapted by most of the American design codes and guidelines, such
as the ATC-40 (1996), among others, where BATC40 is defined by SRA and SRV,
corresponding to constant acceleration and velocity regions, respectively:

 3.21 − 0.68 ln(100ξ eq )

SR A = ≥ SR A T < Tc
2.21
SR = 2.31 − 0.41ln(100ξ eq ) ≥ SRV T ≥ T
 (5.10)

 V 1.65
c

In Equations (5.9) and (5.10) a period Tc’, given by Equation (5.11), should be applied in
order to guarantee continuity conditions with respect to the corner period.

B( vel )
Tc ' = Tc (5.11)
B( acc )

The constraints imposed by Equation (5.10), referring to Table 5.1, depend on the
aforementioned ATC-40 structural typologies (A, B or C), and imply maximum admitted
damping ratios of 37-40% for type A, 28-29% for type B and 19-20% for type C.
Analogously, SRA < SRV.

5.21
Seismic Demand

Table 5.1 – Maximum allowable SRA, SRV (ATC-40) and SRD (present study) values.

Structural Type SRA SRV SRD

A 0.33 0.50 0.59
B 0.44 0.56 0.66
C 0.56 0.67 0.73

In the present study, an approach blending the proposals of Newmark and Hall (1982) and
ATC-40 (1996) is taken into account. The original NH formulation is therefore
complemented by considering the lower limits introduced by ATC-40 for the velocity and
acceleration zones, and introducing a factor SRD (see Table 5.1) with a limitation similar to
SRA and SRV, i.e., maximum admitted damping ratios of 40%, 29% and 20%, for types A,
B and C respectively.

B) Eurocode 8 (herein termed EC8)

Eurocode 8 (CEN, 2005a) recommends the use of the spectral reduction factors given in
Equation (5.12), with a minimum of 0.55.

 T
1 − (1 − η ) T 0 ≤ T < Tb
BEC 8 = b

η T ≥ Tb
 (5.12)

10
η= ≥ 0.55
5 + 100 ξ eq

C) Ramirez et al. (herein termed Ramirez)

Ramirez et al. (2002) proposed a bilinear relationship – Equation (5.13) – between the
reduction factor Bshort and equivalent damping ratio ξeq, valid up to damping ratios of 50%.
Exceeding that value the relation becomes trilinear and dependent on Blong – Equation
(5.14). Tb and Tc are the first and third spectral characteristic/corner periods, whilst Bshort
and Blong can be found in Table 5.2.

5.22
 Seismic Demand

 T
1 − (1 − Bshort ) T 0 ≤ T < Tb
BRam = b
(5.13)
 Bshort T ≥ Tb



1 − (1 − Bshort ) T
T
0 ≤ T < Tb
 b

BRam =  Blong − (Bshort − Blong )

 (T − Tb ) Tb ≤ T < Tc (5.14)
 (Tc − Tb )
 Blong T ≥ Tc



Table 5.2 – Reduction factors for Ramirez et al. approach.

Equivalent damping, ξ (%) Bshort Blong

5 1.00 Bshort
10 0.83 Bshort
20 0.67 Bshort
30 0.59 Bshort
40 0.53 Bshort
50 0.45 Bshort
60 0.43 0.38
70 0.43 0.34
80 0.42 0.30
90 0.41 0.27
100 0.40 0.25

D) Lin and Chang (herein termed Lin-Chang)

In a recent study, Lin and Chang (2003) proposed a period-dependent reduction factor,
Equation (5.15), based on an extensive dynamic analysis parametric study of linear elastic
SDOF systems, using real USA records. Such RF has the advantage that, whilst still
following the general trend of the others with respect to the structural period T, it is
inherently continuous, hence not requiring the addition of any other type of continuity
constraints or conditions.

5.23
Seismic Demand

a ⋅ T 0.3
BLinChang = 1 −
(T + 1)0.65 (5.15)
a = 1.303 + 0.436 ln (ξ eq )

E) Priestley et al. (herein termed Priestley)

Recently, Priestley et al. (2007) proposed the application of Equation (5.16), which was
included in earlier versions of Eurocode 8, but was then subsequently abandoned.

0.5
 0.07 
BPr iestley =   (5.16)
 0.02 + ξ 0 

Figure 5.4 presents results obtained with the above-listed spectral reduction equations for
values of response period from three representative spectral regions; constant acceleration
zone (T=2Tb), constant velocity zone (T=2Tc), constant displacement zone (T=1.5Td).

It is observed that those approaches that feature a lower bound limit (i.e. NH-ATC-40, EC8
and Priestley) tend to yield the higher spectral reductions for low damping ratios (Priestley
across the entire period range, EC8 in the displacement zone and NH-ATC-40 in the
acceleration zone). On the other hand, for higher values of damping (larger than 0.25-0.36)
Ramirez and Lin-Chang equations provide the higher spectral reductions. The dispersion
among the different proposals also increases with damping ratio.

5.24
 Seismic Demand

1 1
NH-ATC40 NH-ATC40
0.9 EC8 0.9 EC8
Ramirez Ramirez
0.8 0.8
Lin-Chang Lin-Chang
Priestley Priestley

Reduction factor (B)

Reduction factor (B)

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 Velocity Zone: T=2Tc 0.1 Displacement Zone: T=1.5Td

0 0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Damping ratio Damping ratio

1
NH-ATC40
0.9 EC8
Ramirez
0.8 Lin-Chang
Priestley
Reduction factor (B)

0.7

0.6

0.5

0.4

0.3

0.2

0.1 Acceleration Zone: T=2Tb

0
0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Damping ratio

Figure 5.4 – Reduction factor variation with damping in the three spectral regions.

Finally, it is perhaps also noteworthy to see that the NH-ATC-40 formulae is highly
sensitive to period values; if one considers a constant high value of damping (e.g., ξeq=0.3),
the spectral reduction factor may vary from 0.44 to 0.67, depending on the response period
of the structure. To shed further insight into this latter issue (i.e. period dependency of
spectral reductions), Figure 5.5 has also been produced.

5.25
Seismic Demand

1 1
10% Equivalent Damping 20% Equivalent Damping
0.9 0.9

0.8 0.8

0.7 0.7
Reduction factor (B)

Reduction factor (B)

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3
NH-ATC40 NH-ATC40
0.2 EC8 0.2 EC8
Ramirez Ramirez
0.1 Lin-Chang 12%B max variation 0.1 Lin-Chang 31%B max variation
Priestley Priestley
0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Period (s) Period (s)

1 1
30% Equivalent Damping 40% Equivalent Damping
0.9 0.9

0.8 0.8

0.7 0.7
Reduction factor (B)

Reduction factor (B)

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3
NH-ATC40 NH-ATC40
0.2 EC8 0.2 EC8
Ramirez Ramirez
0.1 Lin-Chang 50%B max variation 0.1 Lin-Chang 55%B max variation
Priestley Priestley
0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Period (s) Period (s)

1 1
50% Equivalent Damping 60% Equivalent Damping
0.9 0.9

0.8 0.8

0.7 0.7
Reduction factor (B)

Reduction factor (B)

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3
NH-ATC40 NH-ATC40
0.2 EC8 0.2 EC8
Ramirez Ramirez
0.1 Lin-Chang 82%B max variation 0.1 Lin-Chang 112%B max variation
Priestley Priestley
0 0
0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
Period (s) Period (s)

Figure 5.5 – Reduction factor variation with period for different damping ratios.

It is observed that the dispersion among the several RF approaches, indicated in the plots
as a percentage of the maximum RF, increases with the damping ratio. It is also noted that
the NH-ATC-40 variant generally yields the highest reduction factor, especially for periods
higher than 3 or 4 seconds.

5.26
 Seismic Demand

In the parametric study described subsequently, three out of the above-listed five possible
reduction factor approaches have been selected: (i) Lin-Chang (Equation (5.15)), (ii)
Priestley (Equation (5.16)), (iii) a combination of EC8 and Ramirez (Equation (5.17)). This
last hybrid approach, herein termed EC8Ram, aimed especially at creating a more
simplified version of EC8 and Ramirez proposals that would nonetheless envelop the
general period-dependent trend of RF; EC8 is employed for low periods where it yields
lower RF estimates and Ramirez is considered for the high period range where it provides
high RFs.

 BRam
BEC8 0 ≤ T < Tdd = Td B
BEC8 _ Ram =  EC 8
(5.17)
BRam T ≥ Tdd


The somewhat cumbersome trilinear NH-ATCH40 relationship was not considered further,
whilst the equation by Priestley et al. (2007) was employed only in tandem with the
equivalent viscous damping equations proposed by the same authors (Equation (5.8)).

5.2.5.2 Ductility-based spectral reduction

As mentioned before, the reduction of the spectral ordinates may also be achieved by an
alternative type of approach, whereby a reduction factor, directly dependent on ductility, is
used. Amongst the different approaches present in literature, two have been considered in
the current work:

A) Miranda (2000) for firm soils (herein termed Mir2000)

Miranda (2000) observed that, for sites with average shear-wave velocities higher than 180
m/s in the upper 30 m of the soil profile (typically soil types A-B-C-D in ATC and FEMA
guideline documents), inelastic displacement ratios are not significantly affected by local
site conditions, nor by changes in earthquake magnitude, nor by changes in epicentral
distance (with the exception of very near-field sites that may be influenced by forward
directivity effects). As a result, the displacement modification factor expression, in
Equation (5.18), was proposed:

5.27
Seismic Demand

−1
 1  
C Mir ( )
= 1 +  − 1 exp − 12 T µ −0.8  (5.18)
 µ  

B) Vidic et al. (herein termed VidFajFish)

Vidic et al. (1994) suggested a relationship for a ductility-based reduction factor, with a
corner period TC’ dependent of the characteristic spectral period TC, Equation (5.19):

−1
  
µ 1.35(µ − 1) 0.95 T
+ 1 T ≤ Tc '
CVFF =  Tc ' 

[
µ 1.35(µ − 1) + 1
0.95
]
−1
T > Tc ' (5.19)

Tc ' = 0.75µ 0.2Tc ≤ Tc

In the parametric study that follows, a number of diverse spectrum scaling approaches will
be employed, considering combinations of different equivalent viscous damping models
and damping-based scaling factors, together with ductility-based scaling equations. With
reference to the nomenclature introduced above, the eleven cases considered are hence:
ATC40 - EC8Ram, ATC40 - LinChang, TakKow - EC8Ram, TakKow - LinChang, TakGS
- EC8Ram, TakGS - LinChang, DwaKowNau - EC8Ram, DwaKowNau - LinChang,
Priestley, Mir2000, VidFajFish.

5.2.6 Adaptive Modal Combination Procedure (AMCP)

The Adaptive Modal Combination Procedure, proposed by Kalkan and Kunnath (2006), is
fundamentally based on the adaptive pushover procedure of Gupta and Kunnath (2000) in
which the main feature is the modification of the applied lateral loads according to the
changes in the modal attributes of the structure, as well as the system’s response during the
inelastic phase, as the earthquake load carries on. Such new lateral load configuration,
using factored modal combinations, enhances an alternative scheme to represent realistic
lateral force demands, managing to incorporate inherent advantages of CSM and MPA, yet

5.28
 Seismic Demand

avoiding the need for a pre-estimated target displacement. Indeed the procedure makes use
of a displacement-controlled method, in which the demand is determined by individual
adaptive pushover analysis using inertia distribution of each mode, continuously updated.

With respect to the equivalent single degree of freedom system definition, the
corresponding displacement, abscissa of the ADRS format capacity curve, is obtained at
each step through an energy-based approach, computing the work done therein by the
lateral force pattern. The performance point is determined in a Capacity Spectrum Method
similar fashion with the difference of using inelastic spectra, computed for a set of ductility
levels, skipping a preliminary estimate for the target displacement. This so-called dynamic
target point is, therefore, the intersection of the equivalent SDOF capacity curve with the
inelastic demand spectrum corresponding to the global system ductility, which contributes
to a more realistic representation of demand. The match will be better as the ductility
refinement used for computation of inelastic spectra increases. The procedure is carried out
separately for each of the n relevant modes and the total response, similarly to what is done
in Multimodal Pushover Analysis, is simply determined by combining peak modal
responses with an adequate rule (SRSS or CQC). In the end, n individual mode
contribution adaptive pushover analyses and corresponding target point determination
processes (with inelastic spectra computed for possible several levels) will be required
within this procedure.

The following steps synthesize the procedure:

1. Compute modal properties of the structure at the current state of the system.

2. For the nth mode, construct the adaptive lateral load pattern proportionally to the
mass and mode shape; recomputed the load distribution for every step or at a set of
predefined steps.

3. Evaluate the next incremental step of the capacity curve for each equivalent SDOF
system using the energy based approach.

4. If the response is inelastic for the ith step of the nth mode pushover analysis,
calculate the approximate global system ductility and post-yield stiffness ratio,
using a bilinear representation.

5.29
Seismic Demand

5. Generate the capacity spectra in the ADRS format for a series of predefined
ductility levels and plot it together with the inelastic demand spectra at different
ductility levels. The dynamic target point will be the intersection of the equivalent
SDOF system modal capacity curve with the inelastic demand spectrum
corresponding to the global system ductility. The match will be as better as higher
the ductility levels’ refinement is.

6. Repeat steps 1 to 5 for as many modes as deemed essential for the system under
consideration and combine peak modal results using the SRSS combination
scheme.

Several difficulties have been found throughout the implementation of the procedure on
bridges, concerning the computation of the adaptive pushover curves with individuated
modes’ contribution. Additionally, the authors themselves do not claim it eligible for this
sort of structures, potentiating the decision of not to include the AMCP in the endeavoured
parametric study.

5.2.7 Method comparison

As stated along the description, from the six more popular proposed Nonlinear Static
Procedures only four have been selected for bridge application: CSM, N2, MPA and
ACSM. The other two were discarded given their low rationality or applicability to this
sort of structures. All the four methods were applied with the main purpose of making a
comparison of current NSPs, a task that represents an inedited study. Additionally, each
method was preliminary studied in different versions in order to select its best
performance, selecting them for the confronting. Table 5.3 presents a summary of the main
considerations in each of the four methods, in order to clarify the conditions in which the
comparative study was carried out.

5.30
 Seismic Demand

Table 5.3 – Summary of studied Nonlinear Static Procedures

ACSM CSM N2 MPA
Pushover Adaptive
Conventional force-based
analysis type displacement-based
Displacements
Load pattern 1st mode 1st mode All significant
loading
Base shear vs.
Displacement
computed from all Base shear vs. Displacement of a reference/control node,
Capacity curve
nodes, including usually recommended as the centre of mass of the deck
higher modes
contribution
Equivalent viscous damping reduced Inelastic ductility-based reduced
Demand curve
spectra spectra
Six:
Four:
1st mode, uniform
Original CSM and Two:
Eleven: loading pattern or
FEMA improved Central or
Number of Different demand envelope with
with central or Maximum
studied versions spectrum reduction central or
maximum displacement
possibilities maximum
displacement reference node
displacement
reference node
reference node

5.3 Parametric Study

The investigate the applicability of nonlinear static procedures, as well as their individual
calibration, a set of bridge structures has been selected going through different regularity
levels, in terms of piers’ heights and deck length. Further details on the considered case
study can be found in 4.3.

The seismic demand on the bridge models is evaluated by means of nonlinear dynamic
analyses (NDA), assumed to constitute the most accurate tool to estimate the ‘true’
earthquake response of the structures, using the fibre-based finite elements program
SeismoStruct (SeismoSoft, 2008). The same software package was employed in the
running of the force-based conventional pushovers (used in CSM, N2 and MPA methods)
and of the displacement-based adaptive pushover analyses (Antoniou and Pinho, 2004) that
are required by the ACSM procedure.

Results are presented in terms of different response parameters: the estimated displacement
pattern (D) and flexural moments (M) of the bridge deck at the nodes above the piers, and
the shear forces at the base of the piers (V) and abutments (ABT). Then, in order to
appraise the accuracy of the NSPs results obtained with the different approaches, these are
normalized with respect to the median of the corresponding response quantities obtained

5.31
Seismic Demand

through the incremental NDAs; this provides an immediate indication of the bias for each
of the four procedures. Equation (5.20) shows, for a generalized parameter ∆ at a given
location i, how the results from the incremental dynamic analyses (IDA), run for each of
the ten records considered, are first processed.

[
∆ˆ i , IDA = median j =1:10 ∆ i , j − IDA ] (5.20)

The aforementioned results’ normalization consists thus in computing, for each of the
parameters and for each of the considered locations, the ratio between the result coming
from each NSP and the median result coming from NDA, as illustrated in Figure 5.6 and
numerically translated into Equation (5.21). Ideally the ratio should be unitary.

∆ i , NSP
∆i = ⋅⋅⋅ → 1 (5.21)
∆ˆ i , IDA
ideally

∆2 ∆3
∆2 = ∆3 =
∆1 ∆ˆ 2 , IDA ˆ∆ ∆4
∆1 = 3 , IDA ∆4 =
∆ˆ 1,IDA ∆ˆ 4 ,IDA

Figure 5.6 – Normalized transverse deformed pattern.

This normalization renders also somewhat “comparable” all deck displacements, moments
and shear forces, since all normalized quantities have the same unitary target value, thus
allowing in turn the definition of the aforementioned Bridge Index (Pinho et al., 2007).
Recalling the definition presented in Section 4.4.4, the bridge index (BI) is computed as
the median of normalized results for the considered parameter over the m deck locations:
deck displacements (BID), deck moments (BIM) or shear forces at the piers and abutments
(BIV and BIABT), as shown in Equation (5.22). The standard deviation STD measures, on

5.32
 Seismic Demand

the other hand, the dispersion with respect to the median, for each of the procedures’ tested
versions – Equation (5.23).

[
BI ∆ , NSP = mediani =1:m ∆ i , NSP ] (5.22)

( ) 
0.5
 m ∆ i , NSP − BI
∑
2

=  i =1
NSP
STD∆ , NSP (5.23)
 m −1 
 

5.4 Results

In this section, the results obtained from the aforementioned parametric study are
scrutinized and interpreted, with a view to evaluate the accuracy of the different NSPs
considered (recalled in Table 5.3). However, before passing onto a direct comparison
between the four procedures, a preliminary study was carried out to identify which of the
variants of the CSM, N2, MPA and ACSM methods, discussed in previous sections and
summarized in Table 5.3, would lead to the attainment of best results.

5.4.1 Results representation

The relative performance of possible variants within each procedure, or the comparison of
the methods itself, is evaluated in detailed fashion with respect to two variables: intensity
level and bridge configuration. According to such formula, the intensity level of results
consists of the median Bridge Index over the 14 bridge configurations whereas a bridge
configuration level of results represents, for each bridge configuration, the median Bridge
Index across the 6 intensity levels. While an intensity level detail of results will show how
procedures behave when the structure enters the nonlinear range, a bridge configuration
detailed level of results enables the analysis of the influence of symmetry, regularity,
length, abutments type, among other variables. Finally, a global overview may be put up
where the bridge index over all the 14 bridges and 6 intensity levels is computed. In other
words, the median bridge index over all bridge configurations and intensity levels
represents the median of the single BI of every considered bridge configuration, at every

5.33
Seismic Demand

intensity level. The same sort of compound approach can be carried for standard deviation
as well. Figure 5.7 schematically presents the different types of detailing level of results.

BRIDGE CONFIGURATION BRIDGE CONFIGURATION

INTENSITY LEVEL

INTENSITY LEVEL
INTENSITY
MEDIAN
median
LEVEL MEDIAN
median
RESULTS

BRIDGE
CONFIGURATION
RESULTS

BRIDGE CONFIGURATION
INTENSITY LEVEL

MEDIAN
median

GLOBAL
RESULTS

Figure 5.7 – Types of approaches for results: intensity level detailed (top left), bridge configuration
detailed (top right), global (bottom).

The importance, and relevance, of such multiple approaches to represent results is easily
understandable. If a general comparison is intended, especially if the purpose is to elect a
specific procedure or variant of the same method, then an overall perspective may be
needed for the sake of simplicity and clearness of interpretation, making use of
statistically-based reduction. However, the Bridge Index that is computed over all
configurations and intensity levels will certainly conceal significant information that may
help to clarify general results. The observation of prediction in greater detail, from
intensity level or bridge configuration point of view, or even across different locations
inside a specific bridge, is the mean to better understanding and warranty that
miscellaneous judgement is not taken.

5.34
 Seismic Demand

5.4.2 Preliminary evaluation

5.4.2.1 Capacity Spectrum Method

Figure 5.8 illustrates the graphical results for one of the short irregular configurations
(A213) for the different versions of the method, at the end of the iterative procedure, at two
distinct levels of seismic intensity. The plots are in the ADRS format (Acceleration
Displacement Response Spectrum) and two distinct pushover curves are presented due to
the two possible reference nodes. The response spectrum, in turn, is reduced according to
ATC-40 or FEMA-440 and intersected with the two possible capacity curves, rendering
four performance points.

1.4 1.4
A213 (Low Intensity Level) A213 (High Intensity Level)
1.2 1.2

1 1
Acceleration

Acceleration

0.8 0.8

ATC40 - Demand Spectrum - central node ATC40 - Demand Spectrum - central node
0.6 0.6
ATC40 - Demand Spectrum - max disp node ATC40 - Demand Spectrum - max disp node
FEMA440 - Demand Spectrum - central node FEMA440 - Demand Spectrum - central node
0.4 FEMA440 - Demand Spectrum - max disp node 0.4 FEMA440 - Demand Spectrum - max disp node
Capacity Curve - central node Capacity Curve - central node
Capacity Curve - max disp node Capacity Curve - max disp node
0.2 0.2

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
Displacement Displacement

Figure 5.8 – CSM performance points for bridge A213: low and high intensity levels.

It is immediately noticeable that the influence of the reference node position in the capacity
curve is not outstanding. The curves tend to move away from each other as the structure
goes further into the nonlinear range. The performance points will be, thus, more or less
different depending on the region where the demand intersects the capacity. On the other
hand, the effect caused by the employment of the new damping equations proposed in the
FEMA-440 report is quite more pronounced, as well as the corresponding spectral
reduction, which is clearly observed in the plots, especially for high intensity levels.
Therefore, regarding this issue, noteworthy differences in the predictions are expected.

Figure 5.9 and Figure 5.10 show values of Bridge Index and Standard Deviation, with
respect for the median BI, for each intensity level, which is the median across the entire set
of bridges.

5.35
Seismic Demand

2 2
ATC40 - central node
Displacements Piers shear forces
ATC40 - max disp node
1.5 1.5 FEMA440 - central node
FEMA440 - max disp node
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
2 2

Deck moments Abutment shear forces

1.5 1.5
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.9 – CSM median Bridge Index (BI) per intensity level.

The improvements introduced by the FEMA-440 report are clearly observed, especially in
the estimations of displacement, deck moments and abutment shear forces, where results
obtained using the spectrum scaling procedures suggested in FEMA-440 are much closer
to unity (which means NSP estimates equal to NDA predictions) than those obtained using
the ATC-40 equations. Nevertheless, both approaches seem to overestimate the equivalent
viscous damping, hence, the corresponding spectral reduction, as well, which renders
moderate to heavy underestimation of displacements. On the other hand, if shear forces
estimates at the piers are considered, the improvement introduced by FEMA-440
guidelines is barely noticeable, except for the lowest intensity levels, given that predictions
are generally very good, i.e., fairly matching the nonlinear dynamic analysis’ ones. The use
of the location of maximum displacement node as reference will generally yield higher
estimates for nearly all the parameters, thus, closer to unit BI ratios. Again, the exception is
the pier shear forces parameter, to which the difference in choosing one or other reference
node is imperceptible. Generally, performance tends to get better as the intensity level
increases, mainly for pier shear forces and slightly for the rest of the engineering demand
parameters.

5.36
 Seismic Demand

0.5 0.5
ATC40 - central node
Displacements Piers shear forces
ATC40 - max disp node
0.4 0.4
FEMA440 - central node
FEMA440 - max disp node

Median STD
Median STD

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
0.5 0.5

Deck moments Abutment shear forces

0.4 0.4

Median STD
Median STD

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.10 – CSM median Standard Deviation (STD) per intensity level.

Looking at the method’s behaviour from the dispersion point of view, a tendency for better
predictions to be associated to higher scatter levels can be noticed. The FEMA-440
approach presents larger standard deviation with respect to the median BI, normally
oscillating between 0.2 and 0.3, or reaching levels of 0.4, if deck displacements are
considered. The classical approach barely exceeds 0.2 for all the parameters. Shear force
estimates present slightly lower dispersion, barely exceeding 0.2, showing no clear
difference between classical or FEMA-440 approaches, for the case of the piers. The
version that uses the maximum displacement node as reference can be seen as usually more
scattered, which may be due to the fact that its physical location is not constant as the
centre of mass of the deck is. Regarding the dispersion behaviour with increasing intensity,
it is generally either constant or even decreasing, which may be justified with the larger
estimates for strong ground motion intensity being less sensitive to fluctuation of results
coming from distinct locations or configurations.

The influence of structural geometry in the results may be scrutinized by plotting the Bride
Indexes according to the structural configuration. Such representation is presented in
Figure 5.11 considering the median BI over all the intensity levels.

5.37
Seismic Demand

ATC40 - central node

2 ATC40 - max disp node
Displacements FEMA440 - central node
Median BI

FEMA440 - max disp node

1

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

2
Piers shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Median BI

Deck moments
1

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Abutment shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration

Figure 5.11 – CSM median Bridge Index (BI) per bridge configuration.

The main observations, made according to the plot of results with varying intensity, seem
to stand, putting the FEMA-440 version as the most accurate one even if the improvement
coming from the use of the deck’s centre of mass node as a reference does not appear so
general, being more eloquent for the case of long, irregular configurations. Curiously,
lower discrepancy between different versions for the method seems to also occur for
irregular configurations, which denotes the effective improvement carried in by the
FEMA-440 guidelines, given that, for regular configurations, other variables are
eliminated. Regarding shear estimates, the underestimation is higher for irregular
configurations. Having in mind that the method is particularly recommended for regular,
simple structures, vibrating predominantly in the 1st mode, such scenario seems to indicate
that the method is not that unstable for non regular structures.

According to what has been exposed, in subsequent applications, the CSM will be
employed considering its FEMA-440 version (notwithstanding the slender increase in
dispersion), together with the maximum displacement node as reference.

5.4.2.2 N2 Method

Figure 5.12 shows the different performance points, given by displacement targets in the
N2 procedure, for the same chosen A213 configuration for example, two intensity levels,

5.38
 Seismic Demand

low and high, at the end of the EC8-proposed iterative procedure. The plot is in the D*-V*
format (equivalent SDOF system displacement and base shear) and the four distinct
pushover curves correspond to the crossed possibilities of load pattern and reference node.

4 4
x 10 x 10
6 6

5 5

4 4

V*
V*

3 3

A213 (Low Intensity Level) A213 (High Intensity Level)

2 2

Unif - central node Unif - central node

Unif - max disp node Unif - max disp node
1 1
Modal - central node Modal - central node
Modal - max disp node Modal - max disp node

0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
D* D*

Figure 5.12 – N2 performance points for bridge A213: low and high intensity levels.

The SDOF system bilinear elasto-plastic curves are plotted in order to highlight the
differences coming from the possible versions and, for both intensity levels, the divergence
is evident, especially for what concerns the modal pattern. The significance of the
reference node seems, in turn, to be lower, something that will surely be more visible in the
index results that are presented next.

Figure 5.13 and Figure 5.14 show values of Bridge Index and Standard Deviation, for each
intensity level, considering the entire set of bridges.

5.39
Seismic Demand

2 2
Unif - central node
Displacements Piers shear forces Unif - max disp node
Modal - central node
1.5 1.5 Modal - max disp node
Env - central node

Median BI
Median BI

Env - max disp node

1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
2 2

Deck moments Abutment shear forces

1.5 1.5
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.13 – N2 median Bridge Index (BI) per intensity level.

From the evaluation of vertical distances in BI plots, the observation that immediately
stands out is that dispersion among different variants of the method is not particularly high,
except for the deck bending moments. Indeed, whereas in the estimation of shear forces the
differences between the employment of uniform or first mode proportional load
distribution does not influence the results much, when deck displacements or moments are
considered instead the influence of pushover load shape is noticeable (and not always in
the same direction, which somehow explains why EC8 does not recommend the use of one
loading shape over the other). The use of the maximum modal displacement node as
reference yields, in general, worse results than the central node ones. The envelope shape,
on the other hand, seems to somehow “contain” the positive aspects of the two
EC8-recommended distributions, leading to better BI results throughout all response
parameters.

There is a generalized heavy underestimating trend for deck moments and abutments
shears, regardless the considered variant, for all the intensity levels, and for deck
displacements for lower nonlinearity. Deck displacements and piers shear forces estimates
are, therefore, considerably more reliable. Regarding behaviour of predictions with the
increase in intensity level, highly nonlinear stages seem to improve the accuracy of deck

5.40
 Seismic Demand

moments and abutment shear results (even if not sufficiently) and lead to slight
overestimation of deck displacements.

0.5 0.5
Displacements Piers shear forces Unif - central node
Unif - max disp node
0.4 0.4
Modal - central node
Modal - max disp node
Median STD

Median STD
0.3 0.3 Env - central node
Env - max disp node

0.2 0.2

0.1 0.1

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
0.5 0.5
Deck moments Abutment shear forces
0.4 0.4
Median STD

Median STD
0.3 0.3

0.2 0.2

0.1 0.1

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.14 – N2 median Standard Deviation (STD) per intensity level.

Together with the scatter within variants of the procedure, the dispersion with respect to
the median BI results may be evaluated making use of STD plots. Figure 5.14 shows that
dispersion levels are higher for deck displacements, with respect to the other parameters,
reaching maximum values of 0.4, substantiating the trend of better BI estimates associated
to higher scatter. On the other hand, shear forces at the abutments and deck bending
moments usually do not go beyond standard deviations of 0.3 and pier shears of 0.25. With
respect to the different variants, the use of the 1st mode proportional load pattern results in
higher scatter while the opposite trend is precisely found for the uniform load shape. The
envelope technique is expectedly in between the two of them, presenting intermediate STD
quantities. The difference between the two reference node possibilities is not so obvious,
although a tendency for lower dispersion to be associated to the maximum modal
displacement node might be found. Again the connection between good BIs and higher
STDs seems to stand. No great influence of the seismic intensity on the dispersion occurs,
despite a soft increase for displacements and bending moments and the reverse in shear
forces.

5.41
Seismic Demand

Figure 5.15 presents BI results in line with bridge configuration, which allows a better
understanding of the origin the dispersion.

Unif - central node

Unif - max disp node
Modal - central node
Modal - max disp node
2
Env - central node
Displacements
Median BI

Env - max disp node

1

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Piers shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Deck moments
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Abutment shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration

Figure 5.15 – N2 median Bridge Index (BI) per bridge configuration.

In fact, the N2 method may be considered as typically expected behaving NSP, in the sense
that estimates are substantially better for the case of regular configurations, 222, 232 and
2222222, throughout all the parameters but particularly for displacements. Within each
structural type, even if more visible for regular ones, the difference between the
approaches is not so relevant, a pertinent fact denoting that the scatter comes more from
the different configurations rather than the different versions of the procedure.
Nevertheless, estimates for irregular configurations, whether in underestimating or
overestimating fashion, are not completely unacceptable. No significant difference is found
between long or short bridges, apart from slightly less underestimation in the latter.

In agreement with the main conclusions herein drawn, the use of the deck’s central node as
reference point leads to better predictions hence this will be adopted on subsequent
applications, together with the envelope pushover loading shape.

5.42
 Seismic Demand

5.4.2.3 Modal Pushover Analysis

Figure 5.16 shows the different performance points, yielded by the MPA, for the chosen
A213 configuration, two intensity levels, low and high, at the end of each of the modal
analysis. For the selected bridge, the first and second modes only have been considered
relevant. In the end, the quantities will be SRSS combined. The plots are in the
Displacement-Force format (equivalent SDOF system displacement and base shear) and
the two distinct pushover curves correspond to the two reference node location
possibilities.

16 16

14 14

12 12

10 10
Force

Force

8 8

6 A213 (Low Intensity Level) 6 A213 (High Intensity Level)

st st
Max disp node - 1 mode Max disp node - 1 mode
4 st 4
st
Central node - 1 mode Central node - 1 mode
nd nd
2
Central node - 2 mode Max disp node - 2 mode
2
nd nd
Max disp node - 2 mode Central node - 2 mode
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
Displacement Displacement

Figure 5.16 – MPA performance points for bridge A213: low and high intensity levels.

Modal Pushover Analysis has been taken as dependent on the reference node location only
and the main observation that can be made is that the influence is slight. For both intensity
levels the difference in the performance points is minimal, especially for the first mode,
even though the pushover curves corresponding to different reference nodes are equally
distant for both vibration modes.

Figure 5.17 and Figure 5.18 show values of median Bridge Index and Standard Deviation
for each intensity level, concerning the entire set of bridges.

5.43
Seismic Demand

2 2
Central node
Displacements Piers shear forces Max disp node

1.5 1.5
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
2 2

Deck moments Abutment shear forces

1.5 1.5
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.17 – MPA median Bridge Index (BI) per intensity level.

The global picture for the results of this method is not much different from what has been
seen in the other procedures: deck displacements and shear forces at the piers are better
captured, leading to BIs closer to unit with respect to deck bending moments and shear
forces in the abutments, which are underestimated for all the intensity levels, heavily for
the lower ones. For the two mentioned better estimated parameters, the centre of mass of
the deck as reference seems to work better, given that the use of the maximum
displacement node as reference typically overestimates NDA results. The general tendency
is that one, indeed, to have increasing predictions as seismic input intensity increases. In
fact, contrarily to what has been observed in other procedures, overestimation of shear
forces at the piers occurs for higher intensity, which may indicate an exaggerated effect of
the modal combination that is carried out. Furthermore, the estimates of deck bending
moments and shear at the abutments are poor, independently from the reference node
location of seismic intensity.

Looking at scatter of results in terms of dispersion relatively to the median BI, again, lower
STDs are encountered for generalized shear predictions (modest 0.1 or less), whereas the
deck displacements and bending moments predictions are associated to standard deviations
of 0.2 or 0.25. The distinction in terms of reference node type is not immediately visible,

5.44
 Seismic Demand

although the use of maximum displacement node as reference is affected by larger scatter.
Such output is constant and no surprises arise when the structures step in nonlinear
behaviour. No significant pattern is identified with the changes in nonlinearity. Higher
intensity levels do not necessarily yield higher dispersion.

0.3 0.3

Displacements Piers shear forces Central node

0.25 0.25 Max disp node

0.2 0.2
Median STD

Median STD
0.15 0.15

0.1 0.1

0.05 0.05

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
0.3 0.3

Deck moments Abutment shear forces

0.25 0.25

0.2 0.2
Median STD

Median STD

0.15 0.15

0.1 0.1

0.05 0.05

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.18 – MPA median Standard Deviation (STD) per intensity level.

2 Central node
Displacements
Median BI

Max disp node

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

2
Piers shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

2
Deck moments
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

2
Abutment shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration

Figure 5.19 – MPA median Bridge Index (BI) per bridge configuration.

5.45
Seismic Demand

MPA results in terms of bridge configuration are not straightforward to interpret, given
their apparent irregularity or somewhat absence of rule, as showed by Figure 5.19. The
first impression, when looking at displacements, is that the choice for a reference node is
not obvious: whilst for the majority of the bridges, the central is preferable, for some
others, the maximum modal displacement node stands out. With respect to type of
abutments, the performance of type-B abutment bridges seems definitely to be less well
captured, especially for the regular ones. The overestimation found for shear forces at the
piers comes from the long type-A bridges and the type-B ones. The deck bending moments
estimates are generally poor and the same stands for abutments shear, where the procedure
behaves quite inferiorly for long configurations.

On subsequent applications of the method, based on the intensity level results, the centre of
mass of the deck reference node modality was adopted.

5.4.2.4 Adaptive Capacity Spectrum Method

ACSM exposure of results will exceptionally feature the global representation plots, in
order to provide higher clearness, as well as additional useful criteria, to the analysis of the
eleven versions. Figure 5.20 shows the graphical results for the performance points
corresponding to the different possible spectral reductions, at the end of the iterative
procedure for the configuration A213, at low and high intensity levels.

12000 12000

A213 (Low Intensity Level) A213 (High Intensity Level)

10000 10000

8000 5% Demand Spectrum 8000 5% Demand Spectrum

Adaptive Capacity Curve Adaptive Capacity Curve
Acceleration
Acceleration

ATC40-EC8Ram ATC40-EC8Ram
6000
ATC40-LynChang 6000 ATC40-LynChang
TakKow-EC8Ram TakKow-EC8Ram
TakKow-LynChang TakKow-LynChang
TakGS-EC8Ram TakGS-EC8Ram
4000 4000
TakGS-LynChang TakGS-LynChang
KowDwa-EC8Ram KowDwa-EC8Ram
KowDwa-LynChang KowDwa-LynChang
2000 Priestley 2000 Priestley
Miranda2000 Miranda2000
VidFajfFish VidFajfFish
0 0
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800
Displacement Displacement

Figure 5.20 – ACSM performance points for bridge A213: low and high intensity levels.

The plots are in the ADRS format (Acceleration Displacement Response Spectrum) and
contain one single pushover/capacity curve, the adaptive, the original response/demand

5.46
 Seismic Demand

spectrum for 5% damping and eleven performance points, coming from the intersection of
each reduced response spectrum, not plotted for the sake of simplicity, with the capacity
curve.

For the low intensity level no relevant hysteric damping and energy dissipation are
expected, which becomes clear from the concentration of performance points near the
intersection with elastic response spectra. Nevertheless, some approaches, ATC-40 and
ductility based ones, present already considerable reduction, which corresponds to higher
prediction for equivalent viscous damping (elastic plus hysteretic). For higher nonlinearity
levels, even if not so obvious, the same distinction is observed.

Figure 5.21 represents graphically the global median BI values and Figure 5.22 the
corresponding STDs, both referring to each spectrum scaling modality.

2 2
Displacements EC8 Piers shear forces EC8
LinChang LinChang
1.5 1.5
Median BI

Median BI

1 1

0.5 0.5

0 0
ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish

2 2
Deck moments EC8 Abutments shear forces EC8
LinChang LinChang
1.5 1.5
Median BI

Median BI

1 1

0.5 0.5

0 0
ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish

Figure 5.21 – ACSM global median Bridge Index (BI).

Considering the results shown, the first conclusion is that it is the employed damping
model, rather than the chosen damping-dependent spectral reduction equation, that
conditions the results; there is essentially no visible difference between LinChang and
EC8Ram results. Generally, it seems that the approaches TakKow, DwaKowNau and
Priestley lead to the best global indexes, especially for what displacements are concerned,
where major differences are found, with a global median BI fairly close to one and global
median STD, around 0.15, that is not excessive. On the other hand, the two ductility-based

5.47
Seismic Demand

approaches (Mir2000 and VidFajFish) and the damping-based ATC-40 method show the
worst results when compared to nonlinear dynamic analyses. This trend seems to
contradict opposite results, found in previous studies (Bertero et al., 1991; Bertero, 1995;
Reinhorn, 1997; Chopra and Goel, 1999; Fajfar, 1999, 2000), which state that inelastic
spectra-based methods yield better results with respect to their elastic highly-damped
counterparts. However, such findings come from the use of a different NSP from the
Adaptive Capacity Spectrum Method (ACSM), which, within the elastic highly-damped
methods, is expected to yield more consistent results in the assessment of the seismic
response of bridges (Pinho et al., 2007; Pinho et al., 2009) when compared to other NSPs,
and this may explain the difference.

0.3 0.3
Displacements EC8 Piers shear forces EC8
0.25 LinChang 0.25 LinChang

0.2 0.2
Median STD

Median STD

0.15 0.15

0.1 0.1

0.05 0.05

0 0
ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish

Deck moments EC8 Abutments shear forces EC8

0.25 LinChang 0.25 LinChang

0.2 0.2
Median STD

Median STD

0.15 0.15

0.1 0.1

0.05 0.05

0 0
ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish

Figure 5.22 – ACSM global median Standard Deviation (STD).

The underpredicting trend of ATC-40 is probably due to the well known overprediction
that such approach carries out within the equivalent damping estimation in structures
which will not exactly exhibit elastoplastic behaviour.

Similar tendency is found when ductility based reduction factors are used. Mir2000
approach has been primarily developed for elastoplastic behaviour but has also been
widely verified for other models such as Takeda or Clough, proving to work well for
periods longer than 1.2 seconds, which is not the case of the considered bridge structures
with maximum periods of about 0.8 seconds. One would therefore expect overprediction,

5.48
 Seismic Demand

as found by Miranda and Ruiz Garcia (2002) for short periods. However, the low ductility
levels achieved for the considered bridge (see Figure 5.23) structures have lead to barely
unitary displacement modification factors. VidFajFish approach, on the other hand, is
known to be highly dependent on period and ductility for the short period region, which
may apply to the selected bridge structures; for this region, the R factor increases linearly
with increasing period, remaining constant for the rest of the periods, which may cause
some excessive reduction, and, hence, underprediction, given that periods tend to be low,
as early stated, for this sort of structures.

4
IL 0.5
Miranda2000 reduction factor approach
IL 0.75
3.5
IL 1.0
IL 1.5
3 IL 2.0
IL 2.5
2.5
Ductility

1.5

0.5

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

Bridge configurantion

Figure 5.23 – Ductility level achieved using Miranda2000 reduction factor approach.

The Gulkan and Sozen approach, based on the Takeda hysteretic model, is overpredicting
displacements. According to Equation (5.5) and Figure 5.3, it is the one with lowest
equivalent damping prediction, among all the approaches, for ductility levels not higher
than approximately 3. Adding to this the fact that most cases of application correspond to
ductility levels equally distributed between 1 and 3, (Figure 5.24 and Figure 5.25) one may
explain the overprediction by the underestimation of damping-based energy dissipation.

The superiority found when using TakKow, DwaKowNau or Priestley approaches is

firstly, in opposite to the others, due to the intermediate nature on estimating the equivalent
viscous damping. Additionally, the latter two, which are the most recent, have been
particularly developed and therefore, more suitable, to the structural behaviour where
bridges fit in. DwaKowNau’s equation, for instance, depends on the effective period which
is another advantage, whereas Priestley’s expression refers exactly to bridge piers.

5.49
Seismic Demand

5
TakGS damping approach - EC8 RF IL 0.5
4.5 IL 0.75
IL 1.0
4 IL 1.5
IL 3.0
3.5
IL 3.5
3
Ductility

2.5

1.5

0.5

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

Bridge configurantion

Figure 5.24 – Ductility level achieved using TakGS damping approach and EC8 reduction factor.

5
TakGS damping approach - LinChang RF IL 0.5
4.5 IL 0.75
IL 1.0
4 IL 1.5
IL 3.0
3.5 IL 3.5

3
Ductility

2.5

1.5

0.5

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge configurantion

Figure 5.25 – Ductility level achieved using TakGS damping approach and LinChang
reduction factor.

As far as the type of parameter is concerned, there is high coherence of results within
different variants for shear forces at piers whereas the main differences come from the rest
of parameters. A general heavy underestimating trend may be again observed for the deck
bending moments and abutments shear. Dispersion levels are rather low for shear
predictions (and independent of the reduction factor model) and higher for the deck
engineering demand parameters, with maximum STDs of 0.2 or 0.15 for deck bending
moments or displacements. Shear force predictions are associated to version-independent
low scatter without relevant accuracy loss.

All global observations care for scrutiny and, therefore, Figure 5.26 shows, at each
intensity level, the median Bridge Index over the 14 bridge configurations. Given that no
relevant difference has been found between EC8 and LinChang spectral reduction factors,

5.50
 Seismic Demand

for the subsequent plots, and for the sake of simplicity, only the EC8-emplyed RF version
of each ACSM variant has been represented.

2 2
Displacements Piers shear forces
1.5 1.5
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
2 2
Deck moments Abutments shear forces
1.5 1.5
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish

Figure 5.26 – ACSM median Bridge Index (BI) per intensity level.

Observing this new set of results, it turns out apparent that amongst the three approaches
previously mentioned to be performing superiorly, the one proposed by Priestley et al.
(2007) seems to be slightly better, even if only marginally, very closely followed by that of
Dwairi et al. (2007). Such superiority is more evident for deck displacements, deck
bending moments and shear at the abutments. Dwairi et al. approach seems then to
introduce a not relevant dependency of the coefficient CST on the effective period. In
addition, it is also noted that there is a general tendency for the displacements and shear
forces predictions to improve with increasing intensity level, probably due to the fact that
for low levels of nonlinearity, thus dissipation, current damping and RF relationships
overestimate the reduction. For the case of piers shear forces dependence on spectral
reduction approach is practically inexistent. As expected (see Figure 5.21), ATC-40
damping model seems to overestimate damping, leading to underestimated displacements
at each intensity level.

5.51
Seismic Demand

0.3 0.3
Displacements Piers shear forces
0.25 0.25

0.2

Median STD
0.2
Median STD

0.15 0.15

0.1 0.1

0.05 0.05

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
0.3 0.3
Deck moments Abutments shear forces
0.25 0.25
Median STD

0.2

Median STD
0.2

0.15 0.15

0.1 0.1

0.05 0.05

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish

Figure 5.27 – ACSM median Standard Deviation (STD) per intensity level.

As for the standard deviation, Figure 5.27, this did increase with growing intensity, for the
case of deck displacements, to nearly twice the initial one. The differences between
variants arise for deck displacements and bending moments only and again, generally, the
worst modalities predicting Bridge Index, ATC-40 and ductility-based ones, have lower,
but not considerably different, dispersion levels. In other words, the higher scatter
associated to Priestley’s approach, or even very similar TakKow and DwaKowNau, does
not seem to compromise the better median BI estimates. Shear predictions, whether at the
piers or abutments level, stand for the lowest scatter levels, barely reaching 0.1.

Figure 5.28 shows the median BI across all intensity levels, for each bridge configuration.
The response predictions do not appear to be very bridge-dependent, and the
observations/conclusions previously drawn hold for the majority of configurations; e.g.,
the Priestley approach consistently leads to the closest-to-unity BIs. It is also observed that
major variations in the predictions, according to the different variants, occur for irregular,
long bridges. Notwithstanding such variation, the displacements and shear forces at the
piers are fairly well predicted, whilst deck moments and abutments shear forces estimates
are invariantly poor. BI ratios are better in the case of regular bridge configurations and
there is smaller dispersion with respect to the spectral reduction approach. An
overestimating trend for short bridges and underestimating for longer ones can also be

5.52
 Seismic Demand

observed, with the latter cases leading also to larger scatter. These observations generally
stand for all the engineering demand parameters herein considered.

2
Displacements
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

2
Piers shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Deck moments
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Abutments shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration
ATC40 TakKow TakGS DwaKowNau Priestley Miranda2000 VidFajFish

Figure 5.28 – ACSM median Standard Deviation (STD) per intensity level.

5.4.3 Comparative study results

Having identified the “optimum configuration” of all NSPs considered here, it is now
possible to proceed with the parametric comparison of the four approaches, with the
purpose of emphasizing relative advantages and disadvantages and, eventually, coming up
with suggestions for possible preferred choices, if any. The study is again carried out on
the basis of Bridge Index and Standard Deviation comparison, starting from the somewhat
global perspective, where the entire set of results (for al bridges and for all intensity levels)
are first considered together, and then sub-structured in terms of seismic input intensity and
bridge model.

5.4.3.1 Global results

Recalling Section 5.4.1, a global results overview consists in the computation of the bridge
index per NSP over all the 14 bridges and 6 intensity levels and is extremely useful, since
it provides a general picture of the results, enabling an easier interpretation if the choice for
one of the procedures is intended. This representation of results caters for (i) comparison

5.53
Seismic Demand

with dynamic analyses (it is recalled that BI represents the ratio between NSP and NDA
results), (ii) relative comparison of the accuracy of the different NSPs, and (iii)
appreciation of the results dispersion (plots include overall median BI for every method, in
filled markers, and mean BI ± mean STD error bars).

2 2
Median BI, Mean BI ± Mean STD

Median BI, Mean BI ± Mean STD

Displacements Piers shear forces

1.5 1.5

1 1

0.5 0.5

0 0
CSM N2 MPA ACSM CSM N2 MPA ACSM

2 2
Median BI, Mean BI ± Mean STD

Median BI, Mean BI ± Mean STD

Deck moments Abutments shear forces

1.5 1.5

1 1

0.5 0.5

0 0
CSM N2 MPA ACSM CSM N2 MPA ACSM

Figure 5.29 – Global Bridge Index and Standard Deviation.

From the observation of Figure 5.29, it is conspicuous that all nonlinear static procedures,
with the exception of Capacity Spectrum Method, are able to predict displacement
response with effectively good accuracy, evidencing also reasonable dispersion levels. The
traditional CSM underpredicts the NDA estimates, whereas N2 and the more recent
approaches, MPA and ACSM, yield global median unitary ratios, bearing an increase in
the scatter to do so, though. Shear forces at the piers are accurately estimated by all
procedures, despite a slight overprediction of MPA, coupled with highly satisfying
standard deviation. On the contrary, deck moments, mostly, and shear forces at the
abutments are underestimated in relatively heavy fashion by all methods, even if the scatter
levels are relatively low. Still, the inferiority of CSM stands. Regarding dispersion levels,
the observed tendency is to have low scatter in deck bending moments and shear force
predictions (maximum values of 0.5) except for N2, which has approximately the double
of that value. Higher STD values are obtained when predicting displacement. N2 may be
considered, in most of the situations, the method with the highest dispersion levels, whilst

5.54
 Seismic Demand

the opposite part is somewhat assumed by CSM. Another interesting observation is the fact
that, if global mean indexes had been computed, there would have been some slight
improvement, particularly for ACSM.

5.4.3.2 Intensity level results

At each intensity level, the median Bridge Index over the 14 bridge configurations is
computed for each of the four NSPs (see Figure 5.30). The results not only confirm the
observations made in the previous section, but also add some insight on how these may be
influenced by the intensity of the input motion.

2 2
CSM
Displacements Piers shear forces N2
MPA
1.5 1.5
ACSM
Median BI

Median BI

1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level
2 2
Deck moments Abutments shear forces
1.5 1.5
Median BI

Median BI

1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.30 – Median Bridge Index per intensity level.

When global indexes showed, in Figure 5.25, a large similarity between the four
procedures, an intensity level results’ overview will certainly clear out some changes in
their behaviour, especially at high nonlinear stages. It is observed that important variations
are observed in displacement response estimates, with slight underprediction at lower
intensity levels, for all NSPs, evolving to overprediction at high intensity, for MPA and
N2, underprediction for CSM, while ACSM manages to keep a steadier closeness to NDA
results. These differences between the four methods may be justified with the fact that

5.55
Seismic Demand

major conceptual differences exist between them, such as the reference node choice or the
use of an envelope of different displacement shapes for the case of N2.

Regarding shear forces, as the seismic intensity increases, the accuracy in the predictions at
the bridge piers is constant, with the exception of MPA, which overestimates NDA results.
This will probably have to do with overrated higher mode effects, which in turn become
more important as the intensity of the seismic action increases (because the fundamental
period elongates, hence its spectral amplification diminishes, increasing the relative
importance of higher modes). Apart from a general global underestimating showed by
CSM, the rest of the methods behave similarly. Some points of similarity can, however, be
found between N2 and MPA or CSM and its adaptive version, given that both share
ductility or damping based spectral reduction, respectively.

0.6 0.6
Displacements Piers shear forces CSM
0.5 0.5 N2
MPA
0.4 0.4 ACSM
Median STD

Median STD

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Deck moments Abutments shear forces

0.5 0.5

0.4 0.4
Median STD

Median STD

0.3 0.3

0.2 0.2

0.1 0.1

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5
Intensity Level Intensity Level

Figure 5.31 – Median Standard Deviation per intensity level.

For what concerns the dispersion of the results, this did not prove to be much dependent on
intensity level. Indeed, global conclusions are confirmed by Figure 5.31, where STD levels
oscillate in line with the nonlinear static procedure being employed, rather than the seismic
intensity variation. Traditional methods, CSM and N2, yield the predictions with largest
dispersion, reaching 0.4 for deck displacements and 0.3 for the other parameters, whereas

5.56
 Seismic Demand

to MPA and ACSM correspond STDs of 0.2 or 0.1 for deck or shear quantities,
respectively.

5.4.3.3 Bridge configuration results

Herein, for each bridge configuration, the median Bridge Index and Standard Deviation
across the 6 intensity levels is plotted considering each of the four nonlinear static
procedures (see Figure 5.32 and Figure 5.33).

As expected, the response predictions do appear to be very bridge-dependent, even if the

observations/conclusions previously drawn still hold for the majority of configurations.
The best displacement response estimates (in terms of BIs being close to unity, and STDs
being close to zero) are obtained for the regular bridge configurations (e.g., 222, 232 and
2222222), as one would expect, though good displacement estimates are also obtained for
semi-regular and irregular bridges, especially if one considers results coming from higher
modes considering procedures MPA and ACSM; this certainly constitutes good news for
NSPs and their application to assessment of bridge response when the focus is on
deformations.

CSM N2 MPA ACSM

2
Displacements
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Piers shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Deck moments
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
2
Abutments shear forces
Median BI

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration

Figure 5.32 – Median Bridge Index per bridge configuration.

5.57
Seismic Demand

CSM and N2 are many times the most underpredicting ones. An overestimating trend for
short bridges and an underestimating one for longer ones can also be observed in the
results. The same gist, or maybe even more evident, may be encountered for the rest of
parameters where, however, general underestimation occurs. The overestimation of MPA
for the shear predictions at the piers occurs mostly for irregular and type-B abutments
bridges.

The largest scatter is, in general, associated to both long and irregular bridges, an aspect
that is extremely noticeable from the observation of Figure 5.33, mainly in displacements
and abutments shear forces. As for the bridge sort of abutments, no relevant differences
between bridges with abutments of type A (continuous deck-abutment connections) or type
B (deck supported on linear pot bearings) is noticeable. Such finding stands for the
abutments shear predictions, as well as for the remaining parameters.

CSM N2 MPA ACSM

Displacements
Median STD

0.5

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Median STD

1 Piers shear forces

0.5

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Median STD

1 Deck moments

0.5

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111

Abutments shear forces

Median STD

0.5

0
A123 A213 A222 A232 A2222222 A2331312 A3332111 B123 B213 B222 B232 B2222222 B2331312 B3332111
Bridge Configuration

Figure 5.33 – Median Standard Deviation per bridge configuration.

The observation of results at a bridge configuration level enabled the recognition of the
considerable variation that such extensive, statistical comparative study features. The use
of global indexes is extremely useful to draw general conclusions but it does naturally
mask the influence of the different structural characteristics. In order to investigate the
extent of such influence, the results have been plotted again, in the global (Figure 5.34) and
intensity level (Figure 5.35) fashion, filtering the configurations according to the different

5.58
 Seismic Demand

categories: regular (REG), irregular (IRREG), short, long, type-A and type-B abutments.
Given that shear forces at the piers have been estimated with high accuracy by all the
procedures, across all the intensity levels and bridge configurations and heavily
underestimated prediction where obtained for the deck moments and abutments shears,
only the displacements predictions were scrutinized.

The individuation of the results in different categories shows, at first glance, that the
variables that most affect the global predictions (and corresponding standard deviation) are
the length of the bridge and the type of abutments, given that the distinction between
regular and irregular configurations does not yield BIs significantly different from the
results all together. Long configurations, which are certainly more affected by higher
modes, are less well captured by CSM and N2 procedures. Furthermore, even if global
median BI is not necessarily much worse, with respect to short configurations, such
outcome corresponds to an extremely higher standard deviation. The same happens
between regular and irregular configurations: the former feature more accurate predictions
and, at the same time, much lower uncertainty, which certainly highlights its higher
suitability to be analysed through the use of NSPs.
Median BI, Mean BI± Mean STD

Median BI, Mean BI± Mean STD

2 2
CSM N2
1.5 1.5

1 1

0.5 0.5

0 0
ALL REG IRREG SHORT LONG type-A type-B ALL REG IRREG SHORT LONG type-A type-B
Median BI, Mean BI± Mean STD

Median BI, Mean BI± Mean STD

2 2

MPA ACSM
1.5 1.5

1 1

0.5 0.5

0 0
ALL REG IRREG SHORT LONG type-A type-B ALL REG IRREG SHORT LONG type-A type-B

Figure 5.34 – Global Displacements Bridge Index and Standard Deviation according to bridge
category.

5.59
Seismic Demand

Regarding the type of abutment, a constant trend throughout all the procedures could not
be found although, generally, type-B abutments bridges are underpredicted together with
lower corresponding STD.

ALL REG IRREG SHORT LONG type-A type-B

2 2
CSM N2
1.5 1.5
Median BI

Median BI
1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5

2 2
MPA ACSM
1.5 1.5
Median BI

Median BI

1 1

0.5 0.5

0 0
0.5 0.75 1 1.5 2 2.5 0.5 0.75 1 1.5 2 2.5

Figure 5.35 – Median Displacement Bridge Index per intensity level according to bridge category.

Another interesting conclusion, which is quite visible from Figure 5.35 as well, is the fact
that classical procedures, CSM and N2, feature higher variation across intensity levels and
bridge structural characteristics, with respect to recent, improved approaches, such as MPA
and ACSM.

A different perspective is provided by Figure 5.36, in which results are compared, by NSP,
side by side, in terms of the distinct defined characteristics: regularity, length and abutment
type. The relevancy of each of the categories to the relative performance of the nonlinear
static procedures becomes even more visible and, clearly, from a median point of view,
rather than regularity, length and abutment type are the factors that influence the most the
results (extremely higher scatter within results for irregular configurations should,
however, be kept in mind). Indeed, depending on the type of bridges being taken into

5.60
 Seismic Demand

account, the methods can move from under- to overestimation, which strengthens the
importance of looking at the results from a category perspective. Another immediate
observation is that it becomes even clearer that higher modes accounting NSPs, MPA and
ACSM, perform superiorly than the other two.

2
Median BI, Mean BI± Mean STD

1.5

0.5

0
CSM reg CSM irreg N2 reg N2 irreg MPA reg MPA irreg ACSM reg ACSM irreg

2
Median BI, Mean BI± Mean STD

1.5

0.5

0
CSM short CSM long N2 short N2 long MPA short MPA long ACSM short ACSM long

2
Median BI, Mean BI± Mean STD

1.5

0.5

0
CSM type-A CSM type-B N2 type-A N2 type-B MPA type-A MPA type-B ACSM type-A ACSM type-B

Figure 5.36 – Global Displacements Bridge Index and Standard Deviation according to bridge
category.

5.61
Seismic Demand

5.5 Conclusions

This chapter dealt essentially with the application of static methods to estimate the
response, or demand, of the structure when subjected to earthquake loading. The
motivation for using such procedures has come mostly from two aspects: their undeniable
simplicity without expected accuracy loss, constituting a viable alternative to nonlinear
dynamic analysis, and their sense of opportunity in the present-day context, for what
concerns the seismic assessment and design of structures, given that dynamic analysis
seems far from being univocally recommended by codes and guidelines as the main
response prediction tool. Additionally, if buildings have consistently been considered and
validated in the past as object of successful NSP application, bridges still represent, to
some extent, unexplored field, regarding their adequacy to the use of the same NSPs.

The ability of four commonly used Nonlinear Static Procedures in predicting the structural
response of bridges subjected to earthquake action has therefore been appraised and
compared; two pioneering “classical” methods (CSM and N2) were considered along with
two of their more contemporary counterparts (MPA and ACSM). The evaluation was
systematically carried out over a relatively large number of structural configurations,
considering different response parameters and using several accelerograms scaled to a
number of intensity levels. A preliminary study, on the exact same basis, was also carried
out with a view to better understand the terms of each of the employed procedures as well
as to establish their optimum configuration. The following main observations could be
made.

The Capacity Spectrum Method has clearly benefited from the improvements
introduced in the FEMA-440 report, which allowed the attainment of superior
predictions, with respect to those obtained using the antecedent ATC-40 formulae.
The main modification introduced by the FEMA-440 guidelines concerned the
estimation computation of equivalent viscous damping, which was considered in
the traditional ATC-40 version as given by the parallelogram hysteresis loops. This
way of computing energy dissipation has been recognized as overstating, as well as
the Newmark-Hall spectral reduction factors that are used. Seems therefore easily
understandable that the new equations, more realistic, have moved CSM indexes
closer to unity. Such improvements, however, led to still moderately

5.62
 Seismic Demand

underestimated estimates of NDA results for all the considered engineering

demand. Having in mind that CSM is a 1st mode completely based procedure,
parameters such as shear forces or bending moments will probably be more
affected by the absence of superior modes contribution to the response. Dispersion
levels have proved to be substantial. The recommended choice of the centre of
mass of the deck as reference node has proven, globally, to be efficient, even if the
maximum modal displacement reference node has shown some marginal upgrading
for irregular configurations.

The N2 method is favoured by the consideration of an envelope pushover load

shape that bounds the two alternative load profiles suggested by EC8; first mode
proportional and uniform. Such outcome is due to the fact that none of the two
possible load patterns led to consistently better predictions. The envelope version,
however, does not avoid the need for two pushover analyses, which may be seen as
a disadvantage. For regular configurations, one of the load patterns, arbitrarily
chosen, is enough given that displacement estimates are fairly good, as well as the
shear forces ones, for the piers. On the other hand, for the rest of the parameters,
deck moments and shear at the abutments, results tend to be much worse. The
median to actually low scatter associated to the method is a positive aspect. The
demand, and corresponding spectral reduction, has been based on inelastic
smoothed design spectrum, something that has not contributed as decisively as
expected, when compared to overdamped spectra, for the accuracy of predictions. It
is also recalled that the optional iterative procedure, proposed by EC8 to optimize
the equivalent bilinearization process, was employed here and has significantly
contributed positively to the performance of the method. The proposal of such
feature on a non-compulsory basis seems, at least, arguable. Despite the little
difference, the reference node taken as the centre of the deck has worked better.

Modal Pushover Analysis is quite a strictly defined procedure, in the sense that few
windows are left open for improvement or testing. Based on the modal classical
theory, involves the repetition of the main steps for each single relevant mode and
considers spectral reduction as based on inelastic quantities (response spectrum or
displacement empirical ratios) or on Response History Analysis of the SDOF
modal systems. The latter was not tested given that it would imply comparable

5.63
Seismic Demand

effort to pure dynamic analysis of the MODF structure. The reference node location
was, hence, the only tested variable, and difficulties have been encountered to
distinguish the performance of one to the other alternative. The method seems not
to be sensitive to this issue. Underprediction for parameters other than deck
displacements is still found for deck moments and abutment shear forces. Pier shear
forces, on the other hand, come out overpredicted for high intensity levels, which
will probably have to do with the consideration of higher modes. Scatter levels are
still noticeable but the main drawback that can be pointed out, is the substantial
increase in the computational effort of the analysis, brought by the inclusion for an
additionally unclear number of relevant nodes. The user would therefore be faced
with a necessary evaluation of the compromise between accuracy/relevance of
including one more vibration mode, and consequent computation time increase.

The Adaptive Capacity Spectrum Method enables the consideration of higher

modes within a single pushover analysis. Moreover, the reference node choice is
unnecessary as well, given that a displacement-based, step-by-step updated,
equivalent location is obtained from the actual deformed shape of the structure. The
remaining source of debate is the spectral reduction and therefore eleven different
possible approaches for considering the energy dissipation capacity, by applying
spectral reduction factors, has been carried out. Nine of the spectrum scaling
schemes were damping-based whilst the other two were ductility-based. In general,
results have indicated that the choice of an appropriate spectral reduction method is
an important issue within the ACSM, given that considerable differences have been
found when employing different commonly employed approaches, especially for
what concerns estimates of deck displacements. In particular, in case of damping-
based reduction, the employed damping model proved to be much more relevant
than the chosen spectral reduction equation. In general, damping-dependent
methods perform better than their ductility-based counterparts. In particular the
proposal by Priestley et al. (2007) seemed to lead to the best predictions, very
closely followed by those of Kowalsky (1994) and Dwairi et al. (2007). On the
other hand, the ATC-40 damping model tends to overestimate structural
dissipation, leading to underestimation of displacement results. Typically, moments
and shear forces at the abutments were considerably underestimated, as it has been

5.64
 Seismic Demand

observed for all the other NSPs, which seems to indicate that shear amplification
coefficients should perhaps be introduced in NSP formulations. On the other hand,
with the employment of an appropriate spectral reduction factor, excellent response
displacement and pier shear forces estimates may be obtained with such nonlinear
static procedure.

Generally, with the exception of CSM, the NSPs proved to be able to predict displacement
response with relatively good accuracy for all sorts of bridge configurations (regular,
irregular, short, long, etc), something that certainly does lend some reassurance with
regards to the employment of such methodologies for assessing response displacements
and deformations. Given the actual tendency to move the focus of seismic analysis and
design from forces to displacements, such outcome is even more encouraging. As for the
rest of the parameters, the prediction of shear forces at the piers was definitely superior,
apart from some overestimating trend presented by MPA, leading to consecutive unitary
ratios to THA. On the other hand, deck moments and shear at the abutments have been
constantly underestimated, in heavy fashion. The assumption of the deck remaining elastic,
regardless the intensity level and the typically complex modelling of the abutments may
have lead to the poorness of the estimates. Dispersion levels were not negligible, though,
regardless the procedure and, thus, the interpretation of global median results should be
handled with care. Furthermore it has been observed that good median estimates were
typically associated to higher scatter.

Definitely, the endeavour that was carried out was extensive and covered a series of bridge
configurations and intensity levels, so, a statistical approach to the results was needed, in
order to look at the performance of each NSP at a distance that would ease conclusions.
However, at the same time, such reduction of indexes may work against thoroughness,
eventually leaving important information behind, which will constitute the main limitation
of such study. As a consequence, regular, irregular, short, long, type-A abutments and
type-B abutments configurations have been studied in deeper detail. The procedures
proved to be quite affected by changes in deck length and type of abutments, in median
ratios between nonlinear static and dynamic analysis and corresponding standard deviation,
which was much less for short bridges and type-B abutments. The same has been found for
the irregular bridges, featuring extremely high dispersion, when compared to the regular

5.65
Seismic Demand

ones. Moreover, it has been concluded that recent approaches (ACSM and MPA) present
lower uncertainty levels with respect to their application to bridges in a general fashion.

If a single NSP should be recommended over the rest, such choice would be based on two
key aspects. The first is that, undeniably, if a consistent and systematic use of a Nonlinear
Static Procedure is intended, at a reasonable level of trust, then one of the procedures that
take into account higher modes contribution should be used. This issue assumes additional
relevance because bridges are inherently irregular structures. Indeed, the transversal
direction is commonly the vulnerable one, which together with the existence of a usually
long deck behaving elastically, makes superior modes particularly significant for what
seismic analysis is concerned. It is common to find bridges where the second and
subsequent modal participation factors are relevant, when compared to the first. CSM and
N2, pioneering procedures, relying on the 1st mode quantities, presented, on the author’s
point of view, naturally inferior performance for this particular sort of structural schemes.
Between the two more recent proposals, ACSM would probably be the most reliable
choice, since results have shown that, closely followed by N2, it presents a good
compromise between predictions of shear forces and displacements at the piers, assumed to
be the most relevant engineering demand parameters. It has also proved to show lower
dispersion among the different bridge categories. To this extent, MPA could be pointed out
as well but its overestimating trend in piers shear forces make it not so appealing. The
procedure might be compromised by the need for a reliable adaptive pushover analysis,
which is still not as spread out as it would be desirable. In any case, it has the fundamental
advantage, when compared to MPA, of overcoming the repetition of as many pushover
analyses as the considered modes.

Conclusions on the accuracy of different structural response prediction tools enable one to
optimize the probabilistic characterization of the seismic demand, to use in the following
Chapter 6, together with the contribution from Chapters 3 and 4, with respect to seismic
input and nonlinear modelling.

5.66
6. Safety Assessment

As soon as all the relevant elements of a typical seismic analysis procedure are properly
defined, the final step becomes the safety assessment itself. Such endeavour may, similarly
to all the safety problem components, feature different approaching scenarios, even though
it will fundamentally consist in the comparison of the demand, coming from the effects
caused by the seismic ground motion input, with the capacity of the structural elements to
accommodate them, which is characterized according to the geometry, material properties,
nonlinear behaviour models, among others. Putting it simply, the safety assessment of a
single structural system can be carried out through the computation of a safety interval,
that is, the deterministic difference between the capacity and the demand, or a failure
probability, which will require statistical characterization of the variables that are part of
the process.

The deterministic approach is most likely the one that practitioners are most familiarized
with, given that it certainly goes along with the traditional designing mode or structural
safety verification. For a certain limit state, demand and capacity are computed, using
mean or characteristic values for the input variables, and the previously mentioned
difference is determined, implying the fulfilment or not of the limit state. The uncertainty
coming from the several variables may be globally accounted for using a safety factor,
applied to increase the demand and/or reduce the capacity. Such approximate procedure

6.1
Safety Assessment

surely cares for consistency and, therefore, the probabilistic approach tends to gain weight
in the actual safety assessment scene. Indeed, the statistical definition of uncertainty can be
quite simple, either from the capacity or the structural demand point of view, consisting
essentially in replacing a single value by a typical distribution, of known mean and
standard deviation, which will represent the variable at stake in a more accurate fashion.

6.1 Failure probability computation

The failure probability of a structural element, within a single failure mode, may be
obtained according to Equation (6.1), where X is a vector containing the basic random
variables x, in which the structural safety is settled; g(X) is the limit state function
associated to the failure mode under consideration and fX(x) is the joint probability density
function of the vector X, characterizing the way how the variables define the structural
safety problem. This is, according to Borges and Castanheta (1985), a commonly
employed procedure, corresponding to the simplest basic problem of structural safety.

pf = ∫ f (x )dx
g ( X )≤ 0
X (6.1)

Regarding the considered failure mode, failure will occur when g(X)<0, safety will be
verified if g(X)>0 and the failure surface, recalling that one is facing a multiple variables
problem, will be given by the condition g(X)=0. Based on such considerations, the failure
probability can be rewritten as Equation (6.2), where prob[g(X)≤0] represents the
probability that g(X) has to be in the failure domain and Fg is the cumulative distribution
function of g(X).

p f = prob.[g ( X ) ≤ 0] = Fg (0 ) (6.2)

The solution of Equation (6.1) will involve multidimensional integration with the integral
dimension being the same as the number of basic variables in vector X, the physical
variables such as loading, material properties or geometrical data, which incorporate the

6.2
 Safety Assessment

uncertainty associated to the regarded failure mode. Depending on the number of variables
and on whether the expression of g(X) is simple or not, the analytical solution for the
integral will be less or more demanding, if possible.

In a structural engineering context, the safety problem, represented by the expression g(X),
will be essentially dependent on two continuous, independent assumed, variables: R,
standing for a measure of resistance, and S, the structural response. The limit state function
is in this case simply given by, in other words, the difference between the capacity and the
demand, as shown in Equation (6.3)

g(X ) = R − S (6.3)

The corresponding joint probability density function, fX(x), a surface that can be
represented in the plane (S,R) by lines of equal constant density of probability, is given by
Equation (6.4), once that R and S are independent. The failure domain, F, opposite to the
condition that S does not exceed R, will naturally be as expressed in Equation (6.5).

f R ,S (r , s ) = f R (r ) ⋅ f S (s ) (6.4)

F =R < S=R − S < 0 (6.5)

Considering the assumptions in Equations (6.3) to (6.5), the failure probability will be, in
the end, given by Equations (6.6) or (6.7), where fS(s)ds is the probability of S within the
interval [s , s+ds] and FR(s) is the cumulative distribution function of R, the probability of
R being less than the value of S corresponding to s. In addition, as R and S are
independent, the probability of both occurring at the same time is given by the product of
each of the probabilities of occurring separately, i.e. fS(s)·FR(s)ds.

6.3
Safety Assessment

+∞ S
p f = prob.(R − S ≤ 0 ) = ∫∫ f R , S (r , s )dr ds = ∫ f S (s ) ⋅ ∫ f (r )dr ds
R (6.6)
F −∞ −∞

+∞
pf = ∫ f (s ) ⋅ F (r )dr ds
−∞
S R (6.7)

As mentioned by (Freudenthal et al., 1966), the sum for all the values of S yields the
convolution integral of Equation (6.7), illustrated in Figure 6.1.

f (s) , f (r) ,
S R

F ( s). f (s) f (r)

R S R

f (s)
S
F ( s). f (s)
R S

R,S

Figure 6.1 – Basic reliability problem

At this point, it is clear that statistical distributions for the resistance, R, and structural
effects, S, caused by the seismic action, are needed if the failure probability is the intended
outcome of the safety assessment process. As far as resistance is concerned, possible
measures for variable R may be simulated experimentally, numerically or even
complementing one with the other. Numerical simulation will definitely expedite the
computation process, enabling the capacity definition to be incorporated within the global
procedure. On the other hand, experimental assessment, even if inherently limited when it
comes to obtaining large dimension samples, has undeniably accurate relevance. However,
statistical characterization, i.e. distribution fitting for a variable, requires considerable
sampling size, which can be impracticable if based on experimental campaigns. Such fact
will lead, therefore, for experimental simulation to play a complementary role, mainly
useful for calibration or validation scenarios. As a matter of fact, the different parameters
that may be used to represent capacity, such as rotational capacity, ultimate curvature

6.4
 Safety Assessment

ductility or shear strength, are typically expressed as a function of basic material

properties, concrete and steel, which assume their own statistical distributions. Working
out the resistance distribution, represented by FR function in Equation (6.7), does not turn
up, thus, as a difficult task. It will require at most, the use of a simulation algorithm, pure
Monte Carlo or improved Latin Hypercube, for the sampling attainment.

The estimate of the distribution of the structural effects, variable S, is, on the other hand,
dependent on a higher number of analysis steps and variables. The first aspect to bear in
mind is that the statistical layout of the parameter representing the response measure may
not be straightforward to obtain and not necessarily similar to the distribution of the
intervening variables. Indeed, different structural systems, subjected to variable intensity,
and corresponding nonlinearity levels, will have different distributions for the same
response measuring parameter. Moreover, input ground motion will feature variability in
intensity and number, that is, several analyses need to be carried out corresponding to a
sufficient number of intensity levels, capturing multiple nonlinearity levels, and distinct
types of earthquake records.

As a result, if capacity characterization presented itself as intuitive and relatively simple,

demand statistical definition, to confront the former with, involves quite a few nonlinear
structural analysis as well as different types of uncertainty to be accounted for. To skip the
need to define the continuous probability density function of the seismic effects on top of
reducing the number of computational demanding analyses, a numerical standardized
methodology is established and systematically followed.

6.1.1 Seismic action – intensity level probability density function

The intensity level probability function represents, at each seismic intensity level, usually
expressed in terms of peak ground acceleration, the density of probability at each point in
the sample space of that random variable. The probability of the variable falling within a
specific set is given by the integral of its density over the set.

To characterize an event such as the occurrence of an earthquake, believed to be highly

unusual, it is common to use the extreme value theory, a branch of statistics dealing with
the extreme deviations from the median of probability distributions. Within the extreme
value theory, the generalized extreme value distribution has been defined, combining three

6.5
Safety Assessment

types of distinct distribution families, also known as type I, II and III extreme value
distributions. It is believed that such three types of distributions are enough to model the
maximum or minimum of the collection of random observations from the same
distribution. Consequently, if the seismic action is to be defined by means of peak ground
acceleration, a maxima-related extreme value distribution will definitely fit such
endeavour. If a resistance-side variable is to be characterized, a minima extreme value
distribution would, instead, be more suitable.

The generalized extreme value distribution is a flexible three-parameter model that

combines the aforementioned maximum extreme value distributions and has the
probability density function given by Equation (6.8), where µ∈ℜ is the location parameter,
σ>0 the scale parameter and ξ∈ℜ the shape parameter.

1   −1 ξ
 x − µ   x−µ
−1−1 ξ

 exp− 1 + ξ    1 + ξ    k ≠0
σ    σ    σ  
f ( x; µ , σ , ξ ) =  (6.8)
1  x−µ  x − µ 
 exp −   − exp −  k →0
σ   σ   σ 

 x−µ 
1+ ξ  >0
 σ 

The shape parameter, ξ, will define the tail behaviour of the distribution and, therefore, its
type, each one corresponding to the limiting distribution of block maxima from a different
class of underlying distributions. Distributions whose tails decrease exponentially, such as
the Normal, lead to type I, with ξ tending to zero. Distributions whose tails decrease as a
polynomial, such as Student’s t, lead to type II, with ξ positive. Distributions whose tails
are finite, such as the Beta, lead to type III, with ξ negative. Types I, II and III are often
referred to as the Gumbel, Fréchet and Weibull extreme value distribution families, which
can be arguable, in terms of inconsistency. To be exact, the Type I and Type III cases
actually correspond to the mirror images of the usual Gumbel and Weibull distributions,
respectively. The Type II case is equivalent to taking the reciprocal of values from a
standard Weibull distribution. The Type I distribution, Gumbel related to, has early been
used in applications of extreme value theory to engineering problems and, as related to the

6.6
 Safety Assessment

maxima, is frequently the chosen one to characterize seismic action intensity. Indeed the
extreme value theory claims that it is suitable if the distribution of the underlying sample
data is of the normal or exponential type. Its probability density function, following the
general expression in Equation (6.8), is given by Equation (6.9), which can be written in a
more simplified manner, as in Equation (6.10), where z=(x – µ)/σ.

  x−µ   x − µ 
f ( x | µ ,σ ) =
1
exp −   − exp −  (6.9)
σ   σ   σ 

−z
e − z −e
f (z,σ ) = (6.10)
σ

The Type I distribution is unbounded, defined for the entire real domain, and admits a
minima version. Its general shape remains the same for all parameter values. The location
parameter, µ, shifts the distribution along the real line and the scale parameter, σ, expands
or contracts the distribution. Figure 6.2 plots the probability density function, f(x|µ,σ), for
different combinations of µ and σ, where the variable x is the peak ground acceleration,
used to characterize the seismic intensity level, expressed in cm/s2.

-3
x 10
8

µ=100,σ=50
7
µ=200,σ=100
µ=400,σ=150
6

5
f(x|µ ,σ )

0
0 100 200 300 400 500 600 700 800 900 1000
x(cm/s2 )

Figure 6.2 – Extreme value Type I (Gumbel associated) distribution probability density functions
for different combinations of location and scale parameters.

6.7
Safety Assessment

6.1.2 Capacity distribution

The structural resistance can be analysed considering the structure as a whole, or by

scrutinizing its individual elements, studying the resistance of each in separate. When the
structure in study is a bridge, the application of the latter case is more intuitive, given that
the resisting elements are extremely well defined as well as the failure mechanisms. In
other words, considering that the deck will assume elastic behaviour and neglecting
possible soil failure at the foundations, the piers will be the crucial elements, behaving
nonlinearly and causing the bridge collapse as soon as the first attains rupture. The density
of probability will therefore be expressed as a function of the action effects on such
elements, as well as the corresponding capacity.

Regardless of the selected procedure to carry out the safety assessment, as long as a
probabilistic analysis is intended, the capacity of the piers cross sections needs to be
determined and characterized by means of a statistical distribution, quantified in terms of
ultimate ductility permitted by the cross section at the bottom of the piers or the
corresponding top displacement. This is because the input data of the models are often
quite uncertain, requiring them to be considered as random variables. Such capacity
definition must certainly be able to incorporate the uncertainty associated to the variables
on which the ultimate capacity depends. Apart from the geometrical characteristics, for the
case of a cross section made up of reinforced concrete, the resistance will be a function of
the material ultimate properties, the yielding and peak tensions, and corresponding strains,
which are expected to contribute the most to the variability of the capacity controlling
variable.

A proper characterization of the intended distribution will consist on repeated random

simulation of the intervening variables, using their distributions and an appropriate
algorithm, until a sample is obtained. The appropriateness of the simulation procedure will
focus predominantly on the best relation between sample size and associated reliability.
The main goal is to make use of a simulation technique able to yield samples as small as
possible, and consequently faster, that are still good representatives of the underlying
distribution. Among the available solutions the, let us say, traditional approach would be
the Monte Carlo sampling scheme, based on pure random simulation. A more recent and
innovative technique is the Latin Hypercube (McKay et al., 1979; Iman et al., 1981),

6.8
 Safety Assessment

which has been gaining popularity among scientific studies due to the claimed efficient
reduction in the necessary sampling size.

6.1.3 Latin Hypercube Sampling

Safety assessment based on reliability calculations will involve repetitive simulation of

several properties, which may become, mainly if carried out for complex, finite-element
modelled structures, quite time-consuming and demanding in terms of computational
effort. In order to overcome such drawback, advanced simulation techniques are required
based on relatively simple integrated modifications in the existing general procedures. The
Latin Hypercube sampling (LHS) simulation procedure (Olsson and Sandberg, 2002;
Olsson et al., 2003) fits that set of techniques and can be seen as a particular case or
version of the standard Monte Carlo (SMC) numerical simulation, with the need for a
smaller number of runs, tens to hundreds, to achieve a reasonably accurate random
distribution. The LHS may in fact be incorporated into an existing Monte Carlo model
fairly easily and work with variables following any analytical probability distribution. Such
key features are accomplished by means of the stratification of the theoretical probability
distribution function of input random variables. The method is not restricted to the
estimation of statistical parameters of structural response but offers a quite broad usage
domain that goes from sensitivity analysis to Bayesian updating, among other possibilities.
It has been found that LHS is very efficient for estimating mean values and standard
deviations in stochastic structural analysis (Ayyub and Lai, 1989; Florian and Navratil,
1993; Novák et al., 1997) but only slightly more efficient than the SMC for estimating
small probabilities, which can definitely occur when assessing the seismic failure of
bridges. Additionally it has been found that, under some particular conditions, LHS is
considerably more efficient than the SMC version if the probability of failure is dominated
by a single stochastic variable, which again may be verified in the ultimate limit states for
reinforced concrete cross sections depending essentially on the ultimate steel strain. It
seems thus far relevant the application of LHS on such endeavour, expected to perform
from slightly to considerably better than SMC.

The LHS strategy is actually rather straightforward, making use of stratified sampling,
within a simple concept: the probability ranges of probability distribution functions for
input random variables are divided into N equivalent intervals, N being the number of

6.9
Safety Assessment

realizations. Assuming that the structural problem has M input variables Xj, with j=1:M,
the range [0,1] of each cumulative distribution function F(Xj) is divided into N different
non-overlapping intervals of equal probability 1/N, as illustrated in Figure 6.3. Each of
those intervals is then represented by its centroid, Cij, the abscissa of Xj for the average
value of each 1/N probability interval.

F(X
F( yjj))

1.0
(N- 1) /N

(i+1)/N
i/N
(i- 1) /N

2/N
1/N
Ci j yj
X j

Figure 6.3 – Division of the cumulative distribution function F(Xj) in N intervals.

The following step is to use the centroids for the simulation process, this is, the centroids
are selected randomly based on random permutations of integers 1 to N and representative
values are obtained via inverse transformation of the cumulative distribution function. It is
however mandatory that every interval, or centroid, is used, or simulated, once during the
random process. Such condition can be seen as the main advantage of the LHS strategy:
the regularity of probability intervals on the probability distribution function and the
assurance that all of them are accounted for ensures good quality samples, even for a small
number of realizations, N. The final output will be a matrix of random permutations of the
centroids, with dimension N×M, where each column corresponds to one of the input
variables and each row to a particular simulation.

Looking further into the LHS algorithm, where the sampling space will be M-dimensional,
let P denote a matrix [N×M] containing in each column a random permutation of 1, …, N
and R a same size matrix of independent random numbers from the uniform (0,1)
distribution. Using those matrices the basic sampling plan is established and represented by
the matrix S, as follows in Equation (6.11).

6.10
 Safety Assessment

S=
1
(P − R ) (6.11)
N

Each element of S, sij, is mapped according to the corresponding variable distribution F(Xj)
as in Equation (6.12), where FX−1j is the inverse cumulative distribution function for the

variable Xj.

xij = FX−1j (sij ) (6.12)

The vector xi = [xi1 xi2 … xim] will correspond to the input data for one deterministic
computation.

The condition of each centroid being used only once can be observed in a simple example
simulation in Figure 6.4. The sampling space size is 5×2, five realizations of two input
variables. Each column and each line are taken once and all the sampling space is used,
which might not have happened if the SMC simulation scheme had been used.

0.8
Variable 2

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
Variable 1

Figure 6.4 – Latin Hypercube example sampling space for two variables and five realizations.

Nevertheless, there is no assurance that spurious correlation does not appear, even if the
distribution function of each variable is efficiently employed. Figure 6.5 illustrates how an
unwanted spurious correlation in the sample can occur despite the verification of the
simulation LHS conditions. It has, however, been easily demonstrated by Iman and
Conover (1982) that such scenario can be reduced by modifying the permutation matrix P,
used in Equation (6.11).

6.11
Safety Assessment

Correlation
reduction

Figure 6.5 – Unwanted (left) and reduced (right) correlation of sampling plan (Olsson and
Sandberg, 2002).

The procedure is quite simple and basically consists in mapping the elements of matrix P,
divided by the number of realizations plus one, on the (0,1) Gaussian distribution. The
covariance of this new matrix is computed and then Cholesky decomposed, which,
according to proper dealing, will lead to a final alternative matrix, P*, high-level spurious
correlation free, that substitutes the original P one. The only limitation of the procedure is
the number of realizations to be higher than the basic input variables, N>M, so as to allow
the Cholesky decomposition. Such condition, given the type of structural analysis at stake,
will however hardly become restrictive.

6.1.4 Seismic structural effects

Once the resisting structural elements are properly characterized and the seismic action is
defined, the structure will be subjected to the latter and the seismic effects quantified and
appropriately compared with the capacity. The way such comparison is carried out, which
can assume several different versions, will be determined by the type of collapse
probability, depending on how the uncertainty associated to the different variables is taken.

The classical approach, following Figure 6.1 and Equation (6.7), is to build the probability
density function of the seismic effects, submitting the structure to increasing seismic
intensity levels, along the entire domain, and develop the so-called vulnerability function,
the element that enables the conversion of seismic action into seismic demand. This is a
local variability version, given that uncertainty of the action and the capacity are
considered separately, in local, distinct stages.

6.12
 Safety Assessment

If one chooses, still within the local variability approach, to limit the seismic action to a
specific intensity level, the probability of failure is equally possible to get and possibly full
of interest, because directly connected to a particular ground motion level. The seismic
effects can be again characterized again through their probability density function
recurring to different demanding earthquake records, characterized by the same intensity
level, though.

If the goal is to carry out a series of independent realizations, where the uncertainty of each
variable is taken simultaneously and accordingly, then the structural effect consists in the
seismic demand for each run, which is directly compared to the capacity, determined for
the same structural properties.

6.1.5 Failure probability with local uncertainty

The traditional approach, which has been originally developed and applied in the work by
Costa (1989), presents as essential target the definition of the structural effects probability
density function, which will be confronted to the resistance one. In order to do so, it is
evident the need for a relationship between the seismic intensity and the response control
variable, coming from the nonlinear analysis of the structure, or any other reliable tool, for
that specific ground motion level. That relation is usually named vulnerability function and
is essentially a mapping function, establishing the connection between the seismic
intensity, its probability and the actual effect in the structure, confronted with the capacity
to reach the final collapse probability.

The element from where the whole process starts is the seismic action, A, often
characterized in terms of peak ground acceleration (PGA), one of the possible measures of
intensity. The probability density function, fag, of the maximum annual peak ground
acceleration for a specific site is therefore used and comes from corresponding hazard
studies. The Hazard scenario for a specific site may be defined as a function of the peak
ground acceleration or, according to more recent tendencies, acceleration spectral ordinates
for a period that is similar to the fundamental period of the structure. Even if the use of
spectral acceleration is claimed by many as more accurate, given that it centres the analysis
on the structure’s fundamental mode of vibration, the peak ground acceleration seems to
still stand as a reliable intensity measure, as observed in the results of the study in

6.13
Safety Assessment

Chapter 3. Nevertheless, to the same peak ground acceleration there might correspond
different types of ground motion records, concerning nature (real or artificial), duration,
magnitude, epicentral distance, among other characteristics. For that reason a set of distinct
accelerograms must be considered, reflecting wide-ranging frequency content. For a
predefined, not very large, number of intensity levels, corresponding to different return
periods and probabilities of occurrence, the records are scaled to match that specific PGA.
Nonlinear dynamic analysis is carried out using each of the scaled accelerograms, yielding
a sample of the response measure that accounts for record associated variability. For each
ground motion intensity, and corresponding PGA, a statistical value, defined by the mean
or median response measure, is computed and a polynomial function is adjusted to those
points, putting up the vulnerability function, illustrated in Figure 6.6, the function that
turns the seismic action into the demand. The curve will be a 2nd degree, or higher,
polynomial, due to the nonlinear structural behaviour with increasing intensity.

S S

Demand Demand
samples samples

Vulnerability
function

ag ag

Figure 6.6 – Fitting of the vulnerability function to demand points (LESSLOSS, 2004b).

The material properties used for the structural analyses may be considered on a
deterministic basis, through their mean or characteristic values, or making use of their
statistical distribution. If the latter is employed, corresponding to the global uncertainty

6.14
 Safety Assessment

version of the proposed method, the number of structural analyses will necessarily increase
but the statistical meaning of the computed failure probability will be superior. Further
details on the use of such global simulation procedure, taking into account the variability
of the seismic action and material properties all together are discussed in Section 6.1.7.

The vulnerability function may be estimated recurring to pushover analysis, whose validity
in estimating nonlinear response of structure has been recognized in Chapter 5. The
advantages in using such nonlinear static approach are immediate given that N nonlinear
dynamic analyses are avoided, greatly simplifying the entire process, as much as higher the
number of realizations is. The abscissa for the vulnerability curve will the peak ground
acceleration of the response spectra used in the nonlinear static procedure to locate the
performance point in the pushover curve and quantify the control variable. There will be,
therefore, two variants for the failure probability computation version herein accounted for,
based on the distinct ways to get to the structural effects, compared in Chapter 5: nonlinear
static and dynamic analysis.

Once the vulnerability function is properly defined, the statistical distribution of the
response measure can be easily obtained through a numerical procedure based on equal
areas under the probability density function, i.e. probability of occurrence, along with the
changing of the variable from peak ground acceleration to the response measure itself. The
domain of the seismic action probability density function, fag(a), is divided into small
increments, let us say dag. For each increment dag the area A1, see Figure 6.7 (left), is
determined based on the ordinate of the fag function for the centroid of dag. The value
fag(dag) is converted into the value of the structural effects’ probability density function
fS(s) for the centroid of the corresponding ds, guaranteeing that areas A1 and A2 in Figure
6.7 (right) are the same.

6.15
Safety Assessment

R,S
R,S

Vulnerability
function Vulnerability
function
A2

A1

ag ag

Figure 6.7 – Blockwise definition of the probability density function fS(s) (left); Failure probability
computation (right) (LESSLOSS, 2004b).

The variable s, used to measure the seismic effects, as well as the resistance, will refer to
the piers, which are the resisting nonlinear behaving structural elements, and is usually
taken as the ductility at the base or the top displacement. The final step will be to cross the
capacity and the seismic demand distributions, computing the integral in Equation (6.7).

The entire process may be schematically depicted in Figure 6.8, which illustrates the
traditional version of the procedure described so far.

The seismic intensity is characterized by its probability density function, the curve 1.
Analogously, at each intensity level of the seismic action, for which curve 1 is defined, the
effect in the structure is obtained, in terms of a chosen parameter, ductility or
displacement, for instance. The so-called vulnerability function, curve 3, is therefore built,
representing the structural effects of the seismic action long with increasing intensity. This
sort of information is fundamental to cross with the seismic action probability density
function and generate the statistical distribution of the structural effects, curve 4. In other
words, by putting together the probability density associated to a seismic level and its
effect in the structure, one can obtain the probability density of such effect to be verified.

6.16
 Safety Assessment

StructuralAction
effects
Efeito da acção
Effect

Seismic
Acção Action
Seismic Action
Sísmica

Figure 6.8 – Failure probability computation with local uncertainty – traditional approach

At the same time, the distribution of the structural capacity, curve 2, in terms of the same
response parameter used to define curves 3 and 4, ductility or displacement, needs to be set
up. This step is typically carried out by means of a simulation procedure, such as Monte
Carlo or Latin Hypercube sampling methods, based on the statistical distributions of the
variables with relevant uncertainty, usually cross section material properties. Finally, the
intersection of capacity with demand, managed by the convolution integral, will yield the
intended failure probability.

6.1.6 Failure probability for a given intensity level

Additionally, a different meaning can be given to the failure probability coming from the
integral in Equation (6.7), depending on the level of restrictions that is considered within
the adopted procedure. Indeed, the global collapse probability computation method just
described uses the probability density function of the seismic action defined for the entire
intensity domain, taking into account the relative probability of occurrence of different
intensity levels, within a specific hazard local scenario.

On the other hand, if the ground motion intensity is constricted to a specific level, the
computation of the failure probability is still possible, according to Figure 6.9.

6.17
Safety Assessment

R,S

Demand samples
constrained to ag i

ag i ag

Figure 6.9 – Response measure constrained to the ground acceleration agi (LESSLOSS, 2004b).

Such number will be equivalent to a point in the structure’s fragility curve, the probability
of occurrence of the limit state under consideration for a specific intensity level of
earthquake ground motion. The main difference when following this approach will be the
structural effects distribution definition, although the principle remains the same. A set of
different ground motion records is used, all of them with a specific peak ground
acceleration level agi, and the probabilistic distribution of the structural effects, fS,agi, is
defined with the response measure sample coming from that set of accelerograms, whereas,
in the global procedure, the mean or median for each intensity was computed. This sort of
approach will definitely require less computational effort but will, at the same time, call for
a sufficiently large number of records, needed to achieve a reliable distribution of the
structural effects. In the end, Equation (6.7) is solved numerically, replacing fS(s) by fS,agi,
leading to the failure probability pf(agi), the fragility curve ordinate for that abscissa. If the
whole process is repeated, a pointwise definition of the fragility curve of the bridge is
acquired, which can be used, if intended, to compute the global failure probability, making
use of the Hazard local function within the entire considered ag domain.

6.18
 Safety Assessment

6.1.7 Failure probability with global uncertainty

A sort of alternative approach for the failure probability computation described in 6.1.5 is
to consider the uncertainty associated to the different intervening variables in a global
fashion, this is, all at the same time, within a global procedure, carried out for each
iteration step, from the beginning to the end.

It is recalled that, in the procedures described so far, uncertainty has been taken into
account independently; this is, for each eligible variable, a statistical distribution, that is
able to define it, is the element that carries the associated uncertainty and is used all at
once. Examples are, in Figure 6.8, the conversion of distribution 1 into distribution 4
through curve 3 or the intersection of the entire distributions 2 and 4 to yield the
convolution integral.

On the other hand, the approach herein presented uses the statistical distributions of the
different variables in a different manner. Firstly, the distributions are defined,
characterizing each purely input variable, the seismic action and the resistance. Once that
initial step is completed, the assessment procedure is carried out completely, but
individually, recurring to those distributions. Within a global statistical simulation process,
the already described Latin Hypercube, each variable with known uncertainty, seismic
intensity level, type of record and material properties, is randomly simulated, the nonlinear
dynamic analysis is carried out using such parameters, the structural effects and the
capacity are determined and the difference between those two numbers, usually referring to
ductility or displacement, is computed. A statistical distribution will be then adjusted to
that difference, which will lead to the computation of the collapse probability in a different
mode. Data for each realization is obtained through the application of the LHS technique
(see Section 6.1.3 for further detail) in a broader sense, taking as basic input variables the
material properties, the seismic intensity level and earthquake record all together.

Figure 6.10 roughly illustrates the process just described. A vector of input variables,
consisting of the seismic intensity level, ag, type of ground motion record and the relevant
material properties, is put up by random simulation, according to predefined distributions,
and capacity is computed, expressed in available ductility, µR. Nonlinear dynamic analysis
is carried out on the structural model, built with the simulated properties, for the specified

6.19
Safety Assessment

accelerogram with ag level, and the demanded ductility, µS, a point of the previously
mentioned vulnerability function, is obtained.

StructuralAction
effects
Efeito da acção
Effect

2
µR
3

5
µR – µS

µ4S

Seismic
Acção Action
Seismic Action
Sísmica

ag

Figure 6.10 – Failure probability computation with global uncertainty – alternative approach

The difference µR – µS, eventually non-positive, will represent some sort of safety gap. The
main concept behind this procedure is that the entire process, until the computation of the
available ductility margin, can be repeated N times in order to fit a statistical distribution to
that variable. As illustrated in Figure 6.11 the knowledge of such a distribution will enable
the computation of the failure probability as the area under the corresponding probability
density function, fR-S, for a value of µR – µS least than zero.

f (R f−R –SS)

p f = P[(R − S) < 0] = FR −S (0 )
pf = P[(µ R – µS) < 0] = F(µR – µS)(0)

R –− SS
R
Figure 6.11 – Safety margin distribution and failure probability definition

6.20
 Safety Assessment

This way of looking at the safety assessment procedure, within a bridge structures context,
will be largely influenced by two important aspects. The first, and the most important one,
has to do with the random simulation algorithm, used to build the samples for each variable
that cares for statistical characterization. The procedure herein chosen was the Latin
Hypercube sampling method, as an alternative to the pure Monte Carlo sampling scheme,
and will definitely influence the significance of the results, being important to avoid biased
samples, as a warranty of data consistency. The more solid the simulation algorithm is, the
more reliable the findings will be, given that widely representative samples, for intensity
level, for instance, will be needed to assure that the structure will be pushed to its limits.
Additionally, such approach will be extremely sensitive to the type of distribution that is
assumed for the variables, mainly the ones on the seismic action definition side, type of
record and intensity level. Whereas the first may be characterized through a uniform
distribution, the latter is usually described by extreme value distributions, which are known
to be susceptible when working on the tails region, essential for the failure probability
computation – see Figure 6.11.

6.1.8 Reliability index

As a complement to the safety assessment and to soothe eventual lacks of accurateness,

associated to the collapse probability being given by the area under the probability density
function, other safety measuring parameters may be taken into account. An example of
such indirect safety measures is the reliability index, β, which essentially settles on the
distance from the mean R–S to the origin, quantified in standard deviations of the variable
R–S, according to Equation (6.13).

R−S
β= (6.13)
σ R −S

Within such definition, which considers the mean safety margin together with its
dispersion, the smaller the value of the reliability index the higher the probability of failure
will be.

6.21
Safety Assessment

Moreover, the convolution integral computation might not be, for some cases, at least, an
easy task, leading to the need for a numerical solution. The use of this indirect evaluation
parameter becomes, for that reason, at least, appropriate. Reliability index can also be
easily related to central factors of safety.

6.2 Application to a set of bridges

The performance comparison of the several safety assessment methodologies will be,
again, carried out using the set of bridge structures and real earthquake records described
in Chapter 4, Section 4.3. Three different versions for the failure probability computation
will be considered: traditional numerical safety assessment with nonlinear dynamic
analysis of structural effects (NSA-NDA), using plastic hinge models of the bridge
configurations (faster and validated in Chapter 4); traditional numerical safety assessment
with a nonlinear static procedure (chosen from the possible alternatives validated in
Chapter 5) to estimate structural effects (NSA–NSP) and safety assessment based on global
simulation with Latin Hypercube sampling (LHS).

Prior to the safety assessment itself, a brief calibration study is, however, carried out with
the main purpose of defining the parameters for the application of the simulation
technique, the Latin Hypercube, which is fundamental. Among others, the number of
realizations or the relative importance of the different input basic variables on the capacity
computation, are aspects that are worth looking into, so as to optimize the use of the
algorithm. Such sensitivity analysis gains special relevance for the particular context of
structural nonlinear analysis of bridges (Monteiro et al., 2009).

Random simulation is used twice within the current endeavour. First, and commonly to all
the approaches, it is used to build the capacity distribution, based on the repeated
computation of the available ductility of the cross section at the base of the piers, based on
equal number of realizations of the relevant concrete and steel tensions and strains.
Furthermore, sustaining the LHS approach, the variables intensity level and type of record
are added to the hypercube and N nonlinear dynamic analyses, equal to the number of
realizations, are conducted.

6.22
 Safety Assessment

6.2.1 Latin Hypercube sampling in capacity definition

The capacity definition, when made up in terms of available ductility in curvature of the
piers cross section, settles on the geometric characteristics and material properties, the steel
and concrete stress-strain models. For a given cross section, with a certain axial load, the
cracking, yielding and ultimate points of the monotonic moment-curvature diagram are
determined using well known criteria for the mentioned stages definition, by means of a
fibre-type section discretization. The variability in material properties is, however, widely
recognized as typically significantly higher than uncertainty surrounding section
dimensions (Grant et al., 1978; Frangopol et al., 1996). The parameters that are admitted to
assume relevant variability are, therefore, the concrete peak stress, fc, steel yielding stress,
fy, and corresponding ultimate strains, εc and εy. A normal distribution is adopted for each
of the variables with mean values corresponding to the material class. Different scatter
levels are assumed, characterized by increasing coefficients of variation, 5%, 10% and
15%, for all the variables at once, or individually to better scrutinize their relative
importance in the ductility computation. At the same time, the number of realizations
needed to stabilize the result is sought, for each of the dispersion levels. The parametric
study will focus, hence, the significance of the input variable, the corresponding coefficient
of variation and the number of realizations.

The plot on the left of Figure 6.12 illustrates the evolution of the mean value of the
computed ductility for the section of one of the piers, intermediately axially loaded. Five
representatives sampling sizes have been chosen: 20, 50, 100, 200 and 500. Both of the
parameters, coefficient of variation and number of realizations, denote a similar tendency
for the mean ductility to become completely stable for 200 realizations. The same trend is
found when the COV results are looked at. It seems however acceptable the use of 100
realizations as sample size, given that dispersion is not that considerable, neither across
input variables COVs nor sample size.

6.23
Safety Assessment

35 20
COV=5% COV=5%
34.8 COV=10% 18 COV=10%
COV=15% COV=15%
34.6 16

34.4 14
Mean Ductility

34.2 12

COV (%)
34 10

33.8 8

33.6 6

33.4 4

33.2 2

33 0
0 20 50 100 200 500 0 20 50 100 200 500
Number of realizations Number of realizations

Figure 6.12 – Mean available ductility (left) and corresponding COV (right) versus number of
realizations for different coefficients of variation.

The ability of the method to build reliable samples is defined, on the one hand, in terms of
the necessary number of realizations to assure representativity, whereas, on the other hand,
one can question the robustness of the method for successive repetitions of the procedure.
In other words, there is the concern of wondering if two samples of the same size,
independently obtained, will have sufficiently close mean numbers. In order to check on
whether such aspect is pertinent or not, the procedure originating the results in Figure 6.12
left has been repeated ten times, recording the history of mean values for ultimate ductility
along with sampling size. The results for the different considered COVs are plotted in
Figure 6.13.

It is evident form the observation of the plots that, especially for the lower COVs, 5% and
10%, 100 realizations, or even 50, will be enough for the variation of the mean ductility
across repetitions of the simulation algorithm to lose significance. The worst scenario is
definitely for the case of only 20 realizations, for a COV of the input variables of 15%, for
which variation is, anyhow, less than 1%. The main conclusion is thus that, for what
concerns the trustworthiness of the procedure, oscillation of mean values between
repetitions is not relevant. This sort of variability will be further investigated when using
the LHS procedure for direct obtainment of the failure probability through the safety
margin.

6.24
 Safety Assessment

35 35
COV=5% COV=10%
34.8 34.8

34.6 34.6

34.4 34.4
Mean Ductility

Mean Ductility
34.2 34.2

34 34

33.8 33.8

33.6 33.6

33.4 33.4

33.2 33.2

33 33
20 50 100 150 200 300 400 500 20 50 100 150 200 300 400 500
Number of realizations Number of realizations

35
COV=15%
34.8

34.6

34.4
Mean Ductility

34.2

34

33.8

33.6

33.4

33.2

33
20 50 100 150 200 300 400 500
Number of realizations

Figure 6.13 – Mean available ductility versus number of realizations for several sampling
repetitions and different input variables COVs.

If each variable is considered to be the only one with known standard deviation, while the
remaining are kept constant with the mean value, its relative influence in the final result,
the ductility, in this case, can be studied, as in Figure 6.14.

From the observation of the results, the ultimate curvature of the piers cross sections comes
out as fundamentally dependant on the steel ultimate strain. Not only the coefficient of
variation reaches a peak value, but also the scatter among different numbers of realizations
is higher, when just the scatter in εy is considered. Such trend is not found for any of the
other variables, but only when all are considered to have the same dispersion, which
denotes that failure of the section is being caused by the reinforcement.

6.25
Safety Assessment

35 20

34.8 18

34.6 16

34.4 14
Mean Ductility

34.2 12

COV (%)
34 10
20 20
33.8 8
50 50
33.6 100 6 100
200 200
33.4 500 4 500
33.2 2

33 0
all fy εy εc fc all fy εy εc fc
Variable Variable

Figure 6.14 – Mean available ductility (left) and corresponding COV (right) for different numbers
of realizations according to the dispersion of input variables.

A limitation of such preliminary plots is the fact that equal standard deviation, or COV, has
been assumed for all the four basic input variables, tensions and strains, when it is well
known that concrete related variables present larger uncertainty when compared to the steel
ones. Some studies have been carried out in the past with the purpose of quantifying
uncertainty in nonlinear behaviour of reinforced concrete members. One of those studies,
(Kappos et al., 1999), focused, partially, at least, the quantification of variability in input
parameters, namely, the material properties. The model uncertainty in commonly used
confinement models was quantified and introduced in probabilistic modelling of members’
ductility, through fibre model analysis. Such work goes along with the approach followed
herein that uses the available curvature ductility of the cross sections to assess the failure
probability.

For what concrete related variables are concerned, the ultimate concrete strain deserved
particular scrutiny, given that it has been considered by the authors as often the governing
parameter influencing failure, when it comes to ductility. Such strain, predicted by a
confinement model, is assumed as the one corresponding to 0.85fc. The approach that was
followed to estimate variability in such parameter consisted in comparing analytical values
of εcu given by currently adopted models (Park et al., 1982; Scott et al., 1982; Sheikh and
Uzumeri, 1982; Kappos, 1991) with results from experimental tests (Vallenas et al. (1977),
Sheikh and Uzumeri (1980), Scott et al. (1982) and Moehle and Cavanagh (1985)) and get
the uncertainty from there. The amount of scatter that has been encountered in comparison

6.26
 Safety Assessment

ratios is considerable, with COVs ranging from 32 to 36%, even if mean values are very
close to unity, confirming εcu as highly uncertain. As for the concrete compressive
strength, fc, referring to cylinder measures, a study of Barlett and McGregor (1996),
specifically on this matter, has come up with COVs around 18%. For both of these
concrete-related variables, as typically occurs, a normal distribution has been assigned.
Analogously, the same characterization needs to be done regarding steel parameters,
knowing in advance that scatter will be certainly lower. The work carried out by the Joint
Committee for Structural Safety (JCSS, 1995) and by Pipa and Carvalho (1994) on
variability for tempecore steel in various European countries indicate a value of 6% for the
yielding and ultimate steel strengths, fy and fu, COV and 9% for the ultimate strain, εsu, one.
A normal distribution is again currently admitted for the parameters.

Table 6.1 indicates the assumed coefficients of variation for the intervening material
properties, which have been defined according to what is proposed in (Kappos et al.,
1999), except for the concrete ultimate strain, that has been reviewed into a slightly lower
amount.

Table 6.1 – Admitted coefficients of variation (COV) of material properties.

Parameter COV (%)

Concrete ultimate strain (εcu) 30%
Concrete compressive strength (fc) 18%
Steel ultimate strain (εsu) 6%
Steel yield strength (fy) 9%

If the parametric calibration is now conducted considering the dispersion levels just
presented, new plots for the behaviour of the mean ductility numbers for different sample
sizes can be obtained. Even if the admitted dispersion is considerably different, the plots in
Figure 6.15 denote results that are rather smooth. The tendency previously found, that 200
realizations is the needed number of realization for the mean to stabilize, still stands.
Analogously, and even surprisingly, the resulting COV for the available cross section
ductility is extremely steady around 10%. Nevertheless, such findings may be understood
by recalling that, according to results in Figure 6.14, the variable with higher influence is
the steel ultimate, which assumes a significantly low coefficient of variation whereas the
variables with higher uncertainty, the concrete ones, in turn are not so relevant for the

6.27
Safety Assessment

ductility estimation of the hollow sections in consideration. It seems therefore logical that
the results are little or not worsened at all by the changes.

35 20

34.8 18

34.6 16

34.4 14
Mean Ductility

34.2 12

COV (%)
34 10

33.8 8

33.6 6

33.4 4

33.2 2

33 0
0 20 50 100 200 500 0 20 50 100 200 500
Number of realizations Number of realizations

Figure 6.15 – Mean available ductility (left) and corresponding COV (right) versus number of
realizations.

Regarding the variability of the sampling method within several repetitions, illustrated in
the plot of Figure 6.16, the main previous observations can still be made, reinforcing the
conclusion that from 200 realizations on, expect for one or another punctual occurrence,
the steadiness of results is noticeable, in both senses of the variability.

According to the brief parametric study, within the subsequent applications, the capacity of
the piers cross sections has been characterized using samples of 200 occurrences.

35

34.8

34.6

34.4
Mean Ductility

34.2

34

33.8

33.6

33.4

33.2

33
20 50 100 200 500
Number of realizations

Figure 6.16 – Mean available ductility versus number of realizations for several sampling
repetitions.

6.28
 Safety Assessment

6.2.2 Seismic action probability density function

The seismic intensity level has been set in terms of peak ground acceleration (PGA),
which, according to the major observations made in Chapter 3, together with spectral
acceleration intensity measure, leads to the less scattered mean estimates when using real
records. Nevertheless it does indubitably disregard specific information on the earthquake
record, which will sure introduce limitations on following calculations. In line with this
approach, a seismic scenario in terms of probability density of peak ground accelerations
for a specific region needed to be chosen. The seismic action probability density used in
this framework is based on the studies of Duarte and Costa (Duarte and Costa, 1991;
Costa, 1993), where seismic conditions for the Portuguese regions were idealized by means
of a source-zones model with non-radial attenuation functions. Two types of earthquake
scenarios are considered, according to what is expected for Portuguese seismicity,
corresponding to nearby moderate magnitude (type 1) and long distance large magnitude
(type 2) earthquakes (Costa, 1993). A hazard model for a period of 50 years was adopted,
calibrating parameters µ and σ for the type I Gumbel distribution, as detailed in Section
6.1.1, from numerical results obtained from source-zones models: µ = 87.36cm/s2 and
σ = 1/0.00225 for type 1 and µ = 60.5cm/s2 and σ = 1/0.0031 for type 2. Figure 6.17 plots
the assigned distributions.

It is important to bear in mind that the probability density function is defined with respect
to the probability of exceedance of the PGA parameter, which enables that any set of
earthquake records are used, as long as referenced to their PGA. Such approach will almost
certainly be seen as limited, relying on the strength of the defining parameter. The selected
records will be therefore scaled to match the peak ground acceleration level given by the
extreme value distribution.

It seems evident, however that the seismic scenario representing the Portuguese hazard is
not particularly demanding, which will possible not take the bridges response into high
nonlinear stages. The probability density function has therefore been adapted in order to
allow the structural response to be clearly nonlinear, so as to fully understand the
performance of the different assessment techniques. Moreover, the nature of the parametric
study that is herein conducted is essentially comparative, in terms of distinct procedures
and different bridge configurations, rendering the probability density function somewhat

6.29
Safety Assessment

irrelevant, as long as the intended comparison purpose is fulfilled. The final adopted
distribution for the seismic intensity is plotted in Figure 6.18, in comparison with the
Portuguese ones, and was calibrated with the intent of higher probability density peak at
higher peak ground acceleration, higher µ, together with lower dispersion, lower σ.

-3
x 10
1.2

type 1
type 2
1

0.8
f(x| µ,σ)

0.6

0.4

0.2

0
0 100 200 300 400 500 600 700 800 900 1000
2
x(cm/s )

Figure 6.17 – Probability density function (extreme value Type I, Gumbel associated) for two
expected Portuguese earthquake scenarios.

-3
x 10
4.5
type 1
4 type 2
adopted
3.5

2.5
f(x| µ,σ)

1.5

0.5

0
0 100 200 300 400 500 600 700 800 900 1000
2
x(cm/s )

Figure 6.18 – Adopted probability density function (extreme value Type I, Gumbel associated,
µ = 415cm/s2 and σ = 0.012).

6.30
 Safety Assessment

6.2.3 Failure probability with traditional NSA approach

Within this approach, the vulnerability function will be the main needed element, given
that the failure probability is computed considering the uncertainty locally, characterizing
the structural response for the entire domain of the intensity level distribution (see
Section 6.1.5) in terms of measured curvature ductility, µ, at the base of the piers.
According to Section 6.1.4, the vulnerability function may be built up from nonlinear
dynamic analysis or from pushover analysis, within a nonlinear static approach. Results
from the application of both possibilities come in what follows.

6.2.3.1 Vulnerability function from nonlinear dynamic analysis

The domain of the probability density function has been taken as from 0 to 1000cm/s2 and
discretized in steps of 100cm/s2. At each step, that is, each intensity level, each of the ten
real earthquake records (SAC, 1997), identified in Section 4.3, was scaled so that its
individual PGA would match that specific level. Nonlinear dynamic analysis using those
records was run, and the ductility in curvatures was measured at the base of each pier (the
piers in short configurations will be named from P1 to P3 and in long configuration from
P1 to P7). Due to the main drawback in using dynamic analysis with real ground motion
records, which is the scatter associated, for a given intensity level, different response
measures will be obtained from the accelerograms of the selected ensemble and the
dispersion is indeed usually high. The procedure that has been adopted was to take the
mean of the ductility observations and adjust the polynomial vulnerability function to those
quantities. The fitting is carried out to a 2nd degree polynomial by means of the least
squares method. In order to comply with the boundary conditions of the real problem, the
curve will cross the origin of the (PGA,µ) coordinates system, given that nonzero ductility
makes no sense for zero intensity. This condition will surely affect the curve fitting but the
quality of the adjustment is, however, in general, rather good, something that can be seen
in Figure 6.19 and Figure 6.20.

6.31
Safety Assessment

Vulnerability functions - 123 Vulnerability functions - 213

45 55
P1 r 2=0.98967 2
P1 r =0.99344
50
40
P2 r 2=0.9962 P2 r 2=0.99733
2 45
35
P3 r =0.99709 P3 r 2=0.99779
40
30
Ductility demand

Ductility demand
35

25 30

20 25

20
15
15
10
10
5 5

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

Vulnerability functions - 222 Vulnerability functions - 232

25 20
2 2
P1 r =0.9914 P1 r =0.99174
18
P2 r 2=0.99403 P2 r 2=0.99262
20 P3 r 2=0.9914 16 P3 r 2=0.99174

14
Ductility demand

Ductility demand

15 12

10

10 8

5 4

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

Figure 6.19 – Pier vulnerability functions for short configurations.

The observation of such plots lead to the immediate conclusion that the adjustment is fairly
accurate, which sometimes is even more relevant for higher intensity levels. Indeed,
regardless the type of configuration, in terms of length or regularity, or the height, position
or stress level of the pier that is being considered, the correlation coefficient, r2, is always
extremely close to one.

As for the information on the seismic, transmitted by the vulnerability functions, piers in
short bridges tend to absorb higher ductility levels, which is expected, in the sense that a
shorter deck will probably indicate a stiffer bridge. Within the same configuration, the
central piers or the 7m ones, the lower, repeatedly correspond to the upper curves, the ones
corresponding to higher ductility demand. On the other hand, irregular configurations are

6.32
 Safety Assessment

generally more affected by the seismic action, given that the demand is approximately the
double, or even more, that the regular ones’.

Vulnerability functions - 2222222 Vulnerability functions - 2331312

30
2 2
P1 r =0.99382 60 P1 r =0.98035
P2 r 2=0.99038 P2 r 2=0.98424
25 2
P3 r =0.99032 P3 r 2=0.98866
50
P4 r 2=0.98967 P4 r 2=0.99505
20 P5 r 2=0.99032 P5 r 2=0.99564
Ductility demand

Ductility demand
40
P6 r 2=0.99038 P6 r 2=0.99741
P7 r 2=0.99382 P7 r 2=0.99525
15
30

10
20

5 10

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

Vulnerability functions - 3332111

20
2
P1 r =0.98697
18
P2 r 2=0.99317
16 P3 r 2=0.99445
P4 r 2=0.99747
14
P5 r 2=0.99494
Ductility demand

12 P6 r 2=0.99511
P7 r 2=0.98648
10

0
0 100 200 300 400 500 600 700 800 900 1000
2
Peak ground acceleration (cm/s )

Figure 6.20 – Pier vulnerability functions for long configurations.

In one of the long configurations, the 3332111 one, an inversion of the relative placement
between the vulnerability functions occurs. The ductility demand for piers P4, P5 and P6
increases significantly, becoming higher than the P1 one, which was superior for low
intensity. Such apparently minor finding may become actually quite important, depending
on the position of the seismic action probability density function. A pier with higher
ductility demand for the highest intensity levels may be safer than the rest in a seismic
scenario where low peak ground accelerations are more likely to occur.

At this point, it should be recalled that the vulnerability functions are given by polynomials
fit to the mean of the ductility demands, coming from the dynamic analyses with the ten

6.33
Safety Assessment

real accelerograms. It has been, as a matter of fact, encountered high dispersion in such
predictions, of different earthquake records, even if their peak ground accelerations have
been standardized to the same amount. Figure 6.21 and Figure 6.22 plot the ductility
demand, at each configuration, for the pier at highest load, when subjected to each of the
ground motion records.

140 140

120 123 - P1 120 213 - P2

100 100
Ductility demand

Ductility demand
80 80

60 60

40 40

20 20

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

140 140

120 222 - P2 120 232 - P1/P3

100 100
Ductility demand

Ductility demand

80 80

60 60

40 40

20 20

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

Figure 6.21 – Ductility demand for short configurations.

The scatter is rather visible, and seems more noticeable for irregular bridges, where the
ductility demand is higher, which seems expectable. Dispersion increases significantly
with intensity, given that the higher the nonlinearity level, the more diverse the different
earthquake records response predictions will be. The different coloured dots,
corresponding to the different accelerograms, show, additionally, that the records invert
positions for different intensity levels. There is, unequivocally, great variability

6.34
 Safety Assessment

intrinsically associated to the seismic action. This internal uncertainty, referring to record-
to-record or scaling technique variability, is much more significant than the one associated
to the intensity level one. The main analysis outcome, the failure probability, will certainly
be largely influenced by such parameter.

140 140

120 2222222 - P4 120 2331312 - P4

100 100
Ductility demand

Ductility demand
80 80

60 60

40 40

20 20

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

140

120 3332111 - P4

100
Ductility demand

80

60

40

20

0
0 100 200 300 400 500 600 700 800 900 1000
2
Peak ground acceleration (cm/s )

Figure 6.22 – Ductility demand for long configurations.

Another possible glimpse at the seismic action related variability is the plot of the
vulnerability functions that correspond to seismic demand coming from the real
accelerograms. Figure 6.23 and Figure 6.24 illustrate those functions, again for the piers
with highest demand per configuration. The observations previously made still stand, with
the dispersion being highly visible for the irregular and short configurations. In such cases,
the predictions for the ductility demand may exhibit fluctuation between 100% under and
above the mean values, at the maximum considered peak ground acceleration of 1g,
approximately. The mentioned inversion of relative placement between the vulnerability

6.35
Safety Assessment

functions positions is additionally found when moving from one accelerogram to another.
The global picture is the reinforcement of the awareness of the variability related to the
seismic effects.

140 140

120 123 - vf (P1) 120 213 - vf (P2)

100 100
Ductility demand

Ductility demand
80 80

60 60

40 40

20 20

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

140 140

120 222 - vf (P2) 120 232 - vf (P1/P3)

100 100
Ductility demand

Ductility demand

80 80

60 60

40 40

20 20

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

Figure 6.23 – Vulnerability functions for short configurations using different ground motion
records.

6.36
 Safety Assessment

140 140

120 2222222 - vf (P4) 120 2331312 - vf (P4)

100 100
Ductility demand

Ductility demand
80 80

60 60

40 40

20 20

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
2 2
Peak ground acceleration (cm/s ) Peak ground acceleration (cm/s )

140

120 3332111 - vf (P4)

100
Ductility demand

80

60

40

20

0
0 100 200 300 400 500 600 700 800 900 1000
2
Peak ground acceleration (cm/s )

Figure 6.24 – Vulnerability functions for long configurations using different ground motion
records.

6.2.3.2 Vulnerability function from nonlinear static analysis

As an alternative to the long dynamic analysis traditionally used to predict the seismic
structural effect and, consequently, define the vulnerability function, pushover analysis
may be used instead. Within a predefined nonlinear static procedure, for each intensity
level, herein expressed in peak ground acceleration, the structure is pushed until a level
corresponding to the considered PGA, the performance point is found and the ductility in
curvatures is obtained.

As for this matter two sensitive issues arise, possibly compromising the vulnerability
functions estimate. The first has to do with the use and validity of nonlinear static analysis
itself, a topic that has already been largely discussed, whereas the other refers to the need
of associating peak ground acceleration to a specific pushing level. Such task may be

6.37
Safety Assessment

accomplished by means of the response spectrum that is used within the nonlinear static
procedure (NSP), which is the way of representation of the seismic action in such sort of
methodology. For each intensity level, the response spectrum will therefore be scaled so
that its zero ordinate, the peak ground acceleration, matches the intensity at stake. The
validity of such approach is logical, but arguable, and shall be taken into account when
analysing differences in vulnerability function predictions. Recalling the observations
drawn in Chapter 5, any Nonlinear Static Procedure has proved itself as acceptably able to
predict the response of structural systems, when subjected to earthquake loading.
Consequently, one of the traditional methods, implemented in the European code (CEN,
2005b), the N2 method, has been chosen to obtain pushover-based vulnerability curves.
The pushover analysis may be conventional or adaptive (see Chapter 5 for further details)
and both possibilities have been used. According to the recommendations on the
application of the N2 method, two loading distributions, uniform and 1st mode proportional
(mod and uni), shall be tested. Additionally, two reference nodes have been considered:
centre of mass of the deck and maximum modal displacement (centr and max), leading to a
total of four variants using conventional pushover and one corresponding to the adaptive
pushover, which skips the need for a reference node or load pattern.

The comparison between the vulnerability functions, coming from the different types of
static analyses, is represented, together with the dynamic analysis one (NDA), for each of
the tested configurations, from Figure 6.25 to Figure 6.31.

60 THA (r2=0.99) 60 THA (r2=0.996) 60 THA (r2=0.997)

max mod (r2=0.948) max mod (r2=0.995) max mod (r2=0.991)
B123 - P1 B123 - P2 B123 - P3
max uni (r2=0.982) max uni (r2=0.998) max uni (r2=0.999)
50 50 50
centr mod (r2=0.948) centr mod (r2=0.995) centr mod (r2=0.991)
centr uni (r2=0.982) centr uni (r2=0.998) centr uni (r2=0.999)
Ductility demand

Ductility demand
Ductility demand

40 adapt (r2=0.955) 40 adapt (r2=0.978) 40 adapt (r2=0.997)

30 30 30

20 20 20

10 10 10

0 0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

Figure 6.25 – Vulnerability functions according to different approaches – configuration 123.

6.38
 Safety Assessment

90 THA (r2=0.993) 90 THA (r2=0.997) 90 THA (r2=0.998)

max mod (r2=0.646) max mod (r2=0.985)
B213 - P2 max mod (r2=0.987)
80 B213 - P1 80 80 B213 - P3
max uni (r2=0.958) max uni (r2=0.998) max uni (r2=0.971)
70 centr mod (r2=0.777) 70 centr mod (r2=0.989) 70 centr mod (r2=0.99)
centr uni (r2=0.958) centr uni (r2=0.998) centr uni (r2=0.971)
Ductility demand

Ductility demand

Ductility demand
60 60 60
adapt (r2=0.996) adapt (r2=0.975) adapt (r2=0.963)
50 50 50

40 40 40

30 30 30

20 20 20

10 10 10

0 0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

Figure 6.26 – Vulnerability functions according to different approaches – configuration 213.

50 50 50
THA (r2=0.991) THA (r2=0.994) THA (r2=0.991)
45
max mod (r2=1)
45
max mod (r2=0.997) B222 - P2 45
max mod (r2=1)
B222 - P3
B222 - P1
40 max uni (r2=0.999) 40 max uni (r2=0.998) 40 max uni (r2=0.999)
centr mod (r2=1) centr mod (r2=0.997) centr mod (r2=1)
35 35 35
centr uni (r2=0.999) centr uni (r2=0.998) centr uni (r2=0.999)
Ductility demand

Ductility demand

Ductility demand
30 adapt (r2=0.998) 30 adapt (r2=0.994) 30 adapt (r2=0.998)

25 25 25

20 20 20

15 15 15

10 10 10

5 5 5

0 0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

Figure 6.27 – Vulnerability functions according to different approaches – configuration 222.

45 45 45
THA (r2=0.992) THA (r2=0.993) THA (r2=0.992)
40
max mod (r2=0.999) B232 - P1 40
max mod (r2=0.998) B232 - P2 40
max mod (r2=0.999) B232 - P3
max uni (r2=0.999) max uni (r2=0.993) max uni (r2=0.999)
35 35 35
centr mod (r2=0.999) centr mod (r2=0.998) centr mod (r2=0.999)
centr uni (r2=0.993)
Ductility demand

centr uni (r2=0.999) 30 centr uni (r2=0.999)

Ductility demand
Ductility demand

30 30
adapt (r2=0.997) adapt (r2=0.997) adapt (r2=0.997)
25 25 25

20 20 20

15 15 15

10 10 10

5 5 5

0 0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

Figure 6.28 – Vulnerability functions according to different approaches – configuration 232.

The observation of the plots referring to the short configurations denotes the particular
finding that pushover-based vulnerability functions are, generally, conservative,
overpredicting the dynamic analysis ones, regardless the location of the pier that is being
observed or the ductility level. This tendency is, nevertheless, more evident for regular
configurations where, curiously, the nonlinear static predictions and the nonlinear dynamic
ones are less in agreement. Within each configuration, the agreement between different

6.39
Safety Assessment

curves seems to be worse for the pier with higher ductility demand, especially visible for
the configuration 123, 213 and 232, where the differences among piers are higher.

55 55 55
THA (r2=0.994) THA (r2=0.99) THA (r2=0.99)
50 50 50
max mod (r2=0.857) B2222222 - P1 max mod (r2=0.997) B2222222 - P2 max mod (r2=0.995) B2222222 - P3
45 45 45
max uni (r2=0.987) max uni (r2=0.991) max uni (r2=0.992)
40 centr mod (r2=0.857) 40 centr mod (r2=0.997) 40 centr mod (r2=0.995)
centr uni (r2=0.987) centr uni (r2=0.991) centr uni (r2=0.992)
Ductility demand

Ductility demand

Ductility demand
35 35 35
adapt (r2=0.992) adapt (r2=0.995) adapt (r2=0.995)
30 30 30

25 25 25

20 20 20

15 15 15

10 10 10

5 5 5

0 0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

55 55
THA (r2=0.99) THA (r2=0.99)
50 B2222222 - P4 50 B2222222 - P5
max mod (r2=0.992) max mod (r2=0.995)
45 45
max uni (r2=0.994) max uni (r2=0.995)
40 centr mod (r2=0.992) 40 centr mod (r2=0.995)
centr uni (r2=0.994) centr uni (r2=0.995)
Ductility demand
Ductility demand

35 35
adapt (r2=0.994) adapt (r2=0.995)
30 30

25 25

20 20

15 15

10 10

5 5

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

55 55
THA (r2=0.99) THA (r2=0.994)
50 50
max mod (r2=0.997) B2222222 - P6 max mod (r2=0.857) B2222222 - P7
45 45
max uni (r2=0.996) max uni (r2=0.995)
40 centr mod (r2=0.997) 40 centr mod (r2=0.857)
centr uni (r2=0.996) centr uni (r2=0.995)
Ductility demand
Ductility demand

35 35
adapt (r2=0.995) adapt (r2=0.992)
30 30

25 25

20 20

15 15

10 10

5 5

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

Figure 6.29 – Vulnerability functions according to different approaches – configuration 2222222.

As for the differences between the several pushover approaches, the adaptive analysis is
recurrently the most conservative, being actually the only that always overestimates THA,
an effect that is even more pronounced for the piers with higher ductility demand.
Regarding conventional pushover variants, not much relevant pattern can be found, given
that the different versions are not that far from each other, except for the fact that modal
loading shape seems to yield more overpredicting results, closer to the adaptive pushover
curves. For what concerns long, regular configurations, herein represented by bridge
2222222, the results are not so different from what has been observed for short ones, which

6.40
 Safety Assessment

seems, at a certain level, logical, given that the “disruption” introduced by higher deck
length is somewhat balanced by the regularity in the piers. The results are, however,
steadier from pier to pier, with the previous main observations still standing, regardless of
the demand level. Adaptive pushover analysis is again overpredicting, sometimes largely,
whereas the different versions of conventional pushover do not diverge considerably, with
a tendency for the uniform load shape, together with the centre of mass of the deck as
reference node, to work better.

80 THA (r2=0.98) 80 THA (r2=0.984) 80 THA (r2=0.989)

max mod (r2=0.997) B2331312 - P1 max mod (r2=0.997) B2331312 - P2 max mod (r2=0.984) B2331312 - P3
70 70 70
max uni (r2=0.98) max uni (r2=0.996) max uni (r2=0.976)
centr mod (r2=0.996) centr mod (r2=0.998) centr mod (r2=0.999)
60 60 60
centr uni (r2=0.974) centr uni (r2=0.996) centr uni (r2=0.986)
Ductility demand

Ductility demand

Ductility demand
50 adapt (r2=0.99) 50 adapt (r2=0.998) 50 adapt (r2=0.97)

40 40 40

30 30 30

20 20 20

10 10 10

0 0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

80 THA (r2=0.995) 80 THA (r2=0.996)

max mod (r2=0.948) B2331312 - P4 max mod (r2=0.757) B2331312 - P5
70 70
max uni (r2=0.968) max uni (r2=0.992)
centr mod (r2=0.993) centr mod (r2=0.922)
60 60
centr uni (r2=0.996) centr uni (r2=0.998)
Ductility demand
Ductility demand

50 adapt (r2=0.978) 50 adapt (r2=0.839)

40 40

30 30

20 20

10 10

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

80 THA (r2=0.997) 80 THA (r2=0.995)

max mod (r2=-0.036) B2331312 - P6 max mod (r2=-1.528) B2331312 - P7
70 70
max uni (r2=0.951) max uni (r2=0.562)
centr mod (r2=0.64) centr mod (r2=-1.927)
60 60
centr uni (r2=0.859) centr uni (r2=0.246)
Ductility demand
Ductility demand

50 adapt (r2=0.916) 50 adapt (r2=0.969)

40 40

30 30

20 20

10 10

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

Figure 6.30 – Vulnerability functions according to different approaches – configuration 2331312.

If irregularity is introduced when looking into long configurations, some changes are found
in results output, mainly because the differences in the behaviour of the several piers get
more pronounced. Even if the adaptive pushover remains overestimating is many

6.41
Safety Assessment

situations, for some piers, such as short ones, near to the abutments, it is actually the only
able way to represent the ductility demand using pushover analysis, as in P5, P6 and P7 in
2331312 configuration. For the majority of the remaining piers, the 1st mode proportional
load shape version, using deck central node, is the most reliable. It is additionally
noteworthy that the critical pier, corresponding to the higher ductility demand, has
nonlinear dynamic analysis predictions bounded by those two variants, something that has
been verified for all the configurations.

The second long irregular configuration, 3332111, confirms the tendency that has been
encountered for the rest of the analysed bridges. The piers P6 and P7, short and close to the
abutments are expected to present difficulties in predictions and so it is. For the latter,
again, adaptive pushover analysis is the one managing to capture the ductility effects,
whereas, for the P6 pier, no pushover analysis works. This is, nevertheless, the only
pushover complete inability situation that has been found. Two important aspects shall be
taken into consideration herein: overprediction coming form adaptive pushover is not so
notorious, a part that is, for the configuration at stake, taken by the conventional modal
pushover analysis. It is, however, possible, for a large number of cases, to enclose NDA
vulnerability functions by the predictions of the two most promising pushover techniques.
For this specific configuration, worst results correspond, in any case, to the piers where the
demand is lower.

According to the representation of the several vulnerability functions, considering all sorts
of configurations, there is definitely room for the use of pushover analysis for the
estimation of ductility demand in the piers. Except for a couple or so of exceptional
situations, where adaptive pushover has been the only able to provide acceptable
estimations, generally, the simplest approach, that makes use of conventional pushover, is
enough to obtain highly satisfactory results, namely for the critical pier.

6.42
 Safety Assessment

35 35 35
THA (r2=0.987) THA (r2=0.993) THA (r2=0.994)
max mod (r2=0.995) max mod (r2=0.996) max mod (r2=0.998) B3332111 - P3
30 B3332111 - P1 30 B3332111 - P2 30
max uni (r2=0.992) max uni (r2=0.994) max uni (r2=0.995)

25 centr mod (r2=0.99) 25 centr mod (r2=0.991) 25 centr mod (r2=0.994)

centr uni (r2=0.979) centr uni (r2=0.984) centr uni (r2=0.989)
Ductility demand

Ductility demand

Ductility demand
adapt (r2=0.995) adapt (r2=0.997) adapt (r2=0.979)
20 20 20

15 15 15

10 10 10

5 5 5

0 0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

35 35
THA (r2=0.997) THA (r2=0.995)
30 max mod (r2=0.993) B3332111 - P4 30 max mod (r2=0.746) B3332111 - P5
max uni (r2=0.982) max uni (r2=0.777)

25 centr mod (r2=0.997) 25 centr mod (r2=0.91)

centr uni (r2=0.997) centr uni (r2=0.915)

Ductility demand
Ductility demand

adapt (r2=0.939) adapt (r2=0.785)

20 20

15 15

10 10

5 5

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

35 35
THA (r2=0.995) THA (r2=0.986)
max mod (r2=0.373) max mod (r2=-1.059) B3332111 - P7
30 B3332111 - P6 30
max uni (r2=-0.423) max uni (r2=0.795)

25 centr mod (r2=-0.295) 25 centr mod (r2=0.342)

centr uni (r2=0.734) centr uni (r2=0.13)
Ductility demand
Ductility demand

adapt (r2=0.112) adapt (r2=0.694)

20 20

15 15

10 10

5 5

0 0
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000

Peak ground acceleration (cm/s2) Peak ground acceleration (cm/s2)

Figure 6.31 – Vulnerability functions according to different approaches – configuration 3332111.

In order to clarify, and to some extent quantify the quality of the different pushover
approaches, the results may be presented differently, through the computation of an index
able to represent, even if in a median fashion, the accuracy in the predictions for each pier
of each configuration. A measure that is easy to compute and actually quite effective,
especially when it comes to comparison of different versions, is the root mean square
deviation (RMSD). In statistics, the root mean square deviation is frequently used to
measure the differences between values predicted by a model or an estimator and the
values actually observed, being a good measure of precision. RMSD is given by the square
root of the mean squared error, which is given, in turn, by the differences, also called
residuals, between observed data and corresponding model prediction. A small MRSD
indicates a tight fit of the model to the data. Additionally, in some areas of study, the

6.43
Safety Assessment

RMSD may be used to compare differences between two vectors representing different
predictions for the same variable, measuring the distance between oblong objects,
expressed by vectors, as in Equation (6.14).

∑ (µ − µ NDA )
2
PUSH
(6.14)
RMSD = i =1

Herein, the vectors to compare will be seen as the NDA ductility demand values, µNDA, and
the several pushover based approaches, µPUSH. This sort of statistical measure will allow
the direct numerical comparison of proximity between pushover-based vulnerability
functions and NDA ones. Figure 6.32 and Figure 6.33 plot the root mean square deviation
based on the predictions for each pier, of every configuration.

22 22
N2 max mod N2 max mod
20 B123 N2 max uni
20 B213 N2 max uni
18 N2 centr mod 18 N2 centr mod
N2 centr uni N2 centr uni
16 adapt 16 adapt

14 14
RMSD

RMSD

12 12

10 10

8 8

6 6

4 4

2 2

0 0
P1 P2 P3 P1 P2 P3
Pier Pier

22 22
N2 max mod N2 max mod
20 B222 N2 max uni
20 B232 N2 max uni
18 N2 centr mod 18 N2 centr mod
N2 centr uni N2 centr uni
16 adapt 16 adapt

14 14
RMSD

RMSD

12 12

10 10

8 8

6 6

4 4

2 2

0 0
P1 P2 P3 P1 P2 P3
Pier Pier

Figure 6.32 – Root mean square deviation for short configurations.

6.44
 Safety Assessment

When looking at different versions results for short configurations it is clear to notice that
adaptive pushover analysis does not introduce any relevant improvement with respect to
the conventional approaches, given that is never the approach with lowest RMSD,
independently from the pier that is being considered. Within conventional pushover based
approaches, the results from versions that make use of the same load shape come out
typically close, which denotes the greater importance of such parameter with respect to the
reference node choice. In line with that output, the uniform load pattern generally yields
lower residuals, this is, differences between pushover and NDA-predicted ductility
demand. However, for irregular short configurations, major discrepancy occurs and, for the
critical piers, modal load shape does actually work better, as verified in P1 and P2 in 123
and 213 configurations, respectively, due to their small height together with determinant
location. For regular short configurations, the differences between variants are quite less
noticeable.

22 22
N2 max mod
20 B2222222 N2 max uni
20 B2331312
18 N2 centr mod 18
N2 centr uni
16 adapt 16
N2 max mod
14 14
N2 max uni
N2 centr mod
RMSD

RMSD

12 12
N2 centr uni
10 10 adapt

8 8

6 6

4 4

2 2

0 0
P1 P2 P3 P4 P5 P6 P7 P1 P2 P3 P4 P5 P6 P7
Pier Pier

22
N2 max mod
20 B3332111 N2 max uni
18 N2 centr mod
N2 centr uni
16 adapt

14
RMSD

12

10

0
P1 P2 P3 P4 P5 P6 P7
Pier

Figure 6.33 – Root mean square deviation for long configurations.

6.45
Safety Assessment

For the case of the long configurations, especially for the two irregular ones, containing
short piers, the adaptive pushover usefulness in estimating vulnerability functions is more
visible, contrarily to the regular 2222222, where it is always the poorest approach. The best
versions correspond frequently, in long configurations, to lower RSMDs, which seems to
denote that pushover enables a better to NDA in this sort of bridges. Regarding the rest of
the observations that have been made for short configurations, the main conclusions stand.
Conventional pushover with uniform load shape is recurrently the best option, with
reduced RMSD, whereas the reference node is not a decisive aspect, even if the centre of
mass of the deck tends to be a more reliable choice.

The use of pushover analysis in the estimation of vulnerability functions proves itself
practical and useful, at least for a quick estimate of the sort of ductility demand level that
should be expected for a certain configuration. The use of conventional pushover with 1st
mode proportional loading pattern or the consideration of the maximum modal
displacement as reference node do not represent, however, worthy alternatives. Adaptive
pushover may be used in large configurations, in short, abutment-near piers, so as to
complement or even replace unreliable vulnerability functions obtained with conventional
pushover analysis.

6.2.3.3 Structural effects and failure probability

With the definition of the vulnerability function the probability density curve of the
structural effects becomes easily obtainable by means of equivalence of areas, as detailed
in 6.1.5. The statistical distribution of the seismic effects, S, used in the computation of the
collapse probability of Equation (6.7), becomes therefore defined and able to be used in the
numerical process. The cumulative distribution function of the capacity is the crossed with
the probability density of the seismic effects, through the computation of the convolution
integral.

The capacity has been defined as drift-limited or not, which can result in important
changes on the available ductility and consequent safety evaluation. Table 6.2 presents the
failure probability, in numbers, using nonlinear dynamic analysis to compute the seismic
effects, with or without limiting the drift capacity of the piers.

6.46
 Safety Assessment

Table 6.2 – Failure probability using traditional NSA-NDA procedure (local uncertainty).
Failure
Pier 1 Pier 2 Pier 3 Pier 4 Pier 5 Pier 6 Pier 7
Probability
w/o drift control 2.75E-02 3.85E-03 1.71E-09 - - - -
123
w/ drift control 2.75E-02 3.96E-03 3.82E-03 - - - -
w/o drift control 5.02E-05 1.34E-02 1.58E-11 - - - -
213
w/ drift control 6.17E-05 1.34E-02 4.09E-04 - - - -
w/o drift control 5.36E-04 2.48E-03 5.37E-04 - - - -
222
w/ drift control 6.15E-04 2.55E-03 6.16E-04 - - - -
w/o drift control 6.53E-04 4.99E-05 6.54E-04 - - - -
232
w/ drift control 7.50E-04 5.41E-03 7.51E-04 - - - -
w/o drift control 1.13E-06 2.86E-04 1.34E-03 2.32E-03 1.34E-03 2.87E-04 1.14E-06
2222222
w/ drift control 4.68E-07 2.97E-04 1.41E-03 2.43E-03 1.42E-03 2.98E-04 4.70E-07
w/o drift control 5.39E-07 2.64E-08 6.49E-06 1.25E-02 1.81E-10 5.58E-03 1.18E-08
2331312
w/ drift control 1.16E-07 1.01E-03 2.44E-03 1.25E-02 1.98E-04 5.58E-03 3.54E-10
w/o drift control 1.50E-08 1.65E-04 2.74E-04 3.59E-04 2.26E-04 4.96E-05 4.24E-07
3332111
w/ drift control 2.93E-03 1.38E-02 1.27E-02 3.76E-04 2.26E-04 4.96E-05 4.24E-07

The highlighted fields correspond to the pier with highest failure probability, which is
assumed as the bridge failure probability. The critical pier is generally the shortest one,
type 1, 7 meters height, (as in configurations 123, 213 or 2331312) or the central one,
when there are no significant differences in height (as in regular configurations 222, 232,
2222222). Furthermore, for most cases, the behaviour-commanding pier stands, whether
the drift limitation is considered or not. The exception occurs for configurations 232 and
3332111, where the critical pier changes with the drift restriction, herein established at a
maximum of 5%. This shifting takes place especially in type 3 piers, taller, 21 meters
height, given that a relatively small curvature at the base of the pier will correspond, due to
its significant height, to a relatively large displacement at the top, leading the drift to
rapidly reach the imposed limit. Within 232 configuration, the central pier becomes
therefore the decisive one, as opposed to the shorter, next to the abutments, type 2 piers,
which ruled the bridge behaviour when no drift limitation was imposed. In 3332111
configuration, in turn, the critical pier went from the central pier P4, 14 meters height, to a
taller one, closer to the abutments, again a type 3 one. Curious is the fact that a lower mean
ductility capacity, induced by the drift limitation, does not necessarily lead to a lower
failure probability. In fact, looking at the extreme piers P1 and P7, both type 2 ones, in
2222222 and 2331312 configurations, the described situation can be observed, where
carrying out a 5% drift limitation for the piers capacity, the collapse probability

6.47
Safety Assessment

diminishes, in one or two orders of magnitude. The interpretation is quite straightforward

and has to do with two aspects. Firstly, recurring to Table 6.3, it can be noticed that, for the
particular case of such piers, the available ductility becomes little inferior when drift-
limited, with a reduction of less than 4%.

Table 6.3 – Ductility capacity (mean and standard deviation) considering, or not, drift limitation.
Ductility Capacity Pier 1 Pier 2 Pier 3 Pier 4 Pier 5 Pier 6 Pier 7
w/o drift Mean 34.27 34.49 34.00 - - - -
control STD 3.28 3.29 3.30 - - - -
123
w/ drift Mean 34.27 34.03 14.20 - - - -
control STD 3.28 2.73 1.50 - - - -
w/o drift Mean 34.15 34.63 34.00 - - - -
control STD 3.28 3.29 3.30 - - - -
213
w/ drift Mean 32.91 34.63 14.22 - - - -
control STD 2.32 3.29 1.50 - - - -
w/o drift Mean 34.14 34.49 34.14 - - - -
control STD 3.28 3.29 3.28 - - - -
222
w/ drift Mean 32.90 34.03 32.90 - - - -
control STD 2.31 2.73 2.31 - - - -
w/o drift Mean 34.14 34.35 34.14 - - - -
control STD 3.28 3.28 3.28 - - - -
232
w/ drift Mean 32.89 17.30 32.89 - - - -
control STD 2.31 1.63 2.31 - - - -
w/o drift Mean 34.16 34.43 34.35 34.38 34.35 34.43 34.16
control STD 3.28 3.29 3.28 3.28 3.28 3.29 3.28
2222222
w/ drift Mean 32.95 33.88 33.66 33.73 33.66 33.88 32.95
control STD 2.34 2.66 2.56 2.60 2.56 2.66 2.34
w/o drift Mean 34.16 34.29 34.22 34.51 34.23 34.57 34.16
control STD 3.28 3.28 3.28 3.29 3.28 3.29 3.28
2331312
w/ drift Mean 32.95 16.77 16.12 34.51 16.14 34.57 32.96
control STD 2.34 1.59 1.57 3.29 1.57 3.29 2.33
w/o drift Mean 34.02 34.29 34.22 34.38 34.49 34.57 34.29
control STD 3.29 3.28 3.28 3.28 3.29 3.29 3.28
3332111
w/ drift Mean 14.32 16.77 16.10 33.73 34.49 34.57 34.29
control STD 1.50 1.59 1.57 2.60 3.29 3.29 3.28

On the other hand, corresponding standard deviation is nearly one third lower, which alters
significantly the shape of the assumed normal distribution, contracting it, an effect that
becomes more significant than the shifting caused by the reduction in the mean value.
Better understanding is enabled by Figure 6.34, which plots the distributions at stake.

6.48
 Safety Assessment

FR (w/o drift limitation)

FR (w/ drift limitation)
0.8
fS

0.6

0.4

20 30
Ductility
0.2

0
0 10 20 30 40 50
Ductility

Figure 6.34 – Drift control influence in pier P1, configuration 2222222, behaviour.

It is possible to observe that the important region for the failure probability computation,
according to Equation (6.7), is the interval where the functions cross each other, given that
for the rest of the ductility domain the ordinates of FR or fS are residual. Within that zone,
the cumulative distribution function corresponding to the drift limited capacity is under the
non-restrained one, notwithstanding the lower higher mean value. For that reason, the
failure probability, on the latter case, turns out higher, even if mean ductility capacity is
equally superior. Additionally, such detected inversion in the expected behaviour occurs
for the piers with extremely low failure probability, confirmed by the probability density
function of the seismic effects, fS, in Figure 6.34, corresponding to small orders of
magnitude from 10-6 to 10-10, something that definitely calls out some unpredictability on
the numerical computation.

Collapse probability for the several bridges is presented in Figure 6.35, allowing in
particular the relative comparison of the different configurations performance. Results
have been sorted in crescent order of failure probability with drift limitation.

6.49
Safety Assessment

0
10
w/o drift control
-1
10
w/ drift control
Failure probability

-2
10

-3
10

-4
10

-5
10
3332111 232 2222222 222 2331312 213 123
Bridge configuration

Figure 6.35 – Failure probability, according to bridge configuration – traditional NSA-NDA

approach.

Globally, there is agreement between considering the top deformation of the piers
limitation or not, given that the trend of results is similar. The exception occurs when a
type 3 pier becomes the most vulnerable one, such as in configurations 232 and 3332111,
previously mentioned. Together with the numbers, the plot suggests, therefore, that
capacity of piers of 21 meters height, the tallest ones, was overestimated. It can be said, in
addition, that the maximum drift constraint somehow smoothes the results, turning them
more uniform according to more realistic conditions. Moreover, the distinction between
regular and irregular configurations is immediate from Figure 6.35, the latter being more
vulnerable, all in the right side of the plot. A final observation goes to the fact that high
orders of magnitude have been found: 10-2 in irregular bridges and 10-4 to 10-3 for regular
ones, which is, to some extent, significant.

Once more the seismic action will have a preponderant part, and the failure probability
that, traditionally, is computed using mean ductility demand, will vary considerably from
one earthquake record to another. The effects caused by such variability will be, however,
of distinct magnitude, depending on the bridge configuration or the pier characteristics.
Figure 6.36 illustrates, for short configurations, at each pier, the failure probability
computed individually, employing a single vulnerability function, coming from each of the
ten used ground motion records, or computed through the mean vulnerability function. The
same output, for long configurations, is presented in Figure 6.37. Nonlinear dynamic
analysis was used.

6.50
 Safety Assessment

0 0
10 10

-2
10 -5
10
Failure probability

Failure probability
-4
10
-10
10
-6
10
-15
10
-8
10

-20
213
-10 10
10
individual
123 mean
-12 -25
10 10
P1 P2 P3 P1 P2 P3
Pier Pier

0 0
10 10

-5 -5
10 10
Failure probability

Failure probability

-10 -10
10 10

-15 -15
10 10

222 232
-20 -20
10 10
P1 P2 P3 P1 P2 P3
Pier Pier

Figure 6.36 – Failure probability for short configurations.

From the observation of the plots, for both types of configuration, the finding that the
dispersion within the seismic response prediction has spread up itself to the final collapse
probability computation is notorious. Equally confirmed is the perception that larger
variability is found for the critical piers, the ones with higher demand and, consequently,
higher failure probability. This will certainly constitute the essential issue, concerning the
relevance of the dispersion introduced by the seismic action, when the safety assessment is
carried out by means of nonlinear dynamic analysis. Indeed, especially for the case of
critical piers, the question that arises has essentially to do with the validity of using the
mean demand obtained with a certain number of accelerograms instead of the maximum
measured response. EC8 (CEN, 2005b) indicates that the average response effects may be
assumed if at least seven independent records are used. If less than that number of input
motions, with a minimum of three, is used, then the maximum response of the ensemble
shall be assumed.

6.51
Safety Assessment

0 0
10 10

-5 -5
10 10

-10 -10
10 10

Failure probability
Failure probability

-15 -15
10 10

-20 -20
10 10

-25 -25
10 10

-30 -30
10 10

2222222 2331312
-35 -35
10 10
P1 P2 P3 P4 P5 P6 P7 P1 P2 P3 P4 P5 P6 P7
Pier Pier

0
10

-5
10
Failure probability

-10
10

-15
10

-20
10 3332111
individual
mean
-25
10
P1 P2 P3 P4 P5 P6 P7
Pier

Figure 6.37 – Failure probability for long configurations.

Herein, the failure probability computed from the application of a single ground motion
record can be substantially higher than the 10-records average demand one as the just
presented plots indicate, with differences up to ten times more being encountered. Such
findings are, nevertheless, associated to a standardizing criterion, applied to the
accelerograms, based on peak ground acceleration, which definitely represents a part in the
results that can not be neglected. In any case, average results when carrying out dynamic
analysis with real earthquake records, must be used with care.

Furthermore, it can be stated, at least to some extent, that, considering the general
behaviour of the bridges across the several piers, variability in collapse probability is larger
for regular configurations than for irregular ones. In the latter, there is typically a pier, or a
couple of them, considerably more affected, where large discrepancies in predictions are
found, whereas, among the rest, there is general consensus in failure probability estimates.

6.52
 Safety Assessment

The failure probability calculation has been carried out as well as a function of the sort of
vulnerability function origin: nonlinear static analysis or nonlinear dynamic analysis. The
differences that have been found between such vulnerability curves may assume higher or
lower importance, depending on the shape of the seism action probability density function.
With the purpose of continuing to draw attention to those several demand prediction
possibilities, given the major importance of such variable within the traditional
methodology, collapse probability for each pier, using each variant, has been computed.
The failure probability, at the different piers, according to different approaches to estimate
the vulnerability function, is plotted, for short configurations, in Figure 6.38 and, for long
ones, in Figure 6.39.

0.08
max mod max mod
max uni
B213 max uni
0.07 B123
centr mod 0.25 centr mod
0.06 centr uni centr uni
adapt adapt
Failure probability
Failure probability

0.2
THA THA
0.05

0.04 0.15

0.03
0.1

0.02

0.05
0.01

0 0
P1 P2 P3 P1 P2 P3
Pier Pier

0.04 0.04
max mod max mod
0.035 B222 max uni 0.035 B232 max uni
centr mod centr mod
0.03 centr uni 0.03 centr uni
adapt adapt
Failure probability

Failure probability

THA THA
0.025 0.025

0.02 0.02

0.015 0.015

0.01 0.01

0.005 0.005

0 0
P1 P2 P3 P1 P2 P3
Pier Pier

Figure 6.38 – Failure probability, for short configurations, using different vulnerability functions.

The first immediate remark induced by the plots is the intense underestimation given by
nonlinear dynamic analysis, more prominent for regular, short configurations. On the other

6.53
Safety Assessment

hand adaptive pushover and conventional pushover based procedures, with 1st mode
proportional load shape, regardless the reference node, yield the highest and most
overpredicting failure probabilities, with respect to NDA numbers. Larger differences are
expectedly found for the case of the critical piers.

0.07 0.2
max mod max mod
B2222222 0.18 B2331312
0.06
max uni max uni
centr mod 0.16
centr mod
centr uni centr uni
0.05 adapt 0.14 adapt
Failure probability

Failure probability
THA THA
0.12
0.04
0.1
0.03
0.08

0.02 0.06

0.04
0.01
0.02

0 0
P1 P2 P3 P4 P5 P6 P7 P1 P2 P3 P4 P5 P6 P7
Pier Pier

0.09
max mod
0.08 B3332111 max uni
centr mod
0.07 centr uni
adapt
Failure probability

0.06 THA

0.05

0.04

0.03

0.02

0.01

0
P1 P2 P3 P4 P5 P6 P7
Pier

Figure 6.39 – Failure probability, for long configurations, using different vulnerability functions.

With respect to long configurations, results tend to be more uniform, with the main
discrepancies occurring at the piers submitted to higher demand. Even though nonlinear
dynamic analysis is not as overestimated as for short configurations, adaptive pushover
continues to be the methodology yielding higher collapse probability. The tendency for the
pushover based techniques to play largely in a conservative mode is therefore confirmed.
This surely constitutes a motivating feature, given that simplified procedures are certainly
expected to behave cautiously.

6.54
 Safety Assessment

In agreement with the typical definition of the failure probability of a bridge, assumed as
the highest among the piers’ individual collapse probabilities, Figure 6.40 sums the results
for all the bridges in the same plot. Given that the relevance of the reference node in the
final results is minor, such comparison for the entire set of bridges will differentiate
conventional pushover in terms of load pattern only.

A first conclusion that can be reinforced, in agreement with what has already been seen,
when looking at the comparison for the whole set of bridges, is that regular configurations
have the lowest collapse probabilities, even though larger variability has been found from
different approaches predictions in those bridges. Regarding the sort of pushover to use, if
a simple static analysis is intended, modal load pattern does not introduce any significant
advantage when compared to the uniform one, which is in addition even simpler to apply.
Adaptive pushover approach, in general the most overpredicting of all, does not seem
gainful as well. No important differences are found concerning the length of the
configuration: short bridges are slightly less safe than the long ones together with a little
less dispersion within different variants.

mod
uni
0.2 adapt
THA
Failure probability

0.15

0.1

0.05

0
123 213 222 232 2222222 2331312 3332111
Bridge configuration

Figure 6.40 – Failure probability, according to bridge configuration.

The results that have been summarized indicate a tendency for the static analysis approach
to be conservative, useful, even in its simpler form, the conventional uniform loading
shape version, with respect to the dynamic analysis. Within a methodology where the
structural effects need to be estimated through a large number of nonlinear analyses, this

6.55
Safety Assessment

feature constitutes an important advantage, at least for a preliminary analysis. The used
procedure, considering local uncertainty consideration, has proved itself quite influenced
by the extreme variability that characterizes the seismic action. Additionally, the capacity
definition, in terms of practical limitations or statistical characterization plays a rather
important part, easily conditioning the most vulnerable pier. Moreover, the method has the
clear advantage of not needing to recur to distribution fitting of the nonlinear seismic
effects, using an equivalent area under probability density functions based numerical
procedure instead.

6.2.4 Failure probability with global uncertainty LHS approach

Within the alternative that considers the uncertainty of all the variables simultaneously, in
a global fashion, at each repetition of the entire procedure the demanded ductility, µS, is
obtained at the base of the pier and compared to the respective available one, µR. The
difference between these two quantities is computed and, therefore, a sample of N
realizations for such variable is obtained, which will enable achieving the collapse
probability as the zero-ordinate of the corresponding cumulative distribution (µR–µS)
function.

6.2.4.1 Simulation process calibration

Contrarily to the continuous transformation of the seismic probability density function into
the corresponding structural effects one, from a global simulation procedure perspective
herein used, the effects are characterized in terms of a sample, of a specific predetermined
size, of ductility demand numbers. Such discrete data will follow a statistical distribution,
requiring characterization. Goodness of fit tests constitute the best tool for that purpose and
may be carried out for the relevant distributions. Ductility demand is basically a function
of the structural and material properties, which typically follow a normal distribution, and
the seismic intensity, characterized by an extreme value distribution. Consequently, the
structural effects will likely follow one of those distributions, with the goodness of fit tests
focusing on such possibility.

The first parameter to test in this sort of procedure is the size of the sample that will
enhance superior results, in the stability sense, without compromising computational effort.
Similar endeavour focused capacity definition and, according to results that have been

6.56
 Safety Assessment

presented in Section 6.2.1, a minimum of one hundred realizations were necessary to reach
consistency in predictions. For the present case, 10, 20, 50, 100, 200 and 500 repetitions of
the procedures have been carried out, fitting an extreme value distribution to the difference
between the computed capacity and the structural demand, assuring that both variables
were computed using the same nth randomly simulated material properties, seismic
intensity and earthquake record.

Figure 6.41 illustrates the evolution of the distribution fitting to structural effects data for
the short, regular configuration 222. Plotting the results for all the configurations and piers
would be extremely exhaustive, hence, a simple regular pier has been chosen to test the
method. Histograms for the different sampling sizes are plotted for the central P2 pier, the
one with highest demand, as well as the corresponding probability density function of the
extreme value distribution that best fits the data.

The influence of the sampling size is rather evident from the observation of the plots, with
the histograms corresponding to 10, 20, 50 and 100 realizations being typically less
adjustable to the expected extreme value distribution, notwithstanding the fact that
corresponding hypotheses of the data following such distribution were not rejected by
Kolmogorov-Smirnov tests in such cases. Logically, 10 and 20 realizations samples will
hardly have statistical meaning and even 50 as sample size is extremely arguable. In any
case, such numbers have been included in the parametric study not only to confirm such
evidence but to test the method’s toughness and to help finding out its optimum application
mode as well. The turning point is, nevertheless, fairly marked and corresponds to 200
realizations, with no conspicuous advantage in moving on to 500 realizations. It seems
reasonable that, within a failure probability computation process, a superior size of the
samples ought to be needed, in comparison with the hundred stabilizing number found for
the capacity characterization. In addition, the plots refer to the central pier of a regular
configuration, which will expectedly reduce instability levels within the results coming
from nonlinear analysis.

6.57
Safety Assessment

2 4
222 - P2 - 10 realizations 222 - P2 - 20 realizations
1.8
3.5
KS test - 0 KS test - 0
1.6
3
1.4
2.5
1.2

1 2

0.8
1.5
0.6
1
0.4
0.5
0.2

0 0
-5 0 5 10 15 20 25 30 35 40 -5 0 5 10 15 20 25 30 35 40
µR - µS µR - µS

10 25
222 - P2 - 50 realizations 222 - P2 - 100 realizations
9
KS test - 0 KS test - 0
8 20

6 15

4 10

2 5

0 0
-5 0 5 10 15 20 25 30 35 40 -5 0 5 10 15 20 25 30 35 40
µR - µS µR - µS

60 160
222 - P2 - 200 realizations 222 - P2 - 500 realizations
140 KS test - 0
50 KS test - 0
120

40
100

30 80

60
20

40

10
20

0 0
-5 0 5 10 15 20 25 30 35 40 -10 -5 0 5 10 15 20 25 30 35 40
µR - µS µR - µS

Figure 6.41 – (µR - µS) histogram and adjusted extreme value distribution probability density
function for pier P2 in configuration 222, for different numbers of realizations.

The possibility of other configurations, likely to behave in a trickier fashion, requiring

more than 200 realizations is real and must be examined. The variability introduced by the

6.58
 Safety Assessment

heterogeneous seismic action has been herein widely recognized and, therefore, the study
of such effect, together with the needed number of realizations, for all the configurations,
is of high interest. The complete procedure has been carried out including the accelerogram
as an input variable or using each earthquake record independently to compute the collapse
probability. Figure 6.42 illustrates the failure probability obtained for each configuration,
in green, using a single independent ground motion record, or, in red, randomly selecting
it, as a variable. The failure probability is taken from the pier subjected to highest demand.

0 0
10 10
50 realizations 100 realizations
-1 -1
10 10
Failure probability

Failure probability
-2 -2
10 10

-3 -3
10 10

-4 -4
10 10

-5 -5
10 10

-6 -6
10 10
123 213 222 232 2222222 2331312 3332111 123 213 222 232 2222222 2331312 3332111
Bridge configuration Bridge configuration

0 0
10 10
200 realizations 500 realizations
-1 -1
10 10
Failure probability
Failure probability

-2 -2
10 10

-3 -3
10 10

-4 -4
10 10

-5 -5
10 10

-6 -6
10 10
123 213 222 232 2222222 2331312 3332111 123 213 222 232 2222222 2331312 3332111
Bridge configuration Bridge configuration

Figure 6.42 – Failure probability, using each record in separate or considering the record as an
input variable, for different numbers of realizations.

There are two important aspects to retain from the observation of the plots. The first is the
confirmation of the considerable dispersion that this sort of procedures involve, as a
consequence of the disperse nature of the seismic action. The use of ten different records
makes the collapse probability to range from order of magnitude 10-6 to 100, for the lower

6.59
Safety Assessment

sampling sizes. The other finding is that, from 200 realizations on, but especially visible
when using samples of 500 realizations, the scatter reduces approximately one third, with
no failure probabilities under 10-4. It is significant to recognize, at the same time, that the
reduction has occurred upwards, that is, the reduction in the variability has been made by
eliminating the lowest values, which denotes that a higher refinement level of the global
simulation procedure enables a safer assessment of the seismic vulnerability.

When analysing the evolution of results with the number of realizations, a sort of internal
variability of the method, using global Latin Hypercube simulation, is being observed. The
external variation of the results can be seen when carrying out again the whole procedure
more than once, repeating all the simulation process. Three repetitions of the method have
been performed and the changes in the obtained failure probability have been plotted
against each other. Again, the plots for the different series of repetitions are plotted, in
Figure 6.43, for the central pier of the regular, short 222 configuration.

The observation of the plots corresponding to the use of the different ground motion
records individually, indicates that, for all the accelerograms, as expected, there is
considerable variability when using 10 or 20 realizations, whether within the method,
across the number of realizations, or from one repetition to the other. If 50 or 100
realizations are carried out, the scatter reduces to median levels and stability is definitely
reached when samples of 200 or 500 realizations are used. The discrepancy of the results
that has been found between different records is certainly due to their particular
characteristics, given that they are real ones. Similar studies have been conducted in the
recent past (Carvalho, 2009), leading to equally encouraging results regarding the use of
Latin Hypercube sampling in a global safety assessment procedure, where the dispersion in
both directions was similarly lower. Nevertheless, such work was carried out taking
coefficients of variation for the material properties as constant and lower, especially for the
concrete, which will several times govern the structural behaviour, near the rupture.
Herein, a much realistic approach has been adopted for the material properties statistical
models, which enhances even more the Latin Hypercube algorithm performance.

6.60
 Safety Assessment

0 0
10 10

222 - P2 - acel. 1 222 - P2 - acel. 2

-2 -2
10 10
Failure probability

Failure probability
-4 -4
10 10

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
20 50 100 200 500 20 50 100 200 500
Number of realizations Number of realizations
0 0
10 10

222 - P2 - acel. 3 222 - P2 - acel. 4

-2 -2
10 10
Failure probability

Failure probability
-4 -4
10 10

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
20 50 100 200 500 20 50 100 200 500
Number of realizations Number of realizations
0 0
10 10

222 - P2 - acel. 5 222 - P2 - acel. 6

-2 -2
10 10
Failure probability

Failure probability

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
20 50 100 200 500 20 50 100 200 500
Number of realizations Number of realizations
0 0
10 10

222 - P2 - acel. 7 222 - P2 - acel. 8

-2 -2
10 10
Failure probability

Failure probability

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
20 50 100 200 500 20 50 100 200 500
Number of realizations Number of realizations
0 0
10 10

222 - P2 - acel. 9 222 - P2 - acel. 10

-2 -2
10 10
Failure probability
Failure probability

-4 -4
10 10

-6 -6
10 10

-8 -8
10 10

-10 -10
10 10
20 50 100 200 500 20 50 100 200 500
Number of realizations Number of realizations

Figure 6.43 – Failure probability for pier P2 in configuration 222, for different numbers of
realizations, using each of the earthquake records.

6.61
Safety Assessment

Applying the procedure as originally thought, this is, taking the type of record as an input
variable, which follows a uniform distribution, the same plots can be drawn, repeating
again the whole analysis for three independent times. The sampling size has equally been
tested by means of the number of realizations. The behaviour of the failure probability with
the number of realizations, computed for the most vulnerable pier at each configuration for
different series, is hence illustrated in Figure 6.44.

0 0
10 10
123 - P1 213 - P2
-1 -1
10 10
Failure probability

Failure probability
-2 -2
10 10

-3 -3
10 10

-4 -4
10 10

-5 -5
10 10

-6 -6
10 10
50 100 200 500 50 100 200 500
Number of realizations Number of realizations
0 0
10 10

222 - P2 232 - P2
-1 -1
10 10
Failure probability

Failure probability

-2 -2
10 10

-3 -3
10 10

-4 -4
10 10

-5 -5
10 10

-6 -6
10 10
50 100 200 500 50 100 200 500
Number of realizations Number of realizations
0 0
10 10
2222222 - P4 2331312 - P4
-1 -1
10 10
Failure probability

Failure probability

-2 -2
10 10

-3 -3
10 10

-4 -4
10 10

-5 -5
10 10

-6 -6
10 10
50 100 200 500 50 100 200 500
Number of realizations Number of realizations
0
10
3332111 - P2
-1
10
Failure probability

-2
10

-3
10

-4
10

-5
10

-6
10
50 100 200 500
Number of realizations

Figure 6.44 – Failure probability of the critical pier, according to different numbers of realizations,
considering earthquake record as input variable.

6.62
 Safety Assessment

Given that higher scatter has been encountered when using 10 and 20 realizations and for
the sake of computational demand reduction, the extension of such parametric analysis to
the rest of the configurations has been carried out disregarding such sampling sizes.

The immediate perception is that, if the accelerogram is considered as an additional input

variable for the global simulation procedure, the scatter of the method is still barely
noticeable. The methodology proves itself to be resistant to the introduction of a new
variable of high variability, such as the seismic action is, in intensity and type of record.
The difference from one repetition to another is indeed minimal, except for the 2222222
configuration. In terms of sample size, collapse probability stabilizes after 200 realizations.

6.2.4.2 Structural effects and failure probability

Based on the considerations taken from the parametric study carried out on the number of
realizations, the computation of the failure probability has been considered to be taken
from the application of the Latin Hypercube sampling using samples with 200 realizations.
Again, the calculations have taken into account the drift capacity limitation of the piers.
The resulting numbers for the collapse probability, carrying out drift control or not, are
presented in Table 6.4, where the highlighted values correspond to the critical pier.

Table 6.4 – Failure probability using the global simulation LHS approach.
Failure
Pier 1 Pier 2 Pier 3 Pier 4 Pier 5 Pier 6 Pier 7
Probability
w/o drift control 1.37E-02 1.37E-03 1.59E-04 - - - -
123
w/ drift control 1.37E-02 6.63E-04 5.41E-03 - - - -
w/o drift control 1.48E-04 2.06E-02 7.60E-05 - - - -
213
w/ drift control 7.89E-06 2.06E-02 1.63E-03 - - - -
w/o drift control 7.92E-04 2.51E-03 7.92E-04 - - - -
222
w/ drift control 1.21E-04 1.43E-03 1.21E-04 - - - -
w/o drift control 1.44E-03 3.99E-04 1.44E-03 - - - -
232
w/ drift control 5.77E-04 5.19E-03 5.77E-04 - - - -
w/o drift control 2.16E-04 9.05E-04 3.12E-03 5.54E-03 3.12E-03 9.05E-04 2.16E-04
2222222
w/ drift control 7.53E-06 2.55E-04 1.52E-03 3.47E-03 1.52E-03 2.55E-04 7.53E-06
w/o drift control 1.55E-03 6.31E-04 4.88E-04 3.86E-02 9.18E-05 7.78E-03 1.25E-04
2331312
w/ drift control 3.74E-04 5.44E-03 5.24E-03 3.86E-02 4.33E-04 7.78E-03 7.23E-06
w/o drift control 3.45E-04 1.97E-03 1.04E-03 4.05E-04 2.29E-04 1.20E-04 8.59E-05
3332111
w/ drift control 1.51E-02 4.83E-02 3.19E-02 1.93E-04 2.29E-04 1.20E-04 8.59E-05

6.63
Safety Assessment

The reduction of the capacity of the piers, when applicable, through the consideration of a
drift limit does not necessarily turn out into a higher failure probability. This is actually a
tendency that had already been found in the traditional methodology results, related to the
probabilistic nature of the procedures, which involve distribution fitting and convolution
integrals in the sensitive corresponding tails regions. Nevertheless, and again not a novelty,
such situation occurs mainly for medium height piers, usually near to the abutments. Such
piers, as verified previously with Table 6.3, are little affected by the drift control, in terms
of mean capacity, whereas the coefficient of variation is quite reduced. This will contribute
to the changing in the shape of the probability density function, altering the expected
failure probability. The effect caused by the drift limitation in the (µR – µS) distribution is
illustrated in Figure 6.45 for the case of piers P1 and P3 in 222 configuration.

When the drift control is activated, the mean ductility capacity is slightly inferior, as it can
be seen from the maximum of the probability density functions. On the other hand, due to
the reduction in the dispersion, the shape of the probability density function for the case of
drift capacity reduction gets more contracted and becomes superior to the non limited
capacity one for a margin between demand and capacity, µR - µS, around 20.

0.12
f(µ R - µ S) w/o drift control
f(µ R - µ S) w/ drift control
0.1 -3
x 10
1

0.08 0.8

0.6

0.06 0.4

0.2
0.04
0
-10 -5 0 5 10
µR - µS
0.02

0
-10 0 10 20 30 40 50
µR - µS

Figure 6.45 – Drift control influence in piers P1/P3, configuration 222, behaviour.

6.64
 Safety Assessment

Given that failure probability is computed through the area under the probability density
function for safety margins under zero, in the end, drift limitation will hence correspond to
lower vulnerability, as illustrated by the detail of the relevant region on the right.

The effect of considering or not drift limitation when it comes to the accurateness of the
distribution fitting to the variable corresponding to the difference (µR - µS), is not much
visible. Indeed, not all the configurations had their failure probability commanded by drift
limited capacity, as confirmed through Table 6.4, remaining the graphical verification that
distribution fitting is not worsened by the inclusion of such criteria. The configurations
where such is actually relevant are the ones with critical piers of type 2 or type 3: 222, 232,
2222222 and 3332111. Figure 6.46 illustrates the difference in the histograms and
corresponding probability density functions of the adjusted distribution, for the critical pier
of those configurations.

The observation of the plots indicates that the drift limitation of the ductility capacity, µS,
does not induce quality loss, to what concerns the distribution fitting to the variable µR - µS.
when the shifting occurs, typically for the type 3 piers, as already detailed before, the
probability density function can become even better adjusted, which seems to happen for
configuration 232 and 3332111, the latter with the same fitting quality, at least. When the
critical pier is of type 2, the effect of the drift is not so pronounced, no significant shifting
in µR - µS axis occurs and, even though the adjustment is not compromised, the shape of
the best-fitting extreme value distribution becomes visibly more tightened, reducing the
failure probability, which is in agreement to what had been discussed and confirmed in
Figure 6.45.

6.65
Safety Assessment

60 70
222 - P2 222 - P2
(w/o drift control) 60 (w/ drift control)
50

50
40

40
30
30

20
20

10
10

0 0
-10 0 10 20 30 40 50 -10 0 10 20 30 40 50
µR - µS µR - µS
45 50
232 - P2 232 - P2
40 45
(w/o drift control) (w/ drift control)
40
35

35
30
30
25
25
20
20
15
15

10
10

5 5

0 0
-10 0 10 20 30 40 50 -5 0 5 10 15 20 25 30
µR - µS µR - µS
80 80
2222222 - P4 2222222 - P4
70 (w/o drift control) 70 (w/ drift control)

60 60

50 50

40 40

30 30

20 20

10 10

0 0
-30 -20 -10 0 10 20 30 40 50 -30 -20 -10 0 10 20 30 40
µR - µS µR - µS
45 45
3332111 - P2 3332111 - P2
40 (w/o drift control) 40 (w/ drift control)
35 35

30 30

25 25

20 20

15 15

10 10

5 5

0 0
-5 0 5 10 15 20 25 30 35 40 -10 -5 0 5 10 15 20
µR - µS µR - µS

Figure 6.46 – (µR - µS) histogram and adjusted extreme value distribution probability density
function, for configurations affected by capacity drift control.

6.66
 Safety Assessment

Failure probability for the different configurations is summarized in Figure 6.47, sorted in
crescent order of results considering drift limitation.

0
10
w/o drift control
-1
10 w/ drift control
Failure probability

-2
10

-3
10

-4
10

-5
10
3332111 232 2222222 222 2331312 213 123
Bridge configuration

Figure 6.47 – Failure probability, according to bridge configuration – global simulation LHS
approach.

Similarly to the traditional safety assessment procedure, which considers the variability of
each component separately, there is a fairly consensual evolution of the collapse
probability across the different configurations, whether drift limitation is carried out or not.
Major differences occur for the configurations where the critical pier is a type 3 one, 21
meters high, the one that is most affected by the capacity reduction. Again the distinction
between regular and irregular bridges is extremely pronounced, the primer being safer,
together with a tendency for the short configurations to be less vulnerable as well. The
aforementioned inversion in the expected behaviour of type 2 piers is visible for the
configurations 222 and 2222222, where the critical pier fits such profile.

6.2.5 Comparison of methodologies

The final focus of this section will be the comparison of the failure probability results
obtained with the different methodologies, which differ essentially on how the uncertainty
of the variables is accounted for: local or globally. The distinction in terms of drift limiting
the results or not is carried out as well, so as to evaluate to what extent such parameter
influences the agreement of the procedures. The collapse probability obtained for each of
the procedures, drift limiting or not the capacity, is plotted in Figure 6.48.

6.67
Safety Assessment

Major differences for the different possibilities of computing the failure probability occur
for the configurations 3332111 and 232, although such discrepancy has more to do with
the consideration or not of drift limitation criteria, rather than the methodology itself. It has
even been verified that the commanding pier changes for configuration 232, when drift
limitation is carried out, in the global simulation mode (LHS) approach. Within these
configurations, there are type 3 piers, 21 meters tall, in a vulnerable position, which,
together with being highly affected by top displacement limitation, leads to differences in
safety that are more pronounced. There is a general tendency for the global Latin
Hypercube simulation procedure to yield higher failure probabilities, whether the drift is
limited or not, which has certainly to do with the inclusion of more variables, characterized
with high uncertainty, in the global simulation process.

0
10

-1
10
Failure probability

-2
10

-3
10
NSA-NDA w/o drift control
-4 NSA-NDA w/ drift control
10 LHS w/o drift control
LHS w/ drift control
-5
10
3332111 232 2222222 222 2331312 213 123
Bridge configuration

Figure 6.48 – Failure probability, according to bridge configuration, for all the NDA-based
versions.

Nevertheless, it is quite appealing to realize that, except for the two aforementioned
configurations, and despite eventual inversions in the relative position, the four collapse
probabilities, corresponding to the four variants, are of the same order of magnitude.
Regarding the configuration type, irregular configurations are, in general, on the right,
which means, higher vulnerability, whereas there seems to be no connection of the failure
probability with the length of the bridge.

6.68
 Safety Assessment

Filtering the results in Figure 6.48, comparison of the two methodologies is presented in
Figure 6.49, without drift control, and in Figure 6.50, with drift control of the piers
capacity.

0
10

-1
10
Failure probability

-2
10

-3
10

-4
10 NSA-NDA w/o drift control
LHS w/o drift control
-5
10
3332111 232 2222222 222 2331312 213 123
Bridge configuration

Figure 6.49 – Failure probability, without drift limitation, according to bridge configuration.

As previously forecasted, the differences are now much smoother. The failure probability
trends along with the configuration type are notoriously in agreement. Moreover, the
tendency for the LHS procedure to be on the safer side is confirmed, with the exception of
configuration 123.

When drift control is included in the capacity computation, the methodologies get even
closer, which seems understandable, given that such limiting criteria will definitely, at least
to some extent, envelope the capacity. Higher failure probability is still generally obtained
with LHS procedure, despite the reduction already observed and explained in configuration
222. The increase in the collapse probability of configurations 3332111 and 232, for both
methodologies, is clearly observable, as a result of the vulnerable type 3 pier, which is
highly drift-limited.

6.69
Safety Assessment

0
10

-1
10
Failure probability

-2
10

-3
10

-4
10 NSA-NDA w/ drift control
LHS w/ drift control
-5
10
3332111 232 2222222 222 2331312 213 123
Bridge configuration

Figure 6.50 – Failure probability, with drift limitation, according to bridge configuration.

In Figure 6.51 pushover-based methodologies are plotted together with the already
presented nonlinear dynamic analysis based ones. For the sake of simplicity, only
conventional pushover, with uniform load pattern and central reference node, and adaptive
pushover versions are presented.

0
10

-1
10
Failure probability

-2
10

-3
10
NSA w/ drift control
-4 HPL w/ drift control
10
conventional pushover
adaptive pushover
-5
10
3332111 232 2222222 222 2331312 213 123
Bridge configuration

Figure 6.51 – Failure probability, with drift control, according to bridge configuration, for NDA
and pushover-based versions.

The general picture is that pushover-based estimation of structural effects leads, in most of
the cases, to higher failure probability, especially for the adaptive type, which is frequently
associated to a failure probability one order of magnitude above. Such finding had already
been visible from Figure 6.40. Similarity between the two types of pushover is not
noticeable neither between pushover and NDA variants. These results reinforce the belief
that pushover analysis is more likely to play a supporting part in the probabilistic methods,
when compared to dynamic analysis.

6.70
 Safety Assessment

6.3 Conclusions

The analysis of different ways of looking into the safety assessment issue constituted the
main objective of this chapter. Alternative procedures have been proposed, corresponding
to different possible modes of incorporating the variability of the numerous intervening
variables: material properties, type of record and intensity level. An innovative statistical
sampling method has been used to characterize capacity as well as to compute failure
probability within a global simulation procedure. Moreover, failure probability was
computed recurring to two possible demand prediction techniques: nonlinear dynamic or
static analysis. The parametric study of all the necessary elements for the safety evaluation,
the observation of the performance of the alternative methodologies and the final
comparison of results can be summarized by the conclusions that follow.

The Latin Hypercube Sampling method revealed itself as a relying technique, when
used to obtain capacity or demand distribution samples of lower size than the
tradition Monte Carlos or FORM ones. For the capacity characterization of the
piers cross sections, in terms of available ductility, which is a relatively simple
calculation, 50 to 100 realizations proved to be able to provide stable mean results,
without excessive dispersion. In addition, the method proved to be quite effective
and stable within several independent repetitions. The variables that were
considered for the capacity definition consisted of the material properties, only,
with the ultimate steel strain being the most influent one. The sampling algorithm
was equally efficient when the variables were taken with higher, but more realistic
as well, coefficients of variation, which reached 30%, for some cases.

The traditional methodology to compute the collapse probability required the

definition of vulnerability functions, corresponding to 2nd degree polynomials
adjusted to the mean ductility demand for increasing seismic intensity, obtained
with nonlinear dynamic analysis. The curve fitting was extremely easy, together
with high correlation coefficients, denoting the quality of the least squares method
adjustment, for every configuration, every pier and every nonlinearity level.

The use of nonlinear static analysis for the seismic demand prediction has been
confirmed as a valid approach, even if generally overpredicting with respect to
nonlinear dynamic analysis. Curve fitting to demand obtained through such means

6.71
Safety Assessment

was in the same way well succeeded. despite the fact that an adaptive pushover
analysis is a more suitable choice for the prediction of the behaviour of irregular
configurations, a conventional pushover, with a uniform load pattern, has lead to
rather acceptable estimates for regular configurations. This is an important
conclusion, if one keeps in mind the need for convincing methodologies and
definitely less complex and time consuming than nonlinear dynamic analysis to be
incorporated in actual seismic design and/or assessment of structures by
practitioners.

The traditional approach can be quite sensitive in terms of statistical

characterization and a slight change in the dispersion might affect much more the
failure probability than the median available ductility, as has been observed. The
fact that probabilistic methods work over the tails of the distributions, convoluting
number of very low orders of magnitude explains such behaviour and indicates that
distribution fitting must be handled with care.

With the seismic action being introduced as a variable for the global uncertainty
procedure, the needed sampling size increased to 200 realizations, in order to reach
consistent failure probabilities, for all sorts of bridge configurations as well as for
different repetitions. Such increase was expected and is easily understandable,
given that the procedure involves nonlinear dynamic analysis of a sometimes
considerably irregular geometrical configuration. Furthermore, taking the seismic
action as a variable induces a great additional amount of dispersion. The
recommendations on the use of the LHS scheme in bridges safety assessment point
to perform at least 200 realizations, which is, in any case, quite feasible
computationally.

The distribution fitting has been carried out with success and has, in agreement,
been markedly superior, in quality, from 100 or 200 realizations on, with no
advantage in carrying out 500 realizations. The goodness of fit has been evaluated
with Kolmogorov-Smirnov test. However, some distribution samples of size 50
have passed the same test, so, to the author’s belief, further probing, recurring to
other available tests, needs to be carried out. This aspect is underlined by the fact
that the global LHS methodology is quite easily affected as well by slight changes

6.72
 Safety Assessment

in the probabilistic parameters characterizing the fitted distributions, again, for the
same reason of working on the tails.

In both of the methodologies, whether the uncertainty is considered locally or

globally, the seismic action variables, earthquake record and intensity level,
influenced the results outstandingly, especially the accelerogram that was being
considered. The debate on realizing if the seismic action variability does not
excessively outshine the variability of the rest of the variables is therefore settled.
This aspect is certainly increased by the fact that real ground motion records have
been used. It has been, however, interesting to notice that, despite the extreme
variability, the global LHS methodology behaved quite soundly, enabling reliable
results to be reached. The simulation technique seems hence to be able to
incorporate highly different sorts of variables, with different distribution types and,
mainly, distinct levels of uncertainty. Additionally, this conclusion strengthens the
awareness of the need for a proper handling of the seismic action in terms of
earthquake records, in gender and number.

For what concerns the sort of bridge configuration, the trend that has been found is
for irregular configurations to be less safe, whereas, regarding the deck length, no
relevant correlation has been detected. Moreover, regular configurations seem to
put the different procedures closer to each other, whilst, for irregular ones, the
dispersion is slightly superior.

The central piers, or at least the ones that are not close to the abutments, are
typically submitted to higher demand, hence, the most vulnerable ones. Such
conclusion can be easily taken from the 222 or 2222222 configurations, with all
piers of the same height, where the critical piers are the central ones,
notwithstanding the flexibility that has been considered within the modelling of the
abutments. With respect to the remaining configurations, the definition of the
decisive pier will depend on their type: 1, 2 or 3, and how far the drift limitation
criterion affects them. Type 1 piers, 7 meters high, are typically very restrictive,
even when not centrally located, something that is easily explainable by their
naturally higher stiffness, standing for superior action effects, bearing lower
ductility at the base cross section of their piers. As a consequence, the drift control
will not be relevant, once the displacements at the top are already inherently

6.73
Safety Assessment

limited. This sort of pier has been found to condition configurations 123, 213 and
2331312 configurations. On the other hand, type 3 piers have been extremely
influenced by the top transverse deformation restriction, with their capacity being
strongly reduced, becoming the critical piers for configurations 232 and 3332111.
As for the 14 meters high piers, type 2, due to their medium height, no significant
changes in the ductility permitted by the material and cross section properties are
introduced by the drift controlling.

The two key alternative procedures to compute the collapse probability ended up
being very concordant for the majority of the configurations. The methodology that
considers the uncertainty of all the variables, including the seismic action, in a
global fashion, using the LHS algorithm, yielded, however, more severe results.
Such expected higher failure probabilities have definitely to do with the inclusion
of a larger number of variables and corresponding uncertainty. The
acknowledgment of this feature leads to the recognition as well of the major
importance that the seismic action variability has in worsening the vulnerability of
the bridges. The exception goes for the traditional approach, when implemented
with the use of pushover analysis to estimate the demand, which yielded
considerably less safe scenarios, constituting an upper envelope, on the
conservative side.

The distance between the two methodologies gets even lower if the drift
deformation control of the piers is carried out, denoting that such limitation
contributes to standardization of results. Such effect is particularly visible in the
reduction of the dispersion that had been initially found in Figure 6.48 for
configurations 232 and 3332111.

6.74
7. Conclusions

7.1 Concluding remarks

The present work targeted the probabilistic seismic safety assessment of single reinforced
concrete existing bridges. To accomplish so, different methodologies, corresponding to the
use of different methods for structural response prediction, combined with different ways
of incorporating the uncertainty associated to the different variables of the safety
verification problem, have been proposed, calibrated and compared. The computation of
the structural failure probability made use of advanced statistical treatment, featuring
distribution fitting of the relevant variables and numerical random sampling of variables
using the Latin Hypercube sampling technique.

In addition to the distinct approaching methodologies for reaching a collapse probability

that have been considered, independent calibration studies have been conducted, covering
the different aspects around the assessment process, namely, the seismic action input
characterization, the nonlinear structural modelling and the structural response estimation.
Such studies were carried out with the purpose of better understanding the influence of
each issue in the seismic behaviour of bridges, providing useful information and
recommendations for analysis accuracy and optimization.

7.1
Conclusions

All the comparison/calibration studies have been carried out on a rather wide set of
structural bridge configurations, featuring different locations: regular, semi-regular and
irregular, in terms of pier heights and relative position, short and long, with different types
of abutments. In addition, the selected seismic input case study was equally wide ranging,
so as to include relevant seismic dispersion in the analyses. Different intensity levels were
considered, looking into the bridges behaviour in elastic field as well as pushing them into
high nonlinear stages. In order to appropriately organize all the information coming from
all the configurations, earthquake records and intensity levels, results were treated, when
possible, throughout the computation of statistical measures, with the intent of identifying
patterns and drawing as much generalist as possible conclusions.

The study of the seismic action characterization constituted the first logical step and went
over different topics, well known in the literature. The following observations were made.

Focusing the use of accelerograms as the most complete seismic input element
type, the first question remained on the choice between using real or artificial
ground motion records. Looking at the first option, a case study of 20 different
earthquake records was considered within a European integrated project
framework. Different ground motion scaling parameters were tested for the
standardization of accelerograms for use in nonlinear dynamic analysis. The
parameters were all based on quantities taken/computed from the accelerograms
themselves or from the corresponding response spectra. Identification of optimum
scaling techniques, through evaluation of resulting dispersion, has shown that
typically used peak ground acceleration or spectral acceleration for the first
vibration period of the structure being analysed performed better. Such findings
have then been employed in the probabilistic framework developed in Chapter 5,
which made use of peak ground acceleration probability density functions, to
characterize the seismic Hazard.

The issue of the selection of real records for analysis was afterwards addressed, a
task of renowned importance, given the heterogeneity which characterizes the
available real accelerograms databases. Typically carried out following criteria
based on seismological parameters, such as magnitude or epicentral distance, the
work herein undertaken tried to adopt a different perspective, looking at the effects

7.2
 Conclusions

of the earthquakes on the structure, in terms of displacement ductility demand. The

same set of records was used to run nonlinear dynamic analysis on a long, irregular
existing bridge, the Lordo viaduct, used within the mentioned research project.
Selection of the records causing very discrepant results, differing substantially from
the general trend, was carried out by means of clustering analysis, which is based
on likeliness measures, used to aggregate similar individuals. The application of the
clustering procedures allowed the obtaining of a much more homogeneous set of
records, with considerable reduction in the associated scatter. Such a technique may
be easily/efficiently used in reducing dispersion, provided that a proper statistical
software package is available, when selecting real accelerograms to carry out safety
assessment studies, which are largely influenced by the scatter around the records.

A brief incursion to alternative intensity measure approaches, more elaborated, not

simply obtained from the accelerogram quantities or corresponding response
spectra, was carried out. Particularly, an intensity measure based on pushover
analysis and displacement based theory simplified parameters has been proposed
and its potential has been demonstrated for a wide set of bridge cases. The
inclusion of such intensity measure in the study carried out in Chapter 6, as an
additional approach to the estimation of structural demand, as mentioned in Section
7.2, can now be explored.

Artificial accelerograms have been considered as well, generated to be compatible

with the spectrum corresponding to the mean of the real records. Final comparison
with real records scaled according to the two most advantageous techniques has
been carried out. Artificial accelerograms proved to be a valid alternative for the
seismic analysis of the studied bridge, featuring considerable less dispersion as well
as being less conservative than the real records, especially when scaled by means of
spectral acceleration parameter. Their consideration in bridges safety assessment
procedures is, hence, expected to lead to similar results with much less
computational demand.

Modelling of the structural system for seismic analysis has been afterwards looked at, in
terms of the issues of main interest and pertinence, focusing mainly on material
nonlinearity, given that geometrical nonlinearity is usually seen as less relevant. The study

7.3
Conclusions

on material nonlinearity consideration, addressed in Chapter 4, lead to the following

findings.

The main distinction in terms of approaching ways has been made between the use
of distributed or concentrated plasticity models or, in other words, fibre models or
plastic hinge models. Both the modelling alternatives have been tested with the use
of beam-column bar elements 3D models, using two well-known structural analysis
programs: SeismoStruct for spread plasticity and SAP2000 for lumped plasticity.
To the employment of fibre models, the input of the concrete and steel constitutive
laws was enough, whereas, for the concentrated plasticity ones, a characterization
of the plastic hinge constitutive law, based on the same material models, was
required. Consequently, a summary of several available material models has been
presented and the description of the used ones has been carried out, with the
selection being based mainly on past experience on such matter.

Within the use of the plastic hinge based modelling, different important issues have
been addressed. Calibration of elements with plastic hinges development at the
extremities of the piers was done in terms of plastic hinge length, location as
moment-curve defining curve. Both length and location of the plastic hinge proved
to be not greatly relevant; three different approaches for the parameters were tested
in hollow and full reinforced concrete sections (from the piers of the main case
study bridges) and results did not present significant variation. The moment-
curvature constitutive law of the plastic hinges was approximated by a trilinear
curve, which enables optimization, in terms of time and computational effort, of the
concentrated plasticity approach. Furthermore, the fibre model corresponding
moment-curvature was brought in for comparison, with both the curves matching
quite well.

The main comparison between the two modelling schemes was extended to the
comparison of the structural response prediction of the different bridge
configuration of the selected case study. Nonlinear static procedures (NSP) were
the chosen tool for the bridges analysis, using, for this case, the N2 Method,
recommended by the EC8 and confirmed as suitable to the analysis of bridges in
Chapter 5. Moreover, the NSP was applied using capacity curves obtained with

7.4
 Conclusions

different load patterns and reference nodes. The comparison of the capacity curves,
the immediate output of the pushover analyses, obtained for the different
configurations, denoted a fairly good agreement between the two modelling
alternatives, regardless of the employed load pattern or reference node. Slight
differences that were punctually found correspond to slightly higher base shear
prediction for the case of plastic hinges models. Furthermore, comparison on the
structural response of the bridges at their performance point was evaluated through
the computation of ratios of the different models’ response for different response
parameters, across increasing intensity levels. Again, fairly good matching of
results has been encountered, despite minor differences that have, nevertheless,
been found.

The global conclusion pointed out that both modelling solutions provided agreeing
outputs, whether fibre or plastic hinge based. Such agreement was sustained within
the computation of capacity curves as well as structural response estimated by
means of nonlinear static analysis. Such outcome surely brings some reassurance in
the recognition of the validity of more simplified plastic hinge models, as long as
duly calibrated in its parameters, for use in seismic analysis of bridges, when
compared to the certainly more precise fibre models. This outcome also sustained
the use of plastic hinge models in the extensive probabilistic framework carried out
in Chapter 6, accounting for considerable time and computational savings.

The prediction of the structural response of bridges, the scope of Chapter 5, was evaluated
recurring to different nonlinear analysis schemes, namely, nonlinear static analysis and
nonlinear dynamic analysis. For what concerns the use of nonlinear static procedures, the
existing high variability on the possible alternatives lead to an extensive parametric study,
selecting four commonly employed methods for comparison: Capacity Spectrum Method,
N2 Method, Modal Pushover Analysis and Adaptive Capacity Spectrum Method.
Preliminarily, each of the procedures was calibrated with a view to find its optimum
configuration, which was then selected for the final comparison. For the sake of generality
and consistence the comparative evaluation of the different methods was carried out for the
described case study and seismic input and throughout the same response parameters. The
following conclusions are worth underlying:

7.5
Conclusions

Regarding the individual calibration of the methods, different variants for each of
them were tested, mainly in terms of loading pattern for the pushover analysis (N2),
pushover curve reference node (CSM, N2 and MPA), spectral reduction, based on
equivalent viscous damping or ductility (CSM and ACSM). Significant differences
have been found in the NSPs performance, when applied in such different variants.
More than the reference node choice or the loading shape, the spectral reduction
factors issue, together with the variability in the estimation of equivalent viscous
damping, particularly affected the methods, which denoted the need for the
preliminary calibration.

Generally, the nonlinear static procedures provided evidence on their ability to

accurately predict the structural response of bridges, in terms of displacements,
except for the Capacity Spectrum Method, and pier shear forces, for all sorts of
bridge configurations (regular, irregular, short, long, etc), when compared to
nonlinear dynamic analysis results, which is commonly seen as the most reliable
tool for structural performance estimation. Regarding the remaining parameters, an
underestimating pattern has been found for all the methods, in deck moments and
abutment shear forces. Globally, the Adaptive Capacity Spectrum Method behaved
more consistently, followed closely by N2 and Modal Pushover Analysis. For the
case of heavily underestimated deck moments and shear at the abutments, the
Capacity Spectrum Method stood as the poorest one, despite the unsatisfactory
behaviour of all the procedures. With filtering the results according to different
structural categories (in terms of regularity, deck length and type of abutments) less
scattered estimates where obtained using MPA and ACSM. It seems, therefore, that
such methods, including higher modes effects, are definitely more adequate to
accurately estimate the nonlinear performance of structures, as an alternative to
complex nonlinear dynamic analysis. Such conclusion seems understandable,
especially for bridge structures, to which superior vibration modes with important
modal participation factors are typically associated. Moreover, such conclusions
have validated the use of nonlinear static analysis techniques in other topics, such
as the modelling calibration issues in Chapter 4, or probabilistic safety assessment
in Chapter 6.

7.6
 Conclusions

If one method should be chosen over the others, such choice should mostly be
based on two aspects: the demonstrated condition that the method includes higher
modes contribution and the ability to keep a steady accurate level of response
prediction for the set of different deformation/force tested parameters. Under such
circumstances, the Adaptive Capacity Spectrum Method could be the choice,
followed by the N2 Method and Modal Pushover Analysis.

Finally, the evaluation of the actual methodologies for the seismic safety assessment was
carried out in Chapter 6, making use of the observations made in Chapters 3, 4 and 5,
regarding Seismic Input, Nonlinear Modelling and Seismic Demand, respectively.

Two distinct methodologies have been proposed for the safety assessment leading to the
computation of the failure probability of the structure, which differ essentially on the way
of incorporating the uncertainty associated to the several variables of the whole process,
represented by their statistical distributions, which have been whether assumed or
determined. The uncertainty has been considered locally or globally, with a higher number
of variables corresponding to the latter. Moreover, different variants to those
methodologies were tested, including different ways of estimating the structural response,
enhanced by the conclusions in Chapter 5. The following summarizing observations can be
made.

The use of the Latin Hypercube Sampling method in characterizing the structural
capacity has been found efficient and simple, requiring a significantly lower
number of random realizations than other comparables methods. The number of
realizations needed for the sampling results to stabilize was expectedly lower for
the case of capacity characterization rather then for global probabilistic
computation. The estimation of the distribution of the available ductility in
curvatures of the piers cross sections required samples sized of a maximum of 50
realizations, with input variables being restricted to the material properties. No
uncertainty was considered in the geometrical properties.

All the nonlinear dynamic analyses, needed for both the methodologies, were
carried out using plastic hinges models, easier to incorporate within an automatic

7.7
Conclusions

global computation/simulation procedure. Such feature was supported by the

conclusions drawn in Chapter 4.

The traditional methodology, corresponding to the local uncertainty consideration,

has made use of vulnerability functions obtained either through nonlinear dynamic
analysis or nonlinear static analysis, as enabled by the prescriptions in Chapter 5,
on Seismic Demand. The vulnerability functions were obtained by 2nd degree
polynomial curve fitting to the individual ductility demands corresponding to
different intensity levels. Such process was successful and straightforward with
high quality of the adjustment being noted. The use of nonlinear static vulnerability
functions has proven to be valid, even if sometimes too much conservative.

If the seismic action, input record and intensity level, is included as a variable, then
the aforementioned global simulation procedure is being followed, which features
the computation of the collapse probability by adjusting a distribution to the
variable given by the difference between the capacity and the demand. For that
case, the sampling size, within the Latin Hypercube sampling scheme, increased to
200, which continues to be, in any case, computationally low demanding. The
goodness-of-fit of the distributions adjusted to the safety interval variable (distance
going from ductility capacity to ductility demand) was verified by means of the
Kolmogorov-Smirnov test, which provided positive results for almost all the cases.

Irregular bridges have been found to be less safe to the considered seismic action
and resulted in higher agreement between the different methodologies. On the other
hand, no noticeable trend has been found with respect to the deck length. The
central or the shortest piers were systematically the critical ones, sometimes with
huge differences in its failure probabilities, when compared to the more slender
ones. An additional criterion to the probability computation, in terms of maximum
allowable drift for the piers, was added, resulting in quite interesting results,
changing sometimes the critical pier.

The two key alternative procedures to compute the collapse probability have
proven to be quite in agreement for the majority of the configurations. The added
consideration of the uncertainty of the seismic action yielded, however, more

7.8
 Conclusions

severe results, which has to do with the inclusion of a larger number of variables
and corresponding uncertainty. The two methodologies get even closer if the drift
deformation control of the piers is carried out, denoting that such limitation actually
contributes to the results.

Seismic assessment of bridges, within a probabilistic context, is a highly dense and

controversial topic, as has been demonstrated, not only due to the statistical approach to the
safety problem itself but also to all the uncertainty that surrounds its different components.
The work herein carried out over the Seismic Input, Nonlinear Modelling and Seismic
Demand issues, following a natural path that would lead to the optimization of the safety
assessment process, has highlighted several obstacles (and variability) that the engineering
community is still facing nowadays. Hopefully, a contribution to that same community, on
those specific topics, has been provided. Moreover, notwithstanding the difficulties in
standardizing procedures and approaching tactics, it has been observed that conceptually
different probabilistic seismic Safety Assessment methodologies are not as random and
uncertain as many times feared. This surely reaffirms the soundness and the need for
probabilistic seismic assessment (and design) methodologies to become a reality in bridge
engineering practice.

7.2 Future developments

The presented work aimed to contribute with a general overview of all the steps that more
or less generally need to be taken when carrying out the probabilistic seismic safety
assessment of existing reinforced concrete bridges.

Regarding the seismic action characterization, the clustering techniques may be further
probed by extending the case study to a larger number of records and, mostly, to different
structural configurations. The use of artificial accelerograms has not been thoroughly
developed. The use of different types of synthetic records, such as wavelet enriched ones,
can be investigated. Moreover, the effect of artificial records should be tested inside the
probabilistic assessment study as a less demanding alternative to the real records in
dynamic analysis. Several improvements, regarding the proposed preliminary
displacement-based intensity measure, can be carried out as well, characterizing more
thoroughly the advantageous effects of such standardizing technique in nonlinear seismic

7.9
Conclusions

analysis. Furthermore, the simple underlying premises certainly enable its consideration
within a probabilistic assessment study, of a single structure or a whole region network.

Concerning the nonlinear modelling field, the damping consideration can be an aspect that
greatly influences the structural behaviour. Within this work, no parametric calibration
study on the elastic damping was presented, given that typically such portion is considered
to be very low when dealing with bridge structures, where there are few physical
phenomena that may call for such modelling. It is, however, believed that with the
consideration of higher elastic damping values, numerical instability may arise, requiring
additional studies to better understand such phenomenon in reinforced concrete bridges.

Within the structural response topic, different alternative simplified procedures, including
additional variables in the parametric study, can be tested, and further improving of the
prediction of shear forces should be looked for. Additionally, Nonlinear Static Procedures
may possibly start to be employed in the near future by practitioners as the first genuine
nonlinear analysis methods introduced in common design office practice. Consequently,
preparation of such methods to be easily applied, eventually involving some generalization
degree, can be initiated.

Finally, and focusing on the main goal of this work, the probabilistic safety assessment of
bridges, there is still a long way to go before such probabilistic methods can become
widespread. On that matter, the contribution herein given is limited and can be improved in
some aspects. The probability computation of the proposed methodologies is achieved at
the tails of the distributions, which are known to be problematic. Statistical refinement of
the methods can be a possible improving path, in order to bring more confidence to the
vulnerability estimates. In addition, more goodness-of-fit tests can be carried out,
discarding improper distributions, or even different distributions, similar to the extreme
value type ones, herein considered, can be tested, with a view to warrant the attainment of
more solid failure probability predictions.

7.10
References

A
Aktan, A. E., Pecknold, D. A. and Sozen, M. A. (1974) R/C Column Earthquake Response in Two
Dimensions. Journal of the Structural Division, 100(10), pp. 1999-2015

Aktan, A. E. and Ersoy, U. (1979) Analytical Study of R/C Material Hysteresis. Proceedings of the
AICAP-CEB Symposium on Structural Concrete Under Seismic Actions, CEB Bulletin
d'Information no. 132, pp. 97-104

Alavi, B. and Krawinkler, H. (2004) Behavior of moment-resisting frame structures subjected to

near-fault ground motions. Earthquake Engineering & Structural Dynamics, 33(6), pp. 687-706

Alfawakhiri, F. and Bruneau, M. (2000) Flexibility of superstructures and supports in the seismic
analysis of simple bridges. Earthquake Engineering & Structural Dynamics, 29(5), pp. 711-729

Alvarez, J. C. (2004) Displacement-based design of continuous concrete bridges under transverse

seismic excitation. M.Sc. Thesis, ROSE School, European School for Advanced Studies in
Reduction of Seismic Risk, Pavia, Italy

Ambraseys, N., Smit, P., Sigbjornsson, R., Suhadolc, P. and Margaris, B. (2002) Internet-Site for
European Strong-Motion Data. European Commission, Research-Directorate General,
Environment and Climate Programme

Antoniou, S., Rovithakis, A. and Pinho, R. (2002) Development and verification of a fully adaptive
pushover procedure. Proceedings of the 12th European Conference on Earthquake Engineering,
September 9-13, London, UK

Antoniou, S. and Pinho, R. (2004) Development and verification of a displacement-based adaptive

pushover procedure. Journal of Earthquake Engineering, 8(5), pp. 643-661

Arêde, A. and Pinto, A. V. (1996) Reinforced concrete global section modelling: definition of
skeleton curves. Special Publication No.I.96.36, Institute for Systems, Informatics and Safety, Joint
Research Center, Ispra, Italy

ASCE (2005) Minimum Design Loads for Buildings and Other Structures (7-05). American
Society of Civil Engineers, Reston, VA

ATC (1996) Seismic Evaluation and Retrofit of Concrete Buildings, Volumes 1 and 2. Report No.
ATC-40, Applied Technology Council, Redwood City, CA

ATC (1997) NEHRP Guidelines for the Seismic Rehabilitation of Buildings. Report No. FEMA-
273, Federal Emergency Management Agency, Washington, DC

ATC (2005) Impovement of Nonlinear Static Seismic Analysis Procedures. Report No. FEMA-440,
Federal Emergency Management Agency, Washington, DC

R.1
References

Aydan, Ö, Kumsar, H., Toprak, S. and Barla, G. (2009) Characteristics of 2009 L’Aquila
earthquake with an emphasis on earthquake prediction and geotechnical damage. Journal of The
School of Marine Science and Technology, 7(3), pp. 23-51

Ayyub, B. M. and Lai, K. L. (1989) Structural reliability assessment using latin hypercube
sampling. Proceedings of the ICOSSAR'89, August 8-11, San Francisco, USA

B
Baker, J. W. and Cornell, C. A. (2006) Spectral shape, epsilon and record selection. Earthquake
Engineering & Structural Dynamics, 35(9), pp. 1077-1095

Bal, I. E., Crowley, H. and Pinho, R. (2008) Displacement-Based Earthquake Loss Assessment for
an Earthquake Scenario in Istanbul. Journal of Earthquake Engineering, 12(S2), pp. 12-22

Banon, H., Irvine, H. M. and Biggs, J. M. (1981) Seismic damage in reinforced concrete frames.
Journal of the Structural Division, 107(9), pp. 1713-1729

Barbat, A. H. and Canet, J. M. (1994) Estructuras sometidas a acciones sísmicas: Cálculo por
ordenador. Centro Internacional de Métodos Numéricos en Ingeniería, Barcelona (in Spanish)

Bartlett, F. M. and MacGregor, J. G. (1996) Statistical analysis of the compressive strength of

concrete in structures. ACI Materials Journal, 93(2), pp. 158-168

Basoz, N. and Kiremidjian, A. (1998) Evaluation of Bridge Damage Data from the Loma Prieta
and Northridge, California Earthquakes. Technical Report MCEER-98-0004, Stanford University,
Stanford, California

Bazoant, Z. P. and Bhat, P. D. (1977) Prediction of Hysteresis of Reinforced Concrete Members.

Journal of the Structural Division, 103(1), pp. 153-167

Bazzurro, P. (1998) Probabilistic Seismic Demand Analysis. Ph.D. Thesis, Stanford University -
Department of Civil and Environmental Engineering, Stanford, CA, USA

Bazzurro, P. and Luco, N. (2006) Do Scaled and Spectrum-matched Near-Source Records Produce
Biased Nonlinear Structural Responses? Proceedings of 8th U.S. National Conference on
Earthquake Engineering, April 18-22, San Francisco, CA, USA

Bertero, V. V. and Popov, E. P. (1976) Experimental and Analytical Studies on Hysteretic

Behaviour of Reinforced Concrete Rectangular and T-Beams. Technical Report 02/76, Earthquake
Engineering Research Center, University of California, Berkeley, CA

Bertero, V. V., Aktan, A. E., Charney, F. and Sause, R. (1984) Earthquake Simulator Tests and
Associated Experimental, Analytical, and Correlation Studies of One-Fifth Scale Model. Special
Publication, Earthquake Effects on Reinforced Concrete Structures, American Concrete Institute,
SP84(13), pp. 375-424

Bertero, V. V., Anderson, J. C., Krawinkler, H. and Miranda, E. (1991) Design Guidelines for
Ductility and Drift Limits. Report No. UCB/EERC-91/15, Earthquake Engineering Research
Center, Berkeley, CA

Bertero, V. V. (1995) Tri-services manual methods. Vision 2000 Performance Based Seismic
Engineering of Buildings, Vol. 1, Part 2, Appendix J, J1-J8, Structural Engineer Association of
California, Sacramento, CA

R.2
 References

Bommer, J. J. and Martínez-Pereira, A. (1999) The effective duration of earthquake strong motion.
Journal of Earthquake Engineering, 3(2), pp. 127-172

Bommer, J. J. and Acevedo, A. B. (2004) The Use of Real Earthquake Accelerograms as Input to
Dynamic Analysis. Journal of Earthquake Engineering, 8(S1), pp. 43-91

Borges, J. F. and Castanheta, M. (1985) Structural Safety, Curso 101. LNEC, Lisboa (in
Portuguese)

Bozorgzadeh, A., Megally, S., Restrepo, J. I. and Ashford, S. A. (2006) Capacity Evaluation of
Exterior Sacrificial Shear Keys of Bridge Abutments. Journal of Bridge Engineering, 11(5), pp.
555-565

Bracci, J. M., Kunnath, S. K. and Reinhorn, A. M. (1997) Seismic Performance and Retrofit
Evaluation of Reinforced Concrete Structures. Journal of Structural Engineering, 123(1), pp. 3-10

Bradley, B. A. (2011) Design Seismic Demands from Seismic Response Analyses: A Probability-
Based Approach. Earthquake Spectra, 27(1), pp. 213-224

Brancaleoni, F., Ciampi, V. and Di Antonio, R. (1983) Rate-Type Models for Non Linear
Hysteretic Structural Behavior. Proceedings of the EUROMECH Colloquium, October 10-14,
Palermo, Italy

C
Calabrese, A., Almeida, J. P. and Pinho, R. (2010) Numerical Issues in Distributed Inelasticity
Modeling of RC Frame Elements for Seismic Analysis. Journal of Earthquake Engineering,
14(S1), pp. 38-68

Calvi, G. M. (1994) PREC8 - Bridge Models For PSD Testing. Design Documents, Department of
Structural Mechanics, University of Pavia, Pavia, Italy

Calvi, G. M. and Pinto, P. E. (1995) Irregular bridges designed according to Eurocode 8: numerical
and experimental verifications. Proceedings of the 1st Japan-Italy Workshop on Seismic Design of
Bridges, Public Works Research Institute, Tsukuba, Japan

Calvi, G. M. (2004) Recent experience and innovative approaches in design and assessment of
bridges. Proceedings of the 13th World Conference on Earthquake Engineering, August 1-6,
Vancouver, Canada

Carvalho, A. (2009) Avaliação da Segurança Sísmica de Pontes. M.Sc. Thesis, University of Porto,
Porto, Portugal (in Portuguese)

Casarotti, C., Pinho, R. and Calvi, G. M. (2005) Adaptive pushover-based methods for seismic
assessment and design of bridge structures. ROSE Research Report No. 2005/06, IUSS Press,
Pavia, Italy

Casarotti, C. and Pinho, R. (2006) Seismic response of continuous span bridges through fiber-based
finite element analysis. Earthquake Engineering and Engineering Vibration, 5(1), pp. 119-131

Casarotti, C. and Pinho, R. (2007) An adaptive capacity spectrum method for assessment of bridges
subjected to earthquake action. Bulletin of Earthquake Engineering, 5(3), pp. 377-390

R.3
References

Casarotti, C., Monteiro, R. and Pinho, R. (2009) Verification of spectral reduction factors for
seismic assessment of bridges. Bulletin of the New Zealand Society for Earthquake Engineering,
42(2), pp. 111-121

CBSC (2007) California Building Code. California Building Standards Commission, Sacramento,
CA

CEB (1983) Response of R.C. Critical Regions under Large Amplitude Reversed Actions. Bulletin
d'Information No. 161, pp. 255-284, Comité Euro-International du Béton, Lausanne

CEN (2005a) Eurocode 8: Design of Structures for Earthquake Resistance - Part 1: General rules,
seismic actions and rules for buildings. EN 1998-2, Comité Européen de Normalisation, Brussels,
Belgium

CEN (2005b) Eurocode 8: Design of Structures for Earthquake Resistance - Part 2: Bridges. EN
1998-2, Comité Européen de Normalisation, Brussels, Belgium

Chang, G. and Mander, J. B. (1994) Seismic Energy Based Fatigue Damage Analysis of Bridge
Columns: Part II - Evaluation of Seismic Demand. Technical Report NCEER-94-0013,
Multidisciplinary Center for Earthquake Engineering Research (Former NCEER), Buffalo, NY

Charney, F. A. and Bertero, V. V. (1982) An evaluation of the design and analytical seismic
response of a seven-story reinforced concrete frame-wall structure. EERC Report 82/08,
Earthquake Engineering Research Center, University of California, Berkeley, CA

Chopra, A. K. (1995) Dynamics of Structures: Theory and Applications to Earthquake

Engineering. Prentice-Hall, Englewood Cliffs, NJ

Chopra, A. K. and Goel, R. K. (1999) Capacity-Demand-Diagram Methods Based on Inelastic

Design Spectrum. Earthquake Spectra, 15(4), pp. 637-656

Chopra, A. K. and Goel, R. K. (2001) A modal pushover analysis procedure to estimating seismic
demands for buildings: Theory and preliminary evaluation. PERR Report 2001/03, Pacific
Earthquake Engineering Research Center, University of California, Berkeley, CA

Chopra, A. K. and Goel, R. K. (2002) A modal pushover analysis procedure for estimating seismic
demands for buildings. Earthquake Engineering & Structural Dynamics, 31(3), pp. 561-582

Chopra, A. K. and Chintanapakdee, C. (2004) Evaluation of Modal and FEMA Pushover Analyses:
Vertically “Regular” and Irregular Generic Frames. Earthquake Spectra, 20(1), pp. 255-271

Chopra, A. K. and Goel, R. K. (2004) A modal pushover analysis procedure to estimate seismic
demands for unsymmetric-plan buildings. Earthquake Engineering & Structural Dynamics, 33(8),
pp. 903-927

Chopra, A. K. (2005) Estimating seismic demands for performance-based engineering of buildings.

Congreso Chileno de Sismología e Ingeniería Antisísmica, IX Jornadas, November 16-19,
Concepción, Chile

Clough, R. W. and Johnston, S. B. (1966) Effect of stiffness degradation on earthquake ductility

requirements. Proceedings of the Japan Earthquake Engineering Symposium, Tokyo, Japan

Clough, R. W. and Benuska, L. (1967) Nonlinear earthquake behavior of tall buildings. Journal of
Mechanical Engineering, 93(3), pp. 129-146

R.4
 References

Clough, R. W. and Penzien, J. (1993) Dynamics of Structures. McGrawHill, New York

Cofie, N. G. (1984) Cyclic Stress-Strain and Moment-Curvature Relationships for Steel Beams and
Columns. Report No. 1983-06, Stanford University, Stanford

Computers&Structures (2006) SAP2000 - Integrated Finite Element Analysis and Design of

Structures. Computers and Structures, Inc., Berkeley, California

Cooper, J. D., Friedland, I. M., Buckle, I. G., Nimis, R. B. and Bobb, N. M. M. (1994) The
Northridge earthquake: progress made, lessons learned in seismic-resistant bridge design. Public
Roads, 58(1), pp. 26-36

Correia, A. A. and Virtuoso, F. B. E. (2006) Nonlinear Analysis of Space Frames. Proceedings of

the III European Conference on Computational Mechanics, June 5-8, Lisbon, Portugal

Costa, A. C. (1993) A Acção dos Sismos e o Comportamento das Estruturas. Ph.D. Thesis,
University of Porto, Porto, Portugal (in Portuguese)

Costa, A. G. and Costa, A. C. (1987) Modelo Histerético das Forças-Deslocamentos Adequado à

Análise Sísmica de Estruturas. Relatório Técnico, Núcleo de Dinâmica Aplicada, National
Laboratory of Civil Engineering, Lisbon (in Portuguese)

Costa, A. G. (1989) Análise sísmica de estruturas irregulares. Ph.D. Thesis, University of Porto,
Porto, Portugal (in Portuguese)

Crisfield, M. A. (1990) A consistent co-rotational formulation for non-linear, three-dimensional,

beam-elements. Computer Methods in Applied Mechanics and Engineering, 81(2), pp. 131-150

D
Delgado, P. (2000) Vulnerabilidade Sísmica de Pontes. M.Sc. Thesis, University of Porto, Porto,
Portugal (in Portuguese)

Delgado, P. (2009) Avaliação da Segurança Sísmica de Pontes. Ph.D. Thesis, University of Porto,
Porto, Portugal (in Portuguese)

Delgado, R., Marques, M., Monteiro, R., Delgado, P., Romão, X. and Costa, A. (2006) Setting Up
Real or Artificial Accelerograms for Dynamic Analysis. Proceedings of the 1st European
Conference on Earthquake Engineering and Seismology, September 3-8, Geneva, Switzerland

Duarte, R. T. and Costa, A. C. (1991) Identification of Design Seismic Actions Considering the
Nonlinear Behaviour of Structures. Proceedings of the International Workshop on Seismology and
Earthquake Engineering, April 22-26, Centro Nacional de Prevención de Desastres (CENAPRED),
Mexico City

Dwairi, H. and Kowalsky, M. (2006) Implementation of Inelastic Displacement Patterns in Direct

Displacement-Based Design of Continuous Bridge Structures. Earthquake Spectra, 22(3), pp. 631-
662

Dwairi, H. M., Kowalsky, M. J. and Nau, J. M. (2007) Equivalent Damping in Support of Direct
Displacement-Based Design. Journal of Earthquake Engineering, 11(4), pp. 512-530

R.5
References

E
Eberhard, M. O., Baldridge, S. , Marshall, J., Mooney, W. and Rix, G. J. (2010) The MW 7.0 Haiti
earthquake of January 12, 2010; USGS/EERI Advance Reconnaissance Team Report. U.S.
Geological Survey Open-File Report 2010–1048, USGS, California

Elnashai, A. S. (2001) Advanced inelastic static (pushover) analysis for earthquake applications.
Structural Engineering and Mechanics, 12(1), pp. 51-69

Elnashai, A. S. (2002) Do we really need inelastic dynamic analysis? Journal of Earthquake

Engineering, 6(S1), pp. 123-130

Everitt, B. S., Landau, S. , Leese, M. and Stah, D. (2001) Cluster Analysis. John Wiley & Sons
Inc., New York

F
Fajfar, P. and Fischinger, M. (1988) N2 - A method for non-linear seismic analysis of regular
buildings Proceedings of the 9th World Conference in Earthquake Engineering, August 2-9,
Tokyo-Kyoto, Japan

Fajfar, P. (1999) Capacity spectrum method based on inelastic demand spectra. Earthquake
Engineering & Structural Dynamics, 28(9), pp. 979-993

Fajfar, P. (2000) A Nonlinear Analysis Method for Performance Based Seismic Design.
Earthquake Spectra, 16(3), pp. 573-592

Felippa, C. A. and Park, K. C. (2002) The construction of free-free flexibility matrices for
multilevel structural analysis. Computer Methods in Applied Mechanics and Engineering, 191(19-
20), pp. 2111-2140

Filippou, F. C., Popov, E. P. and Bertero, V. V. (1983a) Modeling of R/ C Joints under Cyclic
Excitations. Journal of Structural Engineering, 109(11), pp. 2666-2684

Filippou, F. C., Popov, E. P. and Bertero, V. V. (1983b) Effects of bond deterioration on hysteretic
behavior of reinforced concrete joints. Report No. UCB/EERC-83/19, Earthquake Engineering
Research Center, University of California, Berkeley, CA

Fischinger, M., Beg, D., Isakovic, T., Tomazevic, M. and Zarnic, R. (2004) Performance based
assessment - from general methodologies to specific implementations. Proceedings of the
International Workshop on Performance-based Seismic Design: Concepts and Implementation,
June 28 - July 1, Bled, Slovenia

Florian, A. and Navratil, J. (1993) Reliability Analysis of the Cable Stayed Bridge in Construction
and Service Stages. Proceedings of the ICOSSAR'93, August 9-13, Innsbruck, Austria

Frangopol, D. M., Ide, Y., Spacone, E. and Iwaki, I. (1996) A new look at reliability of reinforced
concrete columns. Structural Safety, 18(2-3), pp. 123-150

Freeman, S. A., Nicoletti, J. P. and Tyrell, J. V. (1975) Evaluation of Existing Buildings for seismic
risk - A case study of Puget Sound Naval Shipyard, Bremerton, Washington. Proceedings of the 1st
U.S. National Conference on Earthquake Engineering, Berkley, USA

R.6
 References

Freeman, S. A. (1998) Development and use of capacity spectrum method. Proceedings of the 6th
U.S. National Conference on Earthquake Engineering, June 1-5, Seattle, USA

Freudenthal, A. M., Garrelts, J. M. and Shinozuka, M. (1966) The Analysis of Structural Safety.
Journal of the Structural Division, 92(ST1), pp. 267-325

Fujii, M., Kobayashi, K., Miyagawa, T., Inoue, S. and Matsumoto, T. (1988) A study on the
application of a stress-strain relation of confined concrete. Proc., JCA Cement and Concrete,
42(1988), pp. 311–314

G
Gasparini, D. A. and Vanmarcke, E. (1976) SIMQKE: Generation of artificial time histories
compatible with a specified target spectrum. Report No. R76-4, Massachusetts Institute of
Technology, Cambridge, MA

Giberson, M. F. (1967) The response of nonlinear multi-story structures subjected to earthquake

excitation. Ph.D. Thesis, California Institute of Technology, Pasadena, USA

Goel, R. K. and Chopra, A. K. (1997) Evaluation of Bridge Abutment Capacity and Stiffness
during Earthquakes. Earthquake Spectra, 13(1), pp. 1-23

Goel, R. K. and Chopra, A. K. (2004) Evaluation of Modal and FEMA Pushover Analyses: SAC
Buildings. Earthquake Spectra, 20(1), pp. 225-254

Goel, R. K. (2005) Evaluation of Modal and FEMA Pushover Procedures Using Strong-Motion
Records of Buildings. Earthquake Spectra, 21(3), pp. 653-684

Goel, R. K. and Chopra, A. K. (2005a) Extension of Modal Pushover Analysis to Compute

Member Forces. Earthquake Spectra, 21(1), pp. 125-139

Goel, R. K. and Chopra, A. K. (2005b) Role of Higher-“Mode” Pushover Analyses in Seismic

Analysis of Buildings. Earthquake Spectra, 21(4), pp. 1027-1041

Grant, L. H., Mirza, S. A. and Macgregor, J. G. (1978) Monte-Carlo Study of Strength of Concrete
Columns. ACI Journal Proceedings, 75(8), pp. 348-358

Grimaz, S. and Maiolo, A. (2010) The impact of the 6th April 2009 L’Aquila earthquake (Italy) on
the industrial facilities and life lines. Considerations in terms of NaTech risk. Proceedings of the
4th International Conference on Safety & Environment in Process Industry, March 14-17,
Florence, Italy

Guedes, J. P. M. (1997) Seismic behaviour of reinforced concrete bridges - Modelling, numerical

analysis and experimental assessment. Ph.D. Thesis, University of Porto, Porto, Portugal

Gulkan, P. and Sozen, M. A. (1974) Inelastic Responses of Reinforced Concrete Structure to

Earthquake Motions. ACI Journal Proceedings, 71(12), pp. 604-610

Gupta, B. and Kunnath, S. K. (2000) Adaptive Spectra-Based Pushover Procedure for Seismic
Evaluation of Structures. Earthquake Spectra, 16(2), pp. 367-392

R.7
References

H
Hall, J. F. (1995) Northridge earthquake of January 17, 1994: reconnaissance report. Earthquake
Spectra, 11Supplement C, 521p

Hall, J. F. (2006) Problems encountered from the use (or misuse) of Rayleigh damping. Earthquake
Engineering & Structural Dynamics, 35(5), pp. 525-545

Hellesland, J. and Scordelis, A. (1981) Analysis of RC bridge columns under imposed

deformations. Proceedings of the IABSE Colloquium, June 2-5, Delft, The Netherlands

Hoshikuma, J. and Nagaya, K. (1997) Stress-Strain Model for Confined Reinforced Concrete in
Bridge Piers. Journal of Structural Engineering, 123(5), pp. 624-633

I
ICC (2006) International Building Code 2006. International Code Council, Washington, DC

Iman, R. L., Helson, J. C. and Campbell, J. E. (1981) An Approach to Sensitivity Analysis of

Computer Models: Part I - Introduction, Input Variable Selection and Preliminary Variable
Assessment. Journal of Quality Technology, 13(3), pp. 174-183

Iman, R. L. and Conover, W. J. (1982) A distribution-free approach to inducing rank correlation

among input variables. Communications in Statistics - Simulation and Computation, 11(3), pp.
311-334

Isakovic, T. and Fischinger, M. (2006) Higher modes in simplified inelastic seismic analysis of
single column bent viaducts. Earthquake Engineering & Structural Dynamics, 35(1), pp. 95-114

Iwan, W. D. (1980) Estimating inelastic response spectra from elastic spectra. Earthquake
Engineering & Structural Dynamics, 8(4), pp. 375-388

Izzuddin, B. A. (2001) Conceptual issues in geometrically nonlinear analysis of 3D framed

structures. Computer Methods in Applied Mechanics and Engineering, 191(8-10), pp. 1029-1053

J
JCSS (1995) Probabilistic model code - Part 3: Resistance models. Joint Committee for Structural
Safety, Working document JCSS-RACK-08-01-95

K
Kaba, S. A. and Mahin, S. A. (1984) Refined Modeling of Reinforced Concrete Columns for
Seismic Analysis. Report No. UCB/EERC-84/03, Earthquake Engineering Research Center,
University of California, Berkeley, CA

Kalkan, E. and Kunnath, S. K. (2006) Adaptive Modal Combination Procedure for Nonlinear Static
Analysis of Building Structures. Journal of Structural Engineering, 132(11), pp. 1721-1731

Kalkan, E. and Chopra, A. K. (2010) Practical Guidelines to Select and Scale Earthquake Records
for Nonlinear Response History Analysis of Structures. U.S. Geological Survey Open-File Report
2010, USGS, California

R.8
 References

Kam, W. Y. (2011) Preliminary Report from the Christchurch 22 Feb 2011 6.3Mw Earthquake:
Pre-1970s RC and RCM Buildings, and Precast Staircase Damage. University of Canterbury,
Christchurch

Kappos, A. J. (1991) Analytical Prediction of the Collapse Earthquake for R/C Buildings -
Suggested Methodology. Earthquake Engineering & Structural Dynamics, 20(2), pp. 167-176

Kappos, A. J., Chryssanthopoulos, M. K. and Dymiotis, C. (1999) Uncertainty analysis of strength

and ductility of confined reinforced concrete members. Engineering Structures, 21(3), pp. 195-208

Kappos, A. J. and Kyriakakis, P. (2000) A re-evaluation of scaling techniques for natural records.
Soil Dynamics and Earthquake Engineering, 20(1-4), pp. 111-123

Karsan, I. D. and Jirsa, J. D. (1969) Behavior of concrete under compressive loadings. Journal of
the Structural Division, 95(ST12), pp. 2543-2563

Kawashima, K. (2000) Seismic design and retrofit of bridges. Bulletin of the New Zealand National
Society for Earthquake Engineering, 33(3), pp. 265-285

Kawashima, K. (2007) Seismic Design of Urban Infrastructures. Lecture Notes, Kawashima

Laboratory, Department of Civil Engineering, Tokyo Institute of Technology, Tokyo

Kent, D. C. and Park, R. (1971) Flexural Members with Confined Concrete. Journal of the
Structural Division, 97(7), pp. 1969-1990

Kowalsky, M. J. (1994) Displacement-based design - a methodology for seismic design applied to

RC bridge columns. M.Sc. Thesis, University of California, San Diego, La Jolla, CA

Kowalsky, M. J., Priestley, M. J. N. and MacRae, G. A. (1994) Displacement-based design, a

methodology for seismic design applied to SDOF reinforced concrete structures. Report No. SSRP-
94/16, University of California, San Diego, La Jolla

Kowalsky, M. J., Priestley, M. J. N. and MacRae, G. A. (1995) Displacement-based design of RC

bridge columns in seismic regions. Earthquake Engineering & Structural Dynamics, 24(12), pp.
1623-1643

Kowalsky, M. J. (2002) A displacement-based approach for the seismic design of continuous

concrete bridges. Earthquake Engineering & Structural Dynamics, 31(3), pp. 719-747

Krawinkler, H. and Seneviratna, G. (1998) Pros and cons of a pushover analysis of seismic
performance evaluation. Engineering Structures, 20(4-6), pp. 452-464

Kunnath, S. K., Reinhorn, A. M. and Lobo, R. F. (1992) IDARC Version 3.0: A program for the
inelastic damage analysis of reinforced concrete structures. Technical Report NCEER-92-0022,
Multidisciplinary Center for Earthquake Engineering Research (former NCEES), Buffalo, NY

Kurama, Y. C. and Farrow, K. T. (2003) Ground motion scaling methods for different site
conditions and structure characteristics. Earthquake Engineering & Structural Dynamics, 32(15),
pp. 2425-2450

R.9
References

L
Lawson, R. S., Vance, V. and Krawinkler, H. (1994) Nonlinear static pushover analysis - why,
when, and how? Proceedings of the 5th U.S. Conference in Earthquake Engineering, July 10-14,
Chicago, USA

LESSLOSS (2004a) European Integrated Project on Risk Mitigation for Earthquakes and
Landslides. http://www.lessloss.org/

LESSLOSS (2004b) European Integrated Project on Risk Mitigation for Earthquakes and
Landslides, State of the Art - FEUP Contribution for Topic 2.3b. http://www.lessloss.org/

Lin, Y. Y. and Chang, K. C. (2003) Study on Damping Reduction Factor for Buildings under
Earthquake Ground Motions. Journal of Structural Engineering, 129(2), pp. 206-214

Luco, N. and Cornell, C. A. (2007) Structure-Specific Scalar Intensity Measures for Near-Source
and Ordinary Earthquake Ground Motions. Earthquake Spectra, 23(2), pp. 357-392

Lupoi, A., Franchin, P. and Pinto, P.E. (2007) Further probing of the suitability of PUSH-OVER
analysis for the seismic assessment of bridge structures. Proceedings of the 1st US-Italy Seismic
Bridge Workshop, April 19-20, Pavia, Italy

M
Mackie, K. and Stojadinovic, B. (2004) Fragility curves for reinforced concrete highway overpass
bridges. Proceedings of the 13th World Conference on Earthquake Engineering, August 1-6,
Vancouver, Canada

Mahasuverachai, M. and Powell, G. H. (1982) Inelastic analysis of piping and tubular structures.
Report No. EERC 82/27, Earthquake Engineering Research Center, University of California,
Berkeley, CA

Mander, J. B., Priestley, M. J. N. and Fellow, R. P. (1988a) Observed Stress Strain Behavior of
Confined Concrete. Journal of Structural Engineering, 114(8), pp. 1827-1849

Mander, J. B., Priestley, M. J. N. and Park, R. (1988b) Theoretical Stress-Strain Model for
Confined Concrete. Journal of Structural Engineering, 114(8), pp. 1804-1826

Marí, A. R. and Scordelis, A. C. (1984) Nonlinear geometric, material and time dependent analysis
of three dimensional reinforced and prestressed concrete frames. UC-SESM Report No. 84-12,
University of California, Berkeley, CA

Martínez-Rueda, J. E. and Elnashai, A. S. (1997) Confined concrete model under cyclic load.
Materials and Structures, 30(3), pp. 139-147

McKay, M. D., Beckman, R. J. and Conover, W. J. (1979) A Comparison of Three Methods for
Selecting Values of Input Variables in the Analysis of Output from a Computer Code.
Technometrics, 21(2), pp. 239-245

McKenna, F. (1997) Object oriented finite element analysis: frameworks for analysis algorithms
and parallel computing. Ph. D. Thesis, Department of Civil Engineering, University of California,
Berkeley,

R.10
 References

McKenna, F. and Fenves, G. L. (2006) OpenSees - The Open System for Earthquake Engineering
Simulation OpenSees. Pacific Earthquake Engineering Research Center, Berkeley, California

Megally, S. H., Seible, F., Bozorgzadeh, A., Restrepo, J. and Silva, P. F. (2003) Response of
Sacrificial Shear Keys in Bridge Abutments to Seismic Loading. Proceedings of the FIB
Symposium on Concrete Structures in Seismic Regions, May 6-8, Athens, Greece

Mehanny, S. S. and Deierlein, G. G. (2000) Modeling of assessment of seismic performance of

composite frames with reinforced concrete columns and steel beams. Report No. 135, John A.
Blume Earthquake Engineering Research Center, Stanford

Menegotto, M. and Pinto, P. E. (1973) Method of Analysis for Cyclically Loaded RC Plane
Frames, Including Changes in Geometry and Non-Elastic Behavior of Elements Under Combined
Normal Force and Bending. Proceedings of the IABSE Symposium on Resistance and Ultimate
Deformability of Structures Acted on by Well Defined Repeated Loads, Lisbon, Portugal

MIDAS (2006) MIDAS Civil - Integrated Solution System for Bridge and Civil Engineering Midas
Information Technology Co., Ltd., Seoul, Korea

Miranda, E. (1993) Evaluation of Site Dependent Inelastic Seismic Design Spectra. Journal of
Structural Engineering, 119(5), pp. 1319-1338

Miranda, E. (2000) Inelastic Displacement Ratios for Structures on Firm Sites. Journal of
Structural Engineering, 126(10), pp. 1150-1159

Miranda, E. and Ruiz García, J. (2002) Evaluation of approximate methods to estimate maximum
inelastic displacement demands. Earthquake Engineering & Structural Dynamics, 31(3), pp. 539-
560

Miyamoto, K., Yanev, P. and Salvaterra, I. (2009) M6.3 L'Aquila, Italy. Earthquake Field
Investigation Report, Global Risk Miyamoto and Miyamoto International

Moehle, J. P. and Cavanagh, T. (1985) Confinement Effectiveness of Crossties in RC. Journal of

Structural Engineering, 111(10), pp. 2105-2120

Moehle, J. P. (1999) Preliminary Observations on the Performance of Concrete Freeway

Structures. National Information Service for Earthquake Engineering, University of California,
Berkeley

Moehle, J. P. and Eberhard, M. (1999) Earthquake Damage to Bridges. Bridge Engineering

Handbook, CRC Press, USA

Moghadam, A. S. and Tso, W. K. (2002) A pushover procedure for tall buildings. Proceedings of
the 12th European Conference in Earthquake Engineering, September 9-13, London, UK

Monteiro, R., Casarotti, C. and Pinho, R. (2008a) Using Nonlinear Static Procedures for seismic
assessment of irregular viaducts. Proceedings of the 5th European Workshop on the seismic
behaviour of Irregular and Complex Structures, September 16-17, Catania, Italy

Monteiro, R., Ribeiro, R., Marques, M., Delgado, R. and Costa, A. G. (2008b) Pushover Analysis
of RC Bridges Using Fibre Models or Plastic Hinges. Proceedings of the 14th World Conference
on Earthquake Engineering, October 12-17, Beijing, China

R.11
References

Monteiro, R., Delgado, R., Crowley, H. and Pinho, R. (2009) Avaliação da segurança sísmica de
pontes segundo diferentes metodologias. 1º Congresso Nacional de Segurança e Conservação de
Pontes, 1-3 Julho, Lisboa, Portugal (in Portuguese)

Monti, G. and Nuti, C. (1992) Nonlinear Cyclic Behavior of Reinforcing Bars Including Buckling.
Journal of Structural Engineering, 118(12), pp. 3268-3284

Monti, G., Nuti, C. and Santini, S. (1996) CYRUS: CYclic Response of Upgraded Sections. A
program for the analysis of retrofitted or repaired sections under biaxial cyclic loading including
buckling of rebars. Report DSSAR No. 2/96, Universit degli Studi G. D'Annunzio, Chieti, Italy

Muguruma, H., Watanabe, S., Tanaka, S., Sakurai, K. and Nakaruma, E. (1978) A study on the
improvement of bending ultimate strain of concrete. Journal of Structural Engineering, 24(Tokyo,
Japan), pp. 109–116

N
Nau, M. and Hall, W. J. (1984) Scaling methods for earthquake response spectra. Journal of
Structural Engineering, 110(7), pp. 1533-1548

Neuenhofer, A. and Filippou, F. C. (1998) Geometrically Nonlinear Flexibility-Based Frame Finite

Element. Journal of Structural Engineering, 124(6), pp. 704-711

Newmark, N. M. and Hall, W. J. (1982) Earthquake Spectra and Design. Earthquake Engineering
Research Institute, Berkeley, CA,

Nielson, B. G. and DesRoches, R. (2007) Seismic fragility methodology for highway bridges using
a component level approach. Earthquake Engineering & Structural Dynamics, 36(6), pp. 823-839

Novák, D., Teplý, B. and Keršner, Z. (1997) The Role of Latin Hypercube Sampling Method in
Reliability Engineering. Proceedings of the ICOSSAR 97, November 24-28, Kyoto, Japan

O
Olsson, A. and Sandberg, G. (2002) Latin Hypercube Sampling for Stochastic Finite Element
Analysis. Journal of Engineering Mechanics, 128(1), pp. 121-125

Olsson, A., Sandberg, G. and Dahlblom, O. (2003) On Latin hypercube sampling for structural
reliability analysis. Structural Safety, 25(1), pp. 47-68

Otani, S. (1974) Inelastic analysis of R/C frame structures. Journal of the Structural Division,
100(7), pp. 1433-1449

P
Padgett, J.E., Nielson, B.G. and DesRoches, R. (2008) Selection of optimal intensity measures in
probabilistic seismic demand models of highway bridge portfolios. Earthquake Engineering &
Structural Dynamics, 37(5), pp. 711-725

Panagiotou, M. and Restrepo, J. I. (2007) Design and computational model for the UCSD 7-story
structural wall building slice. SSRP 07-09 Report, University of California, San Diego, CA

R.12
 References

Papageorgiou, A., Halldorson, B. and Dong, G. (2001) TARSCTHS, a computer program for
Target Acceleration Spectra Compatible Time Histories. State University of New York, Buffalo,
NY

Papaioannou, I., Fragiadakis, M. and Papadrakakis, M. (2005) Inelastic Analysis of Framed

Structures using the Fiber Approach. Proceedings of the 5th International Congress on
Computational Mechanics (GRACM 05), June 29 - July 1, Limassol, Cyprus

Paraskeva, T. S., Kappos, A. J. and Sextos, A. G. (2006) Extension of modal pushover analysis to
seismic assessment of bridges. Earthquake Engineering & Structural Dynamics, 35(10), pp. 1269-
1293

Paret, T. F., Sasaki, K. K., Eilbeck, D. H. and Freeman, S. A. (1996) Approximate inelastic
procedures to identify failure mechanisms from higher mode effects. Proceedings of the 11th
World Conference on Earthquake Engineering, June 23-28, Acapulco, Mexico

Park, R., Kent, D. C. and Sampson, R. A. (1972) Reinforced Concrete Members with Cyclic
Loading. Journal of the Structural Division, 98(7), pp. 1341-1360

Park, R. and Paulay, T. (1975) Reinforced concrete structures. John Wiley & Sons, Inc., New York

Park, R., Priestley, M. J. N. and Gill, W. D. (1982) Ductility of Square-Confined Concrete

Columns. Journal of the Structural Division, 108(4), pp. 929-950

Paulay, T. and Priestley, M. J. N. (1992) Seismic design of reinforced concrete and masonry
buildings. John Wiley & Sons, Inc., New York

Pinho, R. and Antoniou, S. (2005) A displacement-based adaptive pushover algorithm for

assessment of vertically irregular frames. Proceedings of the 4th European Workshop on the
Seismic Behaviour of Irregular and Complex Structures, August 26-27, Thessaloniki, Greece

Pinho, R., Casarotti, C. and Monteiro, R. (2007) An Adaptive Capacity Spectrum Method and other
Nonlinear Static Procedures Applied to the Seismic Assessment of Bridges. Proceedings of the 1st
US-Italy Seismic Bridge Workshop, April 19-20, Pavia, Italy

Pinho, R., Monteiro, R., Casarotti, C. and Delgado, R. (2009) Assessment of Continuous Span
Bridges through Nonlinear Static Procedures. Earthquake Spectra, 25(1), pp. 143-159

Pinho, R. (2011) Seismic Assessment and Retrofitting of Existing Structures. Lecture Notes, ROSE
School, Centre for Post-Graduate Training and Research in Earthquake Engineering and
Engineering Seismology, Pavia, Italy

Pinto, A. V., Verzeletti, G., Pegon, P., Magonette, G., Negro, P. and Guedes, J. P. M. (1996)
Pseudo-Dynamic Testing of Large-Scale R/C Bridges. Report EUR 16378 EN, Joint Research
Centre, Ispra, Italy

Pinto, P. E. and Giuffrè, A. (1970) Il Comportamento Del Cemento Armato Per Sollecitazioni
Cicliche di Forte Intensità. Giornale del Genio Civile, 5 (in Italian)

Pinto, P. E. (2001) Reliability methods in earthquake engineering. Progress in Structural

Engineering and Materials, 3(1), pp. 76-85

R.13
References

Pinto, P. E., Fajfar, P., Chryssanthopoulos, M., Franchin, P., Dolšek, M. and Kazantzi, A. (2007)
Probabilistic Methods for Seismic Assessment of Existing Structures. LESSLOSS Report No.
2007/06, IUSS Press, Pavia, Italy

Pipa, M. and Carvalho, E. C. (1994) Reinforcing Steel Characteristics for Earthquake Resistant
Structures. Proceedings of the 10th European Conference on Earthquake Engineering, August 28 -
September 2, Vienna, Austria

Priestley, M. J. N. and Park, R. (1984) Strength and ductility of bridge substructures. Research
Report no. 84-20, University of Canterbury, Christchurch, New Zealand

Priestley, M. J. N. (1993) Myths and Fallacies in Earthquake Engineering - Conflicts Between

Design and Reality. Proceedings of the Thomas Paulay Symposium: Recent Developments in
Lateral Force Transfer in Buildings, SP-157, ACI Special Publication, UC San Diego, La Jolla,
USA

Priestley, M. J. N., Seible, F. and Calvi, G. M. (1996) Seismic design and retrofit of bridges.
Wiley-Interscience, New York

Priestley, M. J. N. (2003) Myths and fallacies in earthquake engineering, revisited. The 9th Mallet-
Milne Lecture, 2003, IUSS Press, Pavia, Italy

Priestley, M. J. N. and Calvi, G. M. (2003a) Direct Displacement-Based Seismic Design of

Bridges. Proceedings of the ACI Special Seminar on Seismic Design of Bridges, San Diego, USA

Priestley, M. J. N. and Calvi, G. M. (2003b) Direct displacement-based seismic design of concrete

bridges. Proceedings of the 5th ACI International Conference of Seismic Bridge Design and
Retrofit for Earthquake Resistance, December 8-9, La Jolla, USA

Priestley, M. J. N. and Grant, D. N. (2005) Viscous damping in seismic design and analysis.
Journal of Earthquake Engineering, 9(2), pp. 229-255

Priestley, M. J. N., Calvi, G. M. and Kowalsky, M. J. (2007) Displacement-based Seismic Design

of Structures. IUSS Press, Pavia, Italy

R
Ramberg, W. and Osgood, W. R. (1943) Description of Stress-Strain Curves by Three Parameters.
Technical Note No. 902, National Advisory Committee for Aeronautics, Washington

Ramirez, O. M., Constantinou, M. C., Whittaker, A. S., Kircher, C. A. and Chrysostomou, C. Z.

(2002) Elastic and Inelastic Seismic Response of Buildings with Damping Systems. Earthquake
Spectra, 18(3), pp. 531-547

Reinhorn, A. (1997) Inelastic analysis techniques in seismic evaluations. Seismic Design

Methodologies for the Next Generation of Codes: Proceedings of the International Workshop, June
24-27, Bled, Slovenia

Rosenblueth, E. and Herrera, I. (1964) On a kind of hysteretic damping. Journal of Engineering

Mechanics, 90(4), pp. 37-48

R.14
 References

S
Saatcioglu, M. and Razvi, S. R. (1992) Strength and Ductility of Confined Concrete. Journal of
Structural Engineering, 118(6), pp. 1590-1607

SAC (1997) Develop Suites of Time Histories. Project Task 5.4.1 Draft Report, SAC Joint Venture,
Sacramento, CA

Saiidi, M. and Sozen, M. A. (1981) Simple Nonlinear Seismic Analysis of R/C Structures. Journal
of the Structural Division, 107(5), pp. 937-953

Sandhu, J. S., Stevens, K. A. and Davies, G. A. O. (1990) A 3-D, co-rotational, curved and twisted
beam element. Computers & Structures, 35(1), pp. 69-79

Scott, B. D., Park, R. and Priestley, M. J. N. (1982) Stress-Strain Behavior of Concrete Confined
by Overlapping Hoops at Low and High Strain Rates. ACI Journal Proceedings, 79(1), pp. 13-27

Scott, M. H. and Filippou, F. C. (2007) Response Gradients for Nonlinear Beam-Column Elements
under Large Displacements. Journal of Structural Engineering, 133(2), pp. 155-165

SeismoSoft (2008) SeismoStruct - A computer program for static and dynamic nonlinear analysis
of framed structures, available online from http://www.seismosoft.com. SeismoSoft Ltd, Pavia,
Italy

Sextos, A. G., Pitilakis, K. D. and Kappos, A. J. (2003) Inelastic dynamic analysis of RC bridges
accounting for spatial variability of ground motion, site effects and soil–structure interaction
phenomena. Part 1: Methodology and analytical tools. Earthquake Engineering & Structural
Dynamics, 32(4), pp. 607-627

Sheikh, S. A. and Uzumeri, S. M. (1979) Properties of Concrete Confined by Rectangular Ties.

Proceedings of the AICAP-CEB Symposium on Structural Concrete Under Seismic Actions, CEB
Bulletin d'Information no. 132, pp. 53-60

Sheikh, S. A. and Uzumeri, S. M. (1980) Strength and Ductility of Tied Concrete Columns.
Journal of the Structural Division, 106(5), pp. 1079-1102

Sheikh, S. A. and Uzumeri, S. M. (1982) Analytical Model for Concrete Confinement in Tied
Columns. Journal of the Structural Division, 108(12), pp. 2703-2722

Shibata, A. and Sozen, M. A. (1976) Substitute-Structure Method for Seismic Design in R/C.
Journal of the Structural Division, 102(1), pp. 1-18

Shinozuka, M., Feng, M. Q., Lee, J. and Naganuma, T. (2000) Statistical Analysis of Fragility
Curves. Journal of Engineering Mechanics, 126(12), pp. 1224-1231

Shome, N. and Cornell, C. A. (1998) Normalization and scaling accelerograms for nonlinear
structural analysis. Proceedings of the 6th US National Conference on Earthquake Engineering,
May 31 - June 4, Seattle, USA

Shome, N., Cornell, C., Bazzurro, P. and Carballo, J. E. (1998) Earthquakes, Records, and
Nonlinear Responses. Earthquake Spectra, 14(3), pp. 469-500

Shome, N. and Cornell, C. A. (1999) Probabilistic seismic demand analysis of nonlinear

structures. Report No. RMS-35 (Ph.D. Thesis), Stanford University, Stanford, CA

R.15
References

Sinha, B. P., Gerstle, K. H. and Tulin, L. G. (1964) Stress-Strain Relations for Concrete under
Cyclic Loading. ACI Journal Proceedings, 61(2), pp. 197-211

Spacone, E. (2001) A Module for Analysis and Design of Segmental Prestressed Concrete Bridges
(CASI-TR-01-04). Final Report of a CASI FY00 Technology Transfer Grant, Colorado Advanced
Software Institute, Fort Collins, CO

Spacone, E., Camata, G. and Faggella, M. (2008) Nonlinear models and nonlinear procedures for
seismic analysis of reinforced concrete frame structures. Computational Structural Dynamics and
Earthquake Engineering, Taylor & Francis,

Stanton, J. F. and McNiven, H. D. (1979) The development of a mathematical model to predict the
flexural response of reinforced concrete beams to cyclic loads, using system identification. EERC
Report No. 79/02, Earthquake Engineering Research Center, University of California, Berkeley,
CA

T
Takeda, T., Sozen, M. and Nielsen, N. (1970) Reinforced Concrete Response to Simulated
Earthquakes. Journal of the Structural Division, 96(12), pp. 2557-2573

Takizawa, H. (1976) Notes on some basic problems in inelastic analysis of planar RC structures.
No. 240, Transactions of the Architectural Institute of Japan, Tokyo

Tothong, P. and Luco, N. (2007) Probabilistic seismic demand analysis using advanced ground
motion intensity measures. Earthquake Engineering & Structural Dynamics, 36(13), pp. 1837-1860

Tothong, P. and Cornell, C. A. (2008) Structural performance assessment under near-source pulse-
like ground motions using advanced ground motion intensity measures. Earthquake Engineering &
Structural Dynamics, 37(7), pp. 1013-1037

V
Vallenas, J., Bertero, V. V. and Popov, E. P. (1977) Concrete confined by rectangular hoops and
subjected to axial loads. Report No. UCB/EERC 77/13, Earthquake Engineering Research Center,
University of California, Berkeley, CA

Varum, H. S. A. (1996) Modelo Numérico para a Análise Sísmica de Pórticos Planos de Betão
Armado. M.Sc. Thesis, University of Porto, Porto, Portugal (in Portuguese)

Vaz, C. (1992) Comportamento Sísmico de Pontes com Pilares de Betão Armado - Verificação da
segurança. Ph.D. Thesis, University of Porto, Porto, Portugal (in Portuguese)

Vidic, T., Fajfar, P. and Fischinger, M. (1994) Consistent inelastic design spectra: Strength and
displacement. Earthquake Engineering & Structural Dynamics, 23(5), pp. 507-521

W
Wakabayashi, M. (1986) Design of earthquake-resistant buildings. McGraw Hill, New York

Wilson, E. L. (2002) Three-Dimensional Static and Dynamic Analysis of Structures, A Physical

Approach With Emphasis on Earthquake Engineering. CSi - Computers and Structures Inc.,
Berkeley, CA

R.16
 References

Y
Yashinsky, M. (1998a) Cypress Street Viaduct. US Geological Survey Professional Paper No.
1552-8, pp. 19-26

Yashinsky, M. (1998b) Performance of Bridge Seismic Retrofits during Northridge Earthquake.

Journal of Bridge Engineering, 3(1), pp. 1-14

Yassin, M. H. M. (1994) Nonlinear analysis of prestressed concrete structures under monotonic and
cyclic loads. Ph.D. Thesis, University of California, Berkeley, USA

Yeh, Y. K., Mo, Y. L. and Yang, C. Y. (2002) Seismic Performance of Rectangular Hollow Bridge
Columns. Journal of Structural Engineering, 128(1), pp. 60-68

Z
Zeris, C. A. and Mahin, S. A. (1988) Analysis of reinforced concrete beam-columns under uniaxial
excitation. Journal of Structural Engineering, 114(4), pp. 804-820

Zeris, C. A. and Mahin, S. A. (1991) Behavior of reinforced concrete structures subjected to biaxial
excitation. Journal of Structural Engineering, 117(9), pp. 2657-2673

R.17