EBOOKS Multi-scale Reliability and

LU • NOORI
FOR THE SUSTAINABLE STRUCTURAL
ENGINEERING
Serviceability Assessment of SYSTEMS COLLECTION
LIBRARY In-service Long-span Bridges Mohammad Noori, Editor
Create your own
Naiwei Lu • Mohammad Noori
Customized Content
Bundle—the more With the development in global economic and transportation
books you buy, engineering, the traffic loads on brides have been g
­ ­ rowing

Multi-scale Reliability and Serviceability Assessment of In-service Long-span Bridges

the greater your steadily, which become potential safety hazards for e ­ xisting
discount! bridges. In particular, long-span suspension bridges s­upport

Multi-scale
heavy traffic volumes and simultaneous truck loads on the
THE CONTENT bridge deck, and thus the safety and serviceability of the bridge
• Manufacturing deserves investigation. In this book, a multiscale reliability
Engineering
• Mechanical
­method is presented for the safety assessment of long-span
bridges. The multiscale failure condition of stiffness girders Reliability and
Serviceability
& Chemical is the first-passage criteria for the large-scale model and the
Engineering ­fatigue damage criteria for the small-scale model.
• Materials Science
& Engineering
• Civil &
Environmental
It is the objective of this book to provide a more in-depth
understanding of the vehicle-bridge interaction from the
­
random vibration perspective. This book is suitable for
­
Assessment
Engineering
• Advanced Energy
­adoption as a text book or a reference book in an advanced
structural ­reliability analysis course. Furthermore, this book also
of In-service
Long-span
Technologies ­provides a theoretical foundation for better understanding of
the safety assessment, operation management, maintenance
THE TERMS and ­reinforcement for long-span bridges and motivates further
• Perpetual access for
a one time fee
• No subscriptions or
research and development for more advanced reliability and
serviceability assessment techniques for long-span bridges. Bridges
access fees Naiwei Lu is a lecturer of civil engineering at Changsha
• Unlimited ­University of Science and Technology.
concurrent usage
• Downloadable PDFs Mohammad Noori is professor of mechanical engineering and
• Free MARC records ASME fellow in the department of mechanical engineering,

For further information,

California Polytechnic State University and distinguished visit- Naiwei Lu
a free trial, or to order,
contact: 
ing national chaired professor of 1000 Program, International
Institute for Urban Systems Engineering at Southeast University
Mohammad Noori
[email protected] in Nanjing China.

ISBN: 978-1-94708-338-7
 MULTI-SCALE
RELIABILITY AND
SERVICEABILITY
ASSESSMENT OF
IN-SERVICE LONG-
SPAN BRIDGES
 MULTI-SCALE
RELIABILITY AND
SERVICEABILITY
ASSESSMENT OF
IN-SERVICE LONG-
SPAN BRIDGES

NAIWEI LU AND MOHAMMAD NOORI

MOMENTUM PRESS, LLC, NEW YORK

Multi-scale Reliability and Serviceability Assessment of In-service
Long-span Bridges

Copyright © Momentum Press®, LLC, 2018.

All rights reserved. No part of this publication may be reproduced, stored

in a retrieval system, or transmitted in any form or by any means—­
electronic, mechanical, photocopy, recording, or any other—except for
brief quotations, not to exceed 400 words, without the prior permission
of the publisher.

First published by Momentum Press®, LLC

222 East 46th Street, New York, NY 10017
www.momentumpress.net

ISBN-13: 978-1-94708-338-7 (print)

ISBN-13: 978-1-94708-339-4 (e-book)

Momentum Press Sustainable Structural Systems Collection

Collection ISSN: 2376-5119 (print)

Collection ISSN: 2376-5127 (electronic)

Cover and interior design by Exeter Premedia Services Private Ltd.,

Chennai, India

10 9 8 7 6 5 4 3 2 1

Printed in the United States of America

 Abstract

A large number of long-span bridges are under construction or have been

constructed all over the world. The steady increase in traffic volume and
gross vehicle weight has caused a threat to the serviceability or even safety
of in-service bridges. Therefore, ensuring the safety and serviceability of
these bridges has become a growing concern. In particular, long-span sus-
pension bridges support heavy traffic volumes and experience consider-
able wind loads on the bridge deck on a regular basis. Excessive dynamic
responses may cause large deformation and undesirable vibration of the
stiffening girders. In practice, a bridge suffers from multiple types of
loadings in the lifecycle. In this book, a multiscale reliability method
is presented for the safety assessment of long-span bridges. The multi-
scale failure condition of stiffness girders is the first-passage criteria for
the large-scale model and the fatigue damage criteria for the small-scale
model. It is the objective of this book to provide a more in-depth under-
standing of the vehicle-bridge interaction from the random vibration per-
spective. This book is suitable for adoption as a text book or a reference
book in an advanced structural reliability analysis course. Furthermore,
this book also provides a theoretical foundation for better understand-
ing of the safety assessment, operation management, maintenance and
reinforcement for long-span bridges and motivates further research and
development for more advanced reliability and serviceability assessment
techniques for long-span bridges.

KEYWORDS

bridge engineering, dynamics, fatigue damage, reliability evaluation,

traffic load
 Contents

List of Figures xi

List of Tables xvii

Acknowledgments xix
1 Introduction 1
1.1 Research Significance 1
1.2 State-of-the-art Review 4
References8
2 Fatigue Reliability Assessment of Welded Steel Bridge
Decks Under Stochastic Truck Loads  13
2.1 Introduction 13
2.2 Stochastic Fatigue Truck Load Model 16
2.3 Computational Framework of Probabilistic Modeling 21
2.4 Limit State Function of Fatigue Damage Accumulation 25
2.5 Case Study 27
2.6 Results and Discussion 36
2.7 Conclusions 39
References40
3 First-Passage Probability of the Deflection of a
Cable-Stayed Bridge Under Long-term Site-Specific
Traffic Loading 45
3.1 Introduction 45
3.2 Stochastic Traffic Load Simulation Based on WIM
­Measurements 47
3.3 Methodology 50
viii  •   Contents

3.4  Case Study 54

3.5 Conclusions 60
References61
4 Dynamic Reliability Evaluation of the Serviceability
of Cable-Supported Bridges Under Site-Specific Heavy
Traffic Loads 65
4.1 Introduction 65
4.2 Traffic–Bridge Interaction Formulation 68
4.3 Methodology of Probabilistic Modeling of the Extreme
Traffic Load Effects 70
4.4 Probabilistic Modeling of the Traffic Load Effects on Two
Cable-supported Bridges Using WIM Measurements 74
4.5 Dynamic Reliability Evaluation of the Bridge Deflection
­Serviceability 81
4.6 Conclusions 85
References86
5 Lifetime Deflections of Long-Span Bridges Under
Dynamic and Growing Traffic Load 89
5.1 Introduction 89
5.2 Theoretical Bases of the Traffic-bridge Interaction and
Rice’s Formula 92
5.3 Methodology of Evaluating Maximum Load Effects
­Considering Interval Traffic Growth 95
5.4 A Computational Framework of Extrapolating the
Maximum Load Effect FF Long-Span Bridges 98
5.5 Verification Examples 100
5.6 Case Study 104
5.7 Conclusions 114
References115
6 System Reliability Evaluation of in-Service Cable-Stayed
Bridges Subject to Cable Degradation Via a Machine
Learning Based Tool 119
6.1 Introduction 119
6.2 Formulations of Cable Degradation 121
6.3 Framework of System Reliability Evaluation 124
 Contents   •   ix

6.4 Program Implementation of the Framework 127

6.5 Case Studies 128
6.6 Conclusions 137
References138
About the Authors 141
Index 143
 List of Figures

Figure 1.1.  Suspension bridge collapse caused by the combined

effect of winds and vehicles. 3
Figure 2.1.  Fatigue truck load models in the design specifications:
(a) AASHTO with GVW of 240 kN; (b) Eurocode 1
with GVW of 480 kN. 17
Figure 2.2.   PDFs of: (a) GVW of V6; (b) AW64. 
19
Figure 2.3.  Simulated stochastic fatigue truck loads in:
(a) slow lane; (b) fast lane. 20
Figure 2.4.   Proposed computational framework. 21
Figure 2.5.  Dimensions of a steel box-girder: (a) cross section;
(b) U-rib. 28
Figure 2.6.  Finite element model of a half-segment steel
box-girder.28
Figure 2.7.  Stress influence lines of welded joints in two types
of finite element models. 29
Figure 2.8.  Stress–time histories under the V6 truck loads for the
welded joints of: (a) rib-to-deck joint; (b) butt joint
of U-rib. 30
Figure 2.9.  Response surfaces for the 2rd type of vehicle with the
training sample of: (a) U10; (b) U20. 31
Figure 2.10. Comparison between the SVR predictions
with FEA results. 32
Figure 2.11. PDFs of the equivalent fatigue stress in the
rib-to-deck joints of: (a) slow lane; (b) fast lane. 33
Figure 2.12. PDFs of the equivalent number of daily cycles for:
(a) slow lane; (b) fast lane. 35
Figure 2.13.  Fatigue reliability indices of the rib-to-deck joint. 37
xii  •   List of Figures

Figure 2.14. Fatigue reliability indices taking into account the

linear annul growth factors of: (a) traffic volume;
(b) vehicle weight. 38
Figure 3.1.  Probability densities of the traffic flow: (a) proportions
of vehicle types; (b) the time-variant vehicle density;
(c) ADTT; (d) vehicle spacing of the busy traffic flow. 49
Figure 3.2.  Histograms and PDFs of the 6-axle: (a) axle weight;
and (b) GVW. 50
Figure 3.3.  Simulated stochastic traffic loads: (a) the present case;
and (b) the 100th year. 51
Figure 3.4.   Description of Rice’s level crossing theory. 51
Figure 3.5.  Computational framework for first-passage reliability
evaluation of a long-span bridge under traffic loading. 54
Figure 3.6.   Dimensions of a cable-stayed bridge. 55
Figure 3.7.  Finite element model of the cable-stayed bridge. 55
Figure 3.8.  Displacement influence lines of the critical points
of the bridge. 56
Figure 3.9.  Displacement histories of the critical points of
the bridge. 56
Figure 3.10.  Up-crossing histograms and fittings. 56
Figure 3.11. Tail fittings of the extreme load effects on Gamble
paper.57
Figure 3.12. Extrapolated maximum displacements during the
service period with consideration of (a) growth rate
of ADTT; (b) growth rate of GVW. 58
Figure 3.13. Influence of the traffic growth on the first-passage
probability of failure. 59
Figure 3.14. Influence of the threshold overload ratio on the
probability of failure. 59
Figure 4.1.  Physical model of a 2-axle truck: (a) elevation view;
(b) side view. 69
Figure 4.2.  Basis of Rice’s level-crossing theory: (a) crossings;
(b) fitting to the crossings. 71
Figure 4.3.  Flow chart of the computational framework
computational for extrapolation of maximum traffic
load effects. 73
Figure 4.4.  Histograms and PDFs of the 6-axle truck: (a) vehicle
spacing (b) GVWs. 76
 List of Figures   •   xiii

Figure 4.5.  A critical traffic loading scenario identified from the

stochastic traffic-load model. 76
Figure 4.6.  The first vertical mode shapes and dimensions:
(a) cable-stayed bridge; (b) suspension bridge. 77
Figure 4.7.  Deflection histories of the critical points under
one-hour dense traffic loading: (a) cable stayed bridge;
(b) suspension bridge. 78
Figure 4.8.  Statistical characteristics of the two cable-supported
bridges: (a) the mean value; (b) the standard deviation. 79
Figure 4.9.  Deflection histories of the critical traffic loading on the:
(a) cable-stayed bridge; (b) suspension bridge. 80
Figure 4.10. Histograms and Rice’s fittings of the number of level
crossings.81
Figure 4.11.  Extrapolation of the extreme traffic load deflections. 81
Figure 4.12. Probability of failure of the critical girder duo to
displacement up-crossing. 83
Figure 4.13. Influence of growth rate of ADTT on the probability of
failure.83
Figure 4.14. Influence of the threshold of overloading ratio on the
probability of failure. 84
Figure 5.1.  Physical models of a 2-axle truck: (a) elevation view;
(b) side view. 93
Figure 5.2.  Basic principles of Rice’s formula: (a) level crossings;
(b) fitting to the crossings. 94
Figure 5.3.  An interval traffic growth model: (a) interval ADTTs;
(b) a series system model. 96
Figure 5.4.  Flowchart of the proposed computational framework
for the lifetime maximum traffic load effect
extrapolation.99
Figure 5.5.  Analytical results of the first example: (a) annual
crossing rates; (b) daily maxima and fittings
on Gumbel paper. 102
Figure 5.6.   An idealized long-span bridge and crossing vehicles. 103
Figure 5.7.  Analytical results of the second examples: (a) annual
crossing rates; (b) daily maxima and fittings plotted
on Gumbel paper. 104
Figure 5.8.  Statistics of the WIM measurements: (a) hourly traffic
volume; (b) truck spacing of the busy traffic flow. 106
xiv  •   List of Figures

Figure 5.9.  Finite-element model and dimensions of the

suspension bridge. 107
Figure 5.10.  Deflection histories of critical points of the girder
under a 2-axle truck load. 108
Figure 5.11.  An example of identifying the critical loading scenario:
(a) a daily deflection history; (b) a critical loading
scenario. 109
Figure 5.12. An example to show the different between the numbers
of crossings of static and dynamic histories. 110
Figure 5.13.  Histograms and fittings of the numbers of crossings. 110
Figure 5.14. Extrapolations of the maximum deflections
considering the RRC. 111
Figure 5.15. Time-variant level-crossing rates accounting for
traffic growth. 112
Figure 5.16. Extrapolation of the lifetime maximum deflection
accounting for traffic growth. 112
Figure 5.17. Probabilistic assessment of the bridge deflection
under growing traffic loads: (a) CDFs plotted on
Gamble paper; (b) probability of exceedance. 113
Figure 6.1.   A parallel–series system of a stay cable. 122
Figure 6.2.  Strength coefficients of a stay cable due to fatigue and
corrosion effects. 123
Figure 6.3.   Flowchart of the proposed computational framework. 126
Figure 6.4.  Flowchart of the complex structural reliability analysis
software.127
Figure 6.5.  Dimensions and failure modes of a short-span
cable-stayed bridge. 128
Figure 6.6.  Event trees of the cable-stayed bridge: (a) without
cable degradation; (b) with a cable degradation
coefficient of 20%. 130
Figure 6.7.  System reliability index of the cable-stayed bridge
subject to cable degradation. 132
Figure 6.8.  Dimensions of the long-span cable-stayed bridge:
(a) elevation layout; (b) longitudinal layout. 133
Figure 6.9.  Finite element model of the cable-stayed bridge. 134
Figure 6.10. Response histories of the critical points subject to
sudden rupture of the Cm34 cable: (a) cable forces;
(b) bending moments. 135
 List of Figures   •   xv

Figure 6.11. Response surfaces of the Cm33 cable force due to

Cm34 rupture. 135
Figure 6.12. Event trees of the long-span cable-stayed bridge:
(a) T = 0 year; (b) T = 20 years. 136
Figure 6.13. System reliability indices of the cable stayed bridge
subject to cable degradation. 137
 List of Tables

Table 2.1. Vehicle classifications 18

Table 2.2. Parameters of S-N curves in the Eurocode 3 (2005)
specification 26
Table 2.3. Parameters in the GMMs of Δσre 34
Table 2.4. Statistics of the random variables in the limit state
function36
Table 2.5. Fatigue life predication for the rib-to-deck joint
taking into account traffic growth factors 38
Table 3.1. Overview of the WIM measurements 48
Table 4.1. Overview of WIM measurements 75
Table 4.2. Bridge deflection limits 82
Table 5.1. Overview of the filtered WIM measurements 105
Table 5.2. The first five order mode frequencies of the
suspension bridge 107
Table 6.1. Properties of the fourth longest stay cables 133
 Acknowledgments

This book was supported by the National Basic Research Program

(973 program) of China (Grant 2015CB057705), the National Science
Foundation of China (Grant 51378081), the Hunan Natural Science
­Funding (2018JJ3540), and the funding in Hunan Province Engineering
Laboratory of Bridge Structure (16KD03). The book funding provided for
the Changsha University of Science & Technology and the Construction
Project of Preponderant Discipline of Jiangsu Universities in Southeast
University is highly appreciated.
In addition, the authors would like to sincerely express their appre-
ciation to the following researchers: Dr. Yuan Luo, who has c­ ontributed
to Chapter 2; Dr. Yafei Ma, who has contributed to Chapter 3; and
Dr. Qinyong Wang, who has contributed to Chapter 4; Dr. Fanghuai
Chen, who has contributed to Chapter 5; and Ms. Ying Chen, who has
Contributed to Chapter 6. The help from the authors’ research team is also
highly appreciated.
Finally, the first author would like to appreciate his parameters and
the cute wife for their unconditional love and dedication, and to give the
book as a gift for the forthcoming little baby.
 CHAPTER 1

Introduction

1.1 RESEARCH SIGNIFICANCE

Bridges are usually subjected to harsh environmental effects and complex

loading conditions. These effects may subsequently result in changes in
the structural behavior, dynamic characteristics and resistance of a bridge
during its life-time. These important changes that need to be better under-
stood are time variant phenomena [1]. For most existing bridges, support-
ing continuous traffic loads is the basic function. However, in addition to
this load, bridges are also exposed to environmental [2], as well as various
other complex loads, that are mostly random in nature, such as earthquake,
flow-included loadings, wind and so on. All these loading conditions need
to be considered and must satisfy the design criteria. In recent years, rapid
growth of urban systems and the sprawling of large cities, have resulted in
significant increase in traffic volume and the corresponding vehicle weight
on bridges. This new phenomenon, which for most part may not have
accounted for when most bridges were originally constructed, has resulted
in a threat to the safety of bridges [3]. The most frequent causes of bridge
failures were attributed to overloading due to vehicles, besides floods and
scouring [4]. The vehicle overloading is the main human factor resulting
in shortening the service life and even directly causing collapse of bridges
in most counties [5]. In addition to the overloading vehicles, dynamic
problems of long-span bridges have become increasingly significant with
the increment in bridge span and flexibility. The sensitivity to dynamic
wind actions increases with the reduction of modal frequencies [6]. As it
can be seen from the earlier literatures, safety problems of existing bridges
caused by vehicle loads due to sustainable growths of traffic volume and
strong winds are becoming serious issues with the fast development of
urban areas. Therefore, the safety assessment of bridges is extremely
important. If more accurate and more reliable safety assessment meth-
odologies can be developed, it facilitates intervention strategies such as
2  •   BRIDGE RELIABILITY AND SERVICEABILITY

maintenance and reinforcement, which could be adopted to maintain the

performance over certain thresholds according to the safety assessment
results. Furthermore, since most bridges are throat of the traffic systems,
their safety assurance is the foundation of economic development and the
safety and reliability of infrastructure systems.
Suspension, as well as cable stayed, bridges, in particular, are widely
used in highways crossing gorges, rivers, and gulfs, due to their supe-
rior advantages such as mechanical properties, large spanning ability,
and appealing aesthetic appearance. The number and the span of the sus-
pension bridges are increasing gradually along with the advancements of
computational capabilities and the construction technology. However, the
safe performance of these long-span suspension bridges are facing numer-
ous threats such as suffering from the incremental gross vehicle loads,
strong winds, and other natural disasters [7]. There are numerous struc-
tural, mechanical and loading characteristic differences between suspen-
sion bridges and other short-span bridges, such as higher traffic volume,
simultaneous presence of multiple vehicles, sensitivity to wind load, and
inherent nonlinearities [8]. The basic load combination methods, that are
based on the currently used design codes and deterministic analysis meth-
ods may not be suitable for the safety assessment of suspension bridges,
since interactions between the bridge, the loadings, and the environ-
mental factors are ignored in the current analysis methods. On the other
hand, the randomness of these loadings is not appropriately considered.
Furthermore, the environmental conditions of suspension bridges during
their life-time are usually harsh which means that, (1) suspension bridges
are normally located in throat position of highways where busy traffic flow
and heavy vehicles usually emerge, (2) the environmental surroundings
of suspension bridges produce strong winds which may cause a harsh
structural vibration [9]. The large deformation and strong vibration caused
by the increasing vehicle load and strong wind load directly threaten the
safety of bridges and comfort of passengers [10]. Two famous suspension
bridges that collapsed as a result of wind and vehicle load are shown in
Figure 1.1. These historical cases demonstrate the root cause of these cata-
strophic failures which was the result of design flaws by ignoring influence
of multiple loads on bridge dynamic responses. Considering that the two
most important and common live loads acting on a bridge, namely the traf-
fic flow and the wind load, that are the main cause of a bridge failure, are
inherently stochastic in nature, demonstrates the random vibration analysis
of suspension bridge girders possesses great significance for better under-
standing the probabilistic dynamic reliability assessment. Random vibra-
tion analysis of suspension bridge girders possesses great significance for
 Introduction   •  3

(a) Tacoma bridge collapse (b) Maitong bridge collapse

Figure 1.1.  Suspension bridge collapse caused by the combined effect of

winds and vehicles.

better understanding the probabilistic dynamic performance of suspension

bridges. Furthermore, the corresponding dynamic reliability assessment
for in service suspension bridges or similar long-span bridges can provide
theoretical foundation for safety evaluation, operations management, and
maintenance of these bridges.
There are interactions among the bridge, the wind, and the vehicle
since they work together as a system. The impact factor pattern may not
replace the bridge real responses considering the previous interaction sys-
tem [11]. There are three main contents for the interaction among wind,
bridge, and the vehicles [12]. First, there are interactions between vehicles
and the bridge which means that the moving vehicles and road roughness
excitation lead to the bridge vibration which changes the vehicle vibration
subsequently. Turbulent wind loads generate buffeting force on bridges,
which directly change the transient vehicle-bridge coupled vibration.
Finally, the vehicle changes the aerodynamic wind speed. In the proba-
bilistic domain, the stochastic traffic flow and wind load possess promi-
nent randomness and correlation. Furthermore, these loads on nonlinear
suspension bridge structure, the interaction system will cause random
vibration. In conclusion, the safety evaluation of suspension bridges under
combined effect of traffic flow and wind contains two contents: Multi-
load and bridge interaction analysis and the dynamic reliability evaluation
based on random vibration.
In addition to violent vibration of large-scale girders, cumulative
fatigue damage of girder details is another important issue that should not
be ignored. Orthotropic plate which is usually used in steel box girders of
suspension bridges possesses a complex structure and is sensitive to weld-
ing residual stress and construction quality. The stiffness and fatigue life
will deteriorate under the long-term live loads such as vehicle loads and
wind loads [13]. In fact, many bridges suffer a “sudden collapse” during
4  •   BRIDGE RELIABILITY AND SERVICEABILITY

the operating period. This phenomenon can be interpreted as the stiffness

and load-carrying capacity degeneration caused by fatigue damage and
the environmental influence. On this basis, internal forces of several key
components reach their limits under extreme events, and then lead to the
collapse. Above all, a comprehensive understanding of fatigue damage
and random vibration for the purpose of evaluating the dynamic reliability
is more realistic.
The overall objective of this book is to introduce the development of
a comprehensive framework for multi-scale dynamic reliability estima-
tion of suspension bridges under stochastic traffic flows and wind loads.
The framework will contain stochastic traffic flows and winds simulation,
finite element modeling, first-passage reliability based on random vibra-
tion, fatigue reliability based on the criterion of accumulative damage.
Authors hope and envision that the concepts and the approaches presented
in this book will provide a better understanding of traffic-wind-bridge
interaction system in probabilistic domain. Furthermore, it can provide a
theoretical foundation for safety assessment, operation management, and
maintenance and reinforcement for long-span bridges.

1.2 STATE-OF-THE-ART REVIEW

1.2.1 DYNAMIC PERFORMANCE OF LONG-SPAN BRIDGES

UNDER VEHICLE AND WIND LOADS

The literature review related to this important topic, presented in this

section, will provide a necessary and historical background on the devel-
opments of live loads characteristics, vehicle-bridge interaction, and the
wind induced vehicle-bridge interaction.
Compared with short-span bridges, long-span bridges exhibit unique
features such as higher traffic volume, simultaneous presence of multiple
vehicles, and sensitivity to strong wind excitations. Wind loads and vehi-
cle loads are two main continuous variable loads for long-span in-service
bridges. As stated earlier, stochastic traffic flow should be considered in
the analysis of long-span bridges since the random loading caused due
to traffic flow results in a direct and severe vibration of bridges, com-
pared with, and in contrast to, the transient vibration of a single vehicle.
Regarding the probabilistic model of vehicles, the weigh-in-motion (WIM)
technology has been used in the statistical analysis of vehicle character-
istics such as vehicle speed, axle weight, and vehicle type [14]. Usually,
Monte Carlo simulation approach has been adopted to simulate a similar
 Introduction   •  5

stochastic traffic flow with consideration of probability of vehicle type,

axle weight, vehicle distance, and vehicle speed [15]. A comprehensive
Monte Carlo simulation method for free-flowing traffic was presented and
demonstrated by measuring the data of five European highway bridges
[16]. In order to model the vehicle state more realistically, a cellular auto-
mation based traffic flow simulation technique was proposed to simulate
the stochastic live load from traffic for long-span bridges [17].
Research work on the topic of the interaction between vehicles and
bridges originated in the middle of the 20th century. In the beginning, the
vehicle loads were modeled as a constantly moving force, moving mass,
or moving mass-spring. Further progress in this research area led to a fully
computerized approach for assembling equations of motions of coupled
vehicle-bridge, which was proposed by modeling the vehicles as a com-
bination of a number of rigid bodies connected by a series of springs and
dampers [18]. On this basis, a 3-D simulation approach including a 3-D
suspension vehicle model and a 3-D dynamic bridge model was devel-
oped (Shi et al. 2008). The current AASHTO specifications (2010) defined
the dynamic effects duo to moving vehicles by impact factors attributed
to hammering effect and road roughness. The road roughness could be
assumed as a zero-mean stationary Gaussian random process and it could
be generated through an inverse Fourier transformation (Wang and Huang
1992). It was later shown that the foundation settlement and other envi-
ronmental factors would affect the bridge-vehicle interaction due to the
shape of deck [19].
Without considering wind dynamic impacts on the vehicles, the
dynamic wheel load will be underestimated by about 6 to 11 percent [20].
Considering the wind excitations, the vehicle-bridge interaction is more
prominent and complex, and large body of research has been carried out
on vehicle-wind-bridge interaction. A comprehensive framework regard-
ing vehicle-wind-bridge dynamic analysis of coupled 3-D was first pre-
sented by Cai and Chen, [21]. In their framework, a series of vehicles
consisting of different numbers and different types of vehicles driving on
bridges under hurricane-induced strong winds was included. Based on that
framework an equivalent dynamic wheel load (EDWL) approach and the
CA traffic simulation were adopted to analyze the dynamic performance
of long-span bridges under combined loads of stochastic traffic and wind
excitations [22]. A reasonable framework to replicate probabilistic traf-
fic flow, characterize the dynamic interaction and assess the structural
performance under strong wind and heavy traffic was presented to study
the probabilistic dynamic behavior of long-span bridges under extreme
events [23].
6  •   BRIDGE RELIABILITY AND SERVICEABILITY

To the best of the authors’ knowledge, very limited and scarce research
work has been reported in the literature on the probabilistic dynamic anal-
ysis of long-span bridges subjected to combined stochastic traffic and
wind excitations, which is extremely important for the safety of long-span
bridges. Furthermore, most dynamic analyses reported in this area have
focused on time domain analysis, because of the nature of time-varying
differential equations in the interaction system, while very limited devel-
opments have been done in the frequency domain. However, incorporat-
ing the random vibration in the aforementioned coupled system, which
requires and necessitates the use of spectral analysis, is more important
and results in more valuable information in the frequency domain.

1.2.2 FIRST-PASSAGE RELIABILITY THEORY AND

APPLICATIONS TO ENGINEERING STRUCTURES

First-passage principle and fatigue damage principle are two main assump-
tions for structural dynamic reliability evaluation. The first-passage reli-
ability can be described as estimating the probability that a random process
exceeds a prescribed threshold during an interval of time. Knowledge of
this probability is essential for estimating the reliability of a structural
dynamic system whose response is a stochastic process. The fatigue dam-
age criterion should be adopted for the accumulated fatigue damage at
the critical regions of a structure such as connections and joints. This
topic will be discussed in more details in the next section. The classical
first-passage criterion was originally proposed by Rice [24] based on the
random vibration and extreme value distribution theory. Mathematical
formulations for the number of times that the structural responses cross
the limits were also established by Rice. A well-known crossing process
is the Rayleigh distribution as the extreme value distribution of a nar-
row-banded Gaussian stochastic process. The Poisson’s assumption and
the Vanmarcke’s assumption were widely used for general stochastic pro-
cesses in engineering structure [25]. However, these assumptions come
more from intuition or empirical approaches, rather than from theoreti-
cal basis. For this reason, several researchers developed approaches for
the improvement of the assumptions that were based on empirical work.
A joint first-passage probability method was proposed to evaluate the
reliability of linear engineering systems composed of several interdepen-
dent components by Song and Der [26]. For nonlinear dynamic systems,
Cai and Lin [27], investigated the first passage problem using stochastic
averaging. Noori, et al. [28] introduced the first-passage study of a highly
 Introduction   •  7

nonlinear hysteretic system using quasi-conservative stochastic averag-

ing. Bucher and Macke [29], introduced the solutions to the first-passage
problem by importance sampling. A probability density evolution method
which was capable of capturing the instantaneous PDF and its evolution
of the responses was developed by Chen and Li [30].
Applications of first-passage reliability to engineering structures are
very interesting since safety assessment and design can be put forward
to guarantee the structural safety. Park and Ang [31] assessed the prob-
ability of damage for a reinforced concrete structure under the seismic
load. Zhang et al. [32] adopted a pseudo-excitation method and a precise
integration method to compute the non-stationary random response of 3-D
train-bridge systems subjects to lateral horizontal earthquakes. Significant
progress in structural reliability evaluation has been achieved in the last
decades utilizing nonlinear stochastic structural dynamics [33]. Xiang
and Zhao et al. [34] evaluated bridge structural reliability considering the
vehicle-bridge dynamic interaction, where the reliability evaluation method
is the traditional static reliability method which ignores the dynamic ran-
dom vibration of the bridge caused by the road surface roughness and the
bridge-vehicle interaction. Dynamic reliability approach should be consid-
ered to estimate the reliability since a dynamic reliability based analysis
incorporates the random vibration theory and the effect of bridge-vehicle
interactions caused by the random traffic flow are considered.
In bridge engineering, the first-passage reliability method is suit-
able for the safety assessment of existing bridges under vehicle loads.
However, limited work has been done on this issue. Therefore, the pur-
pose of the preset book is to meet this need and fill the gap in this respect.
A framework of evaluating first-passage reliability of long-span suspen-
sion bridges under stochastic traffic flow and wind loads will be presented.

1.2.3 FATIGUE RELIABILITY OF STEEL BOX-GIRDER

BRIDGES

Long-span steel bridges are vulnerable to repeated loads caused by traffic,

wind, gust, and the changing environment. These combined effects can
lead to complex modes of fatigue failure. Fatigue is one of the main forms
of deterioration for structures and can be a typical failure mode due to an
accumulation of damage. Numerous research work has been carried out on
fatigue failure modes, evolution mechanism, and fatigue life assessment
based on numerical simulation, laboratory experiment, and site test. The
two main deterministic analysis methods in this regard are: stress-based
8  •   BRIDGE RELIABILITY AND SERVICEABILITY

approach (S-N curve approach) and linear elastic fracture mechanics

(LEFM) approaches which are applicable to different analyses strategies
[35, 36]. The size effect is considered in the LEFM and is suitable for
structures with initial defects, however, the S-N curve method involves an
abstract model of the fatigue damage, and does not include the analysis
of the crack-tip stress field. Yes, it is widely used in various applications
[37]. Both deterministic and probabilistic procedures have been applied
to estimate the fatigue damage of structures. Since the major load causing
the fatigue of steel bridges is vehicle load, which is a strong stochastic
process, fatigue reliability evaluation has resulted in increasing research
studies in this area. The basic approach utilized in fatigue research stud-
ies of steel bridges under traffic load is to obtain the stress ranges which
are widely used for fatigue analysis of steel bridges. Numerical analy-
sis based on finite element method (FEM) and the condition assessment
method based on the long-term structural health monitoring (SHM) are
commonly used to analyze the structure and obtain the fatigue stress. Chen
[38] assessed fatigue reliability of Tingma Bridge under multi-loadings
based on an SHM system. Zhang et al. 2012 [39, 40, 41], presented a com-
prehensive framework for fatigue reliability estimation of bridges under
combined dynamic loads from vehicles and wind. Guo et al. [42] proposed
an advanced traffic load model taking into account the uncertainties asso-
ciated with the number of axles, axle spacing, and axle weights. Wang et
al. [43] combined the SHM and FEM to assess fatigue reliability of girder
components for long-span cable-stayed bridges. Deng et al. [44] developed
a long-term monitoring data based fatigue reliability assessment method
and applied that to welded details in the steel box girder of Runyang
Yangtze river bridge. Kwon and Frangopol [45] integrated fatigue reliabil-
ity model, crack growth model, and probability of the detection model for
fatigue assessment and management of existing bridges.

REFERENCES

[1] Xie, H., Y. Wang, H. Wu, and Z. Li. 2014. “Condition Assessment of
Existing RC Highway Bridges in China Based on SIE2011.” Journal
of Bridge Engineering 19, no. 12, 04014053. doi:10.1061/(ASCE)
BE.1943-5592.0000633
[2] Frangopol, D.M., and M. Soliman. 2015. “Life-cycle of Structural
Systems: Recent Achievements and Future Directions[J].” Structure and
Infrastructure Engineering 12, no. 1, 1–20. https://doi.org/10.1080/15732
479.2014.999794
 Introduction   •  9

[3] Han, W., J. Wu, C.S. Cai, and S. Chen. 2014. “Characteristics and Dynamic
Impact of Overloaded Extra Heavy Trucks on Typical Highway Bridges[J].”
Journal of Bridge Engineering 20, no. 2, p. 05014011.
[4] Wardhana, K., and F.C. Hadipriono. 2003. “Analysis of Recent Bridge
Failures in the United States[J].” Journal of Performance of Constructed
Facilities 17, no. 3, pp. 144–50.
[5] Deng, L., W. Wang, and Y. Yu. 2015. “State-of-the-Art Review on the
Causes and Mechanisms of Bridge Collapse[J].” Journal of Performance of
Constructed Facilities 30, no. 2, p. 04015005.
[6] Xing, C., H. Wang, A. Li, and Y. Xu. 2013. “Study on Wind-induced
Vibration Control of a Long-span Cable-stayed Bridge Using TMD-type
Counterweight [J].” Journal of Bridge Engineering 19, no. 1, pp. 141–48.
[7] Brownjohn, J.M.W. 1997. “Vibration Characteristics of a Suspension
Footbridge[J].” Journal of Sound and Vibration 202, no. 1, pp. 29–46.
[8] Cai, C.S., J. Hu, S. Chen, Y. Han, W. Zhang, and X. Kong. 2015. “A Coupled
Wind-vehicle-bridge System and its Applications: A Review[J].” Wind and
Structures 20, no. 2, pp. 117–42.
[9] Li, Y., P. Hu, C.S. Cai, M. Zhang, and S. Qiang. 2012. “Wind Tunnel Study
of a Sudden Change of Train Wind Loads Due to the Wind Shielding Effects
of Bridge Towers and Passing Trains[J].” Journal of Engineering Mechanics
139, no. 9, pp. 1249–59.
[10] Aktan, A.E., D.N. Farhey, D.L. Brown, V. Dalal, A.J. Helmicki, V.J. Hunt,
and S.J. Shelley. 1996. “Condition Assessment for Bridge Management[J].”
Journal of Infrastructure Systems 2, no. 3, pp. 108–17.
[11] Deng, L., Y. Yu, Q. Zou, and C. Cai. 2015. “State-of-the-Art Review of
Dynamic Impact Factors of Highway Bridges.” Journal of Bridge Engineering
20, no. 5, 04014080. doi:10.1061/(ASCE)BE.1943-5592.0000672
[12] Chen, S.R., and J. Wu. 2009. “Dynamic Performance Simulation of Long-
span Bridge Under Combined Loads of Stochastic Traffic and Wind[J].”
Journal of Bridge Engineering 15, no. 3, pp. 219–30.
[13] Hao, S. 2009. “I-35W Bridge Collapse[J].” Journal of Bridge Engineering
15, no. 5, pp. 608–14.
[14] Morales-Nápoles, O., and R.D.J.M. Steenbergen. 2015. “Large-Scale Hybrid
Bayesian Network for Traffic Load Modeling from Weigh-in-Motion System
Data[J].” Journal of Bridge Engineering 20, no. 1, pp. 591–99.
[15] Schadschneider, A. 2002. “Traffic Flow: A Statistical Physics Point of
View[J].” Physica A: Statistical Mechanics and its Applications 313, no. 1,
pp. 153–87.
[16] Enright, B., and E.J. O’Brien. 2013. “Monte Carlo Simulation of Extreme
Traffic Loading on Short and Medium Span Bridges[J].” Structure and
Infrastructure Engineering 9, no. 12, pp. 1267–82.
[17] Chen, S.R., and J. Wu. 2011. “Modeling Stochastic Live Load for Long-
span Bridge Based on Microscopic Traffic Flow Simulation[J].” Computers
& Structures 89, no. 9, pp. 813–24.
10   •   BRIDGE RELIABILITY AND SERVICEABILITY

[18] Guo, W.H., and Y.L. Xu. 2001. “Fully Computerized Approach to Study
Cable-stayed Bridge–Vehicle Interaction[J].” Journal of Sound and Vibration
248, no. 4, pp. 745–61.
[19] Ahmari, S., M. Yang, and H. Zhong. 2015. “Dynamic Interaction Between
Vehicle and Bridge Deck Subjected to Support Settlement[J].” Engineering
Structures 84, pp. 172–83.
[20] Chen, S.R., and J. Wu. 2010. “Dynamic Performance Simulation of Long-
span Bridge Under Combined Loads of Stochastic Traffic and Wind[J].”
Journal of Bridge Engineering 15, no. 3, pp. 219–30.
[21] Cai, C.S., and S.R. Chen. 2004. “Framework of Vehicle–bridge–wind
Dynamic Analysis[J].” Journal of Wind Engineering and Industrial
Aerodynamics 92, no. 7, pp. 579–607.
[22] Chen, S.R., and C.S. Cai. 2007. “Equivalent Wheel Load Approach for
Slender Cable-stayed Bridge Fatigue Assessment Under Traffic and Wind:
Feasibility Study[J].” Journal of Bridge Engineering 12, no. 6, pp. 755–64.
[23] Wu, J., and S.R. Chen. 2011. “Probabilistic Dynamic Behavior of a Long-
span Bridge Under Extreme Events[J].” Engineering Structures 33, no. 5,
pp. 1657–65.
[24] Rice, S.O. 1945. “Mathematical Analysis of Random Noise[J].” Bell System
Technical Journal 24, no. 1, pp. 46–156.
[25] Vanmarcke, E.H. 1975. “On the Distribution of the First-Passage Time for
Normal Stationary Random Process[J].” Journal of Application Mechanics
42, pp. 215–20.
[26] Song, J., and A. Der Kiureghian. 2006. “Joint First-Passage Probability and
Reliability of Systems Under Stochastic Excitation.” Journal of Engineering
Mechanics 132, no. 1, pp. 65–77.
[27] Cai, G.Q., and Y.K. Lin. 1994. “On Statistics of First-passage Failure[J].”
Journal of Applied Mechanics 61, no. 1, pp. 93–99.
[28] Noori, M., M. Dimentberg, Z. Hou, R. Christodoulidou, and A. Alexandrou.
1995. “First-passage Study and Stationary Response Analysis of a
BWB Hysteresis Model Using Quasi-conservative Stochastic Averaging
Method[J].” Probabilistic Engineering Mechanics 10, no. 3, pp. 161–70.
[29] Macke, M., and C. Bucher. 2003. “Importance Sampling for Randomly
Excited Dynamical Systems[J].” Journal of Sound and Vibration 268, no.
2, pp. 269–90.
[30] Chen, J.B., and J. Li. 2007. “The Extreme Value Distribution and Dynamic
Reliability Analysis of Nonlinear Structures with Uncertain Parameters[J].”
Structural Safety 29, no. 2, pp. 77–93.
[31] Park, Y.J., and A.H.S. Ang. 1985. “Mechanistic Seismic Damage Model
for Reinforced Concrete[J].” Journal of Structural Engineering 111, no. 4,
pp. 722–39.
[32] Zhang, Z.C., J.H. Lin, Y.H. Zhang, Y. Zhao, W.P. Howson, and F.W. Williams.
2010. “Non-stationary Random Vibration Analysis for Train–bridge Systems
Subjected to Horizontal Earthquakes[J].” Engineering Structures 32, no. 11,
pp. 3571–82.
 Introduction   •   11

[33] Balafas, K., and A.S. Kiremidjian. 2015. “Reliability Assessment of the
Rotation Algorithm for Earthquake Damage Estimation[J].” Structure and
Infrastructure Engineering 11, no. 1, pp. 51–62.
[34] Xiang, T., R. Zhao, and T. Xu. 2007. “Reliability Evaluation of Vehicle–
bridge Dynamic Interaction[J].” Journal of structural Engineering 133,
no. 8, pp. 1092–99.
[35] Guo, T., and Y.W. Chen. 2013. “Fatigue Reliability Analysis of Steel
Bridge Details Based on Field-monitored Data and Linear Elastic Fracture
Mechanics[J].” Structure and Infrastructure Engineering 9, no. 5,
pp. 496–505.
[36] Chen, N.Z., G. Wang, and C. Guedes Soares. 2011. “Palmgren–Miner’s
Rule and Fracture Mechanics-Based Inspection Planning[J].” Engineering
Fracture Mechanics 78, no. 18, pp. 3166–82.
[37] Sim, H.B., and C.M. Uang. 2012. “Stress Analyses and Parametric Study on
Full-scale Fatigue Tests of Rib-to-deck Welded Joints in Steel Orthotropic
Decks[J].” Journal of Bridge Engineering 17, no. 5, pp. 765–73.
[38] Chen, Z.W., Y.L. Xu, and X.M. Wang. 2011. “SHMS-based Fatigue
Reliability Analysis of Multiloading Suspension Bridges[J].” Journal of
Structural Engineering 138, no. 3, pp. 299–307.
[39] Zhang, W., C.S. Cai, and F. Pan. 2012. “Fatigue Reliability Assessment
for Long-span Bridges Under Combined Dynamic Loads from Winds and
Vehicles[J].” Journal of Bridge Engineering 18, no. 8, pp. 735–47.
[40] Zhang, W., C.S. Cai, F. Pan, and Y. Zhang. 2014. “Fatigue Life Estimation
of Existing Bridges Under Vehicle and Non-stationary Hurricane Wind[J].”
Journal of Wind Engineering and Industrial Aerodynamics 133, pp. 135–45.
[41] Zhang, W., C.S. Cai, and F. Pan. 2013. “Nonlinear Fatigue Damage
Assessment of Existing Bridges Considering Progressively Deteriorated
Road Conditions[J].” Engineering Structures 56, pp. 1922–32.
[42] Guo, T., D.M. Frangopol, and Y. Chen. 2012. “Fatigue Reliability Assessment
of Steel Bridge Details Integrating Weigh-in-motion Data and Probabilistic
Finite Element Analysis[J].” Computers & Structures 112, pp. 245–57.
[43] Wang, Y., Z.X. Li, and A.Q. Li. 2010. “Combined Use of SHMS and Finite
Element Strain Data for Assessing the Fatigue Reliability Index of Girder
Components in Long-span Cable-stayed Bridge[J].” Theoretical and Applied
Fracture Mechanics 54, no. 2, pp. 127–36.
[44] Deng, Y., Y.L. Ding, A.Q. Li, and G. Zhou. 2011. “Fatigue Reliability
Assessment for Bridge Welded Details Using Long-term Monitoring
Data[J].” Science China Technological Sciences 54, no. 12, pp. 3371–81.
[45] Kwon, K., and D.M. Frangopol. 2011. “Bridge Fatigue Assessment and
Management Using Reliability-based Crack Growth and Probability of
Detection Models[J].” Probabilistic Engineering Mechanics 26, no. 3,
pp. 471–80.
 CHAPTER 2

Fatigue Reliability
Assessment of Welded
Steel Bridge Decks Under
Stochastic Truck Loads

2.1 INTRODUCTION

Steel girders are commonly used in long-span bridges due to their

load-carrying capacity in proportion to their weight. In general, a steel
girder contains lots of complex welded joints located at the connections
among deck plates, longitudinal ribs, and cross girders. Orthotropic bridge
decks, at the top of the steel girder, including closed ribs and open ribs
[1], are fatigue critical components since they directly suffer from cyclic
traffic loads. An increasing number of engineering investigations indicate
that the welded joints in the orthotropic deck of steel bridge girders are
vulnerable to fatigue damage [2, 3]. In addition, fatigue cracks normally
exist in the critical locations of the welded joints and they grow rapidly. As
a result, the fatigue problem may lead to the decrease of the load-­carrying
capability or even collapse of the bridge. Numerous research studies have
been carried out on fatigue failure modes, evolution mechanism, and
fatigue life prediction based on numerical simulation, laboratory experi-
ment, and site test [4, 5]. Furthermore, inspection and maintenance actions
have been performed to monitor the fatigue damage in structural life-cycle
[6]. Since the traffic load is random in nature, the relative structural fatigue
stress analysis inevitably leads to a probability issue. In addition, uncer-
tainties arising from the structural parameters and the accumulated dam-
age models directly contribute to the randomness of the fatigue damage
[7]. Hence, a probabilistic approach will provide a more reliable evalua-
tion for fatigue analysis of welded steel bridge decks in comparison with
14   •   BRIDGE RELIABILITY AND SERVICEABILITY

deterministic analysis. Therefore, as elaborated earlier, it is appropriate

and necessary to conduct a reliability-based fatigue assessment for welded
steel bridge decks with the consideration of stochastic truck loads.
The fatigue stress spectrum under cyclic truck loads is a kernel task
for fatigue analysis of steel bridge decks. There are mainly two approaches
to obtain the fatigue stress spectrum: (a) structural health monitoring
(SHM), and (b) finite element analysis (FEA). The SHM approach is a
better choice for newly built bridges instrumented with fatigue sensors.
Research on SHM-based reliability assessment has been conducted mostly
for long-span bridges with long-term monitoring data [8, 3]. However,
with the development of advanced computational methods and vehicle
weight measurement technology, the FEA approach exhibits its superior-
ity in efficiency, compared to the costly SHM approach. Wang et al. [9]
presented a framework for combining the SHM and FEA to assess fatigue
reliability of bridges and demonstrated the feasibility of their scheme by
applying that to a long-span suspension bridge. Guo and Chen [10] demon-
strated that the fatigue stress prediction integrating monitored traffic data
and probabilistic numerical simulation approach were in agreement with
those obtained from the monitored data. Ni et al. [11] developed a global
and local FEA model for computing the stress concentration factors of
fatigue critical locations of Tsing Ma Bridge. D’Angelo and Nussbaumer
[12] developed a framework to evaluate the failure probability of a com-
posite steel-concrete bridge in Switzerland by considering both the con-
stant amplitude fatigue limit and the critical fatigue damage accumulation.
Even though several computational frameworks were developed for the
probabilistic modeling of fatigue damage, the time-consuming problem
due to huge repeated simulations in the FEA is still a challenge for achiev-
ing viable numerical solutions. This indicates that much attention should
be focused on improving the computing efficiency, especially for utilizing
the monitored site-specific traffic data.
The critical step for applying the FEA approach to structural fatigue
stress analysis is how to simulate both the structural geometries and the
truck loads more realistically. Truck loading parameters, vehicle config-
urations, axle weights, driving lanes, and driving speeds have varying
degrees of effect on the effective fatigue stress ranges. Numerous research
investigations have been reported on the fatigue truck load model. Some
of those are summarized in the following. Laman and Nowark [13] devel-
oped two fatigue truck load models including a three-axle truck and a
four-axle truck based on the monitored traffic data of five steel bridges,
which were verified and compared with measured results. Subsequently,
several fatigue truck load models were presented in the various design
 FATIGUE RELIABILITY OF STEEL BRIDGES   •   15

codes, such as AASHTO [15] and Eurocode 1 [14]. Recently, a multi-load-

ing model, with the consideration of railway, highway, and wind loads,
for fatigue reliability assessment of Tsing Ma Bridge, was developed by
Chen et al [7]. However, Chen’s vehicle model simplified the vehicles as a
concentrated force, where the multi-axle effect and the dynamic response
were ignored. In response to this limitation, Zhang et al. [16, 17] pre-
sented a comprehensive framework for fatigue reliability estimation of
bridges subjected to combined dynamic loads from vehicles and wind
loads, where the vehicle model was more realistic and the vehicle–bridge
interactions were taken into account. Influence of dynamic impacts due to
deteriorated road surface conditions and vehicle suspension systems was
investigated by MacDougall, et al. [18], Wang, et al. [19], and Zhang and
Cai [20]. Guo et al. [21, 22] developed an advanced fatigue vehicle model
utilizing weigh-in-motion (WIM) measurements to carry out the probabi-
listic analysis. However, the axle weight in Guo’s stochastic vehicle model
followed a standard Gaussian distribution, while the probability density
function (PDF) of actual axle weight mostly exhibited a multi-peaks fea-
ture. This phenomenon can be explained by the general case of no-load
or full-load of a truck in transportation engineering. Given the need for a
more realistic model for the stochastic traffic model, which can incorpo-
rate an appropriate PDF of the axle weight, Xia et al. [23] implied mixture
distribution models with expectation maximization (EM) algorithm and
the Akaike information criterion (ACI) to establish PDFs of fatigue stress
ranges. Chen et al. [24] proposed a 5-vehicle fatigue load spectrum model
to fill in the gap of the fatigue load model in China. However, despite the
fact that a few fatigue truck load models have been investigated and have
been adopted, utilizing several fatigue design specifications, very limited
research efforts have been carried out on probabilistic modeling of a gen-
eral fatigue truck load model, which apparently has a significant influence
on fatigue reliability of steel bridges. Therefore, research on the impact of
the probabilistic model of site-specific truck loads on fatigue reliability of
steel bridge decks is an area that needs to be further investigated.
Therefore, this book aims at developing a stochastic fatigue truck
load model and applying it to fatigue reliability assessment of welded
steel bridge decks. As the first task, the stochastic fatigue truck load model
was established based on long-term WIM measurements. Parameters in
the truck load model, including vehicle configuration, axle weight, and
driving lane, were used for the parametric analysis of structural fatigue
reliability. The second task, presented in this book, is developing a meth-
odology to obtain the fatigue stress spectrum of steel bridges under the
stochastic fatigue truck load model. In the related case study presented
16   •   BRIDGE RELIABILITY AND SERVICEABILITY

herein, a prototype steel box-girder bridge was introduced to illustrate the

feasibility of the proposed framework. Parametric studies demonstrated
the accuracy and the efficiency of the framework. Influence of an increase
in the traffic volume and vehicle weight on the fatigue reliability of the
bridge was investigated. The ultimate goal of this study was to apply
the stochastic fatigue truck model for probabilistic modeling of fatigue
damage and the reliability assessment of welded steel bridge decks.

2.2 STOCHASTIC FATIGUE TRUCK LOAD MODEL

As discussed in the previous section, a fatigue truck load model with

meaningful statistical characteristics is the foundation for the probabilis-
tic modeling of fatigue stress ranges. The stochastic fatigue traffic model
can also be utilized for subsequent reliability assessment of steel bridges
with available traffic data. Herein, we will present how the stochastic
fatigue truck load model was established based on site-specific WIM mea-
surements. First, an overview of the available, or presently used, design
fatigue truck load models is presented.

2.2.1 PRESENTLY USED FATIGUE TRUCK LOAD MODELS

A fatigue truck load is typically used to represent truck traffic at a specific

site with multiple vehicle weights and configurations. The fatigue truck
load is usually evaluated by the equivalent fatigue damage accumulation
criterion, where the fatigue damage accumulation caused by the passages of
the fatigue truck load is equivalent to that of the actual truck load spectrum.
Since the actual truck loads are strongly site-specific and component-spe-
cific, WIM measurements are necessary for establishing an accurate and
site-specific load spectrum. The WIM system mostly consisting of elec-
tric sensors can be used to measure gross vehicle weights, axle weights,
axle spacing, and vehicle speeds in actual traffic flow. As elaborated in
the introduction, research efforts on fatigue truck load utilizing WIM data
have reached great achievements and have been adopted in many design
codes, such as AASHTO [15] and Eurocode 1 [14] shown in Figure 2.1.
As can be seen in Figure 2.1, however, all these fatigue load mod-
els exhibit various deterministic configurations and parameters. Even
though these load models were evaluated based on numerous field inves-
tigations and were provided with generality in the corresponding regions,
they cannot reflect the parametric statistics of actual trucks, which are
 FATIGUE RELIABILITY OF STEEL BRIDGES   •   17

26.8kN 106.6kN 106.6kN

6.0 m

4.27m 9.14m
(a)

1.2m 1.2m

1.8m

0.4m
0.4m

(b)

Figure 2.1.  Fatigue truck load models in the design specifications:

(a) AASHTO with GVW of 240 kN; (b) Eurocode 1 with GVW of
480 kN.

quite important for probabilistic modeling of structural fatigue damages.

Furthermore, these truck models may be outdated and should be updated
with newly measured traffic data. In order to overcome these shortcomings,
the study reported in this book presents a general stochastic fatigue truck
load model and applies that to structural fatigue reliability assessment.

2.2.2 PROPOSED STOCHASTIC FATIGUE TRUCK LOAD

MODEL

A stochastic traffic flow typically includes parameters associated with

vehicle types, vehicle speeds, vehicle spacing, driving lanes, and vehicle
weights. The wide use of both WIM measurements and traffic simula-
tion technologies (e.g., Monte Carlo simulation and cellular automaton)
has supported the development of stochastic traffic modeling [7]. The
stochastic traffic flow model that is widely used in transportation man-
agement and vehicle–bridge interaction analysis [25] is the origin of the
present stochastic fatigue truck load model. For the purpose of fatigue
stress analysis of steel bridge decks, the proposed stochastic traffic flow
model was developed by considering the parameters with higher contri-
butions to structural fatigue stresses. Initially, the vehicle spacing can be
ignored, because the impact of simultaneous truck loads on hot spot stress
18   •   BRIDGE RELIABILITY AND SERVICEABILITY

of welded joints is small. On the other hand, a large number of additional

finite element runs is needed for the consideration of vehicle gap in the
same direction, since the vehicle gap for highway bridges is extremely
larger than the length of the influence line. In addition, the vehicle speed
can be considered as constant. Finally, the gross vehicle weight (GVW)
should be categorized based on the axle weight, since every passing axle
produces different fatigue stress cycles. Note that the vehicles with GVW
less than 30 kN should be ignored, since these vehicles make negligible
contributions to the fatigue damage. As elaborated previously, the sto-
chastic fatigue truck load model, described in the research tasks and uti-
lized in this book, contains three variable parameters: vehicle types, axle
weights, and driving lanes. The modeling procedure can be referred to
the traditional stochastic traffic flow modeling utilizing the Monte Carlo
simulation [26].
A bridge WIM system located in Sichuan Province of China is
selected herein as a prototype. Data corresponding to over 10 million vehi-
cles, collected in five years, were used for vehicle parametric statistical
analysis. The trucks are classified into six types as shown in Table 2.1.
The front axles of all vehicles have single-tire in each side, while the
remaining axels have double-tires on each side, except for the light trucks
with single tires each side on two axles. Taking vehicle type 6, V6, as an
example, both the monitored and approximated probability densities of
the 4th axle weight, AW64,are shown in Figure 2.2. In Figure 2.2, Gaussian
mixture models (GMMs) were used to approximate the probability den-
sity functions (PDFs). It was found that the actual probability density of
axle weight has two peaks corresponding to no-load and full-load states.
This demonstrated the necessity of utilizing the GMM approach. With the
traffic statistics, the simulated fatigue truck loads on both fast lane and the

Table 2.1.  Vehicle classifications

Total Occupancy Occupancy
Vehicle occupancy rate in slow rate in fast
type Description rate (%) line (%) line (%)
V1 Light trucks 34.64 36.64 63.36
V2 2-axle truck 26.12 84.58 15.42
V3 3-axle truck 8.58 91.08 8.92
V4 4-axle trucks 10.24 96.42 3.58
V5 5-axle truck 4.93 92.60 7.40
V6 6-axle truck 15.49 98.08 1.92
 FATIGUE RELIABILITY OF STEEL BRIDGES   •   19

0.02
Measured data
GMM
0.015
Probability density

0.01

0.005

0
0 500 1000 1500
GVW(kN)
(a)

0.12
Measured data
GMM
Probability density

0.08

0.04

0
100 200 300
AW64(kN)

(b)

Figure 2.2.  PDFs of: (a) GVW of V6; (b) AW64.

slow lane are shown in Figure 2.3. The GVW, instead of the axle weight,
was used in Figure 2.3 for more clear and thorough presentation of the
results.
As shown in Figure 2.3, each dot refers to a specific truck with dif-
ferent labeling marks, x-axis shows the arrival time, and y-axis shows the
individual GVW. It is observed that each truck is different due to its spe-
cific characteristics. However, all trucks follow a relatively similar corre-
sponding probability distribution. In addition, the distributions of vehicle
configuration in both fast lane and slow lane are obviously different,
20   •   BRIDGE RELIABILITY AND SERVICEABILITY

1200
V1
V2
900 V3
V4
V5
GVW (kN)

600 V6

300

0
0 0.5 1 1.5 2 2.5
Arrival time (h)
(a)
1200
V1
V2
V3
900
V4
V5
GVW (kN)

V6
600

300

0
0 0.5 1 1.5 2 2.5
Arrival time (h)
(b)
Figure 2.3.  Simulated stochastic fatigue truck loads in: (a) slow lane;
(b) fast lane.

because the heavy loaded trucks have a higher possibility appearing in

the slow lane, while this is the opposite for light trucks. These statistical
characteristics in the truck model directly affect the PDFs of structural
fatigue stress ranges, and are discussed in the case study. The stochastic
fatigue truck load model that contains the aforementioned statistical char-
acteristics of trucks provides a basis for the following probabilistic mod-
eling of the fatigue damage accumulation in welded steel bridge decks.
 FATIGUE RELIABILITY OF STEEL BRIDGES   •   21

2.3 COMPUTATIONAL FRAMEWORK OF
PROBABILISTIC MODELING

2.3.1 PROPOSED COMPUTATIONAL FRAMEWORK

For the purpose of applying the stochastic traffic truck load model to prob-
abilistic modeling of fatigue stress in steel bridge decks, an efficient com-
putational framework should be developed as a connection between the
stochastic load model and the structural effect model. The innovation of
the work presented in this book is the application of stochastic truck model,
presented earlier, instead of the commonly used and the typical truck
model. However, this will be a computationally time-consuming problem
for time history analysis taking into account each truck passage, because
a large number of finite element analysis runs are required to simulate the
fatigue behavior of welded details accurately. Therefore, improving the
computational efficiency, by not ignoring the importance of the require-
ments for the computational accuracy, is a critical step for the application
of the stochastic fatigue truck load in the study presented in this book.
In order to address the exhaustive computation time problem, a
machine learning algorithm was employed by integrating a uniform
design (UD) and a support vector regression (SVR) approach. This pro-
posed learning machine is utilized to approximate the response surface
between the vehicle axles and the equivalent stress ranges for a certain
vehicle configuration. The flowchart summarizing the entire procedure
is depicted in Figure 2.4. There are two main procedures, as illustrated

Stochastic fatigue truck load model

Small-scale sampling: Large-scale sampling:

generate small amounts of truck load samples generate a large number of truck load
for the 6 types of vehicles, respectively. samples to simulate the daily traffic flow.

Hot spot stress range simulation: Estimating Δσre and Ned in the large sample:
compute the stress time history based on FEA estimate the Δσre and corresponding number of
and extract the equivalent stress ranges, daily stress cycles, Ned, for the individual truck
Δσre, of the welded connections under in the large sample utilizing the approximated
individual vehicle load in the small samples. learning machine.

Machine learning: Probabilistic modeling:

approximate the response functions between approximate the PDFs of the Δσre and Ned
the Δσre and vehicle axle weights for each utilizing GMM based on the large sample in
type of vehicles. the stochastic truck load model.
Deterministic simulation Probabilistic modeling

Figure 2.4.  Proposed computational framework.

22  •   BRIDGE RELIABILITY AND SERVICEABILITY

in the flowchart, which need to be elaborated: deterministic simulation

and probabilistic modeling. The purpose of the deterministic simulation
is to establish a regression model to approximate the response surface for
the fatigue truck load and the corresponding fatigue stress ranges using a
small-sample of trucks; while the purpose of the probabilistic modeling is
to establish the probabilistic model for the fatigue damage of the welded
joints under stochastic truck loads. The interaction between the two key
procedures is the approximated response surface model established by the
SVR approach. The details of these procedures are elaborated as follows.

2.3.2 DETERMINISTIC FINITE ELEMENT-BASED

SIMULATION

Fatigue stress analysis of structural welded joints using finite element

method is mostly associated with the structural hot spot stress approach,
rather than the traditional nominal stress approach. Therefore, an accu-
rate finite element model is required for fatigue analysis of these complex
welded details. Besides the local accurate simulation of connections, the
stiffness of the girders needs to be considered. When an individual truck is
passing over the steel girders, several fatigue stress blocks are generated,
especially at the welded locations, such as rib-to-deck and butt joint of
U-rib. Because the influence line of the welded joints is shorter than the
axle spacing of the trucks, which will be demonstrated in the case study,
each axle will generate a peak value of the fatigue stress. Therefore, each
truck passage produces several fatigue stress ranges depending on the con-
figuration of the truck axles. The response surface between the axle weight
and the fatigue stress ranges is implicit and can be approximated via a
machine learning algorithm.
With the previous assumption, the deterministic analysis in the frame-
work involves three sub-steps: uniform sampling, finite element analysis,
and response surface approximation. Note that since each vehicle config-
uration varies, the corresponding response functions should be considered
separately. As the range of vehicle weights are determined, several truck
samples are generated from the stochastic fatigue truck load model uti-
lizing the UD approach [27], which can provide samples that are orthog-
onal and uniformly scattered in the design domain. As the truck samples
are determined, a finite element model of steel girders, including welded
details, is constructed to simulate the fatigue behavior of the welded joints
under the truck load. The stress–time histories are then obtained from the
transient analysis of finite element model. Subsequently, the stress ranges
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  23

and the corresponding number of cycles can be calculated by the rain flow
counting method. The multi-amplitude stress cycles can be equivalent to a
constant-amplitude stress cycle according to the fatigue damage accumu-
lation rule that is illustrated in the next section.
The most essential aspect for the development of the deterministic
analysis in the research presented in this book is the approximating func-
tion linking the vehicle weight and the equivalent fatigue stress ranges.
A simple illustration of the SVR formulations is presented in the follow-
ing. In general, the purpose of SVR is to find a relationship between the
input space and the output space with the given training dataset, {(xi, yi), i
= 1,…,N}, where the xi and yi are the ith input data and output data, respec-
tively and N is the number of total data. The general form of the SVR is
written as [28]

l
f ( x) = ∑ ai È ( x, xi ) + b (2.1)
i =1

where f is the approximating function, È ( x, xi ) is a kernel function, such

as Gaussian, polynomial, and sigmoid kernels; ai is the ith weight of the
kernel function; and b and l are the bias and the total number of the kernel
functions, respectively. These parameters can be obtained by the structural
risk minimizing principle and the Lagrange multiplier optimal program-
ming method [29, 31]. Several advanced SVR method have been devel-
oped in various research fields, such as the Least Square-SVR [32] and
the Wavelet-SVR [33]. However, the application of the SVR in structural
fatigue stress prediction is relatively new.
Considering that the purpose of the SVR is to simulate the relationship
between the axle weights and the fatigue stress ranges, a Gaussian kernel
function is adopted in the work presented in this book. The input training
data are the axle weight of the samples with uniform distribution, and the
output training data are the equivalent fatigue stress range obtained from
the finite element analysis. On this basis, six types of vehicle configuration
corresponds six SVR models in this study. With the representative SVR
model, the fatigue stress estimation from the large-sample of simulated
daily trucks can be obtained by a mathematical analysis instead of the
time-consuming finite element analysis.

2.3.3 PROBABILISTIC MODELING

As the fatigue stress ranges are calculated in the deterministic finite ele-
ment analysis, the next step is the probabilistic modeling of these fatigue
24  •   BRIDGE RELIABILITY AND SERVICEABILITY

stress ranges. The probabilistic modeling results can provide the neces-
sary statistics for the reliability analysis. It can be observed from the sta-
tistics of traffic data that the parameters in the truck load model follow
various types of probabilistic distributions. As a result, the probabilistic
density of the fatigue stress range may not follow a certain ­single distribu-
tion. Regarding the random variables in the stochastic fatigue truck load,
the probabilistic distribution of the axle weight is the critical factor that
directly influences the probabilistic distribution of fatigue stress ranges.
Since the axle weight shown in Figure 2.2 follows a multi-peak distri-
bution, an appropriate probabilistic density function should be considered.
The GMM is also used herein to approximate the probability density of
fatigue stress. The GMM is part of the finite mixture distributions that
are commonly employed for modeling complex probability distributions.
They enable the statistical modeling of random variables with multimodal
behaviors. The basic structure of finite mixture distributions for indepen-
dent scalar y can be expressed as [34]

c
f ( y | c, w, q ) = ∑ wi f i ( y | qi ) (2.2)
i =1

c
∑w
i =1
i = 1 (2.3)

where f ( y | c, w, θ ) is the predictive mixture density function, f i ( y | qi )

is the ith component in a given parametric family of predictive component
densities, wi and θi are the ith component weight and component param-
eter, respectively. For instance, by considering the Gaussian function as
the given approach for predictive component densities, the GMM can be
written as [35]

c
1  1 ( y − mi ) 
f ( y | c, w, q ) = ∑ wi exp −  (2.4)
 2 si 
2
i =1 2À

where μi and σi are the mean value and the standard deviation of the ith
normal mixture parameter. It is observed that the GMM was primly used
for the probabilistic modeling of vehicle weight, and then was used for
the probabilistic modeling of structural fatigue stress range. In addition,
there is a relationship between the vehicle weight and the structural fatigue
stress range. Therefore, the GMM can provide a reliable connection for
the monitored traffic data and the probabilistic modeling of structural
fatigue stress range.
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  25

2.4 LIMIT STATE FUNCTION OF FATIGUE

DAMAGE ACCUMULATION

The structural fatigue damage caused by the truck load accumulates

during the service time. On this basis, the structural component will be
demanded as fatigue failure when the number of stress cycles reaches a
critical value. In practice, it is usually assumed that the fatigue failure
occurs when the structural fatigue damage accumulation reaches the crit-
ical value. The classical method for predicting structural fatigue damage
accumulation is the S-N (stress–life) curve approach together with the
Miner’s linear fatigue damage accumulation theory. In addition, LEFM
approach can also be used to establish the corresponding limit state
function for fatigue cracks which are not the focus of the present study.
The parameters of S-N curves are usually obtained from laboratory test or
extracted from the available design specifications, such as AASHTO [15]
and Eurocode 3 [36].
On the topic of fatigue stress analysis for welded bridge decks under
truck loads, the following two criteria should be considered: (1) the
fatigue strength of welded joints, such as the rib-to-deck and the butt joint
of U-rib, are different and should be considered in the S-N curves; (2)
the low stress cycles should be included since most of the truck induced
fatigue stress cycles have the feature of high-frequency and low-magni-
tude. Herein, the Eurocode 3 specification, which takes both of the preced-
ing criteria contents into account, is utilized in this study and the general
expression of S-N curves is given based on

∆sR3 N R = K C (∆sR ≥ ∆sD ) (2.5)

∆sR5 N R = K D (∆sL < ∆sR ≤ ∆sD ) (2.6)

where ΔσR is the fatigue stress range; NR is the corresponding number

of cycles; ΔσD and ΔσL are the constraint amplitude fatigue limit and the
variable amplitude fatigue limit, respectively; KC and KD are the constant
coefficients for stress ranges greater than ΔσD and that are between ΔσL
and ΔσD, respectively. These parameters for welded joints of rib-to-deck
and butt joint of U-rib specified in the Eruocode3 guidelines are shown
in Table 2.2, which explains why Eurocode 3 [36] was selected for this
study.
The S-N curve is appropriate for calculating the structural fatigue life
with constant amplitude fatigue stresses. However, the fatigue stresses for
these welded details are time-dependent variables due to the fatigue truck
26  •   BRIDGE RELIABILITY AND SERVICEABILITY

Table 2.2.  Parameters of S-N curves in the Eurocode 3 (2005)

specification
Detail
category ΔσD ΔσL
Joint (MPa) (MPa) (MPa) KC KD
Rib-to-deck 50 37 20 2.50×1011
3.47×1014
Butt joint 71 52 29 7.16×1011 19.00×1014

loads. An equivalent fatigue damage accumulation rule is then used to

replace the variable stress ranges by utilizing the Miner’s accumulated
damage rule [19], and is expressed as

ni ∆si3 n j ∆s 5j
D=∑ +∑ (2.7)
i KC j KD
where D is the fatigue damage accumulation, Δσi is the ith fatigue stress
range that Δσi≥ΔσD, Δσj is the jth fatigue stress range that Δσj<ΔσD, ni and
nj are the corresponding number of cycles for Δσi and Δσj, respectively.
With the equivalent fatigue damage accumulation rule, the equivalent
fatigue stress range and corresponding number of cycles are expressed as

ni ∆si3 n j ∆s 5j
∑i K + ∑j K
C D
∆sre5 = (2.8)
N axle / K D

where Δσre is the equivalent fatigue stress range, and Naxle is the corre-
sponding number of stress cycles that equal to the axle number of a truck.
Herein, the variable amplitude stresses can be transformed to constraint
amplitude stresses. Subsequently, the S-N curves in the Eurocode 3 spec-
ification can be used to calculate the fatigue damage.
The previous illustration focuses on fatigue damage expressions
caused by fatigue stress cycles due to an individual truck passage. In prac-
tice, Δσre mainly depends on the axle weight that is random in nature and
should be treated as a random variable. In addition, however, the rela-
tionship between fatigue stress cycles and passing trucks can also affect
the fatigue damage. Thus, additional parameters should be considered in
the limit state function of the fatigue damage. These include the transverse
distribution factor of truck axles, the average daily truck traffic (ADTT),
and the growth factor of traffic volume and the axle weight. Subsequently,
the limit state function can be expressed as
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  27

n
g n (X ) = D∆ − ∑ Di (X )
i =1
n
= D∆ − 365 N ADTT ∆sre5 N ed w / K D ⋅ ∑ (1 + ia )(1 + ib)5 (2.9)
i =1 

where DΔ is the critical fatigue damage, Di is the fatigue damage accu-

mulation in the ith year of life cycle, n is the design life of the bridge,
Ned is the daily equivalent number of stress cycles, w is a coefficient for
transverse distribution of the vehicle tires on the bridge deck, NADTT is the
number of current ADTT, and a and b are annual linear increase factors for
truck volume and GVW, respectively. In the limit state function, the crit-
ical variables associated with the truck load are Δσre and Ned, which have
been appropriately solved by the proposed stochastic fatigue truck model.

2.5 CASE STUDY

The steel box-girders of a suspension bridge are selected herein as a pro-

totype to demonstrate the application of the computational framework and
the stochastic fatigue truck load model. The suspension bridge built to
cross the Yangzi River in China has four lanes with two traffic lanes in the
same traveling direction. The efficiency and the accuracy of the proposed
framework are demonstrated. Impacts of the parameters in the stochastic
truck load model on the structural fatigue reliability are discussed.

2.5.1 HOT SPOT STRESS SIMULATION—

DETERMINISTIC BASIS

Dimensions of the cross section and the U-rib are shown in Figure 2.5.
The half-segment finite element model established by shell63 elements
in ANSYS is shown in Figure 2.6. The deck and the ribs were meshed
with the quadrilateral elements, while the longitudinal stiffening plates,
the diaphragm plates, and the web plates were meshed by triangular ele-
ments. Since the pavement elements have not been considered in the
finite element model, a spreading angle of 45 [27] for a vertical uniformly
distributed wheel load was applied on the bridge deck. For instance, the
thickness of pavement in this case study is 6.7cm, and the load areas are
30cm × 20m and 60cm × 20cm for the front and back wheels, respec-
tively. Therefore, with consideration of the pavement, the corresponding
28  •   BRIDGE RELIABILITY AND SERVICEABILITY

13.8m 13.8m
Sidewalk 1st lane 2nd lane 3rd lane 4th lane
2% 2%
3m

3.54m 11.35m 11.35m 3.54m

Section Ⅰ Section Ⅱ
(a)

Cracks
30cm Bridge
deck

U-rib
28c

Joint I:
m

Rib-to-deck joint
17cm
Joint II:
Butt joint of U-rib

(b)

Figure 2.5.  Dimensions of a steel box-girder: (a) cross section; (b) U-rib.

Slow lane
Fast lane

Figure 2.6.  Finite element model of a half-segment steel

box-girder.

revised load areas are 43.4cm × 33.4cm and 73.4cm × 33.4cm. On this

basis, stress influence lines of two welded joints under the front wheel
load are shown in Figure 2.7, where joints I and II denote the rib-to-
deck and U-rib butt joints, respectively, and the model I and II denote
a one-segment girder model and a three-segment girder model. It can
be concluded from Figure 2.7, the effective fatigue stress is confined to
the region between the two closed diaphragm plates. Since a segment
girder model includes four diaphragm plates in this study, the fatigue
behavior for the joints in the one-segment model is close to that in
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  29

4
Joint I in model I
Joint I in model II
2 Joint II in model I
Joint II in model II
0
Stress (MPa)

-2

-4
Diaphragm plate
-6

-8
0 2 4 6 8 10 12
Longitudinal location in a segment (m)

Figure 2.7.  Stress influence lines of welded joints in two types of finite

element models.

the three-segment model. Even though the segment model has a simi-
lar fatigue behavior with the global model [21], a global model is still
important for a more detail comparison. Nevertheless, this verifies that
the vehicle configuration is sensitive to the fatigue stress, while the vehi-
cle spacing is insensitive.
In order to observe stress–time histories of the welded joints under
truck loads, two V6 trucks with a maximum and a minimum gross weight,
respectively, were added to the finite element model with a constant speed
of 20 m/s. With the transient analysis of the finite element model, the
results are shown in Figure 2.8, where the load case is critical for the
corresponding fatigue stress of the welded joints, and the σmin and σmax
denote the fatigue stress under V6 truck loads with minimum and maxi-
mum GVW, respectively. As observed from the stress–time histories, the
stress amplitudes are time varying. This clearly indicates the necessity of
the equivalent stress range simplification in Equation (2.8). In addition,
the shape of the stress history is associated with the vehicle configuration,
which also demonstrates the importance of the vehicle configuration in
stochastic fatigue truck model. Based on these time–stress histories, the
fatigue stress ranges and the corresponding number of cycles can be cal-
culated by utilizing the rain flow counting method.
30   •   BRIDGE RELIABILITY AND SERVICEABILITY

20
σmin
10
σmax
0
Stress (MPa)

–10

–20

–30 Load case

–40

–50
0 0.2 0.4 0.6 0.8 1 1.2 1.4
(a)
Time (s)

50

40 σmin

30 σmax

20
Stress (MPa)

10

0
Load case

–10

–20
0 0.2 0.4 0.6 0.8 1 1.2 1.4
(b)
Time (s)

Figure 2.8.  Stress–time histories under the V6 truck loads for the welded
joints of: (a) rib-to-deck joint; (b) butt joint of U-rib.

2.5.2 PROBABILISTIC ANALYSIS

The application of the framework in probabilistic modeling of fatigue

stress is illustrated utilizing the stochastic fatigue truck model proposed in
Section 2. Taking the 2nd type of vehicles, V2, as an example, the train-
ing samples were generated with the uniform design sampling scheme.
 FATIGUE RELIABILITY OF STEEL BRIDGES   •   31

U10 and U20, standing for the total number of 10 and 20 uniform distrib-
uted samples, respectively, were selected for approximating the response
surface between the structural equivalent fatigue stress range and the vehi-
cle axle weight. After performing the necessary finite element analyses
for 10 and 20 times, the response surfaces were obtained as shown in
Figure 2.9(a) and Figure 2.9(b), respectively. As shown in Figure 2.9, both
training samples are uniformly distributed in the coordinate space, and the
approximated response surfaces are close to the training data. In addition,
the approximated response surfaces are nonlinear, and the SVR approach
is appropriate for the approximation, since its kennel function is nonlinear.

Training samples with U10 Response surface

Equivalent stress range (MPa)

80

60

40

20

–20
500
400 300
300
200 200
AW22 (kN) 100 100
0 0 Negative AW21 (kN)
(a)

Training samples with U20

Response surface
Equivalent stress range (MPa)

80

60

40

20

0
500
400 300
300
200 200
AW22 (kN) 100 100
0 0 AW21 (kN)
(b)

Figure 2.9.  Response surfaces for the 2rd type of vehicle with the training
sample of: (a) U10; (b) U20.
32  •   BRIDGE RELIABILITY AND SERVICEABILITY

However, when compared with Figure 2.9(a) and Figure 2.9(b), it is easy
to find out that the regions in the corner for both response surfaces are
different, where Δσre with U10 data are negative, while Δσre with U20 data
are positive. The reason for the invalidity of the response surface with
U10 training samples is the lack of training data in the margins. Figure
2.10 shows the comparison between the SVR predictions and the FEA
results, where AW21 was supposed to be 150 kN, and AW22 was supposed
to change from 50 to 500 kN. It can be observed that the SVR predictions
with U20 is in agreement with the FEA results, while two terminals of the
SVR predictions with U10 have a certain difference with the FEA results.
The analysis results indicate that a sufficient number of training samples is
a guarantee for the accuracy of SVR prediction.
With the substitution SVR model, the stress ranges under the
large-sample in the stochastic fatigue truck load model will be easy to cal-
culate. Herein, 30-day stochastic truck data were used for the probability
density modeling for the welded joints in the slow lane and the fast lane.
The results are shown in Figure 2.11 and the corresponding parameters
of the estimated PDFs are shown in Table 2.3. It can be concluded from
Figure 2.11 that the probability density curves of the stress ranges have
multi-peaks and are well fitted with the GMMs rather than the normal
distribution function. In addition, more high-amplitude stress cycles are
included in the fatigue stress spectrum of the welded joints in the slow lane,
while more low-amplitude stress cycles are included in the welded joints
in the fast lane. Note that the shape of the PDFs in Figure 2.11 is different

70
FEA
65 SVR with U10
SVR with U20
Equivalent stress range (MPa)

60

55

50

45

40

35

30
50 100 150 200 250 300 350 400 450 500
AW22 (kN)

Figure 2.10.  Comparison between the SVR predictions with FEA results.

 FATIGUE RELIABILITY OF STEEL BRIDGES   •  33

x 10-3
5
Simulated data
GMM
4 Normal distribution
Probability density
3

0
5 10 15 20 25
Stress range (MPa)

(a)
x 10-3
7
Simulated data
6 GMM
Normal distribution
5
Probability density

0
5 10 15 20 25
Stress range (MPa)
(b)

Figure 2.11.  PDFs of the equivalent fatigue stress in the rib-to-deck

joints of: (a) slow lane; (b) fast lane.

from that calculated by Ye et al. [37] utilizing the SHM data, but they are
similar to the PDFs given by Xia et al. [23]. This result is due to two sim-
plified preconditions in the framework. The first precondition is the vehicle
filters process that vehicle weight less than 30 kN was cut off. The second
precondition is the equivalent stress process that variable amplitude stress
cycles were equivalent to be constant amplitude stress cycles.
34  •   BRIDGE RELIABILITY AND SERVICEABILITY

Table 2.3.  Parameters in the GMMs of Δσre

Mean value μi Standard deviation
Traffic lane Weight (wi) (MPa) σi (MPa)
Slow lane 0.71 3.94 0.33
0.13 7.52 1.52
0.16 15.44 4.75
Fast lane 0.64 1.37 0.24
0.21 7.84 2.35
0.15 14.21 4.20

Since the accuracy of the computational framework was demon-

strated in Figures 2.10 and 2.11, the probabilistic modeling of fatigue
stress ranges can be used to illustrate the efficiency of the computational
framework. After a total number of 150 computer runs corresponding to
approximately four hours by a core 7 computer, the SVRs can be directly
used for the stochastic traffic flow instead of the time-consuming finite
element model. When utilizing the traditional Monte Carlo simulation, all
simulated samples should be directly incorporated into the finite element
model. This results in an explosive growth of computation time due to the
thousands of samples needed for the PDF of the stress ranges. However,
only 150 computer runs is necessary for the training samples of SVRs uti-
lizing the proposed framework. Even though the four hours of computer
runs may be exhaustive, the computational efficiency is superior com-
pared with the computational demand of Monte Carlo simulation.
The number of daily stress cycles can be obtained and the probability
density can be calculated according the probability density of ADTT. Note
that the number of stress cycles due to a truck passage is equivalent to the
number of vehicle axles in Equation (2.8). Figure 2.12 shows the PDFs
of the number of daily cycles approximated with the normal distribution
function taking into account one year traffic data. It is observed that the
probability densities for the number of daily stress cycles are well fitted
by the normal distribution function. In addition, the mean value of the
number of daily cycles in the fast lane is greater than that in the slow
lane. The approximated PDFs for both the fatigue stress range and the
corresponding number of cycles will provide a reasonable basis for a sub-
sequent fatigue reliability assessment of the welded joints. In addition to
the two variables, there are additional variables and constants, as shown in
Equation (2.9), contributing to the fatigue damage. The statistics of these
variables will be illustrated as follows.
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  35

0.004
Simulated data
Normal distribution

0.003
Probability density

m=7927
0.002 s=984

0.001

0
7600 7800 8000 8200 8400
Number of daily cycles
(a)

0.004
Simulated data
Normal distribution

0.003
Probability density

μ=13684
0.002 σ=1130

0.001

0
1.32 1.34 1.36 1.38 1.4
Number of daily cycles 4
x 10
(b)

Figure 2.12.  PDFs of the equivalent number of daily cycles for: (a) slow
lane; (b) fast lane.

The critical fatigue damage, DΔ, in terms of resistance, was assumed

as following a lognormal distribution with a mean value of 1.0 and coef-
ficient of variation (COV) of 0.3 for metallic materials [38]. The fatigue
strength coefficient, KD, was considered as following a lognormal distribu-
tion with a COV of 0.34. The equivalent stress range, Δσre, was computed
from the approximated response functions, and the probability models are
shown in Figure 2.11. The PDFs of corresponding number of daily cycles
36  •   BRIDGE RELIABILITY AND SERVICEABILITY

Table 2.4.  Statistics of the random variables in the limit state function
Mean
Variables Description Distribution value COV
DΔ Critical fatigue damage Lognormal 1.0 0.3
KD Coefficient of fatigue Lognormal 3.47×1014 0.34
strength
w Coefficient of transversal Normal 0.8 1
load distribution of tires
Δσre Equivalent stress range GMM See Table 2.3 and
Figure 2.11
Nde Number of daily cycles Normal See Figure 2.12

are shown in Figure 2.12. The coefficient for the transverse distribution
of vehicle tires on the bridge deck w was assumed to follow a normal
distribution with a mean value of 0.8 and a COV of 1. In conclusion, the
statistics of the variables in the limit state function are shown in Table 2.4.

2.6 RESULTS AND DISCUSSION

Initially, the fatigue reliability assessment is conducted without consid-

eration of the growth factors of both traffic volume and vehicle weight.
On this basis, taking into account the impact of service time and the driv-
ing lane, the fatigue reliability indices of the rib-to-deck joint were cal-
culated based on the limit state function presented by Equation (2.9) and
the ­statistics of the variables shown in Table 2.4. The computed result is
shown in Figure 2.13, where the reliability indices greater than 5 were not
given due to accuracy of the Monte Carlo simulation. It is found, as shown
in Figure 2.14, that the reliability indices for the slow lane and the fast lane
in the 100th year are reduced to 2.9 and 4.3, respectively. Furthermore,
the fatigue reliability indices of the welded joint in the slow lane are lower
than that of the fast lane, because most heavy loaded trucks are prone
to traveling in the slow lane. This result is consistent with the stochastic
truck load model shown in Figure 2.13.
In practice, the traffic volume and the vehicle weight will grow with
the increase in the pace of urban development and the growth of urban
population. Suppose that the annual growth rates of traffic volume and
vehicle weight are both constant. When the linear growth factor of the traf-
fic volume is 0%, 1%, 2%, and 3%, the corresponding reliability indices
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  37

Reliability index 4

Slow lane
Fast lane
2
0 20 40 60 80 100
Service time (year)

Figure 2.13.  Fatigue reliability indices of the rib-to-deck joint.

are shown in Figure 2.14(a). Similarly, if the linear growth factor of the
vehicle weight is assumed to be 0%, 0.2%, 0.4%, and 0.6%, the corre-
sponding reliability indices are shown in Figure 2.14(b). It is observed
from Figure 2.14 that the reliability indices decrease noticeably during the
service time, especially for the case of considering the growth of vehicle
weight. For instance, while the growth factor of the traffic volume is 3%,
the corresponding reliability index in the 100th year decreases to 1.92.
When the growth factor of the vehicle weight is 0.6%, the corresponding
reliability index decreases to 1.35.
The fatigue reliability assessment can also be used to predict the
fatigue life of the steel structures under the target reliability index βtarget.
The target reliability index determines the structural cost of mainte-
nance. Helmerich et al. [30] suggest the βtarget to be in the range between
2.0 and 3.5 according to the calibration study of Eurocode 3 [36]. In the
present book, βtarget is determined to be 2.0 corresponding to the failure
probability of 2.3%. The structural fatigue life under the βtarget is shown
in Table 2.5. It is observed from Table 2.5 that the fatigue life of the
welded joints decreases significantly with the growth of traffic volume
and vehicle weight. When the linear annual growth factor of traffic vol-
ume a is greater than 3%, the corresponding fatigue life is only 97 years,
that is less than the expected design service life. When the growth factor
of vehicle weight b is greater than 0.4%, the fatigue life of the welded
details is only 94 years, which is also less than the expected design
service life.
38  •   BRIDGE RELIABILITY AND SERVICEABILITY

6
a=0
a=1%
5 a=2%
a=3%

Reliability index
4

1
0 20 40 60 80 100
Service time (year)
(a)
6
b=0
b=0.2%
5
b=0.4%
b=0.6%
Reliability index

1
0 20 40 60 80 100
Service time (year)

(b)

Figure 2.14.  Fatigue reliability indices taking into account the

linear annul growth factors of: (a) traffic volume; (b) vehicle weight.

Table 2.5.  Fatigue life predication for the rib-to-deck joint taking into
account traffic growth factors
Fatigue life predication (year)
a=b a= a= a= b= b= b=
Traffic lane =0 1% 2% 3% 0.2% 0.4% 0.6%
Slow lane 195 135 108 96 134 94 72
Fast lane 297 195 161 148 186 143 120

From the earlier parametric study we can conclude, and summarize,

that the traffic parameters, such as the vehicle volume, vehicle weight, and
driving lane, have a significant impact on the fatigue reliability of steel
bridge decks. Thus, it is concluded that a probabilistic model of these traf-
fic parameters should be considered and included in the analytical model.
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  39

The stochastic traffic flow model herein provides a promising solution

for the connection between the measured traffic vehicle parameters and
probabilistic modeling of the structural fatigue damage accumulation.
Furthermore, this study provides an efficient and accurate computational
framework for the fatigue reliability assessment of welded steel bridge
decks under stochastic traffic flows, as illustrated through a case study that
is based on a multi-year collected data. The case study results can also
provide a reference for the highway traffic management in order to ensure
the fatigue safety of existing bridges.

2.7 CONCLUSIONS

As discussed in this chapter, a stochastic fatigue truck load model was

developed for probabilistic modeling of fatigue stress ranges for the pur-
pose of investigating fatigue reliability of welded steel bridge decks.
In order to deal with the uncertainty-induced computational complex-
ity, a framework including deterministic finite element-based hot spot
analysis and probabilistic modeling approaches was presented. In addi-
tion, a learning machine integrating uniform design and support vector
­regression was employed to substitute the time-consuming finite element
model. The developments of both the stochastic truck load model and the
computational framework provide a novel and feasible approach for the
probabilistic modeling of fatigue damage accumulation for steel bridge
decks. A limit state function of fatigue damage was established taking into
account traffic flow parameters, such as growth factors of traffic volume
and vehicle weight. On this basis, the stochastic truck load model and the
corresponding framework were applied to fatigue reliability assessment of
steel bridge decks.
A prototype steel box-girder bridge was presented as a practical
demonstration to illustrate the application of the proposed fatigue truck
load model and the corresponding computational framework in fatigue
reliability assessment. The deterministic analysis results indicated that the
accuracy of the method mostly depend on the number of training sam-
ples, which inevitably affected the computational efficiency. A number of
20 training samples were proved to be reasonable for training the learn-
ing machine of the 2-axle truck, while the total number of 150 was nec-
essary for the six types of vehicles. This demonstrated that probabilistic
modeling approach with the learning machine is much more efficient than
the traditional Monte Carlo simulation. Parametric studies indicated that
growth in traffic volume and vehicle weight has a different influence on
40   •   BRIDGE RELIABILITY AND SERVICEABILITY

the structural fatigue reliability index. A linear annual growth factor of

3% in traffic volume or 0.6% in vehicle weight will result in a fatigue reli-
ability index of less than 2 for the prototype bridge. The result of the case
study can provide a theoretical basis for both future traffic management
and structural maintenance.
Further efforts are needed to improve the framework utilized herein
that is based on both deterministic finite element analysis and probabilis-
tic modeling of the fatigue stress ranges. If an appropriate finite element
model and a fracture limit state function are considered, the proposed sto-
chastic fatigue truck model can also be used to analyze fatigue crack prop-
agation. In addition, the vehicle–bridge interaction analysis should also
be considered in order to deal with the issues of degradation of the road
surface roughness and the variation of vehicle driving speed. Furthermore,
uncertainties associated with structural geometries, material and mechani-
cal properties should be investigated in the future studies.

REFERENCES

[1] Zhang, S., X. Shao, J. Cao, J. Cui, J. Hu, and L. Deng. 2016. “Fatigue
Performance of a Lightweight Composite Bridge Deck with Open Ribs.”
Journal of Bridge Engineering 21, no. 7, 04016039. 10.1061/(ASCE)
BE.1943-5592.0000905.
[2] Liu, M., D. Frangopol, and K. Kwon. 2010. “Fatigue Reliability Assessment
of Retrofitted Steel Bridges Integrating Monitored Data.” Structural Safety
32, no. 1, pp. 77–89.
[3] Deng, L., W. Wang, and Y. Yu. 2015. “State-of-the-art Review on the Causes
and Mechanisms of Bridge Collapse.” Journal of Performance of Constructed
Facilities 30, no. 2, 04015005. 10.1061/(ASCE)CF.1943-5509.0000731.
[4] Sim, H., and C. Uang. 2012. “Stress Analyses and Parametric Study on
Full-scale Fatigue Tests of Rib-to-deck Welded Joints in Steel Orthotropic
Decks.” Journal of Bridge Engineering, 765–73. 10.1061/(ASCE)
BE.1943-5592.0000307.
[5] Ye, X., Y. Su, and J. Han. 2014. “A State-of-the-art Review on Fatigue Life
Assessment of Steel Bridges.” Mathematical Problems in Engineering.
doi:10.1155/2014/956473.
[6] Frangopol, D., and M. Soliman. 2016. “Life-cycle of Structural Systems:
Recent Achievements and Future Directions.” Structure and Infrastructure
Engineering 12, no. 1, pp. 1–20.
[7] Chen, S., and J. Wu. 2011. “Modeling Stochastic Live Load for Long-span
Bridge Based on Microscopic Traffic Flow Simulation.” Computers &
Structures 89, no. 9, pp. 813–24.
 FATIGUE RELIABILITY OF STEEL BRIDGES   •   41

[8] Xu, Y., Z. Chen, and Y. Xia. 2012. “Fatigue Assessment of Multi-loading
Suspension Bridges Using Continuum Damage Model.” International
Journal of Fatigue 40, pp. 27–35.
[9] Wang, Y., Z. Li, and A. Li. 2010. “Combined Use of SHMS and Finite
Element Strain Data for Assessing the Fatigue Reliability Index of Girder
Components in Long-span Cable-stayed Bridge.” Theoretical and Applied
Fracture Mechanics 54, no. 2, pp. 127–36.
[10] Guo, T., and Y.W. Chen. 2013. “Fatigue Reliability Analysis of Steel
Bridge Details Based on Field-Monitored Data and Linear Elastic Fracture
Mechanics.” Structure and Infrastructure Engineering 9, no. 5, pp. 496–505.
[11] Ni, Y., X. Ye, and J. Ko. 2010. “Monitoring-based Fatigue Reliability
Assessment of Steel Bridges: Analytical Model and Application.” Journal
of Structural Engineering 136, no. 12, 1563–73. 10.1061/(ASCE)
ST.1943-541X.0000250.
[12] D’Angelo, L., and A. Nussbaumer. 2015. “Reliability Based Fatigue
Assessment of Existing Motorway Bridge.” Structural Safety 57, no. 7,
pp. 35–42.
[13] Laman, J., and A. Nowak. 1996. “Fatigue-load Models for Girder
Bridges.” Journal of Structural Engineering 122, no. 7, 726–733. 10.1061/
(ASCE)0733-9445.
[14] European Committee for Standardization (ECS). 2003. “Eurocode 1: Actions
on Structure—Part 2: Traffic Loads on Bridges.” EN 1991-2, Brussels,
Belgium.
[15] AASHTO. 2010. LRFD Bridge Design Specifications, 5th ed. Washington,
DC: American Association of State Highway and Transportation Officials.
[16] Zhang, W., C. Cai, and F. Pan. 2013. “Fatigue Reliability Assessment for
Long-span Bridges Under Combined Dynamic Loads from Winds and
Vehicles.” Journal of Bridge Engineering 18, no. 8, 735–47. 10.1061/
(ASCE)BE.1943-5592.0000411
[17] Zhang, W., C.S. Cai, F. Pan, and Y. Zhang. 2014. “Fatigue Life Estimation
of Existing Bridges Under Vehicle and Non-stationary Hurricane Wind.”
Journal of Wind Engineering and Industrial Aerodynamics 133, no. 10, pp.
135–45.
[18] MacDougall, C., M. Green, and S. Shillinglaw. 2006. “Fatigue Damage
of Steel Bridges Due to Dynamic Vehicle Loads.” Journal of Bridge
Engineering 11, no. 3, 320–28. 10.1061/(ASCE)1084-0702.
[19] Wang, W., L. Deng, and X. Shao. 2016. “Fatigue Design of Steel Bridges
Considering the Effect of Dynamic Vehicle Loading and Overloaded
Trucks.” Journal of Bridge Engineering, 21, no. 9, 04016048. 10.1061/
(ASCE)BE.1943-5592.0000914, 04016048.
[20] Zhang, W., and C. Cai. 2011. “Fatigue Reliability Assessment for Existing
Bridges Considering Vehicle Speed and Road Surface Conditions.” Journal
of Bridge Engineering, 443–53. 10.1061/(ASCE)BE.1943-5592.0000272.
42  •   BRIDGE RELIABILITY AND SERVICEABILITY

[21] Guo, T., D. Frangopol, and Y. Chen. 2012. “Fatigue Reliability Assessment
of Steel Bridge Details Integrating Weigh-in-motion Data and Probabilistic
Finite Element Analysis.” Computers & Structures 112, pp. 245–57.
[22] Guo, T., Z. Liu, and J. Zhu. 2015. “Fatigue Reliability Assessment of
Orthotropic Steel Bridge Decks Based on Probabilistic Multi-scale Finite
Element Analysis.” Advanced Steel Construction 11, no. 3, pp. 334–46.
[23] Xia, H., Y. Ni, K. Wong, and M. Ko. 2012. “Reliability-based Condition
Assessment of in-Service Bridges Using Mixture Distribution Models.”
Computers & Structures 106–107, no. 9, pp. 204–13.
[24] Chen, W.Z., J. Xu, B.C. Yan, Z.P. Wang. 2015. “Fatigue Load Model for
Highway Bridges in Heavily Loaded Areas of China.” Adv. Steel Constr. 11,
no. 3, pp. 322–33.
[25] Chen, S., and J. Wu. 2009. “Dynamic Performance Simulation of Long-
span Bridge Under Combined Loads of Stochastic Traffic and Wind.”
Journal of Bridge Engineering 15, no. 3, 219–30. 10.1061/(ASCE)
BE.1943-5592.0000078.
[26] Han, W., J. Wu, C. Cai, and S. Chen. 2014. “Characteristics and Dynamic
Impact of Overloaded Extra Heavy Trucks on Typical Highway Bridges.”
Journal of Bridge Engineering 20, no. 2, 05014011. 10.1061/(ASCE)
BE.1943-5592.0000666.
[27] Liu, Y., N. Lu, M. Noori, and X. Yin. 2014. “System Reliability-based
Optimization for Truss Structures Using Genetic Algorithm and Neural
Network.” Int. J. of Relia. and Saf. 8, no. 1, pp. 51–69.
[28] Suykens, J., and J. Vandewalle. 1999. “Least Squares Support Vector
Machines Classifiers.” Neural Processing Letters 19, no. 3, pp. 293–300.
[29] Moura, M., E. Zio, I. Lins, and E. Droguett. 2011. “Failure and Reliability
Prediction by Support Vector Machines Regression of Time Series Data.”
Reliab. Eng. & Syst. Safe. 96, no. 11, pp. 1527–34.
[30] Helmerich, R., B. Kühn, and A. Nussbaumer. 2007. “Assessment of Existing
Steel Structures: A Guideline for Estimation of the Remaining Fatigue Life.”
Struct. and Infrastructure E. 3, no. 3, pp. 245–55.
[31] Huang, C., Y. Lee, D. Lin, and S. Huang. 2007. “Model Selection for Support
Vector Machines via Uniform Design.” Computational Statistics & Data
Analysis 52, no. 1, pp. 335–46.
[32] Deng, S., and T. Yeh. 2011. “Using Least Squares Support Vector Machines
for the Airframe Structures Manufacturing Cost Estimation.” International
Journal of Production Economics 131, no. 2, pp. 701–08.
[33] Kisi, O., and M. Cimen. 2011. “A Wavelet-Support Vector Machine
Conjunction Model for Monthly Streamflow Forecasting.” Journal of
Hydrology 399, no. 1, pp. 132–40.
[34] Deng, Y., Y. Ding, A. Li, and G. Zhou. 2011. “Fatigue Reliability Assessment
for Bridge Welded Details Using Long-term Monitoring Data.” Science
China Technological Sciences 54, no. 12, pp. 3371–81.
 FATIGUE RELIABILITY OF STEEL BRIDGES   •  43

[35] Jin, S., X. Qu, and D. Wang. 2011. “Assessment of Expressway Traffic Safety
Using Gaussian Mixture Model Based on Time to Collision.” International
Journal of Computational Intelligence Systems 4, no. 6, pp. 1122–30.
[36] European Committee for Standardization (ECS). 2005. “Eurocode 3: Design
of Steel Structures—Part 1–9: Fatigue.” EN1993-1-9, Brussels, Belgium.
[37] Ye, X., Y. Ni, K. Wong, and J. Ko. 2012. “Statistical Analysis of Stress
Spectra for Fatigue Life Assessment of Steel Bridges with Structural Health
Monitoring Data.” Engineering Structures 45, no. 12, pp. 166–76.
[38] Wirsching, P. 1984. “Fatigue Reliability for Offshore Structures.” Journal of
Structural Engineering 110, no. 10, pp. 2340–56.
 CHAPTER 3

First-Passage Probability
of the Deflection of a
Cable-Stayed Bridge Under
Long-term Site-Specific
Traffic Loading

3.1 INTRODUCTION

The rapid growth of global economy has posed a steady traffic growth
in both the volume and the gross vehicle weight (GVW) over recent
decades. Overweight trucks have become a main cause of reducing the
service life or even collapses of existing bridges [1]. In comparison to
short- and medium-span bridges, long-span bridges suffer from higher
traffic loads and the simultaneous presence of multiple vehicles [2]. In
addition, the heavy traffic load, and the steady traffic growth may pose
an additional threat to the safety and serviceability of long-span bridges.
In general, a native design specification of traffic loading has taken
into account the probabilistic extrapolation of traffic loading as well as
the simultaneous presence of multiple vehicles. In addition, the exist-
ing bridges were designed with consideration of the uncertainties of the
traffic load and system reliability [3, 4]. However, the continuous traffic
growth and extremely overloaded trucks have changed the probability
characteristics of the traffic loading [5]. Such changes have been demon-
strated to result in a larger load effect compared to the national design
specifications. For instance, Han et al. [6] found that 4 out of 1,319 trucks
cause larger hogging moment in bridge girders, a bending moment that
produces convex bending, than, for instance, the China’s code’s [7] value.
46  •   BRIDGE RELIABILITY AND SERVICEABILITY

Therefore, instead of utilizing the specified traffic loading, integrating the

site-specific traffic loading is an effective approach for condition assess-
ment of a long-span bridge.
Uncertainties in the traffic flow are the main factors leading to the sto-
chastic feature of the traffic load effects [8]. These uncertainties are usually
quantified by site-specific traffic measurements. With the development of
the sensor technology, weigh-in-motion (WIM) systems, which are gradu-
ally developed for traffic management, has provided a huge database for sta-
tistical modeling of the traffic loading [9, 10]. Fatigue reliability evaluation
of orthotropic steel bridge decks under stochastic fatigue-truck loads sim-
ulated based on WIM measurement was investigated by Lu et al. [11] and
Liu et al. [12]. Advanced statistical approaches have been developed utiliz-
ing probabilistic analysis with WIM measurements, such as the peaks over
threshold (PVT), generalized extreme value (GEV) distribution, and Rice’s
level-crossing approach [13, 14]. The block maxima (BM) load effects for
short- and medium-span bridges presented by Enright and O’Brien [15]
demonstrate that the GEV distribution fits the BM values fairly good. Chen
[16] investigated the influence of the starting point on the crossing rate
fitting. Applications of the statistical model are to evaluate the extrapola-
tions as well as the characteristic load effect. For instance, the live load
in AASHTO [17] specification was extrapolated based on the WIM sites
in Ontario. In addition, the live load specified in Eurocode 1 (2003) was
extrapolated in 1000-year return period based on the WIM sites in France
[18]. Ruan et al. [19] proposed a site-specific traffic load model for long-
span multi-pylon cable-stayed bridges with consideration of traffic volume
growth and heavy vehicle proportion. As elaborated earlier, great achieve-
ments have been obtained on applying WIM measurements for traffic load
modeling. However, to the best of the authors’ knowledge, application of
WIM measurements for reliability assessment of bridges is insufficient.
First-passage probability describes the probability of a scalar process
exceeding a prescribed threshold during an interval of time [20]. In practice,
the first-passage probability has been used in several areas of engineering
structures. Noori et al. [21] investigated the first-passage probability of a
highly nonlinear hysteretic system via quasi-conservative stochastic averag-
ing. Chen and Li [22] proposed a probability density evolution method for
dynamic response analysis and reliability assessment of nonlinear stochastic
structures to estimate the first-passage reliability. Khan et al. [23] evaluated
the first-passage probability of a cable-stayed bridge against the seismic load.
Significant progresses in structural reliability evaluation have been achieved
over the last decades for nonlinear stochastic structural dynamics [24].
Given the nature of stochastic behavior of the traffic loading, the first-pas-
sage probability can be the best description of the load effect crossing the
 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •  47

threshold value considered in the native bridge design specifications. Most

research efforts in this regard have concentrated to seismic loads, while little
effort has been done devoted to utilizing first-passage probability for reli-
ability assessment of long-span bridges under traffic loading.
This work presented in this book aims at implementing WIM
­measurements for first-passage reliability assessment of long-span bridges
subject to heavy traffic loading. As the first step, stochastic traffic loads are
simulated based on the WIM measurements of a highway bridge in China.
A computational framework is presented integrating Rice’s level-crossing
theory and first-passage criterion. Finally, effectiveness of the computa-
tional framework is demonstrated by a case study of a cable-stayed bridge.
Influences of the traffic growth factors on the first-passage probabilities of
the bridge deformation are investigated.

3.2 STOCHASTIC TRAFFIC LOAD SIMULATION

BASED ON WIM MEASUREMENTS

3.2.1 DESCRIPTION OF WIM MEASUREMENTS

A WIM system utilizes scales or pressure sensors embedded into the road
pavement to measure the traffic volume and the GVW of passing vehicles
[25]. In practice, WIM systems can be used for a wide range of tasks,
such as protection and management of highways and other infrastructure
investments. In the present study, WIM measurements are used for statis-
tical analysis of the traffic loading.
A bridge WIM system in Sichuan, China, was chosen herein as a pro-
totype. The WIM system has been working since the bridge was opened to
the public in May, 2012. Details of the WIM system can be found in Liu et
al. [26]. Before statistical analysis of the measurements, a filtering process
was conducted in order to identify and remove invalid records from the
database. The criteria to identify the invalid records are: (1) the GVW is
less than 30 kN; (2) the axle weight is greater than 300 kN; (3) the vehicle
length is greater than 20 m; and (4) the data are incomplete or flagged with
the system error. Overview of the filtered WIM measurements is shown in
Table 3.1. It is worth noting that the maximum GVW for a 6-axle truck is
550 kN according to the traffic laws in China [27]. However, as observed
in Table 3.1, the number of overloaded trucks is approximately 17 per day.
The overload rate of the maximum gross vehicle weight (GVW) is about
200 percent over the threshold value.
Based on these measurements, the trucks were classified into six cate-
gories. Proportion of the vehicle types are shown in Figure 3.1(a), where V1
48  •   BRIDGE RELIABILITY AND SERVICEABILITY

Table 3.1.  Overview of the WIM measurements

Items Values
Time period May 1, 2013 to April 30, 2015
Number of recording days 729
Average daily truck traffic (ADTT) 2,145
Number of traffic lanes 4
Maximum GVW (kN) 1,645
Number of overloaded trucks 12,252

indicates the light car, V2 to V6 indicate the 2-axle to 6-axle trucks, respec-
tively. It is observed that 39.24 percent of the entire vehicles are 3- to 6-axle
trucks, which is the main contribution to the critical traffic load scenario.
The ADTT was fitted by a normal distribution with a mean value of 2,145
and standard deviation of 424 as shown in Figure 3.1(b). The hourly traffic
density in one day is shown in Figure 3.1(c), where the traffic is divided to
free flow and busy flow. It is worth noting that the vehicle density mostly
depends on vehicle spacing, which is another factor contributed by the crit-
ical traffic load scenario. Therefore, the busy flows between 9:00 and 22:00
were used for statistical analysis of the vehicle spacing. The vehicle spac-
ing of busy flows was fitted by a lognormal distribution with a mean value
of 6.21 and standard deviation of 1.40 as shown in Figure 3.1(d).
In addition to the large-scale statistics shown in Figure 3.1, the small-
scale statistics of individual trucks were also analyzed. Taking V6 as an
example, the histogram and PDFs of the axle weight and GVW are shown
in Figure 3.2, where AW64 indicates the 4th axle of the 6-axle trucks. The
Gaussian mixture model (GMM) as shown in Figure 3.2(b) has captured
feature of the bimodal distribution. Both the large-scale and small-scale
information provide statistical parameters for the traffic load simulation.
Based on the previous demonstration, the remaining statistical anal-
yses were conducted to provide the entire probability model for the sub-
sequent stochastic traffic simulation. Most of the extreme values, which
will be used for probabilistic modeling of first-passage failure probability,
are caused by dense traffic flows. Therefore, the vehicle spacing of dense
traffic flow is effective for the traffic load simulation.

3.2.2 STOCHASTIC TRAFFIC LOADING SIMULATION

With the statistics of WIM measurements, Monte Carlo simulation

(MCS) was utilized to simulate the stochastic traffic load in time domain.
FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •  49

V6 15.49%
V1 35.64%

V5 4.93%

V4 10.24%

V3 8.58%

V2 26.12%

(a)

x 10–3
1.4
Monitored data
1.2 Normal distribution
1
Probability density

m=2145
0.8 s=424

0.6

0.4

0.2

0
1000 1500 2000 2500 3000 3500
(b) ADTT

0.8
Probability density

0.6
Busy flow

0.4
Free flow

0.2

0
0:00 4:00 8:00 12:00 16:00 20:00 24:00
(c) Time

1
Empirical data
Lognormal
0.8
Probability density

0.6

0.4

m=6.21
0.2 s=1.40

0
0 1000 2000 3000 4000
(d) Vehicle spacing (m)

Figure 3.1.  Probability densities of the traffic flow: (a) pro-

portions of vehicle types; (b) the time-variant vehicle density;
(c) ADTT; (d) vehicle spacing of the busy traffic flow.
50  •   BRIDGE RELIABILITY AND SERVICEABILITY

0.12
WIM data
0.1 GMM

Probability density
a1=0.38 a2=0.62
0.08
m1=47.22 m2=132.45

0.06 s1=7.48 s2=28.41

0.04

0.02

0
50 100 150 200 250
(a) Axle weight AW64 (kN)

0.02
WIM data
GMM
0.015

a1=0.24 a2=0.76
PDF

0.01 m1=390 m2=865

s1=74 s2=142

0.005

0
0 500 1000 1500 2000
(b) GVW (kN)

Figure 3.2.  Histograms and PDFs of the 6-axle: (a) axle weight;

and (b) GVW.

Assuming a linear growth factor of ADTT to be 0.5 percent, the simulated

traffic loads in 60 minutes for the present case and the 100th year case are
shown in Figure 3.3.
As shown in Figure 3.3, each symbol indicates a truck specified by a
different mark style, x-axis shows the arrival time, and y-axis shows the
individual GVW. It is observed that every individual truck is different from
another, but they follow a relative probability distribution. Therefore, the
stochastic truck load model contains the statistics of the WIM measurements.

3.3 METHODOLOGY

3.3.1 RICE’S LEVEL CROSSING THEORY

The principle of the Rice’s level crossing theory [28] is shown in

Figure 3.4. The precondition of using the Rice’s formula is to make
 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •   51

1000
V1
V2
800 V3
V4

GVW (kN)
600 V5
V6
400

200

0
0 10 20 30 40 50 60
(a) Arrival time (h)
1000

800
GVW (kN)

600

400

200

0
0 10 20 30 40 50 60
(b) Arrival time (min)

Figure 3.3.  Simulated stochastic traffic loads: (a) the present

case; and (b) the 100th year.

Load effect

x1

o t
x2
Level crossing (x>0)
Level crossing (x<0)

Figure 3.4.  Description of Rice’s level crossing theory.

sure that the load effects follow a normal distribution. In this regard,
Ditlevsen [29] has demonstrated that the bridge load effect can be mod-
eled as Gaussian random process under the case of both a long influence
line and a dense traffic. Fortunately, the long-span bridge in the present
52  •   BRIDGE RELIABILITY AND SERVICEABILITY

study has a long influence line. In addition, only busy traffic flows are
used for the traffic load effect analysis. Thus, the bridge load effect can
be assumed as Gaussian random process and the Rice formula is appro-
priate to be used.
Based on the preceding assumption, the mean up-crossing rate v(x)
under the condition of a threshold level is expressed as:

s′  ( x − m) 2 
v( x) = exp  −  (3.1)
2πs  2s 2 

where x is a random process, which represents the load effect in the present
study, m and σ are the mean value and the standard deviation of x, and σ¢ is
the standard deviation of the derivative of x. With consideration of the return
period, the cumulative distribution function (CDF) can be written as [30]:

  1 x − m 2  
F ( x) = exp  −v0T exp  −     (3.2)
  2  s   

where v0 is σ¢/2π, and T is the service period of a bridge. For extrapolation

utilizing the Rice formula, Cremona [30] recommended an optimal level
about how to determine the threshold based on the minimization of the
Kolmogorov–Smirnov statistical equation which is written as:

r + sr
xmax ( Rt ) = mopt
opt 2 ln (v0, optRt ) (3.3)

where xmax is the extrapolation of maximum load effect in a return period

Rt, mopt
r r
, and σ opt denote the mean value and standard deviation of the
optimal fitting starting point, v0,opt is the level crossing rate at the optimal
fitting starting point.

3.3.2 FIRST-PASSAGE PROBABILITY

The first-passage probability is essential for estimating the reliability

of a structural component whose response is a stochastic process. The
first-passage probability of a zero-mean stochastic process crossing a pre-
scribed double-sided threshold during an interval time can be written as:

p(a,τ ) = P(a ≤ max 0≤t ≤τ X (t ) ) (3.4)

 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •  53

where X(t) is the zero-mean stochastic process, a is the prescribed thresh-

old, and t∈(0, τ) is the interval time. In general, there is no exact solution
for the probability. On the basis of Poisson assumption, the probability of
failure is written as:

t
p(a, t) ≅ 1 − A exp  − ∫ v(a, t ) d t  (3.5)
 0 
a
A = P  X (0) < a  = ∫ f X ( x, 0) d x (3.6)
−a

where fx(x, t) denotes PDF of X(t), A is a coefficient that is approximated

to 1 for a Gaussian process, and v(a, t) denotes the mean up-crossing rates
of the process X(t) at the threshold of a.
Considering the growth rate of the traffic flow during the bridge ser-
vice period, the first-passage probability is written as:

n
p(a, T ) ≅ 1 − ∑ exp  −v(a, ti ) ⋅ ri ⋅ ti  (3.7)
i =1

where the service period T is divided into T intervals indicated as ti, and n
is the maximum service period in year. Since the traffic density and GVW
are growing in the service period, v(a, ti) is time-variant.

3.3.3 PROPOSED COMPUTATIONAL FRAMEWORK

Considering the theoretical basis illustrated earlier, a computational frame-

work is presented for the purpose of evaluating first-passage probability
of a long-span bridge under stochastic traffic loading. A flowchart of
describing the procedures in the framework is shown in Figure 3.5.
As shown in Figure 3.5, the framework begins with WIM mea-
surements and the bridge detail geometry and material properties.
Deterministic finite-element simulation is conducted to get the extreme
load effect of the bridge under the traffic loading. With the deterministic
results, Rice’s formula is used to fit the level crossing rate based on the
upper data of the extreme load effect. Subsequently, the extrapolation and
first-passage probability can be estimated by considering the return period
and threshold values, respectively.
During the long-term service period of a bridge, the traffic growth
is considered in the stochastic traffic flow, which is updated at the end
of the framework. The traffic growth model can be considered as an
54  •   BRIDGE RELIABILITY AND SERVICEABILITY

WIM Bridge details

measurements

Stochastic heavy traffic loading Finite-element model

Static influence lines

Extreme load effect simulation
based on finite-element method
Rice’s formula
Level crossing rate fitting
Updating Return period
Extrapolation of extreme load effects
Threshold value
First-passage probability evaluation

No
Traffic growth model?

Yes
Life-cycle reliability evaluation

Figure 3.5.  Computational framework for first-passage reliability

­evaluation of a long-span bridge under traffic loading.

annual compounded growth rate of the traffic volume and the GVW. Since
the static influence lines are used to calculate the traffic load effect, the
computational effort can be ignored.

3.4  CASE STUDY

3.4.1 BRIDGE DETAILS

Hejiang Yangtze River Bridge is a long-span cable-stayed bridge in

Sichuan, China. The span length of the bridge is 210 m + 420 m + 210 m
as shown in Figure 3.6. The material of girders and cables are concrete
and steel, respectively. There are four traffic lanes in opposite direc-
tions. A finite-element model as shown in Figure 3.6 was built with a
commercial program ANSYS. The girders and pylons were simulated by
Beam188 elements. The cables were simulated by Link180 elements.

3.4.2 PROBABILISTIC MODELING OF TRAFFIC LOAD

EFFECTS

The traffic load effects were computed by the static influence lines of the
bridge. The displacement influence lines of the critical points of the bridge
 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •  55

7+33×6+5=210 m 5+34×6+2+34×6+5=420 m 7+33×6+5=210 m

140 m

Figure 3.6.  Dimensions of a cable-stayed bridge.

Figure 3.7.  Finite element model of the cable-stayed bridge.

were computed in the finite-element model as shown in Figure 3.7 by a

moving load of 100 kN. As shown in Figure 3.8, the maximum values for
the quarter-span point (L/4) and the mid-span point (L/2) are 0.0274 m and
0.0182 m, respectively, where L is the length of mid-span of the bridge.
The displacement histories of the critical points of the bridge under one-
hour stochastic traffic loading are shown in Figure 3.9. It is observed that
the L/2 point has a larger vibration amplitude compared to the L/4 point.
This phenomenon can be explained by the first-order mode shape of the
cable-stayed that is symmetric.
A total number of 365-day traffic loading has been used to com-
pute the traffic load effects. Based on the deterministic histories, the
up-crossing rate was fitted by the Rice’s level crossing formula as shown
in Equation (3.1). In order to obtain the extreme value of the traffic load
effect, the 30 percent upper data were used to fit the tail up-crossing rate
as shown in Figure 3.10. Subsequently, the tail fittings of the load effects
calculated by Equation (3.2) were plotted on Gumbel probability paper as
shown in Figure 3.11.
56  •   BRIDGE RELIABILITY AND SERVICEABILITY

0.01
L/4
L/2

Vertical displacement (m)

0

Pylon Pylon
-0.01

-0.02

-0.03
-400 -200 0 200 400
Longitudinal location (m)

Figure 3.8.  Displacement influence lines of the critical points of

the bridge.

0.4
L/4
Vertical displacement (m)

L/2
0.2

-0.2

-0.4
0 10 20 30 40 50 60
Time period (min)

Figure 3.9.  Displacement histories of the critical points of the bridge.

1
L/2 data
Normalized up-crossing rate

v(x)
0.8
L/4 data
v(x)
0.6

0.4

0.2

0
0.2 0.3 0.4 0.5 0.6 0.7
Threshold displacement (m)

Figure 3.10.  Up-crossing histograms and fittings.

 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •  57

6
L/4 data
5 L/4 fitting
L/2 data
4 L/2 fitting

log(-log(P)
3

0
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Threshold displacement (m)

Figure 3.11.  Tail fittings of the extreme load effects on Gamble paper.

It is observed from Figures 3.10 and 3.11 that the L/4 point of the
bridge has a higher up-crossing rate and maximum load effect compared
to the L/2 point of the bridge. In addition, the tail fitting of the up-cross-
ing rate is an appropriate description of probability feature of the extreme
traffic load effect.

3.4.3  RESULTS AND INTERPRETATIONS

Based on the probability model of the traffic load effect, the extrapolation
of extreme load effects and the first-passage probability can be evaluated
with consideration of return period and threshold value, respectively. Since
the mid-span point of the bridge is the critical location for the deformation
failure, the following discussion focuses on the mid-span point. Suppose
the annual linear growth rate of both the ADTT and the GVW (RADTT and
RGVW) to be in the range between 0 and 2 percent. The maximum displace-
ment during the service period was evaluated as shown in Figure 3.12. It is
observed that the maximum displacement is 0.74 m without consideration
of the traffic growth, and the value increases with the increase of RADTT or
RGVW. However, their influences on the maximum displacement are differ-
ent. With the linear increase of RADTT the increase ratio of the maximum
displacement slows down gradually, while the increase ratio of the max-
imum displacement grows linearly with the linear increase of RGVW. For
instance, under the condition of growth rate of 2 percent, the RADTT and
RGVW will lead to the maximum of 0.87 and 0.92 m, respectively.
According to the design code of concrete cable-stayed bridges in China
[31], the threshold in Equation (3.4) is a = L/400 = 1.05 m. The evaluated
58  •   BRIDGE RELIABILITY AND SERVICEABILITY

0.9

Maximum displacement (m)

0.8

0.7 RADTT=0
RADTT=0.5%
RADTT=1
0.6
RADTT=1.5%
RADTT=2%
0.5
0 20 40 60 80 100
(a) Service period (year)

1
Maximum displacement (m)

0.9

0.8

0.7 RGVW=0
RGVW=0.5%
0.6 RGVW=1

0.5 RGVW=1.5%
RGVW=2%
0.4
0 20 40 60 80 100
(b) Service period (year)

Figure 3.12.  Extrapolated maximum displacements during the

service period with consideration of (a) growth rate of ADTT;
(b) growth rate of GVW.

first-passage probability is shown in Figure 3.13. It is observed that the first-pas-

sage probability is 3.9 × 10−7 without consideration of traffic growth, and the
value increases rapidly with the growth of the ADTT and GVW. In addition,
with the sustained growth of RADTT, the impact on the probability of failure
is weakened. However, the sustained growth of RGVW has a constant impact on
the probability of failure. Under the growth rate of 2 percent for the RADTT and
RGVW, the probability of failure is 3.7 × 10−7 and 7.0 × 10−7, respectively.
In addition to the traffic growth, a reasonable traffic management
can reduce the probability of failure. Suppose the threshold overload
ratio to be limited in the region between 0 to 100 percent of the standard
value in MOCAT [27], the probability of failure was estimated as shown
in Figure 3.14. It is observed that the probability of failure has an obvi-
ous decrease with the adoption of a threshold overload ratio. With the
 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •  59

10–6
RADTT
RGVW

Probability of failure
10–7

10–8

10–9
0 0.5 1 1.5 2
Annual linear growth rate (%)

Figure 3.13.  Influence of the traffic growth on the first-passage

probability of failure.

10-6
RADTT=2%
RGVW=2%
Probability of failure

10-7

10-8

10-9
Unlimited 100% 75% 50% 25% 0
Threshold overload ratio

Figure 3.14.  Influence of the threshold overload ratio on the

probability of failure.

traffic growth ratio of 2 percent of RADTT and RGVW the probability of failure
decreased to 4.5 × 10−8 and 9.4 × 10−9, respectively.
The following interpretations can be made based on the numerical
results. First, the maximum traffic load effect is associated with the traffic
density and the GVW. Therefore, the intensive overloaded trucks in the
mid-span point of the bridge forms the critical loading scenario resulting in
the first-passage failure of the deformation of the cable-stayed bridge. The
growth rate of the ADTT increases the probability of exceedance due to the
intensive overloaded traffic flows, and then results in a higher probability
60   •   BRIDGE RELIABILITY AND SERVICEABILITY

of failure. However, the increase rate of the probability of failure is weak-

ening with the continuous increase of ADTT. The increase of GVW has a
constant impact on the probability of failure. This impact can be weakened
by setting a threshold overload ratio for the overloading management.
It’s definitely not enough to only use the deflection of the bridge
girder to represent the entire extreme traffic load effects of a cable-stayed
bridge. The deflection is only a serviceability factor of a long-span bridge.
For the load bearing limit state, some other factors including the cable
force, the bending moment of the towers or the girders are critical as the
extreme load effects. Therefore, further studies are necessary to develop
the proposed methodology for the first-passage probability evaluation.

3.5 CONCLUSIONS

This study presented a methodology for the first-passage probability eval-

uation of a long-span bridge subject to heavy traffic loading based on WIM
measurements. Stochastic heavy traffic loads were simulated based on the
statistical characters via utilizing the WIM measurements. The effective-
ness of the proposed computational framework was demonstrated by a
case study of a cable-stayed bridge. The following conclusions can be
drawn from the case study.

(a) The upper tail fitting of the up-crossing rate is an appropriate

description of probability characteristics of the extreme traffic load
effect, which is the critical factor for the maximum displacement
prediction and the first-passage probability evaluation of a long-
span bridge;
(b) the ADTT growth increases the probability of exceedance due
to intensive overloaded traffic flows, and thus leads to a higher
first-passage probability, but this increase trend is weakening with
the continuous increase of ADTT;
(c) Since the steady growth of the gross vehicle weight has a significant
impact on the probability of failure, setting a reasonable threshold
overload ratio is an effective way as a traffic management to ensure
the bridge serviceability.

Although the proposed approach has been demonstrated by a cable-

stayed bridge, it can also be applied to any other types of long-span bridges.
However, there are challenges for further studies. A more reasonable traffic
growth rather than a linear growth factor should be considered in the future
 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •   61

work. The structural geometric nonlinearity could be considered to make

the simulation more accurate. However, the time-consuming computation
induced by the step-by-step intergradation should be overcome. Subsequently,
the other traffic load effects, such as the cable force, the bending moment of
the girder, and the axial force can also be considered instead of the bridge
deflection. Finally, the Passion distribution assumption in the Rice’s formula
will be improved for a more accurate first-passage probability evaluation.

REFERENCES

[1] Deng, L., W. Wan, and Y. Yu. 2015. “State-of-the-Art Review on the
Causes and Mechanisms of Bridge Collapse.” Journal of Performance of
Constructed Facilities 30, no. 2, p. 04015005.
[2] Chen, S., and J. Wu. 2011. “Modeling Stochastic Live Load for Long-span
Bridge Based on Microscopic Traffic Flow Simulation.” Computers &
Structures 89, no. 9, pp. 813–24.
[3] Liu, Y., N.W. Lu, and X.F. Yin. 2016. “A Hybrid Method for Structural
System Reliability-Based Design Optimization and its Application to
Trusses.” Quality and Reliability Engineering International 32, no. 2,
pp. 595–608.
[4] Liu, Y., N.W. Lu, X.F. Yin, and M. Noori. 2016. “An Adaptive Support
Vector Regression Method for Structural System Reliability Assessment and
its Application to a Cable-Stayed Bridge.” P I Mech Eng O-J RIS 230, no.
2, pp. 204–19.
[5] Li, J., H. Hao, and J.V. Lo. 2015. “Structural Damage Identification with
Power Spectral Density Transmissibility: Numerical and Experimental
Studies.” Smart Structures and Systems 15, no. 1, pp. 15–40.
[6] Han, W., J. Wu, and C.S. Cai. 2015. “Characteristics and Dynamic Impact of
Overloaded Extra Heavy Trucks on Typical Highway Bridges.” Journal of
Bridge Engineering 20, no. 2, p. 05044011.
[7] Ministry of Communications and Transportation (MOCAT). 2004. General
Code for Design of Highway Bridges and Culverts. JTG D60-2004. Beijing,
China.
[8] Li, J., H. Hao, and Z. Chen. 2015. “Damage Identification and Optimal Sensor
Placement for Structures Under Unknown Traffic-Induced Vibrations.”
Journal of Aerospace Engineering 30, no. 2, p. B4015001.
[9] Marques, F., C. Moutinho, W.H. Hu, Á. Cunha, and E. Caetano. 2016.
“Weigh-in-motion Implementation in an Old Metallic Railway Bridge.”
Engineering Structures 123, pp. 15–29.
[10] O’Brien, E.J., and B. Enright. 2012. “Using Weigh-in-motion Data to
Determine Aggressiveness of Traffic for Bridge Loading.” Journal of Bridge
Engineering 18, pp. 232–39.
62  •   BRIDGE RELIABILITY AND SERVICEABILITY

[11] Lu, N.W., M. Noori, and Y. Liu. 2016. “Fatigue Reliability Assessment
of Welded Steel Bridge Decks Under Stochastic Truck Loads via
Machine Learning.” Journal of Bridge Engineering. doi:10.1061/(ASCE)
BE.1943-5592.0000982
[12] Liu, Y., X.H. Xiao, N.W. Lu, and Y. Deng. 2016. “Fatigue Reliability
Assessment of Orthotropic Bridge Decks Under Stochastic Truck Loading.”
Shock and Vibration. doi:10.1155/2016/4712593
[13] Getachew, A., and E.J. Obrien. 2007. “Simplified Site-specific Traffic Load
Models for Bridge Assessment.” Structure and infrastructure Engineering 3,
no. 4, pp. 303–11.
[14] O’Brien, E.J., F. Schmidt, D. Hajializadeh, X. Zhou, B. Enright, C.C.
Caprani, S. Wilson, and E. Sheils. 2015. “A Review of Probabilistic Methods
of Assessment of Load Effects in Bridges.” Struct Saf 53, pp. 44–56.
[15] Enright, B., and E.J. O’Brien. 2013. “Monte Carlo Simulation of Extreme
Traffic Loading on Short and Medium Span Bridges.” Structure and
Infrastructure Engineering 9, no. 12, pp. 1267–82.
[16] Chen, X. 2014. “Estimation of Extreme Value Distribution of Crosswind
Response of Wind-Excited Flexible Structures Based on Extrapolation of
Crossing Rate.” Engineering Structures 60, pp. 177–88.
[17] AASTHTO. 2012. LRFD Bridge Design Specifications, 6th ed. Washington,
DC.
[18] O’Connor, A., and E.J. O’Brien. 2005. “Traffic Load Modelling and Factors
Influencing the Accuracy of Predicted Extremes.” Canadian Journal of Civil
Engineering 32, no. 1, pp. 270–78.
[19] Ruan, X., J. Zhou, X. Shi, and C.C. Caprani. 2016. “A Site-specific Traffic
Load Model for Long-span Multi-pylon Cable-stayed Bridges.” Structure
and Infrastructure Engineering. doi:10.1080/15732479.2016.1164724
[20] Song, J., and A. Der Kiureghian. 2006. “Joint First-passage Probability and
Reliability of Systems Under Stochastic Excitation.” Journal of Engineering
Mechanics 132, no. 1, pp. 65–77.
[21] Noori, M., M. Dimentberg, Z. Hou, R. Christodoulidou, and A. Alexandrou.
1995. “First-passage Study and Stationary Response Analysis of a BWB
Hysteresis Model Using Quasi-conservative Stochastic Averaging Method.”
Probabilistic Engineering Mechanics 10, no. 3, pp. 161–70.
[22] Chen, J.B., and J. Li. 2007. “The Extreme Value Distribution and Dynamic
Reliability Analysis of Nonlinear Structures with Uncertain Parameters.”
Structural Safety 29, no. 2, pp. 77–93.
[23] Khan, R.A., T.K. Datta, and S. Ahmad. 2005. “Reliability Analysis of Fan
Type Cable Stayed Bridges Against First Passage Failure Under Earthquake
Forces.” Journal of Seismology and Earthquake Engineering 7, no. 3,
pp. 147–57.
[24] Balafas, K., and A. Kiremidjian. 2015. “Reliability Assessment of the
Rotation Algorithm for Earthquake Damage Estimation.” Structure and
Infrastructure Engineering 11, no. 1, pp. 51–62.
 FIRST-PASSAGE PROBABILITY OF THE DEFLECTION   •  63

[25] Lydon, M., S.E. Taylor, D. Robinson, A. Mufti, and E.J. Brien. 2016.
“Recent Developments in Bridge Weigh in Motion (B-WIM).” Journal of
Civil Structural Health Monitoring 6, no. 1, pp. 69–81.
[26] Liu, Y., Y. Deng, and C.S. Cai. 2015. “Deflection Monitoring and Assessment
for a Suspension Bridge Using a Connected Pipe System: A Case Study in
China.” Structural Control and Health Monitoring 22, no. 12, pp. 1408–25.
[27] MOCAT. 2004. Limits of Dimensions, Axle Load and Masses for Road
Vehicles GB 1589-2004. Beijing, China: China Communications Press.
[28] Rice, S. 1944. “Mathematical Analysis of Random Noise.” The Bell System
Technical Journal 24, no. 1, pp. 46–156.
[29] Ditlevsen, O. 1994. “Traffic Loads on Large Bridges Modeled as White-
noise Fields.” Journal of Engineering Mechanics 120, no. 4, pp. 681–94.
[30] Cremona, C. 2001. “Optimal Extrapolation of Traffic Load Effects.”
Structural Safety 23, no. 1, pp. 31–46.
[31] MOCAT. 2007. Specifications for Design of Highway Cable-stayed Bridge
JTG/T D65-01. Beijing, China: China Communications Press.
 CHAPTER 4

Dynamic Reliability
Evaluation of the
Serviceability of Cable-
Supported Bridges
Under Site-Specific
Heavy Traffic Loads

4.1 INTRODUCTION

Flexible girders of long-span cable-supported bridges are sensitive to

the simultaneous presence of multiple vehicles that is different from
short-span bridges. In general, a deflection threshold is specified in
a bridge design code to ensure the serviceability of the bridge. For
instance, the critical deflection for suspension bridges specified in the
MOCAT [1] is L/300, where L is the mid-span length of the bridge.
This is defined as the serviceability failure criteria for a bridge while
the girder deflection under the traffic load crosses the threshold [2]. In
the design phase of a bridge, the stiffness of the bridge should satisfy
the deflection requirement under the design live load. However, with
the steady growth of traffic flow and the appearance of the extremely
overloaded trucks, the design live-load model may undervalue the cur-
rent traffic load effect. Han et al. [3] indicated that 4 out of 1,319 trucks
in China cause larger bending moments of short-span bridges than the
design value. In a­ ddition to the direct load effects on bridges, the over-
weight trucks lead to the accelerated deterioration of the road surface
roughens [4]. Even through the impacts of moving vehicles on flexible
66  •   BRIDGE RELIABILITY AND SERVICEABILITY

long-span bridges are negligible, a poor road roughness condition (RRC)

and overloaded trucks may result in a severe vehicle–bridge interaction
(VBI). In addition, from a probability point of view, the dynamic behav-
ior of cable-supported bridge may influence the probabilistic extreme
load effect under the stochastic traffic loads.
The VBI is a major factor leading to the amplification of the traffic
load effect on bridges, especially for the bridges that have girders with
larger stiffness. The current design codes [5, 6] and most researchers
[7, 8, 9] in this area conventionally utilize the dynamic amplification
factor (DAF) and the dynamic load allowance (DLA) to consider the
impact effects on the VBI interaction system in a critical scenario. In
addition, probabilistic approaches have also been developed for descript-
ing the DAF in probability domain. For instance, a set of cumulative
distribution functions of DAF was developed by Kim and Nowak [10].
An assessment dynamic ratio (ADR) was introduced by Caprani [11] to
define the relationship between the characteristic dynamic traffic load
effect and the characteristic static traffic load effect in a reference period.
Subsequently, O’Brien et al. [12] implemented the ADR as a lifetime
dynamic allowance value to investigate the dynamic impact of traffic
loading on short- to medium-span concrete bridges. The Gumbel distri-
bution was demonstrated by Deng and Cai [13] to be the best fit to the
probability model of DAFs of a wide range of bridges. Caprani et al. [14]
utilized multivariate extreme value theory and the ADR to investigate
the dynamic allowance of highway bridges under long-term traffic load-
ing. In addition to the investigations on the probabilistic DAF, reliability
approaches on the VBI have also been implemented by several research-
ers. In this regard, Xiang et al. [15] utilized the first-order reliability
method to evaluate the dynamic reliability of a vehicle–bridge coupling
vibration system. Zhang et al. [16] estimated fatigue reliability of a long-
span bridge with consideration of traffic–bridge interaction. Rocha et al.
[17] presented a probabilistic methodology for safety assessment of
short-span railway bridges. Ettefagh et al. [18] investigated uncertainties
in a stochastic vehicle–bridge coupled system. These studies illustrated
previously mostly emphasized the probabilistic dynamic effects of the
VBI on the short- to medium-span bridges, while the cable-supported
bridges were less discussed.
In addition to dynamic characteristics, simultaneous presence of
multiple vehicles is another issue for long-span bridges compared with
short-span bridges. Therefore, a reasonable traffic load model is critical
for estimating the load effects on long-span bridges. Investigations on
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  67

implementing weigh-in-motion measurements for probabilistic extreme

value extrapolation or interpolation are important topics. In this respect
Miao and Chen [19] utilized WIM data in Hong Kong to develop a lane
and truck load model and compared that with several existing bridge
live-load models. The WIM measurements in Netherlands were utilized
by O’connor and O’brien [20] to extrapolate the maximum traffic load
effect of a short-span bridge in lifetime. Fu and You [21] developed a
reliability-based live load spectrum based on traffic measurements.
A great number of WIM records collected from five European sites with
consideration of overloaded or over-length vehicles were utilized by
O’brien and Enright [22] to simulate the maximum effect of short to
medium span bridges under free-flowing traffic loads. Ran et al. [23]
developed a site-specific traffic load model to evaluate the extrapola-
tion of extreme load effects of a multi-pylon cable-stayed bridge. These
studies have made great contributions to the probabilistic extreme traf-
fic load effect extrapolation of bridges based on WIM measurements.
The conventional approaches in this area are block maxima approaches,
including peaks over threshold and generalized extreme value (GEV)
distribution, and Rice’s level-crossing theory [24]. The level-crossing
theory utilizes the number of level-crossings to describe the probabil-
ity model of the extreme value. It is appropriate to be implemented for
maximum traffic load effect extrapolations of long-span bridges because
the traffic load effect histories are mostly continuous, which can be cap-
tured by the level-crossing approach rather than the GEV distribution.
However, the application of the level-crossing theory for probabilistic
load effect extrapolations of long-span bridges accounting for the VBI is
relatively insufficient.
The case study presented in this book aims at implementing the
level-crossing theory to evaluate the probabilistic extreme deflections
of cable-supported bridges under long-term dynamic traffic loads. The
dynamic traffic load effect is considered in a traffic–bridge interaction
system. Long-term dense traffic load effects are used for level-crossing
rate simulation. The bridge serviceability is assessed based on the prob-
abilistic extreme traffic load extrapolation and the probability of exceed-
ance of the predefined threshold. A concrete cable-stayed bridge and a
suspension bridge are selected as prototypes to investigate the deterio-
ration on the RRC on the maximum traffic load extrapolation and the
probability of exceedance of the threshold deflection. Influences of the
traffic growth and the overloaded trucks on the probabilistic extrapolation
are investigated.
68  •   BRIDGE RELIABILITY AND SERVICEABILITY

4.2 TRAFFIC–BRIDGE INTERACTION
FORMULATION

4.2.1 VEHICLE–BRIDGE INTERACTION SYSTEM

In general, a VBI system is conventionally modeled in two subsys-

tems including the vehicle system with the degree of freedoms (DOFs)
in physical coordinates and the bridge system with the DOFs in modal
coordinates of the concerned mode shapes. The vehicle model is usually
­simulated by several mass blocks, springs, and dampers. A 3D 2-axle truck
model as shown in Figure 4.1 can be simulated by 7 PDFs including the
two-direction rotations (θr1 and θr2) and the vertical translational motion
(Zv) of the vehicle rigid body, and vertical translational motions (Za1L, Za2L,
Za1R, and Za2R) of each vehicle wheel. The equation of motion of a vehicle–
bridge interaction system can be written as

Mb  u b   Cb + Cbb Cbv   u b   K b + K bb K bv   u b 

 +   +  

 M v  
u v   C vb C v   u v   K vb K v   u v 
 Fbr 
= G 
(4.1)
F +
 vr V 
F

where Mv, Cv, and Kv are the mass, the damping and the stiffness matrices
of the vehicle, respectively; Mb, Cb, and Kb are the mass, the damping and
the stiffness matrices of the bridge, respectively; uv and ub are the dis-
placement vectors of the vehicle and the bridge, respectively; Cbb, Cbv, Cvb,
Kbb, Kbv, Kvb, Fbr, and Fvr are the additional terms caused by the expansion
of the wheel-road contact force. All of these time-variant parameters will
change with the moving vehicle on the bridge.
The bridge behavior can be modeled linearly. That allows using
the modal superposition to reduce the number of DOFs of the system.
It is acknowledged that {Φ i }T [ M b ]{Φ i } = 1 , {Φ i }T [ K b ]{Φ i } = wi2 , and
[C b ] = 2wi hi [M b] , where ωi is the frequency of the ith mode shape of the
bridge, ηi is the damping ratio for the ith mode of the bridge. Therefore,
Equation (4.1) can be written in frequency domain:

I  xb   2 É i · i I + ΦTb Cbv  

ΦTb Cbv  ¾
 +   +
b
 M v  
 u v   Cvb Φ b Cv  u v 
ΦTb K bv  xb  Φ b Fbr 
T
ω i2 I + ΦTb K bb Φ b (4.2)
   =  
 K vb K v  u v  Fvr + FVG 
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  69

D1 D2

Zv

qr1

K1uzL C1uzL K2uzL C2uzL

Za1L Za1L

K1lzL C1lzL K2lzL C2lzL

x

(a)

Zv

qr2

K1uzL C1uzL K1uzR C1uzR

Za1L Za3R

K1lzL C1lzL K1lzR C1lzR

y

(b)

Figure 4.1.  Physical model of a 2-axle truck: (a) elevation

view; (b) side view.

where I is a unit matrix, ξb is the generalized coordinate vector of the

bridge, and the bridge deflection vector and can be written by using the
superposition u b = [Φ b ]{xb } .

4.2.2 SIMPLIFIED TRAFFIC–BRIDGE INTERACTION

FORMULATION

For long-span bridges, the simultaneous presence of multiple vehicles is a

unique phenomenon different from short-span bridges. Traffic–bridge inter-
action (TBI) is more complex than a single VBI model, because each vehicle
will impact the equation of motion [25]. However, investigations conducted
by Cai and Chen [7] suggested that the interaction effects between different
vehicles on a flexible long-span bridge are insignificant. Therefore, Chen
and Cai [26] indicated that the dynamic interaction between these vehicles
70   •   BRIDGE RELIABILITY AND SERVICEABILITY

may become negligible, where the traffic–bridge interaction system can be

divided into several independent VBI subsystems without losing accuracy.
Subsequently, Chen and Cai [25] presented an equivalent dynamic wheel
loading (EDWL) approach to simplify the traffic–bridge interaction analy-
sis. The EDWL approach utilizes several moving time-varying loads eval-
uated from a fully coupled vibration analysis of a VBI system to represent
the actual moving vehicles on the bridge [27, 28].
A dynamic load ratio describing the ratio between the EDWL and the
weight of a vehicle is the critical factor to utilize the EDWL approach.
The dynamic load ratio is associated with the physical parameters of the
vehicle and the bridge dynamics. Based on the earlier illustration, the fully
coupled TBI system can be written as

u b + Cb u b + K b u b = Feqwheel (4.3)
M b 

 
nv nb

{ } (4.4)
Feqwheel (t ) = ∑  1 − R j (t )  G j ⋅ ∑ hk [ x j (t ) + ak [ x j (t )d j (t )]]
j =1  k =1 

where Feqwheel is the time-varying force vector on the bridge; Rj and Gj are the
dynamic load ratio and the gross vehicle weight of the jth vehicle; xj, and
dj are the longitudinal and transverse location of the gravity center of the
jth vehicle, respectively; hk and ak are the vertical and the torsional mode
shapes for the kth mode of the bridge model, respectively; nv and nb are the
number of vehicles and the number of adopted bridge modes, respectively.
It is worth observing that the road surface roughness condition is consid-
ered in the relative displacement. The effectiveness of the EDWL approach
has been demonstrated by Chen and Wu [4] with several applications.
It should be pointed out that the total number of vehicles changes with
time depending on the density of the stochastic traffic flow. Nevertheless,
it can be assumed to be stationary for dense traffic flow, and will be con-
sidered in the present study. With the EDWLs, the dynamic load effect of
the bridge under traffic loads can be efficiently calculated.

4.3 METHODOLOGY OF PROBABILISTIC
MODELING OF THE EXTREME
TRAFFIC LOAD EFFECTS

4.3.1 RICE’S LEVEL-CROSSING FORMULA

As mentioned in the introduction, the conventional approach to simulate

the extreme traffic load effects is associated with the GEV distributions.
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •   71

However, the accuracy of the approach mostly depends on the number of

block maxima where only a maximum value in a large number of data of
a block is adopted. Obviously, the GEV approach ignores most of the data
that are simulated in a time-consuming computation for long-span bridges.
Therefore, this study utilizes Rice’s level-crossing formula [29] instead
of the traditional GEV approach to model the extreme traffic load effects.
The level-crossing theory utilizes the number of crossings as shown
in Figure 4.1 to define a probability model. The essence of implement-
ing the level-crossing theory is the Gaussian and stationary assumption of
the traffic load effect process. Diletvson [30] has demonstrated the traffic
load effect on long-span bridges can be assumed as a stationery Gaussian
process or even a white noise because the structural influence line is long
enough compared to the length occupied by individual vehicles. On the
basis of Rice’s level-crossing theory, the level-crossing rate under the
condition of a threshold and a reference period is expressed as [31]:

s′  ( x − m) 2 
v( x) = exp  −  (4.5)
2ps  2s 2 

where x is a random process, m and σ are the mean value and standard
deviation of x, respectively, and σ’ is the standard deviation of the deriv-
ative of x.
In practice, the v(x) can be fitted to histograms of number of cross-
ings in a unit time as shown in Figure 4.2(a, b). By dividing the entire
nonstationary process into several stationary interval processes, the
­level-crossing rate of original process is supposed as a superposition of
the interval processes in proportions. Based on this assumption, the kernel
coefficients can be evaluated by a second-order polynomial function:

Optimal starting point

V (x) V (x)
Up crossing
Rice’s fitting

x
x
Down crossing
Interval length
(a) (b)

Figure 4.2.  Basis of Rice’s level-crossing theory: (a) crossings; (b) fitting to

the crossings.
72  •   BRIDGE RELIABILITY AND SERVICEABILITY

 m p N ( x) 
ln [v ( x) ] = ln  ∑ i i  = a0 + a1 x + a2 x (4.6)
2

 i =1 T 
where v ( x) is an estimated equivalent level-crossing rate of the nonsta-
m2 m 1
tionary process, a0 = ln(v0 ) − 2 , a1 = 2 , and a2 = are second-­
2s s −2s 2
order polynomial coefficients that can be evaluated based on the histograms
of number of crossings.
For the purpose of extreme value extrapolation, the critical step of
the fitting is to find the optimal starting point and the number of inter-
vals. A starting point close to the tail is better for interpolation, while that
far from the tail is better for the extrapolation. A Kolmogorov–Smirnov
test suggested by Cremona [31] is an effective approach to optimize the
starting point. Based on the optimal starting point and class intervals, the
maximum value in a return period can be written as

xmax ( Rt ) = mopt + sopt 2 ln(v0, opt Rt ) (4.7)

where Rt is the return period, xmax is the maximum load effect correspond-
ing to the return period, mopt, σopt and v0,opt, represent the optimal mean
value, optimal standard derivation and optimal crossing rate, respectively.
In addition to the maximum value extrapolation based on the
level-crossing formula, first-passage failure probability evaluation is
another important measure for the assessment of bridge reliability under
traffic loads. First-passage failure is the best description of stochastic pro-
cess crossing the prescribed threshold during an interval time. Based on
the Rice’s level-crossing theory, the probability of failure, Pf , can be esti-
mated by the assumption of a Poisson distribution:

Pf (a, Ts ) = A exp  − ∫ vi (a ) d t  ≅ exp  −Ts v (a )  (4.8)

Ts

 0 

where A is a coefficient associated with the stationarity of a random pro-

cess and can be assumpted as 1 for a stationary process, Ts is usually the
lifetime of a bridge, and a is the threshold value of the bridge, such as
L/400 for the deflection limit of cable-stayed bridges [32].

4.3.2 A COMPUTATIONAL FRAMEWORK FOR

EXTRAPOLATION OF MAXIMUM TRAFFIC LOAD
EFFECTS ACCOUNTING FOR THE VBI

It is acknowledged that the VBI leads to the fluctuation of the bridge

responses. Such fluctuation impacts the level-crossing rate of the extreme
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  73

traffic load effect. In order to take into account the VBI induced traffic
load effect, a computational framework is presented integrating the VBI
and the level-crossing theory to extrapolate the maximum traffic load
effect of long-span bridges.
Flow chart of the computational framework is shown in Figure 4.3.
As depicted in Figure 4.3, the flow chart starts with the statistics of the
WIM measurements. The daily traffic flow can be simulated based on the
statistics of the WIM data. Subsequently, we can identify the critical traffic
loading scenario via the structural influence lines. This loading scenario
is used to compute the dynamic analysis of the vehicle–bridge coupled
vibration system. It is worth mentioning that the effect of vehicle spacing
in traffic density is a significant factor impacting the traffic load effects
on medium- to long-span bridges. The truck spacing instead of vehicle

WIM measurements

Statistics
Stochastic daily traffic loads
Structural influence lines

A daily critical traffic loading scenario

Modular procedures :

Bridge mode shapes and frequencies

RRC EDWL
Convert truck weights into time-
variant moving loads

Transient analysis
Dynamic load effect histories

No
Enough number of days?

Yes
Level-crossing fitting to the
histograms of the number of crossings
Return period
Extrapolation of the maximum traffic
load effects

Figure 4.3.  Flow chart of the computational

­framework ­computational for extrapolation of
maximum traffic load effects.
74  •   BRIDGE RELIABILITY AND SERVICEABILITY

spacing can be utilized to simulate the stochastic traffic flow on a bridge.

The truck spacing in this study denotes the gap between two following
trucks in a traffic lane, which is different to the vehicle spacing. In defin-
ing the vehicle gap between two trucks in an actual traffic stream, a large
number of light cars (GVW < 3 t) were removed.
With the critical traffic loading scenarios, the bridge dynamic
responses can be computed using the EDWL approach with consideration
of the road roughness condition (RRC). Since the modal superposition
approach is involved in the EDWL approach, the structural nonlinearities
cannot be considered in the proposed approach. The time-histories of the
dynamic analysis instead of a constant daily maximum can provide more
statistical information for modeling the level-crossing rate. The dynamic
history provides a way to account for the number of crossings due to
the VBI.
Eventually, the level-crossing rate can be fitted to the histograms of
the number of level crossings. Given a return period, the maximum traf-
fic load effect can be estimated based on Equation (4.7). In addition to
the extrapolation of the characteristic traffic load effect, the probability of
exceedance of the limit value can be evaluated from the CDF of the maxi-
mum traffic load effect as shown in Equation (4.8). It is worth to note that
the effectiveness of the method depends on the number of days for count-
ing the number of crossings. In general, 1000-day simulations are enough
for fitting an effective probabilistic extreme value model. The highlight of
the computational framework is the integration of the VBI analysis and the
Rice’s level-crossing evaluation.

4.4 PROBABILISTIC MODELING OF THE TRAFFIC

LOAD EFFECTS ON TWO CABLE-SUPPORTED
BRIDGES USING WIM MEASUREMENTS

4.4.1 STOCHASTIC TRAFFIC LOAD SIMULATION BASED ON

WIM MEASUREMENTS

A WIM system utilizes scales or pressure sensors embedded into the road
pavement to measure physical properties of passing vehicles such as axle
weight and speed. Axles of travelling vehicles interrupt the magnetic signal
produced by the loop sensors, and therefore, the number of axles, the axle
spacing and the number of vehicles are recorded. In addition to the traffic
management, WIM can also provide a great amount of real traffic data
through loop sensors and piezo sensors. Site-specific WIM measurements
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  75

provide a reliable statistical database to capture the practical behavior of

traffic flows. The WIM data in China were utilized in the present study to
simulate the traffic flow and the critical daily traffic loading scenario that
impact to the maximum traffic load effects.
The WIM measurements in the present study were obtained from a
highway bridge in China. Details of the measurement can be found by
Liu et al. and Lu et al. [33, 34]. Before utilizing these data, filtering pro-
cesses were conducted to remove the invalid and useless data. The crite-
ria utilized were: (1) the GVW was less than 30kN; (2) the axle weight
was greater than 300kN or less than 5kN; and (3) the vehicle length was
more than 20m. The overview of the effective measurement is shown in
Table 4.1. According to the vehicle configuration, all vehicles collected
from the WIM data are classified as six categories.
Probability densities of the vehicle spacing and the axle weights of
the GVW of the 6-axle truck are shown in Figure 4.4. It is observed that
the vehicle velocity of the 6-axle truck follows a normal distribution,
and the GVWs were well fitted by a bimodal Gaussian mixture model
(GMM). In addition to these small-scale parameters, the vehicle spacing
was also evaluated by converting the headway time into the spacing with
consideration of the average driving speed. The other statistical analysis
can be found in Liu et al. [33]. It is worth noting that the vehicle density
mostly depends on vehicle spacing. In addition, most of the extreme val-
ues, which will be used for probabilistic modeling of the extreme value,
are caused by dense traffic flows. Therefore, the vehicle spacing of dense
traffic flow is effective for traffic simulation.
In order to identify the critical traffic loading scenario, a step by step
search approach was utilized based on the static influence lines. A critical
loading scenario is shown in Figure 4.5, where bidirectional four traffic
lanes are involved.

Table 4.1.  Overview of WIM measurements

Items Values
Time period May 1, 2013 to April 30, 2015
Number of recording days 729
Average daily truck traffic 2145
Traffic lanes 4
Maximum GVW (t) 164
Number of overloaded trucks 12, 252
76  •   BRIDGE RELIABILITY AND SERVICEABILITY

1
Empirical data
Lognormal
Probability density 0.8

0.6

0.4
m=6.21
0.2 s=1.40

0
0 1000 2000 3000 4000
(a) Vehicle spacing (m)

0.02
WIM data
GMM
0.015
Probability density

a1=0.24 a2=0.76
0.01 m1=390 m2=865
s1=74 s2=142
0.005

0
0 500 1000 1500 2000
(b) GVW (kN)

Figure 4.4.  Histograms and PDFs of the 6-axle truck: (a) vehicle

spacing (b) GVWs.

V1 V2 V3 V4 V5 V6
150
GVW (tonne)

100

50

0
0
200
400
600 4
800 3
2
Location (m) 1 Traffic lane

Figure 4.5.  A critical traffic loading scenario identified from the

stochastic traffic-load model.
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  77

4.4.2 PROBABILISTIC TRAFFIC LOAD EFFECT

ESTIMATION

A concrete cable-stayed bridge and a steel suspension bridge are chosen

as prototypes to demonstrate the application of the proposed methodology
in the serviceability evaluation of the bridge deflections. Dimensions and
the first order mode shapes of the two bridges are shown in Figure 4.5.
The cable-stayed bridge has 68 concrete girders with segmental length
of 6 m in mid-span and 34 pairs of stay cables in the fan pattern. The
suspension bridge has 64 steel box-girders with segmental length of
12.8 m in the mid-span. It is obviously observed from Figure 4.6 that the
first-order mode shape of the cable-stayed bridge is symmetric and that of
the suspension is asymmetric.
The RRC in this study was defined by the RRC coefficient recom-
mended by the International Organization for Standardization (1995).
The RRC coefficient for classifications of “Good” with the RRC of
128×10−6 was used in this study. Inverse Fourier transform approach [35]
was utilized to simulate the RRC in time domain. The physical properties
of the truck in the case study were from Yin et al. [36]. The bridge verti-
cal deflections under one-hour dense traffic load are shown in Figure 4.7.
As observed from Figure 4.7, following results are derived: (a) for the

210
m

420
m

210
m
(a)

820
m

(b)

Figure 4.6.  The first vertical mode shapes and dimensions:

(a) cable-stayed bridge; (b) suspension bridge.
78  •   BRIDGE RELIABILITY AND SERVICEABILITY

0.3
L/4
0.2 L/2

0.1
Deflection (m)

-0.1

-0.2

-0.3

-0.4
0 10 20 30 40 50 60
Time (min)
(a)
0.4
L/4
0.2 L/2
Vertical displacement (m)

-0.2

-0.4

-0.6

-0.8

-1
0 10 20 30 40 50 60
Time (min)
(b)

Figure 4.7.  Deflection histories of the critical points under one-hour

dense traffic loading: (a) cable stayed bridge; (b) suspension bridge.

cable-stayed bridge, the mid-span (L/2) point has a severe vibration in

comparison with the quarter-span (L/4) point; (b) for the suspension
bridge, however, the quarter-span point has a more sever vibration than
the mid-span point; (c) the suspension bridge has relatively high displace-
ments, compared to the cable-stayed bridge. The first and second phe-
nomenon can be explained by the vertical first-order fundamental mode
shape of the two types of bridge as shown in Figure 4.8, where the cable-
stayed bridge is symmetric, but the suspension bridge is asymmetric.
In order to find out the in-depth difference of the traffic-induced
deformations between the two bridges in probability domain, a 100-
day traffic flow were used to carry out the statistical analyses. The
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  79

0.1
Cable-stayed bridge
Suspension bridge
0

Mean value (m) -0.1

-0.2

-0.3
0 L/4 L/2 3L/4 L
(a) Longitudinal location of the girder (m)

0.5
Cable-stayed bridge
Suspension bridge
0.4

0.3
RMS (m)

0.2

0.1

0
L/4 L/2 3L/4 L
(b) Longitudinal location of the girder (m)

Figure 4.8.  Statistical characteristics of the two cable-supported

bridges: (a) the mean value; (b) the standard deviation.

mean values and the standard derivations of these deformations were

estimated as shown in Figure 4.8. It is observed that The RMS of the
cable-stayed bridge follows a C-type distribution along the girder,
while the RMS of the suspension bridge follows an M-type distribu-
tion. This means both of the mean values are symmetric, but the stan-
dard deviations are asymmetric. Therefore, the deflection-critical point
for the cable-stayed bridge is the mid-point and that for the suspension
bridge is the quarter-point. The following investigations focus on the
two critical points.
With the statistics of the traffic load effects, it is feasible to conduct
the probabilistic extreme traffic load extrapolation.
80   •   BRIDGE RELIABILITY AND SERVICEABILITY

Initially, the histograms of the number of level crossings of the critical

points were counted from the deflection histories as shown in Figure 4.9. The
normalized up-crossing rates of the two critical points of the two bridges are
shown in Figure 4.10, where the one corresponding to the suspension bridge
has a quite higher level-crossing value. With the normalized level-cross-
ing rates, the maximum traffic load effects were extrapolated based on the
Rice’s formula as shown in Equation (4.7). Given a return period of a 1000-
year period, the maximum traffic load effects were extrapolated as shown
in Figure 4.11. It is observed that the maximum deflection for the critical
points of the cable-stayed bridge and suspension bridge are 1.05 and 1.69 m,
respectively. Herein, by utilizing Rice’s level-crossing theory, the statistical
extrapolation of the extreme traffic load effect is obtained.

0.1
Static
Dynamic
0
Deflection (m)

-0.1

1st sample 2nd sample

-0.2 Ns,1 = 2 Ns,2 = 1

Nd,1 = 6 Nd,2 = 4
-0.3

-0.4
0 21 42 63 84
Time (s)
(a)
0.5
Static
0.25 Dynamic

0
Deflection (m)

1st sample
Ns,1 = 1
-0.25
Nd,1 = 2 2nd sample
-0.5 Ns,2 = 0
Nd,2 = 2
-0.75

-1
0 20.5 41 61.5 82
(b) Time (s)

Figure 4.9.  Deflection histories of the critical traffic loading on the:

(a) cable-stayed bridge; (b) suspension bridge.
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •   81

250
The cable-stayed bridge The suspension bridge

200

Annual number of crossings

Rice's fitting:
Rice's fitting:
x0, opt = 0.462 m x0, opt = 0.903m
150 v0 = 161
v0 = 145
m = 0.257 m
m = 0.752 m
s=0.184 m s=0.273 m
100

50

0
0.4 0.6 0.8 1 1.2 1.4
Deflection (m)

Figure 4.10.  Histograms and Rice’s fittings of the number of level

crossings. 

2
Cable-stayed bridge
Suspension bridge
1.7
Maximum deflection (m)

1.4

1.1

0.8

0.5
0 200 400 600 800 1000
Return period (years)

Figure 4.11.  Extrapolation of the extreme traffic load deflections.

4.5 DYNAMIC RELIABILITY EVALUATION OF THE

BRIDGE DEFLECTION SERVICEABILITY

4.5.1 DYNAMIC RELIABILITY EVALUATION OF THE

DEFLECTION SERVICEABILITY

Bridge deflection serviceability limitation under vehicle load effect is a

condition to control the undesirable vibrations of a bridge without causing
severe damages. With the established probability model of the extreme
82  •   BRIDGE RELIABILITY AND SERVICEABILITY

Table 4.2.  Bridge deflection limits

Materials of the
Bridge types bridge girders Deflection limits
Simply supported bridges Concrete or steel L/800
Cantilever bridge Concrete or steel L/300
Cable-stayed bridges Concrete L/500
Steel L/400
Suspension bridges Steel L/350
Note: L is the bridge span length.

load effect, structural serviceability reliability can be estimated with a

predefined deflection limit. Several design specifications have defined
the limit deflection of multiple bridges. The threshold of traffic-induced
bridge displacement in the design specifications of China (MOCAT 2015)
is selected. These criterions are summarized in Table 4.2.
In the present study, the material of the cable-stayed bridge girder
is concrete, and the material of the suspension bridge girder is steel.
Therefore, the deflection limits for the cable-stayed bridge and the sus-
pension bridge are L/500 = 0.84 m and L/300 = 2.34 m, respectively. The
first-passage failure criterion as shown in Equation (4.8) was utilized to
evaluate the dynamic reliability of the deflection serviceability of the two
cable-supported bridges. The critical points of the two bridges are the
quarter-point and the mid-point of the bridge girders. The probability of
failure of each bridge was calculated according to Equation (4.11), and the
result is shown in Figure 4.13.
It is observed from Figure 4.13 that the suspension bridge has a higher
probability of failure compared with the cable-stayed bridge, even though
the deformation threshold for cable-stayed bridge is higher than the sus-
pension bridge as mentioned earlier. The tendency of probability of failure
shown in Figure 4.12 is similar to the extrapolation of extreme load effects
shown in Figure 4.11.

4.5.2 PARAMETRIC STUDIES

Since the traffic volume will grow with the development of global econ-
omy, the average daily truck traffic (ADTT) was selected herein for para-
metric studies. The ADTT is usually affected by numerous factors, and
is difficult to predict for the entire service period of a bridge. This study
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  83

10-4
Cable-stayed bridge
Suspension bridge

Probability of exceedance
10-5

10-6

10-7

10-8
0 20 40 60 80 100
Service period (years)

Figure 4.12.  Probability of failure of the critical girder duo to

­displacement up-crossing.

10-1
Cable-stayed bridge
Suspension bridge
10-2
Probability of exceedance

10-3

10-4

10-5

10-6
0 1% 2% 3% 4%
Annual linear growth rate of ADTT

Figure 4.13.  Influence of growth rate of ADTT on the probability

of failure.

assumes the annual linear growth rate of the ADTT to be 0, 1, 2, 3 and

4 percent. This hypothesis adopts the assumption that the ADTT rapidly
grows to the objective value rather than a gradual growth. Subsequently,
the traffic growth was considered by decreasing the vehicle spacing in
the stochastic traffic model. Influences of the growth rate of ADTT on the
probability of first-passage failure of the bridge are shown in Figure 4.13.
It is observed from Figure 4.13 that the probability of failure of
the two bridges has a higher decrease rate at the initial stage of traffic
growth, and this increase rate slows down with the continued growth of
84  •   BRIDGE RELIABILITY AND SERVICEABILITY

10–1
Cable-stayed bridge
Suspension bridge
10–2
Probability of exceedance

10–3

10–4

10–5

10–6
Unlimited 100% 75% 50% 25%
Overload rate limit

Figure 4.14.  Influence of the threshold of overloading ratio

on the probability of failure.

ADTT. In addition, the growth rate of ADTT enlarges the gap between
the two values of the probability of failure. This demonstrates that a more
realistic traffic growth model is still important for the bridge reliability
assessment.
In addition to the traffic growth, traffic managements will also impact
the traffic flow. In general, an overload control measure will lead to a pos-
itive result. In order to study the influence of the overload control measure
on bridge serviceability, the threshold of overloading ratio is assumed to
be 25, 50, 75 and 100 percent. The thresholds of GVW for the 2-axle
and 6-axle trucks specified in the MOCAT (2004) are 200 kN and 550
kN, respectively. Based on the above assumption, the stochastic traffic
flow load model was updated with the defined overloading ratio, and
then the up-crossing rate and the probability of failure were re-evaluated.
Figure 4.14 shows the influence of the threshold of overloading ratio on
the probability of failure of the bridge.
As observed from Figure 4.14, the threshold of overloading ratio has
a remarkable influence on decreasing the structural probability of failure.
Even for a threshold of 100 percent, the probability of failure has a rapid
decrease in comparison with the unlimited state. In the case of gradual
traffic volume growth, the overload control measure is an effective way to
decrease serviceability risk of flexible cable-supported bridges subject to
long-term traffic loading.
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  85

4.6 CONCLUSIONS

The study presented in this chapter utilized a stochastic traffic flow load
model to estimate the reliability-based serviceability of cable-supported
bridges subject to long-term heavy traffic loading. The heavy traffic flow
was simulated based on site-specific weigh-in-motion measurements via a
cellular automation model. Subsequently, the deterministic load effect and
probabilistic modeling the extreme value were computed via the traffic–
bridge coupled vibration theory and via Rice’s level-crossing theory,
respectively. The case study of two typical cable-supported bridges has
demonstrated the effectiveness of the proposed computational framework.
The conclusions are summarized as follows:

1. The stochastic traffic load model, simulated by the site-specific

weight-in-motion measurements, is appropriate to be utilized to
assess reliability-based serviceability of cable-supported bridges.
Integration of the stochastic traffic load model and the correspond-
ing computational framework provide an effective approach for
reliability evaluation of flexible cable-supported bridges based on
site-specific traffic measurements.
2. Due to the difference between the structural first-order mode shapes
of the two types of bridge, the root-mean-square of load effects
along the bridge girder are different. The root-mean-square dis-
placement of the cable-stayed bridge follows a C-type distribution,
while the one for the suspension bridge follows an M-type distribu-
tion. This means the critical points for the cable-stayed bridge and
the suspension bridge under dense traffic loading are mid-span and
quarter-span points, respectively.
3. The probability of failure decreases rapidly with the initial stage
of the traffic growth, but the increase rate gradually slows down
with the continued growth of traffic volume. In addition, the growth
of traffic volume enlarges the gap between the probability of failure
of the two bridges.
4. Since the extreme traffic load effect is mostly caused by overloaded
trucks, the probability of first-passage failure has a considerable
decrease in spite of taking the threshold of overloading ratio as 100
percent for the traffic control.

In addition to the application for cable-supported bridges, the stochas-

tic traffic flow load model and the proposed computational framework can
also be applied to any other flexible long-span bridges. However, some
86  •   BRIDGE RELIABILITY AND SERVICEABILITY

challenges still remain to be addressed in future work. First, since the prob-
abilistic modeling of traffic load effects needs a large number of data, the
time-consuming traffic–bridge interaction analysis is still a bottleneck for
the computational efficiency. Second, utilizing the critical loading scenario
in the daily traffic flow, instead of the entire dense traffic flow, will save a
considerable amount of time. Third, the probability model and extrapolation
of the extreme traffic load effect should be compared with other approaches,
such as the generalized extreme value distribution and the peaks over value
approach. Finally, as aforementioned, a more realistic traffic growth model
should be considered instead of an idealized traffic growth model.

REFERENCES

[1] Ministry of Communications and Transportation (MOCAT). 2015. “General

Code for Design of Highway Bridges and Culverts.” JTG D60-2015, Beijing,
China.
[2] Jung, K.H., J.W. Yi, and J.H.J. Kim. 2010. “Structural Safety and
Serviceability Evaluations of Prestressed Concrete Hybrid Bridge Girders
with Corrugated or Steel Truss Web Members.” Engineering Structures 32,
no. 12, pp. 3866–78.
[3] Han, W., J. Wu, C.S. Cai, and S. Chen. 2014. “Characteristics and Dynamic
Impact of Overloaded Extra Heavy Trucks on Typical Highway Bridges.”
Journal of Bridge Engineering 20, no. 2, p. 05014011.
[4] Chen, S.R., and J. Wu. 2010. “Dynamic Performance Simulation of Long-
span Bridge Under Combined Loads of Stochastic Traffic and Wind.”
Journal of Bridge Engineering 15, no. 3, pp. 219–30.
[5] European Committee for Standardization (ECS). 2003. Actions on Structures.
Part 2: Traffic Loads on Bridges, Brussels.
[6] AASHTO. 2015. LRFD Bridge Design Specifications, 6th ed. American society
of civil engineers.
[7] Cai, C.S., and S.R. Chen. 2004. “Framework of Vehicle–bridge–wind
Dynamic Analysis.” Journal of Wind Engineering and Industrial
Aerodynamics 92, no. 7, pp. 579–607.
[8] Yang, Y.B., and C.W. Lin. 2005. “Vehicle–bridge Interaction Dynamics
and Potential Applications.” Journal of Sound and Vibration 284, no. 1,
pp. 205–26.
[9] Zhang, N., H. Xia, and W. Guo. 2008. “Vehicle–bridge Interaction Analysis
Under High-speed Trains.” Journal of Sound and Vibration 309, no. 3,
pp. 407–25.
[10] Kim, S., and A.S. Nowak. 1997. “Load Distribution and Impact Factors for
I-girder Bridges.” Journal of Bridge Engineering 2, no. 3, pp. 97–104.
 SERVICEABILITY OF BRIDGES UNDER TRAFFIC LOADING   •  87

[11] Caprani, C.C. 2005. “Probabilistic Analysis of Highway Bridge Traffic

Loading.” Ph.D. Thesis, School of Architecture, Landscape, and Civil
Engineering, Univ. College Dublin, Dublin, Ireland.
[12] OBrien, E., D. Cantero, B. Enright, and A. Gonzalez. 2010. “Characteristic
Dynamic Increment for Extreme Traffic Loading Events on Short and
Medium Span Highway Bridges.” Engineering Structures 32, no. 12,
pp. 3827–35.
[13] Deng, L., and C.S. Cai. 2010. “Development of Dynamic Impact Factor
for Performance Evaluation of Existing Multi-girder Concrete Bridges.”
Engineering Structures 32, no. 1, pp. 21–31.
[14] Caprani, C.C., A. González, P.H. Rattigan, and E.J. OBrien. 2012.
“Assessment Dynamic Ratio for Traffic Loading on Highway Bridges.”
Structure and Infrastructure Engineering 8, no. 3, pp. 295–304.
[15] Xiang, T., R. Zhao, and T. Xu. 2007. “Reliability Evaluation of Vehicle–
bridge Dynamic Interaction.” Journal of structural Engineering 133, no. 8,
pp. 1092–99.
[16] Zhang, W., C.S. Cai, and F. Pan. 2012. “Fatigue Reliability Assessment
for Long-span Bridges Under Combined Dynamic Loads from Winds and
Vehicles.” Journal of Bridge Engineering 18, no. 8, pp. 735–47.
[17] Rocha, J.M., A.A. Henriques, and R. Calçada. 2014. “Probabilistic Safety
Assessment of a Short Span High-speed Railway Bridge.” Engineering
Structures 71, no. 7, pp. 99–111.
[18] Ettefagh, M.M., D. Behkamkia, S. Pedrammehr, and K. Asadi. 2015.
“Reliability Analysis of the Bridge Dynamic Response in a Stochastic Vehicle-
bridge Interaction.” KSCE Journal of Civil Engineering 19, no. 1, pp. 220–32.
[19] Miao, T.J., and T.H. Chan. 2002. “Bridge Live Load Models from WIM
Data.” Engineering Structures 24, no. 8, pp. 1071–84.
[20] O’connor, A., and E.J. O’Brien. 2005. “Traffic Load Modelling and Factors
Influencing the Accuracy of Predicted Extremes.” Canadian Journal of Civil
Engineering 32, no. 1, pp. 270–78.
[21] Fu, G., and J. You. 2009. “Truck Loads and Bridge Capacity Evaluation in
China.” Journal of Bridge Engineering 14, no. 5, pp. 327–35.
[22] OBrien, E.J., and B. Enright. 2012. “Using Weigh-in-motion Data to
Determine Aggressiveness of Traffic for Bridge Loading.” Journal of Bridge
Engineering 18, no. 3, pp. 232–39.
[23] Ruan, X., J. Zhou, X. Shi, and C.C. Caprani. 2016. “A Site-specific Traffic
Load Model for Long-span Multi-pylon Cable-stayed Bridges.” Structure
and Infrastructure Engineering 13, no. 4, 494–504. doi:10.1080/15732479
.2016.1164724
[24] O’Brien, E.J., F. Schmidt, D. Hajializadeh, X. Zhou, B. Enright, C.C.
Caprani, S. Wilson, and E. Sheils. 2015. “A Review of Probabilistic Methods
of Assessment of Load Effects in Bridges.” Struct. Saf. 53, no. 3, pp. 44–56.
[25] Chen, S.R., and J. Wu. 2011. “Modeling Stochastic Live Load for Long-
span Bridge Based on Microscopic Traffic Flow Simulation.” Computers &
Structures 89, no. 9, pp. 813–24.
88  •   BRIDGE RELIABILITY AND SERVICEABILITY

[26] Chen, S.R., and C.S. Cai. 2007. “Equivalent Wheel Load Approach for
Slender Cable-stayed Bridge Fatigue Assessment Under Traffic and Wind:
Feasibility Study.” Journal of Bridge Engineering 12, no. 6, pp. 755–64.
[27] Deng, L., and C.S. Cai. 2010. “Development of Dynamic Impact Factor
for Performance Evaluation of Existing Multi-girder Concrete Bridges.”
Engineering Structures 32, no. 1, pp. 21–31.
[28] Deng, L., W. Wang, and Y. Yu. 2016. “State-of-the-art Review on the Causes
and Mechanisms of Bridge Collapse.” Journal of Performance of Constructed
Facilities 32, no. 2, 04045005. doi:10.1061/(ASCE)CF.1943-5509.0000731
[29] Rice, S. 1945. “Mathematical Analysis of Random Noise.” Bell System
Technical Journal 24, no. 1, pp. 46–156.
[30] Ditlevsen, O. 1994. “Traffic Loads on Large Bridges Modeled as White-
noise Fields.” Journal of Engineering Mechanics 120, no. 4, pp. 681–94.
[31] Cremona, C. 2001. “Optimal Extrapolation of Traffic Load Effects.”
Structural Safety 23, no. 1, pp. 31–46.
[32] Ministry of Communications and Transportation (MOCAT). 2015.
Specifications for Design of Highway Suspension Bridge JTG/T D65-05.
Beijing, China: China Communications Press.
[33] Lu, N., M. Noori, and Y. Liu. 2016. “Fatigue Reliability Assessment of
Welded Steel Bridge Decks Under Stochastic Truck Loads via Machine
Learning.” Journal of Bridge Engineering 22, no. 1, 04016105. 10.1061/
(ASCE)BE.1943-5592.0000982
[34] Liu Y., X. Xiao, N. Lu, and Y. Deng. 2016. “Fatigue Reliability Assessment
of Orthotropic Bridge Decks Under Stochastic Truck Loading.” Shock and
Vibration. doi:10.1155/2016/4712593
[35] Cai, C.S., J. Hu, S. Chen, and W. Zhang. 2015. “A Coupled Wind-vehicle-
bridge System and its Applications: A Review.” Wind and Structures 20,
no. 2, pp. 117–42.
[36] Yin, X., Y. Liu, and B. Kong. 2016. “Vibration Behaviors of a Damaged
Bridge Under Moving Vehicular Loads.” Structural Engineering and
Mechanics 58, no. 2, pp. 199–216.
 CHAPTER 5

Lifetime Deflections of
Long-Span Bridges Under
Dynamic and Growing
Traffic Load

5.1 INTRODUCTION

Numerous in-service highway bridges have collapsed, or have been dam-

aged over the last few decades, due to overloaded trucks. These damages
are mostly resulted from the intense competition in transportation of
goods in the global market [1, 2]. A steady traffic growth in the volume
and the gross weight may pose a safety hazard to in-service bridges [3].
Even though the design live-load model has a confidence level to ensure
the bridge safety in a reference period, such level may be underestimated
by the actual traffic loading scenarios composed of several extremely
overloaded trucks. For instance, a case study conducted by Han et al. [4]
showed that 4 out of 1319 trucks yielded hogging moments of a bridge
larger than the one estimated by the China’s design live-load model.
A review on the cause of recent bridge collapses conducted by Deng et al.
[5] indicated that truck overloading is the major human factor that leads to
the shortening of bridge life. In addition to the general phenomenon of traf-
fic loading, long-span bridges are flexible and ductile especially under the
event of simultaneous presence of heavy-duty trucks [6]. In general, traffic
load effects on long-span bridges have a higher probability of exceed-
ance of a serviceability limit state rather than an ultimate limit state. To
ensure the serviceability of long-span bridges, native design specifications
have recommended the deflection limit, such as L/250 and L/300 for the
­maximum deflection and the maximum deflection range in the Eurocode
3 [7], where L is the effective length of a bridge. Therefore, the lifetime
90   •   BRIDGE RELIABILITY AND SERVICEABILITY

extreme deflection assessment and the reliability calibration of long-span

bridges considering the actual traffic loading are worth investigating.
A weigh-in-motion (WIM) system in a structural health monitoring
system [8, 9, 10] is designed to monitor the traffic loading on highway
bridges. Implementations of WIM measurements in bridge engineering
have been investigated by numerous studies, such as the calibration of
design live-load models [11, 12, 13, 14], bridge fatigue reliability assess-
ment [15, 16] and the characteristic traffic load effect extrapolation
[17, 18, 19]. Conventional methods for the extrapolations of the traffic load
and the load effect are associated with block maxima approaches based
on the generalized extreme value (GEV) theory or Rice’s level-crossing
theory [20, 21]. Nonstationary traffic load effects on short-span bridges
(L<30 m) due to traffic growth has been investigated by O’Brien et al. [22]
utilizing the GEV distributions. In addition to investigations on the static
extrapolation cited earlier, studies on dynamic extrapolation of the traf-
fic load effects have been conducted by several researchers. For instance,
O’Brien et al. [23] presented an assessment dynamic ratio (ADR), a
lifetime dynamic allowance level, to investigate the influence of vehi-
cle–bridge interaction (VBI) on short- to medium-span concrete bridges.
Subsequently, Caprani [13] utilized the ADR in conjunction with a mul-
tivariate GEV theory to investigate the dynamic allowance of a highway
bridge under long-term traffic loading. Even though these developments
are mostly concentrated on short- to medium-span bridges, a solid founda-
tion has been provided for the extended application to long-span bridges.
The simultaneous presence of multi-type trucks and multiple traffic load-
ing scenarios is the special condition leading to the complexity of the statisti-
cal extrapolation of a long-span bridge. In addition, traffic load effects under
individual truck loads can be treated following an independent identical dis-
tribution (IID), but load effects induced by simultaneous presence of multiple
vehicles violate the IID assumption. In order to solve this problem, Caprani et
al. [18] presented a maximum traffic loading event accounting for the simul-
taneous presence of multiple trucks in a period to evaluate the maximum
traffic load effect. Critical traffic loading scenarios on long-span bridges was
presented by O’Brien et al. [19] utilizing microscopic stochastic traffic flows
simulated based on WIM data. The maximum-in-lifetime traffic loading
scenarios on medium-span bridges were identified by Enright and O’Brien
[14] with the consideration of a free-flowing traffic condition. Markov chains
[24] and the cellular automaton [25] are advanced approaches to simulate
the microscopic driving behaviors of individual vehicles (e.g., acceleration
and lane change). Several recent studies have emphasized the traffic loading
behavior on long-span bridges. For instance, O’Brien et al. [26] utilized a
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   91

microscopic traffic model to investigate the congested traffic loading on long-

span bridges with consideration of truck proportions, where the slow-mov-
ing traffic led to greater loading than full-stop condition. Caprani et al. [27]
investigated the influence of microscopic traffic scenarios on the extrapola-
tion of load effects. Ruan et al. [28, 29], used a site-specific WIM data to
estimate the traffic load effect of a multi-pylon cable-stayed bridge and the
maximum static friction coefficient of an anti-sliding model for a suspension
bridge with the consideration of unbalanced traffic loading. Dynamic effects
of traffic loading on long-span bridges due to deteriorated road surfaces lead
to higher dynamic amplification factors (DAFs) [30, 31, 32]. Caprani [33]
demonstrated the DAF of the lifetime characteristic load effect was reason-
able. The studies cited previously contributed to the understanding of the
­statistical extrapolation of the traffic load effects on long-span bridges.
As elaborated earlier, the site-specific WIM data provide a statistical
database for modeling the bridge live load. The stochastic traffic flow is an
effective and realistic model that needs to be considered especially for long-
span bridges under multiple-truck presence events. In practice, stochastic
traffic flows on long-span bridges can be further developed for probabilis-
tic applications including the load effects extrapolation and the reliability
assessment. Since the lifetime of a bridge is much longer than the duration
of the recorded traffic data, some conventional assumptions in this field
might be inappropriate. First, the actual traffic density on a bridge is time
variant, and the traffic volume will grow in the bridge lifetime, which leads
to the virtual non-stationarity of traffic load effects in one day or in the
lifetime. The non-stationarity property apparently violates the stationarity
assumption of the GEV theory. Subsequently, even though the dynamic
effects of traffic loading are not significant for flexible long-span bridges,
the deterioration of road roughness condition will amplify the traffic load
effect. These issues result in the complexity of the probabilistic investiga-
tion of the traffic load effects, as well as the inapplicability of conventional
block-maxima approaches. However, to the best of the authors’ knowledge,
the influence of some actual traffic load behaviors, such as the dynamic
effects and the traffic growth, on the statistical extrapolation is still not clear.
This chapter presents a study that introduces a methodology for the
statistical extrapolation of traffic load effects on long-span bridges consid-
ering several challenging factors including the traffic growth, the dynamic
impact and the actual traffic pattern. The non-stationarity of the traffic load
effects due to traffic growth is considered in a series system compounded by
several interval traffic models. The traffic dynamic impact is simulated by a
traffic-bridge coupled vibration system and its statistical characteristics are
captured by Rice’s level-crossing model. The actual traffic pattern is simulated
92  •   BRIDGE RELIABILITY AND SERVICEABILITY

in stochastic traffic flows based on the statistics of the WIM measurements of

a highway bridge. The proposed methodology is demonstrated and verified
by two numerical examples and subsequently applied to a case study of a
suspension bridge. Influences of road roughness deterioration and the traffic
growth rate on the statistical extrapolation of the maximum deflection and the
probability of exceedance of the deflection limit are also investigated.

5.2 THEORETICAL BASES OF THE TRAFFIC-

BRIDGE INTERACTION AND RICE’S FORMULA

The vehicle impacts on bridges are simulated in a traffic-bridge cou-

pled vibration system. Dynamic statistics can be captured via Rice’s
level-crossing theory. Herein, the theoretical formulations of the
traffic-bridge interaction and Rice’s formula are introduced.

5.2.1 TRAFFIC-BRIDGE INTERACTION

The traffic–bridge interaction originates from the VBI that is convention-

ally used for the DAF estimation. In a VBI system, the vehicle is usually
simulated by the degrees of freedom (DOFs) in physical coordinates. For
instance, a 3D 2-axle truck model as shown in Figure 5.1 can be simulated
by 7 DOFs including rotational (θr1 and θr2) and a vertical translational
motions (Zv) of the vehicle rigid body, as well as vertical translational
motions (Za1L, Za2L, Za1R, and Za2R) of each vehicle wheel. The equation
of motion of the VBI system can be solved by a modal superposition
approach or a step-by-step integration approach in time domain [34].
For long-span bridges, the simultaneous presence of multiple vehi-
cles is a unique phenomenon compared with short-span bridges. However,
investigations conducted by Chen and Cai [12] suggested that the interac-
tion effects between multiple vehicles on a flexible long-span bridge are
insignificant. Therefore, this study utilizes an equivalent dynamic wheel
load (EDWL) approach proposed by Chen and Wu [34] to evaluate the
dynamic traffic load effects. The EDWL approach utilizes time-variant
forces accounting for the mode shapes and natural frequencies of the
bridge to approximate the VBI forces. Eventually, the cumulative EDWL
acting on the bridge can be defined as

nv
 na

{F (t )}
eq
wheel
{ }
= ∑  1 − EDWL j (t ) / G j  G j ⋅ ∑ hk [ x j (t ) + ak [ x j (t ) d j (t )]]  (5.1)
j =1  k =1 
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •  93

Zv

¦È
r1

1 1 2 2
KuzL CuzL KuzL CuzL
Za1L Za2L
1 1 2 2
KlzL ClzL KlzL ClzL
x

(a)
Zv

¦È
r2

1 1 1 1
KuzL CuzL KuzR CuzR
Za1L Za2R
1 1 1 1
KlzL ClzL KlzR ClzR
y

(b)

Figure 5.1.  Physical models of a 2-axle truck: (a) elevation view;

(b) side view.

where EDWLj (t) is the dynamic load of the j-th vehicle at time t; Gj, xj,
and dj are the self-weight, longitudinal location, and transversal location
of the center of gravity of the jth vehicle on the bridge, respectively; hk
and ak are the vertical and the torsional mode shapes for the kth mode of
the bridge, respectively; and nv and na are the number of vehicles on the
bridge and the number of axles of the j-th vehicle, respectively. The num-
ber of vehicles on the bridge changes with time depending on the stochas-
tic traffic flow. The effectiveness has been demonstrated by Chen and Wu
[34]. Note that the road-roughness coefficient (RRC) is considered in the
vertical displacement and velocity of the vehicle.

5.2.2 RICE’S FORMULA

With the simulated deterministic load effects, the statistical parameters

can be estimated by the tail fittings. Rice formula [35] was chosen in the
present study for the purpose of evaluating both the extreme traffic load
effect and the probability of exceedance in the bridge lifetime. Principle of
94  •   BRIDGE RELIABILITY AND SERVICEABILITY

Rice’s formula is shown in Figure 5.2. In practice, influence lines of long-

span bridges are long enough, and the critical traffic loadings are mostly
consistent with intensive vehicles. Thus, the load effects can be assumed
as a Gaussian random process [36]. In addition, interval growing traffic
loading evidently satisfies the stationary assumption. Based on the earlier
assumption, the mean level-crossing rate v(x) under the condition of a
threshold level and a reference period is expressed by Rice [35]

s  ( x − m) 2  1
v( x) = exp  − = (5.2)
2 Às  2s 2  Rt

where x is the traffic load effect, m and σ are the mean value and stan-
dard deviation of the load effect, and s is the standard deviation of the
derivative of the load effects. In general, the level-crossing rate can be
expressed by a normalized rate indicated as the fitted histograms ver-
sus the summation of truncated remaining histograms. The critical step
of using Rice level crossing theory for extrapolating is to determine the
optimal starting point indicated as x0,opt and the optimal number of class
intervals indicated as Nopt [37]. In this regard, the conventional approach
is to utilize the Kolmogorov–Smirnov (K–S) statistics recommended and
described by Cremona [38] to check the confidence level of the predefined
starting point and number of class intervals. Eventually, the statistical

x
Up crossing

Level 1

t
Level 2
Down crossing
(a)
Optimal starting point
v(x)
Rice's fitting

Interval length

x
(b)

Figure 5.2.  Basic principles of Rice’s formula: (a) level

crossings; (b) fitting to the crossings.
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •  95

extrapolation can be evaluated by the derivation of Equation (5.2) consid-

ering a return period Rt.
Note that the extrapolated value x based on the level-crossing rate can
also be defined as the value is exceeded with a probability in a reference
period. Therefore, the cumulative distribution function (CDF) of the max-
imum load effects can be written as

  T 
Fmax ( x, T ) = 1 − a = 1 − 1 − exp  −  
  Rt  

  ( x − m) 2  
= exp −Tv0 exp  −   (5.3)
  2s 2  

where T is a general time period related to a reference period Tref, that

is a 100-year lifetime of a bridge in the present study, a is a probabil-
ity of exceedance that is roughly 10% between Tref =100 years and Rt =
1000 years. It is clear that the CDF can be estimated easily given the
level-crossing function as shown in Eq. (1) for a stationary process. These
equations can only be used for stationary traffic load without considering
traffic growth. The growing traffic case will be shown in next section.

5.3 METHODOLOGY OF EVALUATING MAXIMUM

LOAD EFFECTS CONSIDERING INTERVAL
TRAFFIC GROWTH

As mentioned before, the traffic growth leads to nonstationary density

of the vehicles on the bridge, which directly affects the maximum traffic
load effects. An interval traffic growth model is utilized in the present
study to divide the lifetime traffic loads into several intervals that can
be assumed as nongrowing and stationary in each interval. An improved
Rice’s formula to combine the interval level-crossing models is presented
in a detailed framework.

5.3.1 INTERVAL TRAFFIC GROWTH MODEL

In general, traffic volume grows continually over a period. Such continual

growth is usually defined as an annual growth rate (AGR) that is a com-
pounded rate between two years. Since the traffic density over a bridge
lifetime is nonstationary, the continual growth model is inappropriate to
96  •   BRIDGE RELIABILITY AND SERVICEABILITY

8000
Actual traffic growth

6000 A 10-year interval

ADTT

4000

2000

E1 E2 E10
0
0 20 40 60 80 100
(a) Time (year)

E1 E2 ... En

P1= Fmax, 1 (x, Tint) P2= Fmax, 2 (x, Tint) Pn= Fmax, n (x, Tint)
(b)

Figure 5.3.  An interval traffic growth model: (a) interval ADTTs;

(b) a series system model.

be used for the load effect extrapolation. However, traffic volume could be
assumed stationary in a short period such as one or two years. This short
period is defined as an interval in this study.
An illustrative example of the interval traffic growth is shown in
Figure 5.3, where the curve is the volume of the average daily truck traf-
fics (ADTTs) accounting for the actual traffic growth, and the histograms
are the volumes of the ADTTs in the 10-year interval. It is obvious that
the traffic volume is constant in an interval period rather than growing
with the curve. The advantage of the interval growth model is that Rice’s
formula is appropriate to be used in each interval. The shortcoming of
the interval growth model is that the result is only an estimation and its
accuracy mostly depends on the number of intervals. Obviously, increase
of the intervals will lead to the improvement in the accuracy of the result,
but also leads to additional computational effects.

5.3.2 IMPROVED RICE’S EXTRAPOLATION ACCOUNTING

FOR INTERVAL TRAFFIC GROWTH

Rice’s formula as shown in Eqs. (4) and (5) are unavailable for traffic load
effects under growing traffic load because the traffic density is nonstation-
ary over the bridge lifetime. However, these formulas are effective in each
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •  97

time interval as shown in Figure 5.3(a). The question is how to combine

these individual interval probability models to extrapolate the maximum
value over a specified return period. On the basis of related research con-
ducted by Caprani et al. [18] and Zhou et al. [39], this study utilizes a series
system as shown in Figure 5.3(b) to combine each interval. Therefore, the
CDF of the maximum value in lifetime can be estimated by multiplying
each interval CDF that can be evaluated by Equation (5.3) with T equal to
the interval period. Assuming that the Tref can be divided into Nint intervals,
the lifetime CDF can be estimated by the interval CDF products:

 int  ( x − mi ) 2  
Nint N

  Fmax ( x, T ) = ∏ Fmax,i ( x, Tint ) = exp − ∑ Tint v0,i exp  −

G
  (5.4)
i  i =1  2si 2  

G
where Fmax ( x, T ) is the maximum traffic load effect of a bridge in a refer-
ence period T considering traffic growth, Nint is the number of intervals, Tint
= T/Nint is the interval period, Fmax,i (x,T) is the CDF of the maximum traffic
load effects in the ith interval, and v0,i mi and σi are the mean crossing rate,
the mean value and the standard deviation of the load effects in the ith
interval, respectively. Thus, the maximum load effects in a return period
can be estimated based on the system CDF as shown in Equation (5.4) and
the general form of the CDF as shown in Equation (5.3), written as

−Tref
Rt = (5.5)
ln  F
G
max ( x, Tref ) 

where each maximum load effect corresponds to a return period.

Herein, it can be summarized that the critical steps of combining the
interval traffic load effects for the extrapolation are: (a) estimating the
CDF of the traffic load effects in each interval based on Equation (5.3);
(b) combining the interval CDFs in a series system to formulate the actual
CDF via Equation (5.4); and (c) computing the return periods with respect
to the maximum load effects via Equation (5.5). The maximum load effect
can also be interpolated from the Gumbel paper.
In addition to the maximum extrapolation, Rice formula can also esti-
mate the probability of the maximum load effect exceeding a predefined
limit over a reference period in terms of the probability of exceedance.
Formulation of the probability of exceedance is written as

p ( z , t ) ≅ 1 − Fmax ( z , Tref ) (5.6)

98  •   BRIDGE RELIABILITY AND SERVICEABILITY

where z is a predefined threshold for the traffic load effect, and Fmax(z,Tref)
denotes the CDF of the maximum load effect in lifetime at the z value.
This equation provides a way for the quantification of the probability of
the traffic load effect exceeding a limit.

5.4 A COMPUTATIONAL FRAMEWORK OF

EXTRAPOLATING THE MAXIMUM LOAD
EFFECT FF LONG-SPAN BRIDGES

Based on the interval traffic growth model and the improved Rice’s for-
mula, a computational framework is presented for evaluating the maximum
load effects of long-span bridges using WIM measurements. The flowchart
of the computational framework is shown in Figure 5.4. The entire pro-
cedure outlined in the flowchart is mainly composed of three categories
including the traffic load simulation, the load effect computation, and the
probabilistic extrapolation. Illustrations of the procedures in the categories
are elaborated in the following.
The first module as depicted in Figure 5.4 is the traffic load simulation
based on recorded WIM data. With available WIM data, filtering proce-
dures should be conducted to exclude the invalid data and select the effec-
tive data that contribute to the maximum traffic load effect. In the present
study, lightweight cars were removed from the recorded data, since most
of the critical loading scenarios are usually composed of dense traffic
flows with a high proportion of heavy trucks. The statistical parameters
of the traffic flow can be divided into two groups [17]: (a) those modeling
the individual vehicle feature in small-scale (i.e., the vehicle configura-
tion, the gross vehicle weight, the vehicle spacing and the driving speed);
and (b) those modeling the traffic feature in large-scale (i.e., proportions
of vehicle types and traffic volume). The most critical parameter is the
vehicle spacing defining the vehicle gap, that is, the space between two
vehicles in the same driving lane. The vehicle spacing is a unique and
critical factor for the traffic flow simulation on long-span bridges, and is
usually measured by the headway that is time variant depending on the
traffic density [19]. The extreme traffic load effects are mostly induced
by dense traffic flows with small vehicle spacings. Therefore, the PDF of
the vehicle spacing in dense traffic flows is one of the important factors
impacting the maximum traffic loading on a long-span bridge.
In general, traffic loads on a bridge can be simulated with a math-
ematical model in time domain or in space domain. A stochastic traffic
flow is one of these mathematical models composed of individual vehicles
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •  99

Traffic load simulation:

WIM measurements
Filtering
Statistics of traffic flows
Stochastic traffic simulation
Daily critical traffic-loading scenarios

Load effect computation:

Considering VBI Considering traffic growth

Or

Bridge mode shapes and frequencies Bridge static influence lines

RRCs EDWLs AGR Interval growth
Dynamic load effect histories Daily maxima in each interval

Probabilistic extrapolation:
Counting the
number of crossings
Fitting the level-crossing rates to the
histograms of the number of level-crossings
Return period
Extrapolation of the maximum traffic load
effects
Threshold value
Lifetime probability of exceedance of
the limit

Figure 5.4.  Flowchart of the proposed computational framework for the

lifetime maximum traffic load effect extrapolation.

formulated with statistical parameters. However, there is a problem with

utilizing numerous daily traffic flows to conduct the probabilistic and
dynamic analysis, where the step-by-step integration of the VBI system
solution will be extremely time consuming. Note that the purpose of
using the daily traffic flow load model is the probabilistic modeling of
the extreme traffic load effects. Such maximum value is affected by the
upper tail of the load effects produced by critical traffic loading scenarios.
Therefore, critical traffic loading scenarios identified in a daily stochastic
traffic flow can be utilized for the daily maxima simulation. The principle
of identifying the critical traffic loading scenario is related to the static
100   •   BRIDGE RELIABILITY AND SERVICEABILITY

influence line analysis aimed at finding out the maximum load effects in
the daily stochastic traffic flow. Based on this assumption, a step by step
search strategy is adopted to identify the critical loading scenario. These
procedures are: (1) generating the stochastic traffic flows via MCS and the
statistics of the WIM data; (2) specifying the effective range of the loading
scenario on the bridge according to the bridge length; (5.3) moving the
predefined range forward along the simulated daily stochastic traffic flows
to calculate the static traffic load effect; (5.4) identifying the maximum
load effect and the corresponding loading scenario; and (5.5) repeating
steps (5.3) to (5.4) for the remaining daily traffic flows.
The second module is the traffic load effect computation. There are
two factors considered in this module including the VBI and the traffic
growth. For VBI, the dynamic vehicle load of the bridge under the critical
loading scenario can be evaluated using the EDWL approach with the
consideration of the RRC. The result is a time history that will be used
for counting the number of level crossings. For the traffic growth, the
interval traffic growth model is used to simulate traffic growth with an
AGR. The daily maxima in each interval are estimated using the static
influence lines.
The third module is the probabilistic extrapolation. The first step is
to count the number of level crossings based on the estimated traffic load
history or the interval daily maxima. The level-crossing rate as shown in
Figure 5.2(b) can be fitted to the histograms of the number of crossings.
Subsequently, the maximum traffic load effect over a return period can be
extrapolated based on Rice formula in Equation (5.3) or the interval model
in Equation (5.4). Finally, the probability of exceedance of the predefined
limit can be evaluated from the CDF of the maximum value.
Obviously, there are some key points in the proposed method. Firstly,
more number of traffic intervals will lead to a more accurate extrapola-
tion, but also leads to more computations. Subsequently, since the actual
traffic volume in each traffic interval will grow instead of being constant,
which is assumed in the proposed approach, the extrapolated value will be
slightly under-conservative. In addition, consideration of lightweight cars
should be further developed.

5.5 VERIFICATION EXAMPLES

Two numerical examples including an individual GVW extrapolation and

a deflection extrapolation of an idealized long-span bridge are presented to
verify the effectiveness of the interval traffic growth model.
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   101

5.5.1 INDIVIDUAL GVW EXTRAPOLATION

A numerical example from O’Brien et al. [22] is presented herein to verify

the effectiveness of the interval traffic growth model. In the example, the
block length is 1 day, the total lifetime is 25,000 days, the truck weights
follow a normal distribution with Wi~N(50, 5) in tons, and the initial traf-
fic volume is 1,000 trucks per day. The objective of this numerical study
is to extrapolate the maximum individual GVW over a bridge lifetime
considering the daily traffic volume growth rate of 0.016% (an average
AGR of 4.1%). O’Brien et al. [23] used a day-by-day growth model to
simulate the traffic growth, but the present study adopted a 10-year time
interval. The GVE fitting utilized the entire 100-year daily maxima, while
Rice’s fitting utilized the 10-year 30% upper daily maxima. The follow-
ing discussion focuses on the accuracy of the extrapolation, while the
computational efficiency is the same.
Figure 5.5(a) shows the annual crossing rates of the 1st, the 5th, and
the 10th interval period fitted to histograms simulated by the daily maxima
in 10 years. It is observed that apparently the mean value of the crossing
rate moves to the right hand side, and the mean level-crossing rate shows
a slight decrease. Figure 5.5(b) plots the maxima and fittings in a Gumbel
paper, where the 100-year daily maxima are labeled in point symbols, the
GEV’s fitting are shown in dash lines, and Rice’s fittings are in solid lines.
For the case of nongrowing traffic, both the GEV and Rice’s extrapola-
tions are in agreement with the value (77.50 t) provided by O’Brien et al.
[23]. However, for the case of growing traffic, Rice’s extrapolation more
accurately fits to the reference value (80.60 t). Note that the exact value
(78.842 t) for the case of nongrowing traffic computed by the normal dis-
tribution of GVW to the power of 1,000 [38] is larger than the extrapolated
value. This phenomenon can be explained by the fact that the convergence
of a normal distribution to Gumbel is extremely slow, and both Rice and
GEV’s extrapolation are all under-conservative.
The following inferences can be obtained from the numerical resutls:
(a) the nonstationarity of the extreme distribution of the GVWs can be
observed through the level-crossing curves moving to the left side for a
higher interval, since the higher traffic volume induces larger daily max-
ima as well as the number of higher-level crossings; (b) both the GEV and
Rice’s fittings fit the daily maxima fairly well for the nongrowing traffic
condition, since the GVW is stationary over the reference period; (c) the
GEV fitting deviates the tail of the daily maxima due to the non-stationarity
of the traffic volume during lifetime; (d) Rice’s fitting is close to the tail,
where the nonstationarity of the growing traffic has been captured by the
102   •   BRIDGE RELIABILITY AND SERVICEABILITY

80
1st interval
5th interval
Annual number of crossings

60 10th interval

40

20

0
60 65 70 75 80
(a) GVW (t)

15
Exact values
1000-year return period

No-growth ACGR=4.1%
10
GEV GEV
Rice Rice
–log[–log(F)]

–5
60 65 70 75 80
(b) GVW (t)

Figure 5.5.  Analytical results of the first example: (a) annual crossing

rates; (b) daily maxima and fittings on Gumbel paper.

proposed method utilizing an interval traffic growth model. This example

demonstrates the effectiveness of the proposed interval model for extrapo-
lating a maximum individual GVW considering traffic growth.

5.5.2 DEFLECTION EXTRAPOLATION OF AN IDEALIZED

LONG-SPAN BRIDGE

Since the objective of the proposed method is the application in long-

span bridges, the second numerical example is presented to extrapolate the
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   103

x=L/2=50 m Si ~ N(100, 10) m

GVW=50 t Gi ~N(50, 5) t

50 mm
Deflection: y = Gisin(x¦Ð/L)

Figure 5.6.  An idealized long-span bridge and crossing vehicles.

deflection of an idealized long-span bridge associated with the first exam-

ple. Illustration of this bridge structure is shown in Figure 5.6. The vehi-
cle spacing follows a normal distribution Si~N(100, 10) in meter, and the
other parameters are the same as those in the first example. The objective
of this example is to extrapolate the maximum deflection of the bridge in
a 1,000-year return period. The AGR is supposed to be between 0 and 1%.
The interval period is 2,500 days (effective days in 10 years), and
the lifetime includes 10 intervals. Step-by-step simulations based on the
influence lines of the bridge were conducted to evaluate the daily max-
ima. Three typical annual crossing rates are shown in Figure 5.7(a), where
the crossing rate apparently moves to the right side and has a strong
non-stationarity. Numerical results and a return period line are plotted in
Figure 5.7(b). It is observed from the daily maxima that: (a) the simu-
lated daily maxima for the case of no-growth model is nearly a straight
line plotted on the Gumbel paper; (b) the daily maxima for the case of
growing traffic model moves to the left and forms a curve on the Gumbel
paper. In addition, it is observed from the fittings that: (a) for the case of
non-growing traffic the GEV and Rice’s fittings to the full data or the inter-
val data are all best suited for extrapolating; (b) for the case of growing
traffic load the GEV and Rice’s interval model have better extrapolations
than those extrapolated via the GEV and Rice’s full data model; (c) even
though Rice’s full data model has a relatively better extrapolation due to
the optimal starting point, the interval growth model can provide a much
better extrapolation.
These phenomena can be explained by the following inferences.
First, with the consideration of the traffic growth, the probability density
of the traffic load effects is time-variant, and will move to the higher-value
side. The time-variant probability density violates the IID assumption,
and then results in the worse fitting of the general GEV distribution or
Rice’s formula. Second, Rice’s interval fitting is better for providing the
extrapolation from the starting point of the final interval (105 mm in this
example). Finally, the interval models provide better fittings for the upper
104   •   BRIDGE RELIABILITY AND SERVICEABILITY

50
1st interval
5th interval
40
10th interval
Annual number of crossings

30

20

10

0
50 60 70 80 90 100 110
Deflection (mm)

15
1000-year return period

GEV (full)
10
Rice (full)
–log[–log(Fmax)]

GEV (interval)
No=growth
5 Rice (interval)

AGR=1%
0

–5
50 70 90 110 130
Deflection (mm)

Figure 5.7.  Analytical results of the second examples: (a) annual cross-

ing rates; (b) daily maxima and fittings plotted on Gumbel paper.

data rather than the lower data. This maybe due to the insufficient number
of intervals as shown in Figure 5.3(b) in capturing a higher probability
of exceedance.

5.6 CASE STUDY

A suspension bridge and its WIM data are utilized to demonstrate the
effectiveness of the proposed computational framework. Dynamic char-
acteristics and the traffic growth are considered to investigate their influ-
ences on the probabilistic extrapolation.
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   105

5.6.1 WEIGH-IN-MOTION MEASUREMENTS OF A
SUSPENSION BRIDGE

Nanxi Yangtze River Bridge is a long-span highway suspension bridge in

Sichuan, China. A pavement WIM system composed of scales or pressure
sensors embedded into the road pavement was installed on the bridge.
More details of the bridge and the WIM measurements can be found in Liu
et al. [40] and Lu et al. [16]. A filtering process was conducted to identify
and remove invalid records, where the vehicles with the GVW less than
3 t were removed. An overview of the filtered WIM data is summarized
in Table 5.1. The maximum overload rate was about 200% over the GVW
limit (55 t) in China for a 6-axle truck [41]. A dense traffic flow composed
of such extremely overloaded trucks is literally a safety hazard to long-
span bridges.
Based on the WIM data and the assumption of the stochastic traffic
model, the trucks were classified into 6 categories indicated as V1 ~ V6,
where V1 denotes the 2-axle light trucks, and V2 ~ V6 denote the 2- to
6-axle trucks. The hourly traffic volume per lane was analyzed as shown in
Figure 5.8(a). The hourly traffic volumes of the busy traffic between 9:00
and 19:00 were utilized for statistical analysis of the truck spacing. The
probability density of the truck spacing is accurately fitted by a lognormal
distribution function (LNDF) as shown in Figure 5.8(b).
It is worth to mention that the vehicle spacing in terms of traffic den-
sity is a significant factor impacting the traffic load effects on medium- to
long-span bridges. This study ignores the effect of lightweight cars on the
bridge defection computation by removing the lightweight cars from the
database. Therefore, the truck spacing in this study denotes the truck gap
between two following trucks in a traffic lane, which is different from the
vehicle spacing that defines the vehicle gaps in an actual traffic stream.
The effective truck spacing between two trucks rather than the actual traf-
fic gaps were utilized in the present study. However, the truck spacing

Table 5.1.  Overview of the filtered WIM measurements

Items Values
Time period May 1, 2013 to April 30, 2015
Number of recording days 729
Total number of effective trucks 1,563,921
Maximum GVW (t) 164
Number of overloaded trucks 12,252
106   •   BRIDGE RELIABILITY AND SERVICEABILITY

200

Number of trucks per hour Dense traffic

150

100

50

0
0:00 4:00 8:00 12:00 16:00 20:00 24:00
(a) Time

0.1
WIM data
Lognormal distribution
0.08
Probability density

0.06

0.04

0.02 X ~ LN(6.21, 1.40)

0
0 1000 2000 3000 4000
(b) Truck spacing (m)

Figure 5.8.  Statistics of the WIM measurements: (a) hourly traffic vol-

ume; (b) truck spacing of the busy traffic flow.

makes sense in the context of probabilistic domain, since the PDF of the
truck spacing was fitted to the actual WIM data. Even though the truck
spacing does not represent an actual traffic state, the PDF has captured the
statistical characteristics of the trucks in the actual traffic flows.

5.6.2 EXTREME DEFLECTION CONSIDERING DYNAMIC

TRAFFIC LOADS

As the first task, a commercial program ANSYS was utilized to build the
bridge finite element (FE) model as shown in Figure 5.9. In the FE model,
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   107

Table 5.2.  The first five order mode frequencies of the suspension bridge
Mode frequency
FE Monitored Error
Order model data (%) Illustration
1 0.131 – – 1st asymmetric transversal bending
2 0.1781 0.1849 −3.68 1st asymmetric vertical bending
3 0.2208 0.2492 −11.40 1st symmetric vertical bending
4 0.3106 0.3049 1.87 2nd asymmetric vertical bending
5 0.4074 0.3975 2.49 2nd symmetric vertical bending

WIM system
89 m

L=
82
0m 145 m

Figure 5.9.  Finite-element model and dimensions

of the suspension bridge.

the steel stiffening steel-box girders and the concrete pylons were modeled
by beam elements, the hangers and the cables were modeled by link ele-
ments, the bridge tower elements and the cable elements were separated
and the top joints were connected by a set of coupled degrees of free-
dom. It is acknowledged that long-span bridges are geometrically nonlin-
ear, especially under heavy traffic loads. The present study, however, is
limited to a linear analysis as a first step for implementing the improved
stochastic analysis. The advancement of the improved stochastic analysis
can be extended to the nonlinear case, where the computational efficiency
should be further developed. Table 5.2 summarizes the first 5 fundamental
mode characteristics of the FE model. The vehicle physical properties in
this case study were adopted from Yin et al. [42]. The road surface rough-
ness conditions (RRCs) in this study were defined by the International
Organization for Standardization (1995). The RRCs for classifications
of “Good,” “Average,” and “Poor” are 32×10−6, 128×10−6, and 512×10−6,
respectively. RRC coefficients were simulated by inverse Fourier transfor-
mation approach [3] in time domain.
108   •   BRIDGE RELIABILITY AND SERVICEABILITY

0.02
Static
0.01 Dynamic
Vertical deflection (m)
0

-0.01

-0.02
L/2 point
-0.03
3L/8 point
L/4 point
-0.04
0 L/4 L/2 3L/4 L
Vehicle location on the bridge (m)

Figure 5.10.  Deflection histories of critical points of the girder under

a 2-axle truck load.

As the second task, the dynamic traffic load effects at the critical points
of the bridge were simulated. Initially, a 2-axle truck with GVW = 10 t, a
driving speed v = 20 m/s, and the RRC = “Good” were considered to show
the static and dynamic deflections of the bridge girders. Figure 5.10 plots
the vertical deflections of the three potential points versus truck loading
position on the bridge, where the bridge self-weight effects were excluded.
It is observed that the most critical point is the L/4 (quarter-span) point.
The maximum deflection due to static and dynamic loads are 0.029 and
0.032 m, respectively. Therefore, the subsequent investigation focuses on
the quarter-point of the bridge.
Subsequently, the critical traffic loading scenarios were identified
from the simulated daily stochastic traffic flows. An illustration exam-
ple for generating the critical loading scenario on a 4-lane bidirectional
bridge is shown in Figure 5.11, where traffic lanes of 1 and 2 are in the
same direction, while the traffic lanes of 3 and 4 are in the opposite direc-
tion. A daily maximum deflection was identified from the static deflection
histories in Figure 5.11(a), and the corresponding critical traffic loading
scenario was extracted from the daily stochastic traffic load as shown in
Figure 5.11(b). Obviously, the critical loading scenario is a small part
(0.1%) of the daily traffic flow. As a result, the time-consuming com-
putation due to the step-by-step integration will be greatly reduced by
utilizing the critical traffic loading scenario rather than the entire daily
traffic stochastic flows.
A total number of 1,000-day traffic flows were simulated via
­Monte-Carlo simulation for the current traffic condition. Each critical traf-
fic loading scenario was identified according to the maximum total GVW
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   109

0.6

0.4

0.2

Deflection (m)
0

-0.2

-0.4

-0.6
Daily maxima
-0.8
0 2 4 6 8 10
(a) Time (h)

120

80
GVW (t)

V1
40 V2
V3
0 V4
0 V5
205 V6
410
615 4
3
820 2
(b) Location (m) 1 Traffic lane

Figure 5.11.  An example of identifying the critical loading scenario:

(a) a daily deflection history; (b) a critical loading scenario.

on the bridge. Suppose these critical traffics are free flowing with a constant
velocity v = 20 m/s. Subsequently, the traffic-bridge coupled vibration anal-
ysis was conducted by utilizing the EDWL approach. Figure 5.12 shows
the difference between the number of crossings of static and dynamic his-
tories, where the numbers of down crossings at the deflection level a =
−0.667 m was counted, and Nistatic and Nidynamic are the number of crossings
for the static and the dynamic histories, respectively. The deflection level
is different from the deflection threshold. The deflection level was utilized
to count the number of crossings, but the deflection threshold was utilized
to evaluate the probability of exceedance. It is observed, from Figure 5.12,
that the dynamic histories fluctuate around the static histories. This makes
the numbers of level crossings different. The number of the dynamic cross-
ings is always higher than that of the static crossings. This numerical result
shows the advantage of Rice’s formula in capturing the dynamic effects.
110   •   BRIDGE RELIABILITY AND SERVICEABILITY

0.4
Static
0.2 Dynamic

0
v = 20 m/s
Deflection (m)

-0.2 RRC = Good

-0.4

-0.6 1st scenario 2nd scenario a = 0.667 m

1 2
N Static = 1 N Static = 0
-0.8 1 2
N Dynamic = 2 N Dynamic = 2
-1
0 20.5 41 61.5 82
Time (s)

Figure 5.12.  An example to show the different between the numbers

of crossings of static and dynamic histories.

Static and dynamic histories

2000
Static response
Dynamic response
Static fitting
1500
Number of crossings

Dynamic fitting
x0 = 0.584
v0 = 1001.9
1000
µ = 0.4889
σ = 0.1844 x = 0.584
0
500 v0 = 701
µ = 0.4813
σ = 0.1825
0
0 0.2 0.4 0.6 0.8 1 1.2
Deflection level (m)

Figure 5.13.  Histograms and fittings of the numbers of crossings.

Based on Rice formula, the down-crossing histograms and estimated

crossing rates of the dynamic and static histories were estimated as shown
in Figure 5.13, where x0 is the optimal starting point estimated by K-S test.
It is observed that the dynamic effect has a higher crossing rate, while the
mean value and the standard derivation have negligible difference. With
the fitted level-crossing models, extrapolations of the maximum deflec-
tion were estimated. Figure 5.14 shows the results of static and dynamic
extrapolations in a 1,000-year return period.
As observed from Figure 5.14, the maximum deflections over a return
period of 1,000-year are 1.445, 1.458, 1.464, and 1.475 m for the static
results, depicting “Good,” “Average,” and “Poor” roughness conditions,
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   111

1.5

Maximum deflection (m)

1.4

1.3

Static
1.2 Good
Average
Poor
1.1
0 200 400 600 800 1000
Return period (years)

Figure 5.14.  Extrapolations of the maximum deflections considering

the RRC.

respectively. It is observed that the lifetime dynamic ratio, a ratio between

dynamic and static extrapolations, for the poor road roughness condition
is 1.021. Therefore, even though the worse road roughness condition leads
to a larger number of level crossings, its influence on the maximum traffic
load effect extrapolation of a long-span bridge seems negligible.

5.6.3 LIFETIME MAXIMUM DEFLECTION ASSESSMENT

CONSIDERING INTERVAL TRAFFIC GROWTH

The European commission predicts a sustainable annual growth ratio

of truck traffic volume between 1.5% and 2% (European Commission
2008). Therefore, in the present study, the linear AGR of the traffic vol-
ume was assumed as 0%, 1%, 2%, and 3%, and the traffic was a free-
flow condition. The 100-year lifetime of the bridge was divided into 10
intervals, where the traffic volume was stationary and nongrowing. A total
­number of 1,000-day daily maxima for each interval was utilized to esti-
mate the level-crossing model as shown in Figure 5.15. It is observed that
the level-crossing rate is constant for the nongrowing traffic model, while
the level-crossing curves apparently move to the left considering the traffic
growth, and a higher traffic growth rate leads to a further shift. The traffic
growth for long-span bridges not only results in a large traffic volume, but
also leads to a more intensive traffic density on the bridge. Superposition
of both increases leads to a rapid growth of the extrapolation of the max-
imum deflection. This phenomenon is in agreement with the numerical
results presented in the verification examples.
112   •   BRIDGE RELIABILITY AND SERVICEABILITY

AGR=1% AGR=2%

Annual number of crossings

400 AGR=0% AGR=3%

300

200

100

0
100
80
60
40
20 1.5
0.9 1.2
0 0.3 0.6
Time (years) 0
Deflection level (m)

Figure 5.15.  Time-variant level-crossing rates accounting for traffic growth.

1.9
AGR=0
AGR=1%
Extrapolated deflection (m)

1.8 AGR=2%
AGR=3%
1.7

1.6

1.5

1.4
0 20 40 60 80 100
Service period (years)

Figure 5.16.  Extrapolation of the lifetime maximum deflection

accounting for traffic growth.

Based on the estimated level-crossing models, the extrapolation

of maximum deflection over a 1,000-year return period was estimated
according to Equations (5.6) and (5.7). Figure 5.16 plots the extrapola-
tions in the bridge lifetime accounting for the traffic growth. It is obvious
that the traffic growth leads to a rapid growth of the maximum deflection.
As a result, AGR = 3% leads to the extrapolation of the lifetime maximum
deflection increasing by 18%.
The probability of exceedance of a deflection limit is a serviceability cri-
terion for a bridge under traffic loading. The exceeding criterion was defined
as the maximum deflection of the quarter-point of the bridge girder crossing
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   113

8
AGR = 0
6 AGR = 1%

–log[–log(–Fmax(x,100)]
AGR = 2%
4 AGR = 3%

2 Rt =1000 years
0

-2

-4
1 1.2 1.4 1.6 1.8 2
(a) Deflection level (m)

100
L/400 = 2.05 m
10–2
Probability of exceedance

10–4

10–6
AGR=0
–8 AGR=1%
10
AGR=2%
AGR=3%
10–10
1.2 1.4 1.6 1.8 2 2.2
(b) Deflection level (m)

Figure 5.17.  Probabilistic assessment of the bridge deflection under

growing traffic loads: (a) CDFs plotted on Gamble paper; (b) probabil-
ity of exceedance.

the deflection limit specified in native/local specifications. Therefore, it is

important to define a threshold deflection for the prototype bridge under traf-
fic loading. As mentioned in the introduction, different design codes have
different limits, such as L/350 and L/800 for long-span bridges and for sim-
ply supported bridges, respectively (AASHTO 2010). In the present study,
the deflection limit was supposed to be L/400 according to China’s code [41].
The probability assessment of the maximum deflections is shown
in Figure 5.17, where Figure 5.17(a) shows the CDFs of the maximum
deflections in the lifetime estimated and plotted on the Gamble paper, and
Figure 5.17(b) shows the probabilities of exceedance of the deflection limit
over the 100-year period. The x-axis values at the cross point in Figure 5.17(a)
are similar to the value as shown in Figure 5.16. Therefore, the CDF and
the extrapolations are in agreement. The probabilities of exceedance at the
114   •   BRIDGE RELIABILITY AND SERVICEABILITY

x-axis value of 2.05 m in Figure 5.17(b) are 1.1×10−11, 2.0×10−9, 4.9×10−8,

and 2.7×10−7, for traffic growth ratio of 0%, 1%, 2% and 3%, respectively.
It is inferred that the traffic growth has a significant influence on the
probability of exceedance. Such influence is greater for a lower limit, but
weaker for a higher limit. In addition, the increase of the traffic growth
ratio leads to a higher increase rate of the probability of exceedance. This
phenomenon can be explained by the fact that the traffic growth not only
leads to a larger daily maxima of an individual truck weight, but also
increases the traffic density on the bridge. Therefore, a reasonable traffic
growth model is critical for evaluating the maximum traffic load effect of
long-span bridges in the lifetime.

5.7 CONCLUSIONS

The study presented in this chapter introduced a methodology for estimat-

ing the statistical extrapolation of traffic load effects on long-span bridges
in lifetime. Advancements have been made on several challenging issues
addressing the realistic consideration of the vehicle–bridge interaction, the
actual traffic pattern and the traffic growth. The actual traffic pattern was
embodied in stochastic traffic flows simulated based on weigh-in-motion
measurements. The continuously growing traffic loads were considered as
a series system composed of interval traffic loads. The nonstationarity of
the growing traffic load was captured in this model. The methodology was
verified by two numerical examples and subsequently was applied to the
lifetime maximum deflection extrapolation of a suspension bridge. The
following conclusions are drawn from the numerical studies:

1. In the verification examples, both the conventional and the inter-

val fittings closely match the daily maxima of the stationary and
nongrowing traffic loads. For the nonstationary and growing traf-
fic loads, however, the conventional fitting deviates the tail of the
maxima plotted on Gumbel paper, but the interval fitting of GEV or
Rice represents the tail fairly well. This phenomenon demonstrates
the capability of the interval traffic growth model in capturing the
nonstationarity of the growing traffic loads.
2. In the second verification example, the traffic growth leads to a
remarkable shift of the level-crossing rate and the daily maxima
plotted on the Gumbel paper. Not only does the traffic growth result
in a higher daily maximum GVW, but it also increases the traffic
density on the bridge, which is the main reason leading to the sig-
nificant increase in the extrapolated maximum deflection.
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   115

3. In the case study, the vehicle–bridge coupled vibration due to

the degradation of the road roughness condition leads to more
level-crossings of the bridge deflection, but it does not signifi-
cantly influence the mean value and the standard deviation of the
level-crossing rate. As a result, the lifetime dynamic assessment
ratio is less than 2.1%.
4. For the site-specific traffic condition of the suspension bridge, the
annual traffic growth rate of 3% leads to an 18% increase of the
extrapolated maximum deflection with a 1,000-year return period.
5. The traffic growth has a significant influence on the probability of
exceedance of the deflection limit. Such influence is more signif-
icant for a lower threshold deflection level, but it is minor for a
higher limit. In addition, a higher traffic growth rate leads to a rapid
increase of the probability of exceedance.

Although the proposed methodology was applied to the lifetime

maximum deflection extrapolation of a suspension bridge, it can also
be extended to be used for the extrapolation of the maximum bending
moment, the maximum cable force, as well as the longitudinal displace-
ment of other long-span bridges.
The findings from the presented study provide a basis for extension in
the following direction. Firstly, as an alternative approach of MCS, cellu-
lar automaton and Markov chain sampling can be utilized to simulate the
microscale behavior of vehicles, such as the change of the vehicle spac-
ing on the bridge, more efficiently. Secondly, improvements can be made
through focusing on congested traffic conditions rather than on free-flow
traffic conditions and respective adjustments of the model since congested
traffic conditions was found to be more critical to the maximum deflec-
tion. Thirdly, the approximation of the traffic load by removing the highly
proportioned lightweight cars from the WIM database was found as poten-
tially critical since it may result in the distortion of the vehicle spacing in
the simulated stochastic traffic flow. Finally, nonlinear stochastic analy-
sis can be considered in the proposed approach, but the computational
efficiency for dealing with the demanding nonlinear calculation is a key
problem.

REFERENCES

[1] Fu, G., and J. You. 2009. “Truck Loads and Bridge Capacity Evaluation
in China.” Journal of Bridge Engineering, 327–35. 10.1061/(ASCE)
BE.1943-5592.0000006
116   •   BRIDGE RELIABILITY AND SERVICEABILITY

[2] Zhu, S., D. Levinson, H. Liu, and K. Harder. 2010. “The Traffic and
Behavioral Effects of the I-35W Mississippi River Bridge Collapse.”
Transportation Research Part A: Policy and Practice 44, no. 10, pp. 771–84.
[3] Deng, L., and C.S. Cai. 2009. “Bridge Scour: Prediction, Modeling,
Monitoring, and Countermeasures—Review.” Practice periodical on
structural design and construction 15, no. 2, 125–34. 10.1061/(ASCE)
SC.1943-5576.0000041.
[4] Han, W., J. Wu, C.S. Cai, and S. Chen. 2014. “Characteristics and Dynamic
Impact of Overloaded Extra Heavy Trucks on Typical Highway Bridges.”
Journal of Bridge Engineering 20, no. 2, 05014011. 10.1061/(ASCE)
BE.1943-5592.0000666.
[5] Deng, L., W. Wang, and Y. Yu. 2016. “State-of-the-art Review on the
Causes and Mechanisms of Bridge Collapse.” Journal of Performance of
Constructed Facilities 30, no. 2. 10.1061/(ASCE)CF.1943-5509.0000731
[6] European Commission. Directorate General for Energy and Transport. 2008.
European Energy and Transport. Trends for 2030. Update 2007.
[7] Zhou, Y., and S. Chen. 2014. “Dynamic Simulation of a Long-span Bridge-
traffic System Subjected to Combined Service and Extreme Loads.”
Journal of Structural Engineering 141, no. 9, 04014215. 10.1061/(ASCE)
ST.1943-541X.0001188.
[8] Frangopol, D., A. Strauss, and S. Kim. 2008. “Bridge Reliability Assessment
Based on Monitoring.” Journal of Bridge Engineering 13, no. 3, 258–70.
10.1061/(ASCE)1084-0702.
[9] Li, S., S. Zhu, Y. Xu, Z. Chen, and H. Li. 2012. “Long-term Condition
Assessment of Suspenders Under Traffic Loads Based on Structural
Monitoring System: Application to the Tsing Ma Bridge.” Structural Control
and Health Monitoring 19, no. 1, pp. 82–101.
[10] Xia, M., C.S. Cai, F. Pan, and Y. Yu. 2016. “Estimation of Extreme Structural
Response Distributions for Mean Recurrence Intervals Based on Short-term
Monitoring.” Engineering Structures 126, pp. 121–32.
[11] Nowak, A.S. 1995. “Calibration of LRFD Bridge Code.” Journal of
Structural Engineering 121, no. 8, 1245–51. 10.1061/(ASCE)0733-9445.
[12] Kwon, O.S., E. Kim, and S. Orton. 2010. “Calibration of Live-load Factor
in LRFD Bridge Design Specifications Based on State-specific Traffic
Environments.” Journal of Bridge Engineering, 812–19. 10.1061/(ASCE)
BE.1943-5592.0000209.
[13] Caprani, C.C. 2012. “Calibration of a Congestion Load Model for
Highway Bridges Using Traffic Microsimulation.” Structural Engineering
International 22, no. 3, pp. 342–48.
[14] Enright, B., and E.J. OBrien. 2013. “Monte Carlo Simulation of Extreme
Traffic Loading on Short and Medium Span Bridges.” Structure and
Infrastructure Engineering 9, no. 12, pp. 1267–82.
[15] Guo, T., D.M. Frangopol, and Y. Chen. 2012. “Fatigue Reliability Assessment
of Steel Bridge Details Integrating Weigh-in-motion Data and Probabilistic
Finite-element Analysis.” Computers & Structures 112, pp. 245–57.
 LIFETIME DEFLECTIONS OF LONG-SPAN BRIDGES   •   117

[16] Lu N., M. Noori, and Y. Liu. 2016. “Fatigue Reliability Assessment of

Welded Steel Bridge Decks Under Stochastic Truck Loads via Machine
Learning.” Journal of Bridge Engineering 22, no. 1, 04016105. doi:10.1061/
(ASCE)BE.1943-5592.0000982
[17] O’Connor, A., and E.J. OBrien. 2005. “Traffic Load Modelling and Factors
Influencing the Accuracy of Predicted Extremes.” Canadian Journal of Civil
Engineering 32, no. 1, pp. 270–78.
[18] Caprani, C., E. OBrien, and G. McLachlan. 2008. “Characteristic Traffic
Load Effects from a Mixture of Loading Events on Short to Medium Span
Bridges.” Structural Safety 30, no. 5, pp. 394–404.
[19] OBrien, E., and B. Enright. 2012. “Using Weigh-in-motion Data to
Determine Aggressiveness of Traffic for Bridge Loading.” Journal of Bridge
Engineering 232–39. 10.1061/(ASCE)BE.1943-5592.0000368
[20] OBrien E., F. Schmidt, D. Hajializadeh, X. Zhou, B. Enright, C.C. Caprani,
S. Wilson, and E. Sheils. 2015b. “A Review of Probabilistic Methods of
Assessment of Load Effects in Bridges.” Structural Safety 53, pp. 44–56.
[21] Soriano, M., J. Casas, and M. Ghosn. 2016. “Simplified Probabilistic Model
for Maximum Traffic Load from Weigh-in-motion Data.” Structure and
Infrastructure Engineering 13, no. 4, 454–67. doi:10.1080/15732479.2016
.1164728
[22] OBrien E.J., A. Bordallo-Ruiz, and B. Enright. 2014. “Lifetime Maximum
Load Effects on Short-span Bridges Subject to Growing Traffic Volumes.”
Structural Safety 50, pp. 113–22.
[23] OBrien, E., D. Cantero, B. Enright, and A. Gonzalez. 2010. “Characteristic
Dynamic Increment for Extreme Traffic Loading Events on Short and
Medium Span Highway Bridges.” Engineering Structures 32, no. 12,
pp. 3827–35.
[24] Wang, F., L. Li, J.M. Hu, Y. Ji, R. Ma, and R. Jiang. 2009. “A Markov-
process Inspired CA Model of Highway Traffic.” International Journal of
Modern Physics C 20, no. 1, pp. 117–31.
[25] Chen, S.R., and J. Wu. 2010. “Dynamic Performance Simulation of Long-
span Bridge Under Combined Loads of Stochastic Traffic and Wind.” Journal
of Bridge Engineering, 219–30. 10.1061/(ASCE)BE.1943-5592.0000078
[26] OBrien, E., A. Lipari, and C. Caprani. 2015a. “Micro-simulation of Single-
lane Traffic to Identify Critical Loading Conditions for Long-span Bridges.”
Engineering Structures 94, pp. 137–48.
[27] Caprani, C., E. OBrien, and A. Lipari. 2016. “Long-span Bridge Traffic
Loading Based on Multi-lane Traffic Micro-simulation.” Engineering
Structures 115, pp. 207–19.
[28] Ruan, X., J. Zhou, and C. Caprani. 2016a. “Safety Assessment of the
Antisliding Between the Main Cable and Middle Saddle of a Three-pylon
Suspension Bridge Considering Traffic Load Modeling.” Journal of Bridge
Engineering 21, no. 10, 04016069. 10.1061/(ASCE)BE.1943-5592.0000927
[29] Ruan, X., J. Zhou, X. Shi, and C. Caprani. 2016b. “A Site-specific Traffic
Load Model for Long-span Multi-pylon Cable-stayed Bridges.” Structure
118   •   BRIDGE RELIABILITY AND SERVICEABILITY

and Infrastructure Engineering 13, no. 4, 494–504. doi:10.1080/15732479

.2016.1164724
[30] Calçada, R., A. Cunha, and R. Delgado. 2005. “Analysis of Traffic-induced
Vibrations in a Cable-stayed Bridge. Part II: Numerical Modeling and
Stochastic Simulation.” Journal of Bridge Engineering 10, no. 4, 386–97.
10.1061/(ASCE)1084-0702
[31] Chen, S., and J. Wu. 2011. “Modelling Stochastic Live Load for Long-
span Bridge Based on Microscopic Traffic Flow Simulation.” Computers &
Structures 89, no. 9, pp. 813–24.
[32] Cai, C., J. Hu, S. Chen, Y. Han, W. Zhang, and X. Kong. 2015. “A Coupled
Wind-vehicle-bridge System and its Applications: A Review.” Wind and
Structures 20, no. 2, pp. 117–42.
[33] Caprani, C. 2013. “Lifetime Highway Bridge Traffic Load Effect from
a Combination of Traffic States Allowing for Dynamic Amplification.”
Journal of Bridge Engineering 18, no. 9, 901–09. 10.1061/(ASCE)
BE.1943-5592.0000427
[34] Zhang, N., and H. Xia. 2013. “Dynamic Analysis of Coupled Vehicle–bridge
System Based on Inter-system Iteration Method.” Computers & Structures
114, pp. 26–34.
[35] Rice, S.O. 1945. “Mathematical Analysis of Random Noise.” The Bell
System Technical Journal 24, no. 1, pp. 46–156.
[36] Ditlevsen, O. 1994. “Traffic Loads on Large Bridges Modeled as White-
noise Fields.” Journal of Bridge Engineering 120, no. 4, 681–94. 10.1061/
(ASCE)0733-9399
[37] Beck, A.T., and R. Melchers. 2004. “On the Ensemble Crossing Rate
Approach to Time Variant Reliability Analysis of Uncertain Structures.”
Probabilistic Engineering Mechanics 19, no. 1, pp. 9–19.
[38] Cremona, C. 2001. “Optimal Extrapolation of Traffic Load Effects.”
Structural Safety 23, no. 1, pp. 31–46.
[39] Zhou, X., F. Schmidt, F. Toutlemonde, and B. Jacob. 2016. “A Mixture
Peaks Over Threshold Approach for Predicting Extreme Bridge Traffic Load
Effects.” Probabilistic Engineering Mechanics 43, pp. 121–31.
[40] Liu, Y., Y. Deng, and C.S. Cai. 2015. “Deflection Monitoring and Assessment
for a Suspension Bridge Using a Connected Pipe System: A Case Study in
China.” Structural Control and Health Monitoring 22, no. 12, pp. 1408–25.
[41] Ministry of Communications and Transportation (MOCAT). 2004. “General
Code for Design of Highway Bridges and Culverts.” JTG D60-2004, Beijing,
China.
[42] Yin, X., Z. Fang, and C. Cai. 2011. “Lateral Vibration of High-pier Bridges
Under Moving Vehicular Loads.” Journal of Bridge Engineering 16, no. 3,
400–12. 10.1061/(ASCE)BE.1943-5592.0000170
 CHAPTER 6

System Reliability
Evaluation of in-Service
Cable-Stayed Bridges
Subject to Cable
Degradation Via a Machine
Learning Based Tool

6.1 INTRODUCTION

A cable-stayed bridge is a variant of a suspension bridge and is more

economic than other types of bridges for a medium- to long-span design
scheme [1]. One of the advantages of cable-stayed bridges is the stiffness
provided by cables directly connecting bridge decks and pylons. However,
the cables that are critical components holding external loads are vul-
nerable to atmospheric corrosion and fatigue damage [2]. For instance,
Mehrabi et al. [3] indicated that 39 out of 72 cables of the Hale Bogges
bridge critically were in need of repair or replacement after past 25 years
of service. Cable degradation increases the risk of the cable rupture that
may lead to the failure of other cables, girders, and even collapse of the
entire structure. This phenomenon is summarized as the term of progres-
sive collapse that is emphasized in the design of a cable-stayed bridge
[4]. Even through recommendations for robustness were provided in the
design codes of cable-stayed bridges, uncertainty induced reliability eval-
uation of cable-stayed bridges, especially subject to cable degradation or
cable loss, needs further investigation.
120   •   BRIDGE RELIABILITY AND SERVICEABILITY

Mechanical behavior investigation of cable-stayed bridges subject to

cable corrosion or cable rupture has becomes an important research topic
in recent years. Cao et al. [5] indicated that the ultimate load capacity
of corroded cables was correlated with the nonuniform reduction of the
cross-sectional area. Wolff and Starossek [6] emphasized the dynamic and
nonlinear behaviors of a cable-stayed bridge subject to sudden rupture of
a stay cable. Xu and Chen [7] investigated the mechanical behaviors of
corroded wires with presumed corrosion distribution at the cable cross
section utilizing a series–parallel system. Both numerical and experi-
mental studies on the stress acting during the failure of a stay cable of a
cable-stayed bridge were conducted by Mozos and Aparicio [8, 9]. The
cable loss incident induced dynamic behavior of long-span cable-stayed
bridges under traffic loads was investigated by Zhou and Chen [10, 11]
utilizing a vehicle–bridge–wind coupled vibration model. Aoki et al. [12]
presented a robust design for avoiding the propagation of the cable loss
that is essential for cable stayed bridge design.
In addition to the research progress on the mechanical behaviors illus-
trated earlier, some researchers conducted reliability evaluation of cable-
stayed bridges taking into account uncertainties of structural parameters and
external loads. Cheng and Xiao [13] studied the reliability of the service-
ability of cable-stayed bridges integrating the response surface method and
the finite element method. Li et al. [14] utilized structural health monitoring
data to evaluate the reliability of a long-span bridge. Since cable-stayed
bridges are statically indeterminate structures, the system reliability theory
attracted the attention of researchers. In this regard, Bruneau [15] utilized a
system reliability method to analyze the ultimate global behavior of a cable-
stayed bridge and found nine potential failure patterns. Estes and Frangopol
[16] presented a system model by combining ultimate and serviceability
limit state of a highway bridge. Liu et al. [17] developed an adaptive sup-
port vector regression (ASVR) approach for system reliability evaluation of
complex structures including cable-stayed bridges. However, to the best of
the authors’ knowledge, research progress on system reliability evaluation
of cable-stayed bridges subject to cable degradation is relatively insuffi-
cient. The critical reasons leading to the insufficiency in this regard might be
the complexity of the structural failure modes and the additional computa-
tional effort resulting from cable degradation. Even though the conventional
approaches, such as the β-unzipping method [18], the branch-and-bound
method [19], and the selective searching approach [20] are accurate and
efficient, their applications for long-span bridges are relatively new.
This study aims at evaluating system reliability of in-service cable-
stayed bridges subject to cable degradation. First, strength degradation
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   121

models of parallel wire cables are presented with the consideration of a

parallel–series system. Subsequently, a computational framework and a
computer program integrating the intelligent machine learning technol-
ogy is presented. Finally, two cable-stayed bridges, including an ancient
short-span bridge and a modern long-span bridge, are selected as proto-
types to investigate the influence of the cable degradation on the structural
system reliability. Influence of both fatigue and corrosion induced cable
degradation on the structural failure sequences and system reliability are
investigated.

6.2 FORMULATIONS OF CABLE DEGRADATION

6.2.1 STRENGTH MODEL OF PARALLEL WIRE CABLE

Both parallel wires and strands are conventional types of stay cables in
bridge engineering. A parallel wire cable consists of straight parallel round
wires inside a polystyrene pipe. In addition to the material characteristics,
the strength of parallel wires is also associated with the length of wires,
the number of wires, and the cable degradation [21]. Initially, this study
conducts mathematical formulation of cable degradation due to the effects
of cable length and the number of wires.
In general, a parallel wire cable can be modeled in a parallel–series
system as shown in Figure 6.1, where each wire can be simulated as
a series system and the wires work together as a parallel system. For the
series system, the individual wires of a stay cable can be simulated with
correlative elements depending on the length of the wire. The material
properties and defects in the wires are considered by a correlation length
defined as L0 in the series system. The cable strength decreases result-
ing from a shorter correlation length or a longer wire length. Therefore,
both the correlation length and the wire length should be considered in the
cable strength model. Distribution function of the strength of a wire can be
written by a Weibull distribution function

  zk 
FZ ( z ) = 1 − exp  − l    (6.1)
  u  

where z is the strength of a wire, λ, u, and k are the parameters in the

Weibull distribution that can be estimated from ultimate capacity tests and
the maximum likelihood method. The strength of the wire can be expressed
122   •   BRIDGE RELIABILITY AND SERVICEABILITY

In
pa
ral
0m

el l
10
L=
.5m
=0
L0

s
erie
Ins

λ=L/L0=200 (Corroded wires)

Figure 6.1.  A parallel–series system of a stay cable.

by a function of the scale factor, λ, which is the ratio between the length
of the wire specimen and the correlation length. An experimental study of
a 100 m long wire conducted by Faber et al. [22] indicated that the mean
values of the wire strength for undamaged cable (λ = 3) and the corroded
wire (λ = 200) were 1748 and 1650 MPa, respectively. The variability of
the wire strength is negligible according to Faber’s conclusion.
In addition to the series system of an individual wire, a stay cable con-
sists of numerous wires in parallel as shown in Figure 6.1. In the parallel
system, the increase of the number of wires in a cable leads to a reduction
of the mean strength of each wire, which is the so-called Daniel’ effect.
The general reduction can be up to about 8%, and the deviation of the
Daniel’s effect is negligible.

6.2.2 DEGRADATION OF THE CABLE STRENGTH

Since high stress cables are prone to be corroded [23], the stay cables became
rusty and fractured under long-term influence of corrosion and cyclic
stresses. Thus, cable degradation is a common phenomenon in existing
cable-supported bridges. Deterioration of cable wires takes different forms
including stress corrosion cracking, pitting, corrosion fatigue and hydrogen
embrittlement, which compromise the strength and ductility of wires lead-
ing to a reduced service life of bridge cables [24]. This study considers cable
degradation resulting from atmospheric corrosion and fatigue damage.
In the concept of fatigue damage accumulation, the failure times of
the undamaged wires under a mean stress range can be assumed to be
identical and independent. The wire with the smallest failure times breaks
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   123

firstly in the system. Subsequently, the remaining wires have stress redis-
tributions. Therefore, the probability distribution function of the initial
failure times is written as [25]

 ∆S a a

  eq   N  m 
FN ( Seq , N ) = 1 − exp  (6.2)
 rc   K  
 

where ΔSeq and N are the equivalent stress range and the number of stress
cycles, respectively; α, m, and K are unknown coefficients that can be
estimated by experimental tests, rc is a parameter associated with the
cross-sectional area of the cable. Faber et al. [26] utilized the ultrasonic
inspection technology to investigate the fatigue stress degradation, and
then provided the degradation functions for the main cables, hanger
cables, and stay cables for non-corroded and corroded wires. Based on
Faber’s results, the strength coefficients of non-corroded and corroded
wires were interpolated for 20-year service period as shown in Figure 6.2.
In Figure 6.2, the non-corroded cable is associated with only fatigue
damage, and the corroded cable is associated with both fatigue and cor-
rosion. It is observed that the strength coefficients of the stay cable in the
20-year service period resulting from fatigue and fatigue-corrosion effects
are 0.928 and 0.751, respectively. In addition, the curve for the fatigue
effect is close to linear, while the curve accounting for fatigue-corrosion
effect is nonlinear. Therefore, it is obvious that corrosion results in a rapid
decrease of the cable strength.

0.95
Strength coefficient

0.9

0.85 y1= –1.5×10–5x2–3.2×10–3x+0.998

0.8 y2= –4.7×10–4x2–2.4×10–3x+0.996

0.75 Non-corroded
Corroded
0.7
0 5 10 15 20
Service period (years) s

Figure 6.2.  Strength coefficients of a stay cable due to

fatigue and corrosion effects.
124   •   BRIDGE RELIABILITY AND SERVICEABILITY

6.3 FRAMEWORK OF SYSTEM RELIABILITY

EVALUATION

6.3.1 CHARACTERISTICS OF THE STRUCTURAL SYSTEM

OF CABLE-STAYED BRIDGES

A cable-stayed bridge is a complex system consisting of multiple compo-

nents connected in series or in parallel. These components work together
as a system to support the external loads. As the span of the bridge
increases, the mechanical behavior and the failure mode exhibit unique
characteristics that influence the system safety of cable-stayed bridges. In
addition to the component-level behaviors, such as the cable slag effect
and the nonlinear behaviors [26], the system-level behaviors should be
further considered.
First, due to the high-stress in the stay cables, both the pylon and
the girder behave as beam–column components. Therefore, the beam–
column effect is a significant characteristic for the structural system. The
relationship between the bending moment and axial force will affect the
component stiffness coefficient and the internal forces. The beam–column
interaction is a second-order effect and can be conveniently considered
by utilizing stability functions [27]. Assuming a hollow rectangular sec-
tion, where the neutral axis in the ultimate stays within the webs, from the
sample plastic analysis, the axial bending interaction curve can be written
as [28]

2
M  P  A2
= 1−   (6.3)
MP  PP  4 wZ x

where M is the applied moment, MP is the plastic moment capacity in the

absence of axial loads, P is the applied axial force, PP is the plastic axial
force capacity in the absence of applied moment, w is the web thicknesses,
and Zx is the bending plastic modulus.
Subsequently, since the cable stayed bridge is statistically an indeter-
minate system consisting of girders, cables and pylons, their failure modes
and failure sequences directly influence the system reliability. In general,
the potential failure modes of a cable-stayed bridge are bending failure of
towers and girders, strength failure of cables, and the stability failure of
the pylons. The conventional approach of identifying the failure sequences
is associated with the select search methods. For non-brittle failure, such
as the rupture of a cable, the subsystem is generated by deleting the failed
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   125

cable. For the ductile failure, such as the bending failure of the girders or
towers, the subsystem is generated by adding a plastic hinge at the failure
location. One of the search techniques for the potential failure components
is the branch-and-bound method that is the so-called β-unzipping method
[29]. Suppose a system consists of n components, and k-1 components
denoted as r1, r2,…, rk-1 failed. The conditional reliability index of the kth
component is written as:

( )
br(kk/) = br(kk/)r1, r2,rk −1 = Φ −1  P Er(kk/)  (6.4)

where Er(kk/) is the event of the kth component failure at the kth failure
phase, P() denotes the probability of the event, Φ −1 () is an inverse cumu-
lative distribution function, and br(kk/)r1, r2,rk −1 is a conditional reliability
index of the kth potential component at the kth failure phase. The condi-
tion to be the potential failure component is [30]

β r(kk/) = β min
(k )
+ ∆β (6.5)

where bmin (k )
is the minimum reliability index at the kth failure phase,
and Δβ equals to 3 and 1 at the first and the following searching phases,
respectively.
As observed from the previous reasoning, the beam–column effect is
considered in Equation (6.4), and the failure sequence search criteria is
shown in Equations (6.4 and 6.5). In addition to these unique characteris-
tics illustrated earlier, the cable degradation will add to the time-­consuming
computation of the system reliability. Therefore, special attention should
be paid for utilizing an efficient computational framework.

6.3.2 PROPOSED FRAMEWORK FOR SYSTEM RELIABILITY

EVALUATION OF CABLE-STAYED BRIDGES

Due to the characteristics and cable degradation of the cable-stayed bridge,

an efficient framework should meet the following requirements. First,
since the component failure probability is extremely small, the ­computing
accuracy is an essential requirement. The conventional approaches, such
as the first order second moment (FOSM) approach and the response sur-
face method (RSM) are not appropriate for solving this issue. Second,
since the search for dominant failure sequence is a time-consuming pro-
cess, computing efficiency is another essential requirement. Based on the
preceding formulations, this study utilized a machine learning approach
126   •   BRIDGE RELIABILITY AND SERVICEABILITY

based on an ASVR proposed by Liu et al. [30]. The framework of the

intelligent algorithm was improved for the special application of taking
into account the cable degradation. The flowchart of the framework is
shown in Figure 6.3.
As depicted in Figure 6.3, the main procedures in the flowchart con-
sist of two aspects including the system reliability evaluation based on the
ASVR approach and the updating of the cable degradation model. Details
of the applications of the SVR and ASVR approaches in structural sys-
tem reliability evaluation can be found in Dai et al. [31] and Liu et al.
[30]. The discussion presented herein mainly describes the procedures
associated with the cable degradation. Initially, the cable strengths are
taken as initial values to carry out the system reliability evaluation step
by step. Subsequently, the cable strength model is updated as shown in
Figure 6.2, and then the component and system reliability are reevaluated.
It is worth noting that the event tree should be rebuilt under the case that

Statistics of random variables

Sampling uniformly distributed

training samples
Finite-element model
SVR models approximated by
learning machines

Component reliability
evaluations
Updating the finite-
Cable degradation
element model
model as shown in
Removing the potential failed
Fig. 2
components separately

No
Is the new structure failed?

Yes
System reliability evaluation
based on the parallel-series
model
No, then
T=T+1
T>Ts ?
Yes
Lifetime sytem reliability

Figure 6.3.  Flowchart of the proposed computational framework.

 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   127

cable strength has changed, because the cable degradation may change the
potentially failed components that form the failure sequence. Finally, if the
service period, Ts, is reached, the entire procedure will stop and the life-
time system reliability indices are generated as output. The cable degrada-
tion induced system reliability decreases are reflected by the time-variant
system reliability indices.
The critical step depicted in Figure 6.3 is to return to the sampling
training samples after the cable strength is updated. This procedure sug-
gests that it is inappropriate to only update the reliability index of indi-
vidual cables without reevaluating the failure sequences. Instead, the
seemingly additional computational effort is essential to capture the main
failure sequences. Such deduction will be demonstrated in a case study.

6.4 PROGRAM IMPLEMENTATION OF THE

FRAMEWORK

As mentioned earlier, the introduction of the cable degradation leads

to additional computational efforts. In order to make the computation
more efficient, a graphical user interface (GUI)-based program entitled
“Complex Structural Reliability Analysis software V 1.0” CSRA (2013)
[32] was developed based on the framework. This general program was
developed based on two commercial programs, namely, MATLAB and
ANSYS. The main procedures of the CSRA program are summarized in
Figure 6.4.
In Figure 6.4, a Data Processing System (DPS) [33] is utilized to gen-
erate uniformly distributed samples that will be used for training SVR
models. ANSYS is recommended as the commercial finite-element pro-
gram because of its good connection to MATLAB. The LIBSVM (Library

CSRA

DPS Commercial FE program LIBSVM MCS β-bound function

Sampling training Computing structural Reliability Updating the FE

responses as output SVR models
samples evaluation model
data

Structural event tree

Figure 6.4.  Flowchart of the complex structural reliability analysis software.

128   •   BRIDGE RELIABILITY AND SERVICEABILITY

for Support Vector Machines) [34] is a MATLAB program package. The

MCS (Monte Carlo Simulation) can be the direct MCS or an advanced
MCS. The β-bound function refers to Equations (6.4 and 6.5). At the end
of the route, the finite-element model is updated, and then the component
reliability is reevaluated step by step. Eventually, the system reliability
can be evaluated in a series–parallel system.

6.5 CASE STUDIES

6.5.1 A SHORT-SPAN CABLE-STAYED BRIDGE

An ancient short-span cable-stayed bridge shown in Figure 6.5 is utilized

herein to investigate the influence of the cable degradation on the struc-
tural system reliability. The cable stayed bridge has a single pylon, and 2
stayed cables in each side. The distances between the cable anchors in the
girders or in the towers are 30 m. More details regarding the material and
sectional properties and performance functions can be found in Bruneau
[15]. In this case study, the structura’s mechanical behavior is assumpted
as linear and elastic to be in accordance with Bruneau.
In general, the cables are considered as brittle since the rupture of a stay
cable is momentary. The concrete girders and towers for long-span bridges
were considered as ductile since the prestressed structures were allowed
to have large deflections. The structural system failure was defined by a
plastic collapse mechanism. The plastic failure mechanism was identified
by the plastic-hinge locations and plastic capacities. The potential failure
locations are shown in Figure 6.5. The points G1~G11 of the girders and
the point of T1 and T2 of the pylon are critical to bending failure, and the
points C1~C4 of the stay cables are critical to sudden rupture.
In the system level point of view, if the cable is ruptured, the
cable is directly removed and the load-carrying capability of the new

Rupture
15m×2

C1 T1 C4
C2 C3 Plastic change
T2
G1 G2 G3 G4 G5 G6 G7 G8 G9 G10 G11
25m

15m×12

Figure 6.5.  Dimensions and failure modes of a short-span cable-stayed

bridge.
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   129

s­ tructure is reevaluated. If the bending failure of a girder or a pylon

occurs, a ­plastic hinge is added at the location where the bending failure
occurs. It is acknowledged that the structural stiffness and resistance
are changing at any step, which means the remaining structural ele-
ments reform a new structural system. At the end of the process the
progression along the ­failure sequences will be stopped in case of the
failure probability of the final component is expected to be extremely
high. Note that the ­process should be stopped for saving computational
effort, even though the structure still has the load-carrying capability.
Since Bruneau (1992) has provided the explicit limit state functions,
the ASVR approach was not utilized herein to approximate and update
the bridge model. The strength coefficients of cable shown in Figure 6.1
were directly updated according to the limit state functions provided
by Bruneau (1992). Based on the ­preceding assumptions, the event
trees for of the bridge accounting for nondegradation and degradation
coefficient of 20% were evaluated. Figure 6.6 shows the results of the
evaluation.
The following conclusions can be drawn from Figure 6.6. First, as the
cable strength decreases by 20%, the probability of failure of the C2 cable
rapidly decreases from 0.154×10−7 to 0.243×10−4. Second, the dominant
failure sequence has changed, where the initial domain failure mode starts
at the hinges at G10 and G2, while the cable degradation results in the
C2 cable failure as the start of the domain failure sequence followed by
bending failure at the G6 girder. Finally, the probability of failure of the
structural system increases from 1.53 × 10−6 to 44.6×10−6. As elaborated
earlier, not only the cable degradation decreases the reliability of a cable,
but also has a significant impact on the structural dominant failure mode
and the system reliability.
In order to investigate the influence of the cable degradation on the
system reliability of the bridge, the cable degradation model shown in
Figure 6.1 was utilized to update and to reevaluate the system reliability.
Figure 6.7 plots the results of the time-variant system reliability of the
bridge in the 20-year service period.
It is observed from Figure 6.7 that the system reliability indices have
both similar and different tendencies compared with the cable strength
models shown in Figure 6.2. The similar aspect is that the reliability index
accounting for fatigue and corrosion decreases rapidly compared to that
caused by fatigue. The different aspect is that the fatigue corrosion leads
to a sudden decrease of the reliability index beginning from the 13-year
­service period. This phenomenon can be explained by the fault tree shown
in Figure 6.6, where the cable failure becomes the dominant failure mode
130   •   BRIDGE RELIABILITY AND SERVICEABILITY

Figure 6.6.  Event trees of the cable-stayed bridge: (a) without cable degrada-
tion; (b) with a cable degradation coefficient of 20%.
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   131

Figure 6.6. (Continued)

of the system failure as the cable strength decreases to a critical value. It

can be concluded that in the 13-year service period, the probability of fail-
ure of the cable accounting for both fatigue and corrosion is larger than that
of the hanger at G10 point, which is the former domain failure component.
132   •   BRIDGE RELIABILITY AND SERVICEABILITY

Reliablity index 4.5

3.5

2.5 Fatigue
Fatigue and Corrosion
2
0 5 10 15 20
Service period (years)
Figure 6.7.  System reliability index of the cable-stayed bridge subject
to cable degradation.

Therefore, the continuous cable degradation has resulted in a change of

the domain failure component from the girder to the cable. Such transfor-
mation causes a sudden decrease of the system reliability index. However,
such phenomenon is not observed from the fatigue-induced reliability
decrease, because the cable strength has not decreased to the critical value.

6.5.2 A LONG-SPAN CABLE-STAYED BRIDGE

A long-span cable-stayed bridge [30], crossing the Yangzi River at Sichuan,

China, is selected herein as a prototype to investigate the influence of cable
degradation to long-span cable-stayed bridges. The bridge has two pylons
with double-sided cables following a fan pattern. The pylon and segmental
girders are connected by 34 pairs of cables in double sides. Dimensions of
the bridge are shown in Figure 6.8, where CS1 and Cm1 denote the 1st pair of
cables in the side-span and the mid-span, respectively; G1C and GC1 denote
the 1st pair of girders in the side-span and the mid-span, respectively; and
P1, P2, and P3 denote the critical points of the pylon and the girder.
The physical properties of the parallel steel wires are: the diameter
φ = 7 mm, the ultimate tensile strength fb = 1770 MPa, the modulus of
elasticity Es = 1.9×105 MPa, and the density γs = 78 kN/m3. The physical
properties of the concrete girders and pylons are: the ultimate compressive
strength fck = 50MPa, the modulus of elasticity Ec = 3.45×104 MPa, and
the density γc = 26 kN/m3. The modulus of elasticity of the cables was
estimated by utilizing the Ernst’s equation. The physical properties of the
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   133

Degration
210 m 420/2 m
7m 32×6 m 32×6 m+1 m 30.3 m

141

77 m
1 34
1
Cs34 Cs P1 Cm Cm
P3 P1

40 m 37 m
P2
34
Gs
1
Gs Gm
1 34
Gm
Symmetric P2

(a) (b)

Figure 6.8.  Dimensions of the long-span cable-stayed bridge:

(a) elevation layout; (b) longitudinal layout.

fourth longest cables are shown in Table 6.1, where Ns is the number of
wires in a stay cable, As is the cross sectional area of the stay cable, Es,Ernst
is the equivalent modulus of elasticity of the stay cable, P0 is the cable
force, and Pb is the strength of the cable. More physical properties of the
parallel steel wires can be found in MOCAT [35]. The random variables
include the modulus of elasticities, equivalent densities, cross-sectional
areas and the strength of the girders, pylons and cables. More details on
the stochastics of the bridge can be found in Liu et al. [30].
A finite element model shown in Figure 6.9 was built based on the com-
mercial program ANSYS, where the cables were simulated by LINK180
elements, and the girders and pylons were simulated by BEAM188 ele-
ments. The traffic loads were considered as uniformly distributed forces
in the mid-span. The critical failure components are the longest cables
according to Liu et al. [30]. Taking the Cm31, Cm32, and Cm33 cables and the
P2 and P3 girders as examples, their dynamic response under the sudden
failure of the Cm34 cable were analyzed. The response histories are shown in
Figure 6.10. As observed from Figure 6.10, both the cable forces and the

Table 6.1.  Properties of the fourth longest stay cables

Es,Ernst
Symbols Ns As (m2) (MPa) P0 (kN) Pb (kN)
Cs34, Cm34 253 9.73×10−3 1.864×105 6440 17235
Cs33, Cm33 241 9.28×10−3 1.860×105 6066 16417
Cs32, Cm32 241 9.28×10−3 1.852×105 5665 16417
Cs31, Cm31 223 8.58×10−3 1.841×105 5580 15191
134   •   BRIDGE RELIABILITY AND SERVICEABILITY

Figure 6.9.  Finite element model of the cable-stayed bridge.

bending moments increase and fluctuate under the case of sudden failure
of the Cm34 cable. The dynamic amplification factor for the maximum cable
force and the maximum bending moment of the girder are 1.03 and 1.09,
respectively. In addition, the components close to the failure cables have
stronger responses.
With the deterministic analysis results from the finite-element
model, the response surface function can be approximated by the learn-
ing machines. In this case, the implicit performance functions were for-
mulated by the SVR model. In addition, the SVR model can be updated
by removing the failed components and reevaluating the response of the
new structure. Response function of the Cm33 cable force associated with
the live load, the cross-sectional area of cables and the Cm34 cable rupture
are shown in Figure 6.11, where FC is the cable force of Cm33 cable, Q is
33
m

the uniformly distributed forces of the live load in the mid-span, and As is
the cross-sectional area as shown in Table. 6.1.
Based on the proposed framework, the event trees of the bridge were
established accounting for the fatigue and corrosion effects shown in Figure
6.1. Since there are 34 pairs of cables for each side of a pylon and they have
high correlation factors, only the longest cables including Cm34 and Cs34 were
selected for the first layer of the event tree to simplify the computation. Two
event trees of the bridge in different service periods are shown in Figure 6.12.
As depicted in Figure 6.12, there are two dominant failure sequences.
The first failure sequence is the strength failure of the mid-span cables (Cm34
and Cm33) followed by the bending failure of the mid-span girder (Hinge @
P3). The second failure sequence is the strength failure of the side-span
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   135

7000
33
C
m
6500 32
C
m

Cable force (kN)

6000 31
C
m

5500

5000
0 5 10 15 20 25
Time (s)
(a)

-2

-3 P
3
Bending moment (MN.m)

-4

-5 P
2

-6

-7
0 5 10 15 20 25
Time (s)
(b)

Figure 6.10.  Response histories of the critical points subject

to sudden rupture of the Cm34 cable: (a) cable forces; (b) bending
moments.

34
After C m ruptures
6800

6700
F 33 (kN)

6600
m
C

6500
34
Before C ruptures
6400 m
80
12
70
11
60 10
50 7
Q (kN.m-1) 8 As (×10-3 m2)

Figure 6.11.  Response surfaces of the Cm33 cable force due to Cm34 rupture.
136   •   BRIDGE RELIABILITY AND SERVICEABILITY

Figure 6.12.  Event trees of the long-span cable-stayed bridge:

(a) T = 0 year; (b) T = 20 years.

cables (Cs34 and Cs33) followed by the bending failure of the pylons (Hinge
@ P2). Accounting for the 6 failure sequences and the symmetry of the
bridge, the system reliability indices of the bridge are 7.60 and 6.41 for the
T = 0 and T = 20 years. The system reliability index of the bridge in the 40
years of service period is shown in Figure 6.13.
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   137

System reliability index

6
Target reliability index

5 Designed life of a cable

4 Fatigue
Fatigue and corrosion
3
0 10 20 30 40
Service period (years)

Figure 6.13.  System reliability indices of the cable stayed bridge

subject to cable degradation.

It is observed from Figure 6.13 that the system reliability index of

the bridge declines rapidly in the service period due to the cable degrada-
tion resulting from the fatigue-corrosion effect. According to the reliabil-
ity design specification of China (1999) [36], the bridge reliability index
under the control of brittle failure should be greater than 5.2. Given the
target reliability index βtarget = 5.2, a reasonable service period of the cables
is within 29 years as observed from Figure 6.13. Corrosion leads to the
rapid decrease of the cable strength as well as the system reliability indi-
ces compared to fatigue. Since corrosion leads to the rapid decrease of the
cable strength as well as the structural system reliability, implementing
effective anti-corrosion measures and replacing deeply corroded cables
are necessary for ensuring lifetime system safety of cable-stayed bridges.

6.6 CONCLUSIONS

Cable degradation due to fatigue damage and atmospheric corrosion influ-

ences the structural system safety of cable-stayed bridges in lifetime. This
study presented a framework for system reliability evaluation of cable-
stayed bridges subject to cable degradation. The cable degradation was
modeled in a parallel–series system taking into account both fatigue and
corrosion. Two cable-stayed bridges, including an ancient short-span
bridge with long-distance (30 m) cables and a modern long-span bridge
with short-distance (6 m) cables, were selected as prototypes to investi-
gate the influence of cable degradation on the structural system reliability.
For the short-span bridge, the dominant failure sequence changed
from the bending failure of girders and pylons to the strength failure of
138   •   BRIDGE RELIABILITY AND SERVICEABILITY

cables due to the cable corrosion. As a result, the structural system reli-
ability experiences a significant decrease in the period that the cable reli-
ability is inferior to the one of the critical girder. For the long-span bridge,
the system reliability index decreases to the threshold value in the 29-year
service period taking into account both fatigue and corrosion. Since the
corrosion leads to the rapid decrease of the cable strength as well as the
structural system reliability, implementing effective anti-corrosion mea-
sures and replacing deeply corroded cables are necessary for ensuring life-
time system safety of cable-stayed bridges.
There still remain some challenges in the reliability assessment of
long span bridges that require further study. First, site-specific measure-
ments of the cable strength need to be utilized for establishing a more
reasonable cable degradation model. Second, a more complete event tree
of the cable-stayed bridge instead of taking the longest cables as the first
order, as adopted in the present study, should be formulated. Finally, cor-
relation coefficients need to be considered to make the estimation of the
system reliability more accurate.

REFERENCES

[1] Yang, O., H. Li, J. Ou, and Q.S. Li. 2013. “Failure Patterns and Ultimate
Load-carrying Capacity Evolution of a Prestressed Concrete Cable-stayed
Bridge: Case Study.” Advances in Structural Engineering 16, no. 7,
pp. 1283–96.
[2] Li, H., C.M. Lan, Y. Ju, and D.S. Li. 2011. “Experimental and Numerical
Study of the Fatigue Properties of Corroded Parallel Wire Cables.” Journal
of Bridge Engineering 17, no. 2, pp. 211–20.
[3] Mehrabi, A.B., C.A. Ligozio, A.T. Ciolko, and S.T. Wyatt. 2010. “Evaluation,
Rehabilitation Planning, and Stay-cable Replacement Design for the Hale
Boggs Bridge in Luling, Louisiana.” Journal of Bridge Engineering 15,
no. 4, 364–72. 10.1061/(ASCE)BE.1943-5592.0000061
[4] Marjanishvili, S.M. 2004. “Progressive Analysis Procedure for Progressive
Collapse.” Journal of Performance of Constructed Facilities 18, no. 2, pp. 79–85.
[5] Cao, Y., G.W. Vermaas, R. Betti, A.C. West, and P.F Duby. 2003. “Corrosion and
Degradation of High-strength Steel Bridge Wire.” Corrosion 59, no. 6, pp. 547–54.
[6] Wolff, M., and U. Starossek. 2009. “Cable Loss and Progressive Collapse in
Cable-stayed Bridges.” Bridge Structures 5, no. 1, pp. 7–28.
[7] Xu, J., and W. Chen. 2013. “Behavior of Wires in Parallel Wire Stayed Cable
Under General Corrosion Effects.” Journal of Constructional Steel Research
85, pp. 40–47.
 SYSTEM RELIABILITY OF CABLE-STAYED BRIDGES   •   139

[8] Mozos, C.M., and A.C. Aparicio. 2010. “Parametric Study on the Dynamic
Response of Cable Stayed Bridges to the Sudden Failure of a Stay, Part I:
Bending Moment Acting on the Deck.” Engineering Structures 32, no. 10,
pp. 3288–300.
[9] Mozos, C.M., and A.C. Aparicio. 2011. “Numerical and Experimental Study
on the Interaction Cable Structure During the Failure of a Stay in a Cable
Stayed Bridge.” Engineering Structures 33, no. 8, pp. 2330–41.
[10] Zhou, Y., and S. Chen. 2014. “Time-progressive Dynamic Assessment of
Abrupt Cable-breakage Events on Cable-stayed Bridges.” Journal of Bridge
Engineering 19, no. 2. 159–71. 10.1061/(ASCE)BE.1943-5592.0000517
[11] Zhou, Y., and S. Chen. 2015. “Numerical Investigation of Cable Breakage
Events on Long-span Cable-stayed Bridges Under Stochastic Traffic and
Wind.” Engineering Structures 105, pp. 299–315.
[12] Aoki, Y., H. Valipour, B. Samali, and A. Saleh. 2013. “A Study on Potential
Progressive Collapse Responses of Cable-stayed Bridges.” Advances in
Structural Engineering 16, no. 4, pp. 689–706.
[13] Cheng, J., and R.C. Xiao. 2005. “Serviceability Reliability Analysis of Cable-
stayed Bridges.” Structural Engineering and Mechanics 20, no. 6, pp. 609–30.
[14] Li, H., S. Li, J. Ou, and H. Li. 2012. “Reliability Assessment of Cable-stayed
Bridges Based on Structural Health Monitoring Techniques.” Structure and
Infrastructure Engineering 8, no. 9, pp. 829–45.
[15] Bruneau, M. 1992. “Evaluation of System-reliability Methods for
Cable-stayed Bridge Design.” Journal of Structural Engineering 118, no. 4,
1106–20. 10.1061/(ASCE)0733-9445
[16] Estes, A.C., and D.M. Frangopol. 2001. “Bridge Lifetime System Reliability
Under Multiple Limit States.” Journal of Bridge Engineering 6, no. 6,
523–28. 10.1061/(ASCE)1084-0702
[17] Liu, Y., N. Lu, N. Yin, and M. Noori. 2016a. “An Adaptive Support Vector
Regression Method for Structural System Reliability Assessment and its
Application to a Cable-stayed Bridge.” P. I. Mech O: J. Ris., 230, no. 2,
pp. 204–19.
[18] Thoft-Christensen, P., and Y. Murotsu. 1986. Application of Structural
Systems Reliability Theory. Heidelberg, Germany: Speinger-Verlag Berlin.
[19] Lee, Y.J., and J. Song. 2011. “Risk Analysis of Fatigue-induced Sequential
Failures by Branch-and-bound Method Employing System Reliability
Bounds.” Journal of Engineering Mechanics 137, no. 12, 807–21. 10.1061/
(ASCE)EM.1943-7889.0000286
[20] Kim, D.S., S.Y. Ok, J. Song, and H.M. Koh. 2013. “System Reliability
Analysis Using Dominant Failure Modes Identified by Selective Searching
Technique.” Reliability Engineering & System Safety 119, pp. 316–31.
[21] Nakamura, S., and K. Suzumura. 2012. “Experimental Study on Fatigue
Strength of Corroded Bridge Wires.” Journal of Bridge Engineering 18,
no. 3, 200–09. 10.1061/(ASCE)BE.1943-5592.0000366
140   •   BRIDGE RELIABILITY AND SERVICEABILITY

[22] Faber, M.H., S. Engelund, and R. Rackwitz. 2003. “Aspects of Parallel Wire
Cable Reliability.” Structural Safety 25, no. 2, pp. 201–25.
[23] Yang, W., P. Yang, X. Li, and W. Feng. 2012. “Influence of Tensile Stress
on Corrosion Behaviour of High-strength Galvanized Steel Bridge Wires in
Simulated Acid Rain.” Materials and Corrosion 63, no. 5, pp. 401–07.
[24] Mahmoud, K.M. 2007. “Fracture Strength for a High Strength Steel Bridge
Cable Wire with a Surface Crack.” Theoretical and Applied Fracture
Mechanics 48, no. 2, pp. 152–60.
[25] Maljaars, J., and T. Vrouwenvelder. 2014. “Fatigue Failure Analysis of Stay
Cables with Initial Defects: Ewijk Bridge Case Study.” Structural Safety 51,
pp. 47–56.
[26] Freire, A.M.S., J.H.O. Negrao, and A.V. Lopes. 2006. “Geometrical
Nonlinearities on the Static Analysis of Highly Flexible Steel Cable-stayed
Bridges.” Computers & Structures 84, no. 31, pp. 2128–40.
[27] Yoo, H., H.S. Na, E.S. Choi, and D.H. Choi. 2010. “Stability Evaluation of Steel
Girder Members in Long-span Cable-stayed Bridges by Member-based Stability
Concept.” International Journal of Steel Structures 10, no. 4, pp. 395–410.
[28] Yoo, H., H.S. Na, and D.H. Choi. 2012. “Approximate Method for Estimation
of Collapse Loads of Steel Cable-stayed Bridges.” Journal of Constructional
Steel Research 72, pp. 143–54.
[29] Liu, Y., N. Lu, M. Noori, and X. Yin. 2014. “System Reliability-based
Optimization for Truss Structures Using Genetic Algorithm and Neural
Network.” Int. J. Relia. Safe. 8, no. 1, pp. 51–69.
[30] Liu, Y., N. Lu, and X. Yin. 2016a. “A Hybrid Method for Structural System
Reliability-based Design Optimization and its Application to Trusses.” Qual.
Reliab. Eng. Int. 32, no. 2, pp. 595–608.
[31] Dai, H., H. Zhang, and W. Wang. 2012. “A Support Vector Density-based
Importance Sampling for Reliability Assessment.” Reliability Engineering
& System Safety 106, pp. 86–93.
[32] CSRA [Computer software]. 2013. Complex Structural Reliability Analysis
V 1.0, Changsha University of Science and Technology, Changsha, China.
[33] Tang, Q.Y., and C.X. Zhang. 2013. “Data Processing System (DPS) Software
with Experimental Design, Statistical Analysis and Data Mining Developed
for Use in Entomological Research.” Insect Science 20, no. 2, pp. 254–60.
[34] Chang, C.C., and C.J. Lin. 2011. “LIBSVM: A Library for Support Vector
Machines.” ACM Transactions on Intelligent Systems and Technology (TIST)
2, no. 3, p. 27.
[35] Ministry of Communications and Transportation (MOCAT). 2010. Stay
Cables of Parallel Steel Wires for Large-span Cable-stayed Bridge JT/T 775-
2010. Beijing, China,
[36] Ministry of Housing and Urban-Rural Development (MOHURD). 1999.
Unified Standard for Reliability Design of Highway Engineering Structures
GB/T 50283-1999. Beijing, China.
 About the Authors

Naiwei Lu, Lecturer of Civil Engineering, Changsha University of Science

and Technology, Changsha 410114, China and formerly Postdoctoral
Research Fellow, Institute of Risk and Reliability, Leibniz University
Hannover, Hannover 31509, Germany.
Email: [email protected]

Mohammad Noori, Professor of Mechanical Engineering and ASME

Fellow, Department of Mechanical Engineering, California Polytechnic
State University, San Luis Obispo, CA and Distinguished Visiting National
Chaired Professor of 1000 Program, International Institute for Urban
Systems Engineering, Southeast University, Nanjing, Jiangsu (210096),
China.
Email: [email protected]
 Index

A proposed framework, 125–127

Adaptive support vector regression short-span, 128–132
(ASVR) approach, 120 structural system characteristics
ADR. See Assessment dynamic of, 124–125
ratio CDF. See Cumulative distribution
ADTT. See Average daily truck function
traffic (ADTT) Computational framework
AGR. See Annual growth rate deterministic finite element-
Annual growth rate (AGR), 95–96 based simulation, 22–23
Assessment dynamic ratio (ADR), extrapolating maximum load
66, 90 effects, 98–100
Average daily truck traffic first-passage probability, 53–54
(ADTT), 82–84 probabilistic modeling, 23–24
proposed framework, 21–22
B traffic load effects, probabilistic
Bridge deflection serviceability, modeling, 72–74
81–82 Cumulative distribution function
(CDF), 52, 95
C
Cable degradation D
description of, 119 DAFs. See Dynamic amplification
formulations factors
cable strength, 122–123 Deflection extrapolation,
strength model, parallel wire 102–104
cable, 121–122 Deterministic finite element-based
Cable-stayed bridges simulation, 22–23
advantages of, 119 Dynamic amplification factors
definition of, 119 (DAFs), 91
finite element model, 55 Dynamic reliability evaluation
long-span, 132–137 bridge deflection serviceability,
mechanical behavior 81–82
investigation, 120 parametric studies, 82–84
144   •   Index

E results and interpretations,

Engineering structures, first- 57–60
passage reliability theory, 6–7 description of, 45–47
Equivalent dynamic wheel load importance of, 52–53
(EDWL) approach, 5, 92–93 proposed computational
Extrapolated maximum framework, 53–54
displacements, 57–58 Rice’s level crossing theory,
Extrapolating maximum load 50–52
effects, 98–100 stochastic traffic load simulation
Monte Carlo simulation,
F 48–50
Fatigue damage accumulation, weigh-in-motion
25–27 measurements, 47–48
Fatigue reliability First-passage reliability theory,
case study 6–7
hot spot stress simulation,
27–30 G
probabilistic analysis, 30–36 Gaussian mixture models
computational framework (GMMs), 19, 24
deterministic finite element- Gaussian random process, 94
based simulation, 22–23 Generalized extreme value (GEV)
probabilistic modeling, 23–24 theory, 90–91
proposed framework, 21–22 GMMs. See Gaussian mixture
results and discussion, 36–39 models
steel box-girder bridges, 7–8 Gross vehicle weight (GVW),
stochastic fatigue truck load 18–19
model GVW. See Gross vehicle weight
currently used, 16–17
in design specifications, H
16–17 Hot spot stress simulation, 27–30
proposed, 17–20
simulated, 19–20 I
Fatigue stress spectrum Individual GVW extrapolation,
finite element analysis, 14 101–102
structural health monitoring, 14 Interval traffic growth model
FEA. See Finite element analysis annual growth rate, 95–96
Finite element analysis (FEA), 14 lifetime maximum deflection
Finite element model, cable-stayed assessment, 111–114
bridge, 55 Rice’s extrapolation accounting,
First-passage probability 96–98
case study
bridge details, 54 K
probabilistic modeling, traffic Kolmogorov-Smirnov (K-S)
load effects, 54–57 statistics, 94
 Index   •   145

L computational framework,
LEFM. See Linear elastic fracture 72–74
mechanics Rice’s level-crossing formula,
Lifetime maximum deflection 70–72
assessment, 111–114 on two cable-supported bridges
Limit state function, 25–27, 36 probabilistic estimation,
Linear elastic fracture mechanics 77–81
(LEFM), 8 stochastic traffic load
Long-span bridges simulation, WIM
case study measurements, 74–76
extreme deflection, dynamic Probability density functions
traffic loads, 106–111 (PDFs), 18, 35
lifetime maximum deflection
assessment, 111–114 R
weigh-in-motion Rice formula, 93–95
measurements, 105–106 Rice’s extrapolation accounting,
deflection extrapolation, 96–98
102–104 Rice’s level-crossing formula,
dynamic performance of, 4–6 70–72
serviceability of, 89 Rice’s level crossing theory, 50–52
traffic loading behavior, 90–91 Road-roughness coefficient (RRC),
Long-span cable-stayed bridge, 93
132–137 RRC. See Road-roughness
coefficient
M
Monte Carlo simulation S
for long-span bridges, 4–5 SHM. See Structural health
stochastic traffic load simulation, monitoring
48–50 Short-span cable-stayed bridge,
128–132
N Simplified traffic-bridge
Nonstationary traffic load effects, interaction formulation, 69–70
90 Simulated stochastic fatigue truck
load model, 19–20
P S-N (stress-life) curve approach,
Parallel wire cable strength model, 25–26
121–122 Steel box-girder bridges, 7–8
Parametric studies, 82–84 Steel girders, 13
PDFs. See Probability density Stochastic fatigue truck load
functions model
Probabilistic analysis, 30–36 currently used, 16–17
Probabilistic modeling, 23–24 in design specifications, 16–17
first-passage probability, 54–57 proposed, 17–20
traffic load effects, 54–57 simulated, 19–20
146   •   Index

Stochastic traffic load simulation program implementation,

Monte Carlo simulation, 48–50 127–128
on two cable-supported bridges,
74–76 T
weigh-in-motion (WIM) Theoretical formulation
measurements, 47–48 Rice formula, 93–95
Stress-based approach, 7–8 traffic-bridge interaction, 92–93
Structural health monitoring Traffic-bridge interaction, 92–93
(SHM) Traffic-bridge interaction
fatigue stress spectrum, 14 formulation
weigh-in-motion (WIM) system, simplified, 69–70
90 vehicle-bridge interaction (VBI)
Suspension bridges system, 68–69
case study
hot spot stress simulation, V
27–30 Vehicle-bridge interaction (VBI)
probabilistic analysis, 30–36 system, 68–69
collapse and sudden collapse, Vehicle loads, 4
2–4 Verification examples
description of, 1–2 deflection extrapolation,
dynamic reliability assessment, idealized long-span bridge,
3 102–104
environmental conditions of, 2 individual GVW extrapolation,
System reliability evaluation 101–102
adaptive support vector
regression (ASVR) approach, W
120 Weigh-in-motion (WIM)
cable-stayed bridges measurements
proposed framework, long-span bridges, 105–106
125–127 stochastic fatigue truck load
structural system model, 15–16
characteristics, 124–125 stochastic traffic load simulation,
case studies 47–48
long-span cable-stayed Wind loads, 4
bridge, 132–137
short-span cable-stayed Z
bridge, 128–132 Zero-mean stochastic process, 53
 OTHER TITLES IN OUR SUSTAINABLE STRUCTURAL
SYSTEMS COLLECTION
Mohammad Noori, Editor

Numerical Structural Analysis

by Steven O’Hara and Carisa H. Ramming

A Systems Approach to Modeling Community Development Projects

by Bernard Amadei

Seismic Analysis and Design Using the Endurance Time Method,

Volume I: Concepts and Development
by H.E. Estekanchi and H.A. Vafai

Seismic Analysis and Design Using the Endurance Time Method,

Volume II: Advanced Topics and Application
by H.E. Estekanchi and H.A. Vafai

Momentum Press is one of the leading book publishers in the field of engineering,
mathematics, health, and applied sciences. Momentum Press offers over 30 collections,
including Aerospace, Biomedical, Civil, Environmental, Nanomaterials, Geotechnical,
and many others.

Momentum Press is actively seeking collection editors as well as authors. For more
information about becoming an MP author or collection editor, please visit
http://www.momentumpress.net/contact

Announcing Digital Content Crafted by Librarians

Momentum Press offers digital content as authoritative treatments of advanced ­engineering top-
ics by leaders in their field. Hosted on ebrary, MP provides practitioners, researchers, faculty,
and students in engineering, science, and industry with innovative electronic content in sensors
and controls engineering, advanced energy engineering, manufacturing, and materials science.

Momentum Press offers ­library-friendly terms:

• perpetual access for a one-time fee

• no subscriptions or access fees required
• unlimited concurrent usage permitted
• downloadable PDFs provided
• free MARC records included
• free trials

The Momentum Press digital library is very affordable, with no obligation to buy in future years.

For more information, please visit www.momentumpress.net/library or to set up a trial in the US,
please contact [email protected].
EBOOKS Multi-scale Reliability and

LU • NOORI
FOR THE SUSTAINABLE STRUCTURAL
ENGINEERING
Serviceability Assessment of SYSTEMS COLLECTION
LIBRARY In-service Long-span Bridges Mohammad Noori, Editor
Create your own
Naiwei Lu • Mohammad Noori
Customized Content
Bundle—the more With the development in global economic and transportation
books you buy, engineering, the traffic loads on brides have been g
­ ­ rowing

Multi-scale Reliability and Serviceability Assessment of In-service Long-span Bridges

the greater your steadily, which become potential safety hazards for e ­ xisting
discount! bridges. In particular, long-span suspension bridges s­upport

Multi-scale
heavy traffic volumes and simultaneous truck loads on the
THE CONTENT bridge deck, and thus the safety and serviceability of the bridge
• Manufacturing deserves investigation. In this book, a multiscale reliability
Engineering
• Mechanical
­method is presented for the safety assessment of long-span
bridges. The multiscale failure condition of stiffness girders Reliability and
Serviceability
& Chemical is the first-passage criteria for the large-scale model and the
Engineering ­fatigue damage criteria for the small-scale model.
• Materials Science
& Engineering
• Civil &
Environmental
It is the objective of this book to provide a more in-depth
understanding of the vehicle-bridge interaction from the
­
random vibration perspective. This book is suitable for
­
Assessment
Engineering
• Advanced Energy
­adoption as a text book or a reference book in an advanced
structural ­reliability analysis course. Furthermore, this book also
of In-service
Long-span
Technologies ­provides a theoretical foundation for better understanding of
the safety assessment, operation management, maintenance
THE TERMS and ­reinforcement for long-span bridges and motivates further
• Perpetual access for
a one time fee
• No subscriptions or
research and development for more advanced reliability and
serviceability assessment techniques for long-span bridges. Bridges
access fees Naiwei Lu is a lecturer of civil engineering at Changsha
• Unlimited ­University of Science and Technology.
concurrent usage
• Downloadable PDFs Mohammad Noori is professor of mechanical engineering and
• Free MARC records ASME fellow in the department of mechanical engineering,

For further information,

California Polytechnic State University and distinguished visit- Naiwei Lu
a free trial, or to order,
contact: 
ing national chaired professor of 1000 Program, International
Institute for Urban Systems Engineering at Southeast University
Mohammad Noori
[email protected] in Nanjing China.

ISBN: 978-1-94708-338-7