Vous êtes sur la page 1sur 206

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/36310557

Ultimate strength of highway girder bridges

Book · July 1986


Source: OAI

CITATIONS READS

7 243

1 author:

Hasan Tantawi
Fahad Bin Sultan University
8 PUBLICATIONS   16 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

properties of concrete mixtures implementing recycled wastes aggregate View project

All content following this page was uploaded by Hasan Tantawi on 26 April 2015.

The user has requested enhancement of the downloaded file.


INFORMATION TO USERS

This reproduction was made from a copy of a manuscript sent to us for publication
and m icrofilm ing. While the most advanced technology has been used to pho­
tograph and reproduce this manuscript, the quality of the reproduction is heavily
dependent upon the quality of the material submitted. Pages in any manuscript
may have indistinct print. In all cases the best available copy has been filmed.

The following explanation of techniques is provided to help clarify notations which


may appear on this reproduction.

1. Manuscripts may not always be complete. When it is not possible to obtain


missing pages, a note appears to indicate this.

2. When copyrighted materials are removed from the manuscript, a note ap­
pears to indicate this.

3. Oversize materials (maps, drawings, and charts) are photographed by sec­


tioning the original, beginning at the upper left hand comer and continu­
ing from left to right in equal sections w ith small overlaps. Each oversize
page is also film ed as one exposure and is available, for an additional
charge, as a standard 35mm slide or in black and white paper format. *

4. Most photographs reproduce acceptably on positive microfilm or micro­


fiche but lack clarity on xerographic copies made from the microfilm. For
an additional charge, all photographs are available in black and w hite
standard 35mm slide format.*

*For more information about black and white slides or enlarged paper reproductions,
please contact the Dissertations Customer Services Department

•T Dissertation
Information Service
University Microfilms International
A Bell & Howell Information Company
300 N. Zeeb Road, Ann Arbor, Michigan 48106
8621387

Tantaw i, Hasan Mohammad

ULTIMATE STRENGTH OF HIGHWAY GIRDER BRIDGES

The University of Michigan Ph.D. 1986

University
Microfilms
International 300 N. Zeeb Road, Ann Arbor, Ml 48106
ULTIMATE STRENGTH OF HIGHWAY GIRDER BRIDGES

by

Hasan Mohammad Tantawi

A dissertation submitted in partial fulfillment


of the requirements for the degree of
Doctor of Philosophy
(Civil Engineering)
in The University of Michigan
1986

Doctoral Committee:

Professor Andrzej S. Novak, Chairman


Professor Subhash C. Goel
Professor Niels C. Lind, University of Waterloo-Canada
Professor Stephen M. Pollock
Professor Wadi S. Rumman
RULES REGARDING THE USE OF

MICROFILMED DISSERTATIONS

Microfilmed or bound copies of doctoral dissertations submitted


to The University of Michigan and made available through University Micro­
films International or The University of Michigan are open for inspection,
but they are to be used only with due regard for the rights of the author.
Extensive copying of the dissertation or publication of material in excess of
standard copyright limits, whether or not the dissertation has been copy­
righted, must have been approved by the author as well as by the Dean of
the Graduate SchooL Proper credit must be given to the author if any
material from the dissertation is used in subsequent written or published
work.
DEDICATED TO MY MOTHER AND THE BLESSED MEMORY

OF MY FATHER , TO WHOM I OWE EVERY THING.


ACKNOWLEDGEMENTS

I express cordially my indebtedness, gratitude and

appreciation to Professor Andrzej S. Nowak for his

inspiration to do this work, his generous help, unlimited

kindness and valuable guidance during this work.

I am grateful to Professors Subhash C. Goel, Niels

C. Lind, Stephen M. Pollock and Wadi S. Rumman, members of

the dissertation committee, for reviewing the dissertation

and offering helpful suggestions and comments. The valuable

help and instructions of professor Niels C. Lind (University

of Waterloo- Canada) are deeply appreciated.

I also wish to record my sincere gratitude and deep

thanks to the Jordan Army and Mutah University for giving me

the scholarship which fulfilled my dream of further studies.

My sincere thanks and gratitude are to my wife and my

children for their patience and support during my study.

Finally I would like to thank all my colleagues

specially, Vahid Sattary, Medhat Boutros and Adel El-Tayem

who have helped me in some way or another during the

research.
TABLE OF CONTENTS
DEDICATION

ACKNOWLEDGEMENTS iii

LIST OF TABLES vi

LIST OF FIGURES vi i

LIST OF APPENDICES x

NOTATIONS xi

CHAPTER

I. INTRODUCTION 1

1.1 Historical Development


1.2 Thesis Outline

II. FUNDAMENTALS OF STRUCTURAL RELIABILITY 12

2.1 Limit State Function and Failure Probability


2.2 Safety Analysis Concepts
2.3 Simulation and Approximation Techniques
Used in Calculating the Statistical Moments
of Independent Random Variables
2.4 Simulation and Approximation Techniques
Used in Calculating the Statistical Moments
of dependent Random Variables

III. BRIDGE LOAD M O D E L S ................ 43

3.1 Introduction
3.2 Dead Load
3.3 Live Load
3.4 Dynamic Load
3.5 Load Combinations
3.6 Transverse Location of the Truck Traffic

IV. BRIDGE DECK M O D E L ....................................72

4.1 The Grid Model


4.2 Geometric Properties ofthe Grid Elements
4.3 Strength of the Grid Elements
4.4 Strength Reliability of Single Elements

V. ULTIMATE RESISTANCE MODEL OF BRIDGE SYSTEM . . 104

5.1 General Description


5.2 Ultimate Strength Analysis
5.3 Ultimate Strength Reliability

VI. APPLICATION TO BRIDGE SYSTEM EVALUATION. ... 142

6.1 Evaluation using the Developed Approach


6.2 Bridge Capacity Rating

VII. CONCLUSIONS AND R E C O M M E N D A T I O N S ............. . 157

7.1 Research Results


7.2 Summary of the Conclusions
7.3 Recommendations

A P P E N D I C E S .................................................. 163

B I B L I O G R A P H Y ................................................ 178

V
LIST OF TABLES

Table

2.1 Results of Example 1 .................. 35

2.2 Results of Example 2 . . . . .......................... 37

3.1 Number of Trucks as Function of Time . ........... 49

3.2 Parameters of Live l o a d ................................ 52

3.3 Input Data for Eg. 3 . 3 ............................. 55

3.4 50 Year Live Load Models . ......................... 64

3.5 Dynamic Load F a c t o r .......................... 65

3.6 Typical Percentage Values of Major Load


C o m p o n e n t s ........................... 68

4.1 Statistical Characteristics of Input Data


for Section Analysis .*............................... 100

6.1 Strength of Grid Components............... 145

6.2 Results of Simulation for a Reinforced


Concrete B r i d g e ..................................... 147

6.3 Reliability Analysis without Correlation ........ 153

6.4 Reliability Analysis with Correlation ........... 153


LIST OF FIGURES

Figure

1.1 Flowchart of the Research P l a n .................. 10

2.1 Schematic Illustration of


the Fundamental C a s e ............................ . 14

2.2 Hasofer and Lind Reliability I n d e x ................ 21

2.3 Continous Beam with Three S p a n s .................... 36

3.1 Distribution Functions of Moments for


Various Spans ...................................... 48

3.2 Configuration of the Heaviest Commercial


V e h i c l e s ............................................. 50

3.3 Upper tails of moment distributions from


Truck Survey........................................ 51

3.4 Extrapolated live load d i s t r i b u t i o n s ...............53

3.5 Typical Distribution of Load ratios


(60 Ft. s p a n . ) ........................................ 58

3.6 Distribution of the arbitrary-point-in-time


and 50 year load ratio, LR^, for 60 ft. span . . 61

3.7 Distribution of 50 year live load for


80 ft. s p a n ........................................... 62

3.8 Distribution of 50 year live load for


125 ft. s p a n ......................................... 63

3.9 Typical Curb Distance Histogram


for Interstate Highway 1-94.......................... 71

3.10 Distribution of Curb Distance on Normal


Probability P a p e r ......................................71

4.1 Typical Bridge System and The Corresponding


Grid M o d e l ..............................................73

4.2 Plate Element and the Equivalent Framework Model . 75

4.3 Stress-Strainss Relationships ..................... 80


4.4 Typical Composite Steel Section ................... 81

4.5 Typical Sections Commonly Used in Prestressed


Concrete Bridges....................................... 82

4.6 Typical Strain Diagram in a Composite Section. . . 83

4.7 Typical Moment-Curvature Relationship for


Composite Steel Concrete Section .................. 84

4.8 Actual S e c t i o n ............................ 87

4.9 Idealized Section .................................. 87

4.10 Torsion Versus Angle of Twistrelationship.......... 89

4.11 Typical Thin Section under different state


of torsional stresses................................. 91

4.12 Idealized Thin Section under different state


of torsional stresses................................. 91

4.13 Composite Concrete Steel Section .................. 95

4.14 Typical Torsion Vs. Angle of Twist Relationship


for Composite Steel S e c t i o n .................... . 98

4.15 Typical Moment-Curvature Relationships ........ 101

4.16 Distribution of the Maximum Load per Girder . . . 103

5.1 Coordinate System ................................. 108

5.2 Nonprismatic Beam E l e m e n t .......................... 110

5.3 Reduced Nonprismatic Element


Stiffness Matrix ................................... 114

5.4 Wheel Load D i s t r i b u t i o n ............................ 115

5.5 Clamped Beam Load Versus Deflection Curve . . . . 120

5.6 Plate E l e m e n t . 122

5.7 Plate Grid M o d e l ..................................... 124

5.8 Effect of Mesh Size on the Failure L o a d ........... 125

5.9 Effect of Mesh Size on the Accuracy of the


Results Compared with the Yield Line Theory . . . 125

5.10 Details of the Bridge M o d e l ....................... 126

5.11 Load Versus Deflection (Three Models) ........... 127


5.12 Elasto-Plastic Response for Composite
Steel Girder B r i d g e ................................. 129

5.13 Example of a Series S y s t e m ........................ 131

5.14 Example of a Parallel S y s t e m .......... ... 132

5.15 Reliability Model for Series System in Parallel . 132

5.16 Reliability Model for Parallel System inSeries . 133

5.17 Typical Cross Section of a Composite


Steel Girder B r i d g e ................................. 135

6.1 Reinforced Concrete B r i d g e .......................... 143

6.2 Bridge Grid M o d e l ....................... .......... 144

6.3 Transverse Location of the T r u c k ................. 146

6.4 Composite Steel Girder Bridge .................... 150

6.5 Prestressed Concrete Bridge ....................... 152

A.1 Graphical Rackwirz and Fiessler Algorithim . . . .166


LIST OF APPENDICES

Appendix

A. Second Moment Reliability Index ................. 164

A.l Cornel Reliability Index


A . 2 Determination of the Design Point Using
Rackwitz and Fessler Algorithim

B. Generation of Correlated Random Variables. . . . 170

B.l Transformation to Uncorrelated Space


B.2 Generation of Correlated Lognormal Variables
B.3 Generation of Jointly Distributed Normal
and Lognormal Variables.

C. Reliability of Structural S y s t e m .................. 173

C.l Series System


C.2 Parallel Systems
C.3 Complex Systems

x
LIST OF NOTATIONS

BM : maximum base length of a truck

CDF : cumulative distribution function

COV : coefficient of variation

COV(X.,X..) : covariance of X. and X.


i J

X : covariance matrix

E : Young's modulus of elasticity

E[ ] : expected value of a random variable

: probability distribution function and


FX ,fX
density function of variable X

fxryp : joint probability density function of


random variables X and Y

G : shear modulus

i : moment of inertia of a section

: Truck Moment / HS20 Moment


lra

LR : Truck Moment / Base Length Moment


B

: Truck Moment /OHBDC Moment


LR0

M : bending moment

: plastic bending moment


MP

MTC : Ministry of Transportation and


Communications, Ontario, Canada

OHBDC : Ontario Highway Bridge Design Code

PDF : probability density function

Q : mean value of the combined load effect


: live load

: impact load

Qg :dead load

R :ultimate resistance of the bridge

R :mean value of the resistance

VR :coefficient of variation of resistance

Vq : coefficient of variation of load effect

j3 :reliability index

6r :permanent deflection of the bridge at


ultimate load

S>,0 : cumulative distribution function and density


function for the standard normal variable

u :Poison's ratio

pv v :linear correlation coefficient between


1f 2
variables and ^

T
[ ] :transpose of a matrix

[ I-"*- :inverse of a matrix

I I :absolute value

xii
CHAPTER I

INTRODUCTION

1.1 Historical Development

1.1.1 Reliability Theory

Traditionally, structural design relies on deterministic

analysis. Suitable dimensions, material properties, and

loads are assumed, and analysis is then performed to provide

a more or less detailed description of the structure. The

uncertainty in load and resistance was accounted for by

introducing the so called load and resistance

factors. However, fluctuations of loads, variability of

material properties, and uncertainties regarding analytical

models all contribute to a generally small probability that

the structure does not perform as intended, however, this

small probability has significant impact on other aspects

such as life, budget, etc. In response to this problem,

methods have been developed to deal with the statistical

nature of loads and material properties and, more recently,

a general framework for comparing and combining these

statistical effects has emerged [66].

The structural reliability is defined as the probability

that the structure will perform satisfactorily for a

1
2

specified period of time under a stated set of load

conditions. It has been used to develop economic and

efficient design methods incorporating probability theory

together with methods of advanced structural

analysis. Structural reliability theory has attracted

academic interest since 1960. Lind et al. [64] defined the

problem of rational design of a code as that of finding a

set of best values of the load and resistance

factors. Cornell [18] suggested the use of a second moment

format to define the reliability index. Lind [61] showed

that the Cornell's reliability index requirement could be

used to derive a set of safety factors for load and

resistance. The year of 1974 saw the publication of the

first standards in limit states format based on

probabilistic rationale [14].

However, until recently most of the work was

concentrated on finding the reliability index based on the

behavior of individual elements. Moses [80] presented some

simplified analysis and approximations for evaluating system

reliability. The system reliability is indispensable

information for adjusting the element's safety factors to

account for its presence in the structural assemblage so

that the overall probability of failure remains roughly

equal to what is desired of the element.

Gorman [33] utilized rigid plastic hinge theory to

evaluate the resistance of a redundant building structure


3

subjected to a proportional loading. The mean and the

standard deviation of the system resistance were evaluated

in terms of the statistical properties of their

components. Several simulation techniques were used or

developed. Some of these techniques will be extended for

application to a redundant bridge system.

Kam [52] presented an approximate method for numerically

integrating the probability of failure for a simple framed

structure. He considered three kinds of limit states in the

reliability analysis of frames. The first of these limit

states is first yielding in the structural system; the

second limit state is the formation of the first plastic

hinge in the structure; and the third limit state is the

formation of a local mechanism or instability of the

structure. The incremental load procedure was utilized to

construct the limit state surface. The resistance of the

structure was treated as a deterministic quantity while the

loads were treated as random variables. The method was

demonstrated by analyzing a two-dimensional frame to

construct the three limit states of a structure in the load

space.

Lin [59] developed an approach to determine the

reliability of redundant ductile framed structures. His

approach was based on determining the limit state function

in load space and evaluating the probability of failure

directly for the overall structural system rather than the


4

individual failure modes. Lin determined first the limit

state function in load space for the structure with mean

resistances by performing nonlinear structural analysis

along various load paths. For each load path, a random

variable representing the distance from the origin to the

limit state was introduced to account for the randomness in

the strength of the structure. The probability of failure

conditioned on a particular limit state is calculated by

integrating the joint density function of the load over the

region beyond that limit state. The total probability of

failure was then calculated as the sum of all conditioned

failure probabilities weighted by their respective

probability of occurrence.

1.1.2 Bridge Deck Analysis

The ultimate strength of the bridge system is the sum of

the strength at the elastic stage and the strength reserve

beyond the elastic limit up to failure. The concept of

bridge ultimate strength involves nonlinear behavior under

different loading conditions. The method of modeling and

analysis of the bridge structure should combine accuracy and

simplicity along practical and realistic lines.

Many attempts have been made to simulate bridge system

behavior including: finite element methods Wegmuller [112],

finite strip method Nowak and Boutros [86], finite deference

method Heins and Kuo [44], orthotropic plate theory Heins

and Looney [46], simple yield line theory Reddy and Hendry
5

[100], and the grid analogy [9,51, 8]. The grid analogy was

chosen as structural analysis tool in this

research. Although it is necessarily approximate, it has the

advantage of almost complete generality of applications.

The method of grid analysis involves the idealization of

the bridge deck through its representation as a plane

grillage of discrete interconnected beams. The grid method

allows for any support conditions. The bridge analysis can

be carried out with comparative ease.

The structural gridwork has evoked intense interest

since the earliest stages of development. Several methods

were developed approaching the solution of the grid problem

from various viewpoints. Bares and Massonnet [9] mentioned

that until circa 1955 most of the solutions to the problem

of the bridge grids, with few or man y main girders, were

based on the assumption that the torsional resistance of the

element can be neglected. The torsional rigidity may,

without noticeable error, be ignored in bridge structures

with noncomposite steel girders and reinforced concrete.

The torsional rigidity was considered by several researchers

as mentioned by Bares and Massonnet [9] but this approach

was so complicated and thus making them of limited use in

practice. The general availability of the digital computer

has revived the grid method. A number of computer programs

have been developed using the grid analogy for design

purposes based on elastic analysis [21].


6

Recently, Jaeger and Bakht [51], provided a basic

framework for using the grid model to simulate the behavior

of the bridge system in the elastic stage. Some features of

their work will be used in this study.

The traditional assumption of linear elastic behavior

can produce inaccurate or misleading results that do not

reflect the true structural response. Nonlinear analysis is

more costly to perform than a linear elastic analysis.

However, even a relatively simple nonlinear analysis yields

a more reliable structural response than linear analysis

[93].

Most of the work mentioned earlier deals with the bridge

analysis using the grid model in the elastic condition.

In reliability analysis a programmable analytical model

to analyze the bridge system under the effect of traffic

loading is required. The grid model was found to be suitable

for the purpose. Although the model involves some

approximations, it provides an economic ( in terms of

computing time) and acceptable level of accuracy.

In this study the grid model was extended to include the

nonlinear response of the bridge system in a practical way.

Very early experimental and theoretical work in the area

of nonlinear analysis of a grid system was done by Heyman

(1953). Heyman based his analysis on the assumptions of

plastic hinges and the effect of torsional rigidity. He


7

proved that the interaction between bending moment and

torsion has no importance in predicting the actual capacity

of the grid system. Later, Hollinger and Mangelsdorf [49]

found that an assumption concerning the deterioration of the

torsional rigidity with respect to different levels of

loading could be used to update the torsional rigidity of a

structural element.

The method assuming the formation of plastic hinges has

received a great deal of interest in recent years [84, 48,

67].

Gorman [33] developed a method for identifying

significant failure modes based on rigid plastic hinges.

They applied their method successfully to different cases of

buildings subjected to different loading conditions. They

used approximate methods to calculate the first two

statistical moments of the structural resistance. Some of

these approximation methods will be generalized in this

thesis for application in the reliability analysis of a

bridge system.

1.2 Thesis Outline

1.2.1 Definition of the Problem

Highway bridges are currently designed according to

codes based on the elastic behavior of the material. There

is a considerable amount of strength reserve beyond the

elastic stage and before the ultimate capacity of the system


i
.

is reached. For example, a recent field study done by the

Ministry of Transportation and Communications of Ontario,

Canada, showed that some of the bridges, posted with a

maximum capacity of 8 tons, were loaded up to 300 tons

before reaching the load carrying capacity. Accurate

evaluation of bridge systems is important for decisions

concerning replacement, rehabilitation or posting.

The problem investigated in this study is the

determination of an overall reliability measure for the

structural system. Safety is measured in terms of a

reliability index, 0. The structures considered are bridges

modeled as grillages. Three types of bridges are considered

composite steel girders, prestressed concrete girders and

reinforced concrete beams. The model includes material

nonlinearity, ductility of components and deterioration of

torsional rigidity with the increase of flexural stress in

section.

The basic random variables considered in this study are

load magnitude, load position and component strength. The

component strengths considered in this study are the

ultimate flexural, torsional and shear capacities. These

random variables are characterized by their mean and

covariance values.

1.2.2 Objectives of the Research

The three main objectives of this study are:


9

1. To develop an approach for evaluation of the bridge

load carrying capacity,

2. To develop a practical reliability analysis method

for short and medium span bridges, and

3. To establish practical rating criteria based on the

available bridge loading data, the actual conditions

of the bridge and a predefined target reliability

level.

The calculation of structural system reliability

involves combining the effect of load and resistance. The •

analysis includes a complex interaction between the system

reliability and element reliability. A large number of

variables are required to describe the system. The

reliability of a structural system is investigated using

Monte-Carlo simulation and a discrete distribution methods.

The research plan is presented in Fig. 1.1.

1.2.3 Organization of the Thesis

The work described in this thesis addresses three

separate topics. The first topic is the bridge loading based

on the available models in literature and an a survey

conducted by a research team of the University of Michigan

(see Sec. 3.6). It is described in Chapter III. The second

topic (Chapters IV and first part of Chapter V) is the

evaluation of the bridge ultimate load carrying capacity.

The third topic (the remainder of Chapter V) is the


10

t
RESEARCH PLAN

DETERMINISTIC MODELS

BRIDGE D E C K SECTION ANALYSIS BRIDGE SYSTEM


MODEL MODEL MODEL

RELIABILITY MODELS

LIVE LOAD CO M P O N E N T STRENGTH BRIDGE RESISTANCE


MODEL MODEL MODEL

BRIDGE CAPACITY RATING M O D E L

APPLICATION TO PRACTICAL CASES

ULTIMATE STRENGTH EVALUATION BRIDGE CAPACITY RATING

CONCLUSIONS AN D RECOMMENDATIONS

Figure 1.1 - Flowchart of the Research Plan


11

extension of Rosenblueth point estimation method to model

the ultimate resistance of a redundant bridge system with

various levels of correlation between elements.

The work is organized into seven chapters and an

appendix. In Chapter II the necessary knowledge about

reliability theory and Rosenblueth1s point estimates is

presented. Results of a special survey of the transverse

position of trucks on a bridge is given in Chapter

III. Chapter IV includes modelling the bridge deck into a

grid system, properties of the grid elements and evaluation

of the bridge elements' strength. The reliability analysis

of the bridge system is described in chapter V.

The applications of the developed model to actual

structures is presented in Chapter VI. A discussion of the

results and conclusions can be found in Chapter VII.

Additional Derivations and methods are given in the

appendices.
CHAPTER II

FUNDAMENTALS OF STRUCTURAL RELIABILITY

A highway bridge structure may be subjected during its

life-time to loads (or actions) which may cause a change of

condition or state of the structure. This change may go from

the undamaged state to a state of deterioration, damage in

varying degrees, to failure or collapse. Ultimate strength

analysis involves determining the response of a structure

subjected to an external live load. Predicting the ultimate

strength of a structural system is usually associated with a

considerable degree of uncertainty. This is mostly due to

variability of material properties and simplifications used

in the method of analysis. Predicting the future load

condition that might act on a bridge system is also a

process that involves a great deal of uncertainty. Hence the

risk of unacceptable combinations of load and resistance can

only be evaluated by reliability analysis.

2.1. Limit State Function and Failure Probability

Reliability analysis requires the determination of the

relevant load and resistance parameters, and the functional

relationship among them. If, for a given structure, all the

actions (loads) can be lumped into one single random

12
13

variable, Q, and the strength of the structure into another

single random variable, R, then the state of the structure

can be determined by a variable Z,

Z = R - Q (2.1)

The failure function can be formulated as

g(R,Q) = R - Q (2.2)

The limit state is defined as the boundary between

safety and failure as determined by:

g(R,Q) = 0 (2.3)

The event of failure corresponds to:

g(R,Q) < 0 (2.4)

so that the probability of failure is given by:

OB Q
P f = P(Q>R) = / / fn p (q,r)dqdr = Jdqj fn R (q,r)dr
1 Q>R Q 'R 0 0 Q 'R (2.5)

where f^ R is the joint probability density function of the

random variables Q and R as shown in Fig. 2.1.

If Q and R are statistically independent, then

fQ , R (1-r> * (2.6)

and Eq. (2.5) becomes


14

q=r q

Volume = IJ

Figure 2.1 - Schematic Illustration of


the Fundamental Case [5]

dq = / fQ (q)FR (q)dq
! fR (r)ar
(2.7)

where fg and fR are the density functions of Q and R

respectively, and FR is the distribution function of

R. Reliability P is the complement of the probability of


d
failure,

ps=1-p f (2.8)

In general, R and Q are functions of other random

variables. In this case, the limit state function, Eq. 2.3,

can be expressed in terms of the set of n basic variables as

shown in Eq. 2.9.


Therefore, the probability of failure, Pj, is given by the

multiple integral

• • • f dx.
X I' 2' xn )dxld x 2' n (2 .10 )

where f ^ x ^ X j , ..., xn > is the joint probability density

function of the random vector X, and the integration is

performed over the region where g(X)<0 as illustrated in

Fig. 2.1.

If, and only if, the basic variables x M X - ^ X j , ..., Xn > are

statistically independent, Eq. 2.10 may be replaced by

• • • 9 fx u n)dx
h

2.2. Safety Analysis Concepts

2.2.1 General Concepts

In practice, Eq. 2.11, presents two problems. First,

usually there is insufficient data to define the joint

probability density function for the n basic variables.

Usually, there is hardly enough information to be confident

about the marginal distributions and the covariance matrix.

Secondly, even if the joint density function is known, or in

the case of Eq. 2.11 the marginal densities, the multi-


16

dimensional integration required may be extremely time-

consuming. Analytical solutions do not exist for the

majority of practical problems and the analyst must resort

to the numerical methods by using level 2 or special level 3

methods (explained next) depending on the calculations

performed and the approximations made [107], A short review

of level 2 and 3 numerical methods is presented further in

this chapter.

2.2.2 Level 2 Methods

Level 2 methods provide powerful tools to solve a wide

range of practical problems. At this level only the means,

and the covariances, C^j, of the random variables are

used.

It is often convenient to use the reliability index, j3,

instead of probability of failure, P f , to characterize the

reliability. The generalized reliability index was defined

by Ditlevsen [22] in terms of the probability of failure as:

0 = - $ 1 (Pf ) (2.12)

where $ is the standard normal CDF.

In case of Eq. 2.2, and if both R and Q are independent

and normal random variables, the reliability index, 0,

becomes:
17

(R - Q)
0-

^ R + °Q (2.13)

where

R, aR = the mean and the standard deviation of

the resistance, and

Q, Oq = the mean and the standard deviation of

the load.

If both R and Q are lognormal random variables, a

reliability index can be derived as follows. First,

g = in S = lnR . lnQ (214)

g is a normally distributed random variable with mean

g = InR - InQ (2.15)

In general if X is a random variable, then (see Benjamin and

Cornell [10])

ln(x) = ln(X) - |ln(l + V^) (2.16)

and

2 2
aln(X) = ln(1 + VX J (2.17)

where

Vx = coefficient of variation of X.

Hence
18

g = ln(R) - |ln(l + V*) - ln(Q) + |ln(l + V*)

(1 + v j )
g = In- - Tpln-
(1 ♦ v ‘ >
(2.19)

» P o * 1
g = In (-)
Q
VR + 1
(2 .20 )

But

°g = v/fflnR + alnQ (2 .21 )

hence

R /VQ + 1
In (-)
Q
VR + 1
0 = _ =

A In [(VR + 1 ) (VQ + 1)] (2 .2 2 )

In a more practical way, the following simplifying

assumptions can be made with a reasonable accuracy (Benjamin

and Cornell [10]);


19

ln(x) s* ln(x) (2.23)

(2.24)

Then Eg. 2.22 gives the useful approximation

R
In-
Q

•"S * VQ (2.25)

If R and Q are neither normal nor lognormal, then 0 can

be calculated using a special numerical procedures (Madsen

Krenk and Lind 1986).

The limit state function g(X), equation (2.9), is not

always linear. In case of a nonlinear limit state function,

a Taylor expansion with linear terms can be used. Then, the

first order approximations of the mean and the variance of

the limit state function can be expressed as follows [66]:

G ~ g(X*,X*, • • • f
(2.26)

(2.27)

and the reliability index is


20

(2.28)

where the partial derivatives, gjjp are evaluated at the


•ff ^
linearization point X^, Xj is the mean of the basic variable

X. and COV(X.,X.) is the covariance of X. and X.


1 1 3 1 3

The reliability index 0 evaluated using Eqs. 2.26, 2.27

and 2.28 is approximate and it is called a first-order

second-moment reliability index. If the linearization point

is the mean-value point, then the reliability index is

called a mean-value first-order second-moment reliability

index [66]. A better estimate to the reliability index was

provided by Hasofer and Lind [41]. First, the the set of

basic variables Xj is transformed to a set of normalized

variables with zero mean and unit variance. The new set Z =

(Z^, Z 2 , ..., Zn ) is obtained by applying the following

transformation :

..., n
(2.29)

Where X^ and </x j are the mean and standard deviation of the

basic variable, X^.


21

The Hasofer-Lind reliability index 0 is defined as the

shortest distance from the origin to the failure surface in

the normalized Z-Coordinate system [107], A case where the

failure surface is defined by two variables only is shown in

Fig. 2.2.

Failure Region
Design Point Failure Surface

Safe Region

Figure 2.2 - Hasofer-Lind Reliability Index

In many engineering problems, the assumption of normally

distributed random variables is unreasonable. In such cases,

the reliability index can be calculated using the Rackwitz

and Fiessler procedure [96], Rackwitz and Fiessler developed

an iterative procedure based on normal approximations to

non-normal distributions at the so called design point. The

method is explained in Appendix A.


22

2.3 Simulation and Approximation Techniques Used in


Calculating the Statistical Moments of Independent Random
Variables

2.3.1. Simulation

Monte Carlo analysis is a powerful tool which enables

one to perform a statistical analysis of the uncertainty in

structural engineering problems. It is particularly useful

for complex problems where numerous random variables are

related through nonlinear equations [40].

The fundamental step in Monte Carlo analysis is to use

appropriate random number generators to generate independent

values for each of the basic variables (X^, X 2 , ..., X n )

with known distributions. Let M = f(X^, X 2 , •••# xn )* Then,

the corresponding distribution function of M is determined

by generation of (Xj, x 2 , ..., x n ). In each step the

realization of M, m, is calculated from the following

equation

m = f(xlfx 2 , ..., xn ) (2.30)

By repeating this process many times it is possible to

simulate the probability distribution for M. Although the

sample is generated numerically, it may be treated in the

same way as any other statistical sample.


23

If the limit state function g represents the boundary

between safe and unsafe regions and if M = gtx^, X 2 , .

Xn ), then the failure probability may be expressed as

P- = P( M<0 ) = lim £
£ n->® n (2.31)

where n is the total number of trials, and k is the number

of trials in which g(xlfx 2 , xn ) < 0.

If
However, the ratio — is a random variable whose sampling

distribution, and in particular sample variance, depends on

the number of trials n. For low failure probabilities and/or


If
small n, the estimate of P^ given by — is subjected to a

considerable uncertainty. If k = 0, then a graphical

approach can be used. In the graphical approach, the

approximate cumulative distribution function (CDF) of M can

be plotted on a probability paper. The probability of

failure can be read by extrapolation of the CDF.

In some cases it is not practical to use the Monte Carlo

technique, especially in the case of complicated structural

problems with a large number of random variables. Other

methods are then available which may provide reasonable

results. Some of these techniques are based on reducing the

number of simulated random variables, for example the

Variance-Reduction Techniques, VRT, [6]. Others, like

Rosenblueth point estimates [101, 102] are based on using


24

point estimates for evaluating functions of random

variables. Rosenblueth point estimates were used in

different areas of civil engineering for example Gorman [33]

used it in structural engineering and McGuffey [69] used it

in geotechnical engineering. The results were in good

agreement with some of the existing methods (Monte Carlo

Simulation, Taylor Expansion, Clark's method).

Rosenblueth point estimates are used in this

investigation. The bases of these methods is explained next.

2.3.2. Rosenblueth 2n Point Estimates

Rosenblueth [102] has generalized a 2n point estimate

for a function of n variables which may be correlated. The

joint probability density is assumed to be concentrated at

points in 2n hyperquadrants of the space defined by the n

random variables. The points are defined by all

combinations of each variable at Xj ± o^. Consider a vector

of variables Y = f(X,,X~, ..., X„) with a certain


1 2 n
correlation between the variables. Define a vector

(dimension n) Sk = (Sl k , S2 k , ..., Sn k ) such that S ik = 0 at

the means and ±1 other than that. Then the points are

defined by all 2n possible combinations, k=l, ..., 2n , of Y k

as:

X k = (X1 + Sl k al' X 2 + S2ka2' *•*' X n + Sn k an ) (2.32)


The weighting functions (or probability concentrations),

P k , at each of these points are defined by:

1 n-1 n
l?k = ~r(l + Z 2 S •« Si -iPi-i)
k 2 i=lj=i+l lk 1 3 13 (2.33)

Where p ^ is the correlation coefficient between and

X j . Then the moments of the function can be calculated by

the following equation:

2n
EtY"1] = I Pk [ y k ]m
k=l K K (2.34)

Rosenblueth distribution is useful and it can save a lot

of time in the case of a complicated structure with several

random variables. The correlation between variables is

handled in this method by giving a different weight to each

point of the 2n points.

2.3.3. Rosenblueth 2n+l Point Estimates

Rosenblueth [101] assumed that a function Y = f f X ^ X j ,

..., Xn ) can be approximated by a function H = H(X1 ,X2 ,..

., X n ) [30]? where H is defined by:


26

where,

Yq = Y(x^,X 2 , xn ), i.e. the function evaluated at

the variable means.

Y^ = the function with the value of the ith variable (X^)

shifted one standard deviation

— Y(x, , X5 , ..., x* ± a , ..., x )


1 1 2 1 xi n (2.36)

At the mean values of H and Y, the two functions are equal.

If the random variables are mutually independent, then

Y l ,Y2 f *•*' Y k are mutuallY independent too. Hence the first

two moments can be written exactly as:

H_Ill2
V " f •••I
In_— II“ li
1 Y Y Y i=l Y (2.37)

where,

Y i * YI

Y> = Y (x^f ^ 2 , ••», x. + , ..., x ^ ),

i.e. the function evaluated with the ith variable at its

mean value plus one standard deviation, while the rest

of the variables are at their mean values.


27

• • •»

n ( i + v,2yj ) - i
i=l i (2.38)

where

The difference between this method and the 2n point

estimates is that in this method it is required to work with

one variable at a time while keeping the other variables at

their mean values (in the other method this restraint is not

required). For n greater than 2, the 2n+l point estimate

requires only a small fraction of the number of function

evaluations required for the 2n point estimate.

2.3.4. Three Point Estimates

Gorman [33] derived an expression for a three point

estimate of a random variable, exact for the first four

moments (mean, m; variance o ; skewness, a 3 ; and the

kurtosis, a 4 ). Let Y be a given function, Y =

y(XlfX 2 ,... x n )f of n random variables. To obtain the first

two moments of Y based on the three point estimate method,

the following steps may be followed.


28

1. For each variable determine the three probability mass

concentrations (P+ at x = X + ), (PQ at x = X Q ) and (P_a

= X_) to comply with the first five moments of X:

P + + P 0 +P- - 1 (2.39)

p+x + + P 0 X 0 + P - X - = X (2.40)

p+ (X+ X)2 + p 0 (x 0 X ) 2 + P (X y \2_ 2


X) " X (2.41)

p+(X+ X)3 ♦ P 0 (x 0 X ) 3 + P (X
*»3* a3x°x (2.42)

P + (X+ - X)4 + P 0 (X0 - X ) 4 . P.(X. - X > 4= «4x°x (2.43)

where,

X^ = X + a
X

X = X - a
X

by setting X q = X, solution of these equations yields the

parameters of the three point distribution, P_, P + , P q , X + ,

and X _ , as

3x
1 +
M

a4x “ a 3X
(2.44)
29

a4x " a3X (2.45)

3x
1 -
M

a 4x " a 3X
(2.46)

X_ = X (1 - -^(M-a3x)
(2.47)

X+ = X (1 + -— (M + a3x)
(2.48)

where

2 2
M = v/4(a4x -
a3X + a 3X (2.49)

2. Y _ , Y q and Y + , the function values evaluated at X_ ,Xg

and X + respectively, are now calculated. Next, the mean

is approximated by Y s*Yq .

E[(Y - Y ) k ] « P_(Y_-Y)k+P0 (Y0-Y)k +P+ (Y+-Y)k (2.50)

Results for the mean and variance are not very sensitive

to the skewness and kurtosis of X (Rosenblueth 1981). These


30

statistics are often not available and estimates are

uncertain. It may be considered possible to set skewness

equal to zero and disregard Eq. 2.51, giving the simple

expression:

p* - £•• p0 ■ ! and p- - s (2.51)

X_ = X - /3 ax ; X Q = X and X + = X + /3 ax (2>52)

As before, the three parameters Y + , Y q , and Y_ are

estimated as the functions of X, Y(X+ ), Y(X q ) and Y(X_)

respectively. From those, the mean and the variance are

approximated by

Y = P +Y+ + P0Y 0 + P_Y_

»Y = P+ (Y+ " Y)2 + P 0 (Y0 " Y)2 + P - (Y- " Y)2

2.3.5. Numerical Examples

The following two numerical examples are presented to

demonstrate the procedure and compare the results of the

aforementioned methods.

Example 1

As a numerical example consider the example of

Rosenblueth's (see Gorman [33])


31

with X^ and X 2 are uncorrelated and lognormally distributed.

The mean and standard deviations of function Y will be

evaluated by using the exact method, Simulation, 2n point

estimates and 2n+l point estimates. The following numerical

values are assigned to the parameters:

= 10; = 2? Xj =4 and ax =0.5

Solution using exact m e t h o d . The mean value of Y is

uY = (Xj^2 (X2 )2 (1 + V 2 )(1+V2 )=


1 2
= 100 (16) (1 + 0.04)(1 + 0.015625) = 1690

The variance is

°Y = (* 1 )4 (i2 )4 (1 + VX 1 )2 (1+VX 2 )2 E

where

E = exp I n ((1 + V (1+ V - 1


\

Insertion of the parameters gives oy = 839

The coefficient of variation is

Vy = /(l + 0.04)4 (1 + 0.015625)4 -1 = .4947


32

Solution using 2n point estimate. The 2n possible

combinations of two random variables are four, arranged in a

binary pattern of the sign of the standard deviations,

ro
II
+
(10 + 2)2 (4 2916

o
.
Y1 -

(10 - 2)2 (4 + 0,5)2 = 1296


Y2 =
II

(10 + 2)2 (4 - 0.5)2 = 1764


OJ

(10 - 2)2 (4 - 0.5)2 = 784


II

If the correlation is zero then the weighting function

for each variable equals 0.25.

The mean value of the function Y is calculated using

Eq. 2.34;

Y = | [ 1 2 2 (4.5)2 + 82 (4.5)2 + 122 (3.5)2 + 8 2 (3.5)2] =

= 2916 + 1296 + 1764 + 784 = 169Q

The variance of Y is given by

o2 = E[Y2 ] - (Y)2

= |(29162 + 12962 + 17642 + 7842 ) - 16902 = 621155

The coefficient of variation is

✓621155
Vv - -------- = .4664.
* 1690
33

Solution by 2n + 1 point estimate m e t h o d . From Eq. 2.36,

Yj can be evaluated as

Y* = (10 + 2)2 (4)2 = 2304? Y" = (10 - 2)2 (4)2 = 1024

Y* = (10)2 (4 + 0.5)2 = 2025; Y ~ = (10)2 (4 - 0.5)2 = 1225

2304 + 1024 2304 - 1024


Y, = ------------ = 1664; Vv = = .3846
1 2 *1 2304 + 1024

2025 + 1225 2025 - 1225


Y, = ------------ = 1625; Vv = = .2462
* 2 x2 2025 + 1225

Y Q = 102 (42 ) = 1600

from Eq. 2.37, the mean value of the function H is

1664 1625
H = (1600) ( ) ) = 1690
1600 1600

The coefficient of variation of H is computed using Eq. 2.38

VR = (/(-l + (1 + (.3846)2 ) (1 + (.2462)2 )) = .4663

Solution using a three point estimate. A three point

estimate is made in a manner similar to a 2n+l point


34

estimate, except that the function evaluated at the mean is

used with each variable. By using this method the mean value

of Y and its coefficient of variation were calculated using

Eqs. are found to be,

Y Q = 102 (42 ) = 1600

Y* =16(10 + 2/3)2 = 2900.5

Y~ =16(10 - 2/3)2 = 683.49

Yj =100(4 + .5»/3)2 = 2367.82

Y“ =100(4 - .5j/3)2 = 982.2

Y 1 = |(2900.5) + |(1600) + |(683.49) = 1664

Y 2 = |(2367.8) + §(1600) + §(982.20) = 1625

Y ■ 1600 (if§ffHii§§> - 1690

-§(yI-v 2+!<yo -Yi)2+l(Yl-Yi)2- 417787

4 2 ■ 5(Y2 - V 2 + f(Y0 -y2>2 + i(y2 - y2>2 ' 161245


°Y °Y
Vv = = .3884 and Vv = = .2471
Y1 Yx 2 Y2

Vv = /(I + V 2 )(1 + V 2 ) - 1 = .4703


X *1 *2

Solution bv using simulation. Monte Carlo technique is

used to generate 1000 simulations of the function. The mean

and standard deviation of the sample were calculated and the

values are:
35

Y = 1703.9, and Vy - 0.5166.

For the sake of comparing the accuracy of the

approximate methods compared to that of the exact method the

results of this example are presented in Table 2.1.

Table 2.1 - Results of Example 1

No. Method Mean value Coefficient of variation

1 Exact 1690.0 0.4947

2 Monte Carlo 1703.9 0.5166


1000 points

2n point
3 estimate 1690.0 0.4664

4 2n+l point 1690.0 0.4663


estimate

5 three point 1690.0 0.4703


estimate

Example 2

The maximum negative bending moment, M fi, (over an

intermediate support, B) in a continuous beam shown in

Fig. 2.3, is estimated in terms of the span lengths and load

intensities as follows

+ W 2 (L2.2L3 )l | - W 3L 2l |
mb ----------------------------------------
4(4(L1+L2 )(L2+L3 )-L2)
36

Figure 2.3 - Continuous Beam with three Spans.

Let all the variables in the above equation be considered

as lognormals. Then, the means, standard deviations and

coefficients of variation of these variables are as follows

variable mean standard coefficient


value deviation of variation

30.0 ft. 3.60 ft. 12 %


L1

45.0 ft. 4.50 ft. 10 %


L2

25.0 ft. 5.00 ft. 20 %


L3

4.0 k/ft. .60 k/ft. 15 %


W1

w2 6.0 k/ft. .72 k/ft. 12 %

3.0 k/ft. .54 k/ft. 18 %


W3

The statistics of M fi were calculated using four methods:

Monte Carlo simulation with 1000 points, 2n point estimate,

2n+l point estimate and three point estimate. The results

are presented in Table 2.2.


37

Table 2.2 - Results of Example 2

NO. Method Mean value Coefficient of variation

1 Monte Carlo 870.33 0.1933


(1000 points)

2n point
2 estimate 869.76 0.2030

3 2n+l point 869.80 0.2026


estimate

4 three point 869.78 0.2037


estimate

Examples one and two indicate that Rosenblueth's point

estimates give sufficiently accurate results. Therefore,

Rosenblueth's 2n+l point estimate is utilized in the bridge

system reliability analysis.

2.4 Simulation and Approximation Techniques Used in


Calculating the Statistical Moments of dependent Random
Variables

The random variables dealt with in the previous sections

were assumed to be independent. In practical applications

basic variables will often be correlated. Correlation

sometimes affects the value of the reliability index, 0,

appreciably. In this section the concept of correlation is

investigated.

Let the random variable Y be a linear function of a

number of random variables X.,..., X„ such that:


l n
38

Y = a 0 + a1X 1 +, ..., + anXn , (2.53)

Then the expected value E[Y] is given by

n
E[Y] = an + Z a.E[X.]
U i=l 1 (2.54)

and the variance Var[Y] by

n 5 n n
Var[Y] = Z af Var[X.] + Z^Za.a.Cov[X.,X.]
i-1 1 1 i*j 1 3 1 3 (2.55)

Consider now the case of a set of n mutually correlated

random variables X = ( X ^Xj,..., X R ) with expected values

E[X^], i=l,2,..., n, and covariance matrix Cx . If the Cx is

diagonal then there is no correlation between any pair of

the variables. If general Cx can be written as:

Var(Xx ) Cov(XlfX 2 ) ... Cov(XlfXn )

Cov(X .X,) Cov(X_,x,) ... V a r ( X )


n l n 2 n
(2.56)

A set of correlated random variables (Xlf X 2 , XR )

can be transformed to a set of uncorrelated variables (Y^,

Y 2' ***' Y n^ means of a special transformation X, such

that
39

Y = ST X (2.57)

and

ey = XT ex I (2.58)

where the diagonal elements of Cy are equal to the

eigenvalues of Cx , and %. is an orthogonal matrix with column

vectors equal to the orthonormal eigenvectors of Cx .

The expected values of the vector Y can be evaluated as

shown in Eq. 2.59 [107].

E[Y] = at E[X] (2.59)

Consequently, a set of correlated random variables can

be transformed to a new set of uncorrelated random variables

by using the above mentioned transformation. The reliability

analysis can then be carried out for the new uncorrelated

variables.

The generation of statistically independent random

variables is a direct extension of the methods used to

generate a single random variable using any of the

procedures mentioned in the previous section. This section

describes how correlated random numbers can be generated in

two cases:
40

A. Normally distributed variables. Let Z be a function of n

correlated normal random variables, Z = f(X^, X 2 ,...,

X n ). If the means and covariance of the random variables

{X} are known. Let the mean vector be:

{X} = { X x X 2 , ..., Xn }

and let the covariance matrix of the X variables to be as

given by Eq. 2.56. Then the approximation to find the

statistics of Z is carried out in six steps:

1. Obtain the eigenvalues and the eigenvector of the

covariance matrix of X variables. The eigenvalues

represent the variance vector of an independent set

{Y}.

2. Evaluate the mean values of {Y} using Eq. 2.57.

3. Once the mean and variance of {Y} are evaluated, the

problem becomes similar to that of a set of

independent random variables. In this case any of the

techniques mentioned in Sec. 2.3 can be applied to

obtain a new set of independent random variables Y.

4. The new set of Y is transformed to the original set

variables X. This is done using the inverse of

Eq. 2.57 such that:

{X} = [A] {Y} (2.60)*

5. Evaluate the value of the function Z.


41

6. Repeat steps 3 and 4 to either get enough points to

evaluate the statistics of function Z (case of Monte

Carlo) or the limit of the method (case of

Rosenblueth's methods).

B. Lognormally Distributed variables. In this case, the

random variables can be transformed to normal

variables. Eqs. 2.16-2.17 can be used to find the mean

and the standard deviation of the new transformed

variables. Let Z be a certain function of a set of

random variables, such that, Z = f(Zlf Z2 , ..., Zn ),

where Z ^ , Z 2 , ..., Zn are lognormally distributed. Let X

be another function of X^, X 2 , ..., X n such that X = In

Z, X 1 = In Z^, X 2 = In Z2 and so on then, the covariance

of any pair of the X variables can be written as:

Cy Y “ P v Y ^ Y ^Y
12 12 *1 *2 (2.61)

2 2
Cy y = Py y /ltt(1 + V„ ) In(1 + V„ )
12 *1*2 Z1 2 (2.62)

Where pv v is approximately equal =* p_ „


* 1*2 12

In this transformation the lognormal random variables are

replaced by normal random variables with a new mean,

standard deviation and covariance matrix. The procedure

explained above for normal random variables can now be


42

carried out for the new variables. The mean and standard

de-'iation of the output function will be in terms of the

logarithmic space, i. e. the output will be in terms of In Z

and a(ln Z ) . To get the mean and standard deviation of Z,

the inverse of the transformation mentioned in Egs. 2.16 and

2.17 can be used, namely

Z = e x p [ H T z + |ln(l+v2)] (2>63)

4 = ^ 2 (exp^alnZ^ “ (2.64)

More details on this area are presented in Appendix B.


CHAPTER III

BRIDGE LOAD MODELS

3.1 Introduction

The major load components for highway bridges are dead

load, live load with impact, environmental load (wind,

earthquake), special loads (braking forces, collision). The

load effects for highway bridges vary with the span length

of the bridge. In case of long span bridges, the source of

the most critical loads are strong winds, dead load and

earthquakes. For medium and short spans (up to 300 ft), the

major loads are live load and dead load. The manner in which

heavy trucks are positioned on the bridge is also

important. This includes the possibility of occurrence of

two trucks side-by-side or in one lane at the same time. For

short spans (less than 100 ft), it is particularly important

how the truck weights are distributed on the structure. In

case of very short spans (less than 20 ft), it is the axle

weight along with side-by-side occurrence that dominates the

live load.

In addition to truck weight and configuration, three

important characteristics of bridge load are: The dynamic

factor, the distribution of load to individual members and

43
44

the transverse position of truck on the bridge. The

magnitude of the dynamic effect depends on the bridge

stiffness, roughness of the wearing surface, vehicle

flexibility, vehicle weight and speed. The load distribution

factor depends on several parameters: The stiffness of slab

and girders, the stiffness of diaphragms, the degree of

composite action and the transverse position of truck in the

lane.

The load combination model is based on Turkstra's rule

which is discussed in Sec. 3.5.

Statistical models developed further are based on the

available truck survey (Csagoly and Knobel 1981), survey of

overweight truck files (citations) of the Michigan State

Police (University of Michigan, unpublished), survey of

transverse position of trucks (Al-Zaid, the author and

Nowak) and data available from the weigh-in-motion study

[30].

3.2 Dead Load

Dead load, D, is the gravity load of the structural and

nonstructural elements permanently connected to the

bridge. Because of different degrees of variation and for

convenience, Nowak and Zhou [90] classify the components of

D as follows:

= weight of factory made elements (steel sections,

precast concrete members);


45

Dj = weight of cast-in-place concrete members;

Dg = weight of the wearing surface (asphalt);

= other weights.

The statistical parameters of D are well established for

buildings. It has been found that D is normally distributed

with the mean-to-nominal ratio, X = 1.05, and the

coefficient of variation, VD = 0.05 [24], In safety analysis

it is recommended to use VD = 0.10, rather than 0.05,

because of human errors.

The dead load model for bridges is based on the

available data from literature [90]. The normal distribution

was found to fit the upper tail of all components of the

dead load, D. The mean-to-nominal ratio, XD , is assumed to

vary as follows:

= 1.03 and = 4%
XD1 VD1

= 1.05 and = 8%
XD2 VD2

= 1.0 and = 25%


XD3 VD3

Nowak and Zhou [90] further assumed another factor to

take care of the uncertainty associated with evaluating the

above factors. This factor is called the influence factor,

its mean to nominal ratio, X, is estimated at 1.0 and the

coefficient of variation is estimated at two per cent.


46

3.3 Live Load

Live load, L, covers a range of forces produced by

vehicles moving on the bridge. The effect of live load

depends on wheel force, wheel geometry (configuration),

position of the vehicles on the bridge (transverse and

longitudinal), number of vehicles on the bridge (multiple

presence), stiffness of the deck (slab) and stiffness of the

girders. Because of this complexity, the effect of each

parameter is discussed separately.

Three live load models are considered in this

thesis. The first was developed by the MTC Task Force [35],

in conjunction with the development of the OHBDC [92]. Nowak

and Zhou [90] have revised this model, developing a new

approach to predict the extreme values. The second model

was developed by Ghosn and Moses [30]. The model was based

on a multi-dimensional stochastic process approach that

utilizes the bridge measurement data to obtain the maximum

lifetime distribution of loads on multilane bridges.

Several factors were considered including the occurrence of

side-by-side traffic. The third model is a proposed approach

based on a survey of the overweight vehicles in the State of

Michigan.

3.3.1 Nowak and Zhou's model [90]

In this model the static component of the load effect

is considered. The model is based on Ontario survey data


47

[20]. The concept of Ontario truck formula to determine the

equivalent base length, BM , [19] is used in this model to

develop the bending moment spectra. Values of BM were

calculated for whole trucks and for combinations of axles.

For each pair of weight and equivalent length, the maximum

value of the midspan moment was calculated. The

distributions of the resulting moments are shown in Fig. 3.1

on a normal probability paper. The vertical scale is the

inverse of the standard normal distribution function. The

horizontal scale is the ratio of the moment to the nominal

(OHBDC) moment.

The total number of trucks in life time of the

structure has been established by the committees working on

the OHBDC. The results are given in Table 3.1.

The Ontario surveyed trucks (10000) was considered to

represent an about weekly traffic on class A highway. It was

assumed that the distribution of the bending moment shown in

Fig. 3.1 model the arbitrary point in time moment per lane

for a class A highway.

If the bridge design life time is considered to be 50

year, then the distribution of the maximum 50 year live

load, LgQ, can be considered as exponential based on the

findings of Grouni [35],

Ft (x) = 1 - e“x
50 (3.1)
48

50 y e a r liv e load le v e l

40
60

12, 6 , 18

All spans in m eters


24

M om eiit/O H B D C Moment Ratio

1.0

60
40

30

Figure 3.1 - Distribution Functions of


Moments for Various Spans [90]
49

Table 3.1 - Number of Trucks as Function of Time

Class of Total Numl >er of Trucks


the Road +
Per day . Per 50 Years

A over 1000 over 20 million

B over 250 over 5 million

over 50 over 1 million


C1

less than 50 less than 1 million


C2

+ Classification of roads according to the OHBDC


(1983).

This corresponds to a straight line on the exponential

scale. Therefore, to predict the mean Lgg a line was fitted

to the upper tail of the surveyed data. This approximation

is accepted assuming that the exponential function fits the

upper tail of the load distribution over the region that

contains the design point (see Ditlevsen [23]). However, the

coefficient of variation of Lgg was assumed to be = 11%.

Six heaviest trucks were selected from the Ontario

data. The axle configuration and weights are shown in

Fig. 3.2.

The ratios of the maximum midspan moments were

calculated for different spans. The calculated moment were

divided by the OHBDC nominal moments, so that the ratios


50

42.7 81.0 81.0 86.3 86.3 82.7 46.3 77.4 56.0 M

{ 3.89 J.1.85 { 3.3 ]l.83 ]l.3 ]l.3 [ 3.05 [ 1.91 [ KN

Vehicle No. (1) L=18.41M and W=639.6KN

41.4 92.5 89.0 102.3 96.5 100.8 89.4 M


1 3.51 j.1.4 It 3.76 j l .85 j 3.38 [ 1.85 [ KN

Vehicle No. (2) L=15.75M and W=611.2KN

67.2 64.9 131.7 132.6 49.8 100.1 109.0 M

| 1.83 j 3.43 j l .83 j 2.67.1, 2.74 [ 1.85 [ KN

Vehicle No. (3) L-18.41M and H»639.6KN

44.5 86.7 85.4 72.9 184.6 96.1 M

| 3.48 [l.43 [ 3.96 | 2.46 | 2.44 | KN

Vehicle No. (4) L-18.41M and W»639.6KN

42.7 92.1 90.3 90.3 90.3 101.4 M

[4.06 ll.37 [ 3.00 [ 1.8 j 1.58 j KN

Vehicle No. (5) L«=18.41M and W=639.6KN

49.9 72.9 72.9 111.2 90.7 90.7 M

1 3.53 [ 1.5s[ 3.05 [ 1.8s| 1.37 [ KN

Vehicle No. (6) L-11.35M and W=487.9KN

Figure 3.2 - Configuration of the Heaviest


Vehicles in the 1975 Survey [90]
51

were dimensionless. The ratios were plotted on the

logarithmic scale. The results for several simple spans

ranging from 3m to 60m are shown in Fig. 3.3. Horizontal

scale is the ratio of the moments, while the vertical scale

is the logarithm of the number of trucks exceeding the

moment ratio.
W
The Numbers Refer to -0 o
3
Spans in Meters »-
60, h—
u.
o
i— o
H— 40, S3
o -1 E
3
o
-Q z
E 60. O
3
z E

^ s
O)
o
0.5
Moment/OHBDC Moment Ratio
Figure 3.3 - Upper tails of Moment
Distributions from Truck Survey [90]

Nowak and Zhou [90] assumed that the 50 year population

of trucks is 600 times larger than the number of Ontario

truck survey. They determined the 50 year live load, L 5 q , by

extrapolating a straight line for larger moment ratios.

They determine the mean which corresponds to I n (600). The

distributions of moment due to an arbitrary- point-in-time

truck Fig. 3.1 are extrapolated to predict the maximum 50

year live load on the normal probability paper in Fig. 3.4.


52

Values of the mean and coefficient of variation for the

dimensionless ratio (actual moment due to truck / OHBDC

nominal moment) were determined for different spans and for

both arbitrary-point-in-time and maximum 50 year live load.

The results are summarized in Table 3.2.

Table 3.2 - Parameters of Live Load [90]

Arbitrary-point-in-time 50 year maximum


Span mean/nominal coefficient mean/nominal coefficient
m ratio of variation ratio of variation

3 0.44 0.20 1.0 0.14

6 0.48 0.21 1.05 0.125

9 0.52 0.23 1.12 0.11

12 0.48 0.27 1.12 0.11

18 0.42 0.35 1.12 0.11

24 0.39 0.40 1.12 0.11

30 0.37 0.45 1.12 0.11

40 0.33 0.50 1.12 0.11

60 0.32 0.50 1.12 0.11

3.3.2 Ghosn and Moses' Model [30 3

Ghosn and Moses [30] used a multi-dimensional stochastic

process approach that utilizes bridge measurement data to

obtain the maximum lifetime (50 year) distribution of loads

on multilane bridges. Several factors were included in Ghosn

and Moses' model , including the possibility of side-by-side


53

50 y e a r liv e load le v e l
Function
Normal Distribution

12, 6 , 1 8

All spans in m eters


24
Inverse

M om ent/O HB DC Moment Ratio

1.0

60
40 >

30

Figure 3.4 - Extrapolated Live Load Distributions [90]


54

traffic occurrence. Ghosn and Moses' explained the median of

the maximum bridge moment in a 50 year period in the

following expression:

M “ a m W >95 H g I G r

Where,

m = factor to account for the randomness in the

axle configuration of random truck traffic,

a = the maximum moment effect of the standard

simulation truck with a unit gross weight.

W 9g = truck weight value corresponding to the

upper 5% of the gross weight histogram.

H = factor that relates the moments of the

standard truck (W 9g) to the estimated

maximum lifetime load on a bridge,

g = distribution factor.

I = impact (dynamic) factor.

Gr = growth factor to account for the fact

that truck weights have been increasing

in time.

Values of the parameter are presented in Table 3.3.

The model presented in Eq. 3.2 is used to generate the

load spectrum per girder. The girder load spectrum is used

in this study to calculate the reliability index of the


55

Table 3.3 - Input Data for Eq. 3.3


[30]
r
1 r r..... I .... 1— .. 1.....
jSpan in ft.
Variab .e 30 40 60 80 100 125 150

a (1 ’ a 3.54 4.94 8.4 13.4 18.4 24.4 30.9

m m 1.03 1.08 1.06 1.02 1.0 1.0 .99

vm .2 .18 .17 .12 .10 .08 .07


m

w 12 > W 75.25 75.25 75.25 75.25 75.25 75.25 75.25


.95

H (* ’ H 2.63 2.69 2.75 2.78 •2.8 2.86 2.87

g (2 ) .3 .3 .3 .3 .3 .3 .3
9

1(2’ i 1.2 1.2 1.2 1.2 1.2 1.2 1.2

Q ( 2J 1.15 1.15 1.15 1.15 1.15 1.15 1.15


r 5r

282.1 449.8 806.5 1164.9 1524.0 1973.2 2475.1


V 1’

(1 ’Deterministic variables [30]


( 2 ’Coefficient of variation of these variables do not
affected by the span length. Their values are, .1, .07,
.1 and .1, for Vw , VH , V^, Vj and VGr respectively.

individual girder. Statistical parameters (mean and standard

deviation) of Eq. 3.2 are presented in Table 3.3 for

different spans.
56

3.3.3 Proposed Approach

The 50 year maximum live load model has been developed

in this study based on citations of overweight trucks in the

state of Michigan. Overweight trucks were evaluated as part

of a general plan to collect data on highway traffic

loading. There were 1600 citations available, as the

Michigan State Police keeps them on file for about 6 months

only. The maximum moment and the corresponding critical

position were determined for different spans. Dimensionless

factors, denoted as load ratios, LR, were calculated as the

ratio of the moment due to actual truck, M t , to the

specified moment. Three specified moments were considered:

moment due to AASHTO Truck HS20, moment due to

Ontario Truck, M 0HBDC an(* moment corresponding to Ontario

equivalent base length formula, M fl.

M act M act M act


LR a = — — LRfi = LRq = — -- ■
MHS20 B M'
OHBDC

The moment due to AASHTO HS20 is considered as the basis for

live load model in this study. The other two specified

moments are used for comparison.

Based on the findings of Ditlevsen [23], the cumulative

distribution functions of load ratios LRA , LRfl and LR q , are

plotted on normal probability paper for different

spans. Ditlevsen has shown that any (smooth) distribution of


57

load or resistance may be replaced by its normal tail

approximation for the purpose of evaluating the probability

of failure. The normal tail approximation of a distribution

is the normal distribution that has the same value and the

same density at the design point. The design point is

defined as the limit of an iterative process called the

Rackwitz-Fiessler algorithm [66]. The design point is

normally unique for a pair of smooth distributions. Thus, if

the distribution of load (or of resistance) have the same

normal tail approximation they yield the same probability of

failure and hence the same reliability index. Typical

distributions of live load ratios are plotted on normal

probability paper for 60 ft. span as shown in Fig. 3.5.

These distributions represent the arbitrary-point-in­

time truck loading for three months. The 50 year live load

model considering the live load ratio, LRA , is presented.

The following assumptions were made:

1. The current arbitrary point-in-time live load

distribution is representative for the 50 years to

come.

2. Three months truck loadings are independent random

variables.

3. On short and medium span bridges, only one heavy

truck is considered to cause the maximum bending

moment.
58

CM / M-AASHTC

/ M-BMAX
INVERSE NORMAL DISTRIBUTION FUNCTION

/ M-ONTARIO

CM

o
CO

CM

0.00 0.60 1.20 1.80 2.40 3.00

LOAD RATIO

Figure 3.5 - Typical Distribution of


the Load Ratio (60 ft. Span)
59

4. The surveyed truck is considered as representative

sample of a three month population. The 50 year

population of trucks is therefor assumed to be 200

times larger than the surveyed number.

Hence, the 50 year live load distribution, Fgg(x), is given

by:

F 50*x * " ^F ^x ^ 20° (3.3)

where,

F(x) = three month distribution of live load.

The 50 year load distribution is then evaluated based on

Eq. 3.3. A design point is chosen in the upper tail of the

50 year load distribution. A straight line is fitted

tangent to the original distribution at this point [23],

Hence, the mean and standard deviation are evaluated.

Different spans are considered.

To compare the three models, Ghosn and Moses' model

Eq. 3.2 was modified to obtain the maximum 50 year live load

in terms of Load ratio, LRA , rather than moment. Hence,

Eq. 3.2 becomes

a m W Qc H
ip . »
MRS20 (3.4)
60

where,

T = maximum live load moment in the bridge measured in terms

of the moment due to AASHTO Truck HS20. Other parameters

are the same as in Eq. 3.2.

Simulation was used to generate the cumulative

distributions of the three models. The results are plotted

on normal probability paper, in Figs. 3.6 - 3.8. The three

models are compared in Table 3.4.

3.4. Dynamic Load

Dynamic effect is considered as an integral part of the

live load model. The major factors affecting dynamic load on

a bridge include surface condition (bumps, potholes),

natural frequency of the bridge (span length, stiffness,

mass) and dynamics of the vehicle (suspension, shock

absorbers). Practically it is impossible to predict the

percentage contribution of these three

factors. Traditionally, dynamic effect is measured by an

frequency of the bridge, impact factor. It was calculated as

function of the loaded length of the bridge [2]. OHBDC [92]

specifies the dynamic load allowance as a function of the

natural frequency of the bridge.

Nowak and Lind [89] considered the mean impact effect,!,

as a fraction of the mean largest live load effect, L. They

found that this fraction is a function of the first flexural

frequency of the bridge. They presented an approximate


61

4.20
NOW AK and ZHOU MODEL

GHOSN and MOSES MODEL


3.40
FUNCTION

PROPOSED M ODEL
2.60
DISTRIBUTION

1.80
1.00
0.20
NORMAL

0.60
-
INVERSE

1.40
-
2.20
-
3.00
-

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00

LOAD RATIO

Figure 3.6 - Distribution of 50 Year Load Ratio


for 60 ft. Span (Comparative Results)
62

rvi * NOW AK and ZHOU MODEL

♦ GHOSN and MOSES MODEL


FUNCTION

* PROPOSED M ODEL

CO
CM
DISTRIBUTION

00
NORMAL

CM
<=>■

(O
INVERSE

CM
CM*

CO

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00

LOAD RATIO

Figure 3.7 - Distribution of 50 Year Load Ratio


for 80 ft. Span (Comparative Results)
C\l NOW AK and ZHOU MODEL

GHOSN and MOSES MODEL

co
FUNCTION

PROPOSED M ODEL
o
<o
CM
DISTRIBUTION

00

CM
NORMAL

CO
INVERSE

a
CM

0.00 1.00 2.00 3.00 4.00 5.00 6.00 7.00

LOAD RATIO

Figure 3.8 - Distribution of 50 Year Load Ratio


for 125 ft. Span (Comparative Results)
64

Table 3.4 - 50 year Live Load M o d e l s ' * ’

span in ft.
model statistics 60 80 100 125

Nowak & Zhou mean 1.69 1.77 1.92 2.05

COV 0.11 0.11 0.11 0.11

Ghosn & Moses mean 2.2 2.45 2.55 2.65

COV 0.18 0.12 0.12 0.07

Proposed Model mean 1.85 1.95 2.15 2.21

COV 0.1 0.1 0.09 0.1

'*’ All values in this table are in terms of AASHTO


moment

formula to calculate the first flexural frequency for simple

span bridge such that:

frequency in H z - brTdge length in ft. (3.7)

and for continuous span the same formula, but multiplied by

two. They assumed the coefficient of variation of the

dynamic load, Vj as 45%.

Ghosn and Moses [30] consider impact as part of the

live load model. They increased the live load by a factor I

to count for the dynamic load effect. The mean and

coefficient of variation of I was considered equal to 1.2

and 10% respectively.


65

Nowak and Zhou [90] treated the impact effect separately

from live load model. Their model is based on the data

provided by the OHBDC committee on dynamic load. Further

Nowak and Zhou assumed that impact (dynamic component of

live load) is independent of the static component and follow

the lognormal distribution. The findings of Nowak and Zhou

are listed in Table 3.5.

Table 3.5 -- Dynamic Load Factors

Mean Value Standard Deviation

Type of structure Assumed Assumed


range Value range Value

Steel 0.08-0.2 0.14 0.05-0.2 0.1

prestressed concrete:

AASHTO type girders 0.05-0.1 0.09 0.03-0.07


00
1 0.05
o

o
.
o
.

Box girders and slabs 0.1-0.15 0.14 0.3

Others like frames, 0.1-0.25 0.17 0.12-0.3 0.26


trusses, ... etc.

3.5 - Load Combinations

The total load effect in highway bridge members is a

joint effect of dead load, live load (static and dynamic),

environmental loads (wind, snow, ice, earthquake, earth

pressure and water pressure) and abnormal loads (emergency

braking, collision forces) [90],

Q = D + L + I + E + A (3.7)

Eq. 3.7 represents the joint effect of load components.


66

The distributions of single load components were

developed as described in the proceeding sections of this

study. The distribution of the joint effect is based on the

so called Turkstra's rule.

Turkstra [108] observed that a combination of several

load components reaches its extreme, when one of the

components takes on an extreme value and all other

components are at their average (arbitrary-point-in-time)

level. For example, the combination of live load with

earthquake produces a maximum effect for the life time, T,

when either

1. Earthquake takes on its maximum expected value for T

and live load takes its maximum expected value

corresponding to the duration of earthquake (about 30

seconds), or

2. Live load takes its maximum expected value for T and

earthquake takes its maximum expected value

corresponding to duration of this maximum live load

(time of truck crossing the bridge).

In practice, the expected value of earthquake in any

short time interval is almost zero. The expected value of

truck load for a short time interval depends on class ofthe

road. For a very busy highway it is likely that there is

some traffic at any point in time. Therefore, the maximum

earthquake may occur simultaneously with an average truck

passing through the bridge.


67

For a linear combination of loads, Turkstra's rule can

be written as follows:

< W = ““ (3.8)

where,
II

+ + + E +
o

Dmax L ave ave A ave


M

Iave
ll

+ + + E +
©
ro

D ave Lmax Iave ave A ave


II

+ + + +
©
CO

D ave L ave Imax E ave A ave


II

+ + + +
©

D ave L ave Iave Emax A ave

+ + + + A
Q 5 - D ave L ave *ave Eave max

In all cases the average load value is calculated for

the period of time corresponding to the duration of the

maximum load in that line Eq. 3.8.

The joint distribution can be modelled using the central

limit theorem of the theory of probability [10], If the

number of components is large, and if the average values of

components are of the same order then, sum of several random

variables is a normal random variable. If one variable

dominates (its average value is much larger than any other),

then the joint distribution is close to that of the

dominating variable.

For each sum in Eq. 3.8, the mean and variance of the

sum are equal to the sum of means and sum of variances of

components, respectively.

The distribution of Q is that which minimizes the

overall structural reliability. Usually it is with the


68

largest mean value. If the means are similar, then the

largest standard deviation may identify the governing

combination. In some cases the analysis has to be performed

for several Q^'s to determine the one which governs.

The identification of the governing load combination is

important in the selection of the optimum load factors

(including load combination factors).

Values of D 2 , Dg, L and I as percentages of the

total load are given in Table 3.6.

Table 3.6 - Typical Percentage Va l u e s 1 of Major


Load Components (Nowak and Zhou 1985)

Type of
structure D1 D2 D3 L I <* >

Steel 5-15 25-55 5-10 30-50 10-20

Prestressed
concrete

girders 20-55 5-10 25-50 5-10 10-25

slabs 25-80 5-10 20 10-15 -

(*> All load values are fraction of the total load


except the dynamic load, I, which is a fraction of the
live load.

The parameters (mean and variance) of the joint effect of

the loads listed in Table 3.6 can be calculated using

Turkstra's rule. All components of dead load are time


69

invariant, so their 50 year mean value is the same as

arbitrary point-in-time value.

The 50 year mean maximum combination of L and I can be

calculated by taking the 50 year mean maximum L and adding

average value of I ( as given in Table 3.2 ), according to

Turkstra's rule. The result is somewhat conservative

because L and I are correlated. Dynamic load, as a

percentage of live load, is usually smaller for heavier

trucks.

The formula for the maximum load is

Q ^ x = D + { F U ) ) 200 ♦ I (3>9)

3.6 Transverse Location of the Truck Traffic

A vehicle's position on the bridge is important to

determine the load distribution to various girders. The

wheel load may be applied to various points of the pavement

within the curbs. However, the transverse position is a

random variable. To establish the distribution function of

this variable a special survey was carried out by A1 Zaid,

the author and Nowak [91].

The surveying team measured the frequency of truck

traffic at various distances from the curb or pavement edge

lines. The study was conducted at the interstate highways in

the South Eastern Michigan. The resulting density functions

were very consistent, and they do not seem to depend on

location. A typical curb distance histogram and distribution


70

for Interstate 194 is shown in Figs. 3.9 and 3.10

respectively. This distribution served as a basis for the

reliability analysis.

The frequency histogram for the transverse position may

be affected by presence of special signs, obstacles, etc.

These cases require an individual approach and are not

considered here.
71

Shoulder Right Lane Left Lane Shoulder


h— ►H------ H--H
Figure 3.9 - Typical Curb Distance
Histogram for Interstate Highway, 1-94

1 •

2-

5 .. <i

10 --

20 --

30-•

40- shoulder
50 i—

60-- |

7 0 --

8 0--

90--

954-
Figure 3.10 - Distribution of Curb
Distance on Normal Probability Paper
CHAPTER IV

BRIDGE DECK MODEL

4.1 The Grid Model

The approximate representation of bridge decks by a

grillage of interconnected beams is a convenient way of

determining the general behavior of the bridge under load.

The method of grillage analysis involves the idealization of

the bridge deck through its representation as a plane

grillage of discrete interconnected beams. Fig. 4.1 shows a

typical bridge system and the corresponding grid model.

The transverse beams are located at the abutments and at

intermediate positions spaced at not more than 1.5 times the

spacing of longitudinals [113].

The following assumptions are considered in this work:

1). The members of the grillage lie all in the same

plane,

2). All loads are perpendicular to this plane,

3). Deformations are small compared to the overall

dimensions of the structure, and

72
ACTUAL STRUCTURE

Transverse
\C o m p o n e n t

GRID MODEL

Figure 4.1 - Typical Bridge System and


the Corresponding Grid Model
74

4). The members are rigidly jointed at their inter

sections.

4.2 Geometric Properties of the Grid Elements

The analysis of plates in flexure by means of the grid

framework method was investigated by many researchers [116,

25, 51, 68]. All the researchers used almost the same

approach in simulating the plate by an equivalent grid

framework. Flower [25] assumed in his work that the

differential equations of a plate and grillage are identical

if the spacing between members are small compared to the

overall dimension. Yatteram and Husain [116] simulated the

plate element by six members to form a framed structure as

shown in Fig. 4.2. They derived the properties of the grid

elements by comparing the behavior of the grid frame

elements with that of the plate. Yatteram and Husain [116]

derived a special formulae to define the properties of the

grid elements for different structural cases.

1. Case of isotropic plate with uniform thickness. The

plate structure is divided into several plate panels.

Each panel is represented by six grid members. Two

members lie in the longitudinal direction; there are two

diagonals and two others in the transverse direction as

shown in Fig. 4.2. The properties of each group can be

defined as follows:

Diagonal members:
75

XI

ik
Y
I
4
Q
o
>
(A

i t

Longitudinal Direction

Plate Element Six Members Framed Structure

Figure 4.2 - Plate Element and the


Equivalent Framework Model

_ E u R3 C L
d " K (4.1)

OWhere,

EI^ = flexural rigidity in diagonal direction

L = width of plate element

Ld = length of diagonal members

= length of plate element

R = —

K = —

C
24(1 - m > (4.2)

h = plate thickness
76

GJ =0.0

Longitudinal Memb e r s ;

EI1 = L E (1-K2 v ) C

Where,

EI^ = flexural rigidity in the longitudinal direction

G^ = E L(1-3 m ) C

Transverse Members:

Where,

EIt = flexural rigidity in transverse direction

GJt = E K L (1-3/i) C

2. Case of slab rested on stiff longitudinal girders

In this case the longitudinal members will be coincide

with the main girders. The transverse elements will coincide

with end and intermediate supports. They are spaced not more

than one and a half times the spacing between longitudinal

girders. The diagonal elements will not be considered


77

[51]. The properties of longitudinal and transverse members

can be defined as follows.

Members coinciding with the main girders

The elements in longitudinal directions are composed of

either steel girder, prestressed concrete girder or

reinforced concrete beams and part of the concrete deck

slab. In case of steel or prestressed concrete girder a

complete composite action is assumed to be exist between

concrete slab and the girders. In case of reinforced

concrete beams, it is assumed that the slab and the beams

are cast monolithically. Jaeger et al [51] ignored the

differential rotation in the longitudinal direction. The

same assumption will be adopted in this work. This

assumption will lead eventually to neglecting the diagonal

members. The flexural and the torsional rigidities are

obtained by utilizing the member strength nonlinear analysis

program described in section 4.3.

Transverse Members

The members in this direction are part of the concrete

slab. The flexural and torsional rigidities in the elastic

stage are calculated based on the assumptions of isotropic

plate theory, i.e. the flexural rigidities in the two

orthogonal directions and half of the total torsional

rigidity are all equal [21]. Details of calculations are

given in section 4.3.


78

4.3. Strength of the Grid Elements

The resistance of the equivalent grid elements are

discussed in terms of the following parameters: Flexural

rigidity, El, torsional rigidity, GJ, ultimate moment

capacity, M u , ultimate torsional capacity, T , ultimate

shear capacity, Su , and ultimate Punching shear capacity of

the concrete slab, Sp. These parameters are evaluated for

different types of cross section.

Many researchers investigated the behavior of composite

steel girder sections. Heins [44] derived a formula to

calculate the effective flange width. Yam and Chapman [115]

investigated the effect of slip on the ultimate capacity of

the composite cross section. Hamada and Longworth [36]

considered the ultimate strength of composite beams and the

mode of failure. They found that crushing of concrete in the

positive moment region is the dominant failure mode,

provided the longitudinal reinforcement in the cross section

is at the minimum level.

The principal stresses considered in this work are

flexural, torsional and shear stresses. In this research a

modified and proposed approaches are presented to predict

the moment versus curvature and the torsion versus angle-of-

twist relationships. Shear strength of the section was

evaluated based on AASHTO recommendations. Material

properties (concrete, reinforcing steel and structural


79

steel) considered in this study are based on the stress

strain relationships showed in Fig. 4.3.

4.3.1 Moment-Curvature Relationship

The moment-curvature relationship is used to determine

the linear and nonlinear flexural behavior of the cross

section at different levels of loading. The analysis uses an

iteration process. The actual cross section is modeled by a

set of thin rectangular elements as shown in Fig. 4.4. The

depth of the neutral axis is guessed first and then after

several iterations the internal equilibrium is obtained. The

strain is assumed linear through the cross section. A

specially developed computer program is used to perform the

iterative analysis to get the location of the neutral

axis. The Method is generalized to handel a wide variaty of

cross sections which are used in different types of

bridges. Some of these sections are presented in Fig. 4.5.

Newton-Raphson method is used to form the convergence

criterion which is explained in Eq. 4.7 with reference to

Fig. 4.6,

e i+ l = «i + P 4 ----— )
F. . - F. (4.7)
Where,

ej+1 = the bottom strain to be considered in the next

step;

= the bottom strain resulting in the present step;


80

£(A3

in
(A

b
in

0.80 1.60 2.40 3.20 4.01

Strain /1000

Concrete

r~

3
£5=35e*s»»«

sI
"I
t s
s

3
o’
tn
to
ui 3
oc 3

3
e>
CN

3 ...........................................
=0.00 10.00 20.00 30.00 40.00 S0.00 60.00 70.0

Strain /1000 Strain /!000

Prestressing Steel Structural Steel

Figure 4.3 - Stress-Strain Relationships


Actual Section Idealized Section

Figure 4.4 - Typical Composite Steel Section

= the bottom strain obtained in the previous step;


i-1

= the unbalanced internal force resulting in the

present step; and

= the unbalanced internal force resulting in


F i-1

the previous step;

The procedure is repeated for increasing top strain

level until enough points are obtained to establish the

moment-curvature relationship. A typical curve is plotted in

Fig. 4.7.

Two assumptions were made in the analysis:


82

Composite Section T Section

w m ?
P #
‘"”§ |
w m
if' te f
m
m
l
Double T Section

v.-v.v.,
fe&
.jW
1
L*.«*•••••
>•••••*

w m m m
Box Section

Figure 4.5 - Typical Sections commonly used


in Prestressed Concrete Bridges
83

Top Strain

X
Bottom Strain

Figure 4.6 - Typical Strain Diagram in a Composite Section

1). The effective flange width (bg ) can be used rather

than the total flange width to account for the

effect of shear lag.

The effective flange width was studied by Jaeger and Bakht

[51] and Heins [44], Heins defined the effective flange

width as follows.

For interior girder

2be = b C 61? b + 0.702) (4 ,

For exterior girder

be = b(0.873 - £)
(4.9)
84

7139.99
MOMENT (Kip.In) (XICT1)
9999.99
4800.00
3600.00

84
INCREMENTAL

*© \ ° ;* ■? , ^ / o V ’ O o °‘ n ' 'o '- 0 * < * ' 4

z 2sT-^
2400.00

W3CX135
1200.00

i).00 10.00 20.00 30.00 40.00 50.00


CURVATURE.

Figure 4.7 - Typical Moment-Curvature


Relationship for Composite Steel Section
85

jr = aspect ratio;

b = specified flange width;

L = span length;

bg = half of the effective flange width in the case of

interior girder, or flange width in the case of an

exterior girder.

Based on the above expressions, it was found that the ratio

of the effective flange width to the total width can be

taken as equal to 0.74 if the aspect ratio is within the

range of five percent to seven per cent.

2). There is a complete composite action between

concrete and steel section.

The effect of slip was neglected based on experimental

and theoretical work done by Kurata and Shoda [56] who

observed that slip is not important in the ultimate loading

capacity analysis. Yam and Chapman [115] found that the

ultimate capacity because of slip is reduced by six per cent

to eleven per cent with respect to the ultimate capacity.

4.3.2 Torsion-Twist Relationship

Two methods are considered. One is developed by Michell

and Collins [72] and extended here to be used in the

analysis of composite prestressed concrete sections and

reinforced concrete T-Beam section. The second one is

developed in this study to be applied on composite steel


86

section and denoted as the incremental twist method. The two

methods are explained in the following section:

Extended Michell and Collins Diagonal Compression Field


Theory. (DCFT).

Mitchell and Collins [72] developed the DCFT to predict

the true characteristic of a reinforced concrete section

under a pure torsion. The DCFT is extended in this work to

predict the behavior of a composite prestressed and

reinforced concrete T-beam section. Several assumptions are

made:

1. The elastic theory is applicable to the whole section

in the pre-cracking stage.

2. Torsional shear is resisted by diagonal compression

in concrete and the tensile stresses in the

transverse and longitudinal steel.

3. It is assumed that the behavior of the actual

section, shown in Fig. 4.8 can be modeled as shown in

Fig. 4.9.

4. The over hanging parts of the slab bL and bR

Fig. 4.10 can not carry any more torsion after the

elastic stage. Hence, only the web part will take

any increment in the torsional loading.


87

Figure 4.8 - Actual Section

12.5 "
i r

Figure 4.9 - Idealized Section


88

5. The stirrups are closed and the longitudinal steel is

uniformly distributed.

The procedure of the DCFT is programmed and tested with

some other theories on a simple L shaped reinforced concrete

section Fig. 4.10. The procedure is summarized as:

1. Compute the elastic torsional capacity of the whole

section using the elastic theory,

2. Compute the angle of twist corresponding to this

torsional capacity, and

3. Using the DCFT, compute the torsion vs twist

relationship for the web part of the section under

increasing torsional loading.

The Incremental Twist A p p roach.

This method is suggested in this work to predict the

behavior of a composite steel section under an incremental

torsional stresses. The method is approximate and it is

modified based on the findings of Heyman [47]. Heyman found

that, if the stress-strain relationship in shear is elasto-

plastic, then the maximum shear stress that can occur has a

value of tq , where tq is the shear yield stress in the

structural steel. Further, Heyman assumed that the

phenomenon of torsional stresses can be represented by

imagining that planes of constant slope (equal to the yield

stress) are constructed around the edge of the membrane. As


1000-

80 0-
20
G
•H 4i
I
6 00 -
Torsion

4 00-

200-

rad./in
10 14 10
Twist

Figure 4.10 - Torsion Vs. Angle of Twist Relationship


90

the torsion is increased, then the pressure is i n c r e a s e d * ->■

above that corresponding to the elastic limit, the membrane

will come into contact with the roof slopes. Further

increase in pressure will result in further areas becoming

plastic. In fully plastic condition, the shear stresses

become parallel to the edges. The Phenomenon is illustrated

in Fig. 4.11. The findings of Heyman is further idealised in

this research as shown in Fig. 4.12. This idealization forms

the basis of the suggested method where .

The concrete slab failure in torsion is assumed to be

ductile (i.e., the concrete will retain its yielding

torsion). Hence, any additional torsion beyond the elastic

torsional capacity of the section will be resisted by the

steel section only. There fore the theory of thin walled

section can be applied. A twist angle is applied

incrementally on the section. The response of the section is

calculated in terms of torsion at each incremental step.

The procedure is described in the following steps:

1). Calculate the initial angle of twist (0j) which is

enough to cause the concrete flange to be yielded in

torsion.

$ = Tntax
G X K (4.10)

where,

Tm a V = 5 i/f~
lUoX c as defined by Hsu (1984);
91

k Jl
k
3 » E =
T
Elastic State Initial Plastification

k _L
k
a
T
Partial Plastification Complete Plastification

Figure 4.11 - Typical Thin Section under


Different State of Torsional Stresses [47]

k
a
T
Elastic State Initial Plastification

k H ii. k
1
a
T
Partial Plastification Complete Plastification

Figure 4.12 - Idealized Thin Section Under Different


State of Torsional Stresses (Used in this Study)
92

f = Maximum compressive strength of concrete


c

G = Shear Modulus in elastic stage

X = Thickness of concrete slab:

8 1
K
n2 cosh n2ffxY
(4.11)

where,

Y « total width of the flange;

Y
K 1 if — > 5 and
X

n = number of the series terms

2). Compute the total torsional response of the section

due the initial angle of twist 9 j

N 1 3
T e = 2 (I G i 9 I X i Y i C i }
e i=l 3 1 1 1 1 1 (4.12

where,

N = number of parts forming the section.


1,1.
Xj = width of i part and Yj is its length and

its shear modulus;

9 j = initial angle of twist computed from Eq. 4.10

T g = maximum elastic torsional capacity of the

section.

192 X. xY.
uaau
(4.13)
93

3). Calculate the maximum torsional stress in steel for

a given r using Eq. 4.10.

4). Check the penetration of the yield in the structural

steel by comparing the calculated rmax with the

maximum allowable torsional stresses in steel. If

the calculated is greater than allowable then


max s
calculate the width of the elastic portion X 0 of

each part of the section from Eq. 4.10.

5). Calculate torsional capacity of the steel parts

using the sand-hill analogy. For a thin walled

rectangular section, the torsion capacity of the

yielded part Tp is

T . * 2 rmax (3Y - X)
(4.14)

Where,

T = Maximum torsional capacity of a rectangular

section if the section is completely yielded

the rest of the parameters are as before.

Tmax = yield stress structural steel in shear

/3
94

= flexural yield stress.

The capacity of the rest of the steel is calculated

using the elastic theory.

If the steel parts of the cross section is partially

yielded, then Eq. 4.15 is applied to calculate the torsional

capacity.

(3Y1 - X 1 )
1 max
+ t
e
6
(4.15)

where,

T is the torsion obtained from Eq. 4.11

is the torsion obtained from Eq. 4.10, replacing X and Y

by and Y^.

The procedure mentioned above is repeated until all

steel is yielded. Example 4.1 is presented to clarify the

procedure.

Example 4.1

The procedure will be demonstrated on a composite steel

section as shown in Fig. 4.12. The given data is: concrete

strength, f = 4000 psi steel yield stress, F = 36 ksi


c y
section dimensions are as shown in Fig. 4.14

Solution
95

7.25 "
.74"
I
W 3 3 x 118

31.38 "

.74"
.74"

Figure 4.13 - Composite concrete steel section

The first step is to get the value of r which corresponds

to the maximum torsional capacity of the concrete flange,

t = 5i/4000 = 316.2 psi

ei ~ 1584400 x 8 = 2,5x10 5 rad/inch

192 x 7.25 „
C = 1 e tanh = .946
concrete _5 „ Q . 2 x 7.25
7T5 X 84

192 X .
r* - i Lanu ff x I I »48 _ nr
1
CJ1

steel flanges ~ 1 5 v Afl tann 2 x .74 " *96


11.<
X

192 x .55
Csteel web 1 " 5 „ ,, .Q tanh ^ X ^ S ^ = *"
IT5 X 11 .48

192 x .74
Ccover plate 1 5 g - tanh 2 x ?75 = ,95
7T X 3 .0
96

Tl = - (1584400 x 7.253 x 84 x .946 + 2 x 11153846 x

.743 x 11.48 x .96 + 11153846 x .553 x 31.38 x .99

+ 11153846 x .743 x 9.5 x .95)

= 33.42 k.ft.

The second step is to get the value of t which is enough

to cause initial yielding in the steel parts.

Fv
rsteel = = 20785 P S 1 '
✓3

from Eq. 4.10 it is obvious that first yielding will occur

in the thicker plate i.e. in the cover plate.

&2 = n i 5 3 0 4 6 ^x . 7 4 = 2 4 8 , 5 x 1 0 rad/inch
-5
T2 = T2 + 2 4 8 , 5 x 1 0 — ( 2 x 11153846 x .74 3 x 11.48 x .96

+ 11153846 x .55 3 x 31.38 x .99

+ 11153846 x .74 3 x 9.5 x .95)

= 47.09 k.ft

The third step is to get the value of r when yielding

starts at the web.

# 3 = 111538468JC . 5 5 = 3 3 8 , 8 x 1 0 rad/inch

At this level of twist the flanges and cover plate are

partially yielded. The elastic portion of these parts should

be of thickness = .55 at the core of the element. The

plastified part is at the edges with thickness = * 7 4 2 * ^ *

The total torsion of the section at this stage is


97

T3 = T2 + 338.8x10 5 (2 x 11153846 x #553 x 11>48 x>96

+ 11153846 x .553 x 31.38 x .99

+ 11153846 x .553 x 9.5 x .95) +— ■

(.742 (2(3 x 11.48 - .74) + (3 x 9.5 -.74))

-.552 (2(3 x 11.48 -.55) + (3 x 9.5 - .55))) =51 k-ft.

The above procedure can be continued until enough points are

obtained. The torsion versus twist curve is shown in

Fig. 4.14.

4.4 Strength Reliability of Single Elements

The random nature of element's strength is due to

several factors. These factors can be classified into three

major categories:

1). Material: strength of material,modulus of

elasticity, cracking stress, chemical composition.

2). Fabrication: geometry, dimensions, section modulus.

3). Analysis: approximate methods of analysis, idealized

stress and strain distributions.

The statistical properties of the basic variables were

taken from the available data in the literature [26, 74, 75,

76, 54 and 105] and by engineering judgement. The basic

variables considered are mainly the dimension of the section

and the material properties involved in defining the stress-


98

60

50
Torsion k-ft.

40

84
30

7.25

20

w 33*118

10

0
0 50 100 150 200 250 300
-5
6 x 10 rad./inch

Figure 4.14 - Typical Torsion Vs. Angle of Twist


Relationship for Composite Steel Section
99

strain relationship. Fig. 4.3 present families of stress-

strain curves for structural steel, prestressing steel and

concrete. Mean values, standard deviations and distribution

required to define these curves are presented in Table 4.1.

The nonlinear section analysis is performed using the

basic variables as random variables rather than

deterministic variables. Families of moment-curvature

relationships are obtained for several types of sections,

namely composite steel section, composite prestressed

concrete section and reinforced concrete section. A typical

examples for composite steel and prestressed concrete

sections are presented in Fig. 4.15.

Reliability analysis is demonstrated on a single girder

in the the next example.

Example 4.2

A girder was designed for a 60 ft span bridge. The

bridge is composed of five girders spaced at 7 ft. The

section analysis procedure and simulation were used to

estimate the values of the mean and standard deviation of

the girder's strength. Ghosn and Moses' load model [30] is

generated as shown in Fig. 4.16. The obtained results were:

mean value of the ultimate strength = 2350 k-ft,

standard deviation of the ultimate strength = 190 k-ft,

Load spectra per girder = 875 k-ft,

and the standard deviation of the load = 340 k-ft.


100

Table 4.1- Statistical Characteresties of input Data


for .Section Analysis

Specified Mean Distri


Variable Value Value cov -bution References

Concrete

3000 psi .85 £ .18 lognormal Ellingwood


£'c c
1980
*
4000 psi .85 I .18 lognormal Ellingwood
fc c
1980
*
5000 psi .80 fl .15 lognormal Ellingwood
fc c
1980

Structural
Steel

1.07 Fy 0.065 lognormal Kennedy


Fy fy 1980

0.02 lognormal Kennedy


Es Es Es
1980

- .06 0.1 lognormal <*>


*H

Geometric
properties

Bridth b b 0.014 Normal Kennedy


1980

Dipth d d 0.014 Normal <«>

Prestressing
Steel

1.027 Fy 0.022 lognormal Siriaksorn


fy Fy 1980

F 1.037 Fu 0.0142 lognormal Siriaksorn


u Fu
1980

E - 29320 0.01 lognormal Siriaksorn


ksi 1980

A A A 0.0125 lognormal Siriaksorn


IQon
0.1 Gamma (*)
dps dps dps
with k=3

(1,The data is not available. Assumed parameters are based on judgement.


101

x<

z;

z
uj
si

30.00 30.00 40.00


CURVATURE,
Composite Steel Section

S'
w

25.00 37.50 50.00


CURVATURE.
Composite Prestressed Concrete Section

Figure 4.15 - Typical Moment-Curvature Relationships


102

The reliability index is calculated using the normal model

(Eq. 2.13) as:

2350 - 875
(3 = ------------------------ = 3 . 7 8

l/1 9 0 2 + 34 02

The distribution of the yield moment for each section is

determined. Flexural rigidity and ultimate moments are

evaluated in terms of the section flexural capacity. Other

strength parameters are considered as deterministic

variables.
103

® Ghosn £ Moses Model


FOR 60 Ft. SPAN
INVERSE NORflRL DISTRIBUTION FUNCTION

C3
o

CD

0.Q0 250.00 500.00 750.00 1000.00 1250.00 1500.00 1750.W

MOMENT K.Ft

Figure 4.16 - Distribution of the Maximum Load per Girder


CHAPTER V

ULTIMATE RESISTANCE MODEL OF BRIDGE SYSTEM

5.1 General Description

Bridge engineers are increasingly concerned about the

overload behavior of highway girder bridges. This is due to

age of these bridges (some are very old), weathering

deterioration and continuous demand of the trucking industry

to increase the maximum allowable truck load. The actual

stresses often exceed the original design stresses. Another

problem is issuance of overload permits. Current procedures

involve a considerable risk due to the lack of adequate

analytical methods. Therefore, in this study a practical

method is developed to evaluate the ultimate strength of the

bridge. The approach is based on the grid model with the

direct stiffness method and the incremental load procedure.

The uncertainties in bridge design and evaluation are

involved in the methods of analysis and bridge loading. The

effect of these uncertainties can b$ evaluated through

reliability analysis. Structural safety is measured in terms

of reliability index. A proposed procedure to bridge system

reliability analysis is presented.

104
105

The developed procedure can be used to perform

sensitivity analysis to identify the most sensitive strength

parameters. Effort can then be used to improve the

properties of some critical strength parameters, which will

considerably increase the safety level of bridges.

5.2 Ultimate Strength Analysis

Very little information is available, either on full-

scale testing or on the methods of analysis, to determine

the ultimate load-carrying capacity of highway bridges

[13,12]. Some methods were developed for composite bridge

systems at ultimate loads based on orthotropic plate theory

by Heins and Looney [42], modified yield line theory by

Reddy and Hendry [100], finite difference method by Kue and

Heins, [55] and Heins and Fan, [44] and finite element

methods, [1 1 2 ]. Wegmuller [1 1 2 ] showed that finite element

results agreed with the predicted results. The use of

finite element methods in analysis is limited because of a

considerable amount of computer time involved. The

orthotropic plate theory, the yield line theory and finite

difference methods are based on modeling the bridge deck as

a continuum plate with uniform strength properties in all

directions. This assumption does not apply if the strength

of each girder is a random variable that is independent or

partly dependent on the strength of other girders. In this

study, the grid model and direct stiffness method are

combined together with the incremental load procedure to


106

predict the post-elastic behavior of three types of highway

bridges. The following steps are included:

1. Modelling the bridge deck as a set of grid elements and

computing their properties (Chapter IV).

2. Defining and assembling the general form of the element

stiffness matrix that can be easily updated at each

level of loading.

3. Distributing the wheel loads to grid nodes and Forming

the load vector.

4. Describing the procedure.

5. Presenting numerical examples to verify the efficiency

and accuracy of the proposed method.

In evaluating the ultimate strength capacity of grid

system the following assumptions are included:

1. The members of the grid lie all in the same plane.

2. All loads act perpendicularly to this plane.

3. Deformations are small compared to the overall

dimensions of the structure.

4. Joints between longitudinal and transverse elements

are considered to be rigid.


107

5.2.1 Element Stiffness Matrix

In the direct stiffness method the displaced shapes of the

structure must satisfy compatibility and equilibrium

conditions. The matrix form of the direct stiffness method

can be written as:

[ K ] {D} = {P} (5.1)

Where:

[K] = stiffness matrix, {D} = displacement vector and

(P} = load vector.

The overall stiffness matrix of the structure, [K], is

composed of different element stiffness matrices. Not all

the member coordinates are compatible with the global

coordinate system. Therefore, the stiffness matrix of each

element has to be multiplied by a rotational matrix. The

relationship between element forces and deformation in the

local coordinates and global coordinates are:

i r j = tTHF}; {dj} = [ T H d e } (5>2)

[Ke ] - [ T l X H T ] ( 5 3)

Where,
I f
{Fg } and {dg } element end forces and deformation. 1

indicates local coordinate system.

The element stiffness matrix, [Ke ], is symmetrical about the

diagonal and its general form is


108

Sll

S21 S22

S31 S32 S33


[Kg ] =
S41 S42 S43 S44

S51 S52 S53 S54

S61 S62 S63 S64


(5.4)

The definition of the parameters in Eq. 5.4 is presented

later in this Chapter. The coordinate system considered in

this analysis is presented in Fig. 5.1.

b - Global Coordinates

End (i)

a - Local Coordinates
End (j)

Figure 5.1 - Coordinate System


109

The rotational matrix, [T], with respect to these

coordinates is given by:

Cos0 0 Sin0 0 0 0

0 1 0 0 0 0

-Sin0 0 Cos 0 0 0 0
[T] =
0 0 0 Cos0 0 Sin0

0 0 0 0 1 0

0 0 0 -Sin0 0 Cos0
(5.5)

where,

0 = the angle between the element X-axis

and the global X-axis.

The stiffness matrix for a member with a plastic hinge

is formed as for a nonprismatic beam element. Numerical

integration is used to form the flexibility matrix of such

an element. The element is discretized into ten

segments. The bending moment and the curvature are

calculated at the end of each segment.

The elements of the flexibility matrix, [F], of a

nonprismatic beam element (Fig. 5.2) are:

T . 10 M. M, 9 M. M.
Fll = + 4 2 ^ + 2 2 — i]
EI1 i=2,2 E I £ i-1,2 EIj (5.6)

10 M. M. 9 M. M.
F21 = ^ [ 4 2 -1 + 2 2 -1 ]
i=2,2 El. i»l,2 EX. (5.7)
a - Nonprismatic Beam Element

1 1 5 ^
b - Unit Moment at End i (Ml)

c - Curvature Due to Unit Moment at End i ( H i )


El

lliUII

b - Unit Moment at End j(M2)

c - Curvature Due to Unit Moment at End i (H ?)


J E l'

Figure 5.2 - Nonprismatic Beam Element


Ill

(5.8)

Where,

Fll = the rotation at end 1, due to unit moment at

the same end.

F21 = the rotation at end 2, due to unit moment at

end 1 .

F22 = the rotation at end 2, due to unit moment at

same end.

FI2 = F21.

EIj = flexural rigidity at location i of the beam,

L =length of the beam element,

1 . 0 0 . 0

0.9 0 . 1

0 . 8 0 . 2

0.7 0.3
0 . 6 0.4
0.5 and M . = 0.5
1
0.4 0 . 6

0.3 0.7
0 . 2 0 . 8

0 . 1 0.9
0 . 0 1 . 0

The inverse of the flexibility matrix, [F], is the beam

element stiffness matrix, [C], with respect to two flexural

coordinates. The elements of the 22, [C], matrix are:

F22 F12
Cll ; Cl 2 =
(Fll F22 - FI2 F21) (Fll F22 - FI2 F21)
112

Fll
C21 = C12;C22 = --------------------- .
(Fll F22 - FI2 F21)

The elements of the member stiffness matrix presented

earlier in Eq. 5.4 , can be established as follows:

GJ GJ GJ
Sll = — ; S41 = ---- ; S44 = — ;
L L L

Cll + 2 C12 + C22 Cll + C12


S22 = ---------- 5 -------- ; S32 = ;
L L

Cll + 2 C12 + C22 C22 + C12


S52 ------------ 5 -------- ? S62 = ?
L L

Cll + C l 2
S33 = Cll; S53 ; S63 = C21;

Cll + 2 C12 + C22 C22 + C12


S55 = ---------- 5 ; S65 = -------------
L L

S6 6 = C22.

Where,

GJ = elastic torsional rigidity if no torsional

plastic hinge is formed,


113

- plastic torsional hinge otherwise

An assemblage of the element stiffness matrix is

presented in Fig. 5.3.

5.2.2 Distribution of Wheel Loads

A special consideration is given to load

partitioning. When the position of a load does not coincide

with a grid node, it will be distributed linearly in the

longitudinal direction without taking into account the

moments that are associated with distribution. The load will

be distributed nonlinearly in the transverse direction, i.e

the moments associated with the apportioning will be as

counted for [51]. This idealization is reasonably accurate

if the bridge is divided in the longitudinal direction into

panels of length not more than 1.5 times the spacing of

longitudinal elements [21]. The method of load idealization

is explained in Fig. 5.4. The loads and the moments

associated with the apportioning are listed below:

P b c d2 P b(S + 2c)

(5.10)

(5.11)

P a c d2 P a(S + 2c)

(5.12)
(C11+2C12+C22) (C11+C12) (C11+2C12+C22) (C22+C12)
,2 L ,2 L

Cll Cll
(C11+C12) (C11+C12)
L L

JG JG
L L

(C11+2C12+C22) (C11+C12) (C11+2C12+C22) (C22+C12)


,2 L ,2 L

C21 C22
(C22+C12) (C22+C12)
L L

Figure 5.3 - Reduced Nonprismatic Element's Stiffness Matrix


115

Pb Pa

a - Location of the Load c - Load Apportioning in g - Load Apportioning

Longitudinal Direction the four corners

Figure 5.4 - Wheel Load Distribution

F a d e 2 = E_b - P
4 L S 2 ' 4 L 1 (5.13)

Where,

a, b, c, and d are defined in Fig. 5.4.

The components of the distributed load, M ^ , P^, M 2 , P 2 /

M 2 , P2 , and P4 , form the external actions on the grid

nodes. These actions will be used to form the overall load

vector, {P }.

5.2.3 Analysis Procedure

The elasto-plastic analysis of the composite bridges

under study was performed using a step by step load

incremental procedure. Initially the bridge was analyzed


116

elastically. The minimum value of the load intensity that

will cause any point on the structure to plastify is then

determined.

The load is incremented so that at each step, one or

more additional plastic hinges will form. The formation of

plastic hinges (torsional or flexural) is determined by

comparing the element loads with the element

resistances. The response of each element in the failure

path is considered to be linear within each load

increment. The load pattern is considered constant which is

the AASHTO Truck configuration (this approach has the

advantage of handling various of truck configurations). For

each load increment, the different member stiffness matrices

are updated to account for the newly formed plastic

hinges. The procedure is repeated until an unacceptable

level of permanent deformation has occurred (one per cent of

the span length).

A computer program has been developed to perform the

elasto-plastic analysis. The program flowchart consists of

the following steps:

1. Form the element stiffness matrix, Fig. 5.3,

2. Form the rotational matrix of each element [T],

3. Transform stiffness matrices using Eq. 5.3,


117

4. Place the element stiffness matrices in the overall

stiffness matrix,

5. Form the load vector and solve the system of

simultaneous equations to get the structure

deformations.

6 . Calculate the member end forces using Eq. 5.2

7. Check the end forces of each element to find the

maximum incremental ratio needed to cause forming of

the first plastic hinge using the following equation:

|AF |
r =
F limit state “ lFoldlI (5.14)
Where,

AF= response due to incremental load;

F q 1(j = cumulative actions from previous steps.

F limitstate = e ^t^ler y i el(* moment, ultimate

torsion or ultimate shear

8 . Increase all the structural element forces using the

following equation :

AF
p _ p ^^
new ' old r (5.15)

Where,

F««,, “ total new cumulative actions,


new '
9. Trace the plastic hinges and form the reduced

stiffness matrix of the individual elements.

10. Repeat steps five to nine until either a singular

matrix or an unacceptable level of permanent

deformation is obtained.

5.2.4 Method Verification and Examples

The developed procedure is demonstrated on three types of

structures. The structures were analyzed and the results

were compared with other well known methods, such as yield

line theory, virtual work and the finite element

method. These examples serve to illustrate the degree of

validity of the grid method.

Example 1 :

A clamped beam is loaded unsymmetrically by a point load

(P) as shown in Fig. 5.5. The ultimate load was calculated

using the statical and the virtual displacement method. The

results were compared with the grid method where a good

agreement has been observed.

Static method:
(1) The first plastic hinge

The first plastic hinge will form where moment takes

its maximum value which is at point A.

Mv
P = — If M = 4820 k.ft then P = 334.7 K.
14.4 y

(2) The second plastic hinge

After the first plastic hinge has been formed the

clamped beam can be modeled as a propped cantilever.

M fi = AP . a . b(L + a)/ 21 2

Mg = 16.8 8P

M c = AP . a (2 - 3a/L + a 3 /L 3 )

M c = 17.28 AP

The second plastic hinge will be formed at point C when:

4820 - 3855.74
AP = --------------- = 55.8 kips
17.28

(3)The third plastic hinge

After the first and the second plastic hinges are formed

the beam can be modelled as a cantilever beam .

Mg = AP x b - 60 DP

hence,
120

AP = 4280 - 937.44 - 3213.12 = 11.16 kips

The total cumulative load is:

11.16 + 55.8 + 334.7 = 401.66 kip

Virtual Displacement Method:

The external work, E . W . , is P A.

The internal work, I.W., is 2(1.5 My 9 + M p )


But, E.W. = I.W.

hence P = 401.67 kip.

Grid Method.

The structure has been analyzed using grid program. The

results at different levels of loading are identical to

those obtained by other methods as shown in Fig. 5.5.

450

375

300

2 225

150
j40' 60'

0.1 0.2

Deflection in ft.
Figure 5.5 - Clamped Beam, Load Vs. Deflection Curve
121

Example 2 :

A concrete slab is simply supported on four sides with

loads and geometrical details as shown in Fig. 5.6

Material and section properties:

1.1
reinforcement ratio p = ---
100
I
concrete strength f = 4000 psi

Reinforcing steel yield strength Fy = 60 ksi

Slab thickness t = 6 inches

Effective depth ofreinforcement d = 4.75 inches

The yielding moment per unit length, Mp = 12.5 k.ft.

Yield line theory:

C 1
w = --------------

Ly ( C 2 + C3) 2

Where,

Wn = maximum failure uniform load per unit

area (in k s f ),

W = total collapse load (kips),

M p X = yielding moment in the X direction

Mpy = yielding moment in the Y direction

Lx = Length of element in the X direction

Ly = Length of element in the Y direction

Cj = 24 My
122

Simply Supported Slab

T
t = 6"
J-

As = .6 2 ”s q ./ft.

Typical Cross Section in The Slab

Figure 5.6 _ Plate Element


Hence,

Wn = 1.6927

Total load on the slab, W = W„ L L


n x y

W = 1.6827 (12) (15) = 304.7 Kips.

Grid method

The plate structure has been divided into sub-elements.

Four cases are considered; very coarse mesh, coarse mesh ,

medium mesh and fine mesh. Fig. 5.7 show a typical case of

grid mesh (coarse mesh). The properties of the structural

elements were calculated using Eqs. 4.1 - 4.6.

The load versus deflection for different mesh size is

presented in Fig. 5.8. The conclusion is, as the grid mesh

gets finer the failure load predicted by the grid method

gets closer to that of the yield line theory. The effect of

the mesh size on the results is presented in Fig. 5.9.

McCarth and Traina [1984] observed that the results obtained

using yield line theory form a lower bound to the results

obtained using fine grid mesh. The same conclusion is

obtained in this example where the fine mesh failure load


124

3.75' 3 .7 5 ’

T3’
© —

t3’

Figure 5.7 - Plate Grid Model

was 308.0 kips, whereas the yield line theory gave a value

of 304.7 kips.

Example 3 ;

In this example a 15 ft. composite steel girder highway

bridge model is considered.

The bridge consists of five longitudinal girders. The

girders are 1.5 ft. apart. The concrete deck slab is 1.75

inch thick. The bridge cross section is shown in Fig. 5.10.

In the analysis the bridge is represented by

longitudinal and transverse elements. Strength properties

of the different elements were calculated using the section

analysis methods (Sec. 4.3). The results obtained by using

the grid method agree well with the ones reported by


125

a.

QS

Ui
oc 1 Fine Mesh
u
z 2 Medium Mesh
M
3 Coarse Mesh
4 Very Coarse Mesh

5 Yield Line Theory

0.25 0.50 0.75 1.00 1.25 1.50


DEFLECTION

Figure 5.8 - Effect of Mesh Size on the Failure Load

*\
%

25 \
\

X \
\
\
o 13
W

»- * » f-« » ►
•1 *2 .3 *4 .5
l/(area of the mesh panel)

Figure 5.9 - Effect of Hesh Size on the Accuracy of the


Results Compared with the Yield Line Theory
a -Transverse Cross Section

1.75“? "
HQO
.188

8.0" — ► ««— .135"

.188^-

T 2.281

b -Details ofCross Section

Figure 5.10 - Details of the Bridge Model


127

i i
o

Yield Line Method

o
P in kips

cvi

o
©
x Finite Element
Load

Method /
o
CO

o
CO

o Grid Method
o
c\i

o
d

0.0 0.2 0.4 0.6 0.8

Deflection in Inches

Figure 5.11 - Load Vs. Deflection (Three Models)


128

Wegmuller [112] using finite element approach. The failure

load obtained using grid program 27.4 kips, versus 28.5

using the finite element method. The simplified yield line

[100] gave a value of 27.0 kips. Fig. 5.11 shows a

comparison between the three results.

Example 4 :

In this example a full scale bridge of 60 ft span is

analyzed. The bridge consists of five girders spaced at

6.75ft. Details of a typical section of the bridge is shown

in Fig. 5.12. The bridge grid model and the properties of

its components were calculated as in example 3. The bridge

is analyzed. The elasto-plastic response of the bridge

system is presented in Fig. 5.12.

Observationsi

The following observations can be made from the analysis

in the preceding four examples:

1. The developed grid analysis method showed a good

agreement with the available experimental data,

2. Comparison between the developed method and other

methods (FEM and Yield Line Theory) reveals that the

method is efficient with respect to computer time and

generality of application.
129

800.00
720.00
640.00 Ultimate Strength

Reserve Strength
560.00
(kips)

480.00

Initial Yielding
THE BRIDGE

400.00
320.00
ON
LOAD

240.00

,# 5 « t

* » at 18'
*S a t s"
LIVE

160.00
TOTAL

PL. to * o .a

I Deflection due to Dead Load


80.00

Permanent Deflection
0.00

— 0— 1----- 1----- 1----- 1----- 1----- h-’--- 1----- 1----- 1----- 1


0.00 0.20 0.40 0.60 0.80 1.00
DEFLECTIO N (ft)

Figure 5.12 - Elasto-Plastic Response for


Composite Steel Girder Bridge
130

3. The strength reserve in the bridge system is

considerable and may be as high as 25 per cent

(Fig. 5.12).

5.3 Ultimate Strength Reliability

5.3.1 Background

The preceding approach was devoted to a deterministic

analysis. The ultimate strength of a bridge system was

obtained while all the material properties and load

parameters were at their specified values. In this section

the uncertainty associated with determining the values of

these parameters will be considered.

Traditionally, in highway bridges, reliability was

checked by checking the reliability of single elements. In

practice, a structure is a system with many interacting

components. Failure of a structural system may involve one

or more failure modes. The structural model should consider

all the relevant failure modes.

From the reliability point of view, the fundamental

models of structural system are series, parallel and

combined systems. A series system (weakest link) fails

whenever any of its components fails. An example of a series

system is a determinant truss as shown in Fig. 5.13. In a

parallel system the failure of one component does not lead

to a total failure. An example of a parallel system is a

statically indeterminate frames. Fig. 5.14. Combined system


131

is a combination of parallel and series. An example is an

indeterminate truss with three panels, two diagonals in each

panel. Figs. 5.15-5.16 show typical examples of combined

system.

Classification of a structural system depends also on

other factors. Some of them are: ductility of the system

components, degree of redundancy, load characteristics and

mode of failure. The term ductility is defined in this

section as the ability of the component to hold its load

after its ultimate strength is reached. A review of the

fundamental system classes is presented in Appendix A.

I p

(a) (b)

Figure 5.13 - Example of a Series System:

a Determinate Truss b) System Reliability Model)


132

(a) (b)

Figure 5.14 - Example of Parallel System:)

a Statically Indeterminate Structure)

b System Reliability Model)

1,2 1,3

2.1

K.1

Figure 5.15 - Reliability Model for


Series System in Parallel
133

— Si,n —
sl.l S1.2 Sl,3

— — — s2,n —
S2,l — — —
S2,3
— *f—
• : : •

— SM — sk,2 sk,3
— Sk.n —

Figure 5.16 - Reliability Model for


Parallel System in Series.

5.3.2 Proposed Approach for Evaluating System Reliability

The approach is based on evaluating the two basic

parameters of the limit state function (Eq. 2.2), the

resistance, R, and the load effect, Q. The statistics of the

load effect has been investigated in Chapter III. Statistics

of the bridge resistance will be determined next.

1). Statistics of System Resistance, R.

The resistance of the structure is a function of several

random variables, such that,

R = f(Xlf X 2 , ....... Xn )
134

where, X^, X 2 ,..etc, are the strength of individual

components (deck slab, girders load configuration and

transverse load location). The statistics of the components'

strengths was calculated as shown in Chapter IV. The AASHTO

HS20 configuration will be considered. The statistics of R

(mean and standard deviation) is to be determined. The usual

approximate formulas [10] are obtained from Taylor expansion

of the function about the expectation of the random

variables. This approach imposes excessive restrictions on

the function (existence and continuity of the first few

derivatives) and requires the computation of derivatives. In

case of bridge system analysis with complex interaction of

different components, it is almost impossible to use Taylor

Expansion. This difficulty has been overcome by using

Rosenblueth's point estimates [101]. Rosenblueth developed

his method (2n point estimate) for a function with

independent random variables. This method has been extended

in this study to handle the case of dependent random

variables. This modification together with Rosenblueth point

estimate have been explained in Chapter II.

Demonstration of the procedure

Let the system shown in Fig. 5.17 represent a bridge

deck with five girders. The strength of the components are:

Xj, X2, ... Xg.


135

X, Xj X, x4 ^

Figure 5.17 - Typical Cross Section in


Composite Steel Girder Bridge

Let the covariance of the X.'s be £v (Eq. 2.42). The


j A
VVi
subscript, j, is used to refer to the j component.

The histogram representing the transverse location of

the truck (Fig. 3.9) is discretized into six intervals. The

probability of the truck to be in each location, P^, is

calculated from the histogram. The subscript, i, is used to

refer to the i ^ location of the truck.

The general theory of the procedure is explained in

Sec. 2.3.3. The application is explained in the following

steps:

1. Compute the bridge resistance, Y^, with all Xj's at

their mean values.


Decompose the covariance matrix to transfer the

correlated random variables to a new set of

independent random variables, Z 2 ,....Zg.

The mean values of the independent variables, {Z},


m
are obtained as, {Z} = [A] .{X} (see Section

2.3.3). The variances of the Z^ are equal to the

eigenvalues of the matrix A.

Increase Z. with one standard deviation, o„, other


J
variables are kept at their mean values. Compute the

equivalent set of Xj's where,

{X} = [A]{Z }

Compute the resistance of the bridge system using the

nonlinear analysis program,

Repeat step 3 with Zj decreased by one standard

deviation.

Repeat step 4 to obtain Y T j .

Calculate the average value of the resistance

corresponds to j*"*1 girder such that:


137

(5.17)

9. Calculate the average value of the bridge resistance,

R'-j, for the truck being in location i.

Yij
y. n —
(5.18)

-i + n (i+v?.)
j=n ^ (5.19)

where,

n = number of components.

10. Repeat steps 4 through 9 for other truck locations.

11. The overall resistance of the bridge is calculated

as:

(5.20)

where,

k = number of discreated intervals of the

transverse truck locations

2). Statistics of Load Effects.

The different components of load effects have been

studied in Chapter III. The mean and standard deviation of

load effect has been established for different spans. The


138

distribution of the load effect is considered as Lognormal

[30].

3). Reliability Analysis.

Once, the two variables of the limit state function

have been evaluated, then any of reliability index,/?, models

(normal model, Eq. 2.13, or the lognormal model, Eq. 2.22)

can be used to determine the reliability index, /?.

The obtained reliability index, /?, gives an estimate to

the reliability of a bridge system even if the distribution

of the load and resistance are neither normal nor lognormal

[30].

5.3.3 Proposed Approach for Bridge Capacity Rating

Bridges that have questionable load carrying capacity

are checked by bridge capacity rating. Rating is performed

by a combination of field inspection and analytical study as

guided by AASHTO [2,3]. The application of these guidelines

will be illustrated using the reliability based analysis.

There are two levels of bridge capacity rating [3,114]:

1). The inventory rating which is defined as the load

that produces a stress in the critical bridge

element of 0.55 times the yield stress or the

allowable stress used in design.

2). The operating rating which is defined as the maximum

load that should be allowed on a bridge under any


139

circumstances and should not exceed 0.75 times the

yield stress.

Lathia [57] classified the rating into three levels; in

addition to the two levels mentioned above the third level

which he introduced was, the " Safe Load Capacity", defined

as the actual load that can be carried safely by the

structure on a long-term basis under actual traffic

conditions.

The rating factors defined by AASHTO [3] are:

inv
(5.21)

°-75 Mu - “p
RF
opr
(5.22)

where

RF. inventory rating factor


mv

RFQp r = operating rating factor,

Mu = ultimate moment capacity of girder

= moment due to dead load effect

M L+I = moment created by vehicle load + impact

The AASHTO operating rating procedure addresses the

capacity of an individual girder. In this approach the load

which will cause the first plastic hinge to be formed in the


140

bridge system, Py, will be considered instead of the

ultimate moment capacity of an individual girder. The limit

state function can be formed to reflect the requirements

that the stress does not exceed the yield stress everywhere

in the structure. Using the AASHTO operating rating factor,

Eq. 5.22, gives:

where

Qr - Q d ♦ RF Q l i

dd = dead load effect,

Qli = effect of live load plus impact load

The values of the mean and standard deviation of Py are

determined in a similar way to that mentioned in Sec. 5.3.2.

The mean value of the dead load can be obtained from

actual measurement of the bridge under consideration. The

coefficient of variation of the dead load is recommended to

be = .06 [30].

The truck load model mentioned in Eq. 3.2 is considered

to represent the five year period. The mean and standard

deviation is obtained by using Monte Carlo simulation. The

input data for this model is presented in table 3.3.

The mean value of the limit state function can be

written as:
141

g = py - Qd - RF Qli (5.24)

and its standard deviation is

The reliability index is given by:

9
0 = _
<rg (5.25)

Hence the rating Factor, RF, can be defined in terms of the

reliability index, 0. For a predefined target reliability


•fa
index, 0T , the rating Factor, RF , can be defined as:

* A - / A 2 -4 B
RF
2 (5.26)

where

A
2 Qli(*v - Qp}
c (5.27)

(5.28)

C (5.29)

Demonstration of the procedure on an actual bridge and a

comparison between the proposed approach and AASHTO approach

is presented in Chapter VI.


CHAPTER VI

APPLICATION TO BRIDGE SYSTEM EVALUATION

6.1 Evaluation using the Developed Approach

The following examples are chosen to demonstrate the

proposed approach

6.1.1 Reinforced Concrete Bridge

A reinforced concrete T-beam bridge with the cross

section shown in Fig. 6.1 is considered. The bridge was

built in New Zealand in 1937 and tested to failure in 1977

after being in service for 40 years. It was chosen for this

example to compare the actual failure load with the one

predicted using the grid program. The available material

test data is used in resistance models. Information on some

details (stirrups, compression and temperature steel,

correlation between strengths of the beams and level of

deterioration) is not available so typical values were

assumed. Material and dimensional parameters used in the

analysis are:

span = 32.2 ft.

beams' spacing = 6.67 ft.


n n
beams' section = 2 8 x 14 (as shown in Fig. 6.1-b.)

142
a - General View

. 24 ’ ►

isjsiijjjijinnnfir I
r “

6-8 6-8 6-8


+ +
b - Transverse Cross Section
80"
4 --------------------------

3 *1 0 _________

8"

2*6
* 4 @ 5"
28"

8*10

14"

c - Section Details

Figure 6.1 - Reinforced Concrete Bridge


144

slab thickness = 8

f'c = 7.25 ksi

fy = 45.4 ksi (yield strength of steel

reinforcement)

f =74.7 k s i .(ultimate strength of steel

reinforcement)

Properties of Grid Elements

The bridge super-structure is divided into grid

elements as shown in Fig. 6.2.

6-8

6 » S.4 - 3 2 . 2 4

Figure 6.2 - Bridge Grid Model

Strength of each element is computed using section

analysis program. The results are presented in Table 6.1.

The bridge failure load is computed using the grid

method. The predicted results is compared with the

experimental results. The observed ultimate failure load was


145

Table 6.1 - Strengths of Grid Components

Longitudinal Transverse
Variable components components
Mean COV (%) Mean COV (%)

Flexural rigidity'*’ 1124500. 10 14379. 10


(k-ft)

Torsional rigidity 57201. 32093.


(k-ft)

Ultimate torsion 99.7 46.6


(k-ft)

Yield moment 1257.6 10 72.9 10


(k-ft)

Ultimate m o m e n t ' * * 1375.6 81.


(k-ft)

Ultimate shear _ 208. 118.


(kips)

<*>values are derived in terms of yield moment

330.5 kips compared with 347 kips obtained by using grid

analogy method (within 6%).

The reliability of the tested bridge was evaluated. The

calculation was carried out using the nonlinear analysis and

the simulation methods. Three cases were investigated:

1). No correlation between strengths of girders (p = 0),

2). Partial correlation (p = .5), and

3). Perfect correlation (p = 1).


146

The curb distance histogram was divided into six

intervals. The truck location being in any of these

intervals was considered, Fig. 6.3.

6- 8' 68- ' 6-8r

Figure 6.3 - Transverse Location of the Truck

The computations and the results for the first case (no

correlation) are explained in the following:

The procedure follows the one explained in Sec. 5.3.2.

Numerical application is presented next with reference to

table 6.2. The mean resistance corresponding to truck

location 1 is calculated from Eq. 5.18 as:

836.5 836.5 836.0 834.0


Rx = 838 ( ) ( ) ( ) ( ) = 829.0
838 838 838 838
°Yij
Probability of Mean strength Y_ ♦ y +
Truck Truck Presence Girder Ho. Y* ij Yi
‘l " Yi
+i
location'*) ij 2 2 Yij
(%) (kips)
< v (kips) (kips) (%)
< V
(i) (j) (kips) (kips) (o_ )
< V Vij
ij

1 14.6 838 1 798 875 836.5 38.5 4.6


2 812 861 836.5 24.5 3.0
3 834 838 836.0 02.0 0.2
4 834 834 834.0 00.0 0.0

2 33.8 874 1 831 913 872.0 41.0 4.7


2 849 896 872.5 23.5 2.7
3 871 875 873.0 2.00 0.0
4 874 874 874.0 0.00 0.0

3 16.0 946 1 903 989 946.0 43.0 4.5


2 917 974 945.5 28.5 3.0
3 942 948 945.0 3.00 0.3
4 946 946 946.0 0.00 0.0

« 6.7 946 1 946 946 946.0 0.00 0.0


2 942 948 945.0 3.00 0.3
3 917 974 945.5 28.5 3.0
4 903 988 946.5 43.0 4.5

S 21.S 874 1 874 874 874.0 00.0 0.0


2 871 876 873.5 2.50 0.3
3 849 896 872.5 23.5 2.7
4 831 914 872.5 41.5 4.7

6 7.4 838 1 839 839 839.0 00.0 0.0


2 837 840 838.5 00.0 0.0
3 813 861 837.0 24.0 2.9
4 799 875 837.0 38.0 4.5

<*> The symbols used in this table are explained in Sec. 5.3.2
See Fig. 6.4

Table 6.?. - Results of Points Estimates Approximation


148

Similarly Rg* ... Rg are: 869.5, 944.5, 944.5, 869.5 and

829.0, respectively. The coefficient of variation, V„ for


i
each location are calculated using Eq. 5.19. For example,

VD for location one is given by:


K1

VH1 = t/(l+(.046)2 )(l+(.03)2 )(l+(.002)2 ) -1 = 0.055

The standard deviation of R^ is:

aR1 = VR1 = (.055)(829) = 45.6 kips.

Similar calculations were performed for other truck

locations.

The mean value of the overall bridge resistance is

calculated from Eq. 5.20 as:

6
R = I p. R. =
i=l 1 1

14.6(829)+33.8(869)+16(944)+6.7(944)+21.5(869)+7.4(829) _
100

= 877.6 kips

The standard deviation of the resistance is calculated as:

„ _ 14.6(45.6)+33.8(47)+16.0(51)+6.7(51)+21.5(47)+7.4(45.6)
R ~ 100
149

= 47.7 kips

The mean and standard deviation of the total load effect

is calculated using Monte Carlo simulation and Eg. 3.9.

Q = 355 kips and Oq = 36 kips.

The bridge system reliability index, B , were calculated


s
using Eq. 2.25,

877.6
ln-
355
= = 7.9,
47.7 , 36 ,
Vi - — -r + (— r
877.6 355

The reliability index for the individual girder was

calculated using the procedure presented in Chapter IV. The

means and standard deviations of the resistance and the load

were found to be:

R g = 1058 k-ft.; <rR = 128 k-ft.

Q e = 366 k-ft.; Oq = 59 k-ft.

hence, using Eq. 2.13 the element reliability index is,

Re - Qe
= 4.9
,2 R + °Q
2
150

6.1.2 Steel Girder Bridge

The considered bridge section is shown in Fig. 6.4 . The

main parameters of the bridge are:

| * --------------------------- 3 2 '
_ p - 18" 18

I I I I I

| * ----------- 4 x 6 .7 5 '

Figure 6.4 - Composite Steel Girder Bridge

1. Span is 60 ft center to center of bearings,

2. Structural slab thickness is 7.25 in.

3. 5 girders spaced at 6.75 ft.

4. HS20 loading and no sidewalk live load.


151

5. Structural steel with a nominal yield strength equal

to 36 ksi

6. Diaphragms are C15 x 33.9 spaced at 15 ft.

Monte Carlo technique was used to generate the strength

of each element as mentioned in Sec. 4.3. The flexural

rigidity, El, was derived from the yielding moment, My,

using a linear regression analysis. The major parameter

which controls the bridge response is the yielding

moment. The effect of other strength parameters was found to

be less important with respect to the overall bridge

strength. Therefore, in the analysis, other parameters were

assumed to be equal to their mean values.

The reliability indices were calculated for individual

girders and the whole bridge structure. The results are

presented in Tables 6.3.

6.1.3 Prestressed Concrete Girder Bridge

In this example a 90 ft. prestressed concrete bridge is

considered. The structure was designed according to AASHTO

[2] code provisions. Details of the bridge are given in

Fig. 6.5. Main parameters are:

1. Span is 90 ft center to center of bearings

2. Structural slab thickness is 9 in.

3. Four girders are spaced at 8 ft.

4. HS20 loading and no sidewalk live load.

5. Properties of prestressing steel are summarized in


152

1 28 1

C h C h C h

8 ---- [ - 8 — I— 8 -j— 8 +

b - 96'

#4@18'

#5@5'

23 #5@14'

#5@6

Figure 6.5 - Prestressed Concrete Girder Bridge

Table 4.1.

The results of the analysis are presented in Table 6.3.

The effect of correlation was considered. The results of

the analysis for different levels of correlation are

presented in Table 6.4.


153

Table 6.3 - Reliability Analysis without correlation

Resistance Load Reliability


(kips) (kips) Index, 0,
Bridge
Type Element System
Mean Standard Mean Standard
Deviation Deviation
'V (*s>

Reinforced 878 48 355 36 4.816 7.9


Concrete

Composite 1047 53 537 51 3.78 6.93


Steel

Prestressed 1557 41 1007 79 3.42 5.26


Concrete

Table 6.4 - Reliability Analysis with Correlation.

System Reliability Index,


0S '
Bridge Type
o
I—1

p=0.5 p=0.0
Q.
II
.

Reinforced 7.12 7.48 7.9


Concrete

Composite 6.04 6.32 6.93


Steel

Prestressed 5.0 5.14 5.26


Concrete

The results from these examples revealed that, the

correlation between girders has no significant effect on the

failure load and hence the reliability index.


154

6.2 Bridge Capacity Rating

The composite steel girder bridge in Sec. 6.1.2 will be

used to demonstrate the bridge capacity rating approach. The

bridge will be rated for operating rating using the AASHTO

approach and the proposed approach. The AASHTO truck, HS20

configuration is used as the rated vehicle. The data are:

Bridge span = 60 ft,

Maximum moment due to AASHTO truck HS20 = 806 k-ft,

Ultimate moment for individual girder = 2200 k-ft,

Moment due to dead load = 450 k-ft per girder,

Mean value of the total bridge capacity at the formation of

the first plastic hinge, P y = 848 kips,

The coefficient of variation of P y = 3.5%

The values of the mean and coefficient of variation of the

live load plus impact load using Eq. 3.2 and Monte Carlo

Simulation = 109 kips and .091 respectively.

The values of the mean and coefficient of variation of the

dead load are = 348 kips and .06 respectively

Rating (refer to Sec. 5.3.3):

1). AASHTO approach:

.75 (2200) - 450


RF = ------------------
^(1.27)(|^)

= 1.91
155

2). Proposed approach;

-IT * •
If the target reliability index is selected as 0 = 3.5.

Then by Eqs. 5.26-5.29:

C = (109)2 - (3.5)2 (12)2 = 10117

2 (109) ( 848 - 348)


A = -------------------- = 10.774
10117

(500)2 - (3.5)2 (602 + 242)


B = = 19.65
10117

* 10.774 - t/(10.774)2 - 4(19.65)


RF = = 2.33

*
Clearly the rating factor, RF , depends on the target

reliability index. If the target reliability index is taken


T *
as, 0 = 5 then a similar calculation gives RF = 1.51.

Comparison and Discussion

From this example the following points can be observed:

1). In the AASHTO approach only the capacity of the

individual girder and the distribution factor are

considered.

2). In the proposed approach the bridge is considered as

a system. In other words, if one girder is


156

deteriorated this does not mean that the whole

system is deteriorated.

3). The AASHTO approach looks easier to apply since it

is only required to obtain the ultimate moment

capacity of a single girder. On the other hand, the

proposed approach requires some computational work

involving grid analysis and simulation but this has

to be done only once. Then the bridge can be rated

for any vehicle in terms of an equivalent AASHTO

truck.

4). In the proposed approach there is no need to use the

AASHTO distribution factor since only the equivalent

AASHTO truck weight is required.

5). Incorporating the reliability index into the rating

factor provides the flexibility to assign different

levels of safety to each bridge depending on its

condition, importance, location etc.


CHAPTER VII

CONCLUSIONS AND RECOMMENDATIONS

7.1 Research Results

This study was devoted to the following main topics:

1. Development of a nonlinear analysis procedure

using the the grid analogy to predict the elasto-

plastic response including the failure load of a

bridge system.

2. Development of a practical procedure for the

evaluation of bridges based on the reliability

theory.

3. Development of a practical reliability-based

rating method for existing bridges.

The developed approach has been illustrated by

numerical examples. Special computer programs were developed

for nonlinear section analysis, nonlinear grid system

analysis, simulation, load data analysis and modified yield

line theory.

Three types of cross sections were analyzed: reinforced

concrete, prestressed concrete and composite steel.

157
158

Uncertainties in material properties and section geometry

were considered and the relevant statistical data were taken

from the available literature.

The results of the composite steel section analysis (250

sections were analyzed) indicate a reserve strength beyond

the yield moment of about 15% of the yield moment

(Chapter IV). This amount of reserve strength is not

considered by the AASHTO specifications. The reserve

strength in bridge system was also investigated

(Chapter V). The strength reserve beyond the initial

yielding was as high as 25 per cent (Fig. 5.12). AASHTO

specifications consider the maximum allowable stress in

operating rating as 75 per cent of the initial yield

strength of the individual component. This indicates that

about 40 per cent of the ultimate bridge strength is not

used.

The torsional behavior of the composite steel section

was studied. An approximate approach is suggested to predict

the torsion versus twist relationship. The results indicate

that torsional rigidity drops rapidly if the torsional

stresses in the slab reach their ultimate values Fig. 4.9.

Analysis of prestressed concrete sections indicates that

the reserve strength beyond the yield strength is almost

negligible.
159

Grid analogy has been utilized successfully to predict

elastic response of bridge systems, space trusses and plate

elements. Recently, it has been utilized to predict the

elasto-plastic response of a simply supported slab. In this

analysis the grid analogy is extended to predict the

nonlinear response of a bridge system.

The proposed approach was demonstrated on different

types of structures by numerical examples presented in

Chapter V-VI. The results showed a good agreement with the

experimental or theoretical methods (FEM).

An actual reinforced concrete T-beam bridge was analyzed

and the results were compared with the test results. The

error was within 6% (see Chapter VI). The bridge resistance

was very large although the structure was built 50 years

ago. This is due to the high value of concrete strength

(7.25 ksi, compared to 3-4 ksi usually used in bridge

decks). The statistics of the bridge resistance were

determined using Rosenblueth point estimate technique. The

average value of resistance was found to be nine times the

AASHTO HS20 truck weight. This indicates that there is a

considerable amount of the reserve strength even in bridges

which almost exhausted their design lifetime.

The grid nonlinear analysis was utilized to check the

reliability of bridge systems. The simplified point estimate

method was used to compute the mean and standard deviation

of the bridge resistance. The reliability analysis was also


160

performed for the individual girders (Table 6.4). The

results show a high reliability index which means a very low

probability of failure (Eq. 2.12).

7.2 Summary of the Conclusions

The main conclusions of this study are:

1. Grid analogy can be used successfully to predict the

nonlinear response of a bridge system. In case of

symmetrical loading the error is within six per cent

(Chapter V, Example 3).

2. The amount of reserve strength in the bridges

analyzed in Chapter V (Example 4) and Chapter VI is

approximately 25 per cent of the ultimate load. This

result may be representative of highway bridges with

four or five longitudinal girders.

3. AASHTO specifications consider the maximum allowable

stress in operating rating as 75 per cent of the

initial yield strength of the individual

component. Thus about 40 per cent of the ultimate

bridge strength is not considered.

4. Torsional rigidity drops rapidly if the torsional

stresses in the slab reach their ultimate values.

5. In case of complex structures where the actual

statistics of strength other than the mean and the

variance are not available, Rosenblueth's point


estimate methods can be used to perform the

reliability analysis.

Correlation between the strength of the main girders

does not have much effect on the overall system

reliability (Table 6.4).

Assuming perfect correlation between strength of main

girders (coefficient of correlation, p = 1) yields

conservative results, i.e low reliability index is

obtained (Table 6.3). It also saves a considerable

amount of computing time. The difference in

reliability index, 0, between the case of full

correlation (p=l) and the case of no correlation (p =

0) is within ten per cent (Table 6.3). This amount

of variation in reliability index is common in most

practical cases. As an example, the variation in

reliability index for hot rolled steel columns was

found to be within 15 per cent (Madsen, Krenk and

Lind 1986).

7.3 Recommendations

For more understanding of the torsion versus twist

relationship for composite steel section more work is

required.

Interaction between moment, shear and torsion needs

further investigation.
162

3. Serviceability limit states often govern the design

and evaluation of bridges. Further developments

should combine the ultimate and serviceability

criteria.

4. Rosenblueth's point estimate method needs to be

modified and generalized for using in different types

of bridges.

5. A model to relate the element reliability index to

the system reliability index in case of composite

steel girder bridges is required.


APPENDICES
164

APPENDIX A

SECOND MOMENT RELIABILITY INDEX

A.1 Cornel Reliability Index

In the original formulation of the failure function by

Cornell [see Madsen, Krenk, and Lind 1986] , the failure

function was written as the difference between the

resistance and the corresponding load effect. The

corresponding safety margin is

M = R - Q (A.1)

If R and Q are uncorrelated, Cornell's reliability index can

be explained as in Eq. 2.13.

If the failure surface is a hyperplane, it is possible to

define a linear failure function, g(Z^), [ Madsen et

al. 1986]

n
M = a^ + Z a.z.
° i=l 1 1 (A.2)

The corresponding safety margin in matrix notation can be

written as

M = ao + aTZ (A>3)

where

T
a = row vector with elements a^, and
165

Z = column vector with elements z..

The Cornell's reliability index can be explained as

a + aT E[Z]
B = — — :-------------
Hc _______
/aT c_ a (A.4)
z
Where, E[Z] and C_ are the expected value and covariance
z
matrices of Z, respectively.

A.2 Determination of the Design Point


Using Rackwitz and Fessler Algorithm

Rackwitz and Fessler algorithm is an iteration method to

find the so called design point. The design point is the

point of maximum probability on the failure boundary (limit

state function). The method is based on the principle of

normal tail approximation and it consists in changing the

distribution function of each basic variable to a substitute

normal distribution function. For simplicity of the

presentation , the method will be demonstrated for the case

of two variables only.

Let R and Q represent the structural resistance and the

load effect respectively. The mathematical representation of

the failure boundary is the limit state function given by

Eq. 2.3. The design point, denoted by (R*, Q*), is located

on the failure boundary. Hence, R* = Q*, as shown in

Fig. A.I.

Let FR be the cumulative distribution function (CDF) and

fR the probability density function (PDF) for R. Similarly


166

0 .9 9

0.9

0.841 ----- 1

0.7

0.5
Q =FlR

0-3

0.1

Figure A.l - Graphical Rackwitz and Fiessler Algorithim


167

F q and fg are the CDF and PDF for Q. First, initial value of

the design point R* is guessed first. Next FR is

approximated by a normal distribution, FR ', such that:

fr '<8*) - rR <R*> (A.B)

fR '(R ) = £r (R ) (A_6)

I
The standard deviation of R is

0[$-1 (Fr (R*))]


(Tp ^
fr (R ) (A.7)

where,

4>() = PDF of the standard normal random variable and

$() = CDF of the standard normal random variable.

The mean of R is,

r ' = R* - «r '*-1 (Pr (r ‘ )) (a8

similarly, Fg is approximated by a normal distribution F g ' ,

such that, F q '(Q*) = F q (Q*) and fg'(Q*) = fg(Q*). In this


* * •
case Q = R . The mean and standard deviation of Q are

Q' = Q* - O g ’fc'^FgfQ*))
(A.9)
168

* 0 -1(Fq (Q*))]

fQ (Q*) (A.10)

The reliability index is,

-* -*
R - Q
IJ = ------------ :--------------

Next a new design point can be calculated:

* _» o
R - R - a*'-

^ R ' + °Q (A.12)

* _
- '• 2
Q = nQ - n * ' -

+ a .
(A.13)

The second iteration starts with normalizing the variables

at the design point calculated from Eqs. 2.37-2.38. The

reliability index is calculated and the procedure is

repeated until R and Q do not change in consecutive

iterations. This method is not guaranteed to converge but it

has performed extremely well in a wide variety of practical


169

problems that have been solved in structural reliability

analysis [63].

A convenient graphical version of this procedure has

been given by Nowak and Regupathy [84]. The iterations are

done after plotting the distribution function of R and Q on

a normal probability paper (see Fig. A.l).


170

APPENDIX B

GENERATION OF CORRELATED RANDOM VARIABLES

Generation of dependent and independent random variables

is discussed briefly in Chapter II. In this appendix, some

details are presented on techniques for generating

correlated normal, lognormal, and jointly distributed normal

and lognormal random variables are discussed.

B.1 Transformation to Uncorrelated Space

One approach to the generation of correlated random

variables is to use the so called linear transformation in

which the correlated variables are transformed into

uncorrelated (but not necessarily independent)

variables. There are two linear transformations [33]: (a)

orthogonal transformation, and (b) affine transformation.

a. Orthogonal transformation. This linear transformation

was discussed early in Chapter II.

b. Affine transformation. If the covariance matrix £x is

symmetric and positive definite, it can always be

decomposed into a lower triangular matrix [A], such

that,

eX = [A] [A]T (B.l)


171

£y = U ] T [£x ] [A]
(B.2)

X = [a ]t y (B. 3)

Thus linear transformation Eq. 3 is obtained by finding

[A] in Eq. 1. This can be done by using Cholesky

decomposition.

B.2 Generation of Correlated Lognormal Variables

Monte Carlo methods or the point estimates approximations

can be used to generate correlated normal variables using

linear transformations. In case of correlated lognormal

variables, there is a necessity to transfer these variables

into normal variables first, and then use the linear

transformations to decouple them. The major difficulty is

to derive the relationship between the correlated lognormal

variables and the corresponding normal variables. In Chapter

II, this problem was solved assuming that the correlation

between two lognormal variables does not change after

transferring the variables into normals. This assumption is

practically accurate if the coefficients of variation of the

lognormal variables are less than 25 per cent. If the

coefficients of variation are larger than 25 per cent then

the following expression, which was derived by Gorman [33]

can be used:

plnZ1 lnZ2 (alnZ1 alnZ2 ) = l n ( 1 + P


(B.4)

where,
172

V_ , V„ are the coefficient of variation of the lognormal


Z1 2
variables, and Z 2 , respectively.

B.3 Generation of Jointly Distributed


Normal and Lognormal Variables

Generation of jointly distributed normal and lognormal

variables can be straightforward if the relationship between

the correlated lognormal variables and the corresponding

normal variables is available.

Let X be lognormally distributed with mean and variance,


2 .
ux , ox . Then lnX is normally distributed. Let W = lnX i.e. W
2
is normal N (n1^ , 0^ ) . Let Y be normally distributed N (my,
2
a y). Let the correlation coefficient between W and Y be p

i.e. p(W,Y) = p(lnX,Y) = p. An expression to define the

covariance of W and Y is derived by Gorman [33] such that:

Cov (X,Y)
C o v (ln X ,Y ) « ----------
Px (B .5)

and,

p = VX PY>X
p lnX,Y ___________

>/ln(1 + VX> (B.6)


173

APPENDIX C

RELIABILITY OF STRUCTURAL SYSTEMS

C.1 Series System

For a series system to survive, all possible failure

modes must survive. Therefore, the probability of failure

of an n failure mode system is [Garson, 1980]

Pp = l-p(S1 nS2 n ...nSn ) (C.l)

i V*
where denotes the survival event for the.i mode, and

p( ) is the probability of the events in parentheses. Eq.

C.l can be expressed as

Pp — 1—p(S^/S2n...Sj,j)p(S2/S2fi...nS^) ... p(S^) 2j

In general,

n-1
Pp = 1- n p( S . / S i+1n . ..nSn p(Sn>
(C. 3)

If it is now assumed that the survival events, S^'s,

are mutually independent, then Eq. C.3 takes the form


174

n
p_ « l- n p(s.)
* i=l 1 (C.4)

If the survival events are perfectly dependent, then

the probability of failure of the system is

Pp = 1 - min p ( S i ) i = l,n

The probabilities of failure given by Eqs. C.4 and C.5

are usually called the upper and lower simple bounds for

series systems, respectively [Thoft-Chrestensen, 1982],

C.2 Parallel Systems

In parallel systems, it is important to distinguish

between systems with ductile elements and systems with

brittle elements. An element is said to be ductile if it

maintains its load carrying capacity level after failure,

and brittle if it becomes ineffective after failure. This

criterion of brittle elements complicated the reliability

analysis of such systems. In some cases (e.g., structures

with a low degree of statical indeterminacy), the brittle

failure of one element will usually result in the failure of

other elements. If this is the case, the system behavior is

like a series system.


175

For a perfectly ductile parallel system to fail, all

modes must fail. Therefore, the probability of failure of

an n mode system is

PF = P(Fi nF2n * • •nFV (C. 6)

where F^ denotes the failure event for the ith mode. Eq.

A . 6 can be expressed as

Pp = p(F1/F2 n ...nFn )p(F2/ F 3n . ..nFfi) ... p(Fn ) (C.7)

In general,

n-1
PF iJ1 P(Fi/Fi+in-“ nFn) P(V
(C. 8)

If it is assumed that the failure events, F^'s, are

mutually independent, then Eq. A . 8 takes the form

n
Pp = II p ( F . )
* i=l 1 (C. 9)

If the failure events are perfectly dependent, then the

probability of failure of the system is

Pp = min p ( F i ) i = l,n
(C.10)
176

Eqs. C.9 and C.10 give the lower and upper simple bounds for

the parallel system with ductile elements,

respectively.

C.3 Complex Systems

Complex systems are combinations of series and parallel

systems. The interaction of failure modes and the

definition of failure of the structural system determine its

failure path and the type of combination of simple systems

within the structure.

A complex system in which a series of one of each of n

components is required for operation, and where there are K

of such identical series is shown in Fig. 5.15. If the

components within a series are assumed to be independent,


+•Y\
the reliability of the i series is given by

n
ps = n p(s. .) , i=i t • • • tK
i j=l l '3 (C.ll)

4-h
where p(S. .)“probability of survival of the j component
j
+•h
of the i series. Thus the system reliability is

k
P( i - n [i-P_ ]
S
i=l Si
177

k n
i - n 1 - n p(s. •)
i=l j-l ' 3 . (C.12)

assuming that the k series are independent.

Next consider a system which has n subsystems of k

parallel components in series as shown in Fig. 5.14. The


+* V i
j parallel subsystem will have a reliability of

1 - n [1 - p(S. .)] , j=l,...,n


i=l (C.13)

^ V*
assuming that the k components of the j subsystem are

independent.

If it is assumed that the j subsystems are independent,

then the reliability of the system is

n
ps ~ n Pg
s j-i sj

n
n i - n (i -p(s. .)}
i=l
(C.14)
BIBLIOGRAPHY

A l - Z a i d r R . , "Fatigue Reliability of Prestressed


Concrete Girder Bridges," Ph. D. Thesis Submitted to
the Department of Civil Engineering, University of
Michigan 1986.

American Association of State Highway and


Transportation Officials, AASHTO, "Standard
Specifications for Highway Bridges," Washington, D. C
1983.

American Association of State Highway and


Transportation Officials, AASHTO, "Manual for
Maintenance Inspection of Bridges," Washington, D. C.
1983.

Ang., A. H. and Amin, M . , "Reliability of Structures


and Structural Systems," Journal of-the Engineering
Mechanics D i v ision. ASCE, Vol.Ill, No. EM2, Proc.
April 1968, pp. 671-691.

Augusti' G., Baratta, A. and Casciati, F . ,


"Probabilistic Methods in Structural Engineering. "
Chapman and Hall, London, 1984.

Ayyub, B. M. and Haidar, A., "Practical Structural


Reliability Techniques," Journal of Structural
Engineering. Vol. 110, No". 8, August 1984, Paper No.
19062, pp. 1707-1724.

Bakht, B. and Jaeger, L. G., "Effect of Vehicle


Eccentricity on Longitudinal Moments in Bridges,"
Canadian Journal of Civil Engineering. No. 10, 1983,
pp. 582-599.

Bakht, B. and Jaeger, L. G., "Bridge Analysis


Simplified." McGraw-Hill Book Company, New York, 1985

Bares, R. and Massonnet, C . , "Analysis of Beam Grids


and Orthotropic Plates." Crospy Lockwood and Son Ltd.
London, 1968.
179

10. Benjamin, J. R. and Cornell, C. A. "Probability.


Statistics, and Decision for Civil Engineers" , Mcgraw
Hill book Company, 1970.

11. Bennett, R. M. , Ang, A. H-S. and Goodpasture, D. W . ,


"Probabilistic Safety Assessment of Redundant Bridges,"
Proceeding of the 4th International Conference on
Structural Safety and Reliability. Kope, 1985, Vol. 3,
pp. 205-211.

12. Buckle, I. G . , Dickson, A. R. and Philips, M. H . ,


"Ultimate Strength of Three Reinforced Concrete Highway
Bridges," Canadian Journal of Civil E ngineering. Vol.
12, 1985, pp. 63-72.

13. Burdette, E. G., and Goodpasture, D. W . , "Tests of Four


Highway Bridges to Failure", Journal of the Structural
Division. ASCE, Vol. 99, No. ST3, Proc. Paper 9627,
Mar., 1973, pp. 334-348.

14. Canadian Standards Association, "Standards for the


Design of Cold-Formed Steel Members in Buildings", CSA
S - 1 3 6 , 1974.

15. Chou, K. C . , McIntosh, C. and Corotis, R. B.,


"Observations on Structural System Reliability and the
role of Model Correlations," Journal of Structural
S a f e t y . Vol. 1, 1982, pp. 189-198.

16. Collins, M. P. and Mitchell, D., "Evaluating Existing


Bridge Strictures Using the Modified Compression Field
Theory," American Concrete Institute, SP-88-6, 1986,
pp. 109-141.

17. Committee on Loads and Forces on Bridges of the


Committee on Bridges of the Structural Division,
"Recommended Design Loads for Bridges," Journal of the
Structural D i v i s i o n . ASCE , Vol. 107, No. ST 7,
December 1981.

18. Cornell,C. A., "Some Thoughts on "Maximum Probable


Loads" and "structural Safety Insurance',' Memorandum,
Department of Civil Engineering, Massachusetts
Institute of Technology, to Members of ASCE Structural
Safety Committee, March 1967 Codes,

19. Csagoly, P. F. and Dorton, R . , A., "Proposed Ontario


Bridge Design Load," Ministry of Transportation and
Communications, MTC, Downsview, Ontario, Report RR-186,
1973.

20. Csagoly, P. F. and Knobel, Z . , "The 1979 Survey of


Commercial Vehicle Weights in Ontario," Ministry of
180

Transportation and Communications, MTC, Downsview,


Ontario, Report RR-230, 1981.

21. Cusens, A. R. and Pama, R. P., "Bridge Deck A n a l y s i s . "


John Wiley and Sons, London, 1975.

22 . Ditlevsen, 0., "Generalized Second Moment Reliability


Index," Journal of Structural Mechanics, Vol. 7, 1979,
pp. 435-451.

23. Ditlevsen, 0., "Principle of Normal Tail


Approximation," Journal of the Engineering Mechanics
Di v i s i o n . Proceedings of the American Society of Civil
Engineers, ASCE, Vol. 107, No., EM6, Dec. 1981,
pp. 1191-1205.

24. Ellingwood, B., Galambos, T. D., MacGregor, J. G. and


Cornell, C. A., "Development of a Probability Based
Load Criteria for American National Standards A58", NBS
Special Publication 577, National Bureau of Standards,
Washington, D. C . , June 1980.

25. Flower, W. R. and Schmidt, L. C . , " Analysis of Space


Truss as Equivalent Plate," ASCE.Journal of Structural
D i v i s i o n . Vol. 97, No. ST12, Dec.. 1971, pp. 2777-2789.

26. Galambos, T. V, and Ravindra, M. K . , "Properties of


Steel for Use in L R F D ," Journal of the Structural
D i v i s i o n . ASCE, Vol. 104, No. S T 9 , Sep. 1980, pp. 1459-
1468.

27. Galambos, C. F . , "Highway Bridge Loadings", Public


Roads, Vol. 43, No. 2, Sept. 1979, pp. 53-60.

28. Garson, R. C . , "Failure Mode Correlation in Weakest-


Link Systems," Journal of the Structural D i v i s i o n .
ASCE, Vol. 106, No. S T 8 , Aug. 1980, pp. 1797-1810.

29. Ghali, A. and Neville, A. M . , "Structural A n a lysis."


Chapman and Hall, London, 1978.

30. Ghosn, M. and Moses, F . , "Bridge Load Modeling and


Reliability Analysis," Report No. R 84-1, Department of
Civil Engineering, Case Western Reserve University
1984.

31. Ghosn, M. and Moses, F., " Reliability Calibration of


Bridge Design Code," Journal of the Structural
D i v i s i o n . ASCE, Vol. 112, No. S T 4 , April, 1986,
pp. 745-763.

32„ Goodier, J. N. and Hodge Jr., P. G., " Elasticity and


Plasticity," John Wiley and Sons, Inc., 1958.
181

33. Gorman, M. R . , "Reliability of Structural Systems,"


Report No. 79-2, Civil Engineering Dept., Case Western
Reserve University, Cleveland, May 1979.

34. Grierson, D. E. and Aly, A. A . , " Plastic Design Under


Combined Stresses," Journal of the Engineering
Mechanics Division. ASCE, EM4, August 1980.

35. Grouni, H. N. et. al., "Calibration Task Force-Highway


Loading", Report, the Ministry of Transportation and
Communications, Downsview, Ontario, Canada, 1978.

36. Hamada, S. and Longworth, J., "Ultimate Strength of


Continuous Composite Beams," ASCE, Journal of
Structural Division. S T 7 , July 1976, pp. 1463-1478.

37. Hambly, E. C . , "Bridge Deck Behavior." Chapman and


Hall, London, 1975.

38. Harman, D. J. and Davenport, A. G . , "A Statistical


Approach to Traffic Loading on Highway Bridges",
Canadian Journal of Civil Engineering. Vol. 6, 1979,
pp. 494-513.

39. Harman, D. J., Davenport, A. G. and Wong, W. S.,


"Traffic Loads on Medium and long span bridges",
Canadian Journal of Civil Engineering. Vol. 11, 1984.

40. Hart, G. C . , "Uncertainty Analysis. Loads and Safety


in Structural Engineering". Printice-Hall,Inc.,
Englewood Cliffs, New Jersey, 1982.

41. Hasofer, a> M. and Lind, N. C . , "Exact and Invariant


Second Moment Code Format," Journal of the Engineering
Mechanics Division. ASCE, Vol. 100, 1974, pp. 111-121.

42. Heins Jr., C. P. and Looney, T. G., " Bridge Analysis


Using Orthotropic Plate", Journal of the Structural
D i v i s i o n . ASCE, Vol. 94, ST. 2, Feb 1968, pp. 565-592.

43. Heins, C. P. and Kuo, J. T. C . , "Ultimate Live Load


Distribution Factor for Bridges", Journal of the
Structural Division. ASCE, Vol. 101, ST. 7, July 1975,
pp. 1481-1496.

44. Heins, C. P. and Fan, H. M . , "Effective Composite Beam


Width at Ultimate Load", ASCE, Journal of the
Structural Division. ST. 11., Nov. 1976, pp. 2163-2179.

45. Heins, C. P. and Firemage, D. A., "Design of Modern


Steel Highway Bridges". John Wiley, Interscience
Publications, New York, 1979.
182

46. Heins, C. P., "LFD Criteria for Composite Steel I-Beam


Bridges," Journal of the Structural D i v i s i o n . ASCE,
Vol. 102, ST. 11, Nov. 1980, pp. 2297-2312.

47. Heyman, J., " The Plastic Design of Grillages",


Engineering J o u r n a l . No. 176, Dec. 1953, pp. 804-807.

48. Hodge Jr., P. G., "Plastic Analysis of Structures".


Robert E. Krieger Publishing Co., Malabar, Florida,
1981.

49. Hollinger, B. A. and Mangelsdorf, C. P., "Inelastic


Lateral Torsional Buckling of Beams", ASCE, Journal of
the Structural Division. ST8, Aug. 1980, pp. 1551-1567.

50. Hsu, T. T. C . , "Torsion of Reinforced Concrete". Van


Nostrand Reinhold Company Inc., New York, 1984.

51. Jaeger, L. G. and Bakht, B., " The Grillage Analogy in


Bridge Analysis", Canadian Journal of Civil
Engineering. No. 9, 1982, pp. 224-235.

52. Kam, Tui-Yan, " Reliability of Framed Structures


Subjected to Nonlinear Behavior", Ph. D. Thesis,
Northwestern University 1982.

53. Kennedy, J. B. and Grace N. F . , "Load Distribution in


Continuous Composite Bridges", Canadian Journal of
Civil Engineering. Feb. 1983, pp. 384-394.

54. Kennedy, D. J. L. and Aly, M. G., "Limit State Design


of Steel Structures - Performance Factors ", Canadian
Journal of Civil Engineering. Vol. 7, 1980, pp. 45-77.

55. Kuo, T. C. and Heins, C. P., "Live Load Distribution on


Composite Highway Bridges at Ultimate Load," Civil
Engineering Report No. 53, University of Maryland,
College Park, Md., April 1973.

56. Kurata, M. and Shodo, H . , "The Plastic Design of


Composite Girder Bridges, Comparison With Elastic
Design", Doboka Gokkai-Shi, Japan 1967, Translated by
Arao, S. Civil Engineering Department, University of
Ottawa Canada.

57. Lathia, H. M . , " Bridge Analysis and Rating System,"


ASCE, Journal of the Structural D i v i s i o n . ST4,
April. 1979, pp. 211-220.

58. Lin, T. H., "Theory of Inelastic Structures". John


Wiley and Sons, Inc., New York, 1968.
183

59. Lin, T. S. and Nowak, A. S., "Proof Loading and the


Structural Reliability", The Journal of Reliability
Engineering. No. 8, 1984, pp. 85-100.

60. Lin, T. S., "Load Space Limit State Reliability of


Nonlinear Random Structural System", Ph. D. Thesis,
John Hopkins University, Department of Civil
Engineering, Feb. 1S85, pp. 235.

61. Lind, N. C . , Turkstra, C. J. and Wright, D. T . , "Safety


Economy and Reliability in Structural Design, " in
Proceedings, IABSE 7th Congress, Rio de Janeiro,
Preliminary Publication, 1964.

62. Lind, N. C . , "The Design of Structural Design Norms ",


Journal of Structural M e c hanics. Vol. 1, No. 3, 1973,
pp. 357-370.

63. Lind, N. C . , "Modelling of Uncertainty in Discrete


Dynamical Systems", Applied Mathematical Modelling
J o u r n a l . Vol. 7, June 1983.

64. Lind, N. C. and Nowak, A. S., "Proof Loading and the


Structural Reliability", The Journal of Reliability
Engineering,' to appear in no. 8,1984.

65. Lloyd, D. K. and Lipow, M . , "Reliability: Management,


Methods, and Mathematics," Published by the Authors,
Redondo Beach, California, 1982.

66. Madsen, H. 0., Krenk, S. and Lind, N. C . , "Methods of


Structural Safety". Prentice-Hall, Inc., Englewood
Cliffs, 1986.

67. Majid, K. I., " Nonlinear Structures; Matrix Methods of


Analysis and Design by Computers," Ne w York, Wiley-
Interscience, 1972.

68. McCarth, W. C. and Traina, L., " Equivalent Grid


Elasto-Plastic Analysis of Plates", Proceedings of the
Fifth Engineering Mechanics Division Specialty
C onference. University of Wyoming, Laramie, August
1984.

69. McGuffey, V, Iori, J., Kyfor, Z . , and athanasiou-


Grivas, D . , "Use of Point Estimate for Probability
Moments in Geotechnical Engineering," Transportation
Research R e c o r d . 809, 1981, pp 60-64.

70. Meek, J. L., "Matrix Structural Analysis". McGraw-Hill


Book Company, New York, 1971.
184

71. Melchers, R. E . , " Reliability of Parallel Structural


Systems", ASCE, Journal of the Structural Division.
Vol. 109, No.11, November 1983, pp. 2651-2665.

72. Michael, D. and Collins, M. P., "Diagonal Compression


Field Theory," ACI Journal. American Concrete
Institute. August 1974, pp. 397.

73. Michigan Department of Transportation, "Michigan Bridge


Analysis Guide," Prepared by the Bureau of Highways
Design Division, December 1983.

74. Mirza, S. A. and MacGregor, J . G., " Variations in


Dimensions of Reinforced Concrete Members," ASCE,
Journal of the Structural Division. Vol. 105, St4,
April, 1979, pp. 751-766.

75. Mirza, S.' A. and MacGregor, J. G., 11 Variability of


Mechainical Properties of Reinforcing Bars," ASCE,
Journal of the Structural Division. Vol. 105, St5,
May, 1979, pp. 921-937.

76. Mirza, S. A., Hatzinikolas, M. and MacGregor, J. G., "


Statistical Descriptions of Strength of Concrete,"
ASCE, Journal of the Structural D i v i s i o n . Vol. 105,
St6, June, 1979, pp. 1021-1037.

77. Mohammedi, J. and Longihow, A., "Probability-Based


Design of Highway Bridges- A Review", Proceeding of th
4th International Conference on Structural Safety and
Reliability. Kobe, 1985, Vol. 3, pp. 23-32.

78. Monnier, P. Y. and Schmalz, S., "Probability of Failure


of a Box Beam Using Simulation Techniques", Canadian
Journal of Civil Engineering. Nov. 1980, pp. 22-29.

79. Moses, F . , "Reliability Analysis of Frame Structures",


ASCE, Journal of the Structural Division. November
1970, pp. 2409-2427.

80. Moses, F., "Reliability of Structural Systems", ASCE,


Journal of the Structural Division. September 1974,
pp. 1813-1820.

81. Moses, F . , "Probabilistic Approach to Bridge Design


Loads", Transportation Research Record, No. 711, 1979,
pp. 14-33.

82. Naaman, A. E., "Prestressed Concrete Analysis and


De s i g n " . McGraw-Hill Book Company, New York, 1982.

83. Naaman, A. E., " Advanced Prestressed Concrete", course


notes and recommended readings on CE618 , Civil
Eng. Dept., University of Michigan, Winter 1984.
185

84. Neal, B. G., " The Plastic Methods of Structural


Analysis," rd ed. London, Chapman and Hall, New York
Wiley, 1977.

85. Nolan, C. S. and Albrecht, P., "Load and Resistance


Factor Design of Structures for Fatigue", University of
Maryland, College Park, 1983.

86. Nowak, A. S. and Lind, N. C . , "Practical Bridge Code


Calibration" , ASCE, Journal of the Structural
Division. ST 12, Dec. 1979, pp. 2497-2510.

87. Nowak, A. S. and Grouni, H. N . , "Safety Criteria in


Calibration of the OHBD Code", Proceeding of
International Conference on Short and Medium Span
B r i d g e s . Toronto, August 1982.

88. Nowak, A. S. and Regupathy, P. V., " Reliability of


Spot Welds in Cold-Formed Channels", ASCE, Journal of
the Structural Engineering. Vol. 110, No. 6, June 1984,
pp. 1265-1277.

89. Nowak, A. S. and Boutros, M. K., "Probabilistic


Analysis of Timber Decks". ASCE, Journal of the
Structural Engineering. Vol. 110, N o . 12, Dec., 1984,
pp. 2839-2954.

90. Nowak, A. S., and Zhou, J., "Reliability Models for


Bridge Analysis," Department of Civil Engineering,
University of Michigan.Report No. UMCE 8 5 - 3 . March
1985.

91. Nowak, A. S. and Grouni, H. N., "Serviceability


Criteria in Prestressed Concrete Bridges," ACI Journal,
Proceeding, Vol. 83, No. ..£0, Jan.-Feb. 1986, pp. 44-49.

92. OHBDC, Ministry of Transportation and Communications, "


Ontario Highway Bridge Design Code ," 2nd Edition,
Downsview, Ontario, 1983.

93. Orbison, J. G., " Nonlinear Static Analysis of Three


Dimensional Steel Frames", Ph.D. Thesis Cornell
University, May 1983.

94. Organization for Economic Co-Operation and Development


Road Research Group, "Evaluation of Load Carrying
Capacity of Bridges," Paris, 1979.

95. Park, R. and Gamble, W. L . , "Reinforced Concrete


Slabs" . John Wiley and Sons, 1980.

96. Rackwitz, R. and Fiessler, B., " Structural Reliability


Under Combined Random Load Sequences," Computers and
Structures. Vol. 9, 1978, pp. 484-494.
186

97. Rashidi, M. R. and Moses, F., "Studies on Reliability


of Structural Systems", Report No. R83-3. Department of
Civil Engineering, Case Western Reserve University,
Cleveland, 1983.

98. Ravindra, M. K. and Lind, N. C . , "Theory of Structural


Code Optimization," Journal of the Structural D i v i s i o n .
ASCE, Vol. 99, 1973.

99. Ravindra, M. K. and Galambos, T. V., "Load and


Resistance Factor Design for Steel," Journal of the
Structural D i v i s i o n . ASCE, Vol. 104, No. ST 9, Sept.
1978, pp. 1337-1353.

100. Reddy, V. M. and Hendry, A. W . , "Ultimate Load Behavior


of Composite Steel Concrete Bridge Deck Structures",
Indian Concrete J o u r n a l . Bombay, India, May 1969,
pp. 163-168.

101. Rosenblueth, E., " Point Estimates for Probability


Moments", Proceeding of Nature Academic Scie n c e .
U.S.A., Vol. 72, No. 10, October 1975, pp. 3812-3814.

102. Rosenblueth, E . , " Two-point Estimate in


Probabilities", Journal of Applied Mathematical
Modelling. May 1981.

103. Sabnis, M. G . , "Handbook of Composite Construction


Engineering" . Van Nostrand Reinhold Company, 1979.

104. Sartwell, A. D . , Heins, C. P. and Looney, C. T. G.,


"Analytical and Experimental Behavior of a Simple-Span
Girder Bridge", Highway Record, No. 295, 1969.

105. Siriaksorn, A., " Serviceability and Reliability


Analysis of Partially Prestressed Concrete Beams",
Ph. D. Thesis, the University of Illinois at Chicago
Circle, 1980.

106. Snyder, R. E . , Likins, G. E. and Moses, F., "Loading


Spectrum Experienced by Bridge Structures in the United
States", Report, Federal Highway Administration,
September 1982.

107. Thoft-Christensen, P. and Baker, M. J., "Structural


Reliability Theory and Its Applications". Springer -
Verlag, New York, 1982.

108. Turkstra, C. J., " Theory of Structural Safety, SM


Study No. 2, Solid Mechanics Division, University of
Waterloo, Ontario, 1970.
187

109. Umakanta Behera, A. M. and Fergson, M . , "Torsion Shear


and Bending on Stirruped L-Beams", ASCE, Journal of
Structural Division. ST 7, July 1970, pp. 1271-1285.

110. United States Department of Transportation, Federal


Highway Administration, "Recording and Coding Guide for
the Structure Inventory and Appraisal of the Nation's
Bridges," January 1979.

111. Warner, R. F. and Kabraila, A. P., "Monte Carlo Study


of Structural Safety ", ASCE, Journal of Structural
D i v i s i o n . Vol. 94, No. ST12, Dec. 1968, pp. 2847-2859.

112. Wegmuller, A. W . , "Overload Behavior of Composite


Steel-Concrete Bridges." ASCE. Journal of Structural
D i v i s i o n . Vol. 103, No. ST9, Sep. 1977, pp. 1799-1818.

113. West, R . , " Recomendations on the Use of Grillage


Analysis for Slabs and Pseudo-Slab Bridge Decks,"
Report 46.017, Cement and Concrete Association and
CIRIA, London, 1973.

114. White, K. R . , Minor, J., Derucher, K. and Heins, C. P.,


" Bridge Maintenance Inspection and Evaluation". Marcel
Dekker, Inc., New York and Basel, 1981.

115. Yam, L. C. P. and Chapman, J. C . , "The Inelastic


Behavior of Simply Supported Composite Beams of Steel
and C o ncrete", Proceeding of the Institution of Civil
Engineers. London. Dec. 1968, pp. 651-683.
**
116. Yatteram, A. L. and Husain, M . , "Grid Framework Method
for Plates in Flexure, " ASCE, Journal of the
Engineering Mechanics Division. E M 3 , June 1965, pp. 53-
64.

View publication stats

Vous aimerez peut-être aussi