l

Preface

Engineering decisions must be made in the presence of uncertainties which

are invariably present in practice. In the presence of uncertainties in the
various parameters encountered in analysis and design, achievement of
absolute safety is impossible. It is now more than twenty five years since it
was proposed that the rational criterion for the safety of a structure is its
reliability or probability of survival. In structural reliability, the probability
of failure (which is taken as one minus reliability) is taken as a quantitative
measure of structural safety. Probabilistic concepts are used in reliability
analysis, and in the design of structures. Using structural reliability theory,
the level of reliability of the existing structures (structures designed by
existing structural standards) can be evaluated. It can also be used for
developing a design criterion, that is, calibrating codes and developing
partial safety factors, the use of which will result in designs with an accepted
level of reliability. Structural reliability has been applied to many decision-
making problems, such as development of partial safety factors, establishing
inspection criteria, taking suitable decisions for improving the capability of
existing structures, development of maintenance schedule etc., in the field of
engineering.
Presently, only four or five books are available on this topic. These
books, written by foreign authors, are very expensive and beyond the reach
of Indian students and engineers. A book on this topic giving information to
readers on the results of the reliability study of reinforced concrete structural
elements and frames, with the field data pertaining to Indian conditions, is
also presently not available. There has been an overwhelming need among
students (present and past), fellow teachers in engineering institutions,
scientists in research organizations, and field engineers for such a book on
reliability analysis and design of structures giving fundamental concepts of
structural reliability theory and illustrating its applications to practical
problems. Teaching the course to postgraduate students for the last eighteen
years, delivering a series of lectures given periodically to the participants of
short-term courses, and research experience have motivated me to write a
book which treats the topic in a simple manner so that structural reliability is
easily underst:Dod and appreciated by readers.
The main aim of this book is to introduce the probabilistic bases of
structural reliability, the techniques and methods of evaluating the reliability
/I

of structural components and systems, the methodology in the development

of reliability-based design criteria, and the evaluation of partial safety factors
(code calibration). The whole field has been developed in such a way that
it is easily understood and followed by readers. Proofs and mathematical
derivations are given only if they serve to explain basic ideas, otherwise,
the original literature is cited for proof. Another important aim of this book
is to focus the attention of academicians and engineers on the importance of
an awareness of the need for a reliability-based design criterion, which is
being followed in Norway, Canada, USA, UK and other countries which are
in the process of modifying their standards. The book takes care of begi1mers
as well as experienced specialists . Though the book in general deals with
reinforced concrete and steel structures, the theory presented and the
methods indicated could be applied to other structures also. The book will
serve as a useful text and reference book for students, teachers, scientists and
engineers. It is hoped that the book will provide enough foundation for
further research work.
This is the only book which deals with the reliability analysis of
reinforced concrete frames, adaptive sampling method (for component and
system) and response surface method to estimate reliability, and fatigue
reliability evaluation of bridges. The reliability analysis of structural
components as well as systems is covered in a single volume. The book
gives the results of analysis of field data on basic variables and the reliability
study of concrete structures for Indian conditions. Nearly 110 examples are
worked out. Problems with answers are given under "Exercise" at the end of
each chapter.
Basic concepts of structural safety are introduced in Chapter 1. Certain
inadequacies in the conventional safety checking methods are exposed and
the need for a probabilistic criterion is emphasized. A history of the
structural safety is also briefly traced in Chapter 1. The necessary
background on statistics and probability, required for understanding the
subsequent chapters and reliability analysis, is included in Chapters 2 and 3,
respectively. A number of examples are given, illustrating the applications
of probability theory in civil engineering.
The collection of field data on basic variables, and a statistical analysis
of the same, is a very important part of reliability study. The statistical
analysis of resistance variables and load variables is presented in Chapters 4
and 5, respectively. The established statistics of basic variables for Indian
conditions are also given.
The computation of the structural reliability for the fundamental case of
two variables, load and resistance, is treated in Chapter 6. Difficulties
encountered in the probabilistic analysis of structures are brought out, and
how the Monte Carlo technique can be used to tackle such problems is
 vii

explained in Chapter 7. Applications of the Monte Carlo technique in

structural engineering problems are outlined and illustrated.
Simple methods of computation of reliability using ftrst-order second-
moment mean-value methods, and Level 2 and advanced Level 2 methods
are treated in detail in Chapter 8. Statistics of basic variables, established in
Chapters 4 and 5, are used to illustrate the methods of evaluating structuraJ
components' reliability. A number of problems are solved. The results of the
reliability study of the existing reinforced concrete designs as per the present
IS 456-1978 code ate also given. Chapter 9 deals with the computation of
the partial safety factors for a specified or required level ofreliability. The
method of fixing optimal partial safety factors, which wilJ ensure on i~ em
target reliability under all design situations, is introduced. The mcthodoJoAy
of the calibration of code is also treated in the same chapter. Results of U1e
study of the evaluation of partial safety factors for Indian conditions are also
presented.
The system performance and its reliability are of more concern and
ii'Jlportance to engineers. The modelling of structures for the computation of
reliability is demonstrated in Chapter 10. Bow1ds on the reliability of
structural systems are introduced. Methods of generation of domi rumt
modes, and the reliability analysis of steel and reinforced concrete frames
are presented. Results of the reliability study of reinforced concrete fran1es,
designed as per the Indian Standard Code, are also given.
Considerable research work has been done in developing methods to
improve the estimate of reliability and also better sampling techniques in
simulation methods. Second order reliability method is briefed in Chapter
11. Advanced simulation methods to calculate reliability, based on
importance sampling method (ISM) and adaptive sampling method (ASM),
are explained in detail and illustrated with examples. Response surface
method is also presented which can be clubbed with ISM and ASM.
Application of ISM and ASM to system reliability evaluation is included
with examples.
Fatigue is one of the principal modes of failure in bridges, offshore and
ship structures, pressure vessels etc. In Chapter 12, evaluation of reliability
of joints I details under fatigue is explained in a simple manner using S-N
curve approach. Reliability evaluation using lognormal format and Weibull
format ru;e presented. For the desired reliability level, estimation of design
stree range for fatigue criterion and partial safety factors are also dealt with.
Number of exampies are solved to demonstrate the fatigue reliability
evaluation. Application to offshore structures and bridges are explained.
Fracture mechanics (FM) approach is also introduced. The method of
evaluation of reliability, based on FM approach, is also presented.
While teaching the course, advanced topics discussing the generation
of dominant modes in frame structures, reliability analysis of reinforced
viii

concrete frames and advanced reliability methods may be omitted. Spending

only about 5 to 6 hours on Chapter 3, a course on first ten chapters can be
completed in a 35-hour lecture series. In case if the students have
background on basic statistics and probability, saving 6 hours in Chapters 2
and 3, Chapter 12 can also be completed.
I express my gratitude and indebtedness to Prof. P. Dayaratnam, former
professor, Indian Institute of Technology, Kanpur, who initiated me into the
field of structural reliability and suggested that I write a book on the said
field. He encouraged me throughout with his useful comments and
suggestions. I have used many of the results of the research work of students
who have worked under my supervision. I sincerely thank them, viz. A. G.
Deshpande, Padmini Chikkodi, Neville Kumar Shetty, David Arulraj,
Bajare, C. P. Joshi, Ravi, Kulkarni, Potkar and Prabhu for their contributions
to the book. Thanks are also due to the students who attended my lectures
and who, through their participation and comments, led to the development
of the book. My wife's encouragement throughout the hard days of writing
this book is greatly appreciated.

R. RANGANATHAN
 Contents

Preface v

List of Symbols xii

1 CONCEPTSOFSTRUCTURALSAFETY 1
1.1 General 1
1.2 Design Methods 1

2 BASIC STATISTICS 9
2 .1 Introduction 9
2.2 Data reduction 11
2.3 Histograms 14
2.4 Sample correlation 17

3 PROBABILITY THEORY 22
3.1 Introduction 22
3.2 Random events 23
'I[Random variables 43
3ijunctions of random variables 51

----
3.5 Moments and expectation 60

--
3.6 Common probability distributions 69

4
--
3 .7 Extremal distributions 79

RESISTANCE DISTRIBUTIONS AND PARAMETERS 91

4.1 Introduction 91
4.2 Statistics ofproperties of concrete 91
4. 3 Statistics of properties of steel 97
4. 4 Statistics of strength. of bricks and mortar 100
4.5 Dimensional variations 101
4.6 Characterization ofvariab1es 101
4. 7 Allowable stresses based on specified reliability 105

5 PROBABILISTIC ANALYSIS OF LOADS 112

5.1 Gravity loads JJ 2
5.2 Wind load 132
 X

6 BASIC STRUCTURAL RELIABILITY 143

6.1 Introduction 143
6.2 Computation of structural reliability 146

7 MONTE CARLO STUDY OF STRUCTURAL SAFETY 156

'.......--7.1 General 156
7.2 Monte Carlo method 158
7.3 Applications 164

c;JLEVEL 2 REL~ILITY METHODS 179

8.1 Introduction 179
8.2 Basic variables and failure surface 180
8.3 First-order second-moment methods (FOSM) 182

9 RELIABR.ITY BASED DESIGN 225

9.1 Introduction 225
9.2 Detennination of partial safety factors 226
9.3 Safety checking formats 239
9.4 Development of reliability based design criteria 242
9.5 Optimal safety factors 252
9.6 Summary of results of study for Indian
standards-RCC design 260

10 RELIABILITY OF STRUCTURAL SYSTEMS 268

10.1 General 268
10.2 System reliability 268
10.3 Modelling of structural systems 273
10.4 BoWlds on system reliability 283
10.5 Automatic generation of a mechanism 293
10.6 Generation of dominant mechanisms 301
10.7 Reliability analysis ofRCC frames 315
10.8 Structural safety in other fields 335

11 ADVANCED RELIABILITY METHODS 340

11.1 Introduction 340
11.2 Seoond order reliability method 340
11.3 Importance sampling method 342
11.4 Adaptive llllllpling method 351
11.5 Response aurface method 358
11.6 ASM and ISM in system reliability 366
11.7 Application of ASM • Rooftruas 371
 xl

1l FATIGUE RELIABU.,ITY 378

12.1 Introduction 378
l.2.2 s-N curve approach 383
12.3 LRFD format 393
12.4 Applications in bridges 396
12.S Applications in offshore and ship structures 404
12.6 Fracture mechanics approach 414
I

APPENDIX A: Standard normal tables 424

APPENDIX B: Partial safety factors for RCC members 430

Index 435
 List of Symbols

a Crack width
ai Initial crack width
ar Final crack width
b Breadth of section or member
D Dead load, cumulative damage
Dr Cumulative damage at failure
Dn Nominal value of dead load
d Effective depth
daldN Crack growth rate
EA Axial rigidity of a member
EI Flexural rigidity of a member
Es Young's modulus of steel
Esc Secant modulus of concrete
Etc Tangent modulus of concrete
epr Error in the estimate of pr
f, Modulus of rupture of concrete
F:\{ ) Cumulative distribution function of X
f\{ ) Probability density function of X
fv Yield strength of steel
g( ) Failure function
K S-N curve parameter
L Live load intensity or load effect
Lapt Arbitrary point in time live load
Lm Lifetime maximum live load
Ln Nominal value ofL
M Safety margin or plastic moment
m Slope ofS-N curve
N Number of cycles in fatigue studies
Pr Probability of failure
Pli Probability of failure of the structure under failure mode i
Prs Probability of failure of the system
Ps Probability of survival
Pss Probability of survival of the system
xiii

R Resistance
RD Design value of R
Rn Nominal value ofR
Ro Reliability
Rr Theoretical resistance
s Stress range, action
Se Equivalent stress range
Srd Design stress range
v Wind speed
Vmax Lifetime maximum wind speed
w Wind load
Wapt Arbitrary point-in-time wind load
Wm Lifetime maximum wind load
Wn Nominal value ofW
X Median of X
Directional cosine
fJ Reliability index
r Partial safety factor
Yo Partial safety factor for dead load
YL Partial safety factor for live load
Ymc Material reduction factor for concrete
Yms Material reduction factor for steel
/'R Partial safety factor for resistance
rw Partial safety factor for wind load
Ox Coefficient of variation of X
¢ Probability density function of standard normal variate
or null set
Cumulative distribution function of standard normal
variate
Px Mean value of X
ux Standard deviation of X
P.rr Correlation between X and Y
ASM Adaptive sampling method
BSS British standard specifications
FORM First order reliability method
FOSM First-order second-moment
ISM Importance sampling method
 xiv

ISS Indian standard specification

PWLEP Piece-wise linear elastic-plastic
PDF Probability distribution function
RCC Reinforced cement concrete
PMF Probability mass function
SDM Standard deviation multiplier
SORM Second order reliability method
 1
Concepts of Structural Safety

1.1 GENERAL
r:The evaluation of the safety of structures is a task of much importance. It
has been one of the subjects of interest for engineers. T~e safety of a
,syuctute_<k e ds O,!l ,the r~nce 8, Qf the s tru cture and the action , S,
(load or loa9 ~ff~t) QO the. structure. The ac iop. ·s ,a Juncti _n . of lo ads
(live load, wind load , etc.), which are rando m vari ables. Sim ilarly, t he
resistanc~ p_r response of the structure depends on the J?hysical properties
of materials, and the geometric properties of the structure which are also
's~ to stati ~tical v.ariations, and are probabilis itic Evcn though it was
7
~ nown that the above para meters were random variables, no seri us attempt
was made to consider their random variations, till 1960, in the analysis,
and design, and evaluation of safety. It was, probably, due to the reason
that engineers were not confident of applying probability theory or statis-
tics or other mathematical tools. It was only around 1960 that engineers
and research workers started realising the need for the evaluation of safety,
taking into account the random variations of the design parameters.

1.2 DESIGN METHODS

General principles for checking safety define a method for calculating the
behaviour and strength of structures subjected to loadings. Design methods
may be classified in the following ways.
1. By the way the coefficients related to safety are introduced:
Permissible Stress Method
This is also called the working stress design (WSD) method. Here, stresses
occurring under maximum service loads (working loads) are compared with
fractions of the strengths of materials. These fractions of the strengths of
materials are called permissible stresses. A structure is assumed to have
failed if stresses developed at any point of the structure are greater than the
permissible stresses. The safety is defined in terms of the factor of safety,
which is given by
fa ilure stress
Factor of safety = permtsst
. 'bl e st ress-
For ductile materials, viz. steel, the yield stress is taken as the failure stress,
and for brittle materials, viz. concrete, the ultimate stress is taken as the
2

failure stress. In this method, the elastic beh il vi \lllf of the material is
considered, (i.e. Hooke ' ~ Jaw is valid) and the ln<~d ddlcction curve of the
structure is linear.
For structural steeL the factor of safety is about I .67. What does thi s
mean in connection with the safety of a steel structure'! [t does not also
convey how much load the lructure will with tan I. lf the l'a ct r of safe!
is doubJed, docs it mean that the capacit ' of the ..,tr ucture i · al o do ubled?
Definitely it is no t because the behavi ur f th e.: mater ia l and , tru lme is
inelasti near th e c !lapse load . Ju st because th e :-; trcs. al n p( int i more
than the per mi siblc stress, it does not nccc ari ly ca use the c Jl a p e ol' the
stn.1cture cs pt' ia lly in the case ind etermin ate st ru cture . In the case of
reinforced cement co ncret e (R .) stru lure, th e use of per missi ble stre s
mctho 1 by intr du cin' tw dHTcrcnt fa t . rs or safe ty-o ne to ·on rete
{a bo·ur 3) and an mhcr t .rei nrorcin g steel bars (u buut 1.7 ) in vite nt re
critici m. What i their c mb incd ciTect in defi ning tlw nfely f R
structure ? Th e point s that wer ra ised wuh rc. pc 1 ll lh ted st ru 'turc
arc more pertinent to RCC structure nlso where the behaviour is nonlinear
and inelastic .
Whenever combinations of loads arc considered , viz. , de:1d load + li ve
load wind load o r dea d I ad + li e I ud + earthqua ke lo HI , an in rcasc
in th e all wa le stresses (33!· per ce nt) is cun siucrcd sincr. the ltkclihoou f
all the load r ac hing their mnx imu1n va lue : itttldt anc u~l y , i· rem He.
H wever, th er is no rati onal busis t'or the ck lll n of th e 'lluc, vi;,. k
per cent. Tt muy be said th at th e afet dcfineu in the r ·rmissib l • stress
d cs not rcft e t the true safely , r the actu a l ~a fety that L avai lable. T he
structure designed by the permissible stress method is safe under service
load and is assumed or expected to carry the ultimate load.
Merits of WSD are:
,.(1) s11nplicity and
.Yf) familiarity .
Deme1 its of WSD are:
(i) A ,eiven set of permissible stresses will not guarantee a constant level
o f sa fe! I'M nil structure ·. lW c ·amrlc. if two ronf s1 ruct ur e~-(u) R T
. hcl'l tvr t.: and (b R beam a nd slah type . de igned for th e sa 111 C I I\'C
load ti:>i nl! tile , nmc permi s ihle Ires cs. are c n idcred , the rn tio nt the
d ad load I<' li c loa d f 1r t li e ~ hell type will b co nsidcrn hly 11111 ·h io WC I
titan lh l· ra tiO ror tit -: slnb and be<tm t pe. Si nce the dead I wd ~a n b~.: csli·
mat ·d <11ld r m li l' t ·d more nccll ratc l th :1n the li \' c load,' hi It i. ubjcclc l
!1 mnrc pr11bubiitslt c \' <III<~ti o n . the shell r< t's tnu.: lu rc will hHvc a higher
ch un l' 11f f:Ji llll c th:1n the hcavtc r sinh and bl.'am ty c 1 oof ~ l rut· t u rc. Th nt
is to s:1 ' · twn ~ 1ru ' 1111 •• dcsi 'ltcd for Ihe :.amc live I rtJ u ·in th e amc
permiss ible ~ II"l' \Sl'S will ha v difTc r •nt leve ls t' ur •ty.
(ii) Th e workin , . trc~s ·h · ·k111,:t l'llrmat mny be un. afc when ne lnad
counte-racts the other load . For cx:unplc, consider a column, shown in
 3

Fig. 1. 1, subjected to dead load /) and wind load , 11 ·. The column h~s been
designed by the working stress method hy limiting stresses u .. uer service
loads in tension and compression to lirty per cent of Lh~.:ir respective strength
values, 3 N/mm 2 in tension and 20 N/mm 2 in compression. The stress
distribution under service loads 1.0 D +
1.0 W is shown in Fig. 1.1 c. When
the wind load is increased by twenty six per cent, it can be seen that the
stress at the point B reaches its failure level. Therefore, using the WSD
method can lead to designs with safety less than conceived adequate under
normal conditions, when loads counteract each other.

Stresses in Ntmm2

1111111 WLilllll! m
4 25
(a)
. Due to D

5·75 ~
(b)
Due to W
10 o --~ I~!Tj
(c l · ~~ 575
Due to 1·00 + 1·0W
'Ql(] l·50

(d)

Due to 1·00 + 1·26 W

3 00
FIG. 1.1 Working stress design with one load counteracting the other

Ultimate Strength Design and P/a.~tic Design Method

In these methods, the safety is ensured by magnifying the service loads (or
load effects, such as bending moments, etc.) and checking the structure at
this magnified load called collapse load. 'The magnification factor is called
the load factor, defined as the ratio of th~ ultimate load to service load) In
these methods, the safety is atlcast related to the capacity of the structure.
: -Th~y take into account the inelastic behaviour of the materiat: In the
ultimate strength design, the elastic analysis is first carried out, and only
sections are designed for the factored load. Hence, moment redistribution
is not taken into account in the ultimate strength design applied to RCC
structures. However, the plastic design applied to steel structures takes into
account the redistribution of moments, and the analysis of the structure is
4

carried out at the collapse load. The plastic design as compared to the
ultimate strength design, relatively gives a better picture of the true safety
of the structure.
In these methods, separate load factors are assigned to different loads.
Specifying a larger factor to the live load or wind load than the dead load,
reflects that the variability in the live load or wind load is known to be
larger than in the dead load. However, these factors have been selected
more or Jess only on the subjective judgement without any rational ba5is.
Fact red_l_()_a9. i~ ~n _i~agi11ary load which never C()!lles on t.h~ structwe.
These load factors, like factors of safety, are not related to the life of the
~!r~·cilire. !~_the load factori, ·assumed to ensure a 50 ye(lr life of -the
structure, is increased by 50 per cent, it does not mean that .the life t.o. the
si-~~cture also increases by the saine amount (i.e. life of the structure need
not be 75 years). The structure designed by the ultimate strength design
method or the plastic design method is safe against collapse load and the
same structure is assumed or expected to perform satisfactorily under
service load.
Limit State Method
A limit state is a state beyond which a structure or a part of a structure,
becomes unfit for use, or ceases to fulfil the fi.mction or satisfy the condi-
tions for which it has been designed.
The limit states are placed in two categories:
(i) Ultimate limit states-these correspond to the maximum load carrying
capacity (i.e. strength of the structure).
(ii) Serviceability limit states-these correspond to the criteria (durability)
under normal load (service load) conditions.
The coefficients of safety are related to ultimate load conditions and service
load conditions. That is, increased loads (or load effects) are compared
with the relevant resistance of the structure where effects of the service load
are compared with specific values. This method is definitely better than the
previous methods as the safety is ensured under collapse load and service
load conditions.
2. The second way of classifying the design methods is based on the
safety conditions.
(i) Deterministic design methods where basic parameters (e.g. loads,
strength of materials, etc.) are treated as non-random.
(ii) Probabilistic design methods where design parameters are considered
as random.
In the conventional deterministic design method, it is assumed that all
parameters are not subjected to probabilistic variations~: f:rowever, it is well
known that loads (live load on Ho . rs, wind load, oceari waves earthquake,
etc.) coming on structures arc random variables. Similarly, the strengths of
materials (strength of concrete, steel, etc.) and the geometric parameters
 5

(dimensions of section, effective depth, diameter of bars, etc.) are subjected

to statistical variations. Hence, to be rational in the estimation of the
structural safety, the random variations of the basic parameters are to be
taken into account. Since load and strength are random variables, the safety
of the structure is also a statistical variable)
In overcoming the uncertainties in the design parameters.( the safety
factor is ensured by taking the smallest value of the strength (R.) and the
largest value of the load (SJ~~ The safety factor, v, is taken as Rs/SJ. This
way of fixing the safety in design is very conservative and leads to un-
economical design.
The second way of fixing safety is as follows:
Let LJR be the allowed deviation from R and LJS the allowed deviation from
S. For the safety of the structure,
R>S
R - LJR > S + LJS
R(l- ":) > s( I+"%)
-~ > ( l + "ff)/( I - LJRR)

Hence, the minimum, value of the safety factor is

If the maximum variations in R and S are 10 per cent and 20 per cent of ,.
their respective computed values, i.e.

LJS
s = 02
. and ,:j:= 0.1

then the minimum value of v is

V=
(l + 0.2)
(1 - 0.1)
1.33
The safety can also be expressed as the ratio of the mean values of R
and S. This safety factor is called the ~Lg\_fu!x f~ct~r, vc, defined as

Vc = mean value of R
----,------;~
mean value of S
Definitions of safety factors vary widely and are probabilistically inaccurate.
To understand the drawback in defining the safety by central safety factor
consider Fig. 1.2 where probability density functions of Rand S are plotted.
When R and S are plotted, it will be seen that both distributions overlap.
The shaded portion (overlap) in Fig. 1.2 gives an indicative measure of the
probability of failure of the element or structure.
6

r, s
FIG. 1 .2 Overlap of acti on and resistance distributions indicating failure
probability

It will be seen now that for the same vc, the value of pr will be different.
Consider Fig. 1.3 where mean action and mean resistance are increased m
the same proportion keeping their standard deviations constant. Thus

v.
~
= krR
-krS·- R
=-~ -
S

Resistance
,., ...
I \
Action I \
I
I
\
\
\
\
\

''
r,s
FIG. 1.3 Effect of failure probability due to proportional changes in action
and resistance

It is observed f rom Fig. 1.3 that even though vc remains the same, the
overlap of th e lw curve · change, meaning the change of pr. Same things
hold g od when kr i. < I .
If the mean values of R and S are kept constant and dispersions in R
and S are changed (Fig. 1.4), it is seen again that the overlap of the two
curves changes, indicating a change in the value of pr. Since the mean values
of R and ,",' arc not changed, vc remains the same; but pr is different. The
probability of failure is affected by (i) the mean values R and S, (ii) the
standard deviations of R and S, and (iii) the point of intersection of the
two curves. d:his clearly shows the inadequacy of defining afety ~y the
central sa fety fa ctor. The best way t ~ delrne sa fely 1 by the probability of
failure or reliabilitY., Frcll lcnth ul ( 1.1 ) said: "Because the design of u
structure embodies uncertain prcdic\inns or th e perform ance of s,t ructural
 7

R~sist ance

r, s
FIG. 1.4 Effect of failure probability due to changes in dispersion of action
and resistance

materials as well as of the expected load patterns and intensities, the

concept of probability must form an integral part of any rational analysis
or design; any conceivable condition is necessarily associated with a nume-
rical measure of the probability of its occurrence. It is by this measure
alone that tbe structural significance of a specified condition can be
evaluated". Since, the achievement of absolute safety or reliability in the
um:ertain w rid IS unpo sible, <!Pro babilistic appr ach t~ the -evaluation of
safety becomes a sensible solution. The parameters encountered in civil
engineering - problems are subjected to random variations. There is a need
for a rational approach to the evaluation of structural safety, taking into
account these random variations. 't_he _stud.y_,<?,_f_~a,r}~~i!!ty _comes under the
domain of statistics _a~<;l prQ~.a~ility. Using the r babilistic approach, ~_here
is a p ssi6Tiity of obtainitW uniform relta 1ty (!.!_!liform performance in
S rucltir~S 11 er I J:i.en .design S.iUIJ ti ,I ) which- may probably Jead to
econ mical designs. Hence, probabilistic approach must be used. Hence-
forth ilfety will be defined by reliability which is defined as the probability
of survival of a structure under given environmental conditions. It is
nothing but the ability of the tructurc to fulfil it ns igned funcli( 1is
---- -
atts actorny forsome SQecifieJ time. In structural analysis 11nd design, it is
.
the probabi li ty tHat a stnl~Jure willnot'attain each specified limiL during a
pecified reference period. F r co nvenience the reliability is defined in
terniSOfthe probability of failure (probability of un atisfnctory per-
formance) which is equal to l-thc reliability of the stmc1urep Wh en
probability lheory is used in the limit tale de ign, the method 1s cu ll ~ d
probability-based limit stale design.
Probability-based Limit States Design
ln this design method, probabilistic methods are used to guide th~ selection
of the partial safety factors to loads and resistances of the structure or
•. structural element or materials ofthe structure, and they result in the de-
~ired overall safety. The principal advantages of this design method
are ( 1.2):
8

(i) more consistent reliability is attained for different situations as the

different variabilities of the various resistances and loads are considered
explicitly and independently
(ii) the reliability level can be chosen to reflect the consequences of
failure
(iii) it is a tool for exercising judgement in non routine situat ion
(iv) it provides a tool for updating standards in a rational manner.
The conceptual framework for the analysis of structural reliability anu
probability-based design is provided by the classical reliability theory.

REFERENCES
1.1 Freudenthal, A.M., "Safety and the Probability of Structural Failure", Transac-
tions, ASCE, Vol. 121, 1956, pp. 1337-1375.
1.2 Ellingwood, B.R., T.V. Galambos, J.G. McGregor and C.A. Cornell, "Develop-
ment of a Probability Based Load Criterion for American National Standard
A58", National Bureau of Standardr, Special Publication 577, Washington, D.C.,
June 1980.

EXERCISE
1.1 Is it possible to \!Ccount for the uncertainties in loads in the working stress
method?
1.2 Is it possible to account for the uncertainties in loads and material strengths in
the ultimate load method?
1.3 Wh<~t do you understand by limit state design?
1.4 What is central safety factor?
1.5 Whnt factors all'ect the probability of failure of a structure?
1.6 Whnt do you understand by uniform reliability in structure?
1.7 Do you think that the usc of factor of safety is related to the life of structure?
1.8 Do you think that a design obtained using the ultimate load method with a set of
load factors will ensure a particular life of the structure?
 2
Basic Statistics

In most engineering problems experiments are generally conducted. Experi-

ments may be carried out to study a particular roperty ofi 1natedal~ ·s-uch
as strength or to study a natural phenomenon like win ve octly, earth-
quake mtensity or, to assess t e s ren t l a eam,_;!c. ect 1 ni are to
be made on tile basis of these ex riments Experiments or observations are
usually repeated several times under umform or similar conditions. Even
though great care is taken to keep the conditions of experiments as uniform
as possible, the individual observations exhibit an intrinsic variability that
cannot be eliminated.
Consider the production of concrete. If a concrete mix is prepared and a
set 0 three or five cubes are macfe'out of this concrete mixal!.~ they are
tested for the CQJ.n ressive stren th it will be found that each cube will give a
dT ereilt-sl;n th. If another batch of conCrde is prepared for the samernix
ratio under the same conditions and a set of cubes are made out of this
concrete and tested, it will be found again that these cubes will give another
set of values fgr the .strength of concrete. Tbe average strength ~f concrete,
calculated for each set, will also be diftcr~nt. A typical set variaTt nof
average strenglb of M 20 concrete, obtained from a project (2.1), is hown
in Fig. 2.1. It is found that the resu t the st~ne;th obtf!ioed var,taqd do
---=--------.:..---_:;______
not give the same value repeatedly for the same mix. This --~-- ·····
means that the

4001-

1 ~,L-~6~~,,~~,s~~z~,--~2b6--~J~,--~3~6--4f.,~~4~6~cst,--•s~s:-.~~
Set nucnber

FIG. 2.1 Set variation of cube strength of concrete

 10

strength of concrete of a mix is ~ut,jcu to ranll om variations and it rs nol

possible to predict the exact--ou tcnrm or tire tesT.'"'
If a surl'cy on Jive load ('fl buildings is conducted, it may be observed
that the intensity of Jive load varies from bay to bay. A typical variation ol
floor load intensity (FLI) of an ofticc building (2.2. 2.3) is shown in Fig 2.2 .
The occurrence of liv~ loa~ 1_s purely a random phenomenon. It varies with
iltile-.- It !1-~;;, bc_;1 fou;1d that there is a variation from room 'to room in the
same !lll-o r. from floor to floor in the same building, and from building to
building.

4 8 12 16 20 24 28
Bay nurnber
FIG. 2.2 Variation of floor load intensity

Normally there i. sufficien t contrnl in the prod uction of a particular size

of a ma terial or the raw materia ls in the pi'Oclucts. However, th re will be
s ,me variations. This ca n be observed in the production of bricks or steel
bars or casting of concrete members. The variation of the mean deviation
of co lumn depth, observed in .a...,_building pr<~J~~t (2 ~4), i shown in Fig. 2.3·.
Variations are gener_rtlly smalLJ Varia(i ns arc more pronounced 10 natural
p~enom-enoil; e.'g~- wi_nd, rainfall, stream flow, height of ocean waves, etc.
Figme 2.4 giv_es the observed data on yearly maximum wind speed at
re
B_a n&al (2.5). It can i_,e aga in observed that one cannot definitely tell
whaL will be the maximuw-.. .wltid speed in the coming year. The wind speed
( is-probabilistic in nature. .._ ·

Colurnn number
FIG. 2.3 Variation of colun1n depth

It can b· c ncllldcd that 111 ·rc ex ists a ce rtain uncertainly in many of the
van . blcs witlt whi h t: r 11 'll.' im·c• :1re concerned . There arc inherent varia·
tic ns in all the phy I a! pr pcr11 · · or materia ls, loads, natural phenomena,
viz. wind, eartbqunkc, rainl'all . nullll cr ()f vchide cto si n' a road junction,
12

Sample Mean
Sample mean X of a random variable X, is defined as
-- I •
X=- I: x, (2 . 1)
n i-t

where x 1, x2 , . .. , x, is the sequence of the observed values. For illustration,

'! ~et of data on the compressive strength of brick is given and its mean
\'alu e w rkcd ut in Table 2.1 . The mean is a m easure of the ce ntral
tende n y {central va lue). Thi s is by ar the best ~tatislic t. num erically s um~
'•mt rizc ::1 di strib uti n a nd the ce ntre of gravity of the data . Fo r a given data,
if o ne is asked t give only a s ingle number, he wo uld pro ba bly use this
sample mean as his best prediction of the variable. The mean value is
liighly susceptible to extreme values of the observed data . Other measures
of the central tendency are the mode and the median.

TABLE 2.1 Computation of mza11, SD and CV of a set of data

Sl. Strength of brick I x1 - X' I (x1 - X)'

No . (N/mm 1)

I. 29.0 3.1 9.6

2. 27.7 1.8 3.24
3. 29.7 3.8 14.44
4. 21.4 4.5 20.25
5. 24.7 1.2 1.44
6. ~5.2 0.7 0.49
7. 20.0 5.9 34 .81 J
8. 27.1 1.2 1.44
9. 29.7 3.8 14.44
10. 30.8 4.9 24.01
II. 20.6 5.3 28.09
12. 20.6 5.3 28.09
13 . 30.1 4.2 17.64
\4. 28.0 2.1 4.41
IS. 23.9 2.0 4.00

II = }5 E o= 388. 5 E = 206.40
-
X'= -
3H8.5
~ 5-
vanaoce
. = .tt206.40
-- - = 14 .74
1
= 25.9 N /mm• s = V 14.74 = 3.84 N /mm 1
3 84
8 = ·
25.9
= 0 148
.

Mode _is the m;)s\ frequently observed data whereas the median is the
middle value or
the observation when the values &re arranged in the ranked
order of magnilUJe. If the number of observations are even, then the
average of the two middle observations is taken as median. Mode is not
unique.
For the given set of data, it is desirable to specify a number which gives
 13

an idea of the dispersion variability of the observations. Th~ range, the

standard deviation (SO) am! the coefficient of variation (CV) arc the general
measu_'::~. o_f_ d!!_p~rsion. ~ ,, .
Range

The range R is given by

R = X ! - Xs (2.2)
where Xt and Xs are the largest and smallest values of n values of the obs~r­
vations respectively, It is seldom used as a descriptive parameter of popula-
tion since it indicates very little about the way the distribution appears
inside the interval of values. However, this measure is attractive mainly
because it is computationally convenient and simple.
~_amo unt of sc11tt~t ~s -~lt:<! rly_q~p~nden Lon ho\y__rnllch the set of values
~vi!t_~s _[r_<?!!l_ ~h~_<;entr'!.L~a.!.l1~.: The greater the scatter, the _l~rger the ~ I
~~i_<!_tis>!l · The standard deviation is a measure of dis ersion.

Standard Deviation

This is defined as the positive square root of the average squared deviation
from the mean, i.e., ·
I n
s2 = - 1.: (x; -- X) 2 (2.3)
11 1= 1

where s is the standard deviation. The above formula gives an estimate of s.

An estimator whose expected value is not equal to the parameter it has
estimated is said to be a biased estimator. The unbiased estimate of sis
given by

S2 - I }; (Xi - -· X) 2 (2.4)
.-· ;;--=:[ i=l

Variance
This is defined as the square of the standard deviation. It is difficult to say,
purely on the basis of standard deviation or variance whether the dispersion
is large or small. T~i~jL~<::an i ngfu! Q.nly r_ell!~ive to the central val!Je. For
this reason, the coefficient of variation (CV) is often preferred and it is a
convenient measure for comparing the relative dispersion of more than one
kind of data.
Cf!dfi
.... ,;;;,'"of Variation
--This is defined as

(2.5)

where 8 is the coefficient of variation.

L The calculations of variance, SO and CV, are illustrated in Table 2.1.

14

2.3 HISTOGRAMS
The preceding section was mainly concerned with the collecti on of data and
the calculation of mean, SD and CV of a set of observations. Next step is
the presentation of the collected data in a useful form . The observations
are made and noted down as they occur and hence the collected data will be
in an unorganised form. Th_is tmorganise~ data is arranged properly. The
values are marked in increasing order. These ordered values are then divid-
ed.. Into int rvnl and the .,number of observ ations (frequency of observatio ns)
in each interval is plotted as bar. The plot obtained is called a histogram.
For plotting histograms, the approximate number of intervals may be select-
ed by using the following formula (2.6):
a = 1 + 3.3 log10 11
where a '"'= number of intervals between the minimum and maximum
values and
n =·c sample size (number of observations)
If the proper interval for drawing a histogram is not taken, the plot may
not give the correct picture of the underlying distribution of the variable.
Let the length of the brick be considered as a variable . A sample of 400
bricks are tested . Using Eq . (2 .6a), the number of intervals for drawing a
histogram for a sample size of 400 is
a = 1 + 3.3 Jog1o 400 = 9.59 (2.6b)
The grouped data on the length of bricks is given in Table 2.2. For this
grouped data, the histogram of the length of brick is shown in Fig. 2.5a.
The histogram gives the investigator an immediate impression of the range
ot' the data, its most frequently occurring values and the degree to which it
is sea ltercd .
TARLE 2.2 Grouped data 0 11 lcng llz o( brick for drawing histogram

Rangr (111111 ) Frequency Rel a ti ve frequency Cumul a ti ve frequency

221 221 0 .0025 I 0.0025

22J l25 0.0075 0 .0100
225 227 25 0 .0625 0 .0725
227 224 71 0 . !775 0 .2500
229 211 'l2 0 .2300 0.4800
2J I -211 RR 0.2200 0.7000
2JJ 21~ 75 0 . 1875 0.8875
2:l5 2:17 )J 0.0825 0 .9700
237 2:\'l 10 0 .0250 0.9950
239-241 2 0 ,0050 1.0000
11 400 1.0000

Relative frequency_ is obtained by dividing the number of obs~rvations in

an interval by the total numhcr of observations. The calculation of relative
frequency is illustrated in Table 2.2. In Fig. 2.Sa, the relative frequency is
 15

·~T '•I
0 ·25
n :400
r--
--
~ aor ~
1--
- 0 ·20

~ I
I
~ tiO !-
~ I 0
(!J

"'
.0
0 I
...._ I
0 l,Q f-- 0
-
~

(1J I
.D r-

~ 2+
r---
0 -05

O
L--=r ; 221 225
[l_ l 0
229 233 237 241
Length of brick (mm)
,. ,'
1·0
p-·
lbl /

r:/
I
0·8 I
I
I
~
j'
c: I
~ 0· 6 I
0"
...
411
I
0
411 I
-~ 0 ·4
., I
I
3 I
E
::J
"...)
02 - I
r/
.
I
I
,
.P
0 =..:::1'-: '"'
221 225 229 233 237 241
Length of brick (mm)
FIG. 2.5 Histogram and cumulative frequency of length of brick

also marked on the Y-axis on the right side. The relative frequency yields
the investigator an immediate idea that what is the chance of the variable
lying within a specified range. From Table 2.2 and Fig. 2.5a, it can be seen
that the chance (probability) of a value for a length L lying between
229 and less than 231 is 0.23. That is
P(229 ~ t < 231) = 0.23
where P(X) should be read as the probability of X .
16

In Table 2.2, cumu lut.ivc frequency ha. also been computed. The cu mul a-
ti e frequ•ncy- " lcss than a particular value''- i obtnincd by add in g the
freque ncies one by nest. rting from the l p of the frequ ency table. imilarly,
cumul ativ relative frequcn y can be omputed as shown in Table2.2. Fr 111
the table il C<lll be interpreted th at the chance f geLLing a va lu e for the
length of brick less th f1n 23 1 mm is 0.48. That i
P(L < 231) = 0.48
Th e umtllativc frequency diagram of the length of brick i shown Fig. 2.5b.
Fr m !hi ~ <lia ram one can qu ickly ay what i th e chance of ge tting n
va lue f r a length less than a particular value. For instance, this is equal to
0. 7 for the spe ified value of a length equa l t 233. Frequency di stribution
nf the field da ta n the trength of M 15 collcrete fl oor live load in office
building, yearly maximum wind speed and th e tren gth of over-rei nforced
prcstre .eel oncretc beam are shown in Figs. 2.6, 2.7, 2.8 and 2.9 res·
pectively (2. 1, 2.3, 2.5, 2.7).
The hi stograms shown in Pigs. 2.6 to 2. 10 have differe nt shapes. It can
be see n in ·ig. 2.5 that the hi togram is sy mmetrical about the mean whereas
other hi togram. are not· th at i 1 t e arc s kew~; W 1cther a hi togram is
~ nHnolrica l or not can be ou nd by computing the coefficient of skewnes .

108 0·27
n = 399
- r-- Mean= 24·03N/mm 2
~ SO = 5·76 N/mm2
Ill
c:
0
:;:.72 ~ -
.
Ill
>
Cll
Ill
.Q ,.__
-
0
0
.. 36
.8
- 0·09·~
Cll
>

'ii
E a::
::J 1--
z
r---
~
20 28 44
Cube strength (N/mm2)
36 52 °
FIG . 2 .6 Frequency distribution of M 15 concrete

Coefficient of Skcwnus
The sumple coelllcicnt of skewness is related to the third moment about the
mean. The coelllcient of skcwnes e1 is given by

I' I
.1'
I [I
· -1 -
II
.En (xt --
1-1
.fp] (2.7)

The coefficient of sk.ewnes~ iH u measure o;· skewness or asymmetry about

 17

"= 165
Mean :714 ·3 N/m2 0·30
SO =267·2 Ntm2

0·24
>.
u
t:1
a
0·18
....~

020 0·12 'i

..
ou
Q; a:
.0
e
:1
z 0 ·06

379 584 789 994 1199 1404

Floor load intensity ( N/m2)
1609 1814 2019 °
FIG. 2.1 Frequency distribution of floor live load in office building

n :33
Mean= 86·332 kmph .Jo&
SD :6·92kmph
>
u
~12
0
.J6 t:1
:;:;
....w
ou
> ~
a.
Ill
.0 6
0 Ty~ 1 extremal
....0 {largest)
...
a.
.0 4
E
:1
z

0
65 75 85 95 105 115 125
Wind speed (krnph)
FIG . 2.8 Frequency distribution of yearly maximum wind speed at Colaba

the mean. The coefficient is positive for histograms skewed to the right (i.e.
with longer tails to the right) and negative for those skewed to the left (i.e.
with longer tails to the left).

)'J*' SAMPLE CORRELATION

Engineers on many occasions may have to deal with two variables of
18

0 ·32 n =1052 r--

Meanc 110·4tm
SO: 6JtrTY
r- r--

,-- I--
Ill
>
~ 0·08
a: ,--

0
80 6 89 ·4
J
98·2
107·0 115·8
n
124 6
Resisting moment 1 Mro, ( tm)
FIG. 2 .9 Frequency distribtttion of resisting moment of ;;~, over-
reinforced nrestressed concrete beam

20
n = 60
r "0 ·66

18

•
•
"E • •
E
214
•
....01
~ •
c
<1>
• •• •
•':: 12
Ill

•• •
•
••
• •
8 •

I I 1. I ___.1
16 18 2G 22 24-----:-'
26~---'
28- 30
Cube str.?ngth ( N/mm2)
FIG . 2.10 St:utli"llt:lllt of the dat<t connecting cube ~trength and cylinde1 strength

related interl'~t ; nne v:tl i:thil- 111 :1y depend nn the other. When pairs of data
of two vari<1 blcs arc plot ted as shown in F;g. 2. 10, <1 plot called scattergram
 19

is obtained. A numerical summary of the tendency of the high values of one

variable X pairing with the high values of the other variable Y or, the high
values of X pairing with the low values of Y is given by the sample
covariance Sxr• which is defined as
] n
Sxy = n _ /}. (x1 - X)(y; - Y) (2.8)
1 1

If sxr is positive, it means that the high values of X pair with the high
values of Y and if sxr is negative, the low values of X pair with the high
values of Y.
Sample Correlation CoelficUnt
The sample correlation coefficient is obtained by normalizing the sample
covariance with standard deviations. The sample correlation coefficient, rxr•
is given by

'xr
Sxy
= sxsy = n -1 /~1
'l n (X;-sx x)(y;-Sy Y) (2.9)

rxr is a dimensionless quantity and its value varies from -1 to + 1. The

correlation coefficient gives a measure of the degree of the linear depend-
ence of the two variables. If rXY is equal to I, variables are perfectly
positively correlated, and if r equal to -1, variables are perfectly negatively
correlated. If rxy = 0, there is no linear dependence between the two
variables. Calculations of the sample covariance are illustrated in Table 2.3.

TABLE 2.3 Computation of sample co1•ariance and correlation coefficient

Cube Cylinder
Sl. strength strength (x 1 - XJ (y,- YJ (XI- X)(Y;- YJ
No. XI Y1
(N/mm•) (N/mrn•)

I. 15.17 9.86 -6.565 -3.955 +25.965

2. 17.92 11.29 -3.815 -2.525 + 9.633
3. 20.13 12.48 -1.605 -1.335 + 2.143
4. 22.54 . 14.65 +0.805 +0.835 + 0.672
5. 24.80 15.38 +3.065 +1.565 + 4.797
6. 18.67 11.95 -3.065 -1.865 - 5.716
7. 22.91 14.43 +1.175 +0.615 + 0.723
8. 27.70 18.00 +5.965 +4.815 +28.721
9. 29.24 18.42 +7.505 +4.605 +34.561
10. 18.27 11.69 -3.465 -2.125 + 7.363

E 217.35 138.15 108.862

X = 21.735; y = 13.815; sx = 4.533;

~)(lo8.861)
1
sxr Co _ = 12.096

12.096 093 -; ..
'xr 4.533 x 2.868 = · ,..
20

From Table 2.3, it is noted that rxy = 0 .93, indicating that the cube strength
and the cylinder strength of concrete are linearly positively correlated.

REFERENCES
2.1 Dayaratl(am, P. and R. Ranganathan, "Statistical Analysis of Strength of Con-
crete" Building and Environme11t, Vol. II, Pergamon Press, 1976, pp. 145-152.
2.2 Prabhu, U.P., "Stochastic Analysis of Live Loads in Office Buildings", .M.
Tuh . Tlresis, I.I.T., Bombay, 1984.
2.3 Prabhu. U .P. and R. Ranganathan, "Stochastic Analysis of Live Loads in Office
Buildings", Proceedings of the Natior.al Conference on Quality and Reliability,
I.I.T., Bombay, Dec . 1986, pp. 275-291.
2.4 Ranganathan, R. and C.P. Joshi, "Statistical Analysis of Strengths of Concrete
and Steel and Dimensional Variations", Report No . DS and T: 4(1)/83/STP-III/2,
Civil Engineering Dept., I.I.T., Bombay, March, 1985.
2.5 Ranganathan, R., "Statistical Analysis of Wind Speed and Statistics of Wind
Load for Probabilistic Criterion", Report No. DS and T: 4(1)/83/STP-Ill/4, Civil
Engineering Dept., I.I.T., Bombay, March 1986.
2.6 Benjamin, J.R. and C.A. Cornell, Probability, Statistics and Decision for Civil
Enigneers, Mc-Graw-Hill, New Yo.-k, 1970.
2.7 Ranganathan, R. and P. Dayaratnam, "Reliability Analysis of Prestressed Con-
crete Beams", Bridge and Structural Engineer, Vol. 8, No.2, June 1978, pp. 11-24.

EXERCISE
2.1 The test results of the compressive strength (N/mm•) of 50 concrete cubes obtain-
ed from a building project are given below:

17.24 !6.18 16.53 15.20 18.40

19.73 17.24 20.53 19.38 23.42
17.60 18.76 20.()1) 20.36 20.27
19.82 20.09 21.78 19.82 19.11
21.42 22 .31 21.L6 21.15 20.36
13.60 14.98 15.08 18.01 14.93
13 96 15.64 15.56 16.09 13.96
13.87 15.75 12.11 17.18 16.20
15.65 16.27 14.83 13.24 15.03
13.96 15.58 17.36 16.29 16.71
Calculate the mean, the standard deviation, and the coefficient of variation of the
strength of ::oncrete for the given data. Plot a histogno.m. Determine the chance of
ge!ting a value less than 15 N,'mm 1 .
(Ans. X= 17.41, s = 2.76, P(X < 15) = 0.37)
2.2 Samples of soil are collected from various depths below ground level and tested in
the laboratory to determine their shear strength. The collected field data are given
below :

Depth (m) 2 4 5 6 7

Shear (kN/m1 ) 14 8 20.3 32.2 39.0 42.0 56.4

strength
 21

Determine the sample covar' lltee and correlation coefficient between the depth of
I hesoiJ and its shear strength . What do you infer?
(Ans. sxy = 27.99; '..tY = 0.987)
For the data gi\·en in Exercise 2.1, determine the coeffiCient of sk.ewnes>. What do
you infer? (Ans. 0.27)
2.4 What do you understand when you get a negative correlation for a given set of
data? Give an example in a civil engineering field where negative correlation
appears?
 3
Probability Theory

3.1 JNTRODUCTJO!\
In every walk of life people make statements that are probabilistic in nature
and that carry overtones of chance; example, we might talk about the
probability that a bus will arrive on time, or that it may rain tomorrow,
or that a child to be born will be a son, or a flood may occur in a river this
year ... , and so on. What is the characteristic feature in all the above pheno-
mena? It is that they all lack a deterministic nature. Past informations, no
ml;ltter how voluminous, will not allow us to formulate a rule, and to deter-
mine precisely what will happen when the experiment is repeated. Phenomena
of the above type are called random phenomena. The theory of probability
involves their 5tudy. Variables in engineering problems can be classified as
shown in Fig. 3.1. In a deterministic study, parameters may be considered
as a function of time (time variant) or in some problems they may be inde-
pendent of time (time invariant). Similarly, in a probabilistic study variables
may be treated as time invariant or in many cases time variant (e.g. wind
~<)'ad ocean-wave heig ht, earthq~ake, etc.). When a random varjable as-
~um ·s values as a fun c tion of time, the variable is C'llied a stochastic
variable. The probabilistic study of stochastic variables is called stochastic
process or random pro~ess. In most engineering problems, random variables
of interest are stochastic in nature. However, for simplicity, variables are
considered as time invariant. This chapter deals with random variables
which are not stochastic.

Time invariant

Dl.'lerministic
Time variant

Time invari~t

--......__ Time vanant

(StaLonM'/ c.r>d nor.stationary)

fIG :l 1 Var111l>l" s ;, rHI classlirc<Jtion

 23

3.2 RANDOM EVENTS

Preliminary Ideas
Before discussing probability theory a few preliminary ideas that are used
in subsequent discussions are introduced in this section. The first is the
concept of an event.
Sample Space and Events
Consider a number of persons boarding a bus at a particular bus stop.
A survey is carried out daily at the same time. The capacity of the bus is
60. Let X be the number of persons boarding the bus. It can be seen that
X can assume value 0, I, 2, 3, ... , 60. Each one is a possible outcome of
the experiment (experiment in a general sense; here experiment is the
counting of the number of persons boarding the bus). Each of these out-
comes is called a sample point. The collection of all these pos,ib<e outcomes
of the experiment is called a sample space. l-Ienee this sample space
consists of a set S of points called sample points. Each or these outcomes,
a sample point, is called a simple or elementary event. Let a simple event be
denoted byE;, the c;ubscript i, here, denoting the number of persons. Then
for this example, there are sixty-one simple events denoted by Eo. Et, E2, . .. ,
E,, . . . , E60, where E; is an event representing the occurrence of the varinble
X taking a value i. The s~m1ple space for this experiment is

S = {Eo, Et; E2, ... , Es9, E6o}

One may be interested in the collection of n set of outcomes in an
experiment. In this case, one may note down the number of persons board-
ing the bus-(i) less than 4 or (ii) greater than 5 and lesslhan 10. Such events
are called compound events or simply events. If A is 'the event representing
the number of persons less than 4, then sample points in the event A are
A .- {Eo, E1, Ei, EJ}
and if the event B is defined as the number of persons bonrding greater
than 5 and less than 10, then sample points in the event Bare
B -· {Er, , h, ... , E9}
lf Cis an event denoting the number of persons boarding the bus greater
than or equal to 4, then
C = {£4, E~ . ... , EGo}
when events A and C arc compared , it can be seen that event C consists of
all points that arc not included in event A. Such two events are called
complementary events. A is the complement of C
The formal definitions of various events arc given below:
(i) Simple event: An event consisting of a single sample point. A simple
event cannot be decomposed into a combination of other events.
(ii) Compound event: An event made up of two or more sample points.
24

(iii) The complement event Ac of the event A consists of all sample points
in the sample space not included in the event .4. In the cited example
Ac "'·~ C

(iv) Certain event: An event constituting all sample points in the sample
space.
(v) Null event is the complement of a certain event and is generally
designated as </>.
In the experiment-number of persons boanling a bus-the sample points
have individually discrete entities and arc countable. Such a sample space
is called a countably discrete sample space. If such a sample space has
finite nu ·nbcr of points, it IS called countably finite discrete sample space.
A second example of the finite sampk ~ pace i' the observation of the
number of days in a year wilh l·~nlperature higher than say JOoC at a
purticular location. fach day of the year is il possible sample point. The
sample space con~ists of 365 sample point~. This is a discrete finite (count-
ably) sample space. Another example i~ the observl:tion of a successful bidder
among the number of contractors bidding for a parti~:ular job.
Sometimes a discrete sample space may have sample point which arc
countably infinite. for example, the number of persons undergoing an car
operation in a year in the wlloie world. ln this case the number of persons
could he thcor<:::tically any integer from zero to infinity. Such a sample space
is calkd a c!iscretc (countably) infinite sample space. Another example is
the ob~e' '- ation of the number of accidents along a busy road during a
yca1
Many e11gineering problems or physical situations involve measurements.
Consider the experiment, the measurement of a deflection during the load
test of a reinforced concrete beam. It may be ro~sible to get ,,;1y value
(noninteger) of the deflection starting from zero with the in:;trumcnt (dial
gauge). If the least count ('l the dial gauge is 0.001 mm, deflections could
be obtained at an increment of 0.001 mm starting from zero. The sample
points may be 0.000, 0.001, 0.002, 0.003, .. n where n is the number of
points which may be etfcctively large. The o.ample space will have a conti-
lllllllll ol' sample points. Such a sample space is called a continuous ~ample
~p:IL'C. l·or convenience. a continuous sample <;pace is defined from 0 to co,
l.l'. any value g1cater than zero is assumed as a possible outcome of an

L'\IH.'IIIIH.'III l11 'ome situations, the variable of interest may assume negative
vallll.:'. I"' nan1ple, the deflection of a simply supported prestressed
conu-clt' hc:1111. During the initial loading stages of the beam, the beam
will havl' : 111 UJHValll (negative) deflection, and after a certain Tevel of the
att..:rn:d load, a dt>ll'llll'<lrll (positive) deflection. In this case, the interval
--- oo to ! co become' Ihe sample space. Another example of this case is
the measurement ol' l'l nor.
In som,~ f-!ijsical situ:lll<lJIS . il 111:1\' he known rrom physical conditions
that a continu,lliS v:u iahlc (II iniL'ITSI c:~n :1s,umc a \';lluc within a tinite
 25

interval only. For example, (i) wind directions can be observed from 0° to
360c and this finite interval becomes the sample space, (ii) the strength of
M 15 concrete if one assumes that the strength value cannot exceed 40
N/mm 2 or be less than 8 N/mm 2 • Then the interval 8 to 40 N/mm 2 is the
sample space. Events in a continuous sample space can also be defined. A
few examples are given.
ExAMPLE 3.1 Consider a traffic engineer noting down the number of vehicles
on a small bridge at u particular instant. The maximum number of vehicles
that can be at a time is I0. Sketch the sample space and show the events
(i) observing less than four vehicles, and (ii) observing greater than 5 and
less than 9 vehicles.
Solution: Let
E; = the event observing i vehicles.

Hence, the sample space is

S =-= {.Eo, Et, Ez, ... , £10}
and this i~ shown in Fig. 3.2. This is a finite discrete sample space.

Event A / Event B

r ----- -- - --, r- ------,

lEO El E2 EJI E4 I E6 E7 Es I
L_ _ _ _ _ f" ____ _J
L - -- - -- - J

FIG . 3 .2 Sample space and events- Exampl e 3.1

Let
A = the event observing < 4
B = the event observing > 5 and <9
These events are also shown in Fig. 3.2.
EXAMPLE 3.2 An engineer at an airport is measuring the wind speed at
regular intervals of time. Sketch the sample space and mark lhe event A
observing the wind speed less than 40 kmph and the event B observing the
wind speed 60 kmph.
Solution The possible outcomes of the measurement of wind speed can be
from 0 to co. Hence, this is a continuous sample space. This is sketched in
Fig. 3. 3. Events A and B arc also marked in the same figure.
In many problems we are interested in events which arc actually the
combinations of two or more events. A It hough the read er must surely be
familiar with these terms, let us review them briefly.
Let us now define relations between events. Consider counting the number
26

0 20 40 60 r.o 100 -a:

f-- A
FIG. 3 .3 Sample space und events-E xample 3 .2

ol· vehicle~ l'll a bridge at <I lime (Example 3.1). The sample ~race is
S · {Ln, 1~·1. b, EJ, .... E,o}
Let A · 'the e\cnt ohscrving < 4 vehicles
C , .. the event observing> 7. and < 7 vehicles
Then we have the following relations between A and C:
(il Both A and C occur together. This situation will happen when the
simple event EJ occurs. This is \1 rilten as An
C and it is read as A
intersection C.
(ii) Either A or Cor both occur. In the present example, this is the event
having sample points (Eo, L1, E2, E\ E4, Es, EG). This is written as AU C
and is read as A union C. i.e. sample points in the event A U C arc
AU C = (Eo, E1, E2, b, £4, Es, EG)
Let D be the event observing more than 7 vehicles. Then it can be seen that
the events A and D have no points in common. The event AnD is impos-
sible . This is written as
AnD= 0
and A and D are called mutually exclusive or disjoint events. Relations
AU C and An Care marked in Fig. 3.4.

AUC

E vent 0
FIG. 3.4 Relationship among events

The union and the interscdion of events are best understood by drawing
Venn diagrams shown in Fig. 3.5. In Venn diagrams, the sample space is
 27

represented by a rectangle while events are represented by regions within

the rectangle.

(a) (b)

(c) {d)
FIG. 3.5 Venn diagrams for (a) intersection, (b) union. (c) mutually
exclusive and (d) complement of events

Consider the observation of the direction of wind speed at an airport.

This has a continuous sample space with the variable taking any value
from 0 to 360°. Let
A be the event observing wind direction ~ I 00°
and B is the event observing wind direction > 100°
If the accuracy of the measuring instrument is 0.1 o, then
s= (0.0, 0.1' 0.2, ... , 100.0, 100.1, ... , 360)
A = (0.0, 0. \, ... , \00)
B = ( 100.1, 100.2, ... , 360)
It can be seen that the union of the two events A and B contains all sample
points in the underlying sample space and these two are collectively
exhaustive. In general, two or more events are collectively exhaustive if the
union of all these events constitutes the underlying sample space.
A listing of a number of important laws obeyed by the combination of
the events is given below without formal proofs.
Identity laws : A U<P =A, A nS = A
AUS = S, An<P = <P
Idempotent laws : AU A = A, A nA = A
Complement laws : AUAc = S, AnAc = <P
Commutative laws : AU B = B U A
AnB = BnA
28

De Morgan's Law The complement of the union and intersection of events

is the intersection and union of their respective complements.
(AUB)<' = Acnnc (3.1)
(An B)< ~" ACU nc (3.2)
A.~.wciative Lau•.1:
A U ( B U C) = (A U B) U C (3.3)
A n<sn c)= (An B)nc (3.4)
Disiribuiive LaiVs:
A U(Bn C)= (AU B)n(A U C) (3.5)
A n(BU C)= (An B)U(A nc) (3.6)
Venn diagrams for associative and distributive laws are shown in Fig. 3.6.

(AUB)UC
FIG. 3.6 Venn diagrams for associative and ~istributive laws

Probability Measure and Axioms

The empirical notion of probability is that of relative frequency; the ratio
of the total number of occurrences of a situation to the total number of
times the experiment is repeated. When the number of trials is large, the
relative frequency provides a satisfactory measure of the probability asso--
ciated with a situation of interest.
A random experiment is a repetitive process or operation that in a single
trial, may result in any one of a number of possible outcomes such that a
particular outcome is determined by chance, and is impossible to predict.
Under a given set of conditions, a random experiment has N exhaustive,
mutually exclusive and equally likely outcomes A1, A2, ... , AN. If M of
the outcomes are associated with the occurrence of an event A and N-M
outcomes with the nonoccurrence of A, the probability of the occurrence
of A is (M/N), i.e.,

P(A) = ~
II' an experiment has a sample space and an event A is defined on S, then
P(A) is a real number called the probability of the event A, or the probabi-
lity of A, and this J>(A) must satisfy the following axioms:
(i) For each event A <'f 8
0 ~ P(A):::; I
 29

~ii) P(S) ,= I
(iii) If A1, A2, . . . are denumerable mutually exclusive events defined on
S, then
P(A1UA2UA3U ... ) = P(AI) + P(A2)-+ P(A3) + .. .
For a finite number of mutually exclusive events, say k,
k
P(A1UA2U .. . l)Ak) = .E P(A;) (3.7)
i-1

The probability of any event is the sum of the probabilities assigned to the
sample points within which it is associated.
From the axioms of the probability theory, the following formulae can be
obtained (students are expected to prove them) by drawing Venn diagrams:
= P(A) + P(B)- P(A n B)
P(A U B) (3.8)
P(AUBUC) = P(A) + P(B) + P(C)- P(AnB)
--P(BnC) --- P(AnC)-+ P(AnBnC) (3.9)
This can be extended to the union of a number of events.
EXAMPLE 3.3 During the route survey of a transport mini bus, 100
observations of the total number of persons travelling by the bus on a
particular length of the route yielded the following results (Table 3.1).
Observations have been made at random.
30

Define
A =---" more than 15 persons travelling by the bus
B , ' > 12 and < 18 persons travelling by the bus
Solution Assuming the number of persons travelling are mutually exclu-
sive events, we can use the relative frequencies given in Table 3.1 to repre-
sent the corresponding probabilities.
Hence,
P(A) = 0.08 + 0.16 -!- 0.14 + 0.17 + 0.16
= 0.71
P(B) = 0.01 + 0.05 --1- 0.06 + 0.08 + 0.16
= 0.36
The verification of Eq. 3.8 is
(An B) = (EJ6, E11)
Hence, P(A n B) =-= 0.08 + 0.16 = 0.24
According to Eq. 3.8,
P(A UB) = 0.71 + 0.36 --- 0.24 = 0.83
This can also be calculated as
(AU B) = (£13, £14, . • . , E19, E2o)
= 0.01 + 0.05 + 0.06 + 0.08 + 0. 16
+ 0.14 + 0.17 + 0.16
= 0.83
Hence, the theorem (Eq. 3.8) is verified.
Two Dimensional Sample Space
Consider the same experiment discussed in Sec. 3.2, namely the number of
persons boarding a bus at a bus stop. Instead of counting the total number
of persons, one is interested to note down how many males and females
board the bus. Let
l~tj "- the event representing i men and j women boarding a bus
Then the sample space for such a case can be sketched as shown in Fig. 3. 7.
The experiment in two-dimensional space involves an observation of 2
numbers at the same time.
Another example is that an airport engineer may be interested to note
down the wind speed and the wind direction for the orientation of an
airport. This is a continuous two-dimensional sample space which is shown
in Fig. 3.!\.
In some situations it is also possible to have a discrete-continuous
samplu space. 1-' 11 ".'Wmplc, in Example .I if the traffic engineer re ords
not only the nurnhcr ul' vcltich:s < 11 the bridge, but al o the total weight of
the vehicles on the bridge at the same time, f r s-uch a case, assuming t11e
 31

•
Eo.o Eo,1
• • •
Eo.z Eo, 3 - - - - - - - - - - - - - - - E 0, 60
•
• E1,1•
E1,0
•
E1,2
•
E1,J ------------E1,59
•
• • • • •
E2,0 E2,1 E2,2 Ez 3 -- - - - - - - · E2 58
' .

• • Ess,2
Esa.o Ess,l
• /
• •
E59,o Esg, 1

FIG. 3.7 Two-dimensional sample space-number of men and

women boarding a bus (Note i +
j > 60)

-,. 360
2'
'0

,-----
/
----
~

c
-~
.Q 240 Sampl~ spat~

'U
'U
c
3:
120
)
00 20 40 60 80 100 --('(.

Wind sp~~d ( k mph)

FIG. 3.8 Continuous two-dimensional sample space-wind
speed and direction

minimum weight of a vehicle is 30 kN and the maximum 300 kN, and the
maximum number of vehicles on the bridge at a time is 6, the sample space
will be as shown in Fig. 3.9.
Conditional Sample Space
If one is interested in the possible outcomes of an experiment, given that
some event A has occ'-lrred, the set of events associated with the event A can
be considered a new reduced sample space. In Example 3.3, given that 15 or
more men have been observed, the number of women boarding the bus will
have a reduced sample space as shown in Fig. 3.10. This is a conditional
sample space.
Suppose that in sampling the number of persons bording a bus at a bus
stop we restrict our observations only to women. Here, there is a new sample
space including only part of the elementary events in the original sample
space. This new reduced sample space is also a conditional sample space.
 6~
I
Vl
Ql
s I~
u 4!
E
Ql I'
> '
- 3~
o I
(i;
0
2~; -----
E :
::: 1 > - -
z
0 o~-·--;3t------g--- - -g---- -,~ - \~ ----T8-
Total w~1ght Xh)O(k~J)

FIG. 3.9 Discrete cont1nuous S'lmple sp~ce: number nf vehicles

and their total Wf'ighl nbservf'd nvP.r a smull bridge
nt an ins~ant

20
c:19
Ql
•

0
F. 16 ••
(i; 17
.rl
•••
516 •• ••
z
15 0
._ ---' - - - - - ' - - - - -- J
5 10 15 20
Nu:...,ber of women

Fig 3.10 Conditional sample space-number 11 women

in a bus given 15 or more men

Conclitio,al Probability
As the title s11ggcsts, we nrc interested in the probabil;ties ol rvenh. gi\cll
some condition . The conditional probabilit y or :111 event A. gi\en the occur-
rence or an event IJ, is defined by

P(4/
,
8),c:
'
P(A n
P(IJ)
B) ( 3.I())

provided /'(/1) 1- 0. P( A I B) is not defined if P(B) = • 0.

ExAMPI.I-. .lA In the rrevious Example 3.\ let us assume th<H 11 is given that
()person s <~rc travl'lling in the bus. Now under this condition, it is required
lo find out what is tht: clwnce of observing 3 or fewer women travelling in
the hu s.
Sol11tion Let
T the event observing (l persons travelling
F thl: event observing 3 or fewer women
 33

The sample points in Tare

(E6,o, Es,1, E4,2, £3,3, £2,4, Et,s, Eo,6)
Let it he assumed that the probability of observing a man travelling in the
bus is equal to that of observing a woman traveJJing in the bus. All proba-
bilities in the reduced sample space must add up to one. The probability of
observing each sample point in the reduced sample space is equal to 1/7.
Sample points in the event (T n F) are
(E6,o, Es,1, £4,2, £3,3)
Hence.
P(Fl T) = !!_(T n F) = 47
P(T)
~c that all pr babilities in the reduced sample space mu st add up to I.
~X AM Pt. 3.5 From n ertai n lot, 100 mild steel bars were .eJec ted at random
and tc ted for their yield stren gth and ullimale trength. Tf a spe imen has
an yield strength less than the guaranteed yield strength and less than the
guaranteed ultimate strength specified by code, we may define those cases as
failures. Under this condition, it was found that 25% of the specimens had
failed against yield strength, 20% against ultimate strength and 10% in
both.
(i) If a specimen had failed against yield strength, what is the probability
that it had also failed against ultimate strength?
(ii) If it had failed af!ainst ultimate strength, what is the probability it had
ulso failed against yield strength?
(ii) What is the probability that a specimen failed either against yield
strength or against ultimate strength?
Solution Let
Y (specimens which failed against yield strength)
=

Z = (Specimens which failed against ultimate strength)

Given.
P(Y) ~,.--. 0.25 P(7.) 7 0.~0

P( Y n Z) = 0. 10
(i) The probability that a specimen also failed against ultimate strength, ·
given that it had failed against yield strengt~ is
P(z I Y> = P(Y n z> =o o.w
P( Y) 0.25
= 0.4
(ii) The probability that a specimen also failed against yield strength,
given that it had failed against ultimate strength is
P(Y
1
z> = P(Y n z) = o.1o
PCZ) 0.20
= 0.5
34

(iii) The probability that a specimen failed either against yield strength or
against ultimate strength or against both is
P(YUZ) = P(Y) + P(Z)- P(Y fl Z)
= 0.25 + 0.20 - 0.10
= 0.35
AMPLE 3.6 Two vehicles are approaching a road junction. The action of
the driver of the following vehicle is dependent on the action of the leading
vehicle. The probability of the leading vehicle turning right is 0.3 and the
probability of the following vehicle turning right is 0.6. The probability of
both the vehicles turning right is O.l. Determine (i) the probability of the
following vehicle turning right if the leading vehicle turns right.
Solutim1 Let
L = the event that the leading vehicle is turning right
F = the event that the following vehicle is turning right
Given:
P(L) = 0.3, P(F) = 0.6
P(L n F)= 0.1
(I") P(F I L) =
0.1
0.3 = 3
I

P(LUF) = 0.3 + 0.6 - 0.1 = 0.8

(ii) What is the probability of the following vehicle not turnmg right
when the leading vehicle is not turning right? i.e. to determine P(F" I U) .

P(F' I U) = P(P
P(U)
U)n
From DeMorgan's Law,
P(fC n U) =: P(F U L)c
= I - P(F U L)
- I -- 0.8 = 0.2
Hence

P(Fc I U) :-=
02
·
2
(I - 7
(iii) What is the probability of the following vehicle not turning right
when the leading vehicle turns right? i.e. determine P(F i L)

P(rc In =
-
P(P
P(L)
L) n
But
P(F () L) ,. ~ P(L) -- P(L (I F)
- 0.3 -- 0.1 = 0.2
 35

Hence

Rule of Multiplication
The probability of the joint occurrence of two events is equal to the margi·
nal probability of one of the events multiplied by the conditional probability
of the other, given that the first event has occurred.
We can rewrite the formula, Eq.( 3.10), to yield
P(A n B)= P(li)·P(A-Im'- (3.11)
This is called the generalrulc of multiplication of the probabilities and is
extremely useful in many instances to find the probability that two events
will occur simultaneously.
The above theorem can be extended to the joint probability of a number
of random events At, A2, ... , A,.:
P(A1 n A2, ... , n A,.) = P(At)P(A21 A,)P(AJ \At n A2) ...
P(A,. I AJ n A2, ... , fl An-t)
EXAMPLE 3.7 Twelve concrete cubes are being cured in the laboratory.
Out of them, 9 cubes were prepared from a batch of M I5 concrete mix and
the other three belonged to M 42 concrete mix. During curing, the marking
face of the cubes have been kept at the bottom by mistake. Now three cubes
are drawn at random from the curing tank one after the other. Find the pro-
bability that all the three cubes belong toM I5 concrete.
Solution Let
At = the event that the first cube isM 15 concrete
Similarly A2 and AJ are defined.
The probability that the first cube is M 15 concrete is 9/12 since 9 out of
12 cubes are M 15 concrete, i.e.
. 9
P(AJ) = -
12
If the first cube is M 15 concrete, then the probability that the next cube is
M 15 concrete is 8/11 since only 8 of the remaining II cubes are M I5
concrete, i.e.

P(A2 I Ar) = TI8

Similarly, it can be written that

(A In Al) imp Iies that the first two cubes selected are M 15 concrete. Hence
the probability that the first three cubes selected one after the other at
random are M 15 concrete is
36

P(AJ n A2 n AJ) = P(AJ)P(A2 1 AJ)P(AJ 1A1 n A2)

= (tz)(A)C~)
21
=55
Probability Tree Diagram
In practice, a finite sequence of experiments are conducted. Each experiment
will have a finite number of outcome. with associated probabilities. A con-
cnien t way of describi ng uch a process and computing the probab il ity of
any cv nt is by a pmbability tree diagram, ill ustra t d below. The mu ltipli-
ation theorem is u ·cd to c mpu te the probabili ty.
ExAMPLE 3.8 Consider a reinforced concrete rectangular beam. The ulti-
ma te stren"lh or benm i a funcli n of the cube trcngth < f co ncrete feu,
the yield strength f steel, r.
Bnd other parame ters. If the cube strengt h of
co ncrete and steel nnJ other pora mct r arc t.: nsidercd <ts not subjected to
random variations, the given beam i under-reinforced dctermini ti all y.
Tl oweve r, if r•.,
ami;;, are subjected to random variation.' then the beam
may be under-reinforced or over-reinforced, depending on the values assumed
by f,.;, and_(;.. When the beam is subjected to an external bending mo ment,
tho beam may tail or surviv depending on whether the external momen t is
greater than or lcs · Hwn the resisting monh:nl f the beam. lt is gi en that
un lcr a given ex ternal m mcnt, the pr babi iity of the bea m becoming undcr-
rci nf<>r ·cd i 0.6 and the chance of failure of the beam i 0. 1 under thi
given event. The probability of failure of the beam is 0.2 if the beam is
over-reinforced. Assume the events under-reinforced and over-reinforced as
independent. Compute the probability of failure of the beam .
Sol!i t ion Let
A =' the event that the beam is under-reinforced
8 ~.-= the event thilt the beam is over-reinforced
F ·= the e\·e11t failure
S --· the event ~un•ivai
the pn>h;1hility tree diagralil is shown in Fig. 3.11 . The probability offailurc
or the hcam, ,.,-.is
{'( P(A)P(F I A) + P(B)P(F I B)
. (0.6)(0.1) -1- (0.4)(0.2)
().()6 + 0.08 '-~ 0.14
.'lt(l t i.\·t ical IIIIII'J'l'lldl'll<'t'
Ir 1he nee un cncc o I" an even 1 A is not affected by the occurrence of another
event B. then 1t i.'> said that the two events A and Bare statistically indepen-
dent. Mathematically, two events are said to be independent if and only if
P(A \ B) ,= P(A) (3. 12)
 37

FIG. 3.11 Probability tree diagram-Example 3.8

From this definition, it can aslo be written

P(A n B) = P(A)
P(B)
P(A . n B) = P(A)P(B) (3.13)
or P(B I A) = P(B)

Equation (3. I 2) or (3.13) is generally used to define the independence of two

events. By Eq. (3.13), it is meant that if the two events are independent, the
probability of their joint occurrence is equal to the product of their indivi-
dual probabilities of occurrence.
Extending Eq. (3.13) to a number of N events, A, B, ... , N are mutually
statistically independent if and only if

I
) P(A- n JJ n ... n N) = P(A)P(B) ... P(N)
.
(3.14)
In practice, an engineer may postulate that two events are independent,
or it may be clear from the nature of experiments, or he may be able to con-
clude after sampling that there is no apparent relationship between the two
events .
EXAMPLE 3.9 Two lakes a and b supply water to a city. The probability of
lakes a and b becoming dry in summer is 0.2 and 0.1 respectively. Lake a
can supply 60% of the city's full requirement when b fails (i.e. becomes dry),
and b can supply 80% of the city's full requirement if a fails. The proba-
bility that both will become dry is 0.05. Calculate the probability that the
city will have its full supply of water during summer, if there is a failure of
the lake.
Solution Let
A = event lake a becomes dry
B = event lake b becomes dry
38

Then
P(A) '= 0.2 P(B) = 0.1
I'( A n IJ) ,=, o.o5
From the rhysica1 situation , it can be said that the chance of A becoming
dry docs not depend on H.
Hence

P(A I ll) = ~~~~ = o.s

0
P(B I A) = 0~~ = 0.25
When there is failure. the conditional probability of the lake a not becom-
ing dry, p1, is given by
P(Acn B)
PI P(X08)
- P(ACI B)P(B)
- P(A) + P(.B) - P(B I A)P(A}
- (I - 0.50)(0. 1) 0.05 = 0 2
- (o.2f--F (O.t) - (0.25)(0.2) '= 0 .25 ·
Similarly, when there is failure, the conditional probability of the lake b not
becoming dry, p2, is given by
P(.Bc n A)
Pl = P(A U B)
P(Bc I A}P(A}
:- P (A um
= ( I - 0 .25)(0.2) = O
0.25 .6
If there is a failure of the lake, the probability that the city will have its full
supply of water during summer is
Pl X 0.6 + p2X 0.8 = 0.6
Total Probability Theorem
Suppose JJ is an event which is accompanied by a set of evCIHS A1, A2, .. . , A.,
which partition the sample .S such that they are mutually exclusive and
collectivcl_v cxhausti ve as shown in Fig. 3.12. One is interested in finding out
the probability of the event B, P(B), which pro ably is not possible to obtain
directly. This is obtained as follows:
From fig. 3.12, it is dear that
P(B) ,.. I'(IJ n A1) + P(B n A2) + ... + P(B n A.)
n
"- 1.: l'(IJ r1 A, l (3.15)
i-1
 39

(BOAt.> -

FIG. 3.12 Venn diagram for total probability

Expanding each term using the conditional probability theorem, we get,

P(B) = E" P(B I Ai)P(Ai) (3.16)

1~ 1

This is called the total probability theorem. This is illustrated with examples.
EXAMPLE 3.10 Two cities, 500 km apart, are to be connected. Alternatives
are: connecting them by rail (R), highway (H) and air (A) by constructing
airports at 2 cities. The government wi II decide on the basis of the cost and
merits of each. The chance of selecting R, H and A is 0.4, 0.5 and 0.1 res-
pectively. However, if the government decides on constructing a railway line
the probability of completing it in 3 years is 0.3; similarly, for highway and
air link, the corresponding probabilities that they will be completed in 3 years
are 0.7 and 0.4 respectively.
(i) What is the probability that the two cities will have the means of
transportation in 3 years?
(ii) If some transportation facility between the two cities is completed in
3 years, what is the probability that it will be a rail transport?
Given:
P(R) = 0.4 P(H) = 0.5
P(A) = 0. 1
Solution Let B = the project completed in 3 years.
Then it is given,

P(B I R) = 0.3 P(B I H) = 0.7 and P(B I A) = 0.4

(i) Using the total probability theorem, the probability that the cities will
have a transport facility in 3 years is
P(B) = P(B I R)P(R) + P(B I li)P(H) + P(B I A)P(A)
== (0.3)(0.4) ~- (0.7)(0.5) + (0.4)(0.1)
= 0.51

J
40

(ii) It is given that transportation facility is completed in 3 years. Tlnder

this condition, the probability that this will be a rail link is given by

P(R I B)= PCR n IJ)

P(B)
P(B I R)P(Rl
- P(B).
=.-. ~0.3) llAl = () 235
0.5 1 ..
(iii) If the government rejects outright the proposal of air connection,
what is the probabiltty that the final decision will be a highway?
This is given by
PI II (!? U 1-nl
P(llj R U H)
PR U H)
P(fl)
P(R) + P(ll)
0.5 5
0.4 -;- 0.5 9
LXAMPLE 3.11 A water ~upply system is to be dcsigneJ to meet the demand
during any given day during a summer. There are three demand levels, Dt,
1>2 and J)J, being equal to 200,000, 300,000 and 400,000 litres/day tespec-
tivcly. The probabilities (lf meeting these demand levels are 0.7,0.2 and 0 I
respectively. If the dcnwncl level is 200,000, the probability of the supply
being ~1dequatc during an} given d~lY in the summer is I and the cut rcspon<J-
ing values f,,r 300,000 anJ 4DO,fJOO litr..:s/!Liy :1re 0 Sand 0.6 rcspccti\dy
during :111y .~i,en day in the sunu>1er.
( i) Find the pmbability tbat lhl· supply will be adequate during any gi ..~11
day in the summer.
Given:
f'(Dtl == 0 7
.\olution l.et
·I tilt.: Slipp!Y ;s <tdequatt.: during any day in the summer
111~.:11 il ,, !'.1\'l:ll,

/'( , I iJ; ; . = I
l'(A I D1) • 0.8
c and P(A i DJ) = 0.6
using tilt.: ll>liil ptub~ibilil} theorem, the probability or the supply being
adequate on :11ty one day in the summer is calculated.
P(A) /'( il \ /),)/'(/) tl :- P(A l D2)P(D2) + P(A I DJ)P(DJ)
(I )(tl.7) : (tU)(O :) : (0.6)(tl, I)
= 0.7 i \l.lh i ().()(, 119~
 41

(ii) If the adequate wpply is observed, what is the probability that the
demand i~vel is 300,000 litres/day?

P(D 2 1 Al = PCD2 n AJ
· P( A )

P(A I Dl)P(Dl)
P(A)
•
(0.8)(0.2) 4
= '----'-:-o~
0.92 23
Bayes' Theorem
If A 1, A2, ... , An are mutually exclusive and collectively exhaustive events
of the sample space S, and B is any event in S as shown in Fig. 3.12, then
for any event A;,

P(B L A;)P( A;) ~ (l '. r i\ , ; (3.17)

P(A; I B) ---' R

1..: P(B I A;)P(A;) \ . r,..

i= I '

This can be considered as a converse problem of total probability theorem.

Bayes's theorem is quite useful in updating the available data.
ExAMPLE 3. I 2 Many government projects are executed by a contractor.
The chief engineer know3 from his previous experience that the chance of
getting a good quality of construction from the contractor is 0.8 and a bad
quality of construction 0.2. The evaluation of the quality of construction is
decided by the hammer test (nondestructive testing in situ). rr the strength
of concrete in situ obtained from the hammer lest is > 20 N/mm?., it is
decided that the quality of work is good. However, il is known that the
hammer test is not very reliable. The probability of a good quality work
passed by the hammer test is 0. 7 and that of a bad quality work 0.2.
After a project is completed by the contractor, there is a dispute between
1 • the contractor and the engineer about the quality or construction. Th~
hammer test is then conducted. If the good quality construction has passed
the test, what is the updated probability of expecting a good quality work
from tl1e contractor?
Solution Let

G = good quality of work

B = bad quality of work
C = the work passes the test
Given: P(G) = 0.8
P(B) = 0.2

P(C I G)= 0.7 P(C I B)= 0.2

42

The updated probability of expecting a good quality work from the

contractor is

I'(G I C) .. C P(G)P(C I G)
P(C I G)P(G) 1-P(C I B)P(B)

(0.8)(0.7)
- (0.8)(0.7) +
(0.2)(0.2)

~ ~:~ = 0.93

[n future, the engineer will use this value (0.93) as the probability of getting
a good quality of work from the contractor.
If another project is executed by the same contractor and that work also
passes through the hammer test, then

3.3 RANDOM VARIABLES

The random variable is a numerical variable whose specitic "alue cannot be
predicted with certainty before an experiment. The value assumed by a
random variable associated with an experiment depends on the outcome of
the experiment. This value is associated with every simple event defined on
the sample space, but different simple events may have the same associat•;d
value of the random variable, e.g. the strength of concrete, the wind speed
observed at a location, the number ol' persons waiting at a bus stop, etc.
Sometimes artificial values may be assigned to a random variable associated
with simple events. for example, a random variable, of the quality or a
product, may assume different .-Late~: poor, satisfactory, good, very good,
etc. Then each slate may be ;Jrtilicially assigned value as I, 2, J, ... , ek. A
random variable X on a sample space S is a function from the sample space
to a set ol' real numbers. The probability law of X, describing its behaviour.
is characterized by the probability distribution of X.
Di~-crete Variable~

The probability law of a discrete random variable is described by its pro-

bability mass !'unction (PM!:'). For a random variable X, it is written as

Px(x) = P(X ~= x) (3.18)

P(X cc.= x) is read as the probability of X, taking a value x. The PMF of a

random variable must satisfy the three axioms of the probability theory.
Hence
(i) 0 !( px(x) ::.; I J'or all X
(ii) E px(x;) "= I
all x;
 43
all .v 1 .;; b
(iii) Pla ::::; X ::::; b\ o-= E px(x;)
all x 1 >a

If the PMF of a random variable is ~"~ iven or known , one can immediate~
tellthe pro a 1 1ty of the random variable X assuming a-val~e x.
The probability distribution -~fa random variable X is alsOdescri bed by
its ~umtilatlv-ntlstr115u I OilfuncTioll(CDF)~-·px(:X)."-Thi~ is defined as- · -
( x(x)== P(X ~ :X) - -- for all x· (3.19)
For a discrete rand m variable,
Fx(x) = E px(x;) (3.20)
a ll x 1 .;; x

ExAMPLE 3. 13 Let X be the number of days in a week at a place having a

rainfall greater than 5 em. The following probabilities are assigned to the
possible va lues that X can assume.

r o.os x=O
0.10 x= I
0.15 x=2
0.30 X= 3
px(x) =
0.20 x=4
0.10 X= 5
0.08 x=6
L o.o2 X= 7

~ 1.00
Note that the axioms of probability are satisfied. Plot the PM F and CDF of
X. Find the probability of observing
(i) two or fewer days having a rainfall greater than 5 em
(ii) 3 or more days having a rainfall greater than 5 em.
Solution The plots of PMF and CDF of X are shown in Fig. 3.13
From CDF, it is easy to calculate the probabilities. Thus,
(i) P(2 or fewer days) = Fx(2)
= px(O) + px(l) + px(2)
= 0.3
(ii) The probability of observing 3 or more days having a rainfall greater
than 5 em:
P(3 or more days) = 1 - Fx(2)
= l - 0.3 = 0.7
44

Px (x)

PMF
0·3
02

01 .

Oo 2
__,_r I
·~

3 4 5 6 7 )(

Fx(x)
10 ~----

0·8 ~ -J
I

06 ,.___J I
COF
I
I
I
r-J
I
0 2 r-1
I
0 _J... _ _j __ -

0 1 2 3 4 5 6 7
FIG. 3.13 PDF and CDF of X-Example 3.13

The PMF can be recovered from CDF. For example,

px(x = 4) = Fx(4) - Fx(3)
= 0.8 -- 0.6 = 0.2
That is,
px(x;) = Fx(x;) - Fxtx; - E) (3.21)
where Eisa positive integer which is equal to l in this problem.
Continuous Variable
This is a function which can assume a continuum of points in a given inter-
val. The probability of such a variable; X, assuming a particular value is
zero. Its probability law is described by its probability density function,
fx(x). The probability of X in the interval is given by

P(a ~ X ~ b) = J: fx(x) dx (3.22)

It is to be noted that/x(x) itself does not give the probability. [tis only a
measure of the density of proba bility at the poinl. Probabilities are given
by integrals only . - -·
The PDF of X, in fact , is defined by
r ( ) _ dFx(x ) (3.23)
!XX -- ~

where Fx(x) is the CDF of X .

 45

If X can take values right from -- oo to +co, then

Fx(x) = P[X ~ x] = r"" fx(x) dx (3.24a)

If X can take values only from 0 to oo, then

,, ~) dx 1 (3.24b)

For Fx(x) to be a roper distribution function, the follbwing conditions

must be satisfied:
(i) Fx(-oo) = 0
(ii) Fx(oo) = 1

(iii) J:"' .fy(x) dx = I

(iv) fx(x) ~ 0
(v) Fx(x) ~ 0 and is nondecreasing with x

r
It is obvious that

-
< X ~ b) = ~"" fx~ -
"- Fx(b) - Fx(a)
r"' fx(x) dx

This is displayed in Fig. 3.14J -

fx (x) Fx (b)- Fx (a>

EXAMPLE
FIG. 3.14 -·
Continuous random variable

3.14 The bearing capacity, Y, of a soil below a foundation is

known to vary from 200 to 400 kN7m 2• Its PDF is given as

=
k(l -- ...L) 200 ~ y ~ 400
Jy(y)
{ 0 400
elsewhere
where k is a constant.
Determine the probability of failure of the foundation if the uniform load
on the foundation is 300 kN/m2 •
Solution For the given function to be a proper distribution function,
400 •

J200 Jy(y) dy =i
46

k
J40:l (
200
I -
400
\'
-
)
dr ""' SOk
-
k = 1/50
Therefore,

200 :( y :( 400

elsewhere
(i) The probability of failure of the foundation is
P( Y < 300) = FY(300)

-- J::: 1
50( 1 -- 4~ 0 ) dy = 0.75
Jointly Distributed Discrete Variable.1·
Here, two or more random variables are treated simultaneously. Consider
a number of persons travelling in a mini-bus having a maximum capacity
of n. One is conducting a survey and finding out how many men and
women are travelling in the bus for every kilometre length. Let X be the
number of men and Y the number of women travelling in the bus. These
rnndom variables on a sample space with respective image sets are
X(S) = (XJ, X2, . . . , Xn)

Y(S) = (y1, y2, ... , Yn)

The product set
X(S) · Y(S) = {(Xt, Yt), (x2, y2) ... }
is made into a probability space by defining the probability of the ordered
pair (x;, yj) to be P(X = x;; Y = yj) which is written as pxy(x;,-YJJ,Ttrts-- --
furttt1on Tscalled the joint probability mass function of X and Y. Hence,
the joint PMF is
pxy(x, y) = P[(X = x) n(Y = y)] {3.25)
the joint CDF is defined as
Fxy(x, y) = P[(X :o::;; x) n (Y :o::;; y)] (3.26)
For discrete variables X and Y,
Fxy(x, y) = I: 1.: pxy(x;, Y;) (3.27)
X; ~X Yj ~ y

The conditions to be satisfied are

PXY(Xt, YJ) ~ 0
E E pxy(X;, y;) = 1
all.\ allyj

The joint PMr and the joint CDF describe the joint probability Jaw or the
joint probabilistic behaviour 'or the variable£.

-- ----~-
 47

ExAMPLE 3.15 Five RCC beams are tested in the laboratory to determine
the load at the first crack, and at failure. If the load at the first crack is
greater than 20 kN, it is classified as safe against cracking. Similarly, if the
failure load is greater than 35 kN, it is taken that the beam is safe against
collapse. Let
X = number of beams safe against cracking
Y = number of beams safe against collapse
•
Because of the random behaviour of the beam, if the beam is safe against
cracking, it is not necessary that it should be safe against failure. The joint
PDF of X and Y is given in Table 3.2 and displayed in Fig. 3. 15.
TABLE 3.2 Joint PMF-Examp/e 3.!5

0 2 3 4 5

0 0 0.02 0 0 0 0 E = 0.02 = Px(O)

0.01 0.05 0 0 0 0 E = 0.06 = Px(l)
2 O.Q2 0.01 0.1 0.01 0 0 E = 0.14 = Px(2J
3 0.03 0.015 0.01 0. 15 0.015 0.00 E = 0 .22 = Px(3)
4 0,04 0,02 0.015 0.01 0.25 0.01 E ,= 0.345=flx{4l
5 0.05 0.02 0.02 0.015 0.01 0.10 E = 0 . 215 ~~px(5l

E 0. 150 0. 135 0.145 0.185 0.275 0.110 1.000

Py(O) Py(ll Py(2l Py(3) J!y(4) Py(5)

1 )C

FIG. 3.15 Joint PMF-Example 3.15

Determine the probability of the event E which is defined as the count in

which the number of beams safe both in cracking and in collapse are the
same. (It does not imply that the .same beam is safe in cracking and
collapse).
48

Sol!lfion
P(£) -=-- P( Y "= X) = l' P(X =" x;, y = y;)
= pxr(O, 0)+ pn( I, I) !· pxr(2, 2) +
Pxr(3, 3)+ pxr(4, 4) + p.n-(5, 5)
= 0 + 0.05 + 0.1 0 + 0. 15 + 0.25 + 0. I0
= 0.65
Marginal Distribution
From the joint distribution of t variables, it is possible to get the distri-
bution of individual variables. he distributions of individual variables arc
called marginal distribution he marginal distribution of X is found by
summing over all the valut: of the other variable Y. That is,
px(x) = P X= x] = 2 PXY(X, Yr) '- (3.28)
all .v 1
The derivations of marginal distributions of X and Yare shown in Table 3.2
by adding values horizontally and vertically. Marginal distributions of X
and Yare displayed in Fig. 3.16.
~ ~(x) Py (y)

QI · 04
0 ·3 OJ
02

01

0
0

FIG . 3.16
3
x-
__l
'· 5
02

01

0
(1

Marginul distributions of X and Y-·Example 3.15

ll_
Y---
) 4 5

Fx(x) ~ P[X ~ x] = E px(xr)

X;~X

E PXY(Xt, Yi) (3 .29)

x 1"x all YJ
Similar expression~ can be written for the marginal distribution of Y.
It is to be noted that marginal distributions alone are not sufficient to
define the joint distribution. In Example 3.15 (Table 3.2), there are 36
points describing the joint distribution while two marginal distributions
have only 12 points.
Conditional Distribution
Given two discrete random variables X and Y with values x and y,, the
conditional probability mass function of X given that Y takes on the value
Yi is defined as
· pxr(x, · I)
PXIY(X I Yl) = 0

(3.30)
py(yJ)
 49

The denominator is the marginal distribution of Y evaluated at the given

value 0f yr. For a proper conditional PMF of X,
0 ~ PXIY(X, y) ~ 1
~ PXIY(X~o y) = l
all X;

ExAMPLE 3.16 Consider the previous example and derive the conditional
distribution of X if Y takes a value of l.
Solution The conditional PMF of X, given Y = I, is
_ p,n(x , I)
PXIY (X I I) - PI{ I)

From Table 3.2,pr(I) = 0.135

pxr(O, 1) 0.02
PXIY(O, I) = 0.135 = 0.135 = 0.148
pxrO, 1) 0.05
PXIY(I, J) = 0.135 = 0. 135 = 0.371
? I) pxJ(2, I) O.Ol O
PXIY ( -• = OJJS --= 0.135 = .074
px t(3, I) 0.015
PXIY(3, I)= 0.135 = 0.135 = 0.111
pxr(4, 1) 0.02
PXJY(4, I)= {[135 = O.J) 5 = 0.148
Pxt(5, I) 0.02 0. I 48
PXIY(S, I) = 0. I 35 = 0.135 = L'l.OOO

'The plot of pxw(x I I) is shown in Fig. 3.17.

Px 1v (llll >

04

03

02 ..

x-
FIG. 3.17 Conditional PMF of X·-
Example 3.1 5

Jointly Distrihllted Continuous Variables

1i If X and Yare continuous random variables, their joint probability law is
-\'. described by their joint probability density function, /XY(X, y), defined as
11'
 50

(3.31)

This is the volume under the function over the region.

For a proper joint PDF, the following conditions are to be satisfied:
(i) (yy(x, y) ?: 0 for all values of x, y

(ii) J:<X) J:<X) .fyy(x, y) dx dy --: I

The joint PDF is also given by

o2
r:n{x· '·v) = Bxoy
·
';}""1)""Fxr(x J')
'
(3.32)

where Fxr(x, y) = P(X ~ x; Y ~ y)

The probability density function of one variable, i.e. marginal density
function can be obtained by integrating out the other variable. Symbolically,
it is written as

fy(x) ~ J:oo fi(Y(X, y) t~l' (3 .33)

The marginal CDF of X is obtained as

Fx(x) ~P(X ::::;; x) =- rc.o t:y(x) tf.,· 0 34)

= Fxr(x. co)
The conditional PDF of X, given that Y has taken a value ,1'1. is def1ncd

· ( ) fxr (x · Yt )
fx1r xlyt = fy (y,) 1.3. 15)

The conditional CDF of X is defined as

F.lw(.\· 'yd =-= f."' /ltrL\·! yl) dx \.1 . .16)

fi~AMI'l.l 3. 17 H two random variables. X and Y, have a join! di~11 ibution

given by
() < < 2
l~
O<x<l l'
0\')'
/n(x,y) ·
elsewhere
determine (i) the joint CDF of X Y, (ii) the milrginnl distribution ol' A.', and
(iii) the conditional PDF or Y.
Solutio11 til The joint CDF of XY is

1 1 l Lr. I') Lt ( 1 . ,1') dx dy

II·· r< 1: O· : y<2

 51

) (ii) The marginal density function of X is

.fr(x) = J: /H(x, y) dy

= J: X)' dy
= 2x 0 < x <
(iii) The conditional PDF of Y is

fl'ix(y I x) = fn(x, y~
.fx(x)
X)' )'
=2x= 2 0<y<2
Independent Random Val'iables
Two random variables are independent if and only if
FXY(X, y) = Fx(x) · Fr(y) '-~'- (3.37)
,,
fur all values of t~.e random variables for which the respective functions
are defined.
Two discrete variables X and Yare independent if and only if
PXIY(X I y) = px(x)
PXY(X, y) = px(x) · py(y) (3.38)
Two continuous variables X and Yare independent if and only if
/XY(X, y) = .fx(x)(y(y) (3.39)
/YiY(X I y) = .fx(x)
.fn y(y I x) = fr(J•)
FxiY(X I y) = Fx(x)
The assumption of independence of two events permits one to get a joint
distribution from marginal distributions.
In the case of jointly distributed variables, only two variables have been
considered; however, whatever that has been done can be extended to
multiple variables.

3.4 FUNCTIONS OF RANDOM VARIABLES

Civil engineering problems often involve the functional relationships, which
predict the value of one variable (dependent) from the value of another
basic (independent) variable. For example, (i) the lateral pressure on a wall
is a function of the density of water and the level of water in the tank,
(ii) the intensity of wind pressure is a function of the drag coefficient and
the square of the wind velocity. If the basic variable (wind speed or the level
of water) is random, the depend-ent variable (lateral pressure or the intensity
of wind pressure) is also a random variable. This section deals with the
detennination of the probability law of orie variable from the other.
52

Functions of Sing~Ranilom Va.·iah/e

Consider the case of one random varia hie X. It is given that
Z =-= g(X)
Where g(X) is a monotonically increasing function and Z is a single valued
function or X. For such a function, = ;.: ,: zt if and only if x ;.::,: x,, namely
that the value for which zt = g(x,), as shown in Fig. 3. I 8, or when Z = z,
x = g- 1(:) where g- 1 is the inverse function of g, then
P[7. = zl = P[X ,_~ xl
= P[X = g- 1(z)]

FIG. 3.18 Relation between random variable X and random

variable Z

Hence if X is a discrete random variable, the PMF and CDF of Z are

givett by
pz(z) = px[g-l(i)] (3.40)
Fz(z) = E px(.,·;) / (3 .41)
al'.<; .;; ~-l(z)

If X is a continuous random variable, the CDF of Z is

Fz(z) = P[Z ~ :] = P[X ~ x]
= Fx[g- 1(::)] .

Hence
g-1(=)

Since
Fz(z) ::=.'
f=
x
-co
fx(.,·) dx .·'/

g- 1(z),
(3.42)

d[ u-l(z)]
dx = od; dz

Now Eq. (3.42) becomes

f7.(z) ="' fx fy[g-l(z)] I[R~' (z)] dz

-«> z
 53

Hence the PDF of Z is

d
fz(z) =' d::: [g- 1(:::)]Jx[g- 1(z)]

~ (dlxz .f:r[.rr '(::)] .,. ;

..........-
In general, the above equation is written as
dx ~-" ------
(£(::) = - fx[g-- 1(z )] ; (3.43)
-· - I dz · --~
J

The absolute value of dx/dz is necessary since for some functions g(X), a
positive dx corresponds to a negative d: and vice versa (i.e. the function
may be a monotonically decreasing function).
Note: If each value of z corresponds to 11 values of x, i.e. the inverse
function x = g- 1(z) is multivalucd, then

/z(z) = I~~ lnf.,-lg-l(:)j (3.44)

·AMPLE 3.18 A column is to be designed for a load W which is equal to

its self weight sand a fraction of the live load Lon the beam supported by
the column. That is,
W = s -1- cL
where c is a constant (positive). Assume that Lalone is a random variable.
Find the PDF of W if the PDF of L is

/L(/) = .} - exp (-/2/2) I~ 0

v 277
Solution When
W=11·,
I = (w - s)/c = g-t(z)
dl l
dw = -;
Using Eq. (3.43),

I
c .(L
Jiv(w) = - (II'--
-c-s)

lV ;;;.: S

The CDF of W is

I
ll'

Fw( w) = , fw( w) dw
54

Tim can also be obtained if the CDF of Lis given:

Fw(w) = P[W ~ w]

=-~ p [L ~ w ~ s] = FL (w ~- s)
In case, in a particular physical situation, the relationship is
W= g - cL
then,
l=g-w_ dl
c , dw = -- ~

The PDF of W is

fw(w) =I+ j.h (g ~ w}

and the CDF of W can be obtained as follows:
Fw(w) = P[W ~ w]
=--= P[g -- CL ~ IV]

'·"' p [ L ~
g-
--c- w]
. (g· --;;-
'"= 1 - - FL
... IV)
ExAMPLE 3.19 In Example 3.18, if

- 1<1<+1

elsewhere
what is the PDF of W?
Solution From Example 3.18,

/iv(11') = + (w /L ~ s)
1
_, - . s - c <
2c
11' < s +c
Skcl\.:hes of/i.(/) and /iv(ll·') arc shown in Fig. 3.19.
fw(wJ I

- _,
FIG. 3.19
112

( )_ c)

Probabil1ty density !unctions of Land W-Example 3.19

b
· ( s +- c_,),---w
EXAMPLE 3.20 The intensity of wind pressure, W, is given by the relation
W = 01:V 2
where 01: is a constant (equal to 0.006 as per lS code) and V is tne annual
JnllXimum wind speed. If the PDF of V, following Type II extremal (largest)
distribution, is given as

fv(v) = -k ( - u )k·il exp [ -(ufv)k] v ;;.: 0

u v
determine the PDF and CDF of W. k and u are parameters (constant)
of' V.
Solution When W assumes a value w, then

v = ± (: r2
Idwdv I = \!1 lXII'

Hence, using Eq. (3.43), the PDF of W is

.fiv(W) = 1-2V!'I.w
l -I (/v( VtviOC) + /v( -- y' 1; /IX )J 1 (3.45)

For the given PDF of V,

fv(v) = 0 for v < 0
Hence Eq. (3.45) becomes

fiv(w)-=
2
J IY.!II
fv('\lw/t:~.)

- -
1
-- 2 VlXlV
--(!.(-V 11 - II i1•/1X
)k+l exp {-(u/V wf-x )k}]

Let
Then the PDF of W simplifies to

/iv(lv) k
== --- ---E
(II' )k/211 cxp [ -·(11·,/u•)k'~J w~O (3 .46)
2wc IV

(Note that W also follows the Type IJ extremal (largest) distribution with
parameters We = cu 2 and k/2).
The CDF of W is derived as follows:

Fw(w) '-" J~· Jiv(1v) dll'

Substituting for .!iv(w) in the above equation and putting (lvc/w) :;= y,
we have
_kL_
J
l'
Fw( 11') = "' 2
I' exp [ -- ykl 2] (~V
56

Let X = y"/2
then

Fw(w) = - f~ e-x dx = c-"

Hence the CDF of W is

Fw(w) = exp [ -(w)w)"ll] w>O (3.47)
The CDF of IV can be obtained by directly using the CDF of V, which
is given by
Fv(v) = cxp [-(ufu)k) v ~ 0
Fw(ll') '-'-' P( W G 11')
= P(rt.V 2 ~ w)

= p(v ~ /'v)
"~
+J>(v ~ - ~
/2~}
~

The second part is equal to zero as V c; ;,,Jot take a negative value. Hence,
(using Eq. 3.47)
Fw(w) = P(V ~ y~-;j;j

Let ll'c ~= rt.u 2• Then,

Fw(ll') =-= cxp [ -·(wc/w)k/2) w~O

ExAMPLE 3.21 Given

Z =a sin X

(x(x) ~ {2~ 0< X< 21T

otherwise
find the PDF of Z.
S olution Equation (3.43) bas been derived on the assumption that Z is a
s!n r.,IL- valuec.J function of X. In this case X is a double valued function for
each value of Z. Hence for such a function, Eq. (3.43) becomes

fz(z) = 2/x[g- 1(z)] ·Idzdx I

In general, if each value of:: corresponds to 11 values of x (i.e. the inverse
function or.\ ,-_ gw 1(:) is mullivalued), then
 57

for the given function,

x =sin- I (:)

dx
- a < z <a
dz = va?. - zl

Hence,

= ---;-~=:::=- 2/x [sin- 1 (_!_a-)]

1
/z(z)
V a2 - z2

- -V---:a=::2='=
:_ z2:::= U~)
1
/z(z) =
ya2 - z2
(_!__)
~
-a < z <a

Functions of Two Random Val'iab/es

In many situations, nn engineer may have to deal wit h cases where one
vari able depends on two or more variables. For example, (i) the tot<t l
momen t induced on a column may be the sum of the moments du e to li ve
load and wind load. Since live load and wind load are random variables, the
total moment on the column is also a random va riable. One has to derive
the PDF of the total moment from the known di stributions Wtnd -load of
and live load. (ii) Strain in a tension member is the ratio of the force in the
member to its area of cross section. Jf area and force are random variables,
strain is also a random variable whose PDF is to be obtained from the
known distributions of force and area. In general, functional relations may
be of the following types:
Z=X -1 Y Z = X -- Y

Z=; Z=XY

Case (i): Z = X Y + (3.48)

Let Y take a particular value y, i.e. Y ~ y. Then,
Z=X+y
The conditional PDF of X, given Y = y, is
. ( f.VI'(X , J) ./
.I xir x, y) c." · .fr(y) ,_:/

Treating y temporarily as a constant,

x=z-y and ~=Ill

dz
Hence,
/7. ~ r(z, y) -= I I ! .h I I'(~ ·- · y , y) -f
= /Zii'(Z ·-- )',}') (3 .49)
 58

But it is known that

/zl (z, y) ='-' .fL 1y(Z, y)/i·(y) (3 .50)
u~ing Eq . (3.49) in Eq . (3 .501, the joint distribution of ZY is obtained as
/z1 (::, y) -= [x 1 r(z - y, y){l'(y)
~-= fxr(z - )', y) (3 .51)
fxr(z - y , y) is nothing but the joint probability of X and Y evaluated at
X ~~ : --- y and Y = y.
From the joint distribution lEq. (3 .51)], the marginal distribution of Z can
be obtained as ---

fz('::.) = r"' /Xl'{: __ y. y) dy // (3.52)

When X and Yare independent,

fz(z) = ]:., fx(:: - J'!fr(y) dy ,

Similarly for other cases, the marginal distribution of Z can be obtained .

Case (ii): Z = X - Y

fz(z) = J:., !xr(z + y, y) dy (3.54)

Case (iii): Z= X
y

fz(z) = J:., I Y I fxr(zy, y) dy (3.55)

Case (iv): Z'-"XY

/z(:) = J:J ;, I fyy (; , Y) dy (3.5n>

Notf!: All the above equations, considering X and Y as continuous random

variables, are valid for discrete variables also, keeping in mind that instead
of integ ra ti on, summation is to be carried out.
~AMJ'LI 3.22 Let the stress in a member, X, and the area of section, Y.
be ind cr cmlcnt random variables. The force Z in the member is then
given by
Z=XY
It is given that

fy(x) = g1 x O~x~4

fr(y) -.- __!__

a
Determine the PDF of Z.
 .•.
• 59

,..,'o/ution Since X and Y arc indepenJent, Eq. (3.56) becomes

h1r the given PDF of Y,

//.(z) '· J." _..!_y -(y(.!..)-

0 y u '
1
- dv

In the above equation substitute for fx ( f,) using the gi\'Cil /r(x). For the
relation_:_ =--= x with limits for x, 0 ~ x :( 4 implies that 0 ~ .-=-- ~ 4.
y y
Corresponding limits for y arc
z .
4 ~ )' ~ 00,

Hence,
1 1 (.,) I c/1'
(z(z) =
f a
::4
- - _:_-
y 8 y a ·

~ ~J~4 .1~ dy
~~ 2H 1 -- ;J 0:(z:(4a~ /
ExAMPLE 3.23 A water tank is supplied with water through an inlet pipe
at a constant rate for a period of time X. The water flows out through the
outlet pipe from the tank at the same rate for a period of time Y. If X and
Yare independent with distributions
/x(x) ~ .\e-x~ x ~ 0
JY(y) = {3 c-fly y ~ 0
determine the PDF of Z """ X -- Y, the change of the amount of water in
the tank after one cycle or inflow and outflow, assuming that the tank can-
not become dry, or overllow.
Solution J t is given tlwt
Z ==X-- Y
Since X and Yare independent, Eq. (3.54) becomes

/L(:) -'= J~.,.- /x(z ·i· y)/r(y) dy

Since jy(y) = 0 for j• < o..

fz(z) = J~ /x(z -1- y) /r(y) dy

;·" J"' f.-..:(z + y ) {3 e-fl.v dy

0 .•
60

In tile above equation substitute for /r(z + y) using the given /x(x).
Since f'<(x) = 0 for .Y < 0,

t:~(:: + yj is zero for z +y < 0

.For : < 0, y shouiJ be > -- z.

/z(z) = ]:, A e-A(z-l-yJ ~ e-#y dy

= ( 3!._ ) e~' :: < 0

"- + P
For z > 0, _I'> 0,

Hence the PDF of Z is

~ } cP'
fz(z)
{( " + f3
= (_M_ ) e--Az
:: < 0
;; > 0
" +fJ
3.5 MOMENTS AND EXPECTATION
The complete description of a random variable requires a probability
uistributi n in one of' its arious forms. However, in many applications,
the form of the di lribution funcli n i not known in all details. In such
situ ati ons, c ncise d crip tor whid1 describe the dominant features of the
function may be valuable, and enou •h for engineering applications. These
descriptors may be expectation (mean), variance, etc.
The expected value of a discrete random variable X, denoted by E(X), is
dclincu as
E(X) = 2; Xt px(Xt) (3.57)
all x 1

If X is continuous, then

E(X) = [:., x fx(x) dx (3.58)

The same quantity, L(X), is also called the mean of X or the first moment
of the dtstributio u of •. This h uld not be c nfu ed with the am pl e
mean whic.:l1 is · Hnputcd fr m the data and ba · ·tatistical entity. An
expcctati n is <'~tkulatcd from th e pr babilily distribution . lt can be c n·
siJercd u. til ..: w..:i)!.hlcd avl:rag..: < r th • values of X in which each possible
value is wei ht..:d by the pl'l)habilit of its currence,
 61

A family of averages, called moments, define the probability distribution

nf a variable as

m1 ~¥ 1 ) = J:"' x 1fx(x) dx 0.59)

where m; is the i''' moment of X about the origin . The first moment is the
mean value of X and is designated by f.'. That is when i = I,

mt = E(X) = ft = J:., x.fx(x) dx

tn the mean and are called

(3.60)

The first four moments are commonly used. The first centrnl moment is
zero. The~e:9~~ central. m~~,~~~!_!~~,~~!· iancc, given_hy
Var(X) = c2 = E[(x - ftFJ

= J:., (x - 11F f-,:(x) ~ (:Ui I)

The third central moment is related to the symmetry of the distribution

and is incorporated in the dimcnsio~ less coefficient of skewness, n, given· fiy -
(,·,_:_v
~, --a;
(3.(i2)

where a is the standard deviation .

. If the distribution is symmetrical, rt = 0. If rt is positive, the distribu-
tion is called positively skewed and will have a long tail (upper tail) at the
right . If rt is negative, the distribution is called negatively skewed and will
ha\·e a long t~i l (I wer tail) on the left. The varlnti n of the shape of the
density function with n is shown in Fig. ~.20.

FIG. 3.20 Variation of PDF with coefficient of skewness

The fourth central moment is related to the flatness is the coefflG.ient of

kurtosis, r2, given by
C4
rz = a4
- 0.63)
It is often compared to a standard value of 3 for normal distribution. Ir
r2 > 3, the distribution is said to be flatter, and if <3, the distribution is
more peaked than normal.
62

ExAMPLE 3.24 The lateral strength, S, of a frame is subjected to random

variations . The PDF of Sis given as

~ (2s -- 1) (2 -· s) J<;s~2
{s(s) o"' -
{
0 elsewhere
. Determine the mean and vnriancc of S.
Solution Th e mean value of Sis obtained using Eq. (3.58)

1~ =- J: s[-~(2s- 1)(2- s)] ds

= ~ [5;3 -- -s42 - s2]2 I

7
= -
5
The variance of Sis given by

Var(S) ~~ f[s ~ r[ ~ (2s - I) (2 -- s)] ds

G
=- 100
Algebra of Expectations
The expectation has a number of convenient properties which are useful.
For any random variable X, and constant a,
E(aX) = a E(X)
E(X + a) ""' E(X) +a /
For any two random variables X and Y,
E(X + Y) = E(X) -t- E(Y) "/
Expectation is a linear operation and the expected values of a sum is the
sum of the expected values. The same relation can be extended to sums of
more than two variables, and is · valid whether the variables are indepen-
dent or not. Whenever X and Y are independent, the expected value of the
product will decompose, that is,
E(XY) = E(X) E(Y) .
The above relation is not valid when x·and Yare dependent.
The expectation of an arbitrary function of a random variable X is
easily expressed. If X is discrete, then
l:'[,t:(X)l ~ 1) g(Xt) px(xt) ,. (3.M)
:til X1 1

and if X is continuous, then

"' .
E[g(X)] 0
"'

J -00
• g(x) /x(x) dx
' ../
I (3.65)
 63

In other words, g(x) merely replaces x in the definition of expectation. These

expressions are not a new defin ition, but are derived by considering a
rnndom variable Y ~-= g(X) nnri relating the distribution of Y to the distri-
bution of X. Tfg.(X) nn.d g2(.Y) are any two functions of X, then
~fg.Cn + g2(X)] 7
= ~[gl(Xl] + £L1!'2(X)l-~
Contlitional E.rpectatio11
The conditional expectation of a random variable X, given the value of a
related random variahlc Y, is defined as
E(X I y = y) = E
all X;
.\'; (p.-.:IY(;;
__ _ --
I;.y (3.M)

when X and Yare di_screte, and

E(X I y ~= y) __ r"' .\"~ f.\ (.1 .67)

when X and Y are continuous.

If X and Yare independent, then
E(X Iy = .rl ~~ I:(X) - ~­
The expectation of marginal distribution of X is
[(X)-=--~ E E(X I y ,.---, y) py(y)
:til,.

when X and Y arc discrete, and

0.69)

when X and }'arc Ct)ntinuous.

A brief way to express these is
/:"(,\'} ~= £ [ £(X J )" o.:: y)l
Note: E[X ; Y ""-' _rl is a constant and
E1 X\ Y] is a rand</lm variable.
Propel'tie.~ of Val'iancc
As giYen earlier [Eq . 0 .61llthc variance of .X is f!ivcn by
2
Var(X) = E[(X -,u) ]

..,.= n X2 + ft~- 2Xp)

=' E(X 2) -\- E(!-1- 2 ) - - 2tL E(X)
-= £( .¥2) -! tt'l -- 'Jpl
Hence the variance of Y is
Var(X) = a~' = E(.''(2) - 1-1-2 (3.70)
or ( r:(xi)
__
....
·~-- ~; --r:·~
_____ .............
-- -- -~- ---

The linearity property of expectation is not valid for variances.

 64

Jr a and b are constants, then

(i) Var(a) = o
(ii) Var(aX) =c= a2 Vnr(X)
(iii) Var(a + bX) = h2 Var(X)
The conditional variance of X, given Y, is defined [-lS

Var(X I Y ""' y) = E[(X --- JLXtr) 2 1 Y = y] (3.71)

For discrete X and Y,
Var(X I Y =' y) ,.--, I: (x,- - f'X t}')2 P>.w(x; I y)
all .\";

For continuous variables X and Y,

Var(X \ Y = y) ~- J'·
-00
(x - - JLXtYf (r;rC\ ! y) dx

'file t:once.pts of expectation and moments can be extended to jointly dis-

ribuled random vnriahles. 1f Z is a function oftwo continuous random
( variables X and Y, i.e.
7 = g(X, Y)
then the expectation of Z is
E(Z) = E[g(X, Y)]

=
f"' -rn I"'
-co g(x, y) fn(x, y) dx dy (3.72)

The joint moments of the order of m + 11 of a joint distribution of X and

Yare defined as

E[X111 Y"] =

The central moments,

reo r. x"'y".fn(:x, y) dx dy

c,,, are similarly defined. Thus,

(3.73)

c," = Er(x - /.I.J)"' (Y- M)"J (3.74)

where /l! and /L2 correspond to the first order moments obtained by putting
(m c' I, 11 -~ 0) and (m = 0, n = 1) respectively in Eq. (3.73). That is, for
example

''l =I"" J:""

-oo
xfn(x, y) dx dy

x[f:oo fxr(x, y) dy dx 1
x fx(x) dx "'""' E(X)

Si 1ilarly, 1'2 = I~(Y)

Co1•ariance and Correlation Coefficient
The central moment obtained by putting m = ).and n = I in Eq. 3.74 is
 85

called the covariance of X and Y.

Cov(X, Y) = axy = E[(X- t-'1) (Y - t-'2)]

o--, J:, J:«> (x -- ftl) (y - t-'2) fxr(x, y) dx dy

The above expression can be shown to be equal to

Cov(X, Y) = E(XY) - E(X) £'0:1 (3.75)
Relating to mechanics, the variance corresponds to the moments of inertia
about axes x andy passing through the centroid of a plate and the cova-
riance corresponds to ihe product moment of inertia with respect to the
uxes x and y, mentioned earlier.

r
The correlation coefficient, a dimensionless quantity, is obtained by nor-
lalizing the covariance with tandar~l deviations of the corresponding pair
f variables. That is the correlation coefficient between the variables X nnd
' i. defined as

Px r - Cov(X, Y); : (3 .761

O',yOy

Some important points about the correlation coeflkient Pare:

(i) The value of P lies bet ween ·- I and +I, i.e.
-· 1 ~ p ~ -1-1
,'ii l When P is between 0 and I , th e higher values of · will match with
the h' •her va lue nr y [Fig. J.2 l(a l. Variables arc positively c rrelated .
· i When Pi between - I and 0, lhe higher values of · will match with
' the smaller values of J Fig. rJ .2 1( b)l. Variable. u rc nega tively correlated.
(iv) p is a measnre of the linea r <lep ndence between two varinblcs.
v) If 0 < P < I or - I < P < 0, it is said th ut at least some depen-
dence exi l between X and Y.
_ / / 1) lf P is close to I, it is said thot a good linear relationship exi ls
Jctwccn X and Y.
) vii) lf P -· I or - I, it is said that there is a perfect linear relationship
between and Y . [Fig.. 3.21(c) and 3.2J(d)l.
(viii) If X and Y arc independent, P = 0.
(ix) If P -= 0, it does not mean that X and Yare independent (unrelated) .
Jt means that the linear relationshi doc. not exist between X and Y. hut
tl-;ere;Tiay he a perfe t non 1near relationship (depen( ence) b~tw·e~~; X and
YTFig~ T2T(c) . -- -··
(x) In engineering problems, the independence of variables is assumed
(i.e. P = 0) to simplify the problem.
Mean ami Variance of Functions of Variables
If Z is a linear function of variables X1, X2, ... , Xn, say
II

Z = E h;X;
i~1
66

v 0</<1 y -1<.!< 0

v f~O
•
• • •• • •
• • •• •
• •
•• •
• • ••
• • • •
~----------------·- x

(e) Zero Correlation; But lfl Zero Correlation;

Nonlinear Relationship No Relatfonship
FIG. 3.21 Significance of correlation coefficient

then the expected value of Z is

II

£( Z) =~ 1: b; E( Xi) (3 77)
,_,
The ab q· relation is valid whether variables X; are independent or not.
If X; art' corrc laled, then
II fl II

Var[Zl ,.." L' b~ Var(X,) " L' 1' h;b1 Cov(X;. Xj) (3 .78)
i-· ·1 i -- I }>i
 67

If X; are independent, the above equation simplifies to

Var(Z) = E" b;2 Var(X;) (3.79)

1-1

When X1 and X2 are independent, their expectation of the product will

decompose. That is, if

z = x.xz
then, E(Z) = E(X•) E(Xz) (3.80a)

(3.80b)

and
whert: ft;, a,
and 8, are the mean, the standard deviation and the coefficient
of variation of X1 £espectivcly.
If Z is a nonlinear function of several variables the approximate meanx,,
and variance ol' Z are obtained by using Taylor's series expansion and
truncating the series to the required approximation. If
Z = g(X1, X2, ... , X,.)
the first order approximations_of E(Z) arutYar(Z) are given by
. ----
(3.81)

Var (Z) = oz, ~~~" /!." ax,

ag ,. ax,
1
I I
ag ,. Cov(X;, Xi) (3.82)

::, Lmeans that the derivative is evaluated at the mean values of the
yaria b~s.
If X; are uncorrelcted, then

I·
Var(Z) = 1 ~" [aai, I,.]2 Var(X;) )· (3.83)

~AMPLE 3.25 A simply supported beam is subjected to loads P!, P2 and

PJ as shown in Fig. 3.22.

'
f h _, f-- '-'4 -
~ ~ --+-- - ~ -
I 2 2

FIG. 3.22 Simply supported beam-

Example 3.25
88

It is given that
E(P,) = 20 kN Var(PJ) = 2 (kN)2
E(P2) = 40 kN Var(P2) = 4 (kN) 2
E(P3) = 50 kN Var(PJ) = 10 (kN)2
Determine the expected value and standard deviation of the shear force at
the left end if (i) loads P1, P2 and PJ are statistically independent and (ii) if
loads are correlated with correlation coefficients
Pl2 = 0.7 P23 = 0.8 PJI = 0.6
Solution The shear force Vat the left end of the beam is
V = 0.75 P, + 0.5 P2 + 0.25 PJ
The expected value of y, using Eq.(3.77) , is
E(V) = 0.75X20 + 0.5X40 + 0.25.><50
= 47.5 kN
Case (i) Loads are independent
The variance of Vis calculated using Eq. (3.79):
Var(V) =- (0.75 2)(2) + (0.5 2)(4) + (0.252)(10)
= 2.75

The standard deviation of Vis equal to = V2.75 = 1.658 kN.

Case (ii) Loads are correlated
The variance of Vis calculated using Eq. (3.78). Defore using Eq. (3.78), the
covariance between the variables is to be calculated. The covariance is given
by Eq. (3.76):
Cov(P,P2) = P,2a1a2

= (p12) [Vai(PJ)] 112 [Var(P 2)11.' 2

= (0.7)(2 112)(41/ 2) ,... 1.98

Similarly,
Cov(P2P3) = (0.8)(4 112)(1011 2) = 5.06
Cov(PJPI) = (0.6)(10 1i 2)(21t 2) = 2.68
The variance of Vis
Var( V) ,-c (0.75 2)(2) + (0.5)2(4) + (0.252)(10)
--f-2[(0.75 X 0.5)( 1.98) + (0.5 X 0.25)(5.06)
+ (0, 75 X 0.25)(2.68)]
c- /..75 + 2(2.763) =6.505
The standard deviation of I· is equal to 2.55 kN.
 ••
3.6 COMMON PROBABILITY DISTRIBUTIONS
There are a number of discrete and continuous probability distributions
which are used in engineering applications. It is always convenient to have
a mathematrcal functiop (PDF or CDF) to describe a random variable.
Before an engineer u es or proposes a probability distribution (probabilistic
111 del) it is necessary and better that he knows how these models have arisen
and wh;H physical situation has given ri e to the distribution. Many of the
· rnm on distribution are tabulated for convenience and ready use. Out of
the several probability distributions, only some of the models which are
often useu in reliability analysis and design of structures are dealt with. The
other models which are not discussed are tabulated at the end.
J niform Distrih11tion
Thi i a continuous distribution. Here the random variable X is equally
likely to have any value between the lower limit I and the upper limit u.
The PDF of X is given by

=~
1
/x(x) - I
{ -u --- ICx~u (3.84)

0 elsewhere
The mean and the variance are
JL "··'-"
I+ u
-2-

a2 = -
(11 ·- l)2
..,..-,---'--
12
When the uniform distribution is described between the JimHs 0 and I, it is
c:.~lled th e tandard uniform distribution. In the case of the standard
unif, rm distribution, th cu mulative probability of the variable Y, taking
a value Yl, is equal to the value of Yl itself. That is,
Fy(y,)= y,
Thi properLy is u ed in the inver ·e Lransformation technique applied for
generati ng the random variates (Chapter 7). The sketches of the uniform
probability di tribution and the standard uniform distribution are sh wn
in Fig. 3.23.
fy (y)

1
u:r

y
(a I (b)
FIG. 3.23 (a) Uniform dtstribution and (b) standard uniform distribution
70

Normal Distribution: N(p,, a)

If a phenomenon (a random variable) arises because of several factors and
if the effects of these several factors act in an additive way to result the
phenomenon, then the model arising out of such a situation will be a
normal distribution. In short, this model arises out of an additive mecha-
nism. This distribution is also known as the Gaussian distribution. The PDF
of a normal variate is given by
2
fx(x)
·
= 1
uy2 77
exp [-- .!.(
2
x -- !L
a
) ] -oo:::;;x:::;;oo (3. 85)

where p, and a are the parameters, mean and standard deviation of the
distribution respectively . ln futme, this distribution will be designated as
N(p,, a).
A normal distribution with parameters p, = 0 and a = l is called a
standard normal distribution and is designated as N(O, 1). The PDF of the
standard normal variate U is given by

fu(u) = V~ 77 exp( - ~ u2 ) -·-oo ~·u ~ oo (3.86)

Because it is so frequently used, the standard normal density function and

its CDF are given special notations, ¢>(u) and (})(u) respectively. Hence 1/>(u)
is the cumulative probability of a star.dard normal variate. That is,
1/>(u) = Fu(u) = P(U :::;:; u)
The PDF and CDF of U are shown in Fig. 3.24. Referring to Fig. 3.24,
we have

'u tul

0 02275

-J 3 u
(-J ()) (3 (1")

FIG . 3 .24 Standard normal density function

Conversely, lhc value of 111 at a cumulative probability of Pt is given by

The standard normal v:11 iate is widely tabulated. It is to be noted that

</J( II!) ~ I -· r/J(u2)
 71

If <1>(- u2) = p2, then

Ul =- ' -- rJH (p2)
because of symmetry. The CDF of X with distribution N(p,, a) is written
as

Fx(x) = ~j·'
(]v'21T - <X>
exp [- J.(x-
2 (]
P-} ]dx
2
(3.87)

X .... (J-
Let u=

Then du = dx/a
Using these in Eq.(3.87) , we have
I J(x-,.)/" exp ( -u2/2)du
Fx(x) = -
·v1 br -oo
A

(x-p,)
= <J) - ( ] - (3.88)

Hence, using normal probability tables, probabilities of any other normal

distribution can be obtained. Modern computers have built-in functions to
compute <P(u). A polynomial is used to evaluate <J>(u).
AxAMPLE 3.26 The cube strength of concrete, X, follows the normal
distribution with parameters , p, = 30 N/mm 2 and a =--' 4.5 N/mm 2
Fig. 3.25). Calculate the probability of getting a value for a strength
(i) less than 25 N/ mm 2 and (ii) less than 40 and greater than or equal to
30 N/mm 2 •
X is distributed as N(JO, 4.5).

FIG. 3.25 PDF of X-Example 3.26

Solution (i) The probability or getting a value less than 25 N/mm 2 is

25 30
P(X < 25) = Fx(25) = <1> (
4
-:-; )

= <1>(- - I. 11)
=-I ·- tf>(l.ll) ~I- O.XCi67 =-= 0.1333
72

(ii) The probability of getting a value less than 40 and greater than or
equal to 30 is
P(30 < X < 40) = Fx (40) --- Fx!30)

= tP( 404.5- 30) -- !J> ( 304.5- 30)

= 0.9869 - 0.5 = 0.4869
Some properties of normal variables are:
(i) The distribution is symmetrical; hence the coefficient of skewness 1s
zero.
(ii) The mean, median and mode are the same.
(iii) The coefficient of kurtosis is equal to 3.
(iv) The normal distribution is reproductive, that is the sum and the
difference of two or more normally distributed random variates is itself
normally distributed.
If Z = X1 ± X2 ± ... ± X, and X; are independent normal variates
with parameters /LI and a,, Z is also a normal variate with parameters J.l.x
and az, given by

f.'z = /LJ ± f.'2 ± ... ± """

a~ 0~ -1- a~ - \
. 2
c_; ••• -j (J!l

If X; arc conclated, then

cr; = fa?
i --1
+2 f f Cov (x; ,XJ
i ,j j>\

At this stage, it is very useful to know the remarkable result estnhlishcd

by the Central Limit Theorem which says, when stated loosely, that the
sum of a large number of arbitr.;rJiy distributed ranuom variables wi!i te-nd
to be normalJy distributed. Hen-.:c:, physical process which i.' the result of
the combined effecis of several factors (irrespective of their individual
distributions) woulll tend to be normally distributed.

Lognol'mal Distribution: Ll\.(Z, a1o z)

This model arises out of .., multiplicative mechanism acting on <1 number of
factors. Such mechanisms are expected to occur in the crushing of aggrc·
gates ami the Jittigue strength of materials. Let the random variable
X= lnZ (3.89)
be normally distributed with parameters N(p.x, ax); then the random
variable Z is said to follow the lognormal distribution whose PDF is
given as

cxp [- 2-'{lnt:/Z)}] <IJu 7

::>0 (3.90)
 73

where Z, the median of Z, and UJn z , the standard deviation of In Z are

the paran1.:ters of the distribution. This distribution is designated as
LN(Z, UJn z). The parameters ar~ calculated by the following equations:

Z- = p.z cxp ( - 1 2 z)
2a'." (3.91),.........

and '
aj, z = In (8z2 + I) (3.92) /
where 8z is the coefficient of variation of Z. The cumulative probability of
a lognormal variate can be calCulated using standard normal tables r/J(u).
Let
u = -I- In (z/Z) -
UJn Z

1
du = - -
. - dz
ZUJnZ

Substitution of the above values in Eq. (3.90) yields

I Fz(z) =. 1J- fu exp (-tN2) du

v 27T -<»

= r/J(u) = rp[ In (z/Z) ] (3.93)/

· Utn Z
Similarly the PDF of Z can be connected to the PDF of the standard
normal:

fz(z) = _1_
z
rp [In (z/Z) ]
ZO'J n O'Jn Z
(3.94)

the lognormal distribution for various values of a1n z is plotted and shown
in Fig. 3.26. It can be observed that as the coefficient of variation decre-
ases, the curve approaches the normal distribution.
f z(Z)
l4
JJ. = 1.50 1

, 8

1 2

06

FIG. 3.28 Lognormal density tunctit'ns

 74

vExAMPLE 3.27 The compressive strength, Z, of M 15 concrete follow s the

lognormal distribution . It is given that
flz ""'~ 24.04 N/mm 2 az - 5.76 N/mm 2
•
Determine the probability of getting a strength less than the specified value ,
I 5 N/mm 2 •
Solution The coellicient of variation of Z is equal to
8z =~ 5.76/24.04 =~ 0.24
Using Eqs. (3.91) and (3.92), we have
arn z =' In (0.24 2 + 1) = 0.056
UJn z =-"' 0.236
(Note: For 8z :::;; 0.25,

Z =---= 24.04 exp ( 0.056/2)

= 23.37 N/mm 2
He1ke Z is distributed as LN (23.37, 0.236) . The probability of getting a
value less than 15 N/mm 2 is
P(Z < 15) = rz (15)
= f/J[Ln__ (_I~{f~}_?)
0.236
) = o 03
.
Some properties of the lognormal variate Z are:

(i) If X = In Z, then flx = In Z

(ii) i is always less than fLz
(iii) Z is positively skewed
(iv) CTJn z ~ 8z for 8z ,; 0.25
(v) If Y = Z1 ·Z2· ... ·Z,
and if the Z; are independent and lognormally distributed with para-
.meters z, and CTJn z_, , theu Y is also lognormalfy distributed with parameters
(3.9 ..-:J
n 2 ] l/2
CTJn y = [ ,_,
_E CTJn z 1 (3. 96)

If W is the quotient of the two independent lognormally distributed variables

Zr and Z2, i.e.

W=z'
Z2
then W is also a lognormal variate with parameters

w= z1(i2 (3.97)
11Jn w = [urn zr + afn z2]ll2 (3.98)
 75

Gamma Distribution G(k, >.)

The sum of independently and identically distributed exponential random
variables results in the gamma distribution. If the occurrence of an event
constitutes a poisson process, (Ref. 3.1), then the time until the kth occur-
rence of the event, is described by the gamma distribution. Let Xk denote
the time till the kth event. Then the probability density function of the
gamma variable Xk with parameters k and >. is given by

x~O (3.99)

Hereafter, the suffix k for X is removed.

Parameters k and >. are connected to the mean and variance by the following
equations:
k
!J.x=-r (3.100)

2
ax= -rzk (3.101)

The gamma distributed variable X with parameters k and >. is designated as

G(k, >.).
The parameter k need not be integer valued. For a noninteger valued k,
the PDF of X is written as
fx(x) = >.(>.x)k-le- >.x x~O (3.102)
T(k)
,\ ~ 0, k ~ 0

where (3.103)

The gamma distribution function is widely tabulated as the incomplete

gamma function, given by

F(k, x) = J: e-ttk-l dt

This can be used to evaluate the CDF of X:

Fx(x) = J:!x(x)dx

= ),k J"'o e-~xxk-1 dx

T(k)

Substituting y = >.x, the integral becomes

Fx(x) = T~k)J:x e-Yyk-l dy

T(k, Ax)
= f(k) (3.104)
76

Equations (3.1 00) anu (3.1 0 I) arc valid for non integer values of k also.
The shape of gamma distribution is shown in Fig. 3.27 . This distribution is
widely used because, Iike observed data from many phenomena, the vari-
able is limited to positive values and is skewed to the right. The gamma
distribution is used to describe the maximum river flows, the yield strength
of the reinforced concrete members (3.2), the sustained floor load in build-
ings, etc. For an integer valued k, the gamma distribution [Eq. (3.99)] is
also known as the Erlang distribution. The gamma distribution [(Eq. 3.102)]
is also called the Pearson Type Ill distribution.

FIG. 3.27 Shapes of gamma distribution

The tables for an incomplete gamma function have been tabulated by

Karl Pearson (3.3). This table directly gives the cumulative probability of an
incomplete gamma variate. The algorithm AS32 given by G.P. Bhattacharjee
(3.4) can be used to compute the incomplete gamma function. However,
modern computers have built-in functions to compute the cumulative
probability of an incomrlete gamma variate.
Pearson tables give values of /(u, p), where l(u, p) is the cumulative proba-
bility of the variate. One enters Pearson tables with p ~-= k - I and
u = >.xlvk. and finds the value of I(u, p).
EXAMPLE 3.2!\ The floor live load, X, on an office building is found to
follow the gamma distribution with parameters k and >. being 3.86 and
7.55 :< 10- 3 respectively. Calculate the probability of the floor load exceed-
ing the value 1500 N/m 2 •
Solution The mean und tandard deviation of X are calculated using
Eqs. (3.100) and (3 . 101 ). Thus

JLx ~ ~ = 7 _ 5 ;:~ 0 _3· = 511.6 N/m2

ax= - A-
kv =
'\1'3.86
7.55 X IO· 3 = 260.35 N/m2
 77

The probability of the floor load exceeding the value 1500 N/m2 is
P(X > 1500) = 1 - P(X ~ 1500)
= I - Fx(ISOO)

I _ T(k, i\x)
T (k)

T(3.8 6, 11.3 25)

=I
r cJ .86)
= I - 0.9968 = 3.2 >< I0-3
Some properties of the gamma distribution are:
(i) It can take only positive values.
(ii) It is positively skewc£1.
(iii) Tf x, is G(k, , i\) and Xz is G(b, i\), nnd if Y = x, + X2, then Y is
also gamma distributed with parameters k1 + b and .\.
~a Distribution: BT( a, h, p, q)
Many of the random va ri ables in practice, say the strength of steel or
concrete, take values within ce rtain limits. Under these co ndit ions, the
appropriate probability distriburi onfor a ra nd om ariable whose possible
value. lie in a res tricted interva l, say betwee n limits a nnd h, is th e beta
distribution .
A standard beta distributed random variable, X, is defined over the range
0 ~ x ::;; I. Tts PDF is
xP- I( J _ x}q ~ l
.fx(x) = B(p, q) (3.105)

where B(p, q) is the beta function which is tabulated directly or may be

obtained from tables of the gamma function from the relation

B(p q) = T(p)r(q) (3.106)

' T( p q)+
The mean and variance of X are

p.x= _P_ (3 . I 07)

p+q

a~= pq (3.108)
(p -1- q)'2(p +q+ I)
The standard beta distribution is designated as BTx(p, q). When a beta
distributed random variable, say Y, has a range a ~ y ~ b, the simplest
approach is to transform Y according to

X= Y - a
b- a
78

Then the PDF of Y is

_h(y) =--= _1 _!x(y -

b-a b-a
a)
_ (y - a)P· '(b -- y)H
(3.109)
B(p, q)(b - a)P+q- 1

Fy(y) = Fx ( iJ~a
y - a)

The mean nnd variance of Yare given by

p
p.y""' a+ --(b --a) (3.11 0)
p+q

a~=(b-a)2 [ (p + q)2(p-\-
PfJ
q + 1)
] (3.111)

Depending n the parameters ol' pant.! q, the den it y function of th e bcw

d istributi n wi ll have different hapc as h wn in Fig. 3.2 . Whenever fJ
an I q take nonintegcr v!l lues, the bela fun ction i. cu ll ed th e inco mplete
beta function. The ·umulative pr babi lity of the incomp lete beta fun tion
is tabulated by Pear on {3.5) us .Bs{ p, 11 . l-Ienee, Pearson's tables can be
met! to ~aku l ate the cum ulnti e probabi lity of a beta variate Y. 1t mu t be
noted tlwl the ttLb le are given for p ~ q. For p < q,
Rlp, q) = I - Bo-,>(q, p)

,P= '1, q :4 p:4,q:: 2

0·20

0-15
p:1 t q:1
0 ·10

0·05

0 o~----¥----c~----+-----+-----~----~----~r

FIG. 3.28 Shapes of beta distribution

For examrle, if p = 2 and q = 4,

Bo.3(2, 4) = I - Bo.7(4,2)
Modern co mputers have bui lt-in fu ncti n to compute th e cumulative proba-
bility of an in ·om plctc hcta ariute. The nlg rithm AS 63 given by
Maj umd er and Dlulltacharj ·e (3 .6), and modified by Cran, Martin and
Th ma (3 .7) . ·an he u~cd to eval uate the incomplete beta variate.
E XA MP U3 3.29 It is given that the strength Y of M 35 concrete fqll ows the
beta di stribution. The 111'1111 and str1ndard deviation of Y are 42.50 N/mm2
and 6.25 N/mm 2 rcspc~: t iv cly. lt is,. und from data that the minimum and
 79

maximum values of Yare 30 N/ mm 2 and 55 N/mm 2 respectively. Calculate

the probability of the strength of concrete being less than 35 N/mm 2 .
Sol11t ion It is given:

J.Lr "'"' 42.5 N/mm 2 ay ~ 6.25 N/mm2

2 b =-= 55 N/mm2
a= 30 N/mm
Using Eqs. (3.110) and (3.111), we have

4~ . 5 ,...., 30 + p__-f!_t-__q(55 - 30)

6.25 2 = (55 -- 30) 2 [ - - --· ·- ___ pq_ _· -- . ]

(p + q)2(p + q +I)
Solving the above two equations, p and q are
p = 1.5
Hence the strength of concrete is distributed as BT (30, 50, 1.5, 1.5). The
· probability of the strength of concrete being less than 35 is
P(Y < 35) c" Fr(35)

rn terms of standard beta variate X,

F r(35) =~ Fx ( y
---
b -·a
a)
~ FxG~ _ ~~) -~ Fx(0 .2)

As per Pearson's tables,

Fx(0.2) = Bo.2(1 .S, 1.5)
= 0.1423 (from tables)

3.7 EXTREMAL DISTRIBUTIONS

Civil. engineers are more concerned with the occurrence of the largest or
the smallest of a number of random variables in the analysis and design
of structures. The structural safety of a determinate truss (system) may
depend only on the extremes, for example, on the strength of the weakest of
many elementary members (components). A civil engineer may be interested
to know the value and the distribution of the likely maximum wind
speed, or the floor load acting on a building during its lifetime.
Let X be the largest of the 11 random variables Y1, Y2, ... , Yn. The
probability that all the values in n variables will be less than a specified value
x, is
Fx(x.) = P(Y ~ x,)
= P (all n of the Yt ~ x,)
 80

If the y,. are independent,

Fx(xs) = c P( Y1 <( x,)P(Y2 <( Xs) . . . P( Y" <( x.)
~~ F y,(x ,)Fyb:s) .. . Fy,(Xs)

Jf all the Yt are identically distributed with a common distribution Fy(y),

then
Fx(Xs) ""' jFy(Xs)l" (3.112)
If the X; are continuous random variables with a common PDF, fx(x), then

fr(x ,) ~ ~~.Fx(x,) = n[Fr (x,)ln-lf'y(x,)

_From past experience if an engineer knows the distribution of the max imum
wind speed Y; observed in any one year, he may be able to determine the
distribution of the largest wind speed in a particular lifetime of the struc-
ture, say 50 years.
It has been found that for some parent distributions of specific general
types, the extreme value distribution can be approximated by certain
theoretical distributions, ·called asymptotic distributions, for large n. As 11
increases, it is more and more accurate. It is not necessary to know the
underlying distribution of Y; precisely. It is enough if the general trend of
the tai I portion of the Y; is known. There are three asymptotic distributions
proposed by Gumbel. They are described below:
~ype 1 Extremal (Largest) Distribution: EX1,L (u, IX)

This distribution requires that the upper tail of the parent distribution that
contains the extreme value be exponential in nature (normal, Weibull,
exponential, gamma, and other similarly shaped density functions). The
distribution of X, the largest of many independent random variables with a
common exponential type of upper tail distribution (Fy(y) = 1- exp ( -h(y)),
has the form of Type I extremal (largest) distribution, given by
fv(x) = !X exp [ -CY.(x -- u) - cxp{ ·-IX (x ·- u)}l -co ~ x ~ oo (3.113)
Fx(x) = ~xp l-exp f- :z(x ·- u)}l - oo <( x ~ oo (3.\14)
The parameters u (locacion, here it is median) and o: (dispersion) are given hy
0.5772
/-LX =co II -;- - - (3.! 15)
IX

i
(3. 116)

This distribution is also called the Gumbel distribution and is positively

skewed. The coefficient of skewness is 1.1396. The distribution is designated
as EX1,I.(u, IX). The shape of the distribution for u = 0.275 and o: = 2.566
is shown in Fig. 3.29. This model is usee' for describing the maximum
annual flow in a river, the maximum annual wind speed at a location, etc.
 81

10

u: 0·275

-o.e lC

FIG. 3.29 Shape of Type 1 extremal (largest) distribution

EXAMPLB/oThe yearly maximum wind speed, X, observed at Pune

follows the Type I extremal (largest) distribution. It is given:
/LX = 83.67 kmph ax= 15.97 kmph
Calculate the parameters of the distribution and determine the probability
of the wind speed exceeding I I 7 kmph.
Solution Parameters of the distribution are calculated using Eqs. (3.115)
and (3.1 16).

(X = 0.0803
0.5772
ll = 83.67 - 0.0803
= 76.48 kmph
Hence the CDF of X [Eq. (3.114)] is
Fx(x) = exp [ -exp {--0.0803(x- 76.48)}] -oo::;;x~oo

The probability of the maximum wind speed exceeding 117 kmph in any one
year is
P(X > 117) = I - Fx(I17)
= l - exp [--exp {-0.0803(1 17- 76.48)}]
= I - 0.962 = 0.038
~e 1 Extremal (smallest) Distribution: EX1,s(u, (t,)
This distribution is similar to Type 1 extremal (largest) except that the
lower tail of the parent distribution has an exponential form. The distribu-
tion of Z of the smallest of many independent variables with a common
82

unlimited distribution with an exponential lower tail has the form of Tyre
1 extremal (smallest) distribution given by
fy(y) = IX exp [1X(y - u) - exp {1X(y - u)}] -- oo ~ y ~ w (3. I I7)
Fy(y) = 1 - exp [ -exp {1X(Y- u)}] --oo ~ y ~ oo (3. I!~)
The parameters u and IX are given by
0.5772 (3 . I I 9)
fl-y= u - -oc-

2 1T2
ay = 6<X2 (3.120)

This distribution is negatively skewed and the coefficient of skewness is

-1.1396. A typical shape of the Type I extremal (smallest) distributi0n is
shown in Fig. 3.30.
f v(y)

1·0

0·8 u :0·725
c(. ,.z S66
0·6

·1· 0 0 ·5 1·0 2 ·0 y
FIG. 3.30 Shape of Type 1 extremal (smallest) distribution

EXAMPLE 3.3 I The minimum annual tlow Yin a river is assumed to follow
the Type I extremal (smallest) distribution. The me:1n and slandard deYi:t-
tion of Yare

Calculate the probability of the minimum annual llow in a year being less
than 2 m3/s.
Solution The parameters of the distribution an~ [Eqs. (3.1 I 9) and (3.120)1
1T
IX= ,- "'-~ 0.64l
V 6/2
- - s _,_
11 .
~).577_2 - ~ 9
- ' 0.64 I - -.
Hence the C'DP of minimum annual flow is
F~(l') = I - exp 1--exp {0.64J(r --- 5.9)}]
The probability of the minimum annual flow in a year less than 2m 3/sis
P(Y < 2) Fr(2l
· C'Xp 1-- exr {0.641(2- 5 Q)}] = 0.079
 83

Type 2 Extremal (la1·gest) Distribution: EXz,t(u, k)

This is anotlier model for the largest value of many independent identically
distributed random variables. Here, the form of the parent distribution is
not generally defined. This model is generally selected on the basis of an
empirical fit to a set of data. The PDF and CDF of the variabltl X, the
largest of many Y; are

/x(x) = k [ u 1k+J
ux exp [ -(u/x)k] x~O (3.121)

Fx(x) = exp [ -(u/x)k] x~O (3.122)

where u and k a~e parameters of the distribution. They are connected to the
mean, variance and coefficient of variation of X as follows:

k>l (3.123)

a~ = 112[r( I - ~ ) - r2( 1 - ! )] k > 2 (3.124)

r(I - f ) - t k>2 (3.125)

r2(r - !) ••
The above equation has been solved and the values of k for the corres-
ponding values of ;) are given in Table 3.3. Type 2 extremal largest distri·
bution, designated as EX2,r.(u, k), is used to model the annual maximum
wind speed, the maximum annual flood, the maximum atmospheric tempe-
rature, etc. A typical shape of Type 2 extremal (largest) distribution is shown
in Fig. 3.31.

TABLE 3.3 Valun of k for correspo11di11g values of 3 for Type 2

extrl'lllflf (largest) distribution

8 J.:. 8 k II k

0.30 5.15 0.21 6.95 0.12 11.62

0.29 5.29 0.20 7.255 0.11 12.65
0.28 5.45 0.19 7.59 0.10 13.88
0.27 5.61 O.IR .. 7.97 0.09 15.425
0.26 5.79 0.17 8.395 0.085 16.35
0.25 5.98 0.16 8.87 O.OR 17.4
0.24 6.195 0.15 9.415 0.075 1R.62
0.23 6.42 0.14 10.04 om 20.03
0.22 11 675 0.13 10.77 0.06 23.72
84

fx (x)

1-0
u :1·727
0·8 k :5·150
06
0·4

0·2

00 J 4
FIG. 3.31 Shape of Type 2 extremal (largest) distribution

EXAMPLE 3.32 The yearly maximum wind speed X, observed at New Delhi
follows the Type 2 extremal largest distribution. It is found (from data)
that
p,x = 100 kmph ax= 23 kmph
Calculate the probability of the annual maximum wind speed exceeding
120 kmph.
The parameters of the distribution are first calculated. Thus
23
8x=-,..,.... 0.23
100
From Table 3.3,
k = 6.42 for 8i = 0.23
Using Eq. 0.123),
100
I. I} = 89.29 kmph

The probability of the wind speed exceeding 120 kmph in any one year is
P(X > 120) == I - P(X ~ 120) = I - Fx(i20)
Using Eq. 0.122),
Fx(l20) ,..- exp r---(89.29/12W' 5]
-., O.R64
Hence,
P(X > 120) ceo I - 0.864 """ 0.136
Type 3 Extremal (smallest) Di.~tl'ihution: EXJ,s(u, k)
This model is for the sm:tllest ol' tht: many random variables. This distribu-
tion is also called the Wcibull distribution which is extensively useLf in
 86

reliability studies. The PDF and CDF of a variable X following Type 3

ext rem a I (smallest) distribution are

/v(x) 'o= -k- -·- ('" - ')k·Jcxp [- (\:· -•--I/)k]

11 ··- I II - I II -
(3.126)

rr(X) ,_-~ I - exp - [ (;,\: -·-· ')k]

I x;-:1 (3.] 27)

The parameters 11 and k arc given by

JL_,. = 1 + (11 - nr( 1 + ! ) (3.128)

u~- :c~ (11 - n{r (' + ; ) - r2 ( 1 + ! )] (3.129)

This moue! has been used to represent the material strength in tension and
fatigue.
For many practicable problems, it may be reasonable to assume I = 0. If
I=' 0, Eqs. (3.126) and (3.127) simplify greatly. The CDF of X for I= 0 is,
Fx(x) = I - cxp [ --(x/u)kJ x ~ 0 (3.130)
with

(3.131)

(3.132)

(3.133)

The values of k corresponding to the values of8x arc given in Table 3.4.
Gumbel has studied droughts using this model with I = 0. A typical shape
of Type 3 extremal (smallest) distribution is shown in Fig. 3.32.

TA8LE3.4 Va lue.r of k for cot-respotrditrg 3 for Type 3

ex/remal (smlllleJt) distributio11

8 k ll k 3 k

0.300 3.72 o.rxo 6.54 0.10 12.45

0.250 4.56 0.170 6.97 0.095 13.18
0.240 4.77 0.160 7.45 0.090 14.00
0.230 5.00 0.150 7.99 0.085 14.92
0.220 5.25 0.140 8.61 0.080 15.97
0.210 5.52 0.130 9.34 O.o7 18.59
0.200 5.83 0.120 10.19 0.065 20.25
0.190 6.17 O. t 10 11.22 0.06 22.21
86

tx <)(>

0·5 .

04 U: 6-32
k =5· 83
0·3

0·2

01

FIG. 3.32 PDF of Type 3 extremal (smallest) distribution

ExAMPLE 3.33 The flexural strength X of an over-reinforced prestresse(

concrete section has been found to follow the Type 3 extremal (smallest
di lribulion with the I wcr bound I = 0. It is given :
p.x = 825.8 kN m ax = 48.5 kN m
Calculate the probability of failure of the beam when the external momen
on the section is 437.5 kN m.
Solution The parameters of the distribution are first calculated as follows:
Sx = 48.5/825 .8 = 0 .0587
Using Eq. (3.133),

r (I+-2)
(0.0587F = kl - 1
r2( 1 + T)
Solving the above equation or using Table 3.4, the value of k is found to be
22.86. Using Eq. (3.131),
825.8
u = ( I ) = 846 kN m
r I + 2 1.6
Hence the CDF of X is
Fx(x) = I - cxp [ -<.x/846) 21 '~] x~O

The beam l'aib when the strength of the section is less than the extern~l
m mcnt acti ng on the section. Hence the probability of failure, Pf, of th~
sec tion is
/ 11 -~ I'(X < 437.5)
= I -- cxr l--(437.5/846)2Hr•J
- o. 284 · w-r.
The PDI (lf :111 the l'lllllmnn distributions is listed in Table 3.5 also fo
 TABLE 3.5 Common probabilistic nwdels a11d their parameters

Relation between parameters and

Distribution and designation PDF
mean and variance

l+u
Uniform u-=-, I~ X ~ II l'x = 12
2 (u - /)•
CX= ~

l\'ormal (Gaussian) N<p,x. axl 1

-= exp
0X~2~
[ - '(x-p.x)']
-::;- - - -
- ax
CIO~X~CIO ~X and a2
X

Lognormal LN (X. "In x)

1 ~{ln
.\"<11n XV:!:t exp [ - -
(>·{X)}•] x~O
-X= 11-xexp ( - I 2 )
TalnX
aln.\'

a~n X = ln (&i + J)

Gamma G!k. A1 A!A.r)k- le- Ax x ~ O: k. >. ;;>- 0 k

p. X = ...,...; a\ =
k
,.--

(x -
rck l
a)p-J(h - xiq-l
" "'
Beta BT (a. b. p, t/) u ~ x ~ h '
I'•' = a , - - (b- a)
p .
B(p. q)th - u) p+q-l ,, p +q
s [ pq ]
"X = (b - a)% tP -;- tl)' (p + q + I)
Type I extremal (largest) O. ST12 • ,..
EX 1,L (11, «)
"'exp [ -Ot(x- u) - exp { -tz(x -til}] -oo ~ x ~ co !1-x = II -;- - «- ; ax = ~
Type I extremal (smallest) 0.5772 t :t•
« exp [tz(x - u) - exp {<L(x - u)}] -co ~ x ~ GO ~...- ~~ 11 - -"-; ax = (;i
EX~,s(u, «)
Type 2 extremal (largest)
-k ( -u )k+l exp [ -(u/x)kj x ~ 0 11-x = ur(t - -}) k > 1
EX 2,L (11, k) II X

c~r = u=[r(l - ~) -r•(t - +)] k > 2

CD
(Collld.)
""'
 TABLE 3.5 (Coutd) CD
CD

Relation between parameters and

Di~tributiL'Il and Designa!ion PDF
mean and variance

Type 3 .:xtn:mal (smJIIes!)

EX 3 , 5 (u, In u-lu-1
(x - f)k-t
-k- - - - exp [ - (x---/)k]
u-1
x~l l'x = I + (u - I)F (I + ~)
cr\=(u- l)•[r {1 + !) -ro(t +f)]
(Ar ).re- >.r
Po:sson P(),t) X= 0 . J. 2 . . . . l'x = >.t
,,-•
crx =At
?

l
E'.ponenlial EX(.\) ~e-Ax X~ 0 P-x =y
l
.. \ .. = ~·

R;~yieigh ;o
X exp [ - 2] ( -;-
\" )"] x ~ 0 l'x = a.-J-:-:!2

cr} = a.•(z - ; )
 89

ready reference. The distributions, (i) !-distribution (ii) chi-square distri-

bution and (iii) F-distribution, which arc generally used for statistical tests
(hypothesis testing), are not given.
Throughout, it has been assumed that the parameters of the distribution
arc known. They are to be estimated from the data using (i) the method of
moments or (ii) the method of maximum likelihood. Readers are suggested
to refer any book on probability and statistics for parameter estimation.
For a given data, the suitability of a probabilistic model is checked using
chi-square lest or Kolrnogorov-Smirnov test (Refer 3.1 ).

REFERENCES
I 1 Benjamin, J.R. and C.A. Cornell, Probability, 5itatistics and Decision for Cit·il
Ent;iltccr.<, M•·Graw-Hill, New York, 1970.
1 ., Ticky, M . ami M . Vorlkek, "Safety of Reinforced Concrete Framed Structures",
l'roccedi11gs, lntl'matiunal Symp. 011 Flnuml Mecltcmics of Reinforced Concri'IC,
Miami, 1964. ASCE-ACI, pp 53 84.
3,3 !'carson. K .. Tabln of tltc lllcolltJ!f!'lc Gamma hmctiu11, Cambridge University
l're~s. London, 1922.
3.4 llhatlacharjee, G.P., "Algorithm AS 32--Incomplele Gamma Integral", Joumal of
Hoyul Statistical Society, Vol. 19, No. 3, 1970.
J.5 l'car<>on. K., Tahlcs of tlte brcumpletc Beta Fu11ctioll, Cambridge University Press,
Cambridge, England, 1968.
3.6 Majunuler, K.L. and G.P. Bhattacllarjee, "Algorithm AS 63-The Incomplete
Beta Integral", Applil'd Statistics, Vol. 22, 1973, pp, 409-411.
J.7 Cran, G.W., K.J. Marlin and G.E. Thomas, "Algorithm AS 109-A Remark on
Algorithms AS 63: The Incomplete Bela Integral", Applied Statistics, Vol. 26,
1973, pp 111-114.

EXERCISE
J.l During Ihe monsoon sc:1~u n in Uombay, a Mro ng winll may co me l'rom uny dircc-
lion h •I ween 9 = 0 ( o uthJ and 8 = 90 (WO. t) . The maximum wind speed cannot
he greater lhan :wo kmph. Sketch I he sa mple pace for th e wind direction and the
wind speed. Show the event , the wind s peed greater than 30 kmph, and the wind
direction, 20 < 9 < 60, in the sketch.
3.2 A si mply supported bcum of span I is to be designed for hcur. There are two
load. Q, 20 k N ;rnd Q1 = 50 kN which ca n come o n the beam; but they can act
o nly at discrete points. 0.25/, 0.5/ and 0. 75/ on the beam. Lt is not necessary that
both load s should act lll I he SillllC time. Sketch the sa mple space for the shear at
the left end of the beam.
3.3 The completion of a water l<~nk involves the successive completion of four stages.
Let ,
A ~ cxcaV<ftion (;omplcted on lime; P(A) ~ 0.9
JJ ~~" foundation completed on time; P(B) ~ _0.8
C ~" columns and bracings completed on lime; P(C) = 0.7
D ·" tank cumpleteJ on time; P(D) = 0.7
If !he events :nc statistically in<jcpcndcnt,
(i) what is the probability of the whole structure completed on time?
(A11s. 0.3528)
(ii) whnt is the prohability of the tank portion completed on .time and atlcast one
of the other three works is not completed on time? (Ans. 0.3472~
90

3.4 !"here arc three member~ in a dctcrrninalc tnt's subjected loa ~;ivcn ~yslcm lo;~d-
ing. If P; is the probability of failure of the mcmb~.:r i, it is given as: /' 1 0.1,
p, - 0.2 and p 3 - ' 0.3. The performance ul' a member dcpcnJs on other 1ncmlm'
II is given:
/'ll-'1 I F,nlal 0.8
Determine the reliability of the truss.
(:lu.r. 0.784>
3.5 A policy dcci;ion, like limiting the maximum salary of an lnJian to Rs . 1.500. ;,
10 be raken hy the Govcrnnwnl. This det end on the elec tion rcsulb Supr ~c lt
tile party ·I wins, the probabil ity f imp lementing th e Jccis ion i flO%. while it IS
20~ u fm lh ' p:r rt y flnnll 40 ~~ fur the Pill'tY ( ', ' ~~t l lllC th CI'C OirC nly three r~rties .
Wl llwul "now ing whil: h part y will 1 in in the elect ion, one Cllllll 1 ay tilt' t: hunc •
of inlrot.lucing the decision. If the chance of .1 winning the cle~.:lion is 0.6. of B
0.1, and of C 0.3, determine the chance of introdu~:ing the t.lccision.
(A111. 0.62)
3.6 fhc probability density function of ntinfall in a t.lay during the monsoon ;cason
is given by
fx(x) ~= 32e-4.< x~O

Calcul<ttc the mean and the variance.

(Am. IL ·~ ~; a• "' 25)
3.7 Two variables, X ant.! Y, follow the lognormal distribution. It' Z -'~ ,\")' and vari-
ables X and r arc statistically indl·pemlcnt, prove that Z follows the lugno1 n1:tl
t.listribution.
3.8 The cube strength of M 35 concrete. X. follows the normal distribution with
parameters p, ~ 42.28 N 1mm• and a ,_ , 5.6 N 'mm•.
li) What is the probability or X < 35?
(.4/1.1. 0.0983)
(iiJ What is the probability of 30 ~ X~ 50?
(Ails . 0 .90 \SJ
3.9 The yield strength or steel. X, follows the lognormal t.listribution with 111CIIIl
~ 15M N/mm 2 and a ~ 48.8 N/mrn 2 • What is the probability <~f getting a ykld
strength value less thanl500 N,mma?
(Ans. Cl.t\793)
3.10 'I he fatigue life of a structur;tl component, rneusurcd in terms ol the number of
cycles of a particular lout.!. is mot.lellcd having Wei bull distribution" hich is ~,:iven
by
f~:(x) ct.ft.r~-1 cxp (-ct..\')~ .\',<X, ft > 0

ct. and fJ arc parameters of the distribution given by 0.001 and 0.5 respectively.
The mean value of X ;1nd the parameters arc rc.latcd by the CC(Uation

l'x .co 0(-1/ll/'(t -r- f)

(il How long t:an such structural component be expected to last.
(Ans. 2 x IO" cycles)
(ii) What is the probability that such a component will last more than 3 x 10'
cycles'/
tAns. 0.1769)
 4
Resistance Distributions and
Parameters

4.1 INTRODUCTION
The first step in the reliability analysis and design of structures, is to study
the variability of the strength of the structural (RCC, steel, prestressed
concrete, masonry, etc.) members in flexure, shear, compression, bond, tor-
sion, e~c. The strength of a structural member may vary from the calculat-
ed or 'nominal strength' due to variations in the material strengths and in the
dimensions of the members, as well as variabilities inherent in the equations
used to calculate the strengths of members. One has to identify the sources
of variability and quantify (statistics) the same. The fundamental require-
ment in the reliability study is the collection of data on strength and other
physical properties of the materials of the structures, and the geometric
parameters of the sections and the statistical analysis of the same.
The structural designer specifies the characteristic strengths of materials
and the builder tries to procure the materials satisfying the specifications,
and thereby, trie5 to achieve the same strength as assumed by the designer.
However, if the quality control is poor, then the strength of th..: structural
member will be less than that assumed. This may endanger the c;afety of the
structure. Hence for providing a design with an assured level of reliability
the systematic identification of the uncertainties in the strength of materials
and the dimensional parameters and statistical analysis of the collected
data becomes an important task.
In this chapter, information on statistics of basic variables, vil. physical
properties of concrete, reinforcing steel bars and bricks, and dimensional
variations of RCC members, based on actual field data, are furnished.
Methods are also indicated to account for other uncertainties and thus to
determine the allowable stresses of materials for a given reliability or
probability of failure.

4.2 STATISTICS OF PROPERTIES OF CONCRETE

The cube strength (compressive strength), the modulus of rupture (flexural
strength), and Young's modulus (initial tangent modulus and secant modulus)
are the properties of concrete that are generally required in the design of
concrete structures.
92

The strength of concrete in a structure may be diflercnt from the specified

strength, and also, the strength may not be uniform throughout the structure.
There are several sources of uncertainty which contribute to the total
variation in the strength of concrete. Sources arc (4.1 ):
(i) variations in the quality of materials
(ii) variations in the placing of concrete
(iii) variation in the supervision
( iv) variations in weighing
(v) variations in the mixing proccdun:s
(vi) variations in the transporting methods
(vii) variations in the testing procedures
(viii) variations due to the actual strength of concrete in a structure
being different from the control specimens (cube or cylinder) and
(ix) variations in the methods of curing
In construction projects, samples of concrete cubes or 150 :< 150 ;..: ISO mm
size are generally cast during every hatching of concrete. These cubes are
tested in a laboratory at the end of the 28th day of curing. The mean value
and standard deviation of each set of a concrete mix can be obtained . The
computed mean value of the strength of each set can be plotted as shown in
Fig. 2.1. It is generally found that the in-batch variation, which may be
considered as a variation in the testing procedures, mixer inefficiencies anu
the actual concrete strength, varies from 3 to 10 percent. All the lest results
of a particular concrete grade belonging to a project can be clubbed and a
histogram can be drawn. A typical plot of a histogram of the cube strength
of M 15 concrete belonging to a project is shown in Fig. 4.1, (4.2). Concretes
of' the same strength and with the ame quality control may be prepared in
din'ercnt projects. All these samples are combined to form a class of con-
crete. Figure 4.2 gives the histogram of a typical class of M I 5 concrete. It
can be observed, as expected, that the variation and coefficient of skewness
for the class is more than for a group shown in Fig. 4. I. As more and more
groups are combined, the distribution may become more and more skewed.

~
..
.Q

>
0
20 n =63
M•an .. 20·29N/mm2
SO :'l' 49N/mm2
0·32
>-
u
G; 15
Ill
0·24 ~
.Q r---
~
0
..... 10 - 0·16 ....
0

-- I'---\

I I
I I
0(18

0
Gl
>
:sli
a:

16·0 18 .5 . 21·0 23·5 26·0 28·5 31·0

Cube str•noth ( N /m m2)

FIG. 4.1 Histogram of M15 concrete for a typica 1 proie~! group

 93

A national building code must specify the coefficient of variation for a par-
ticular class of concrete, irrespective of source. It is also felt that this
specified coeffic ient of va ri nti on mu st be rel ated to lhc quality co ntro l. l n
the present c de IS: 456- 1978, the cod t: spec ifi s th e valu es f ta nda rd
dev ia tio n for various grad e of concrete; but the degree of qua lity co ntrol
is not a ttached to these values.

n = 399
Mean= 24·03 N/mm2
SO: 5·76N/mm2
106 - -
r-
r--
Ill
c:
0
-
~72 -
Ql
Ill
~

-
0

0
I--"

~
~ 36 . -
~
z
~
f--

0
0--, - 0
12·0 I 20·0 ) - l 28·0 ~ i 36·0 A' • 44·0 52·0
Cube strength (N/mm2)
FIG. 4.2 Histogram of a typical class of M 15 concrete

After drawing the histogram, a mathematical probabilistic model is fitted

to the data. The different types of models that are normally used to describe
the compressive strength of the concrete cube are:
(i) normal distribution
(ii) lognormal distribution and
(iii) beta distribution
The suitability of a probabilistic model to fit the data is arrived at after
applying the chi-square or the Kolmogorov-Smirnov goodness-of-fit tests
(4.3) . Th" chi-~q uare test is briefly explained below: ~~

~uflre Test
I. Draw the histogram for the observed data.
2. Assume the model with its parameters calculated from the data.
3. Select the level of significance 11.. Generally 11. is taken as 5 or I per cent.
4. Calculate the value of chi-square as '

X~al = X ( QI - e,)z (4.1)

1-J er
 94

where
·-
X~nl = the calculated value of chi-square
o; ,-, the observed frequency in the ith interval
e; -, the expected frequency corresponding to the assumed distribution in
the ith interval
a = number of intervals considered
5. Compute the number of degrees of freedom N given by
N =,a-- r -- 1 (4.2)
where r is the number of parameters estimated from the data.
6. For the assumed a, and computed N, find the value of chi-square from
the standard chi-square table available in any text-book on stntistics (4.3).
Let this be designated as X~. 1- a
7 If X~,11 is Jess than the vaiue obtained from tl)e tables, accept the dis-
tribution with its parameters at the assumed level of significance. Otherwise,
reject the hypothesis.
The chi-square test is demonstrated in Table 4.1 for a set of data of M 15
concrete (4.4). Readers should read a specialist's book (4.3) for this topic.

TABLE 4.1 Deii/OIIIfratioll of chi-square 11'.>1

Sl.
Interval oi ('·
I
(o 1 - e1)2 /e 1
No.

< 14 1() 19.6 0.66

2 14-16 53 53 9 0 02
3 16--18 RS 75.1 2.22
4 18-20 45 57.3 2.64
5 20-22 30 28.9 0.04
6 22-24 15 10.9 1.54
7 > 24 3 4.2 0.34

250 250 7.46

(i) Model assumed-lognormal with parameters

X= 17.36 N/mm 2 and OJn x = 0.152

(ii) rx assumed ~ 5%
(iii) Calculation of e;:
e, o=o (n)p; where n = sample size

p1 --, P(X < 14) = rp .Jln

,
° /X)} =
O(n X
4
0.0785

e1 = (250)(0.0785) = 19.6
 95

Similarly, e2 = (n)p2,

pz = P(X < 16)- PCY ~ 14)

= <~>{ 1 " Oln<t 6}{Ji>}- o.o1ss = o.21so

e2 = (250)(0.2156) ::= 53.9

Similarly, other values of e, are calculated and given in Table 4. I.

(iv) Degrees of freedom N = a -- r - I = 7 - 2 - I = 4
(v) From chi-square Tables (4.3),
2 2
XN,fl-rx) = X4. o•>~ = 9.49
(vi) X~al = 7.46 < 9.49
(vii) Hence accept lognormal model with parameters at oc = 5 per cent.
The procedure that i. c plained for the collecti on of samples for the com-
pressive stre ngth of concrete. can be f0l lowed for the collection of amples
of cylinders (15 em din . X30 em height) and beams (10 em X 10 em X 50 em)
belonging to different grade. 0f concrete. ylinders can be tested to get th e
dnta on the initial tangent modulus, E, ., and the secn nt modulus, E.,r. or
concrete. ollected beam specimen can be tested in the lrtboratory at til~:
end of the 28th day of curing to get datl't on th e modulus of rupture of
concrete, f;.. Table 4.2 give. the result. of the tali tical una lysis of the data
on various properties of concrete c llected by the author 4.4, 4.5 , 4.6) at
various places in lndia . Frcquen y distribution of E,r and f, of M 15
concrete are shown in Figs. 4.3 and 4.4.

100,.---- - - - - - - -n- :: -1-50-:--- - - - - - - - . , 0·4

Mean:: 2·5 X10 4 N/mm 2
SO 3·4X 104Ntmm 2
0·3 >-
u
c:
Gl
:J
g-
0·2.!::
...
0
Gl
>
·~
t
~25 0 ·1~
i
0 ~----L--L--L--L~2~~~2~9~~~3~3~--~ 3·7°
1·3 1·7 2·1 ·S · ·
2
Initio! tQnQrl'lt modulus X 104(N/mm )

FIG. 4.3 Frequency distribution of Initial tangent modulus of M 16

concrete
 TABLE 4.2 Results uj statistical analysis of •·arious propcrticr of cmun•!e .,
CQ

-
Specified strength !-' cr 8 Probability Quality
Variable and source Mix
(N/mm') (N/mm•) (N/mm') ~~~) distribution control

Cube Strength
liT, Kanpur M 15 15 24.03 5.76 23.96 LN Nominal mix
M 20 20 29.16 5.49 18.83 N
M 25 25 30.28 3.77 12.45 N Design mix
M35 35 42.28 5.60 13.24 N
REC , Calicut M 15 15 22.67 5.ot 22.10 LN Nominal mix
TIT, Bombay MIS 15 17.56 2.69 15.33 LN Design mix
M20 20 26.80 4 04 15.07 N, LN
Cylinder Strength
liT, Bombay M 15 11.10 1.92 17.28 N,LN
M20 17.21 3.34 19.40 N,LN
Initial Tangent Modulus
liT, Bombay M 15
.. M20
22,076
25,491
25 ,147
34,100
3,398
5.009
13.51
14.65
::-r,LN
N,L::-r
Secant Modulus
liT, Bombay M 15 - 19.606 3,397 17.07 N
M20 - 28,031 4,951 17.66 N, LN
Modulus of Rupture
IIT,,Bombay M 15 2.71 3.682 0.871 23.64 N
M20 3.13 5.893 0.603 10.26 N,LN
 97

100r - - - - - - - - - - - - - - - --.0·4
n :2.50
Meon:J·682 N/mm2
S0:0-871 N/mm2

..
Cll
>

~:I
25 0·1 'i
0
a::
z

Q11-:•Q,..-
1·5~-L-,J2·L5__JL.._...J
J·L 5- 4· 5 5· 5
Modulus of rupture (N/mm2)
FIG. 4.4 Frequency distribution of modulus of rupture of M 15 concrete

4.3 STATISTICS OF PROPERTIES OF STEEl~

T he yield strength , _(;.and the modulus of elast ic ity of steel, Es, arc the two
main phys ica l proper ties of steel that are used in th e des ign of RC and steel
st ructu res. I n the ca se of prestressed concrete structu res, the ultimate st rength
of high tensile steel wires is used in the design. The variation in yield stren gth
is due to the variation in (i) material strength, (ii) cross-sectional area,
(iii) rate f loadin g d uring testing . and (iv) I he effect of t rai n at wh ich t he
yield is de fi ned (4.7). The amount o f varia tion in trenglh wi th in (l sinp.h:
bar cont inu ously ca.L for a pa rt icul ar length in u ingle cast is \!cry · m .:~ ll .
(less tha n one per ent ) and may he negligible as sh wn in Fig. 4.5. How-
ever the in-b. lch variatio n fo r n gi ven hea t i slightl y !u rge r. For a on. truc-
tion w rk, the reinforc ing bar. may be su ppl ied by o pH rti ~,; ul ur ma nu rnctl lr-
ing firm havi ng a num ber r ·tccl ro ll ing mil t . Hen c.: e, the supp lied bar.
may be frc1111 difl'erent rolling mi lls. If the chel11ical compo itlc n of s1e I is

"'
~ 1850
4mm 4JHTS
.c: w i rt'
~1750
Q,

~ 1700

Specimen number
FIG . 4.5 Variat io n of ult imate strength in a single cast length
98

well controlled during the production, it is reasonable to expect the vari-

ations in strength from cast to cast to be small; otherwise it will be signi-
ficant. If the bars are supplied to the site by different manufacturers, the
variation in strengths may be high due to different rolling practices and
quality control adopted by different manufacturers. The variation in the
strength of steel due to the change in the mean strength with bar diameter
is significant. From these discussions, it is evident that there are several
sources which contribute to the overall variation in the strength of bars.
pccimens can be collected from varioliS rolling mills belonging to a parti-
cular firm <1 nd these can he te ted in a laboratory to determine _(~ and E.,.
Test results of all samples irrespective of the dian1eter are clubbed, and the
me<Jn value and standard deviation of such a data belonging to a particular
mill arc calculated. A histogram can he drawn for such a data. Figure 4.6

n = 745
Mean:447·61 N/mm2
50:17 26Nimm2
~ 222 0·3
0
0
>
~
~148
0
Cll
>
....0
0·1 ~
a:

Vield strl'ngth ( N fmm 2)

FIG. 4.6 Frequency distribution of yield strength of Fe 415

grade steel from a rolling mill

shows a typical histogram of yield strength or high yeild strength deformed

b:1r (1-IYSr) belonging to a mill. The procedure is repeated by collecting
specimens from various rolling mills or at various stage of a construction
project. To know the statistics of the yield strength of steel, irrespective
of the sou1ce :~nd diameter, all samples belonging to rolling mills and from
the construction project sites (field specimens) arc clubbed and the mean
value, standard deviation and coefficient of variation of Fy or r,
for such a
data can be obtained. A histogram is drawn for such a data (Figs. 4. 7 and
4.8). Suitable probability distribution can be fixed for the collected data
using any one goodness-of-fit tests. Table 4.3 gives the results of the statisti-
cal analysis of the data on the strength of reinforcing bars collected h\'
lhe author at various places in India !4.2, 4.4, 4.8).
100

4.4 STATISTICS OF STRENGTH OF BRICKS AND MORTAR

For the reliability study of brick masonry construction and reinforced brick
masonry structures, the statistics of the strength of brick, strength of mortar,
thickness of bed joints, water absorption, reinforcing steel bars etc. are requi-
red . The strength of masonry depends on several factors, such as the
strength of the brick unit, strength of mortar and thickness of bed joint, the
type of supervision given in construction, etc. The effect of the uncertainty
or variability of different parameters is responsible for the variability of
masonry strength. The statistical variations in brickwork depend very much
on the constructional practice and the degree of quality control.
Dayaratnam and Ranganathan (4.9) have collected samples of l;lricks
manufactured by different firms around Kanpur(U .P.) and Calicut (Kerala),
and have done detailed study on the various properties of bricks. Results
of the statistical analysis of the strength of bricks are presented in Table 4.4.
The strength of the prism decreases at a faster rate with increase in the
joint thickness for perforated bricks as compared with the solid bricks. So it
is expected that the variability of joint thickness will affect the strength of
brickwork . Results of the statistical anal ysis of the thickness of hori zontal
and vertical joints in existing structures are also given in Table 4.4.

TABLE 4.4 Rt•stdts o./ statiltica/ analysii o/ hri< k s. mortar .\ ll'rngt!t

and joint thicknesr (4 . 9)
- - - - - - - - - ---------·-
Probability
Source Pa rameter p. WJ distribution

Bricks
Kan pur Zon l: Length (rnm) :!28 .5 1.6 N
Arendth (mmJ 109.4 2.3 N
Height (mmJ o3.3 3.3 N
Wat~r ahsorp1inn ( ~) 0
15 I 22.3
Compre"i'- e str<·ng th
IN mm'J ILJ 9 3 l.ll
Calicut Znnc Len!,\th (lllllll ::'.32 .7 l ,tl "N
Breadlh lmrn) 116.0 2.3 N
Hcighl 11111111 75 .7 4.8 N
W~ilCI' absorpliOll (~ 0 ) I (•.4 12 ')
Compres,;ive strengt h
tN;mm'J lJ.~ 29.X N
Mo1l<11
HOI i/'.lllta I J <>illl 1 hi~knc~s (1111111 12 .4 13.5 N
Ve1 tic~1l Join I I hicknc" (1111111 12.tJ 16 4 N
M ix I. J' Stre ngth (N :mm'J 2 1. 5 13 .3 i'i. LN
Mix 1.4'' Strength <N 111111'l 1-U 10.0 N, LN
Mi~ 1.5* StJ<:IIgth tN 111111') 1114 18.1 N,LN

*L.aboJ<IlOI) made spc~imcns

Nott•: ~--Normal; LN --Lt>!.\IH•IIII;II
 101

The mortar is used as the binding material between brick units. The
results of the statistical analysis of the strength of cement mortar cubes
belonging to different mixes- 1.3, 1.4 and 1.5- are presented in Table 4.4.

4.5 DIMENSIONAL VARIATIONS

The dimensions of RCC members may not be the same as specified. There
may be deviations from the specified values of the cross-section shape and
dimension, which may be due to size, shape and the quality of formwork,
and concreting and vibrating operations. Variations also occur in the effec-
tive depth of members. The actual effective depth available may be different
from the specified values because of the improper placement of reinforcing
steel bars, not providing proper cover blocks and change in values when
needle vibrators are used during casting of members. The amount of varia-
tion in dimensions vary from place to place and structure to structure depend-
ing on the quality of construction techniques and the training of the site
personnel. Mirza and MacGregor (4.10) studied the variations in dimensions
of RCC members for American conditions.
The difference between the nominal and the built-in dimensions are best
characterized by the mean and standard deviation of the error. The coeffi-
cient of variation of the error increases as the size of the member decreases.
Nineteen multistoreyed buildings have been visited during construction, and
actual field Jata have been collected on the various geometric parameters of
RCC members (4. 1I). The data are to be collected during the construction,
and the measurements of members are to be taken in an unplastered
condition. The results of the statistical analysis of variations in dimensions
of slabs, beams, columns and foundations, carried out by Ranganathan and
Joshi (4.4, 4.11 ), are presented in Table 4.5. The relationship connecting the
coefficient of variation and nominal size of the member has been found to be

lh - 4.9/11, {4.3)

where It, is the nominal size of the member in mm. The frequency distribu-
tion of deviation in beam rib depth is shown in Fig. 4.9. All variables follow
the normal distribution.

4.6 CHARACTERIZATJON OF VARIABLES

The basic informati on required to describe behaviour of a rand om var iable
' is the proba bilit y ui:.tributio n with 1L · parameters. Howe er, in the ca e of
lir t-order·scconJ moment n1cthod of reli ability , aria ble are characteri zed
by Lheir mean · and coeffi ienl or variation . The co nce1 ~ of uncertninty i
c nveyc I through the coeflkie nt or variat ion. l n I'Ciit-tbili ty study, ~ill unccr·
tainl ic · whi h affect the de. ign reliabi lity must be ncco unlcd for . The e
uncertai nties must inc lud e th e inherent stati stical variability in the basic
\'n riable and the add itional wrcc~ nf uncertainties nri si nl! due l model·
ling. Modelling uncerlnintic '' uld incluJ errors in the ~:sli1111lion ot'
102

TABLE 4.5 Results of statistical analysis of variations in dimensions

of RCC members

Mean Standard
Size range
Type deviation deviation
(mm)
(mm) (mrn)

Slab (13 slabs) 100 to 110

Overall depth + 7.89 5.43
Top cover -19.75 6.89
Bottom cover -1- 3.27 7.8
Effective depth + 1.87 6.8.
Beam (252 beams)
Breadth +10.29 9.47 200 to 350
Overall depth +14.37 9.38 250 to 700
Effective depth + 6.25 3.79 270 to 370
Top cover - 0.56 8.41 30
Column (364 columns)
Breadth - 0.25 5.69 259 to 300
Depth + 0.11 7.89 250 to 1000
Cover (for 62 columns) -19.09 12.13 40
Distance d, * fi.24 11.89 360 to 710
Footi11g (6 footings)
Length -40.25 46.50 1500
Breadth +37.73 32.28 1300

Note : *d, is the distance from one end of the column to the centre of bars on the
other side.

----n·=-2::-cs=-::---
2 -- ---,o ·io
Mean :14·37mm
SO :9·38mm

-5 25
Devial1on in· beam rib depth ( mm)

FIG. 4.9 Frequency distribution of deviation in beam rib depth

parameters, probability distribution, idealizations, testing procedures, human

errors in calculation , etc.
Let
X h..: a ba ~ i~· ,·ariuhlt!
1'-.r he the liul' llll'atl nl· X
 11
I
J

103

ox be the true coefficient of variation of X

X be the sample mean
ox be the sample coefficient of variation
X and Sx are calculated from the data collected unde( carefully controlled
conditions. Hence, llx describes the inherent statistic~! variability. If the bias
and coefficient 0[ variation of uncertainties, attributed by other factors, are
considered as M and 8M, then 1-'X and 8x are estimated as (4.12)
(4.4)
-o 2
&x = (ox + oM) 1t2 (4.5).
If 8!11 is due to 11 factors, 8/11 can be broken and written as
= (ll~ + 8~ + ... ll~)l/2
OM (4.6)
If the model is unbiased, M is taken as I. In this approach, what is done is,
is predicted by f(. Using this approach, the overall variation in the basic
1-'X
random variables can be nxed. This is illustrated below.

4.6.1 Compressive Strength of Concrete in Structure v

11
Let
Y be the cube strength of concrete
X be the strength of concrete in structure
The mean value or the strength of concrete in structure is taken as 0.67
times the mean value of the cube strength of concrete. That is,
/1X = 0.67/LJ
'i
ln section 4.2, the ClH!Ilkicnt of variation of Y, rcprc~entating inherent
variability, was obtained. Taking into ac~:ount the uncertainties involved in ij '
the testing procedure o;.,t,st) and in siru \' ~tria lion of the strength of concrete I;
(8;" •'"'), th e e 1imate qf the coefficient of variation of the strength of con-
crete in structure can be written as (4.1).
\
~2 < 2 . "~
OX = or + Oiii!C~l
+ Oitr
,2
Hill

If
O;n situ = 0.1
then,
si- = oi- + o.oJ 25
From Table 4.2, for M 15 concrete (design mix),
Sr = 0.1533
1-Ienc.:e the total \'ariation in the strength of concrete is
8x = I(0.1533)2 + (0.05)2 + (0.1 )2]!12
= 0.18
104

Similarly, the mean value and coefficient of variation of the compressive

strength of concrete in structure for various grades are calculated and given
in Table 4.6.

TABLE 4.6 Statistics of strengths of concrete and steel

Me1tn Coefficient of Probability

Variable
(N/mm1 ) variation distribution

(i) Compressive strength of

concrete in structure
(a) Nominal Mix M 15 15.19 0.24 LN
M 20 19.54 0.21 N
(b) Design Mix M 15 11.78 0.18 LN
M 20 17.96 0.15 N
M 25 20.29 0.15 N
(ii) Initial tangent
modulus of concrete
Design Mix M 15 25147 0.187 N,LN
M 20 34100 0.206 N,LN
(iii) Modulus of rupture
of concrete
Design Mix M 15 3.682 0.246 N
M20 5.893 0.125 N. LN
(iv) Yield strength of
steel
Fe 250 320 0.10 N
Fe 415 469 0.10 N
(v) Modulus oi elasticity
of steel
2.04 x 10' 0.091

4.6.2 Yield Strength of Steel

In the case of steel, we must include the variation that may occur due to the
testing procedures and the method of specifying yield point. If the coefficient
of variation in the :esti ng procedures (ointcst) is taken as 5 per cent anJ the
coefficient of vari&i'ion in the mcth d of specifying yield (8ap. level) is also
taken as 5 per cent, the total variation in tlie y1eld str ngth o f Pe 4 15 grade
steel is
• 2 2 2
atot~t = 8actuat + 8iutest + 8sp. level

From Table 4.3, Sactual = 0.073. Hence,

Ototal = [(0.073)2 + (0.05)2 + (0.05)2] 1/2
= 0.102
For mild steel (Pe 250), it is expected that
o,p. level C:.!. 0
f
I
 I 1

1 O!i

for Fe 250 grade steel, ?>"'' " is

For J;,dian conditions. the >tatistics of the strength ol' cuncrek and steel,
giYen in Table -l.6, may be used for code calibration.

4.7 ALLOW.\BLE S J'RESSES U:\SED ON SPECIFlEO /

HELIABILITY (4.1Jl
The 1\tlrking ~Ires' de.-ign ( WSD) :lnaly,es :1 \trul'lut·c i"ur \\'Urking load-..
ami dc~igns the members such that the :1ctual :-tresses in the members :11C
limited lo :1 portion o[' the ) icld S(I'C 'l '> or critical Or ultimate 'tre ' ~ that l\111
be carried b~ the material.
Dc>ign criteria in WSD can be ..,pccilied a'
/i(!'vl, (it-:, GS, DL, LL. WLJ < 1;" tM, .!::_6.. GE . GS. DL, LL. WL) (~.7)

where;; is the 'Ires-, developed in the structure and is the ;tlltnl able 'tress. t>
The sub ~ cript i rei"e rs 111 tethiPn or .comprc s,iun 1>1" 11c.\lll e or shear or buml
stress, etc.
The stress ,leveiL'Ped in thc structure can bt: axi:tl •lr lxnding nr shear. It
is a function of the material properties (MI. geLllllCtr) of dements fGU,
geometry of ~tructure~ (GS), dead lu,td ( DL), li \'e lu:1d (LL), wind f,);td
(WLl, etc. The allowable stress is a Cunction ol' the material. It also depend'
on the specifications for testing the m:1terials. A liberal spccilicatilln llll
material standards has to be compensated by lower allowable st1 C'>S . 1,,
depends on functional aspt:ct (FA) of the structun:. A high pre ~ sure \e ~s l'l
for liquid is to be treated diffcJently from a !ugh pressure I'C sscl t'nr the
containment :-hell of a nuclea1 reactor .
Allowable stresses in compression are go verned hy the buck Iing cri t<.:rion,
which Jepends on the geometry 11f the clement and strul'lure. :\llm\·:1hk
~tresses in a single load condition arc dillcrcnt t'ron1 tlwsc in a combined l(laJ
conditi.on and hence it is a function ol· load combination. Expressiun'
simibr to Eq. (4.7) can be written for other Jcsign criteri :! bcl>cd on allm~­
able deflection and cracking. The code spccilicati•11t fur pcrmi~sihlc strc~sc-,
has to lake care of the many complicated situatil'llS.
Pl'ohability of Failw·e of M11terial in WSD V
When the stress developed in the material is greater than the <illowablc s1rc''·
it is defined as a failure. Hence the probability or t'ailur~: of material, /'J,
can be written as
{11 C-'O
- ~ - · · ··"" ..
P(X
____< _fa)
,._... ~ , ..
(4.8)

where X is the random variable, namely the srcngth of the malcnal.

If X is normally distributed.

flf = ,, ( {;,
Cj) ' - - - -
/lX) (4.9)
. ax
 106

EXAMPLE f The cube strength of M 20 concrete, Y, follows normal distri-

buti n. tven:
J.tY = 26.8 N/mm2 8y~-=0.18

/a = 0.34 X (cube strength of concrete)

= 0.34(20) = 6.8 N/mm2
Determine the probability of failure of concrete in the structure.
Solution The strength of concrete in struct'.lre, X, is equal to 0.67 times the
cube strength. Hence
p.x = 0.67 p.y = 17.96 N/mm2
u,y = 0.67 uy
= 0.67(0.18)(26.8)
= 3.23 N/mm2
Using Eq. (4.9), the probability of failure of concrete in the structure is,

Pr = f/>( 6.80 -=.._!_7.96)

.23
= f!>(- 3.455) = 2. 75 X )0-·4

EXAMPL~ The yield strength of HYSD bars (Grade Fe 415), X, follows

the normal distribution. Given:
= 468.9 N/mm2
P.x s; = 0.1
[a= 190 N/mm 2
Determine the probability of failure of steel.
Solution
ux = 0.1 ·-. 468.9 = 46.89 N/mm 2
Using Eq. (4.9),

::z (/) ( 190 - 468.9)

p; . 46.89
~" •P - 5.948) = 1.4X I0-9
/Determination of Alloll'ablc Stress
The allowable stress can be fixed for a given reliability or probability of
failure of the material. lf the strength of the material follows the normal
distribution, then Eq. (4.9) can be rewritten as

./~ -·· P.x = f!>-l(p1) = k (4 .10)

vx
Substituting ax =--= 8x~-tx, the above equation becomes
/tx I
J,, ~-= I+ k c_;.
 107

/Lx and ax are obtained from the field data and hence they are known.
Knowing JJ. ¥ and ~x and k for a given Pf, the allowable stress can be fixed.
Factor of safety, v, is defined by the convention of WSD as the ratio of
the ultimate stress or yield stress to the working stress of the material.
Hence

V=- =
/LX I (4. ll)
fa (l + k8x)
The fixing of allowable stresses for a given reliability is illustrated with the
following examples.
EXAM,P-!:E~t is given that the ratio or the mean value of the cube
stren'gth of M 15 concrete (design mix) to its characteristic strength is 1.4
and the coeflicient of variation of the strength of concrete is 0. 18. Determine
the allowable stress for the probability of failure of concrete equal to to· 3 •
SoltJtioll In the case of concrete, the allowable stress is fixed as a fraction
of the characteristic cube strength of C"'llCrete. For flf '-" Io-J, k "-' -3.091
(from tables).
Let
fCII =o the characteristic cube strength of concrete
X == the strength of concrete in the structure
The mean value of the cube strength of concrete is gi ven a~ 1.4 .fw. Hence
Eq. (4.11) becomes
[~ -'~ 1.4(1 + kf>x)
f eu
As the allowable stress in the element of the structure is to be fixed, the
prism strength (that is the strength of concrete in the structure) is to be
used and the above equation can be written as

!a
}CU
= (0.67)(1.4)[ 1 - - (3.091 )(0.18)]

= 0.416
If the specified cube strength of concrete is 15 N/mm 2 , the allowable stress
for PI = 10-3, is
fa = 0.416 X 15 = 6.24 N/mm 2
Similarly, for various values of Pf, the allowable stresses can be calculated
for a particular characteristic strength. They are given in Table 4. 7.

TABLE 4.7 Factor of a safety aud allowable stress for A! 15 concrete (design mix)
for different values of probability of failure

to-• JO-' IQ-6 to-•

-3.091 -3 .7t9 - 4.265 -4.754
v 2.40 3.22 4.59 7.39
_, fa (N/mm1 ) 6.24 4.65 3.27 2.03
108

It can b\! seen 1'1 0111 the above tabk that as fi t decreases. the ;lilo\vablc
stress also decreases as expected.
EXAM~4.4 In the case of steel, the allowable stress is lixcd as a fraction
\lf theyield stress. From the data it is found that the ratio of the mean
'alue \)r the yield strength of steel to its characteristic strength is 1.13 (for
Fe 415- Table 4.3). The overall variation in the strength of steel has been
found as 0. 1 (Sec . 4.6). Determine the allowable ~tress for steel for
Pr ' I0- 3 •
Solution The factor of safety for HYSD bars. using Eq. (4.11 ), can be
written as follows:
Let
x· '--" yield stress of the material
f.;. ·- the characteristic yield strength of steel
lt is given that
!lX ·- I.JJ_f;,
Using Eq . (4.1 I)
t; l
v--' j~ = 1.13(1 .! kSx)

for PI = JO- J, k ~- - 3.091

and 8x '-'" 0.1. Thus,
fv I
v '-= '--'-
.f;,
= Cl3 ~3"'.09I)(0.1 >J
g
fnfv -= 1.2 ]
For Fe 415 --grade steel, J.;. ·-· 415 N/mm 2• Hence the allowable stress f(•r
Pl =' 10 3 is

r. a- m
41 5 ' ~'
, 4 N ,mm
.l2 ' 2
'

If p1 '-~ 10--4, the value of allowable stress can be similarly calculated and
it is equal to 294.5 N/mm 2 •
If one takes the guaranteed yield strength itself as its mean strength, then
the value of allowable stress for f'f ~= I o- 3 is
fn '-"' ,!;.[! --- (3.091 )(0.1 )]
'·= (415){1 - 0.3091) = 286.7 N/mm 2
It must also be noted that the safety has been calculated based on the
yielding of steel (i.e. if steel yields, it is considered as a failure). However,
the actual failure (that is by breaking of steel) occurs at a vaiue of 1.2 times
the yield stress of material. Hence, the actual safety available is more.
The collection of different data on the strengths of different materials
T 109

and on ge metric parameters have been discussed, and the statistics of

variables based on the actual field data and the published works for Indian.
c nuit i ll S have bee n pre ented . By and large, the talis1ica l descripti ons
suggest I r. re baed on pu blished work . . It is re ognizeu th at th e knowledge
of the behaviour f materinl is c ntinually evo lving and the mean , vari-
ances, and distri butio n. of the variab le may be changing a more a nd more
data is collected or when da ta is updated. Re earch wo rkers also have not
used th e ame model und parameters fo r reliab il ity st ud ies during the las t
tw decades. For the amc data there may be a number of di stributi on.
wh ich appea r to li t the data equall y well. Ex trem e ca uti n hould be
exerci eel if the type 0f distributi on is chose n on the basis .o f ample da ta.
A better or preferab le appro·lch i. to make u e f physical rea oning about
th e nature or each variable to guide th ehoi cc o f rhe distribution. In
engineering r>roblem , most of the tim e we may have to resort to empiri-
cally fi tted di tri buti ns. It is to ben lted th at the var iables t hat have been
discussed are the basi var ia bles of a re istance variable of the structure.
Hence it is important. a nd to be recogni zer!, th a t lhe elected models mu st
be simple, convenient , a nd reasonably good for th ese bas ic vari ables.
Modell ing of th e resistance varinble of ·1 . lructural element and a stru -
wre is a diflieult tn k. The rcsi tance is a function f the se ba. ic vnriables,
viz. strength of mater iu Is, geometric parameters, etc. Getting field data for
the res i lance of an R column, beam or frame or steel elements and
structure in civil engineering is quite expensive and impossible . One may
have to resort to the simulation technique (to be discusseJ in a later chapter)
or physica l reasoning to choose appropriate models. We have already dis-
cussed diliercnt models in Chapter 3 and also the conditions und er which
they arise. Th y may be helpful in choosing a model. Normal, lognormal,
Wci bull. bew, and sometimes gamma distributi ons are generally u cu to
characterize the resistance of a structure. Again, it is important that the
selected models must he simple and convenient, otherwise it will lead to
difficulties in evaluating the reliability 0f a structure.
The estimation of parameters is important as the accuracy of prediction
depends on the parameters estimated from the data. The methods that are
generally used are:
(i) method of moments
(ii) method of maximum likelihood
(iii) mean rank plot-graphical procedure
The method of moments is the simplest. The graphical procedure is easy to
apply for simple probabil ity distributions. The method of maximum likli-
hood is difficult t apply as it often involves iterative calculations. However,
it is supposed to be the best method as the estimators have all the desirab le
properties, viz. unbi asedness, efficiency, and consi tency. The description of
the methods are beyond the scope of this book. Readers should study and
refer to any of the standard books (4.3. 4.14).
110

REFERENCES
4.1 Mirza , S.A .. M. Hatzinikolas and J.G. MacGregor, "Statistical Descriptions on
Strength of Concrete", Journal of Struct. Div., ASCE, Vol. 105, ST6, Ju.1e 1979.
pp, 1021-1037.
4.2 Ranganathan, R., "Reliability Analysis and Design of Prestressed Concrete Beams
a l Different Limit States", Ph .D. TMsis, l.f.T., Kanptlr, May, 1976.
4.3 llcnj umin, .I .R. and .A. Cornell , Prohahility, Stllli.rtic.r and Decision for CM/
F:nr: inct'l's, McG raw-Hill . New York. 1970.
4.4 Ranganat han, R. and .P. Joshi, "Stali tical Analysis of Strengths of Concrete
and Steel and Dimensiona l Variations", Report No. D.S. and T: 4(1)/8.1/STP ..
J/1/1. , ivil Engg. Dept., f.l.T., Bombay, March 1985.
4.5 Dayaratnnm, P. and R. Ranganathan, Statistical Analysi.r of Strenutll of Concrete,
J)uilding and Em•ironment, Vol. 11, Pergamon Press, 1976, pp. 145- 152.
4.6 Alex Mathew, "Probabilistic Analysis of CE:mcrete Poles", M.Sc. (Engg.) Tlresis.
Calicut Re gional Engg. College, 1980.
4.7 Mirza. S.A. and J.G. MacGregor, "Variability of Mechanical Properties of
Reinforcing Bars', Joumal of Simer. Div., ASCE, Vol. 105, STS, May 1979,
pp . 921- 937.
4.8 Rangannth an. R . an d C.P. Joshi " Vnriations in Strength of Reinforcing Steel
Hars", Joumfll of /Itt> lnstifrltion of Enginet>r.r (India). Civil Engg, Div., Vol. 68,
Muy 1988, J'P. 309- 312.
4.9 Dayaratnam, P., R. Rangnnathan, et at., "Report on Brick and Reinforced Brick
Masonry" , Project Report No . DST/427/4, Nov . 1982, Civil Engg. Dept., IIT,
Kanpur.
4.10 Mirza, S.A. :~nd J.G . M:~cGregor, "Variations in Dimensions of Reinforced
Concrete Members", Joumal of Stmct. Dil•., ASCE, Vol. 105, ST4, Apri11979,
pp. 75 l·-766.
4. 11 Rangnn athan , R. a nd .P. Joshi , " Variations in Dimensions of R Members",
Joumtrl of the Bridge anti Stmctuml Engimu.lr, Vol. 16, ept. 1986, pp. 1- 10.
4. 11 Ang., A. H. and C.A . Cornell , "Reliability Ba sis of Structural Safety nnd
fZ ign" . Joumal of Struct. Dil'., .1SCE, Vol. 100, ST9, ept. 1974 pp. 1755- 1769.
4.13 Oaynr:HJ Him , P. a nd R. Ranganathan , "Allowable tresses and Load Factors
Based on P~obability Theory", Jotll'lltd of the Institution of Engineers (India), Civil
Engg. Div., Vol. 58, July 1977, pp. 20-25.
4.14 Siddall. J.N., Probabilistic Engineering Design, Marcel Dekker, New York, 1983.

EXERCISE
4. 1 The cuhe I rength of M 20 concrete foll ows the normal distribution with par:~ ­
mc t ~r~ t' = -9.16 N /mm~ and " = 5.49 What is the characteris tic strength of
concrc1e? (A11s. 20.16 N /mm 1 )
4.2 The yie ld strc n ~;th of steel follows the I gnorma l distribution with I' = 295.3
N/mm• 1111d" = 16.24 N/mm•. lf the specified strength of steel is 235 N/Qim 1 ,
detc nni nc the chnrac teristic strength of steel. ( A11s. 269.4 N/mm')
4.3 If the ratio or the mcw1 value of lhe cube strenglli of MIS concrete to its characteristic
strength i~ 1.5 1. rutd the coollicicnl of variation of the strength of concrete is 0.24 ,
dctennim: th e nllowablc stress for a reliability of 0.9999? (Ans. 1.61 N/mm 2)
4.4 If the yield strength of steel follows the no rmal distribut ion wilh I' = 468.9 N/m mD
and" = 46.R9 N /mm' , determine the nllowoble stress for a reliability of 0.9999.
(A ns. 294.5 N(mm'l
4.5 The flexu ra l strength (ultimate) of n prestressed concrete beam follows the normal
d istribution wic h lhe coemcicnt of variation being 0.05. The beam is subjected to
dead load and live load . Assume the loads are deterministic. If the combined load
 111

factor, r,. is defined as the ntlio of the mean value of the strength of hc: urn to the
moment due to working loads. what is the value ofF" for a desired re liability of
0.999Q? (!Ins. 1.22!!)
46 If !he ratio or d.:ad load to live load is 0 .5, and load factor for dead load is 1.2.
''hat is the load factor for live load for a desired reliability of 0.9999'1
(Alii". 1.193)
 5
Probabilistic Analysis of Loads

5.1 GRAVITY LOADS

5.1.1 Introduction
The accurate evaluation of gravity loads and the proper assessment of the
maximum loads that a structure will have to carry during its lifetime are
very important for a safe and economical design. After the advent of high
speed digital computers, accurate techniques are available to analyse and
design any complex structure under given loads. However, the state of
knowledge about the analysis of loads is not comparable. The loads remain
an estimate based on experience, judg.e ment, traditi.on, trial, and error.
Recently, during the past 15 years, considerable attention has been drawn
to the measurement, nnalysis, and modelling of loads because of the increa&ed
familiarity of the engineers with the probabilistic and statistical methodology
necessary to treat the load phenomenon in the quantitative manner, which
engineers expect.
Loads on structure are stochastic in nature. They vary with space and
time. This spatial and temporal variability is to be taken care of in the
design. In recent years, a significant amount of live load survey has been
conducted in many countries (5.1-5.9). At the same time, the trend has
been set up to develop probabilistic limit state design and reliability based
codes. The characteristics of the loading is probably the most important
parameter to a reliability based analysis and design . In the formulation of
reliability based codes, considerable attention will have to be focussed on
·the acquisition of reliable load data of a form suitable for the estimation of
key statistical parameters. Concurrent to this, there is a growing awareness
to develop probabilistic models and estimate the statistical parameters. The
study of floor loads in buildings with respect to how live loads are measured,
analysed and modelled, is presented.

5.1.2 Load as a Stochastic Process

Loads or actions in general are the forces acting on the structures due to
external influences (self weight, superimposed loads, snow, wind and wave
loads) and imposed deformations (differential settlements and temperature
variations). Loads are subjected to random variations in magnitude and
position with time. Loads are, therefore, d.<.,;,ribcd lls time varying, free
 113

positioning, and dynamic effect producing and hence loads are to be

modelled as a stochastic process.
A single time history representing a random phenomenon is called a
sample function. When this evolves in time, it leads to a process. A stocha-
stic process is the collection of all possible sample functions, which the
random phenoti:enon might have produced.
A sample function of a continuous time varying stochastic process of
load X(t) is shown in Fig. 5.1, in which x(!t) is the magnitude of a time
varying load X(t) at time 11. This x(tl) is called the arbitrary point-in-time
load . It is simply the load that would be measured if the load process were
to be sampled at some time instant, e.g. in a load survey. This load is a
random variable. If this is designated as X, the PDF of X is shown in
Fig. 5.1. In the same figure , if Xmn' is represented by the random variable
Z, then the PDF of Z.fz(z), will be as shown in Fig. 5.1.

~co
..:
N

I
:~max
I

t,
Timeo • t

FIG. 5.1 Continuous tima varying load

In the case of reliability study, the treatment of load as a stochastic

process is inconvenient. For practical reliability analyses, it is necessary to
work with the random variable representation of load rather than with the
random process representation (5.10). Again, in the case ofreliability study,
the designer is interested in the value of the maximum load that is likely to
occur during the life of the structure. This load is called lifetime maximum
load. Ultimately one is interested to know the probability distribution of
this load . This may be physically interpreted as the distribution that would
be obtained if the lifetime maximum load were measured in an infinite
number of identical structures (5.11). In later sections, we will see how we
achieve this. ,
Gravity loads are divided into dead loads and live loads. Live load is
again divided into (i) sustained load and (ii) transient load or extraordinary
load.
 114

0.1.3 Dead Load

Permanent loads are considered as dead load. This is mainly the weight of
the structural system. This may undergo a little reduction because of wear
and tear during its lifetime. This is negligible and can be ignored. Dead
load may undergo increase because of the addition of some partition wall
or covering during the life of the building. These may be rare events. This
also induces a modest change only. Hence the dead load· can he assumed
to remain constant in time throughout the life of the structure. This is
depicted in Fig. 5.2(a).
x x(t)

,._
~
0
..-><
(a) o~ad load

X X \)

-
~
><
><
< (b) Sustained Load
: )( (t,)
I

t1

X (t)

I c l Extraordinary Load
FIG . 5.2 Types of loads

The total dead load to he supported by a structure is generally the sum

of self~weights of many parts. Hence the dead load is modelled with a
normal probability distribution . The variability in deaJ load is strongly
 115

affected by the weights of nonstructural items, such as roofing, partitions,

etc. As there is a tendency to underestimate the total dead load, it is
assumed (5.10) that the ratio of the mean load to nominal load is 1.05, and
the coefficient of variation is 0.10 for code calibration.

·...6.1.4 Live Loads

Live loads may in general, be defined as any load produced by the occu-
pancy of the building. Nonpermanent gravity floor loads arising during the
service life of the buildings are considered as live loads. That means, live
loads include the weight ofpeople and their possessions; furniture, movable
partitions and other portable fixtures and equipment. The total live load on
a floor is considered under two components, viz. (i) a sustained load compo-
nent (long term), (ii) extraordinary load (transient load) component.
Sustained Load
A sustained load is the load of furniture, equipment and other loads needed
for the activity and the normal personnel involved in the activity. Sustained
loads shown in Fig. 5.2(b) may change at discrete times, but inbetween
changes, remain relatively constant. A change at discrete times may be
thought of as change due to change of occupancy (tenancy). The variation
of load inbetween changes is due to the changes which a normal activity
brings. New pieces of furniture may be added or exchanged or shifted, and
the contents in desks and cabinets and other storage places vary. The
persons who are involved in the activity are not present all the time which
brings a variation of the load. As stated earlier, this variation between two
load changes is limited and small compared to the total load. Hence a cons-
tant load between load changes is assumed in the load analysis. It may be
noted in Fig. 5.2(h) that sustained loads may be entirely absent for a certain
duration. This may be considered as the ~ime gap during change of tenants.
The sustained load is the load usually measured in Jive load surveys. This
is referred to as the arbitrary point-in-time- load, Lapt· The PDF of this load is
also shown in Fig. 5.2(b). This load is a spatially varying random function.
This is assumed constant in time within a particular change of occupancy.
It is therefore known as the long term load. The load changes with change
of occupancy are assumed to occur as poisson arrivals.

E:rt1·am·dinary Load
This arises from infrequent clustering of people above and beyond normal
personnel load. That is, the extra personnel load. This extraordinary load
(EL) is also due to the event when 111any pieces of furniture or equipment
have been gathered together in one place at some instant of time, for
example, at remodelling events. The EL is very unpredictable and it occurs
with relatively high intensities and in short durations (in most cases a CO\Iple
of hours). The term short duration is used in the sense that their durations
are very small relative to permanent and sustained load. Hence they create
a spike on the lifetime history of the load as shown in Fig. 5.2(c). It is very
116

difficult and almost impossible to get data on EL. It cannot be measured in

the regular live load survey.
The total load is therefore split up into three parts. This is a simplified
model. The division is mainly on the difference in the time history between
the loads. As the dead load, already discussed, has been considered as con-
stant in time and can be modelled with a probabilily distribution, it can be
combined with other loads. In future, live load only will be discussed.
Li1•e Load Survey
The development of new codes, based on the reliability theory or prob-
abilistic limit state design, needs more and more information about loads
based on the actual field data. This has initiated the conduct of load survey.
During the past decade, numerous load surveys have been conducted in the
U.S.A., Europe, Canada,' Sweden, Australia, India, etc.
J . Bryson and Gross (5.2) have developed the methodology of load
surveys. The live load survey is the process of measuring the actual floor
loads, Lapt, and collecting the extensive scientific and systematic data, and
information, such as (i) building data which includes geographic region,
location, height and number of storeys, age, type of occupancy, floor plans
of building, layout of framing systems, number of rooms/bays, floor area
of building, etc., (ii) occupancy data giving information about the types of
firm, spatial orientation and duration, (iii) room/bay data, which incor-
porates details about the floor level, room number, location of the room,
room use, room size, floor area, openings, surface finishings, floor cover-
ings, occupants including number and weight, item description including
location, contents and weights, etc., (iv) extraordinary load information
about occasions of persons gathered, frequency, fmniture stacking occasions,
painting and remodelling, etc. - 1
The scientific live load survey provides a sound statistical basis for (i) the
adoption of an appropriate probability model for live loads, (ii) the proper
assignment of parameters to the probabilistic model, (iii) the refinement of
probabilistic load models, (iv) better understanding of the randomness of
live loads, and (v) the modification of the existing loading standards.
Simple Stati.~tical Analysis of Lb•e Loads
Before we consider the rigorous statistical analysis of floor loads as an area
dependent random process, let us first understand the simple treatment of
the load analysis.
Assume that live load ~urvcy has been conducted in a building and the
position and magnitude of loads are known on each bay (or room) of the
building. Assume constant area. The floor load intensity (FLI), Q, is the
total load acting on a bay (or room) in a floor divided by the floor area of
the bay (or room). The actual live loads (measured in load surveys) may
have any random positions and distributions. From the design point of view,
the effects of actual live loads (i.e. stress resultants) developed in the floor
slab or supporting beams and columns, are important. Therefore, it is
 117

necessary to convert the survey loads into uniformly distributed loads. This
uniformly distributed load intensity, which would produce the same load
effect as the actual set of loads, is called the equivalent uniformly distri·
buted load (EUDL). Let EUDL be designated as L. Hence the set of point
loads on each bay, with actual magnitudes and positions measured in load
surveys, must be transformed to EUDL by using influence surface methods
or energy methods or finite element methods, taking into account the
boundary conditions and the configuration of the supporting systems. Once
a probability model is selected and the parameters established for L, the
characteristic load, LA, can be fixed. If Pk is the probability ofa load greater
than Lk, then
Pk = P[L > Lk] = l - P(L :::>; Lk)
= I - I·L(Lk) (5.1)
1
or Lk = Fi (1 - Pk) (5.2)
If the occupancy does not change during the lifetime of the building, the
above calculated load gives the lifetime maximum live load with a prob-
ability of its excecdencc equal to pc
For live load on buildings, it is usually assumed that the occupancy
varies a few times during the lifetime of a building, T, in a completely inde·
pendent way. Assuming that the whole building is occupied by only one
tenant (i.e. single tenant model) at a time, Jet the building be occupied by
N tenants during the lifetime of the building. The live load during each
occupancy is a random variable. Let L1, L2, ... , Lt ... , LN be the random
variables representing the maximum live load intensity (EUDL) during each
occupancy. lt is assumed that the live load does not change with respect to
time during each occupancy. If FL;( ) is the CDF of L; and FL111( ), the
CDF of the lifetime maximum live load, L,,, then the probability of L, less
than or equal to a particular load, say characteristic load Lk, during the
lifetime of the building is given by
P( Lm ~ L") "'" Pl(LI ~ Lk) n {1.2 :S: Lk) n ... n P(LN :::>; L~c)]
If L; urc as ~ umeJ as statistically independent, the above equation becomes
P(L, 'c. L1..) = P(LI :S: Lk)P(L2 :S: Lk) ... P(LN :S: Lk)
li,(Lk) = Fi. 1 (Lh)FiiL~;) ... l·L 1,,(Ld
If L; are identically distributed, the above equation simplifies to
h111 (L~;) = [f'LI(L!c)]N (5.3)

where FL (Lk) is the probability distribution reflected in a histogram of

1

live load data measured during a short period of time (initial fitted dish+
bution for L nvt). If FL 1( ) has an inverse at Lk = (I -pk)IIN, then
(5.4)

The above L~c 1s the lifetime maximum live load for l\' tenancies and P;. is
 118

the probability of live load exceeding Lk. Hence for a given number of
tenancies and a specified value of pk, the value of Lk can be calculated from
the initial fitted distribution for the live load. This is illustrated with the
following example .
.J r~· AMP~ From the statistical analysis of live load survey, it is found
th:1t live load follows the lognormal distribution with parameters

L= 1217 N/m2 Gin L = 0.368

Determine the characteristic load for Pk = 0.05 if (i) there is no change in
tenancy and (ii) the building is going to be occupied by 5 tenants during
the lifetime of the building.
,')'olution
Case (i):
Using Eq. (5.2), the characteristic load is given by
L1, -'--' Fii(l - ph)~ FL'i(l --- 0.05)
FL(Lk) "-= 0.95
Since L follows the lognormal distribution, using Eq. (3.94),

w[ In (Lk/L)] = 0.95
Gin L

L = f exp [aio LW- 1{0.95)]

~"' 1217 exp l0.36Wl~"~(0.95)J

c= 2220 N/m 2
Case lii ):
It is given th<Jt N •- 5 and p~o · 0.0). UsingEq. (5.4), the value of L1. during
the lifetime of the building is
1
Lk ~ Fi l(l -· 0.05)ii 5 1

-' Fi 1(0.9898)
= L exp [alnt<P· 1(0.9898)]
- - 12 I 7 cxr l0.368cP 1(0.9898)]
• ltl60 N/m 2
Similar!), the values of /.1, for different numbers of tenancies arc calculated
and given in Table :'i.l. It is seen from the table that Lk increase~ for a
gin' II value ur fil,, :11HI /.1, decrease~ as fil, increases for a giYcn value of N.
Area l>epelltit'llt Su.l'tuincd ],ot/11 /ntell.l'ity Model
ln the last section, it has been :1ssumed that the bay or room area is cons-
tant and the llul'r luad doe'> nt't depend on the are:1, i.e. 110t as a function
or the :1rea. However, it i' 11elle,,ablished that the noor lo:td depends on
 119

TABLE S.J Lifetime maximum live load for different number of

tellallcies-Example 5.1

Period Lifetime of Lifetime maximum load for

of building N Pk- 0.05 Pk- 0.10
tenancy (years) (kNjm•) (kN/m 1)

5 25 5 2.86 2.57
5 30 6 2.92 2.65
5 40 8 3.04 2.76
5 50 10 3.17 2.76
5 100 20 3.43 3.13
10 50 5 2.86 2.57

the area. Live loads vary from building to building, floor to floor, bay to
bay, point to point, and also time to time. To quantify these variations and
uncertainties, to some extent rationally, the instantaneous live load survey
data of arbitrary point-in-time loads on floors of selected bays of selected
buildings have to be analysed to model live loads with certain assumption s
and simplifications.
Statistical Assumption
The load intensity on a floor can be characterized as a stochastic process
which is assumed stationary both in space and time.
The ~ssumption of stationarity in space implies that the load in build-
ings, used for the same type of occupancy, can be represented with the same
statistical distribution. This assumption is generally used, and is necessary
so that with a proper selection of the buildings out of the whole population,
good estimations of the statistical properties can be achieved.
The assumption of stationarity in time implies that the statistical distri-
bution of the load from one point in time to another is the same. This
assumption is needed. It is not possible to conduct a continuous load survey.
The procedure of analysis of live load is to start with the preposition of
a probability model for the load intensity. From this, a probability model
for the load elfcct or the equivalent uniformly distributed load (EUDL) is
Jerivcd.
Load Intensity
The sustained load intensity at any location on a floor of a building 1s
modelled as the superposition of
· (i) the main trend,
(ii) the periodic components, and
(iii) the random fluctuations
According to the assumption about stationarity in space, a constant mean
load intensity is chosen. Hence the main trend is the mean load intensity,
which is assumed to be constant for a type of occupancy. It is to be noted
that the mean load intensity will be diiTerl.!nt for diiTcrent types of occupancy.
That is, say between hospillll buildings and otlkc buildings.
120

The periodic components are the variations in the load intensity around
the mean due to different buildings, different floors, and different bays.
Random fluctuations take into account unknown uncertain deviations
from the mean load intensity.
The load intensity model is assumed to be noncorrelative. That is, the
correlations between load intensities from floor to floor, and bay to bay, and
point to point which have a very little effect on the total load (or load effect)
are not considered. Therefore, the correlations arc neglected for simplicity
and hence the load intensity model is assumed to be noncorrclative. With
the above assumptions, Pier and Cornell (5. 12) proposed a model for the
load intensity as
w(x, y) = m +· r + D(x, y) (5.5)
where
1\'(X, y) '= the load intensity at any Jomtion Oll a bay Of a floor of a
building
m = overall mean load
r = a zero-mean random variable which can be split up to
represent di!Ierent variations
IJ(x, y) = a zero-mean random process which represents unknown
spatial variations.
The above model has been applied by various research workers (5.9, 5.13-·
5.17) anu it is expected that this will be the general method for the analysis
of sustained load. The r term may be split up into
!'bldg- representing builuing variations
I}'- representing floor variations
l'booy- representing bay variations

The split is justified if every building is occupied by one organization. Thai

is, a single tenant model is assumed. This is the case in most of the office
buildings. In case if a building is occupied by many organizations (this will
l•c in the ~:asc of tall buildings), then l'bldG can be considered as l'or 11 repre-
senting variations between organizations.
The sm;illcst structural unit used in the load intensity model is a struc·
tural bay. Hence the load intensity is integrated over the bay area to get
the Iota! load. It is recalled that the spatial load intensity has been assumed
as a noncorrclativc random process. However, the total spatial load over an
area is nssu1ned tn be dependent on the area. Hence the model is an area
dependent random process. Since D(x, y) is a zero-mean random process,

E[ (J JD(x, )') dr dy)j A] = 0 (5.6)

,I

' !Ill~ \ arian~:e ol' D(.Y, _r) i~ given by

Var (!>) =~ JJJJ (\,v IJ>(x. r), D(u, ;·Jl dx (rl' dud<·
A I
 121

Since the spatial load intensity has been assumed as a noncorrelative random
process, we have

Var (D) =~ Var [JJ D(x, y) dx dy]

A
2
= liD -¥ A (5.1)
A

where u~ is the spatial variance.

Let LD be the total spatial load over the area. This is dependent on the
area. Hence

E[Ln(A)J = E [f f D(x, y) dx ely]

A

=- 0 (5.8)

Var [LD(A)] = Var [JJv(x, y) dx dy]

A

(5.9)

Cov [LD(AJ), LD(A2)] = Cov [fJD(x, y) dx, dy, JJD(x, y) dx dy]

A1 A1
= 0 if A1 nA2 = ,P (5.10)
c/>, here, means null set.
Statistical properties of w(x, y) over an area can be written by using the
above derived results for the spatial load. Let L be the Iota! load, i.e. the
sum of the load intensity over any finite area. Then the mean value of L is

E[L(A)] = E [JJ
A
w(x, y) dx dy]

= E[mA] + E[rA] + E[D(x, y)A] (5.11)

As r is a zero-mean..random variable,
E(r) = 0 (5.12)
Using Eqs. (5.8) and (5.12) in Eq. (5.11),
E[L(A)] = mA (5.13)
The variance of L is

Var [L(A)] = Var [lj w(x, y) dx dy 1

= Var [mA] + Var [rA] + Var [D(x, y)A]
122

Using Eq. ,(5.9),

Var [L(A) ! ~~' 0 + a; ..J2 + a7JA / .-f 2

(5.14)
where a;is the variance or/',
The covO.Jriance of loads between two difYcrent influence area ~ :f1 and . b is

Cov [L(A,), L(A2)] = Cov [f J

.4,
w(x, y) dx {~~·. JJ ll'(x, y) dx dy]
,.,,

(5.15)
To obtain the unit load, VL, the total load over the area 1s uivided by the
area. :-fence
VL(A) ~- L(A)
A
Moments of unit load are

ELLh(A)] = E [ ~1_)] = m (5.16)

Var LVL(A)I ;;-"" Var [ L~:)] "- ' ~2 Var LL(A)]

Using Eq. (5.14), the above equation bec.:omcs
2
. 2 an
Var lUL(A)] = aA + A (5.17)

ov LL(A ,). L(A2)J

AIA1
Using Eq. (5.15), we have
Cov[UL(A ,), U L(A2)] = a~ (5.1 !))
So far we have not considered the load effect. This can be taken care or
by determining the coefficients with which the load should be multiplied to
get the load effect. The load effect is to be obtained by the influence surfaces.
Instead of integrating over the influence area, the influence surface is used.
As every load effect has its own influence surface, the theoretical load effect
can be obtained for any case.
The correct solution for the influence surface is very complicated. To
simplify the solution, two dimensional extension of influence Jines is used
(5.5).
Let
H = the load effect
Then the load effect over the influence area is

H(A) = JJ w(x, y) I(x, y) dx dy (5.19)

 123

The equivalent uniformly distributed load, L, thul produces tho same load
effect is obtained by di viding load etl'ect by the integral under the member's
influence surface. (This load L is also a function of A. Howe er, for con-
venience A is removed in the notation).

JJw(x, y), l(x, y) clx dy

L -- A (5.20)
- JJ l(x, y) dx dy
A

where I(x, y) is the influence surface function for the particular load effect
sought and A is the influence area over which l(x, y) assumes nonzero

1
values. The statistical properties of L arc ·

ElL] ~ E Jlj/(x, y)w(x, y) dx dy

l !fl(x, y) dx dy J
rJJ I(x, y) dx dy l
= m ifJ
LA
I(x, y) dx dy r
J
=m (5.21)
The variance of L which is a function of A, is

I
r JIV(X, y)I(x, y) dx dy l
J
Var [L] = Var

t A
~JI(x, y) dx dy

=
Var(H(A))
[JJJ(x, y) dx dy
A
r (5.22)

The variance of IJ(A) is

Var (H(A)) ~= Var [fIw(x, y) I(x, y) dx dy]

A

= Var [Ifl(x, y){m + r+ D(x, y)} dx dy]

A

= 0 + a~(JII(x, y) dx dyr
A

+ Var [JJ
A
I(x, y)D(x, y) dx dy] (5.23)
124

It can be derived (5.5) that

Var [J,,Il(x, y)D(x, y) dx ely] = (]~ I

.I
If2(x. y) """ dy (5 .24)

Hence the variance ot' L is

2
fJfl(x, y) dx dy
Yar [L] = n; + av ~
A~--------~ (5.25)
A [f J l(x, y) d.\' dy
2

Using Eq. (5.19)

Cov lL(Ad,L(;h)J = u~ (5.26)
Let
IJ l 2(x, y) dx t(l'

k = ~A.-------~~ (5.27)
(ff I(x, Y> dx dyy
A

The coefficient k is the mean squared influence divided by the square of the
mean influence; k is always greater than or equal to I. It depends on the
type of member, its structural configuration and boundary conditions, and
the type of response sought. k can be obtained for any load effect and it is
rclatively insensitive to load effect type (5.5).1t has been found by McGuire
and Cornell (5.13) and Sentler (5.5) that the values of k arc
k =" 2.04 for end moments in beams (interior bay)
k = 2.2 for column axial loads
k = 2.76 for mid-span beam moments
k = 1.98 for mid-span beam moments if the beam IS simply
supported
k =-~ 1.34 to 1.5 for mid-span moment of' a slab
Ellingwood and Culver (5.15) have taken an average value of 2.2 for their
analysis of loads. The analysis carried out by Rao and Krishnamoorthy
(5.7) shows that considering all load etl'ects, k varies from 1.92 to 2.46.
Hence we can write
E(L) = m (5.28)
2
Var (L) = ail = a; + a; k (5.29)

If we are considering the load effect for beams, the statistical properties of
L of a beam are
I:.:(L) = m
 125

as tlie influence area for a beam is twice the area of the structural bay; the
value of k corresponds to the corresponding beam effect (mid-span beam
moment, end moment ; mid-span shear, etc.) . Similarly, the statistical pro-
perties of EUDL of one storey interior column loading is
E(L)-= m
2
2 2 ao
a1. =a,+ 4A
as the influence area for one storey interior column is four times the struc-
tural bay area A. The value of k for column loading is 2.21 r5.5, 5.13). If
the interior column supports 11 floors, then
E(/.l = m
2
2
uL =
2
u, + 4uonAk
It is generally found that lognormal and gamma distributions closely fit the
data (Lapt) from load surveys (5.5, 5.8, 5.9, 5.18). However, since a constant
mean load intensity model has been assumed, the probability distribution
characterizing the sustaiucd load should have a reproductive property. The
gamma distribution has this property but not the lognormal distribution.
Maximum Sustained Loatl Intensity Model
The maximum sustained load, L.,, is the maximum of the various sustained
loads supported by a, given area during the lifetime of the building. That is,
this is the maximum load which will occur during the lifetime of the
building. This is also called the lifetime maximum sustained load.
The following assumptions are used in the stochastic analysis of L,
(5.12):
(i) The sustained load (SL) during each occupancy is constant, but this
value is random.
(ii) The stochastic load process of SL is homogeneous in time and space.
(iii) One tenant and one floor model is adequate.
(iv) The successive sustained loads on any area are independent and
represented by a probability distribution over the ensemble.
(v) The probability distribution of occupancy durations are independel\t
of each other and do not change with time.
(vi) . When an occupancy change occurs, it occurs simultaneously every-
where over the area A.
(vii) The successive sustained loads follow the gamma distribution.
(viii) The load changes occur according to the Poisson process.
(ix) The duration of occupancy is exponentially distributed.
(x) A fixed number of changes occur during the lifetime of the building.
Let
Lm = maximum sustained load during the lifetime T of the building
126

T = duration of sustained load, i.e. lifetime of the building

L(t) = sustained load on the floor at timet, i.e. instantaneous SL
Then lifetime maximum load is
Lm = max [L(I)] 0 ~ I~ T (5.30)
Lm is also a random variable . The cumulative distribution of Lm is
h 11 k/.) = P[Lm ~ ll] {5 .31)
If the number of occupancy changes is N, then (N + I) is the number of
occupants (tenants) who haveoccupie I the building during T years. Hen ·e
(N ·!- 1) load values occur during the load history. It has been assumed
that the SL is constant during each occupancy of the building and has a
distribution h(oc). Hence if the building is subjected to N occupancy changes
during the time T, then (N + I) values of L will be observed during T.
This set of(N -1 I) values can be considered as a rand m sample. If it is
assumed that the number of occupancy changes, N, is known, i.e. a constnnt
or fixed, then
F1. 111(oc) ~ P[max. load ~ IX]
~ P[all (N + I) loads ~ o:J (5.32)
Since successive sustained loads are assumed to be independent, and identi-
cally distributed, the above equation becomes
FJ.m(a) = [FL(a)]N+I (5.33)
However, the duration of an occupancy of the building is not determi-
nistic, i.e. varies randomly. In such a case, N is a ,random variable and the
CDF of Lm is
co
FL 111 (a) = E P[(N
n~o
+ 1) loads ~a IN = n]P(N,...., n)

As successive sustained loads have been assumed to be independent,

<0
FL111 (a) = E [FL(1X)]"'+ 1P(N = n) (5.34)
n=O

It has been assumed that the number of load changes in a period of time
(0, t) occur according to the Poisson process with mean rate of arrival, v.
Hence
e- v'(vt)"
P(N = n) = - ,-, - 1
(5.35)

using this in Eq. (5.34), the CDF of Lm during lifetime Tis

oo e-vt(vT)n
FL (IX) = E [F£(1X)]"~' - -'---'--
"' n~o n!
= FL(1X) ie-vT [FL(~X)(vT)]n
n-o n!
= FL(iX)e-vT exp [vTFL(1X))
= FL(oc) exp [ ~vT{l - h(1X)}]
FM high cumulative levels; the above equation ·can be writte1las · 1:_-,.:,
FL,(rJ.) = exp [-:vT{<I ·- FL(oc)}] (5.36)
~ -- IIT[I - FL(ot)] (5.37)
Hence if the ptobability ~listripU:tion of t~e sustained load at any arbitrary
point-in-time (obtained from load survey) is known with its parameters, the
c~mulative probability distributiqn of maximum SL can be obtained.

EXAMPLE 5.2 From the analysis of the live load survey data, it is known
that (5.9)
E(L) =m= 717.3 N/m2

a2L = 2661
. •
-1 · ~690000
A
k

Calculate -the maximum sustained load at 0.932 fractile o:e. FL,;,(«) =- o~932)
for the following given,conditions:
(i) FL(«) follows the lognormal distribution
(ii) 11 = 1/8, T := 64 yrs
(iii) A ~ 27m 2 ·
(iv) k = 2.2
Solution It is given thnr
Using Eq. (5.36)

,,
Substituting the values of 11 and T,
exp [ -8{1 - FL(oc)}l = 0.932
FL(Ii) = 0.991
Using the ~iv~n probability distribution and parameters of L,

. . FL(f/.) = [ In (ocfi.)] - 0;991

· Ufn L

The parameters L and a1n z. are estimated as fo llows:

For A = 27 m2 , k = 2.2
1 69
a}. = 2663 + ( ~~ 00 ) 2.2
=;: 140366;7 (N/tn2)2
aL = 374.6 N/mi

aL = 374 •6
717.3
· o s·22
•
128

Using Eqs. (3.91) and (3.92), the paramete rs of L arc

Din 1. =lin (0.522 2 + I )li/ 2
= 0.491

~/, = 717.3 exp ( (0.4291 p)

= 635.8 N/m 2

Using the calculated values of Land Din L ,

rp [In (7./635.8)]
0.491
-= 099 1
.
"'- = 2048 .9 N / m~
Th[s is the IHaXimum \' alue of the lifetime sustained load with the pro-
b:lhility oi' its exceede1Ke during the lifetime of the building being
(1 -- 0.932) = 0.068
ExAMPLE 5.3 For the same t:'(ample , calculate the maximum sustained load
if L follows the gamma di stribution .
Solution The parameters of L following the gamma distribution are
( Eqs. 3.100 and 3.101)

>. = _!!!___ 7 17 · 3 ,---, 0.0051

(D ,y (374 .6)2
k = >.m = (0.005 1)(717.3)
= 3.658 N/m 2
From the previous example,

F,.(r~-) =- r~(k~«) = o.991

Using Pearson's table, it is found that the value of rx = 1870 N/m 2.

For the code calibration or the reliability analysis of structures, it may be
necessary to know the probability model of Lm with its parameters. It is
also of interest to know the expected value and variance of Lm for the
purpose of structural design. Approximate formulae for the mean and vari-
ance of Lm may be derived (5.15) by fitting a Type 1 extremal (largest)
distribution to the upper load fractiles and calculating the mean and vari-
ance of the fitted Type 1 distribution. This involves
(i) the calculation of the values of Lm at two fractile levels in the upper
tail, say Lnr = 0.932 and Lm = 0.992 for various values of A
(ii) the calculation of the parameters u and rx of the assumed Type 1
distribution for each area
(iii) calculation of the mean and variance from the calculated values of
u and rx for each value of A
 129

(iv) the plotting of the values of E(L,.) and Var(Lm) with the correspond-
ing values of A.
(v) the fitting of a suitable curve to these points (may be the least square
fit) connecting (a) E(Lm) and A and (b) Var(Lm) and A.

Transient Load
This load includes (i) the weight of the probable assembly of persons dur-
ing the ofr1ce party or get together functions or some othe r activity, (ii) the
weight of the probablr. accumulation of equipment and furnishin g during
remodellillg of the premises, and (iii) the weight of the probable storage of
the materials. Normally, the concentration of people in combination with
the sustained load causes the highest load. Because of this only, the activity
of persons is generally considered. Again, the clustering of people ab ve the
normal personnel load only is considered as the normal per~o nnel load,
which is the load of persons n rmally present in th ' ucli~ity already t:on-
sidered as one part of the ustained load.
The knowledge of transient load is very limited . Very few transie nt live
load surveys have been carried ou t because o f thl: difficu ltie. ill\o lvcd in
this type of survey. Transient I nels are tu b~ ob tained by wnducting
surveys continuously in time. This would give ne•;c .. a ry dal<• ab1 ut the
magnitude anq the time a peel of transient loads. Thi rrocedu rl! is, lww-
ever, difficult to employ. The other way of coll.ccting the data i. through
questi oning about the transient load events in the past. This method may
be easier but less accurate and may bring many uncertainties. The trans ien t
load occurs for a short time and is commonly modelled as a Dirac delta-
function with magnitude equal to the intensity of the maximum lt1ad
applied during the event. The transient load occurs instantly and is assum-
ed to arrive as a Poi son event. Each event is modelled by a random
number of randomly positioned and sized load cells. occurrin g random ly in
space. The EUDL as o iated with a n e tra rdinary load B is assessc I by
modelling the load evenl as a erie of r 111domly distrib uted load ce ll s,
each of whiob co ntains a lu ter of loads. The mod\l l i. h:l~cd on Poisson
occurring independent events, each f neg li gib le tlur(lti >n. Ba ic compo n-
ent loads Q (weight of ingle concentrated load in the c II, i.e. weight of
sin gle person) are assumed with specified mean value I'Q and va rian e v~.
Each load cell contain a random number R of compom:nt loads (i.e. R is
the number of loads per cell, i e. the number ol' person~ in one load cell)
with mean fi-R and variance a~. The number of load cells in a gi\'en area A
is assumed to be Poisson distributed with parameter,\, which is the mean
rate of load cells in A. Q is generally assumed to he independent of A. If it
is assumed that Q and Rare independent, the mean and variance of Bare
given by (5.13)

EiB] = E[QRl" (5.3~)

. A
130

H it is assumed that Q and R are independent [using Eqs. 3.80(a) and

3.80(b)], then

E[Bl = Ap,~P,R (5.39)

Var[B] = aBl =
kA(p,l. aR2
Q
+ i-Ll a2 l
R Q - - ao R
2 o 2)
(5.40)
A2 -

It may be noted that eve11 though the transient load events are probably to
a certain degree area dependent, a constant mean load intensity model is
assumed and the random process is made dependent on the area to reflect
the fact that a high concentration of people is more likely to occur in
;,;mall areas than in larger ones. The probability distribution of B is gener-
ally assumed to be gamma (5 .5, 5.15) as the gamma distribution has a
reproductive property. An exponential distribution has also been suggested
by Sentler (5.15).
Life Time Maximum Transient Load
The distribution of the lifetime maximum transient load, B,, is obtained in
the similar way used for the sustained load . The occurrence of B is assumed
to be Poisson with mean occurrence rate of v. Hence the CDF of B, during
the lifetime Tis given by
Fam(oc) """Fo(oc) exp [- vT{l - Fs{oc)}] (5.41)
where Fam(a.) = CDF of Bm
Fa(a.) = CDF of B
Maximum Total Load Model
Two types of live load, namely sustained load and transient load have been
discussed. The total live load, which is some combination of the above-
mentioned live loads at any instant , is of interest. Based on certain assump-
tions, the total live load is derived.
It is assumed that the sustained and the transient loads are independent
of each other in time and space.
As the live load has been considered in two parts, of which one is conti-
nuous in time, the total load is a two dimensional stochastic variable . The
assumption of independence simplifies the problem as the joint density
function is the product of the individual density functions.
Chalk and Corotis (5.14, 5.16) have suggested a load model combinin,!!
all possible load cases, each weighted by its respective likelihood of occur-
rence. The maximum total loftd during the lifetime of a building may arise
from one of the following situations:

Case I : Lt = Lm + Bt }
Case II : Lt = Bm + L
Case III: L, -,.., Lm + Bm (5.42)
Case l V : L, = L -1- B
 111

where Lrn = the maximum sustained load, Bm = the maximum extra-

ordinary load, Bt = largest extraordinary load occurring during the duration
of Lm, L = instantaneous sustained load, and B = instantaneous ex.tra·
ordinary load. The case IV is not considered as the probability of its
occurrence is small.
If E()..) is the average duration of the sustained load and Tis the lifetime
of the building, then the probability that Case I or Case II occurs is
[T - E()..)]/T, and the probability that Case IH occurs is E()..)/T. The
probability of the maximum total load can be written as

P[Lt < I] = P[(Lm + Bt) < I]· P[(Bm + L) < I] [T-TE()..)]

E()..)
+ P[(Lm + Bm) < /] --T
If it is assumed that Lrn, Brn, Bt, (Lm +
Bt) and (Lm + Bm) follow the
Type I extremal distribution, the CDF of Lt is

FLt(l) = exp [- cxp (- Wt)] exp [- exp (- w2)l [T-TE()..)]

E(>t)
., + [- exp (- WJ)] T (5.43)

where Wt, w2 and Wl are reduced variates corresponding to (Lm Bt), +

(Bm + L) and (Lm + Bm) respectively.
In conclusion, the analysis of live load is complicated. The probabilistic
analysis of live loads to predict the mean of lifetime maximum total load at
desired reliability level is based on the live load survey data collection,
data reduction , and the probability models of sustained, eJCtraordinary, and
total loads. The procedure of the analysis is summarized as (i) the estima·
lion of parameters m, a~, a~ and v from the survey results, (ii) establishing
the tati tics of sustained load and extraordinary load that are obtained
fr m the respective load models, and (iii) the estimation of mean and
variance of the ma)timum sustained load, the maximum extraordinary load ,
and the maximum total load by fitting Type 1 extremal (largest) distribu·
tion to the respective cumu lative distributions.
Live load survey has been carried out on three office buildings in
Bombay (5.9, 5.18). These buildings are modern office buildings occupied
for a sufficien t length of time for normal occupancy consolidation, and the
age of the buildings varies from 20 to 40 years. All the three buildings a(e
multistoreyed. The total area and tbe number of bays covered in the survey
are 1800 m2 and 386 re pectively. The bay areas in the buildings vary from
27 to 67 m1 . [t has been found that the floor load intensity varies from 0.1
to 4 kN/m 2 • The resu lts of the suitability of the mathematical model for
FLl are given in Table 5.2. The collected data for all the buildings has
been combined, and for the combined data, the mean and the coefficient of'
variation of FLI are 0.717 kN/m 2 and 0.52 respectively. Using the model
132

TABLE 5.2 Suitabi/iiY of mathenwtical model fur bay FU of o./fice buildings (5.18)

Distribution and Rem8rks of

Sl. No. Description
parameters chi-square lest

I. Adn1illistrativc Building,
I.I.T, Bombay
t<=- 0 5% k N 'm' LN(0.54, 0.42) Accepted"' ~ 5%
8 = 0.446 G(5.027. 8.435) Accepted "' ··" 5%
2 Hc~d Office Walchand
Ruildin~. H0mbay
!'- = 0 7:!8 kN ,' m' LN(O. 713 , 0.202) Accepted "' = S%
8 = 0.207 G(23.34, 32.06) Accepted rx ~ 5~,~
3 ( c'nlr,d R:til ~v:\\
!\dmitti'irali'c
Building. Homhay
tJ. = 0. 7 .15 1-N m" LN(0.620, 0.618) Acce pt ed ot =~ 50,~
8=-0,67
4 All Ru ;l dings comb i ned
!0gc•t ht' l
t< -0 .717 kN m' L~(0.(,3fi, 0 488) .'\cccpted "' = 5%
~ =- 0 52

proposed lw Pier :1nd Cornell (5. 12), and the method of analysis explained
in the text, and the <lppro::tch used by Ellingwood and Culver (5.15), the
colkcted tbta have been analysed and the following values for the mean and
c0etTicic-nt 0f variation of T.mn• have been suggested by Ranganathan (5 . 18)
for buildings. The value of v has been taken as 8. For lifetime maximum
total live load, Lnnx.
Model : Type 1 (extremal largest)
Mc<~n : 2.4R kN/m2
Coefficient of variation : 0.283

~- = 2•4 = 0.62 I
\)fllln::tl 4 .0
For arhitr<try roint-in-timc \ar~ing li\e load, Lap,,
Model : lognormal
Mean : 0. 717 kN/m 2
('octllcient of variation : 0.52
: 0.7417 = 0.179
' n min al

5.2 WlNil LOAD

5.2.1 lntrodmtion
The wind load. W, acting o n :1 str11cture can be written in the form
11·· = BV 2 (5.44)
 133

where B is a parameter covering all components of the wind luad (except

the basic wind speed), i.e. pressure coefficients, area reduction factors,
velocity multipliers for height and exposure, etc. V is the " ind speed,
generally referred to a height of 10 m. Wind loads are random in nature
due to random variations of wind speed and uncertainties in the estima·
tion of the pressure coefficients, the exposure factor, and the rust factor.
The modeJiing of wind load is much more complex and difficult than the
modelling of speed. Because the velocity appears in the equation as a
squared value, its statistics is very impo:'lant. However, the uncertainties in
the various factors contained in B contribute to the overall variability in
the wind load.

5.1.1 Wiacl Speed

The wind velocity is stochastic in nature. It has spatial and temporal vari·
ation during a storm. Wind speed, V(X, t), in a given direction in a point
of position vector X, at time t during a storm is generally considered as the
sum of two terms (5.19), viz.
V(X, t) = Vo(X) + V1(X, t)
in which Vo(X) is the steady componcnt equal to the average velocity dur-
ing the storm and v,(X, t) is a zero-mean process describing the gusts. The
above model is useful when the structure under investigation behaves
dynamically under wind excitations. However, many structural engineering
· problems arc concerned with structures in the static field. lf only the static
behaviour of the structure is involved, the velocity is expressed in the form
V = V*«(z)G (5.45)
where v• is the steady (average) velocity at a reference height (10m), at(z)
the multiplication factor for height, and G the gust factor. The maximum
value of V over an appropriate time interval Tis of interest in structural
reJaability analysis. For this purpose, the mean arrival rate (or the mean
occurrence interval T = 1/A) of V must be specified. Hence, it is necessary
to associate return periods T with the values of wind speed. This can be
done on the basis of cumulative distribution of yearly maximum wind
speed.
Wind velocities are measured in a horizontal plane with the aid of
anemometers or anemographs, which are installed at the meteorological
observatories at heights generally varying from 10 to 30m. The different
types of anemometers are (i) pressure anemometer (ii) rotation anemo-
meters, and (iii) gust measuring anemometers. The one which is usually
used in India is the cup anemometer which falls in the catego,.Y of rotation
anemometer. Very strong winds (greater than 80 ~ph) are generally
aaooiated with cyclonic storms, dust storms, or vigorous monsoons. A
cyclone is one in which the wind speed exceeds 80 Jcmpb. The wind velocity
recorded at any locality is extremely variable and in addition to steady
134

wind at any time, there are effects of gusts which may last for a few seconds.
Wind forces acting on structures are significantly large only during strong
winds and these occur only during storms. Hence only these extreme wind
forces are of interest to the structural engineer. Attempts are, therefore,
always made to collect data on extreme wind speeds and suggest a suitable
probabilistic model for the same.
The continuous recording of wind velocities is generally carried out in
meteorological stations. Out of these values, one is interested in the extreme
or the maximum. From the continuous recording, it is possible to obtain
daily, monthly, and yearly maximum wind speeds. Figure 5.3 shows the
14 0385
n = 36

..~
111
~ 10
12 u :97 6 kmph
a:= 0 066
O·JJO

0-275 ~
>
Ill Qj
ell ::l
.a 8

0
0

6
Type r e~trema I
(largest J
0 220

0 165
-g
...
Qj
:>

~ 4 0 110 ~
.a Ill
E a::
::>
z 2 0 055 J

0
80 95 110 12S 140 155 170'
W1nd speed (kmph)
FIG. 5.3 Frequency distribution of annual maximum wind speed at New Delhi
(Safdarjung)

vanatton f annual maximum wind speed observed at New Delhi ( afdar-

jung). Since the yearly maximum wind speed can be interpretated a the
largest of daily values or perhaps the largest of many gusts, velocities, the
statistical behaviour of the yearly max imum wind speed is represented by
two types of extremal distribution with unlimited upper tail. One is tbc
Type 1 extremal (largest) distribution, o cal led Gumbel di stribution and
the other one is the Typ 2 extremal (largest) distribution also ca lled
Frechet di ·tribulion. The choice of the underlying distribution can be made
after the analysis of fitting closeness to the Jata. ll was sugge red that
Type 2 di ' lributi n is an appropriate model to employ in mo t of N rth
American Region (5.20), although recent tudies (5.21, 5.22 have indicated
that Type I is more appropriate. In Japan (5.23) and Australia (5.24),
T pc I is r und l be m re suitable n the basis of statistical ana ly is. Ln
Germany, ciJUeJier and Panggabean (5.25) have fitted Type I and Type 2
di tribution to maximum yearly gu t and average velocities. The Type 1
distribution has been used to lcscribe the tatistical behav iour of the yearly
maxin 10m wind peed in India 5.26 5.27). The mean rank pi tting
(Fig. 5.4) of th e data on the year ly maximum wind speed observed at
Delhi shows a good straig ht line lit , en ouraging the use of Type I extre-
 135

mal (largest) distribution. The parameters of the selected distribution are to

be estimated using any-one of the methods (5.28). However, Simiu, Bietry,
Filliben and Grigoriu (5.22, 5.29, 5.30) have proposed an improved techni-
que for the analysis of wind speed data.

3·5t
J·O u : 97 ·2 kmph
oL : 0 06 Y: o((~-u) _
2·5
~ 20 -
0
...
0 1·5
>
~ 1·0
u
:I
'0 05
"
ct
0
-05 -

-1 0

-1 5
0 20 40 60 140 160
Annual wind speed ( kmph)
FIG. 5.4 Mean rank plot for Type 1 extremal (largest) distribution for Vann
observed at New Delhi (Safdarjung)

5.2.3 Return Period

A very common problem in wind analysis is to assume the return of an
observed extreme wind speed or cyclone. For design purposes, one often
attempts to estimate the magnitude of an extreme wind of a particular
return period. The return period, R, which is called the mean recurrence
interval, is defined as
I I
R=-
p
= I - F11(vs)
(5.46)

where u. is the specified design wind speed, F11(v) is the CDF of yearly
maximum wind speed, V, and p is the probability of wind speed V exceed-
ing Vs in any year.
The return period is to be understood as the interval between events.
Hence a 10-year return period wind (i.e. p = 0.1) is the wind which could
be expected to occur in the long term, about once in every I 0 years. It does
not mean that there will be a period of 10 years between winds of a parti-
cular size. The occurrence of wind in time is a random process and so it is
quite possible that I in to-year wind could be exceeded more than once in
one year, or in successive years, or that there may be a period of more than
30 years in which no heavy wind as large as in the 1 in 10 year wind
 136

occurs. The one I in 10 year event is the event that could be expected to be
equalled or exceeded about I 0 times in a 100-year period. Hence the I in
10-year wind has a frequency of 0.1, that is, there is a 10 percent chance
th at it will be eq ualled r exceeded in any year.
XAMPLE 6.4 The yexrl max imum wind peed follows the Type I distri-
. _ but ion ~i th pa ra meters
u = 97.6 kmph IX = 0.066
Determine the return period of the design wind speed 158.1 kmph.
Solution It is given that V follows the Type I distribution. Hence
Fv(v) = exp [ -- cxp {-·- a.(u -- u)}]
Then
Fv(us) = Fv(I 58.1) = exp [- exp {-- 0.066( 15R.I - 97.6)} I
= 0.9817
Using Eq. (5.46), the return period of the design wind speed is
I
R =" _ _ Rl ? = 54.7 years
1 09
In case if one wants to find out the 20-year return period Wind speed,
then
I
p = 20 0.05

I
I - Fv(z·) p 0.05
Fv(v) - ·• 0.95
Then the wind speed corresponding to this probability is given by
exp I --exp [- O.Ofi6(t•- 97.6)}1 = 0.95
v = 142.6 kmph

Hence the 20-year return period wind speed is 142.6 kmph.

1n the current design procedures, wind loads arc treated semi-proba hi 1~>­
tically. The annual maximum wind speeds arc reco rded and an appi\lj1rialc
probability distribution is fitted lo the data. A wind with some speciliecl
probability of exccedencc in any one year is then selected for design pur-
poses. Usually, a 0.02 cxceedcnce probability for 50-year return period is
used. Although 50-year return period has attained a somewhat mystical
status in civil engineering, ils use doc> not hold up well under closer
examination. ln fact, the 0.02 excecdcncc level for a Type I extreme value
distribution, normally used for wind spcl:ds. corre~ponds to an exceeclence
level of 0.63 in a lifetime of 50 years (5.31 ..
 137

5.2.4 Estimation of Lifetime Design Wind Speed

Occasionally, it is necessary to design a structure against wind load for a
fixed period from the period of construction. For example, if a structure is
built which is only to be used for 3 years following construction, or is to
be removed at the end of 3 years. the risk ol" damage e.-.;ists only for this
period. Thus v:hat is required for design is wind speed associated with a
probability of being exceeded in the fixed period starting with the building
of the structure. This design wind ~reed is designated as lifetime design
wind speed.
Jr l',t is the lifetime design wind ~peed, I - h(vc~) is the probability of
the annual extreme wind speed exceeding the design value I'"· Hence, the
probability of no extreme wind exceeding v,1 in the first m years is (Fv(r,J)]"'.
(This derivation is similar to the one that is derived fnr the lifetime deoign
live load). The probability of atleast one extreme speed exceeding 11rl is
f?m ·= I - 1Fv(rc~)] 111

( 5.47)
Here Pm and m are chosen by the designer. For example, if m = 50 years
and the designer has chosen a chance of the design wind speed being
exceeded to be pso = 0.05 or one in twenty, then the value F computed by
Eq . (5.47) becomes equal to 0.9989746. (It is to be noted that this
corresponds to a return period of 975 years). The characteristic wind speed
for the ultimate limit state is defined (5.31) as the wind gust speed with an
estimated probability of exceedence of five per cent in a lifetime period ot
fifty years of the structure. Based on this definition, substituting m 50
and pso "-= 0.05 in Eq. (5.47), the computed design speed , ."become~ the
characteristic wind speed for the ultimate I imit tal e.
AMfLF.f.i' For the same data given in Example 5.4, calculate the lifetime
de igr~ speed for rn = 50 years and p, = ().05 .
Solution Using Eq. (5.47),
Fv(vtl) = [I - Pm]l/m

= [I - 0.05] 1150 = 0.9989746

Since V follows the type I extremal distribution,
Fv(vtl) = exp[-exp{-ot(v- u)}J
That is,
exp[ -:-exp{ -0.066(vrt - 97.6)}] = 0.9989746
Vtf = 201.88 kmph
This is the characteristic wind speed to be used for the design under ulti-
mate limit state.
Similarly, the design wind speed can be calculated for different values of
m and pm. The variation of lifetime design wind speed with the service
138

period of structures for different probabilities Pm, is shown in Fig. 5.5 for
New Delhi (Safdarjung) station. As expected, for a given value of pm, the
design ~peed increases with the lifetime of the structure, and for a given
lil'ctim c of the structure, it increases with decrease in the values of risk
( i. C. {lm) .

II
220r-
l Pm=O·OS
I
~ 2001
i Pm=0·10
-t, I
E ! Pm 0·20
_,. i
'i 180
~
Ill
"0
.~ 150
~
c
01
en
~ 140

120
o~--~2o=----40
-!=------=s:":
o---=s'::-o----:,~o-=
o--
Dt>Sign life (years)
FIG. 5.5 Variation of design wind speed with fixad life period of the
structure

5.2.5 Probability Model for Wind Load

Recalling Eq . (5.44), the wind load on structure can be written in the form
W= BV 2
where B is a parameter covering all components of the wind load except
the basic wind speed . The parameter may be assumed to be made up of the
product of the number of variables as follows
(5.48)
where K is the analysis factor, C is the pressure coefficient depending on
the geometry of the structure, E is the exposure coefficient depending on the
location (e.g. urban area or open country), G is a gust factor depending on
the turbulence of wind and the dynamic interaction between the structure
and wind, and D is a directionality factor to take into account the effects
of the wind direction. Hence the wind load may be written as
W ,;, K C E G J>V 1 (5.49)
 139

If one wants to find out the probability model for W, the Monte Carlo
simulation (dealt in Chapter 7) technique can be used, which requires the
probability distribution and the parameters of individual variates. To deter-
mine the lifetime maximum wind load model, the probability distribution
and parameters of lifetime maximum wind speed must be known. If V
follows the Type I distribution, the lifetime design speed for m years, Vm,
also follows the Type t distribution. The mean and coefficient of variation
of Vm are given by (5.32),

r"' = r( I + -.v-,.,6 Sv ln(m) ) (5.50)

Ji'
8v, = 8v o-
P'm
(5.5 I)

Ji' nr and 8vm are the mean and coefficient of variation of Vm.
The approximate mean and coefficient of variation of W can be found
out by the following expression assuming all variables in Eq. (5.49) as
independent:
W = K C E a D Ji'1 (5.52)
(8w) 2 = (Sx)l + (8c)2 + (8s)2 + (Sa)2 + (So)2 + (2Sy) 2 (5.53)
Since W is the product of the number of random variables, the probabilis-
tic model for Wmay tend towards the lognormal distribution. However,
Ellingwood (5.32) has proposed Type l extremal (largest) distribution
(based on Monte Carlo technique) for W for the assumed mean and co-
efficient of variation of the different variables in Eq. (5.49).
The author bas collected data on the annual muimum wind speed
observed at 48 stations, and the daily maximum wind speed observed at 4
stations in India, and has statistically analysed the collected data. The
Type 1 extremal (largest) distribution, in general, is found to fit the data on
annual and daily maximum wind speed. Using the results of the analysis of
wind speed, the analysis of wind load has been carried out taking into
account the uncertainties in various parameters affecting the wind load,
and statistics of wind loads have been fixed for a probabllistic criterion.
The analysis of wind load is carried out for the maximum wind load, Wm•••
corresponding to the lifetime maximum wind speed, annual maximum wind

TABLE5.3 Statistics of wind load

.,.'
Variable Mean a u Ill Probability
distribution

WmuiWa 0.804 0.334 0.683 4.1S Type 1 extremal

(largest)
w.twa 0.349 0.392 0.287 9.31 "
w.p,IWa 0.04S2 0,743 0.030 3B.IS
140

load, Wa, corresponding to the annual maximum wind speed, and the daily
maximum v.ind load (which is considered as an arbitrary point-in-Lime
varying wind load), Wapi, corresponding to the daily maximum wind load.
The final statistics of wind load established for Indian conditions are given
in Table 5.3 (5.27).

REFERENCES
5.1 Dunham, J.W. "Design Ltvc Loads in Buildings", Tra/1.\'0ctiuns, ASC£, Vol.
112. 1947. pp. 725-744.
5.2 Btyson. J 0 . anti D . Gross "Techniques for the Survey and Evaluation of Live
Floor Lo;tds and Fire Loads in Modern Office Buildings", NBS Buildi11g Science
Se1ics, 16, U.S. Dept. of Commerce, Washington, D.C., 1968.
5.3 Mitchell, GR. and R.W. Woodgate, "A Survey of Floor Loadings in Office
Buildings". CIRIA Report 25. Building Research Station, London. England,
Aug. 1970.
5.4 Culver, C.G., "Live Load Survey Results for Office Buildings", Journal of the
Stmctural Division, ASCE, Vol. 102, No. STII2, Proc. Paper 12615, Dec. 1976, pp.
2269-2284.
5.5 Senller, Lars., "A Stochastic Model for Live Loads on Floors in Buildings",
Report60, Division of Building Tcrhnology, Lund Institute of Technology, 1975.
5.6 Ranganathan, R. and P. Dayaratnam, "Statistical Analysis of Floor Loads and
Reliability Analysis", The Bridge and Stmctural Engineer, Vol. 7, No. I, March
1977.
5.7 Rao, P.S. and G.S. Krishnamurthy, 'Imposed Live Loads-Their Evaluation',
Dept. of Civil Engineering, Indian Institute of Technology, Madras, IndiJ, 1982.
5.8 Prabhu, U.P. and R. Ranganathan, "Stochastic Analysis of Live Loads in Office
Buildings", Proc. of the National Conference on Quality and Reliability, held at
I.I.T., Bombay, Dec. 1984, pp. 275-291.
5.9 Ranganathan, R., "Statistical Analysis of Live Loads in Office Buildings", D.S.
and T. Report No. 4f/f83-STP-1l!(3), Civil Engineering Dept., I.I.T., Bombay,
Oct. 1985.
5.,10 Ellingwood, B., T.V. Galambos, J.G. McGregor and C.A. Cornell, 'Develop-
ment of a Probability Based Load Criterion for American National Standards
ASS', NBS 577, U.S. Dept. of Commerce, Washington, D.C., June 1980.
5.11 Christensen, P.T. and M.J. Baker Structural Reliability Theory and its Applications,
Springer-Verlag, Berlin, New York, 1982.
5.12 Peir, J.C. and C.A. Cornell, "Spatial and Temporal Variability of Live Loads",
Joumal of the Structural Division, ASCE, Vol. 99, No. ST5, Proc. Paper 9749, May
1973, pp. 903-922.
5.13 McGuire, R.K. and C.A. Cornell, "Live Load Effects in Office Buildings",
Journal of the Structural Division, ASCE, Vol. 100, No. ST7, Proc. Paper 10660,
July 1974, pp. 1351-66.
5.14 Corotis, R.B. et a/., "Area Dependent Processes for Structural Live Loads",
Journal of the Structural Division, ASCE, Vol. 107, ST5, Proc. Paper 16266, May
1981' pp. 857-872.
5.15 Ellingwood, B.R. and C. G. Culver, "Analysis of Live Loads in Office Buildings",
Journal of the Structural Division, ASCE, Vol. 103, No. ST8, Proc. Paper 13109,
Aug. 1977, pp. 1551-1560.
5.16 Chalk, P.L. and R.B. Corotis, "Probability Models for Design Live Loads",
Journal of the Structural Division, ASCE, Vol. 106, No: STlO, Oct. 1980, pp.
2017-2033.
 , 41

5. 17 H~sofcr, A.M .• "Statistical Model for Live Floor Loads", Jour11al at' lht! Stmc·
Jural Division, ASCE, Vol. 94, No. ST'IO, Proc. Paper 6/46, Oct. 196M, pp. 2183-
2196.
5.1 R Ranganathan, R., •· Reliability Antilysis und Design of RCC Slabs, B~ams and
Columns and frames-Code Calibration", D.S. a11tl T. Report No. 4il/8.1-STP Ill
(5), Civil Engg. Dept., J.I.T., Rombay, Sept. 1987.
5. 19 1\ugusti . G ., A. B11ralta, and F . Casciuti, Prnbabili.rtic Meihod~ in Stmcltnal
DIJ:ineering, Chapman and Hall, New York, 1984.
5.20 lhom, H.C.S., ''New Distributions of Extreme Winds in the U.S.". Jaurnal of
StmNural Divi.rio11, Proc. ASCE, Vol. 94, ST7, July 1968, pp. 1787-JSill.
5.21 Simiu, E. and J.J. Fillihen. "Probability Distributions of Extreme Wind Speeds",
Journal of Structural Division, Proc. ASCE, Vol. 102, ST9, Sept. 1976, pp. 181li-IR78.
5.22 Simiu. E.. J. IJietry and J.J . Filliben, ''Sampling Error in Estimation of Extremr.
Winds'', Journal of Structural Divisio11, Proc. ASC£, Vol. 104, ST3, March 1978,
pp , 491-502.
5.23 Ito, M. and Y. Fujino, "Design Wind Speed and Wind Load Fa.:tor Based on
Prt•babilistic Ration:1l~:", Proceedings of the Fiftlr 111/ernationa/ Conference on Wind
l:ngineerinJ:, Fort Collins, Colorado, U.S.A., July 1979, pp. 1271-121!0.
5.24 Dorman, C.M .L., "Extreme Wind Gust Speeds in Australia , exCluding Tropical
Cy,lones". Cil•il Engineering Transactio/Is, Institution of Engineers, A.tstralia,
Vol. Cl:J R, No.2, 1983, pp. 96-106.
5.25 Schuelkr. and Panggabean, "Probabilistic Determination of Design Wind
Velocity in Germany", Proc. of brstitution of Civil Engineers, London (U .K.),
Pa1t 2, Vol. 61. Dec, 1976, pp. 673-683.
5.26 Rang;wathun, R ., ' 'Stalislical Analysis of Wind Speed and Wind Load for
Probubilistic Criterion", D.S. and T. Report No. 5/1/83-STP-111(4), Civil Engg.
Dept. I.I.T .. Bombay, March 1986.
5.27 Ranganathan, R. , "Wind Speed and Wind Load Statistics for Probabilistic
Design", Journal of tire Institution of Engineers (India), Civil Engg. Div., Vol, 68,
May 19!l8, pp. 303·30S.
5.28 Ang , A .H .S. and W.H. Tang, Probability Concepts in Enginet?ring Plmminf( and
Design, Vol. I, John Wiley and Sons, Inc., New York, 1975.
5.29 Grigoriu, M., "Estimates of Design Wind from Short Records", Journal of Stru·
ctural Division, Proc . ASC£, Vol. 108, STS, May 1982, pp. 1034-1048.
5.30 Grigoriu, M .. "Estimates of Extreme Winds from Short Records", Journal of
Strttclttral Division, Proc. ASCE, Vol. 110, ST7, July 1984, pp . 1467-1484. ·
5.31 Holmes, J .D. ' ' Wind Loads and Limit States Design", Tire Civil Engineering
Transactions, Tire Institution of Engineers, Australia, Vol. CE 27, No. I, F~b. 1985,
pp . 21-25.
5.32 Ellingwood, B., "Wind and Snow Load Statistics for Probabilistic Design",
Journal of Struclrlral Dil'ision, Proceedings ASC£. Vol. 107, ST7, July 1981, pp.
1345-1350.

EXERCISE
5.1 The live load on a building follows the lognormal distribution with mean
~~ 1.3 kN/ma and 3 ~~ 0 .381. If the specified design load is 2 .5 kN /m 1 , what is the
probability of exceeding the specified design load? IAns. 0.0256)
What is the value of I ive load with a probability of cxceedence of five per cent?
(Ans. 2.22 kN/rn')
5.2 The live load on a building follows the lognormal distribution with mean
= 1.3 kN/m' and 8 = 0.381. The lifetime of the building is 50 years and the
period of tenancy is 5 years. What is the lifetime maximum design live load for
the buildin11 with a probability of exceedence of five per cent during the lifetime?
(Aru. 3.17 kN/m')
142

5.3 The annual maximum wind speed observed at a station follows the type 2 extrcm;il
(largest) distribution with parameters 11 = 81.00 kmph and k ~c 7.05. What is the
Jclurn period of the design wind speed = 182.5 kmph? (Ans. 309.6 yrl
At the same station a temporary structure is to be designed to serve for a
period of 3 years only. If the engineer takes a risk of five per cent, what value of
design speed will he choose for the design of the structure?
(Ans. 144.2 kmph)
5.4 Tf the annual maximum wind speed at Bombay follows the Type I extremal
(largest) distribution with parameters 11 = 81.4 kmph and 11 = 0.126, determine
the characteristic wind speed for the ultimate limit state.
(Ans. 136 kmph)
What are the mean value and coefficient of variation of the 50-year lifetime
maximum wind speed? (Ans. I 16.9 kmph, 0.086)
5.5 The model for wind load is given by Eq. (5.50):
W = K C EG D Jl9
If the variations in K and D arc neglected, and if llc = 0.12, liE = 0.16, 80 = 0.11
and llv = 0.114, determine llw. (AilS. 0.322)
 6
Basic Structural Reliability

6.1 INTRODUCTION
The performance of a structure is assessed by its safety, serviceability, and
economy. The information about input variables is never certain, precise,
and complete . The sources of uncertainties may be (i) inherent randomness,
i.e. physical uncertainty, (ii) limited information, i.e. statistical uncertainty,
(iii) imperfect knowledge, i.e., model uncertainty, and (iv) gross errors. In
the presence of uncertainties, the absolute safety of a structure is impossible
due to (i) the unpredictability of (a) loads on a structure during its life,
(b) in-place material strengths, and (c) human errors, (ii) structural idealiza-
tions in forming the mathematical model of the structure to predict its
response or behaviour, and (iii) the limitations in numerical methods.
Therefore, some risk of unacceptable performance must be tolerated. With
respect to risk of life, the structural safety is important. In the conventional
deterministic analysis and design methods, it is assumed that all parameters
(loads, strengths of materials, etc.) are not subjected to probabilistic vari-
ations. The safety factors provided in the existing codes and standards,
primarily based on practice, judgement, and experience, may not be
adequate and economical.
T he concept of reli abilit ha been applied to many fields and has been
in terpreted in many ways he most common definition, and accepted by
all , of reliability is that re 1a ility is the probability of an item perfor ming
its intended function over a given period of time under the operating condi-
tions encounter . It is important to note that the above definition stresses
· -'
four · significant -elements, viz. (i) probability, (ii) intended function,
(iii) time, and (iv) operating conditions. Because of the uncertainties, the
reliability is a probability which is the first element in the definition. The
second point, intended function, signifies that the reliability is a performance
characteristic. For a structure to be reliable, it must perform a certain
function or functions satisfactorily for which it has been designed, i.e.
safety against shear or flexure or torsion, etc. The reliability is always related
to time. In the case of structure, it is related to the lifetime of the structure.
During this specified life of the structure, it must perform the assigned
function satisfactorily. The last point is the operating conditions. This
establishes the actions or stresses that will be imposed on the structure.
These may be loads, temperature, shock, vibrations, corrosive atmosphere,
etc. Reliability also changes with respect to quality control, workmanship,
production procedure, inspection, etc.
144

As stated in Chapter 1, in structural analysis and design, reliability is

defined as the probability that a structure will not attain each specified limit
(flexure or shear or torsion or deflection criteria) during a specified reference
period (life of the structure). For convenience, the reliability, Ro, is defineu
in terms of tl1e probability of failure, PI· which is taken as
Ro ~-co. 1 - p1 (6.1)

In the case of the classical reliability theory, for reliability prediction

informations on life characteristics of the system, operating conditions and
the failure distribution are needed. Life characteristics are measured by the
failure rate or the mean time between the failures or the mean time to
failure. Assuming the failure rate is constant over time, the failure rate A is
defined as

(6.2)

where f is the number of failures during a specified test interval and Tis
the total test time. That is, A is a ratio of the number of failures during a
specified test interval to the total test time of the components or items. Tb~
smaller the value of A, the higher is the reliability.
If the failure rate is col1stant during the operating period, the mean time
between the failures is the reciprocal of the constant failure rate.
If there are n components with failure times It, 12, ... , tn, then the mean
time to failure is defined as
J n
MF = - .E 11 (o.Jl
n {c~l

Let a set of N items (structures) be repeatedly testerl. After a time 1 (this

may be considered as the time eiapscd since the structure is put intl) service.
i.e. the age of the structure), let n components f<til VI structures in a faibi
condition). Then the probability of failure at time 1 can be express~d as

F(t) = .!!_ (6.4)

N
This F(t) is called tht failure function or the lifetime failure distributim'
function for the set . and the reliability function or survival function, R(t1
is given by
R(t) = 1 - F(t) (6.5)
The failure rate function is given by the derivative of the failure function.
That is,

f(t) = d~~t) (6.6)

The hazard rate or hazard function is the instantaneous failure rate as the
interval length tends to zero. It is defined as the probability of failure per
 146

unit of time given that the failures have not occurred prior to time t. That
is,

(6.7)

0
,.
u

~
t
0"
•...
I&.

t •
FIG. 8.1 Various reliability functions

Hence, if the informations on failure rate, time between failures or break-

downs, reliability function, and hazard function are available, based on
actual data, many predictions can be made about the system performance
and decisions can be taken based on that.
For structural systems it is difficult to predict the expected life or the
expected failure rate or the expected time between breakdowns. In the
reliability format, it is assumed that structural failures are not due to dete-
rioration . The structures cannot be assumed to be nominally identical. The
structural failures cannot be expressed in terms of the relative frequency.
Thus, the structural reliability theory differs from the classical reliability
theory in many such aspects except in the probabilistic nature because of
the uncertainties. The probability of failure of a structure is a subjective
probability. The reliability of a structure is not a unique property. It chang-
es as the state of knowledge about the structure changes. ·
The acceptable probabilities for structural failures are very low, e.g. (i) of
the order of IQ-3 for serviceability limit states, meaning thereby that on an
average, out of 1000 nominally identical structures, one may deform exces-
sively or (ii) of the order of 10-6 for ultimate limit states, which means that
out of one million identical structures, one may collapse. In practice,
structures are never identical in a large number. Moreover, these low p~­
babilities are to be estimated from the statistical properties extrapolateM
from the available statistical data around the central values of the random
variables. Therefore, it will be proper to consider these probabilities as
146

conventional, comparative values without having much statistical signi-

ficance. In the light of this, probabilistic methods play an important role in
making rational comparisons between alternative structural designs. The
currently developed reliability analysis of structures aims at evaluating
the probability of failure (or reliability) of a structure.

6.2 COMPUTATION OF STRUCTURAL RELIABILITY

Consider a simple structure with one element only. Lel R be lhe resistance
(capacity or strength) of the structure and S the action (load or load effect,
viz. bending moment, shear force, etc.) on the structure. The 'structure is
said to fail when the resistance of the structure is less than the action. That
is,
PI= P(R < S)
= P(R- S < 0) (6.8)
Or P.r = P(R/S < I) (6.9)
where PI is the probability of failure of the structure. If /R(r) is the proba-
bility density function (PDF) of Rand if Sis assumed as deterministic, the
hatched portion shown in Fig. 6.2 gives the probability of failure. This is
ex pressed as

{Jf = roo _(R(r) dr --- w~r~w (6.10)

fR(r)

FIG. 6.2 Determination of probability of failure for

deterministic action

Fundamental Case
In real situations, both R and S are random variables. The plots of the
density functions of R and S are shown in Fig. 6. 3. The hatched portion
shown in Fig. 6.3 is an indicative measure of the probability of failure. The
probability of failure is computed as follows (rj,l):
The probability of S assuming a value s, is equal to the area A 1 marked
in Fig. 6.4.
 147

s,r

FIG. 8.3 Probability of failure for random variations of S

and R

fs{s),
fR(r) U~r tail ofS
I

FIG. 8.4 Determination of reliability

P(s - ~ < S <s + d;) = fs(s) ds = At

The probability that R >s is equal to the shaded area Az under the
resistance density curve:

P(R > s) = J~ /R(r) dr = Az

l dRo = /s(s) ds r f.{r) dr

the reliability of the structure, Ro, is the probability of R being greater than
all the possible values of S:

l = f dRo = J:, /s(s)U: /R(r )dr] ds 1 (6.11)

-------
Rp

~
 148

Hence the probability of failure is

/ Pr = I - Ro = I - J:. ,fs(s)[l - FR(s)] ds

= roo .fs(s)FR(S) ds (6.12)

The reliability can also be found by considering whether the structure sur-
vives when the action remains less than the given value of the resistance.
Following the same procedure given above, Ro and PI can be expressed by
the following equations also

Roo If,(.-Hf,(s)d} (6,13)

PJ = I -- J:,., .fR(r)Fs(r) dr (6. 14)

It must be noted that the integrals in Eqs. (6.12) and (6.13) are to be
evaluated numerically. Except for a few cases, the closed form solutions are
not available.
The closed form solutions for the evaluation of PI, when both R and S
are normal and both R and S are lognormal, are given below:
. .£ase 1: Both R and S are normal
'---' The proba bility of failure of a structure is given by Eq. (6.8):
PI = P[(R - S) < 0)
Let
M=R-S (6.15)
where M is defined as the margin of safety. When R and S are independent
and normally distributed, M is also normally distributed. The mean value
of M, /LM, and the standard deviation of M, aM, are given by
/LM '= P.R - /LS and aM = (a~ -1- a~)l/2
Hence the probability of failure is given by
PI= P(M < 0)

. [ !LS - /.I.R ) .,
flf = (/) - /,. (6.16)
I. 2 - 0'~· )1 12 /
\CIU ".
If R and S arc correlated with correlation coefficient, r and if the joint
distribution of Rand Sis normally distributed, the value of PJ is given by

(6.17).
 149

Let

{J = J.'M (6.18)
CIM

Then the value of Pi corresponding to {3 is given by

PI = til(- {J)
and. the value of {3 corresponding to a given PI is
{J = -~-l(pl) (6.19)
Hence {3 is related to the probability of failure and is called the 're · lbllity
in ex'. T te va ue o JS a ected by the mean va ues a n standard devia-
tions of R and S, and also by the level at which the distributions of R and S
intersect with each other.
~ase 2: Both R and S are lognonnal
The probability of failure of a structure is given by Eq. (6.9):

Let

lz = ~ (6.20)

When R and S are indepen ent and lognormally distributed, it is known

,...
that Z is also lognormally distributed with parameters Z and a1n z, where Z
is the median of Z and 17Jn z is the SD of In Z . Thus
PI= P(Z <I)

Pf =en[ In~ I z)] ~ (6.21)

O"~a z

When RandS are distributed as LN(R, - and LN(S, - s) respectively,

-
the parameters Z and OJnz
a1n R)

of the lognormally distributed Z are given by

a1o

- R (6.22)
Z=;:;:;-
S
and (6.23)

Substituting the above equations in Eq. (6.21), we get

PI= ~[ 2
ln (S/R)2 -
] ./' (6.24a)
(aln R + Clln s)l /l /

- rr>[ In (S/R)
-- ] (6.24b)
- [In {(8~ 1)(81+ + l)}]112
150

But if X is lognormally distributed, then

- I 2
In (X) = In /LX - 2a1n X

) [ ttx ]
= n o~ + J) l/2

Using similar equations for In Rand In S and substituting them in

Eq. (6.24b), we have

{~~J ~; -7 J '/
1
PI = <P[ In 1}
(6.25)

{In [(13~ + 1)(13~ --1-- 1)]}"2

lt is to be noted in the above equation that PI has been written in terms of
the mean values and the coefficients of variation of R and S only. When oR
and Bs are less than about 0.3, Eq. (6.25) becomes

(6.26)

If R and S follow exponential distributions with parameters An and As

respectively, it can be easily proved that (6.2)

AR
Pr = (An +As)

For other combinations of distributions of RandS, Eq. (6.12) or (6.14) is

to be used to compute the probability of failure. The closed form solutions
are generally not available.
ExAMPLE 6. 1 Derive an expression for the probability of failure when S
(say action due to wind) follows the Type 2 extremal (largest) distribution
and R (say strength of steel) follows the lognormal distribution. Given

Fn(r) = IJ'>fln (r/R)] ,. ~ 0 (6 .27)

l U! n n

and /s(s) = -k ( -u )k+l exp [ -(u/s)k] s~O (6.28)

.!/ s
Solution As random variables can assume only positive values, Eq: (6.12)
for the probabii:ty o f failure becomes

Pr ·. c J'"' fs(s)Fn(s)
()
ds (6.29)

Equation (6.2R) is rcwril!cn as

f s(s) -k·(
u
)-(k I S
ll
I)
exp [ --(s/u)-k]
 161

Substituting the above equation and Eq. (6.27) in Eq. (6.29), and putting
s/u= v, we get

PI= kJ<rJ ~ [ln(uv/R)) (v)-<k+O exp [ -v-k] dv

o <Tro R

Let

R
-={3 v-k = t
and
u
Then

PI = Jo rp[ln (u/{3)]e-t( -dt)

oo <Tin R

=
"' [- ! In
J0 ~
t - In fJ] e- 1 dt (6.30)
!7Jn R

This can be evaluated using the Laguerre-Gauss quadrature formula.

Similarly, for other combinations of probability distributions of R and S,
expressions (integral form) for the reliability or probability of failure can
be develo
0 xAMP .2 The ax:ialload carrying capacity of a column, R, i normally
distri d with f.tR = 1000 kN and aR = 200 kN. The column is subjected
to a axial load , S, which is normally di tributed with fJ.S = 700 kN and
as = 300 kN. Calculate the reliability of the column assuming' R and S are
independent.
Solution The margin of safety is given by
M= R -- S
Since R and S are normally distributed, M is also normally distributed.
Using Eq. (6.16),

PI = d>[(~s ; ~~; 112]

700 - 1000 ]
= <P [ (300 2 + 200 2) 112

~-~ tl>( -0.832) = 0.2027

Ro = 1 - 0.2027
/ = 0.7973
~XA~PLE 6.3 A prestressed concrete pole is subjected to wind load, which
is as shown in Fig. 6.5, lognormally distributed as LN (1000 N/m 2 , 0.2).
Determine the mean depth of the pole at the limit state of deflection for a
reliability of 0. 999. It is given that
(i) allowable deflection: span/325
152

Wind
load(w)
l f-d---.j
200mm I J
f
St>cl ion on x x

FIG. 6.5 Concrete pole-Example 6.3

(ii) Young's Modulus of concrete (E):

LN (2.6 x 104 N/mm 2, 0 .2)
(iii) breadth of pole: 200 mm
(iv) variations in depth (d) and breadth of the pole are negligible.
Solution The maximum deflection, i.e. at the top ofthe pole, is given by
(w:< 200)(6000)4
Ymax == - SEJ - -
where w is the wind pressure in N/mm 2 and I is the moment of inertia in
mm 4 .
Since,

H'
,
}·mox -- -~
£d ' · 1• 94 ,,,, I 0 15
3A

At the limit state of deflection, the failure will occur when the allowable
deflection, Yan, is less than }'max, i.e.
}'all <}'max

or the probability of failure of the pole is given by

PI= P[(;~~~ )< l]

Let
z '-- _Y•II = 6000 ( £d
3
)
- Ymox 325 >~ 1.94:: 10 15 W
As E and w are lognormally distributed and dis deterministic, Z is also
lognormally distributed. The parameters of Z are given by

:-
z ="' 325
Gooo
< 1.94·>, ]QIS- -:-
(£d3)
W

= 2.4S .·, 1o- 7 d 3

 163

2 2
UJoZ = [alnE + UJnw] 1/ 2

= (0.2 2 + 0.2 2)1' 2 = 0.28

Since Z is lognormally distributed, we have

PI= tl> [In (1/Z) ]

GJnZ

= I - 0.999 = I0- 3
In ( 1/2.48 X IQ-7 dl) = a 10zcJ>-l(lQ-3)

= (0.28)(-3.1)
~lving the above we get d = 212.56 mm
...hxAMPLE 6.4 A reinforced conctete beam of an effective span, 8 m, is
subjected to live load. The cross sectjon has been designed with M 25 con-
crete and steel grade Fe 250. The area of teel (Aat) is 1400 mm 2 and the
self-weight of the beam 3 kN/m. It is given that the random variables, the
cube strength of concrete (feu) and the yield strength of steel (jy) are
normally distributed.
Breadth of the beam (b) = 240 mm
Effective depth of the beam (d) = 480 mm
Mean value offcu = 30.28 N/mm2
Mean value ofh = 320 N/mm2
SO Of feu = CTc = 4.54 N/mm2
SO of/y = CTs = 32.0 N/mm2
Calculate the probability of failure of the beam if the live load (L) is
normally distributed with mean·, 6 kN/m and standard deviation, 3 kN/m.
Solution The action, here, is the bending moment at mid-span due to dead
load (D) and live load on the beam. Assuming the dead load and span
length as deterministic, p.s and crs are calculated as follows:
The mean value of S is

P.s = 3X8
2
+ 2
P,L(8 )
8 8
= 24 + 6X 8 = 72 kN m

crs = CT£ (~)

= 3x8 = 24 kN m
The resistance, here, is the ultimate resisting moment of the beam. This is
given by (as per lndi~n Standard Code),

R = f,Att d [1- 0 ·~7/[c~']

154

Tn the <lbClve equation , o nl y /~u and/~ are considered as random variubles.

The appr xinHliC , 11aluc~ ul' mean und standard devia tion f R are
calculated using Eqs. l3 .!)1) and (3.83) . lt is a. sumed lhlll fy and }~ u are
independent.
0. 77 .• 1400 320 ]
i'u = 320 1400 4~0 [ 1 - 240 ,, 480 ~· 30.::! i{

193.774kN m
Using Eq. (3.~41.

, (O·RI )l , ( 1)2 ,
oI~
a'R '--= fl./.~/~ a;+ ~{r," ~~ a-c (6.31)

~R I ~~ Ast d [I _. l .54.Ast/lfy]
iJjy /l hd /lfcu

[t- 2404 480

= 1400 ·· 480
.
l.5 X
A
1400
>.: 3 2 ~]
:30.28
== 0.54 10 6

.~ R I = (CL77 A~t ~~y)

/cu ~ ~~f~ u

= (Q: 77~ l40~ 320

240 ,; 30.28
2

2
2
}

= 0.70 :< 106

Using th~ abuve values in Eq. (6.31)
a1 = (0.54 : 106) 2(32)l + (0.7 ~~ 106) 2(4.54)"
aR = 17.56 kN m
Since live load is normally distributed, S is also normally Jistributcd in this
case. Assuming R is normally distributed, the value of pr is obtained using
Eq. (6.16), i.e.

REFERENCES
6.1 Haugen, E. B. l'robabilistit! Approach to Design, John Wik) , N.:w York, 19(,8 .
6 .2 Kapur, K. C. and L. R. Lambt:rson, Reliability in E11 •in("tillg De.l'i[/11, John
Wi!ey, New York, 1977.

EXERCISE
6.1
'
If the probability d<!nsily functions of resistance Rand actionS are
!R - ~R exp ( -~Rr)

and
 161

derive an expression for the reliability, R0 , and prove that it is given by

As
Ro = (AB + ~s>
6.2 If Rand S follow gamma distributions, given by
)." n- 1e - J..r
[R(I') = /' 11, .\, Y;;<: 0
r(n )
Dnl m- t - 1'•
fs(s) " -;:;",.,---,,e _
= --'- Ill, (J' s ;;,: 0
F(m)
derive an expression for R 0 •
rem +
II) fl/(l +IJ
R0 = r(m)F(n} 0 (I - u)m-Jun-1 du

=
r(m +
n)
F(m)F(n) B 1(m, n) where t = PIA
)

6.3 If the resistance of a structure follows the lognormal distribution, and the action
on the structure due to wind follows the Type 1 extremal (largest) distribution,
• derive an expression for R 0 and show how will you solve it numerically.
6.4 It is assumed that the strength of a RCC column is given by the sum of the
strengths of concrete, C, and reinforcing bars B1• C and B1 follow normal distri· ~
butions with parameters given by ,- 0 '1-"1
1
I /~'c= 25N/rrtm• ~= 5N/mm• .J.tl\=-r{)t.-;Ut.- 1-" ....:J.. . ,~
Jk. \... I'Bt = 460 N/mm 1
a81 = 46 N/mm 2
O'lt"';f r._. '~--+-,;~>..!..-
If the size of the column is 250x 400 mm and if it is provided with four 20 mm
diameter bars, determine the mean value and standard deviation of the strength
of the column. The column is subjected to a dead load, D, and live load, L, with
distributions N(1500, 200) kN and N(500, 200) kN respectively. Compute the
reliability of the column. (Ans. Ro = 0.96638)
6.5 The strength of a column, R, is given by
1
R = ff E/
a'
where E is the Young's modulus, I the moment of inertia and a the length of the
column. It is subjected to load Q. The mean values and coefficient of variations
of all the variables are given below:
P.E = 2.03 xI ()I N/mm• llE = 0. 1
P.r = 12.5 xI ()I mm• 81 = 0.05
p.,. = 5000 mm a.= o.o5
I'Q = 700 kN aa = o.3
If all the variables are lognormally distributed, determine the probability of
failure of the column. (A.ns. 0.11365)
6.6 A tension member of a steel truss is subjected to an axial load, Q. The strength
of the member is given by fy A, where fv is the yield strength of steel and A is the
area of cross section of the member. Given:
I'Q = 20 kN sa= 0.4
v./y= 286 N/mm• 3/y = 0.1
Find the area of the member for the specified reliability of 0.99865. That is,
Pt = 1.3Sxto-•. Assume variation in area is negligible. (A.ns. 167.8 mm1)
 7
Monte Carlo Study of
Structural Safety

7.1 GENERAL
In the process of giving predictions about s me phy ical system, the follow-
ing four ste p arc involved: (i) observation of a physica l system, (ii) formula-
tion of a hypothesis, (iii) prediction of the behaviour of the system on the
basis of the hypothesis, and (iv) performance of experiments to test the
validity of the hypothesis. Sometimes it may be either impossible or
extremely costly to observe certain processes in the real world. It is evident
that there are many situations which cannot be represented mathematically
due to the stochastic nature of the problem, complexity of the problem
formulation, or the interactions needed to adequately describe the problem
under study. For such situations defying mathematical formulation, simula-
tion is the only tool that might be used to obtain relevant answers. Even if
a mathematical model can be formulated to describe some system of inter-
est from the limited data available, it may not be possible to obtain a
solution to the model by straightforward analytical techniques and in turn
make predictions about the behaviour of the system. For example, let us
consider the probabilistic behaviour of a prestressed or reinforced concrete
flanged beam. We want to determine the reliability of the beam.

7.1.1 Failure of a Flanged Section (7.1, 7.2)

Prestressed concrete members usually have symmetrical or unsymmetrical I
sections. Because of the random variations of the parameters of the resis-
tance of a section, the failure of a flanged section can take place with the
occurrence of any one of the .following events:
y, - the section is under-reinforced with the neutral
axis in the flange
Y2 --the section is under-reinforced with the neutral
axis in the web
YJ- the section is over-reinforced with the neutral axis
in the flange
Y4- the section is over-reinforced with the neutral axis
in the web
 167

The occurrence of each event has a certain probability. The probability

tree for the failure of a section at the limit state of strength is given in
Fig. 7.1. If the above events are assumed mutually exclusive, it can be seen
that the probability of failure, PJ. of a flanged section is the sum of the
conditional probabilities of failures of the section under each given event,
and the same can be written as
4
PI 1: P(F I Yt) .P( Yt)
= 1-1 (7.1)

FIG. 7.1 Probability tree diagram

where P(F I Yt) denotes the conditional probability ofF for a given event Y1.
F denotes the event 'failure'. In Fig. 7.1, Ro represents the event 'reliable'
(i.e. safe). The conditional probability of failure of a section for any given
event (say Yt) is given by
P(F I Yt) = P[(R-S) < 0 I Yt] / (7.2a)
or P(F I Yt) = P[(R/S) < I I Ytlf (7.2b)
where S is the action (load or bending moment) on the section and R is the
resistance of the section. The resistance of a section is a function of the
various material and geometric properties of the section:
R = g(Xt, X:z, .. . , Xn) (7.3)
Because parameters X1 are usually random variables, the resistance is also a
random variable with density function /R and cumulative distribution FR.
If X1 are correlated, their joint distribution must be known. Assuming the
X1 in Eq. (7.3) are statistically independent, their joint density function is
~_.,.-,

n '1
/Xu Xu • . • , Xn (Xl 1 X2, .• , , Xn) ll /XJ(XJ) (7.4)
J- 1 j
and its cumulative probability is

FR(r) = P(R ~ r) = fa ..• f n"

J~t
!xJ(XJ) dXj (7.5)

The restrictiDn R ~ r defines the region of integration Gin Eq. (7.5). The
integral contained in the equation cannot be evaluated in a closed form.
158

A2ded to this, the evaluation of PI requires the evaluation of the probability

of occurrence of each given event Y;. Defining:
B,- the event that the section is under-reinforced
R2 - the event that the section is over-reinforced
83 - the event that the neutral axis lies in the flange
8~ the event that the neutral axis lies in the web

the probability of occurrence of the event Y1 is

(7.6)
The events 8, and BJ are dependent on each other and the density function
of each is again the function of the material and geometric properties of
the section. Hence, the evaluation of the probability of occurrence of each
event Y; is difficult. Finally, to calculate the conditional probability of
failure for the given event, Eq. (7.2a) or (7.2b) is to be used which involves
numerical integration. The evaluation of PI thus becomes a formidable task
even when adequate statistical data are available. In such cases, simulation
becomes a satisfactory substitute for finding solutions. Simulation is a
process of creating the essence of reality without ever actually attaining the
reality itself. As defined by Naylor (7.3): "Simulation is a numerical techni·
que for conducting experiments on a digital computer, which Involves
certain types of mathematical and logical relationships necessary to describe
the behaviour and structure of a complex real world system over extended
periods of time".

7.2 MONTE CARLO METHOD

7.2.1 Introduction
The Monte Carlo method is a simulation technique. One of the usual
objectives in using the Monte Carlo technique is to es timate certain para-
meter and probability distributions of random variables whose values
depend on the interactions with random variables whose probability distri-
butions are specified. As it is known that the ultimate resisting moment,
Mr, of a section is a function of several random variables, the probability
distribution of M, depends on the equation connecting these random
variables. As explained in the previous section, as closed form solution for
the calculation of the cumulative probability of Mr is not possible, the
Monte Carlo method can be used to study the statistical prc!_";,rties of Mr.
Secondly, as explained in Sec. 7.1.1, the failure of a flanged section can
take place under different events. Hence to study and simulate the complete
random behaviour of the section at the limit state of strength, the Monte
Carlo technique is the best suited method.

7 .2.2 Monte Carlo Method (7 .4)

Provided high speed digital computing facilities are available, a simple
 159

Monte Carlo technique can often be useful in obtaining the distribution

FR(r). Let R be a function of n independent random variables Y;:
R = g(YJ, Y2, ... , Y,) :-
..-
The technique consists of three steps:
I. Generating a set of values y;k for the material properties and geometric
parameters Y; in accordance with the empirically determined or assumed
density functions fn The suffix i is used to denote the ith variable and
suffix k is used to represent the kth set of values (Ytk, y2k, ... , .l';k, ....
y,k) ofthe corresponding variables (YJ. Y2, ... , Y;, ... , Y,).
2. Calculating the value rk corresponding to the set of values y;k obtained
in step I, by means of the appropriate response equation for resistance of
the section. That is
,-:f"rk = g(yJk, Y2k, ... , y;k, ... , Ynk)

3. Repeating steps I an 2 to obtain a large sample of the values of I?

and therefore, estimatin I n(r) .
This method can also be used to obtain distributions for M and Z where
·:> M=R-S (7.7)
R
'-v '>· Z = -
S (7 .8)

Here, R is the resistance and S the action. It is then only necessary to

obtain additional sample values for S in accordance with the density func-
tion fs and to combine the equation for resistance with Eq. (7 .7) or (7 .8) to
provide the direct means of calculating the values of M or Z.
The procedure for generating a random deviate from a specified distribu-
tion generally follows this pattern:
I. Generate a random number from the standard uniform distribution.
2. Perform a mathematical transformation of the standard uniform
random number {or numbers) wiJ_i.£!1 produces a rand m deviate fromt]le
desj~ed distribu iQJl. 1
· --- "' ·
3. Use t 10 transformed deviate in the experiment as required.
Various methods have been developed for the generation of uniform
pseudo-random numbers. Subroutines for this purpose are readily available
(7.3, 7.5). Built-in programmes are generally available in all the computer
centres to generate uniform random numbers. The transformation of Jb.~
uniform random number to the random varia e of the desired.distrib~1tion
ts o tained by the inverse transformation method, if ossible.
ln,erse Transformation Technique
Consider the cumulative distribution function, Fy(y), of the distribution to
be simulated. Fy(y) is defined over the interval (0, 1). Consider the standard
uniform variate V, which is also defined over the interval (0, 1).
160

Generate a value v for the stanuard uniform random deviate. For a

standard uniform variate V, the cumulative probability of V ~ vis equal
to i ' . That is, Fv(ll) = v. Hence if we set
Fy(y) = ;•

Then y is uniquely determined by the relation

Fy(y) = v* y = FY" 1(v)
Th is is pruphicnlly hown in Fig. 7.2. The PDF f Vis shown on the left
-;ide. The gcnemlcu uniform random number v is projected on the curve of
1he DF <'I' Y. The p int on the curve i projected down on the horizon-
1
ltd uxis '" ge t Ihe co rresponding va lue y. Hence, y ...,.., F r (11) is the variate
desired from the given distribution of Y.
fy(y)

PDF of V

-Y
> I
f I
I
~---1(')
- r-
>
....
LL.
0
a..
Q

- >---v -
_,I __ _ c CDF of V

'
' y
_, - v
- > y =Fy (v)

FIG. 7 .2 Inverse transformation technique

1
Wh en the inver e f Fy(y) i.e. Fy (v), does not exist or, it is so compli-
cated a 1 I e impracticable, other techniq ues such as rejection technique,
compositi on metho I, and approximation methods (7.5) are to be used.
Hence, the suggested procedure for drawing the ktb set of input values y1k
from the c rrespondin distribution of Frt is to generate first a set of n
random numbers, llik, with uniform density in the range 0 ~ v ~ 1.0. The
values of y;k are then obtained from
Ylk = Fit'(vlk) (7.9)
 161

The method of obtaining a random deviate of FYI ( ) using the inverse

transformation technique is illust~ated in the following example.
EXAMPLE 7.1 Using inverse transformation technique, develop expressions
for generating random deviates of Y having the following distributions:
(i) Uniform distribution, (ii) Exponential distribution, (iii) Weibull distri·
bution, (iv) Type 1 extremal distribution, (v) Type 2 extremal distribution,
and (vi) Type 3 extremal distribution.
Solution Uniform distribution:

Given
elsewhere

Then Fr(y) =J:(b ~ a) dt

=(~)
b- a

Set v = Fy(y) (Y~E)

b-a
I The inverse transformation is
y=Fi 1 (v) =a+ (b- a)v (7 .10)
where v is a uniform random number with uniform density in the range 0
and 1. .
· (ii) Exponential distribution:
Given fr(y) = ~e->.y y ~ 0
Fr(y) = 1 - e->.y
Set v = Fr(y) = 1 - e->-Y
-In (l - v)
y= (7 .11)
A
However, one can straightaway use the following equation:
- In (v)
y = A (7.12)

rather than Eq. (7.11), since (1 - v) is also from the uniform distribution.
(iii) Weibull distribution:
Given Jy(y) = (l,~yll-le-a.ytJ y ~ 0
-ayP
Fy(y)=l-e
Set

Hence (7 .13)
162

(iv) Type I Extremal (largest) distribution:

Given Fy(y) = exp [ - exp { - (1.(y - u)}] -- oo :::::; y :::::; oo
Set v =---= exp [ - exp { - (1.(y - u)}]
In [ - In (v)]
Then y=u- cc
- (7 .14)

(v) Type 2 Extremal (largest) distribution:

Given Fy(y) = exp [- ( ; rJ y~O

Set v = exp [- (~ rJ
Then y = u/( - In v)l/k (7. 15)
(vi) Type 3 Extremal (smallest) distribution:

Given Fy(y) = 1 - cxp [ - (~fJ y~O

Set v = 1 - exp [ ·- ( ~ rJ
Then y = u[ - In ( 1 - v)]llk (7 .16)
One can straightaway use the expression
y = u[ - In (v)]l/k (7 .17)
since (I - v) is also from the uniform distribution.
For normal distribution, the Box and Muller technique is used to
generate normal variates. Here, standard normal deviates are obtained by
generating two uniform random numbers v1 and v2 (with a uniform density
range between 0 and l) at a time. Then the desired standard normal variates
are given by (7.5)
Ul = [2 In 1/u.] 1i 2 cos (21T v2) (7 .18)
1 2
uz = [2 In I/v1] ' sin (21T v2) (7 .19)
ExAMPLE 7.2 (Normal distribution) Generate normal variates from the
distribution or Y following the normal distribution with mean J.L and
variance a 2 •
Solution First generate two uniform random numbers v1 and v2 in the
range 0 and I. Then, the standard normal variates are given by Eqs. (7.18)
and (7.19). We know that the standard normal variate is connected to the
normal variate Y as follows:
Y-f..l.
--=U (7.20)
a
where U is the standard normal variate. Hence we can get two normal
variates Y• and y2, using Eqs. (7.18)- (7.20). Thus.
Yl = Ot/( + jL
Y2 = OU2 + jL
 163

That is

.. Yt = I' + a[2ln l/vt] 112 cos (2'11'112) (7.21)

yz = I' + a[2 In 1/vt] 1/ 2 sin (2.,.oz) (7.22)
ExAMPLB 7.3 (Lognormal distribution) Generate the lognormal variates
from the distribution of Y following the lognormal distribution with para·
~

meters Y and a1n y,

Solution As in the case of normal distribution, here also, we first generate

two uniform random numbers Vt and vz and get two standard normai
variates using Eqs. (7.18) and (7.19). Using the following transformation

In (y(Y)
--"'-'----'- = u (7.23)
O'JnY

for transforming the lognormal variate to the standard normal variate, we

get two values of the lognormal variate Y:

Yt = Y exp (UtO'tn y)
Y2 = Y exp (u2a1n y)
Using Eqs.(7.18) and (7.19),

Yt = Yexp [a1n y(2 In I ivt) 112 cos (2'1TV2)] (7.24)

(7.25)
EXAMPLB 7.4 (Beta distribution) The PDF of the standard beta distribution
is given by Eq. (3.105) with parameters p and q, i.e.,

/x(x) - 1-q) xr (1
= B(p, 1
- x)q-t 0 ~ x ~ 1

The procedure to generate beta deviates is as follows (7.5):

Generate two standard uniform random numbers v1 and v2.
Set g = (vt) 11P and
Check whether g+h~l

If g + h ~ I, the standard beta deviate is given by

X= g (7.26)
(g + h)
If we want to generate a random deviate from the beta distribution of Y,
given by [Eq. (3.109)]
(y _ a)P- t(b _ y)q- t
Jy(y) = B(p, q)(b - a)P+g 1 0 ~Y ~ b

then use the transformation to transform the beta variate to the standard
beta variate, i.e.
(y - a)
x = -i':(b;-_- a+-
) '"
184

Hence, the required beta random deviate is given by

y = x(b- a)+ a (7.27)
EXAMPLE 7.5 (Gamma distribution) We are interested in generating gamma
distributed random deviates. The PDF of the gamma distribution is given
by Eq. (3.102), i.e.
,\ (Ax)k - l e-~ -'
.fx(x) = T (k) x;:?:O

A,k;:?:O
where ,\ and k are parameters of the distribution. The procedure to generate
gamma deviates is as follows (7.6):
(i) Let k' ~ I be the integer part of k.
(ii} Generate k' + 3 standard uniform random numbers, i.e. 111, v2, ..• ,
Vk' +3, satisfying the condition
(7.28)
(iii) The gamma distributed deviate is given by
I k'+J I v\'k
x = - · - I: In v;
,\ i-4
+ ,(-·In
11
v3) Ilk+
Ill
11(1-kJ
V2
(7.29)

7.3 APPLICATIONS
The Monte Carlo method has a variety of applications. It can be used to
study the distribution of a variable, which is a function of several ~andom
variables, to simulate the performance or behaviour of a system, and to
determine the reliability or probability of failure of a system or a component.
The simulation technique has been used in the reliability study of structures
by several research workers. Some of the applications are illustrated through
the following examples.
ExAMPLE 7.6 The strength of an axially loaded short column is given by
R == 0.67 CAc -+ As
where C is the cube strength of concrete, F the yield strength of the
reinforcing bars, Ac the area of concrete and As the area of steel. Given:
Size of the column = 250 mm x 500 mm
11-c "~· 19.54 N/mm 2 ac = 4. I N/mm 2
1-'I ~~ 469 N/mm2 ap = 46.9 N/mm 2
A, ·--~ 1250 mm 2
C and Fare normally distributed. The problem is to determine the distribu-
tion of R using the Monte Carlo method.
Solution A rea of concrete (A c) =•-= 250 X 500 - 1250
•= 123750 mm 2
R = 0.67 >~ 123750 C + 1250 F
= 82912.5 C + 1250 F (7.30)
 115

Using Eqs. (7 .21) and (7 .22), the random deviates of the normal variates of
C and Fare first generated. Using these values in the prediction equation
for R, random deviates for Rare generated using the Monte Car1o method.
The mean value and standard deviation of R, calculated after the genera-
tion of 500 and 1000 values, are given below:

/kR = 2.216X 106 fiR = 3.466 x 105 (after 500 values)

P,R ,: 2.207 X 106 fiR = 3.460 x 10 5 (after 1000 values)
These values, when verified with the theoretical exact values

/kR = 82912.5 p.c + 1250 P.F

= 2.206 X 106 N
and fiR = [(82912.5 flc)2 + (1250 fl£)2]112

J\ = 3.4496 X 10s N
agree very well. The error on the estimates of mean is almost nil and on
standard deviation about 0.3 per cent.
The frequency distribution of generated R is shown in Fig. 7.3; The
coefficients of skewness are -O.ot and+ 0.016 at the end of500 and 1000
generated samples respectively. Coefficients of Kurtosis are 2.637 and 2.989
at the end of 500 and 1000 simulations respectively. R, being normal, the
theoretical values of the coefficient of skewness· and Kurtosis are zero and
3 respectively. The normal distribution fits very well for the generated data.
Theoretically also, R should follow the normal distribution.

r 7 .3.1 Sample Size

We have seen in Example 7.6 that the generated data is used for estimating
the mean ·and standard deviation of the resistance of the column. As
larger and larger samples are used, the estimates are closer- to the population
values. The minimum size of the sample depends on the desired accuracy
of the estimates.
For the estimate of the population mean of a random variable X, the
minimum sample size is specified (7.3) such that the probability of the true
mean falling within the confidence interval

(7.31)

is (I - oc) per cent where Xm and s.~ are the sample mean and standard
, deviation of X, and oc is the level of significance. ~~12 is the value of the
standard normal variate at a cumulative probability of rx/2. If em is the
specified acceptable error in the estimate of the mean value of X, then

em= ~~/2 ( ~;;) (7.32)

166

240 ,......._
Ia) n :1000
IJR=220·7kN
~
.2 180 1--
17R = 34-6kN
§....
Gl
.----
~
....0 120
t--
j:I
60 r-
z

0 ,....._..r- ~
108 152 196 240 284 328
ResistancE' (k N)

328 (b)
n :1000--..
n =500 --tr-
284

.....
z
~ 2t.O
~
c
0
; 196
·:
a:

152

lOB 1
0·0005 0{)227 Q 1586 Q 5 0 841 0·977 0·9986 l' ( U)
-3 -2 -1 0 2 3 u
FIG. 7.3 Generated distribution of R: (a) frequency distribution of R and
(b) CDF on normal probability paper-Example 7.6

then the minimum size of thy sample for the estimate of the population of
X is given by

n
= cp211./2(Sx ]2 (7 .33)
em
For a large sa mple size say n > 120), the stand ard deviati n C>f sx is
equal lo 'Sx/ '\1 2n. Hence ftl r Ihe estima te of the statH.Inrd deviation of X,
Lhe minimum ~ izc is speci fic I sul.!h that th e probabi lity of th e true standard
deviation falling within the confidence interval (7.3)

Sx ± cpot/2( ; ; J
2
(7.34)
 '167

is (I - ex) per cent. Specifying

e, = ~«1 2(J;n) (7.35)

the minimum sample size for the estimate of the population standard devia- 1

tion of X is given by

n-
_ 1
2 ~«12 Sx ]2
2 [-;; (7.36)

where e. is the acceptable error in the estimate of the standard deviation of

X. Generally, the acceptable error = 5% and ex = 5% are taken.
If the Monte Carlo technique is used to generate straightaway the samples
for the margin of safety and determine the probability of failure, Shooman
(7.7) has proposed the following expression for the percentage of error on
the estimated probability of failure:

- p ]1 /2
o/
to
Error = 200 ( 1 liP! 'I (7.37)

Using this equation, the sample size can be calculated for the required
,1 accuracy.
ExAMPLE 7.7 Calculate the sample size, required .for the case study in
Example 7.6, to estimate the mean and standard deviation for an acc~pt­
able error of five per cent on the estimates of the mean and standard devia-
tion, and a level of significance equal to five per cent.
Solution If the mean value and standard deviation of the generated samples
for R (say after 500 samples) are
R. = 2.216 >~ t06 SR = 3.466 X 105

then the sample size required to terminate the simulation process, using
Eq. (7.33), is

11 = 2 ( ]2
f[>«/2 - s'·
e,
For ex = 0.05, confidence level = 1 - oc = 0.95. Thus
~a/2 = ~0.025 = ~- 1 (0.975) =--= 1.96

. 5 -
The allowable error on the mean = 100
R

Hence the sample size required to estimate the mean with oc = 5% and
e, = 5% is
_ 2[ 3.466 X 105 ] 2
11
-
96
(1. ) 0.05 X 2.2 1 X 106

----= 38
188

The sample size required to estimate the standard deviation with at = 5%

and e, = 5%, using Eq. (7.36), is

n = _!_ 1/>~,2 ( SR ]2
2 · e,
_ _!_ 2 ( 3.466X lOS )
- 2XI, 96 0.05X3.466XIOS

= 768
Considering both, the minimum sample size required is 768. In Example
7.6, it can be seen that ~t the end of 1000 simulations (n = 1000), the
error on the estimate of the standard deviation of R is less tban five
per cent.
EXAMPLE 7.8 Consider the column in Example 7.6, the strength of which
is given by Eq. (7.30). The column is subjected to an axial load Q. Given:
fJQ = 1.2 X 106 N ao = 0.35 X 106 N
~tc =
19.54 N/mm2 ac = 4.1 N/mm2
/tF = 469 N/mm 2 aF = 46.9 N/mm2
Variables Q, C and Fare n:Jrnudly distributed. Determine the probability
of failure of the column using the Monte Carlo method.
Solution The resistance of the column is given by [Eq. (7.30)]
R = 82912.5 C + 1250 F
The safety margin equation is
M = 82912.5 C : 1250 F- Q (7.38)
Using the given distributions and the corresponding parameters of C. F
and Q, the simulation is carried out and 20,000 samples are generated for
M. During the process of generation, the number of values of M falling
below zero are counted. At the end of 20,000 simulations, the number of
sample values of M falling below zero is obtained as 417. Hence, the
probability of failure of the column is
417
:-- (. ~ ' ' PI = 20000 = 0.02085

Using Eq. (7.37), if we want to have an estimate of Pf (say 0.02) with an

error ± lO per cent, the sample size required is
2002 ( I - 0.02
n = 0.02 >~ 102
= 19600
We have generated 20000 samples. Hence there is a 95 per cent chance that
the percentage error in the estimated fJJ is less than I 0 per cent.
The theoretical.va!Ltcs of {tAJ and aM, u~;ng Eqs. l3.77) anJ (3.79), are
 169

Since C, F and Q are normal, M also follows the normal distribution.

Hence the theoretical value of p1 is
II PI = P(M < 0)

=~ [0 -::M]
= ~(- 2.0488) = 0.0202
From the Monte Carlo method, the value ·o r p1 has been obtained as 0.02085
at the end of 20000 samples values of M.
During the process of code calibration, reliability analyses of existing
designs as per the current codal provisions are carried out for variou·s limit
states criteria. For this, the probability distribution and statistics of the
strengths of members (say, in flexure, tension, shear, torsion, etc.) for
various failure criteria are to be known. Statistics of the strengths of
members are established using the Monte Carlo method (7.8, 7.9). The
determination of the statistics of the flexural strength of RCC beam is
illustrated below.
EXAMPLE 7.9 A simply supported reinfo,rced concrete beam of span I is
subjected to a uniformly distributed live load L and a dead load D. The
breadth, effective depth, and area of steel on the tension side are b, d and
Aa respectively. It is given:
b = 300 mm d = 550 mm A, = 1039.5 mm2
P.c = 17.58N/mm2 ac =
3.164N/mm2
P.F = 469 N/mm2 ap = 45.9 N/mm2

b: mean deviation = + 10.29 rom a = 9.47 mm

d: mean deviation = 6.25 rom a = 3.79 mro
C and F are the cube strength of concrete and the yield strength of rein-
forcing bars respectively. Their nominal values are L5 N/rom 2 and
415 N/mm 2 respectively. The above data are based on the actual field data,
given in Chapter 4, for Indjan conditions (7.9). C follows the lognormal
distribution and all other variables are normally distributed.
Study of Distribution of Strength in Flexure
The theoretical model for the ultimate resisting moment of a RCC beam is

R = FAs a[ 1 - 01
· Za%A,] (7.39)

This equation is obtained when the material reduction factors attached to

the strengths of concrete and steel are removed in the equation given by
IS: 456-1978 for computing the design strength of a singly reinforced
beam. There will be, in general, a certain roo,del error associated ~ith every
prediction equation for the strength of a member. If B is the model para-
meter, let p.s and 1111 be the mean and standard deviation of B. For flexural
170

strength, fLB = 1.01 and aB = 0.0465 (7.8). Attaching this model parameter
8 to the prediction equation, Eq. (7.39) becomes

R ""c BFAs d[ I - _9_.7? FA ,

bd
J (7.40)

From the given data, f-tb = 300 + 10.29 =--= 310.29 mm and f-td = 550
-, 6.25 ,-, 556.25 mm.
Using the Monte Carlo technique, random deviates of various variables
are generated ( B is assumed to follow normal) and then, using the same in
the prediction equation, sample values of Rare generated.
Generally, the values of Rare normalized with its corresponding nominal
value Rn, so that the statistics of R of different designs could be compared.
R, is obtained by substituting the nominal values of the variables in the
prediction equation. For this problem,
07
Rn = (1.0)(415)(1039.5)(550)[ 1- · J0~~~;5~~0I3i' 5 ]
= 2.055, 10 8 N mm
Hence, instead of studying the distribution of R, the distribution of R/R.
is studied. It is to be noted that Rn is deterministic and is constant for a
particular design. The frequency distribution of the generated samples of
R/ Rn and the statistics of R/R,. are given in Fig. 7.4. It is found that the
normal distribution fits the generated data well (based on the chi-square
test at five per cent level of significance).
7 ~-------------------------------------------------,
n :30000
M
0 6 Mean= 1160
><
oJ)
so= 0·105
§ s.
~

~ 4
a;
1/)
g J .

0 2
(i;
.Q
E
::;)
z 0~--~~~~~~L__L~~~---L--~-L--~~-=:L---
0 ·720 1·061 1·175 1·289
R/Rn
FIG. 7.4 Frequency distribution of the resistance of RCC beam-Example 7.9

During the reliability analysis of the present designs, the statistics of the
strengths of members for various combinations of basic variables for each
failure criteria (shear, flexure, torsion, etc.) are studied in detail using the
Monte Carlo technique, and then fixed. To be consistent, Ellingwood, et al.,
(7 .8) have fitted a normal distribution to the lower tail below five per cent
fractile of the generated strength distribution, and the statistics (mean and
 171

standard deviation) of R/ R, are established. Typical values of resistance

statistics of RCC member , established for Indian conditions, are given in
Table 7.1 (7 .9) . In Table 7. I, YR is the ratio of the design value of R to its
nominal value.
TABLE 7.1 Typical resistance statistics of RCC members
Steel Concrete
Member 8
grade grade ILRtR 11 "'R
Slabs
One way (SS) Fe 250 M 15 1.433 0.124
Fe 415 M 15 1.275 0.124 Range
Two way (SS) Fe 415 M 15 1.281 0.124 0.835-0.865
One way (C) Fe 415 M 15 1.263 ... 0.136 Average
Two way (C) Fe 415 MIS 1.286 0.129 0.85
Beams (flexure)
Singly reinforced Fe 250 M 15 1.288 0.104
Fe 415 M20 1.179 0.103 Range
Fe 415 M 25 1.169 0.101 0.835-0.845
Fe 415 M 15* 1.197 0.105 Average
Doubly reinforced Fe 415 M 15 1.151 0.103 0.84
Beams (shear)
Fe 250 M 15 1.355 0.166 Range
Fe 415 M 15 1.277 0.165 0.855-{),865
Average
0.86
Columns
Compression failure Fe 415 M20 1.29 0.152 Range
Fe 415 M 20* 1.38 0.224 0.68-0.79
Average
0.725

Tension failure Fe 415 M20 1.19 0.13 Range

Fe 415 M 20* 1.22 0.15 0.68-0.89
Average
0.8

Note: SS = Simply supported

C = Continuous ,
* = Indicates nominal mix
Sometimes in engineering problems we may have to deal with situations
while studying the performance of a system under two failure criteria or
two different designs when they are correlated . Under such conditions the
correlated sampling technique may be used. This is illustrated in the
following example.
EXAMPLE 7.10 Consider the portal frame shown in Fig. 7.5. Consider the
two failure modes shown in Figs. 7.5b and 7.5c. It is given that
I'M3 = P,M4 = = 300 kN m
P,Ms
CIMJ = CIM4 = = 30 kN m
CIMS
1-'Ml = P,M2. = P,M6 = I'M1 = 50 kN m
CIMI = CIM'J. = CIM6 = 11M1 = 5 kN m
172

1-<a = 180 kN UQ = 40 kN
J.LH = 40 kN U!f = 10 kN
a a

-r
H J
rt---- --.1.
4- -
5
--+.... 6

4m

l, 7
1---6m--~

(a l Frame (b) Failure Mode 1

a
H

FIG. 7.5
r (c) Fai lure Mode 2
Correla ted failure modes-Example 7.10

where M; is the plastic moment capacity of section i. All variables are

normally distributed.
Using the mechanism method of analysis (7.10), sefety margin equations
for the two failure modes can be written as
Zt = M2 -t- 2M4 + M6 - 3Q (7 .41)
Z2 = Mt + M2 + M6 + M1 - 4H (7.42)
The probability of failure of the frame under failure mode i is
Pfi = P(Z; < 0)
The probability of failure of the frame under failure modes Zt and Z2 is
P/12 = P(Z1 < onz2 < o) (7.43)
The problem is to generate the joint distribution of Zt and Z2 and then
calculate P/t2.
It can be seen that Zt and Z2 are correlated as they depend on the same
basic variables M2 and M6. The correlation sampling technique can be used
to generate the joint distribution of Zt and Z2, and to calculate P/12· The
procedure is to generate normal deviates of Mt, M2, M4, M6, M1, Q and H
using their respective parameters. Substituting the generated deviate of each
variable in the equations for Zt and Z2, the random deviates of Zt and Z2
are generated. While generating random values z1 and Z2 for Zt and Z2, a
 173

count is made when z1 < 0 and z2 < 0 are simultaneously observed. The
process is repeated for generating a number of samples. The procedure is
outlined in the flow chart given in Fig. 7.6, where the number of simula-
tions has been fixed at 20000.

E•r
r:. -
i Input: paramt'ters and distribution
I of t'ath variablro M, M2,M4,M6
L_ M7 ,a
and H. Selt'tt n:. 20000

Gt'nt'rate 7 valut's
v1 , v 2 , v 3 . v4 . v5 • v6 • v7

Transform to
dt'viatE's m 1 ,m 2 • m 4 ,m 6 ,m7,
q, h of torrE'Spondin g distri-
butions of M1 1 M21 M4• M6 1
M7, Q and H.

Us£' Eqs 7·39 and u.o

ZJ= m1 +2m2+ m4- 3q
z 1 =m1 t m2+ m4+m 6 -4h
Gt'nt'ratE' z 1 and z2.

Yes
Is and z 2 <0
No

No

FIG. 7.6 Flow chart-Example 7.10

174

Result:
Number of samples for the condition
(zt < 0 and z2 < 0) = 374
Hence

fiJI~ ,-~ P(Zt < onz2 < O) = ~~~2 0= o.o187

Let us compare this with the theoretical value:

/-LZI = 160 Uz[ = 134
/-L7.2 = 40
The probability of occurrence of failure mode l is
Pit= P(Zt < O) =--' cJJ(-160/134) = 0.1!62
Similarly,
Pf2 = P(Z2 < O) =-= f/1( -40/41.2) = 0.!658
The correlation coefficient between Zt and Z2 is (Eq. 3.77)

p = Cov
-· (Z1, Z2)
Uz t11Z2

= (1)(1) Var (M2) + (1)(1) Var (M6)

Uzt<1Z2

= 0.009
This being negligible, and assuming Zt and Z2 are statistically independent,
we have
P/t2 = (PJ"t)(PJ2) = 0.0193 l

The value 0.0187 obtained from the Monte Carlo technique agrees well with
the theoretical value.
In engineering problems, quite often we come across situations when
variables in the safety margin are correlated. Let the safety margin M be
M = Xt- X2

The variables Xt and X2 are correlated. We want to determine the joint

distribution of Xt and X2, i.e. the distribution of M. In such situations, the
correlated variables are first transformed to uncorrelated variables Yt and
Y2 using the transformation matrix [T].
Y = [T]'X (7.44)
where each column of matrix [T] contains an eigen vector corresponding to
the eigen value of the covariance matrix [Cx].

Var (Xt) Cov (Xt , X2) ]

[Cx] = [ Cov (X1, X2) Var (X2)
 175

Jf ~~ and ~2 are the eigen values of the matrix [Cx] and e1 and e2 are the
corresponding eigen vectors,

~,
11 12
I Tl = [c1, e2l [ e e ]
e21 e22
The expected vallle and covariance of variables Y1 and Yz are
l E(Y) = [T] 1E(X) (7.45)
'I (7 .46)
[Cy] = [T]t[Cx][T]
where [Cyl is the covariance matrix of the variables Y1 and Yz. The dia-
gonal elements of [Cy] are Var (Y) which are equal to the eigen values of
ICxl. [T] 1 is read as the transpose of [T].
Var (Y) is nothing but a matrix having diagonal elements equal to the
eigenvalues and other terms zero. That is
~~ 0 ]
Var (Y) = [ (7.47)
0 "2
Since rTJ is an orthogonal matrix,
X= [T]Y (7.48)
Hence, the given equation forM can be written in terms of the uncorrelated
variables Y. Knowing the mean and standard deviation of )', the sample
values for M can be generated using the Monte Carlo method. This is
illustrated in the foJlowing example:
EXAMPLE 7.11 Consider the safety margin equation
M = X1Xz- X3
where X1 and Xz are correlated. The covariance matrix is given as
' [ 0.0222 0.0111 0 ]
[Cx] = 0.0111 0.011 0
0 0 0.0308

1.222 ] I 0.149 l
P.x =
[
0.620
1.050
0.1755
ax= l 0.105 J
It is given that all Xt are normally distributed. The problem is to determine
the distribution of M.
Eigen values of the matrix [Cx] are t\1 = 0.02903; t\2 = 0.004167;
t\3 = 0.0308. (Note: . The computation of eigen values is illustrated in

-
Example 8.11).
The corresponding normalized eigen vectors are

J
l
0.8516 ] -0.5242
e1 = 0.5 242 ez = 0.~16
[
0
176

Hence the transformation matrix is

0.8516 -0.5242
[Tl =: 0.5242 0.851(1
[
0 0
Using Eqs . (7.43) nnd (7.44),

J
l
0. 1704
l
11.591 ]
!Joy= 0.2536
0.62
ay = 0.06456
0.1755
Using Eq. (7.46),
X1 = 0.8516 Yt - 0.5242Y2
X2 = 0.5242 Y1 + O.R516 Y2
X,= YJ
Hence
M = (0.85l6ft --- 0.5242Y2)(0.5242Yt + 0.8516Y2) -- YJ
Now Y1, Y2 and YJ are independent variables. Since X1, X2 and XJ arc
normal, Y1, Y2 and YJ are also normal. Knowing the mean and standa1d
deviation of Y;, the normal deviates of Y; can be generated. Using the usual
Monte Carlo technique, the required samples forM can now be genera tell
to study the distribution and establish the mean and standard deviation nl'
M. Figure 7.7 shows the generated cumulative distribution of M.
10 r-------------------------------~~--------~

0·8
M~n= 0·671
so= 0·313
~0 ·6

0·2

Safety margin ,M

FIG. 7.7 CDF of safety mrJrgin with correlated variables-Example 7.11

 177

REFERENCES
7.1 Ranganathan, R., "Reliability Analysis and Design of Prestressed Concrete
Beams at Different Limit States", A Ph.D. Thesis, Civil Engineering Dept., liT,
Kanpur, May 1976.
7.2 Ranganathan, R. and P. Dayaratnam, "Reliability Analysis of Prestressed Con-
crete Beams", Journal of Bridge and Structural E1~gineer, Vol. 8, No. 2, June 1978,
pp. 11-24.
7.3 Naylor, T.E., Computer Simulation Experiments with Models of Economical
System1, John Wiley, New York, 1971.
7.4 Warner, R.F. and A.P. Kabaila, "Monte Carlo Study of Structural Safety",
Journal of Struct. Div., Proc. ASCE, Vol. 94, ST-12, Dec. 1968, pp. 2847-2860.
7.5 Philips Don T., A. Ravindran, and J.J. Solberg, Operations Research: Principles
and Practice, John Wiley, New York, 1976.
7.6 Ang, A.H.S. and W.H. Tang, Probability Concepts in Engineering Planning and
Design- Vol. 11, John Wiley, New York, 1984.
7. 7 Shooman, M.L., Probabilistic Reliability: An Engineering Approach, McGraw-Hill,
New York, 1968.
7.8 Ellingwood, B.R., T.V. Galambos, J.G. McGregor and C.A. Cornell, "Develop-
ment of a Probability Based Load Criterion for American National Standard
ASS", NBS special publication S77, U.S. Deptt. of Commerce, Washington, D.C.,
June 1980.
7.9 Padmini Chikkodi and R. Ranganathan, "Partial Safety Factors for RCC
Design", International Journal of Structures, Vol. 8, July-Dec. 1988, pp. 127-149.
7.10 Neal, B.G., The Plastic Methods of Structural Analysis, Halsted, 3rd edition, 1977.

EXERCISE
7.1 It is given that
Y= xlx.
where X 1 and X, are statistically independent lognormal variates. Given the
parameters

X1 = 10 x. = 5
"lnXI = 0.3 alnX2 = 0.05
determine the distribution of Y using the Monte Carlo method and check whether
it is lognormal with parameters

y = so and alnY = 0.304

7.2 If the variable Y is
Y = xl + x.
where X 1 and X 1 are exponentially distributed independent variates with respective
parametres .\ 1 and .\1 , where being 6 and 12 respectively, determine the distribution
of Y using the Monte Carlo method and check whether it follows the exponential
distribution with parameters,\ =i 4 and ILy = 1/4
7.3 The annual maximu m wind speed obscrvedal'a station follows the Type I
extremal (largest) distribution with parameters
u = 81.4 kmph IJt = 0.126
Determine the distribution of a 20 year maximum wind and lhe probability of the
lifetime maximum wind speed exceeding the specified design speed= 120 kmph.
Use the Monte Carlo method.
178

7.4 The total load Yon a structure is given by

Y~D+L

where D and L are correlated with the correlation coefficient O.S. It is given that
I'D = SO I'L = 100
aD= S DL = 40
Generate the distribution of Y if D and L are normally distributed. Check whether
it is normal.
7.5 The distribution of Lapt follows the gamma distribution with parameters

A= 23 .87 k = 0.328
Determine the distribution of the lifetime maximum live load for 10 occupancy
changes during the life of the building using the Monte Carlo technique.
7.6 The ultimate strength of an axially loaded short RCC column is given by
R = kCA + Ys
where k is a constant, Cis the cube strength of concrete, A is the area of concrete ,
Y is the yield strength of steel , and sis the area of steel. It is given that k = 0.67
and s = 1250 mm 2 • Variables C, Y and A follow uniform distributions as given
below:

fc(c) = -c- -c
1 -
c, ~ c ~c.
1

/y(J') = ----
Yt- Ya
Yt ~ Y ~ J's

/A(a) = --- a1 ~A~ a,

a,- a,
where r1 = 18 N/mm 2 c2 ~ 25 N/mm•
)' 1 = 420 N/ mm• Ya ~, 460 N/ mm 2
a1 ~ 1000 em• a2 ~' 1100 mm 2
Determine the distribution of R using the Monte Carlo technique.
I

lI
j
8
Level 2 Reliability Methods

8.1 INTRODUCTION
The Joint Committee on Structural Safety (8.1) classified the structural
reliability analysis and the safety checking into three groups. They are term-
ed as Level I, Level 2, Level 3 methods. The levels are defined as follows
(8.1, 8.2).
Levell
A design method in which appropriate levels of structural reliability are
provided on a structural element basis (exceptionally on a structural basis)
by the specification of a number of partial safety factors, related to some
predefined characteristic values of the basic variables.
/Leve/2
A design method incorporating safety checks only at a selected point (or
· points) on the failure boundary (as defined by the appropriate limit state
equation in the space of the basic variables)- rather than as a continuous
process, as in Level 3.
Leve/3
Safety checking based on 'exact' probabilistic analysis for whole structural
systems or structural elements, using a full distributional approach based
on failure probabilities, possibly being derived from optimisation studies or
assessed by other approach criteria.
The present structural design (8.3) with explicit consideration of the
number of limit states (being called as limit state design) is nothing but Level
I design. It is advocated that the present design be called as Level 1 design .
The limit state is a criterion to define a particular failure or performance
condition. In Level 2 methods, certain idealisations and assumptions are
used. Mean values an iances of the random variables only are required.
In advanced Level 2 methods, distributions also can be taken care of in-an-
approximate way. Reliability levels are defined by safety indices or equivalent
"operational" or "notional" probabilities. Level 2 methods are approximate
compared to Level 3 methods where full joint probabilistic description of
the randcm variables are used, and they are purely probabilistic methods
and are exact in estimating the reliability. It is recognised that Level 3
methods will be used rarely- for checking special structures or at research
level. Level 2 methods are more practical-oriented and are quite suitable for
180

design . They are suitable for calibrating codes on reliability basis. Level 2
methods will be used by committees engaged in calibrating codes for the
evaluation of partial safety factors in a rational manner. It is realised that
structural designers will be working with Level 1 methods of checking. It is
also to be understood that Level 1 method is not a reliability method.
This chapter deals with Level 2 methods (including advanced Level 2
methods) of reliability analysis.

8.2 BASIC VARIABLES AND FAILURE SURFACE

In any engineering problem, several random variables are involved. In
structural engineering problems, geometric parameters of the section (i.e.
dimensional variations), physical properties of the materials (cube strength
of concrete, yield strength of steel, Young's modulus of steel and concrete,
etc.) and loads (live load on floors, wind load, etc.) coming on structures
are subjected to random variations. If the coefficient of variation of a
random variable is very small (e.g. dimensional variations in many cases),
probably this may be ignored and the variable may be considered as deter-
ministic. Hence in any engineering problem, the parameters which are to be
considered as random variables are initially fixed and those random variables
are called as basic variables. Let these basic variables be Xt, X2, ... , Xn .
Any equation that is developed for a particular limit state condition (failure
condition) of the structure will be interconnecting these basic variabies and
hence it is a function of these variables.
Let this function be
(8.1)
This function is called a failure function. This is nothing but representing
the margin of safety, M, which can be written as
M=R-S (8.2)
where the resistance R and the action S will be in terms of the basic
variables Xt, X2, .. . , Xn. Hence,
M = g(Xt, X2, ... , Xn) (8.3)
When this failure function is made equal to zero, i.e.
g(Xt, X2, ... , Xn) = 0
it is called a failure surface (or limit state surface). The safety is ensured by
specifying a small value for the probability of reaching a particular limit
state. The magnitude assigned depends on the serviceability of the conse-
quences of reaching the: particular limit state. If fx. (x) is the probability
density function of the jointly distributed variables X1, X2, ... , Xn, then
the probability of failure (or probability of reaching the limit state) is

(8.4)
 181

where X = (X1, X2, XJ, .•. , X")../

X= (XJ, X2, XJ, ..• , X,) ~

dx = (dx1, dx2, ... , dx,)~

The multiple integral is to be evaluated over the region g < 0.
The failure surface equation divides the design space into two regions,
viz. (i) safe and (ii) unsafe failure regions. For the two variable case, i.e. if
the failure function is g(X1, X2), this is shown in Fig. 8.1. It may be noted
that the same failure surface may be represented by different equivalent
failure functions.

Foilur•
g(xl ,x2 )<0

FIG . 8.1 Concept of design space . failure surface

and failure and safe regions

Recall the fundamental case-A structure with resistance R subjected to

an action S- discussed in Chapter 6:
M = g(R, S) = R - S
the failure surface equation is
g(R, S) = R - S = 0
r: It has already been derived [Eq. (6. 12)], assuming R and S independent, i.e.

FJJ.<·~ dx ' \ 'V (8.5)

Equation (8.5) is a particular case of Eq. (8.4) and differs in two main respects.
Equation (8.5) is not expressed in terms of the basic variables X1; but in
terms of state variables R and S. Equation (8.5) is concerned with a specific
failure mode related to the form of R and S. In general, R and S
will be in terms of the basic variables X1. The PDF of R and S wilt depend
on the PDF of individual basic variables and the nature of functions relat-
ing them to particular state variables R and S. In many practical cases, R
and S will be related to some of the same basic variables and hence will be
corre la ted . l ni tiaJi y, the st ructural sa fety was assessed using Eq. (8.5), igo ting
correlnti~m GeiW'CenRandS, i(Udts. "'Ose OfEq. (8.5) is not satisractory
"-- ---
oecause o the lack of.,......__ sta tistical data fo r the varia bles R and . Lf rhe
 182

distributions of R and S are directly known and if they are uncorrelated,

Eq. (8.5) will give the exact value of the probability of failure.
The probability of failure provides a basis for quantifying structural reli-
ability. All uncertainties in the joint probability law of all basic variables (in
the fundamental case, R and S) must be known. However, in practice, these
probability laws are seldom precisely known due to general scarcity of data.
In many cases, the probability Jaws of individual basic variables will not be
known and it may be difficult to obtain. The joint distribution of all the
basic variables, in general, is impossible to get in the field. If the failure
function is highly nonlinear, it may be difficult to numerically evaluate the
integral [Eq. (8.4)] even if the marginal distributions of the variables are
known. These difficulties have motivated the development of approximate
methods of evaluating structural reliability.

/ 8;'3 FIRST-ORDER SECOND-MOMENT ~ETHODS (FOSM)

In these metods, the random variables are characterized by their first and
second moments. In evaluating the first and second moments of the failure
function (i.e. say, the mean and variance of M which is a nonlinear function
of the basic variables), the fir st order approximation is usecL That is why
these methods are called first-order second-moment methods. In the case
of nonlinear failure functions, linearisation is performed using Taylor's series
expansion in the reliability analysis.
Consider the fundamental case with only two basic variables R and S:
Pf = P(l< < S)
M = g(R, S) = R -- S (8.6)
The failure surface equation is
R-S=O (8.7)
Cornell (8 .3) first detlned the reliability index ~as

(8.8)

where fl.M and aM are the mean value and standard deviation of M. That i~,
~ IS the reciprocal of t11e coellic i~ nt r VClrlation in M . The concept of~ is
illustrated in fig . 8.2a which -sh w the PDF of M for the fundamental
case - two variable problem. The safety is defined by the condition M > 0
and therefore, failure by M < 0. The reliability index may be thought of as
the distance from the origin (M "'"' 0) -to the mean fl.M nu:_<m.\~ed in standard
deviation units. As such, ~ is a 111 ~,1 s ure of' lhc pr.o..llll.bilil¥JllaL11L\Y.ilLb
less thun zero . If
!l :lf - f3a .-., ? 0 (8.9)
then the reliability in terms of the safety inde x is atleast ~ .
When both R ami .)' are normal an ~d e nt,
I'M ..~. i'R · l' s aM = (a71 + a])i !Z
 183

D J.ioR -J.ioS ( )
,.. = (a~ + cr~)l/2 8.10
When both 1J. and S are lognormal and independent, the alternative
formulation for failure [Refer Eq. (6.9)] is taken. That is, for failure

(;) < 1

In ( ~) < 0
The failure surface equation is

M =In ( ~) = 0

Using the small variance approximations,

aXt = Var [1n ( ~)] ~ (8~ + 8~)

f3 = In (P.R!P.s) ~ (8.ll)
(8J + 8~)1/l
The above format of Eq. (8.11) (the corresponding reliability concept depict-
ed in Fig. 8.2b) has been used for the development of probability based
load and resistance factors for the design of steel structures (8.5).
M<O
-

ln(r/s)

(b)

FIG. 8.2 Concept of reliability index (a) M = A - S; (b) M = In (R/S)

lf the safety margin is a linear function of basic variables and if basic

variables are normally distributed, the safety margin M is also normally
distributed.
Let
~· (8.12)
 184

Using Eqs. (3.77) and (3.78),

n
/-LM = bo + i-1E bt/-Li (8.13)

2 n 2 2 n
aM = E
i=l
bt a, + 2 i=lEn E
J-1+1
Pub;bja;aj (8.14)

where bo and b; are constants and PU is the correlation coefficient between

X1 and Xh and /-Li = /-LX; and a; = ax,.. The probability of failure is related
to the reliability index as follows:
Pr"'-= cP(-~) ..:/ (8.15)
or ~ = --<P-l(pJ) {/ (8.16)
For a linear combination of the normally distributed variables, using {3 the
true_value of reliability can be obtained.
v,..£XAMPLE 8.1 Calculate the re liabil ity index of the beam (against the limit
state of collapse in flexure), shown in Fig. 8.3, subjected I<> a self-weight
QJ and a live load Q2. The flexural resisting moment capacity of the beam
is R. It is given that
fto 1 = 400 N au 1 ~ -, 10 N
!tQ
2 ~~o 5000 N UQ2 = 2000 N
/-LR = 10000 Nm a11 = 1000 Nm

FIG. 8.3 Simply supported beam-Example 8.1

.";'ol111ion Maximum bending moment due to external loads is

I I
Me= Qt -8- -+· Q24

CCC Qt (:) + Q2 (:)

Hence, Action S = ~1 -!- Q2
The failure function (R -· - S) is

This is a linear function of variables R, Q 1 and Q 2•

.M =-~ R -- Q, - Qz
2
 185

Using Eqs. (8o 13) and (So 14) and assuming R, Q1 and Q2 are mutually
independent,

Substituting the given data, we have

1
P-M = 10 ·- 2(0.4) - 5 = 4o8 kN

u~ = (1)2
.
.+ (..!.)\o.01)2
2
+ .(2)2
UM = 2.236 kN
Hence the reliability index is

~ = (2~~6) = 2.147
[t has so far been assumed that the failure function is a linear combina-
tion of the basic variables. However, this may not be t~ue most of the Limes
in practlca cases. 1e function for M is nonlinear, the approximat
va lues of P-M and UM are obtained using Tay!or"s series expansion of lincaris-
ed safety margin M. Let
oM = g(X1, X2, . o ., X,)

Using Taylor's series expansion abo~~· the point

X
.= ( . .
x,,Xz,. o ., X,)
-•

M • •
= g(X1,Xz, • 1:• (oX
o. o' X,)+ ag
1-1 1
Ix•) .
· (X1- X1)

'' ( ()2g
+I: - 2
/ )(x, -2 x,•)2 + .. (8.17)
1-1 ax/
0

x•

Recall that ( :.:.) x• means that

0~ is evaluated at X*.

Retaining only the linear terms, we get

M e:t g(Xt~ ,Xz, • o •• , Xn) • 1:" (oX

+ 1-1 ag) (X, --X1)• (8.11))
1 x•

In the case of mean value methods, the point = ~" = That · • the x;
expansion is about the mean 01nt.Tn such a case, for Eq. (8.18)-
P-M = E[g(X)] e:t g(p.i, f'-2, p.,) +0 (8.19)

--
o • o,
 186

Var [g(X)] ~ Var [g(/lJ, ... J.l·n)J

-1· Var '·g I )(X;

[ ;_,E" (a-; -- ft;) J
''""
(·g:(lx, II'- means that cgjoX; is evaluated at f.I.Xp flX2' . . . , flX,.
Since Var r.dtiJ, fL2, .. . , f.l.n)l = 0, and assuming X; are uncorrelated,
2
aM = Var [g(X)] ~ _1..:" [ -:>X·
ag ) ]2(a;) 1 (8.20)
·~I c; I 1'-

11 here a; ·-= ax;. It is to be noted that both flM and aM arc only first order
:IJ)~.!.lns.
If the second order terms in Eq. (8.17) are taken into account, the second
order approximation of flM is obtained as

iJlg I 2
fLM = g(ftl. fL2, .. . , fln)-,
' II
E --, (a,/2) (8.21)
;.. , axr '""
Even in the second order approximation of /-LM, only the mean value and
'aria nee are required. Hence in practice, the second order approximation for
ft_\t nnd first order approxin~ation for aA-i are used. H~vet:_,_ in Level 2
methods the nonlinear function is linearized retaining only linear .terms in the
T~ylor's series expansion, and hence the first order approximate values of
1111 and aM are used. The extent to which the values for /-LM and aM obtained
h~ using Eqs. (8 .19) and (8.20) are accurate, depends on the effect of neglec-
lillg higher order terms in the Taylor's series expansion and the magnitudes of
1he coefficient or variation of X;. If g( ) is linear and the basic variables
:11-c uncorrelated, Eqs. (8.19) and (8.20) are exact. If X; are correlated, the
lir~t order approximation of a.\1 is obtained as

I
n~, · ~ i;; (a~, /J(a: IJcov (X,
1 1
X1)l (8 .22)

J l' \1 1'1.1. l' ,_ Determine the r liability intlcx ~ r a !ilecl tension member,
l11n in!! ten~lie st reng th R, ubjected to a tensile lood Q. Given:
1-LR = 280 N/mm 2 aR = 28 N/mm 2
flQ ~, 5000 N aQ = 2000 N
/-LD = 6 mm an= 0.6 mm
The member is circular in cross-section of diameter D.
Solution The induced stress in the member is 4Q/1TD 2 and this is the action
(i.e. load effect). Hence the safety margin is
4Q
M=R - -
-rrDl
 187

Using Eqs. (8.19) and {8.20)

I
..1
/1-M ~ /1-R - ..i.(/1-Q)
rr 11-~
II
= 280 _., !( ~~ ) 5 0
= 103.16 N/mm 2

aL ~ (~~/J <a~> + (~~IJ <a~> + (~~tr (ab)

2 2

= (1)2>~ (282) + (.,~ 2 ): (a~) + (!2 3): (a~)

= 282 + (- 4 )2 2 (8fta) 2 (av)2
2 (a 0 ) + - 3
rr~-tn '"'-'D
= (28 2) + (0.00125)(2000) 2 + (3474.7)(0.6)2
,
a:it = 784 + 5003.5 -\- 1250.9
aM '= 83.9
I
I fJ = {103.16/83.9) = 1.23
l
• ExAMPLE 8.3 The reliability index for the beam given in Example 8.2 is
calculated using a different failure function.
Solution Let us consider Q as the action and the capacity of the section
as (R)(rrD 2/4). Then the margin of safety M is

M = rrD2)
(R) ( 4 -- Q

The failure occurs when M is less than zero. The mean value and variance
of Mare calculated using Eqs. (8.19) and (8.20):

/1-M= (/1-R) (TT~b) - /-!Q

= (280)t~·r) _ 5000 = 2916.8 N

alt ==(~~):(a~) + (~~): (ab) + (~~): {ab)

=('"~by (ak) + (rr~rD r (ab) + a~

= ("! Y(28
2
2) + (rr 280 X 6/2)2 (0.62) + (20002)
<1M= 2670.9 N
Hence the reliaQility index {J is
fJ = (2916.8/2670.9) = 1.0918
188

In the last two examples, the safety margin M is a nonlinear function of

the basic variab les. The mean value method has b · d and lioeari z~­
tion of M is ab ul the mean value . ll can be observed that differen.t value
of f3 haveoeei1o btamed- for the same problem. That is, f:J cha_ng~s when
different but equivalent nonlinear failure functions are used. This can also
be demonstrated again. --- - · --- .
For the fundamental case when R and S are uncorrelated,
M=R--S

~. = ~: = [(af~-"D~' 2 ]
(8.23)

If the equivalent failure function, given below,

M = In ( : ) = In R - ln S (8.24)

is selected, we get

{3 = /t tntR /S)
2
UJu(Rt SJ

If linearization of the safety margin

M =In R -- InS

is done about P.R and P.s, then

(8.25)

It is clear that f3, and ~2 are not equal. Hence the reliability index {3,
defined by the equation {3 = P.M/aM, is thus not invariant with regard -to
the choice of the failure function. If the linearization is done about the
mean value, the method can give different values of ~. that is different
values of Pf, for the same problem. When the failure functions are linear
functions of the basic variables, they will yield same values of~. and hence
the same p1. In general, an expansion of M about the mean point should
not be used. Mean value FOSM methods have two basic shortcomiv.gs:
(i) g( ) is linearized at the mean value of basic variables. When g is
nonlinear, significant errors may be introduced at increasing distances from
the linearizing point by neglecting higher order terms in the Taylor's series
e~pansion. In most structural engineering prqblems, the mean point is, in
fact, at some distance from the failure surface g( ) = 0 and thus there are
likely to be unacceptable errors in approximating the equation
M = g(Xt, X2, ... Xn) = 0
by the equation

M !~ g(X~. x;, . .. , x;) + i-tE (~xg) · (X;-

ox, x•
Xt)
 189

M=R-S
The failure surface equation for a set of the realization of values of R and
Sis
r-s=O (8.26)
The above Eq. (8.26) is shown in Fig. 8.4(a).
Let
Z2 = (S - P.s)
as
For a set of realization of R and S,
(r - P.R) (s - Its)
Zl = - Z2 = (8.27)

Hence, the safety margin Eq. (8.2~) be~omes ~

. f_l
ZJUR + /LR - Z2US - /LS = 0 6" fi- (8.28)
ZICTR - ZzCTs + /LR - /LS = 0
The above equation is represented in Fig. 8.4b. T..;.h:..:.i.:;.. :; s ~';.;;:n_.;.;~~..:;..;..;..~-- ....
s . . ',;;-
coordinate system since R and S have b'een no~d -"w.....,..._...,
etr cor ndin mean va e The mean values Zt anci Z z ar
zero and their variances are equal to one. In Pig. 8.4b, OD is drawn per-
pendicu ar to the at ure sur ace and it can be proved easily that OD = {3.
Proof: In Fig. 8.4b,
OB = (J.'R - /LS) ~
C1R V
zz
s

(al Original Coordinate Syste111 (bJNorlllaliud Coordinate Syste111

FIG. 8.4 Linear failure surface
190

fTR
ta n !._ O BC -~- -
as

Sill L OBC '"" (~~ + a~) ll~

on =--= OB sin LOBC

= (~~~~?;-~)Iii[ c~R au /lS )]

Hence it is proved that~ is the shortest distance to the linear failure surface
from the origin 0 in the normalized coordinate system. This is ltsed in the
deflnition of the reliability index defined hy Has0fer and Lind.

8.3. 1 Hasofer and Lind's Method (8.6)

Let the failure function g he a function of independent basic variahles
Xt , X2 ... . , X", i.e . g(X,, X2 , .. . . X"). The basic variables are then
normalized using the relationship
X, - ----
Z;= -·--- i';
I == I' 2, .. . , fl (R.29)
a;

where /Li ,-, /l.l'; and a; =ax; · In the::: coordinate system , the failure surface
is a function of z;. Using Eq. (8.29) in the failure function and equating it
to zero, the failure surface equation is written in the normalized coordi-
nate system , i.e. the :: coordinate system . This failure surface also divides
the design sample space into two regions. safe and failure. Because of the
normalizati on of the basic variables,
i'Z; ~· () and (8.30)

It is also to be noted that the ::: coordinate system has a rotational

symmetry with respect to the standard deviation and the origin 0 will
usually lie in the safe region . A two dimensional example is shown in
Fig. 8.5. It is to be noted that as the failure surface g(:::t, ::2) moves away
from the origin, the reliability, g(Z) > 0, increases and as it moves closer
to the origin, reliability decreases. Hence, the position of the failure surface
with respect to the origin in the normalized coordinate system determines
the measure of reliability .
Hasofer and Lind (8.6) defined the reliability index ~ as the shortest
distance from the origin 0 to the failure surface in the normalized coordi-
nate system . The point D (Fig. 8.5) is called the design point, and it is on
the failure surface. This point is also called the check point for the safety
of the structure. Now ~ is related to the failure surface (and not to the
 191

FIG. 8. 5 Formulation of safety analysis in

normalized coordinates

failure functions). The )U sure obtained is invariant to the failure

functio.n. since equiva)ent failure functions will - res-uJtth~- ·~~;;~--r;TI~~e
;·;·.~race. Th~ r~i~bility index, ~ = J.I.M;ro:~,d~firled'bY Cornell~ will~cide
with the value obtained by Hasofer and Lind when the failure function is
a linear function of basic variables. Hence in this method also (Hasofer and
Lind), the important relation,

(8.31)
can be used, provided the failure function is a linear function of the nor-
malfv.ilis.tributed basic variables.
from the above discussions, it is obvious that ~ defined by /.I.M/aM can
be obtained for a nonfinear unct10n by expa ilding tiiel'U nction abQ.t!l •.u<
design oi 1 This correspon s to approximating the nonlinear. failure
surface by its tangent plane at the design point D as shown in Fig . 8.5.
For a nonlinear failure surface, the shortest distan.ce of the origin (nor-
n1 alized coordinate system to the. fai ure surface is not unique as in the
C!lse of a linear failure surface,_ The computation of the probability of
failure involves numerical integration. For practical purposes, an approxi-
mation to the exact value is required. Shinozuka (8.7) ha s proved that the
point D. on the faflure surface with ' minimu~l •stance tot eon innorma-
fized coordinate system ~the m OS! p,rQba,.ble f@u ~poinl. The tangent
plane to the design point D may then be used to approximate the value of
{3. If the failure surface is concave towards the origin, the approximation
will be on the safer side, while for the surface convex towards the origin it
will be on the unsafe side.
The roblem therefore re~uces to finding out the minimum value of the
distance OD Fig. 8.5). -- ·- -
192

Let
(8.32)
he a nonlinear failure surface in the normalized coordinate system and
D '-"= z·
. ..
= (.:::,, Z2, . . . 'Zn)
. (X.33)
be the design point on the failure surface. That is
g,(z•) = 0
Tht: distance from a point z ,.---, (z,, z2, ... , Zn) on the failure surface to the
origin is

(8.34a)

(8.34h)

The problem is to minimizer subject to the constraint gt(z) = 0

Using the Lagrange multiplier method, the problem can be solved. The
Lagrange function L is
L = r + l.g!(z)
-= (z' z)''2 + !.g,(z) (8.3.'i)

For minimum

~L
oz;
= ~ + /. uz;
~g,
(z 1 z) 1 '
""" 0 i = ' 1, 2.... n (8.36)

~~' = gt(ZI, 22 , . . . , Zn) '-~ 0. (8.37)

There are n + l equations. In matrix notation, then equations [Eq. (8.36)1

can be written as
z
(z' z)''2 + /.G = 0 (8.38)

ag, ag, ag,)

where Gt = (-az, . -a-·· .... -~-
~2 rz,
(8.39)
... ...;,r ;1
The solutions for z~ and ,\' are obtained as c\''
( ,
, -:-."A (8.40)
r (• ~;: '

-'• = (G! G·)-112 (8.41)

Premultiplying both sides of Eq. (8.40) by G~ and using Eq. (8.41), it ts

obtained as
z~ G.
r---r---....-
- (G! G. )tt2 (8.42)

This r is the minimum distance and is equal to ~. G. is the gradient vector

194

the comparison of Eqs. (8.43) and (8.52) indicates that both are same. The
ratio, defined by Eq. (8.52), is also the distance from the tangent plane of the
failure surface at the design point D = z* to the origin in the normalized
variate coordinates.
The problem of finding the minimum value of {3 for a nonlinear failure
surface is solved iteratively. The problem can be solved in many ways. One
simple method is solving the following 11 equations [Eq. (8.46)]

i = 1, 2, ... , II

with (n + l)th equation

where !<./ --
_ (E " (ag-1 )2 ] 112 (8.53)
i=J az, *

and searching for directional cosines which minimize {3.

--T he fc,llowit g _s_teps arullYul.ved in the- method:.-

I. Write the limit state equation, g(x;, x1, ... , x,.) = 0, in terms of the
bi1sic va riablcs. . '/-' _ }J '1
2. N nnnlize the basic variables using Eq. (8.29). li -:. ~
J. Write the (limit state) failure surface equation in terms of norma-tfei
lized coordinatt> system. i.e.

Write expressions for agi/()z; i = I' 2, ... , II

At the design point z; = C1.Jf3. Using this, write gt(z) in terms of f3 and 11.1.
Write the equation such that
{3 = g({J, ex 1, 11.2, ••• , IX,.)

:·or computation purpose~ .

4. Select a value for {3 and values for
. f'. 2
:x1, IX2, ••. , ex,. satts ymg E :t.; = I .
While choosing values for 7.;, select positive values for load variables and
negative values for resistance variables.
5. Start the iteration. Calculate the new value of fJ using the equation
{3 =, g({J, (/.1, ·:X1 • . . ., ex,)
6. Calculate
/-- [II ( O~ J )2]1'
f<. . . .I=EI ~ * 2
_,

7. Determine new values of a.;

i=l,2, ... ,1l

 111

. With these new values of f3 and llt," start the next iteration. Go to

,. step 5
Jstop the procedure when the values of f3 obtained from two successive
'terations is within the acceptable error.
(Note: If the equation g,(z) is linear or quadratic, it may not be necessary
to start the procedure assuming a value of {3.)
The procedure is explained with examples.
/ExAMPLE 8.4 Determine the reliability index of a simply supported I beam
at the limit state of shear. The beam is subjected to a point load Qat mid·
span. Jt is given that
/lQ = 4000 N UQ = 1000 N

Ill. = 95 N/mm 2 ufs = 10 N/mm 2

ud = 2.5 mm -twd = 40 /ld = 50 mm

where d is the depth of the beam, lw is the thickness of the web, J. is the
shear strength of the material. The coefficient of variation of lw is negligible.
Solution
Maximum shear force = ~
It is assumed that the web resists the whole shear. The beam fails in
shear if
Q
.f.Twd- 2 ~ 0

Hence the failure surface equation is

g(X) = .fsfwd- ~ = 0
As variation in fw is negligible, fw is considered as deterministic.
Let
(/s --- Ns)
ZJ =
(d - - lld)
Z2 =
(Q - · !Jo)
ZJ = -·- -- -
11(1

Substituting them in the equation for g(X ), we gel

g,(:) '-''· lw(af,'::i -\ /LJ9)(adZ2 ~- /Ld) - · ~ C1QZ3 -- ~Q = 0

~- . 50[
~ '=' ~ , -::;:
40 °fsZJUd !2 f·
0 JoZI/Ld + fLfsad Z~ + 1-'!slld J"

1..\0
rTQZl -· fLO = O
2 2
196

Substituting the given data, we have

g1(z) = 625ZI + 296.88Z2 + 31.25ZIZ2 - 500Z3 + 3937.5 = 0
At the design point, Zt = f3a, [Eq. (8.45)].
g1(z) = 625{3al +
296.88{3a2 +
3 I .25{3 2(1.1(1.2
-500{3a3 +
3937.5 = 0

f3 -3937.5 (8.54)
= 625aJ + 296.88a2 + 31.25f3aJ't2 - 500aJ
Taking partial deviatives of g1(z),

( ~~: ) * = (625 + 31.25z2).

= 625 + 31.25f3a2 (8.55)

( ~;~ ) * = (296.88 + 31.25zl)*

= 296.88 + 31.25{3al (8.56)

( ~/!l ) = -500 (8.57)

OZJ *
Start with
f3 = 6 (1.1 = -0.58 (1.2 = -0.58 (1.) = +0.58
Using these in Eqs. (8.54) to (8.57), we have
--3937.5 .
f3 = 625(- 0.58) + 296.88( - 0. 58) + 31.25(6)(- 0.58)( --0.58)- 500(0.58)
= 3937.5/761.62 = 5.17
Using Eqs. (8.46) and (8.53)

a;= - ~erzJ*
(1.[ = - ~[625 + 31.25(5.17)(-0.58)] = -
53
~ 29

a2 -- - _!_l"'~9(i
K - ) Xo" , 31 .25(5 . 17)(-0 ·"''8] =
__j_ -
203 18
K·
I _ 500
IX.J -= - K-l- .:-.00] = K-
K2 = (- 531.29)2 + (-203.18) 2 + (500) 2
= 573551.17
K = 757.33
Hence

tXJ = - ~;~·~~ = -0.702

 197

203 .18
IX2 =- 757.33 = -0.263

500
IXJ = 757 .33 = +0 ·66
With these new values of {3, «1, «z and IXJ, the cycle is repeated till {3 con·
verges to the minimum. Summarized results are given in Table 8.1.

TAQLE 8.1 Computation of ~-Example 8.4

Iteration
Variable
Start 2 3

fJ 6 5.11 4.82 4.796

«t -0.58 -0.702 -0.738 -0.741
ac, -0.58 -0.263 -0.241 -0.234
«• +0.58 +0.660 +0.63 +0.629

The solution is: fJ = 4.796 PJ= 111- 1(-4.796) = 6x 10- 7

--
ott= -0.741 «1 = -0.234 ot3 =0.629
The design point is: z* = (/h 1, fJot2o fJac 3)

~AMPLE 8.5 For the same failure case, in _Example 8.4, determine the
mean depth of the beam for a reliability index of 5. The beam is subjected
to a point load Qat mid-span . It is given that
N! = 300 kN ao = 80 kN
P./1 = 95 N/nim 2
afs = 10 N/mm 2
d
ad= 5 mm -=40
fw

Coefficient of ·variation of lw is negligible.

Solution As the coefficient of variation of lw is negligible, it is considered
as deterministic. '
The failure surface is

h!wd- !£2 = 0 (8 .58)

Lel
'( J~ - fll s
ZJ =

Z2 =
d-
--
fld
ad

ZJ = Q -aQ ILQ
198

Substituting the above equations in Eq. (8.58), we get

tw [( OJ5Zl + /kf )( ad Z2 + ftd )]

5 -
OQZ3
- -
2
--
P,Q
2 = 0

Using the given data, we have

ttd[50zsz2 + lOttdZs + 475z2 + 95ttd}- 1600 x 103 zJ- 6000x 103 = 0
(8.59)
At the design point, using z; = rx;{3, the above equation becomes f?5
tt~(50rxs + 95) + iJ-d(l250rxsrx2 + 2375rx2)- 8000 , l0 3rx3 - 6000 X 103 = 0
(8.60)
Using Eq. (8.46),

cx1 = - ~ [~~(250rxz + 10ttd)] (8.61)

I

rx2 = -- ~[~~(250cxl + 475)] (8.62)

40 x 10 3
OtJ = -=---
K
(8.63)

Start with 1X1 = -0.58 'X2 = -0.58 OCJ = 0.58

Substituting the values of cx1, IX2 and OtJ in Eq. (8.60),
(ttd) 2(66) + /-Ld(2795.5) - 10640 X 10 3 = 0
Solving the above equation,
t-td = 380.9 mm
Using this value of ftd = 380.9 mm, new values of rx1, Otz and OCJ arti
obtained.
Using Eqs. (8.61) to (8.63),
1 [380.9
OCJ =- K 40(250X -0.58 + lOX 380.9) ]

= _ c4~90)

()(2 4()(250 X (0.58) +

= - K1 (380.9 475) ]

=- c~2)
40 x 10 3
OtJ = - - x -

using the relation IX~ + IX~ + IXi = I'

K = 51371
Hence oq = -0.656 IX2 = -0.059 lltJ = 0. 752

..
 199

Now the whole process is repeated till the maximum value of JLtl is obtained.
Summarized results are given in Table 8.2.

TABLE 8.Z Summarized results-Example 8.5

Iteration
Variable Start
2 3

-0.58 -0.656 -0.729 -0.733

-0.58 -0.059 -0.083 -0.082
+0.51! + 0.752 1-0.688 +0.685
ILd 380.9 420.0 422.8 423.0
(mm)

The solution is :
lld "' 423 mm

8.3.1 Non-normal Distributions

So far, the mean values and standard deviations of basic variables X; only
have been used in evaluating the reliability index. Probability distributions
of the variables X; have not been considered. If the safety margin equation
is linear and X; are normally distributed, the evaluated reliability index can
be connected to the true value of the probability of failure of the structure
(Eq. 8.15). as M is normally distributed. However, in practical situations,
n1any of the basic variables are non-normal, · e.g. wind speed, live load,
strength of low strength com:rctes, etc. In such cas.es, the value of f3 (or PJ)
L'an be obtained using cg uivalent norma l distributions (8.8, 8.9) at the
d~point. The ti·an sformalion of a non-normal variable to a normal
variable at the design point is done as follows:
At th e failure point (i.e .. the design point D)xi.
(i) the probability density ordinate of the original non-normal variable
X; is made equal to the probability density ordinate of the equivalent
normal \ariable x:. That is
(8.64)
I
(ii) the cumulative probability of the original non-normal variable X; is
made equal to the cumulative probability of the equivalent normal variable
X1. That is
(!t65)

If ft\-; and a'r; are the unknnwn mean and standard deviation of XI, then
Lq . (~.651 becomes

(8.66)
. 200

The above equation leads to

,.,.}1 = - o~ 1 «'-I[Fx1(x;>] + x~ (8.67)

Considerina Eq. (8.64),

fx, ~:)=-~ J xt -.J4~] (8.68)

ux, 1 ux,
since XI is a normal variable.
Substituting Eq. (8.67) in Eq. {8.68), we get
~{<J>- 1 [Fx1 (x;)]}
o_K1 = {.tj(X;) (8.69)

Since Fx1 and/x1 are given or known, the values of ,.,.} 1 and 1 of equiva- ox
lent non-,normal- can. be obtained· using Eqs. (8.67) llfte (8.6Q), The pro-
cedure of determining fl for the failure surface having non-normal basic
variables involves the following steps:
(i) Write the limit state equation in terms of the basic variables, i.e.
g(Xt, X:z, .. ., X,)= 0
(ii) Normalize t~e basic variables using Eq. (8.29).
X; - ,.,. x,
For normal variable X1, z,= - -ax,
-

For non-~ormal variable XJ,

where ,.,.x1 and ax1 are the unknown mean and standard deviation of equi-
valent normal X) of non-normal X1 at the failure point.
{iii) Write the limit state equation in terms of the normalized variables
and unknown values of
,.,..rj and ox)
(iv) Select values for fl and ott, «:z, .•. , otn as explained in the pre.vious
section and values for ,.,.}1 and o'xr
(v) Start the iteration. Calculate new values of {l, otJ, oc.2, .•• , otn as ·
explained in the previous section.
(vi) For non-normal variable~ (say XJ), the design point is
XJ• = I
,.,.x + Q I
a.J,.ax
r
1 1

(vii) At this design point xj, find new values of P.'x1 and a'x1 using
Eqs. (8.67) and (8.69). I

Go to step S and repeat the procedure till {l converges to the minimum.

The procedure is illustrated with the following examples.
·',
EXAMPLE 8.6 A cantilever steel beam (ISLB 450)
.I
of span I is subjected to
 ·201

a load P at the free end. The resisting moment capacity of a section is taken
as FyZ, where Fy is the yield stress and Z is the section modulus. 'Hence at
the limit state of collapse in flexure, the safety margin can be written as
M = FyZ- PI
Given:
For Fy, J.Lt = 0.32 kN/mm2 ot == 0.032 kN/mm 2
z' fl2 = 1400x 10 mmJ 3
oz = 70 X 10 3 mm 3
P, fk3 = 100 kN OJ= 40 kN
Fy and Z are normally distributed and Pis lognormally distributed. Calculate
~if I= 2m.
Solution Denote
Xt = Fy
J.LI =fiX;

Then the failure surface equation is rewritten as

g(X) = X1X2 - · 2000XJ = 0
I
Let p,) and cr3 be the meaJl value and standard deviation of the equivalent
normal X! of the non-normal XJ at the design point. Normalizing the
variables
I Zt=--
x,- fit X2 - 11-2
Z2 = ---
UJ 02

XJ - -- P-3
ZJ =
J. O'
J

Substituting these in the failure surface equation and using the given data;
gt(z) = Ot02ZtZ2 + 0Jjh2Zl + U2P,tZ2 + /h1/h2
-- 2000( o5z3 + p,J)
= 2240ztzz + 44800zt + 22400zz - 2000aSzJ
+
448000 - 2000p,i (8. 70)
At the design point, z; = IX;{3 and gt(z) = 0. Using these in Eq. (8.70),
f3 = . 2000p,) - 448000 (8.71)
2240~ t 1X2~ + 44800rx t + 22400G<2 - 2000a51X3
Taking partial derivatives of gt(z),

1Xt = - l (~~~.) ~ - l
(_
(22401XzfJ + 44800) (8.72)

IX2 = - .!.( Ogt ) = :_ j_(22401Xtf3

K 8z2 K
--j- 22400) (8. 73)

IX) = - J..( ag. ) = - J..<-

K 8z3 K
2oooa5) (8.74)
202

As XJ is lognormal, the parameters of XJ arc lir~t calculatecl.

u1n .\'J --' lin (oL > I)J 112 o 0.385

XJ - /1_,, cxp ( --kafn 1 ,) c_ 92.85 kN

U~ing Lq. (8.69) and CDFof XJ lEq. (3.93)].

~ q,un (x;/ XJ)/aln .nJ

-
/ r.,(x;)
llul the PD 1- of lognormal XJ is given by

.t:r,(x~>. c~ - .-
1
.Tl <T Jn.r,
- -- rJ,tiJ~:y;;,rJ)/aJ" .,,J (8.75)

Uslll!! Eq. (8 75) in the above equation for (/r,, we have

'
ax, = .
XJ<TJn Xa (8.76)
The mean value i
'., is calculated using
-===-Eq. (8.76) in Eq. (8.67).
J.i~, =-~; O'lnx~<l>-- 1 f<l>{ln(x;ix3)/0'Jnx, }J + x;
c·• x;( I - In x; + In XJ) (8.77)

Assume
f3 -.. 5'
IXJ ·--~ · 0.5 IX2 = -- 0.5 OtJ =co= + 0. 707
tJ.; -· • /J,J
~
== I 00
. _u; = a3 = 40 kN
~

and start the procedure. UsingEq. (8.71),

(2000)(100) -· 448000
/J --• (2240)(0.5)(5) -- (44800){0.5) 22400{0.5) - -2000(40)(0. 70)
=· 2.839
Using Eqs. (8.72), (8.73) and (8.74),

IXJ '"" -- KI [2240( --0.5)(2.839) + 44800]

r

41620
,:
= --
K

<X2 = - ~ [2240(- 0.5)(2, 839) + 22400)

19220
K
 203

'XJ .c -- k[- 2000(40)]

8000
=---x
Using iXr + IX~ + :Xi = 1, K = 92200. Hence
IX( = - 0.451 iX2 = . . :. . 0.209 IXJ :-...c 0. 868

The design point, xi= p.\·, ·: Ct.Jf3a\·,

== 100 -1· 0.86R(2.839)(401

= 198.5 /
As X.1 follows the lognormal distribution, using Eqs. !X.76) and (8.77). new
values of 1-'J and a.l are calculated:
1 = (198.5)(0.385) '--' 76.42
!tJ --~ ( 198.5)(1 -- In 198.5 + In 92.85 I
= 47.75
Carry out the second iteration with the new vnlues or {3, :XJ, cx2, :XJ, a) and
p.). The whole process is repeated till the convergence is achieved. The
results of each iteration are given in Table 8.3.

TABLE 8.3 Cumpututimr 4 8-J::xanrp/e 8.6

lleration
Variable Start
2 3 4

fl 5 2.839 2.247 2.192 2.192

«t -0.500 -0.451 -0.273 -0.260 -0.264
otz -0.500 -0.209 -0.126 -0.124 -0.126
«a + 0.707 +0.868 +0.954 +0.958 J-0.956
"'3
• 198.5 211.5 208.4 208.2
.}
"j 40 76.42 81 .40 80.21 80.12

l'j 100 47.75 37.51 40.04 40.23

Result.> are:
fl = 2.192 Pf = tfl- 1 (-2.192) =• 0.0142
Design point: (;: 1 Zi, z)) = {J ( -0.264, - 0.116, 0.956)

~ = ( -0~579, -0.276, 2.095)

(xj, .ti, .\:)) = (0.3149, 1381000, 208.2)
(Note: xJ = 40.23 + (0.956)(40.23)(2.192) = (208.2)

The same problem has been solved for various values of the coefficient of
variation of P, and corresponding .values of f3 have been computed. The
204

variation of f3 with Sp is shown in Fig. !l.6. As expecteu, {:3 decreases (i.e.

reliability decreases) as Dp increases.

rJ
2
-
0 01 02 &p 0 J 04 OS

FIG. 8.6 Variation of P with 3p --- Example 8.6

ExAMPLE 8.7 An RSJ section is used as a column. The height H of the

column above ground level is 10m. Jt is subjected to a wind load W which
follows the Type 2 extremal largest distribution. The allowable deflection
at the top of the column is H/250.
Given:
For Young's Modulus(£): (Normal)
tt• = 2.041 x 102 kN/mm 2 CTJ = 0.156x 102 kN/mm 2 (S = 7.62%)
For Moment of Inertia (!): (Normal)
J.l-2 = 315 ;~ 106 mm 4 CT2. = 15.75 :;.; 10 6 mm 4 (8 ,_, 5% !
For wind load (W): lType 2 extremal (largest)]
J.I-J = 6 kN CTJ = 1.38 kN (8 = 23%)
Parameters: u = 5.358 k = 6.42
Compute the reliability of the column at the limit state ut' dellection.
Solution For a uniformly distributed wind load,
WH 3
Maximum deflection = SE!

The failure surface equation is

H
250
His considered as a deterministic variable. Substituting the given value of
H, the above equation becomes
40- (IOQOO)l W = 0
8El
El- 0.3125x 10 10 W = o
Let
Xt =E X2 = I and XJ = W
 205

Hence the failure surface equation is

x.x2 - (0.3125 x I0 10) XJ = o (8.78)
a;
Let ,_,; and be the mean and standard deviation of equivalent normal X~
at the design point. After normalizing the variables, the above equation
becomes
g1(z) = UJC12ZIZ2 -j- C1JfL2Zl -j- U2fliZ2 -j- /lJfL2 -0.3125 X 10 10(aSzJ -j- 1-LS)
Using the given data,
g1(z) = QZJZ2 -j- bz1 -j- CZ2 -j- 6429 X 107 - da;z3 X d/-'i
where a= 245.7 X 106 h = 4914 X 106 c = 3215X 106
d= 3125 X J06
At the design point, z; = «;{3 and g,(~) =0
Using these, it can be written that
f3 = dp.3 - 6429 X 107
aa.,cx2f3 + b,,
+ c1X2 - das!XJ (8.79)

The directional cosines are given by

I
IXJ ,..= -- K (a1X2{3 + h) (8.80)

I .
IXl =- K (a1XJ{3 + c) (8.81)

I>:) = --- KI (- dG§) (8.82)

where K=[E (ag,azl )2]'• 12

I-I

Start with
~ = 5 = - 0.5
IXJ C1.2,...., - 0.5 IX) = 0.707
,.,; = /L3 =6 (1; = C!) = 1.38
Substituting these values in Eq. (8.79), we get
f3 ='-" 6,691
Using Eqs. (8.80), (8.81) and (8.82),
1
lXI ":"- K (4092 >< J06)

1
~>:2 = - K (2393X J06)

IXJ = }c (4313 x 106)

Using oci + oci + oci = I, K = 6408 :< I 06 • Hence

CXi ='= - 0.6385 «2 = - 0.3734 «3 = 0,6730
 208

The design point x; is given by

x; = JL; + 1X3~a~ = I 2.21
XJ follows the T c 2 extremal largest distribution. The PDF and CDF of
X3 are [Eqs. (3 ."121) and 3.122)1

Fxix;) = exp [- (;; rJ

/Yix;) = : [:; r+J exp [ -- c~ rJ
'·
Using x; = 12.21 u= 5.358, and k = 6.42, we get
fx,(x;) = 0 002637
I f
Using Eqs. (8.67) and (8.69), the new values of a3 and JLJ are calculated:
I - 4>ffl>-1(0,995)} 5.507
a3 - 0.002637
JL~ = x; - ci>- 1 [Fx3 (x;)]a~
12.21 -- f/>- 1(0.995)(5 '\[)7)
- - I .961
Using these new values of ~. a,,
1)(2, p.; and a;,
successive iterations are
carried out. Results are summarized in Table 8.4 .

EXAMPLE 8.8 An under-reinforced concrete bt>am of breadth (b) 240 mm

and effective depth (d) 480 mm is reinforced with steel bars (grade Fe 250)
of area (A,) 1400 mm 2• The grade of concrete used is M 15 (nominal mix).
The beam is subjected to a moment M. Given:
Variable Fy: Normal
~ = 320 N/mm2 a = 32 N/mm2
 207

1~ = 22 .67 N/mm ~

Parameters: .f~u -~ 22.04 UJn/cu = 0.237

·1- Type I extremal (largest):' ,.
ft , . 72x 106 N mm a= 24 :-:106 N mrn
P:rrarneters: II -,- 61.2 >< JOS ot ~, 0.534 ~< I0-' 8
Determine the reliahility of the beamat the limit state of collapse in fle)!:ure.
Solution The ultimate strength of thb beam is given by

R ,,_, AsFy d [ 1 - O.~L.~y~s ]

The failure surface equation is given by

g() = R - - !If

-~
0
Ad"y d [ I - ·~:u ~Y~•]- M (8.83)
l.et
F.v
Using thr given value~ of As, b and d in Eq. (8.83), the fa,ilu.re surface
equation becomes
g(X)={IJXrX2 - · mXf - mX1X1 =0 (8.84)
where at ~· As d = 672000
0 7
a2 = (m)( ·; / ·') = 6288 o3 = 1
1 I I I
Lelt.t2 and a2, and l-'3 and a3 be the va lues of the mean and standard devi·
ation of the equivalent normals X~ and Xi at the design point.
Normalizing the. variables Xt, Eq. (8.84) becomes
, I
.!,'J(Z) """ 0J(OJU2ZtZ2 + 0t!-'2Zt + 02/'!Zl) -
I I (
02 OtZI
2 2
+ 2aiZIP.t)
-- m( a~a;Z2Z3 + tJ~P.;Z2 + u;JAolZ3)
2
+ 01!-'JJA-2 - I
021-'1 -
I
03!-'2f-l3
I

At the design point, z; = ot;{J .and gt(z) = 0.

Using these, the above equation can be rewritten as
{ ' 2 • •)
fJ = - \atJ.lrJ.lz - a2J11 - 03J.l2J.l3 (8.85)
h, -b2 -h3
where
I
br = a1(ara2~Jrt.2fJ +· ar!-'2~1 + a2}AoJ«1)
I I

b2- = 02(oioci,B + 2uJOtJP.t)

208

The directional cosines are given by

«1 = - ]: [aJ(uJu~iil/3 + au-';) - 2a2(u~«1/3 + O'JJ.tJ)] (8.86)

ll2 = - ~ [OJ(O'JO'~otJ/3 + O'~J-'J) - 0)(0'~0'~(/.3/3 + O'~J-';)J (8.87)

where K = [ 1:
1-1
(agazl )2]• 112
1

Start with
fJ=5 IXJ =- 0.5 ll2 = -0.5 IX) = 0,707
1-'l = 1-'2 = 22.67 u2 = u2 = 5.44 ·
1-'~ = /-'3 = 72X 106 u) = O'J = 24 X 106
Substituting these values in Eq. (8.85), we have
f3 = 4.449
Using Eqs. (8.86), (8.87) and (8.88), we get
1
IXJ = - K(l27.1 X J06)

I
IX2 =o- K(I07,3X J06)

I
«3 = - K(- 253.6X J06)

Using «:+ cxi + 1Xi :.-:: I, K = 303.3 X I 06

the directional cosines are
IXJ = -·0.4)9} ill = -0.3537 IX) = 0.8362
The design point x2 and x; are given by
x2 = p.2 + rx2fJa~
= 22.67 + (-0.3537)(4.449)(5.44) = 14.11

= 72 X )06 + 0.8362(4,449)(24 X 106) = 161.3 X 106

It is given that £2 follows the lognormal distribution. Using Eqs. (8,76) lind
(8. 77), new values of p.~ and u2 are calculated:

o2 = 3.336 J-'2 = 20.4

 209

X3 follows Type 1 extremal (largest) distribution. The PDF and CDF of X3

tro [Eqs. (3.114) and (3.115)]
fx.(x;) = ot exp [- oc(x; - u) - exp {- oc(x; - u)}] . (8.89)

Fx.(x;) = exp [- exp {- ot(x; - u)}] (8.90)

Using the above equations, at the design point X3y

•
F4x;) = 0.9953 fx.(x;) = 0.253 X I0-9
Using Eqs. (8.67) and (8.69), the new values aj and p~ are

a) = 1/l[~(~x,~;))]
Xa X3
= 54.50 X 106

p~ = x; - tZ>-1[Fxix;)]ax,
= 161.3 X 106 - <1H(0.9953)(54.50 X 106)
= 19.89 X 106

Using these new values of {J, IXJ, oc2, IXJ, p2, a2, p] and a3, the whole process
is repeated and successive iterations carried out till the required convergence
is achieved. Results are summarized in Table 8.5.

TABLE8.S Computation of reliability-Example 8.8

Iteration
Variable Start
2 3 4

fJ 5.000 4.449 3.310 2.932 2.904

0(1 - 0.500 -0.419 -0.261 -0.241 -0.259
(t1 -0.500 -0.358 -0.0518 -0.084 -0.104

.
oc a
.\'2
+0.707 +0.836
14. 11
+0.964
19.83
+0.967
20.77
+0.960
20,52
a'2 5.44 3.34 4.69 4.917 4.858
1.1.~ 22.67 20.40 21.93 22.00 21.99
•
.\'3 161.3 x to• 193.7 x 10' 174.1 x to• 172.2x 1QG
a~ 24 x to• 54.SO x 10' 63.88xiO' S8.34 x to• 57.78 X 10'
1.1.) nx to• 19.89 x 10' -7.0x 108 -9.S x 10' -11.70x 101

Result : Reliability index fJ = 2.904

Design point : {J( -0.259, -0. 104, 0.96)

8.3.3 Determination of {J for Present Designs

During the process of code calibration, the reliability analysis of structural
components designed as per the present code are first carried out, and then
the reliability levels of the present designs under different design situations are
 210

established. Different design situations may mean different load combina-

tions, viz. D + LorD + WorD + L + W. Again, different ratios of
load under each load combination may also be considered. The process of
establishing the reli ability level is explained in the following section.
Consider the case: the reliability study of RCC beams at the limit state of
collapse in flexure. Let R be the ultimate strength of a RCC section in flexure.
This is the true or actual strength based on the theoretical model to compute
the ultimate strength. After identifying the basic variables in the theoretical
model, the statistics of R are established using the Monte Carlo method,
(explained in Chapter 7) taking into account the model error also. Let R,
be the nominal strength of the member.
When the nominal values of variables are used in the theoretical model,
the strength obtained is called the nominal strength. When strength is calcu-
l:ucd substituting nominal va lues of variables in the equation given by the
ct\tle it is called the design strength Rn. Let YR be the strength reduction
far tor given by Rv/ R". If the loCid combinatir-11 D -:- L is considered, the code
(8.3) specifies that member is tfl be designed for YDD"-+ YLLn, where Yv and
YL are the partial snf~:t~ fa ct r. f r d~ad load and li,:c J nd rc pee-
lively, and D, and L. are tlllminnl val ues of·D and I. tc ~ pcctively. A s per
IS Code I .3), 'Yo = 'Yt - 1.5. Hence the de ign streng th of the member is
/?ll - 1.5(Dn -f- [ 11) (8 91)

Since Rn

R,. _
D,
u( I +L")
D"
(8.92)
Yn

Let the failll'c surface equation under the D +L load cme·be

R -- D -- L ~, 0
Dividing by D", the above equation can be written as

(.B..)(R")
R. D,
_ (.!2)(D")
nil [),. -· (J:..)(L"
Ln D
)-- o II -

Using Eq. (8.92),

( _B_
Rn
)[ 1.5( I + LiD,)]
'Yn
_ (.!!..)
D,
__ (!::.Ln )(L"),.,...
D11
O (8.93)

For a kn9wn value of 'YR, the reliability analysis c.:u n be carried out and ~
computed fN Yarious ratios L./D". When the reliability is estimated nt the
limit state of collapse (ultimate limit states), the statistics of the lifetime
maximum live load is to he used. This i~ illu.'tr:tted in the following example.
~,ExAMPLE 8.9 From the statistical study of the flexural strength of doubly
· reinforced sections, with M 20 grade of concrete (nominal mix) and Fe 415
grade of steel, it has been found that the mean value and standard devia-
tion of R/Rn are 1.222 and 0.149 respective!:;. 'YR = 0:844.lf L 11 /D,. = 0.5,
-----
----------------------------------
 211

determine the reliability of the beam under the load combination D. + L.

.f
Olven:
For
.f

f
~~,) :'p. =: 1.05 u = Q.IOS

e / (L) :P- = 0.62 (] = 0.1755

.I
R and D are normal and L is Type 1 extremal (largest).
II Let

X1 = R X2 = J2. XJ = .!:..
Rn Dn Ln
e
al =1.5(1+ Ln /Dn) aJ =Ln
YR Dn
For "'R = 0.844 and Ln/Dn = 0.5, at = 2.666. The safety margin given by
I
Eq. (8.93) becomes
(8.94)
In the present case, at = 2.666 and OJ = 0.5.
Let p.) and u) be the mean 'and standard deviation of equivalent normal
Xi at the design point. The failure surface equation, being a.linear equation,
#LM = OJ#LI - fl-2 - {/Jf'~

Ullf = [aTuf + u~ t (aJu.l)2]1/2

R fLM _ Ql/1. 1 - 1''1.- (/JJ.Li

(8.95)
1-' =
uM
- 2 2
[a1 u 1 + + (a3a))2i'/2
'1.
u2

The above equation can be verified by normalizing the variables and follow-
ing the usual procedure in the previous examples. Start the procedure assum-
ing values for the unknown a~ and fL).
Start with rr) = 113 = 0.1755
J.I.J = f1.J = 0.62
Using Eq. (8.95),
~ 2.66(i( 1. 222) - ( 1.05).-L 0.5(0. 62)
= l2.Cl66 2(0.1 49)2 + 0.1052 0.5 2(0.1755)2)1/2
1. 898 -
= 0.4201 = 4.517

(0.5)(0. 1755) = 0 2089

0 .420 1 .
212

The design point x; is

X3
• = /LJ, -f- a '
IXJ/-'173

= 0.62 + (0.2089)(4.517)(0.1755)
= 0.7856
XJ follows the Type I extremal largest distribution.
Following the procedure in Example 8.8,
Fx 3(xj) = 0.8458 .fxbn = 1.035
Using Eqs. (8.67) and (8.69), new values of aj and ~~ are

a:l = 0.2466 fLJ = 0.5327

The whole process is repeated till {3 converges. Summarized results are given
in Table 8.6.

TABLE 8.6 Computation of reliability-Example 8.9

Iteration
Variable Start
2 3 4

4.517 4.529 4.526 4.525

0.1755 0.229 0.247 0.255 0.259
0.62 0.552 0.533 0.522 0.517

Result : Reliability index {3 = 4.525 ·

Probability of failure p1 = tll(-4.525)
c.~ 2.465 X 10-e

The same problem has been solved for various values of L,jD, and the
variation of {3 with Ln/D, is shown in Fig. 8.7. It may be observed that the
values of {3 range from 4.33 to 4.66, which are high. Normally, for compo-
6 ----- -----
Do'ubl y reinforct'd beam

3o - - ois- ·--0-.1.5-0--0--'7L....5---,,-':.oo:-:::---,..J2~5=----:-,'='5~0--'
Ln/Dn
FIG. 8. 7 Vmiation of fJ for doubly reinforced beam with L,!D,.
-Ex~<np!e 8 ,9
 213

:fitnt failure, a value of~ about 3(8. tO) isconsidered for code calibration.
The high value obtained is due to the fact that for office buildings the value
Jtf nominal live load specifi ed by IS Code (8.11) is quite high .
· The fo rmulation for determining f3 for the load combination D + W is
••me as for the L +D case. Lis to be replaced with Win all Equations
l'rom (8.91) to (8.93). The failure surface equation i.s
R-D - W=O

or bt(:J- _Q_ - bJ (
Dn
w) = 0
Wn

where b3 = w"
D"
bt = 1.5 ( 1+ ~~~JD,) (8.96)

The value 1.5 in the above equation is as per the present IS Code {8.3). If
we consider the shear strength of the beam (limit state of collapse in shear),
it has been found that the statistics of R/R" for a RCC beam (with M 15
design mix and Fe 250 steel grade) are I" "'~ 1.355, a = 0.225, and 'YR =
0.85 (8.12). The normal distribution has been fitted to the tail region. Usiqg
the statistics of R/R", DfD, and Wm/ W,, given in Table 5.3 , the reliability
analysis can be carried out for various ratios of Wn/Dn. The variation of f3
5 with W"/D, for the case of a beam in shear under the load combination
D + W case is shown iti Fig. 8.8 (Ref. 8.12). Values of fJ vary from 3 to
7 3 .5. Instead of the steel grade Fe 250, if Fe 415 is used, the statistics
of R/R, for the same case change to I"' = 1.277 and a = 0.2105. The results
of the reliability analysis for beams in shear using Fe 415 grade is also
shown in Fig. 8.8 . It can be observed that beams with stirrups of Fe 250
grade have higher reliability than those with stirrups of Fe 415 grade. It is
mainly becau~e the ratio of the mean value of the yield strength to its
specified strength for the Fe 250 grade steel is much higher than the ratio
for the Fe 415 grade steel (Table 4.3).

~ r-----------------------------------~
1. t.41S(design) +Fe- 250
2. t.41S(design) +Fe 415
4

.P
.I
2

0·5 1·0 1·5 2·0 2·5

Wn/Dn

FIG. 8.8 Variation of fJ with W) D, for RCC beam in shear

under load: D + Wm
214

The reliability analysis of members under load combination D +L+ W

can be formulated as follows. The failure surface equation is
R-D-L-W=O (8.97)
Dividing each term by D,, the equation can be rewritten as

(:,,)(~:) - (g)(~::) - (L)(t:)- (:;,,)(~::) = o (8.98)

As per the IS code (8.3),

RD = 1.2(D, + L, + Wn)
Dividing by [),, and using the relation R, = RD/'YR,

R, ="' !. 2 ( I + L,/D, + W,/Dn ) (8.99)

{), "/R

Hence the failure surface equation becomes,

(8.100)

where C1 = ! .2 (I + L,JD, + I
}' ,, ,/DI')

.. w,
C4 = -
D,
Now for different combinations ol' L,/D, and JV,/ D, ratios, the reliability
analysis can be carried out.
EXAMPLE 8.10 It is desired to determine the reliability of a column under
a load combination of gravity plus wind loads, viz. D + L + W. From the
statistical study of redangular RCC columns subjected to axial load and
uniaxial bending, il has been found that (8. I 2) for compression failure,
'YR = 0.725

!J.R / RII =-= ).22 "R / Rn = 0.171 (S = 14% J

1\:lean and standard deviation ol' D/D, and L/L, are
/)
For
0, I•· = I 05 rJ = 0.105 (8 = 0.1)

rt>r
"
I"
I' = U.Cl2 u = 0. 1755 (b = 0.28)

w
hll'
w,. I' = ll. 804 I) = 0.26!) (D = 0.334)

I? :1•1d fJ lt~lluw rwrm:tl. I :1r1d W l'o ll r1W Type I extremal (l.trgest). Deter-
rninc· {j il L,/U, = 0.5 <111d II',.'[),, = 1.0.
 216

. olulfon ·The failure ·surface is given by Bq. (8.100) Let '

7) Xr ..;.,Ji. X2 = ·D,
I)
Rn

. XJ -:c. Ln
L
X4 -w
Wn
Ih r ilure surface Eq. (8.100)- becomes
M = CrXr ~ X2- C3X3 -C4¥4 = 0 (8.101)

here ( ,I -_I ' 2 0 + Ln/Dn + Wn/Dn)

)IR

Ln
') C3 = -
D,
/..,
I or D
/1
= 0..s, w,
D,
= 1.0

' .'
1)
c, = 4.138, qJ -~- Q.~. C4 = 1.0
It i given that the live load and Wiri\i load follow the Type 1 extremal
II tribu tion. Let /.l.J and aj be.Ahe tnean value and standard deviation of the
c uivalenl 11:ormal XS -nt -the desi-gtr poii1t. Simihtrly, 1-'1 ahd b~ are for Xt
'he fail~re. surface eq\tati<>n bejpg litre~r
1w = c,·;,·- cw.!- C4P.4
P.2---:-
'I
a,,t =- L(Cw;) 2 +a~ + (CJu~) 2 + (C4u4)2J'' 2

{3 = ILM
O'M

The proce~ure ofcGmput~t.ioi' i~ s.imil~u to the one explained in the previous

~~mple. Summarized results. al'e given in Table ~. 7.
The variation of {3 with ·w,{D, and L,JD, for RCC columns under; the
load combination D +
Lm + Wm is shown in Fig. 8.9.

8.3.4 Correlated Variables

It; all the previous problems, we have assumed that all variables in tlie failure
surface equation are statistically independent, and have computed reliability
index based on .this assumption. In practice, we may have to deal with cases
whe\1 all variables or some of the variables are correlated. The procedure of
solving such· cases is explained below.
216

TABLE 8.7 Computation of reliability-Example 8.10

ltera tion
Variable Start
2 3 4 5

fJ 3.755 3.728 3.698 3.&89 3.687

~3 0.1755 0.1955 0.185 0.182 0.181 O.lSl
!J.j 0.6200 0.5795 0.585 0.586 0.587 0.587
(J4 0.268 0.387 0.455 0.495 0.516 0.527

1'4 0.804 0.657 0.555 0.481 0.438 0.416

Result : Reliability index~ = 3.687

C: Comprfssion fait ur ~ 1:ln/Dn::05

T: Tension failure 2:Ln/On.,.1·0
J: l n I 0 n =1· 5
Duign mix
4·0 ,_--e=:::::=:=--o---

CJ
C2
T3
._,._ .... ,C1
J·O .
-- - --.T1
..,.T 2

2· 5~-----:;:-:-----::-';;-----:-'-::------:-l~
0 0-5 , 0 1·5 2·0
Wnf Dn

FIG. 8.9 Variation of fJ with W,ID, and L,ID, for RCC columns under
load: D , L,, W,,

Let Xt, X2, ... , X, be the set nf correlated variables appearing in the
failure surface equation . let [CxJ he the covariance matrix of the correlated
variables. That is
.- Var (XtJ Cov (X1, X,)

Cov <.r2,XJ) Var (Xz) Cov (X2, Xn)

!Cx] =

Cov (X11, X1) Cov (X11, X2) Var (X11l

(8.1 02)
l.d ,\1, ,\~, . . . , A, be lh<! eigen values and [V] be the matrix having each

__ .-;._
 217

c lumn represented by an eigen vector corresponding to each eigen value.

"l
Th nl is, if C;j is the eigen vector for Ah then the elements of matrix [ V] are
e11 e12 .. ) Clj

Cll t'22 C!2j £12u

7 lVI= (8.103)
I
:7 (',1) Cn2 e,/j t'mr J
:7 This matrix [ V] is an orthogonal transformation matrix. Then the required
6 ~t . of uncorrclatcd variables Y is given by (8.2)
Y = LV]'X '(i!.l04)

where Y = {Yt, Yz, ... , Y,} 1 and X = {Xt, Xz, . . , X,}'. The . superscript
t denotes the transpose. Since [V] is orthogonal, l V]- 1 = I V.l'. Hence,
X , ,~ I.V.IY (X.I05)

Tne expectt·d values or Y arc given by [Eq. (~ . I 04)1

E(Y) = IV]'E(X) (8.106)

The variance matrix of Y, [Cy"l, is given by
[Cr] = [V]'[Cx]LV]

r At 0 0 0 l
0,
=[A}= (8.107)

. I
0 0 A, J
That is, the eigen values of tv] arc also the variances of respective variables
Yt, Yz, . . . , Y,. Knowing the mean values and standard deviations of Y,
the variates Y; can be nonnalized as usual, i.e.

c (8.108)
d
and fJ can be determined following the procedure given in the previous
sections. Hence the following steps are involved for correlaLed variables:
(i) Determine the eigen values and the corresponding eigenvectors of the
covariance matrix. That is, determine the matrix [V).
(ii) Write the safety margin equation in terms of the uncorrelated
variables using Eq. (8.1 OS).
(iii) Determine the mean values of Y using Eq. (8.106) and the variances
~) ofY using Eq. (8.107).
(iv) Normalize the variables Y; using Eq. (8.108) and write the safety
h
218

margin equation in terms of the normalized variables Z;. Note that using
Eqs. (8.105) and (8.1 08), we have

X = [V][ far]Z + !J.y]

= [V][ ay]Z -1- [ V]!J.y (8. 109)
where
r IJy, 0 0 ... 0 l
I

[o1•l =
l) ay, 0 0
II I
(8.110)

L o .. . .. . ... ... ay" J

!J.y = (/1-Yu fLy,, • ., fL~·~~)I

Z = (Z,, Z2, . . ., Z")l

(v) Determine {3.
With the orthogonal transformation of Eq. (8 .104), it can be shown (8.7)
that the reliability index of Eq. (8.44) becomes
-- z~G•
{3 = (G '·[C]G:)I/2 (8.111)

EXAMPLE 8.11 For the same problem in Example 8.9, assume that R and
D are correlated. The correlation arises because both depend on the
dimensions of the beam. Assuming Cov (R, D) = 0.0111, determine the
reliability index.
Solution The failure surface equation is [Eq. (8.94)J
{/JX1 -- X2 - (/)X3 = 0
where
C/1 ~~ 2.666 OJ = 0.5
The covariance matrix is ll-- [I L

[CxJ =o
1oom -- ~-01 J I
0.01 I I

0.01 tO
0

0
/

. 0 o..-~ 0.03bB J
The corresponding determinant equation is
(0.0222 -- ,\) 0.0111 0 J-,
Det 0.0111 (0.0 l I - ,\) 0 = 0
[
0 0 (0.0308 - ..\)
l'he characteristic equation of [Cx] is
(0.0222 - A)L(O.OI I - A)(0.030g- A)] - 0.0111[0.0111(0.0308 - A)] =0
,\ 3 - 0.064,\ 2 + 11.436 X 10- 4,\ - 3.725 X r'o- 6 = 0
Thil'eigen values, given by the roots of the equation are
At = 0.02903 A2 = 0.004167 t\3 = 0.0308
The corresponding normalized eigcn vectors are
0.8516 J
et = _ 0.5~42
[

Hence, the orthogonal t~ansfQrmation matrix is

~J
0.5242

~(
0.8516

lVJ 0.5242 0.8516

0 0
Hence, using Eq. (8.104),
Y = [VJ' X

/A- Ya
l[
The expected values of variates y/arc lEq. (8.106)]

II ~Y, =
0.8516
·-0.5242
0.5242
0.8516
0
1.222
1.050
0.620
J
- f'Ya 0 L..

This yields
fi.Yt = 1.591 , /kY~ = 0.2536 1-'Ya = 0.62
The variances of Y, given by eigenvalues, are, \..
2
Uy 1 = 0.02903 uy 1
2
= 0.004161 u}. = 0.0308 X.'\...~
that is ~ ~
Uy 1 = 0.1704 Q'y~ = 0.06456 uya = 0.17
Using Eq. (8.109), it is obtained as
Xt = 0.1451Zt - 0.0338Z2 + 1.222
..--- (8.112)
X2 = 0.0893Zt + O.OSSZ2 + 1.050 (8.113)
_./~3 = 0.1755Z3 +~ (8.114)
Substituting these in the failure surface equation, the equation in terms of
the uncorrelated normalized variates becomes
at(O.l451Zt - 0.0338Z2 + 1.222) - (0.0893Zt + 0.055Z2 + 1.05)
-- a3(0.1755Z3 + 0;62) = o
220

Substituting the values of at = 2.666 and m = 0.5 in the above equation,

we get
0.2975zt - 0.J452Z2 - 0.0878ZJ + 1.898 = 0
The above equation being linear, and since t.tzi = 0 (normalized variates),
1.898
f3 = [(0.297 5 )2 + (0. 1452)2 + (0.0878)2J l /2

= 5.54
The same problem can be solved straightway without using the trans-
formation matrix, since the failure surface equation is linear. The given
equation is
M = atXt - X2 - a1X1 = 0
The reliability index is

ftM = att.tx, - t-txa - U3t.tx. = 1.8978

aM= [(atux.F + aL + (max.F- 2at Cov (Xt, X2)] i 12

= [(2.666 X 0.149) 2 -\- (0.1 05)2 + (0.5 .>< 0. 17 55) 2

-2 >< 2.666 >·: 0.0111]!
= 0.342
f3 = 1 .897~ = 5.54
0.34_
EXAMPLE 8.12 Consider the same problem of the cantilever beam, given tn
Example 8.6. The limit state failure equation is of the form
X1X2 - XJ ~-~ 0
If
/.L.\, '~ 14 ;.; 10 2 ax, = 0.7 ;< 10 2
ftXa = 32 ax, = 3.2
20 /, 10 3
/.LX 3 o...; ax,--" 8:< 103
Cov (X1, X2) "= 1.12 X 10 2 Cov (X2, X1) = 8 :-: 1Ql
Cov (X2, X3) = 45 X 10 4

determine the reliability index if all, Xt, x~ and XJ follow tbe normal
distribution.
Solution The variance matrix is

[Cxl =
l 4900
112
45 ;; 104
112
10.24
8000
45 X 10 4
8000
64 X 10 6 J
 221
\.
The eigen values are
.\1 = 64 X 106 ,\2 = 1738

The corresponding [ V J matrix given by Eq. (8.1 03) is

. [ 0 .7031 X IQ- 2 0.9995 - 0.03 224 ]

fVl = o.J25 x 1o-J o.03224 o.9995

1.0 -0.7032xl0-2 0.1017XI0- 3

The uncorrelated variables Y are given by [Eq. (8.1 04)1

Y = fVl'X
The expected values and standard deviations of Y; are [use Eqs. (8 .106) and
(8.107)]

/LY 1 = 20010 ILl', -= 1260 /LY; = - I 1.]1

ay 1 ·= 8000 ay• = 41.69 ay3 cc= 2.728

Since all X; follow normal distributions and Y; is a linear combination of

Xt, Y; also follows the normal distribution with the corresponding para-
meters JLY1 and ay1• The original failure equation is written in terms of the
uncorre1ated variables using
X= IV]Y (8.1 05)
as Xt = 0.7032 x J0··2Yt -1- 0.9995Y2 - 0.03224Y3
X2 = 0.125 'C I0- 3Yt + 0.03224Y2 + 0.9995YJ
XJ = Yt --· 0.7032 x 10- 2 Y2 + 0.1017xJ0- 3 fJ
Normalizing the variables Y;,

x, can be written in terms of z,.

X = [V][ay]Z + [V]JLy (8.115)

where
0
[ 8000
[a1•J = 0 41.69
0 0

Py ~ [
20010]
1260
- 11.11
222

Substitution of [V], [ay] and fLY in tbe above equation yields

X1 = 58.48Z, + 41.6722 - 0.0879523 (8.II6)
X2 = Z1 + 1.34422 -1- 2.72723 (8. I I 7)
X3 = 8000Z1 - 0.29322 -{- 2.782 X I0- 4z3 (8. I 18)
The given failure surface equation is
x,x2- x3 = o
Using Eqs. (8. 107) to (8.109), the failure surface equation in terms of the
uncorrelated normalized variates becomes
g,(z) = 56.2Szi + 56.0z~ - 0.24z~ + I 17.26ZIZ2 + I53.3z,z,
--1-II3.5Z2Zl -- 4799zl + 32I7Z2 38I6zJ + 24850 +
The procedure of determining f3 is the same as explained in Example 8.6.
Results of iterations are summarized in Table 8.8.

TABLE 8.8 Camputation of {J-Example 8.1:!

Iteration
Variable Start
2 3 4

p 5.000 -7.519 3.721 3.533 3.5.13

0(1 -0.5 -0.624 -0.714 -0.699
()(. -0.5 -0.457 -0.424 -0.437
"a 0.707 +0.634 0.557 0.565

From Table 8.8

f3 = 3.533
the design point: ::;• = Ct.;*f3
(z;, ::i, zj) = (-2.47, -1.544, 1.994)
xi = 58.48zt + 41.67z~- 0.08795z3
= 208.7R
x~ = zi + I .344z; .!_ 2.727z!
= -4.544
x~ = 8000zj - 0.293z; + 2.782 >~ J0- 4zj
= I9759.5

REFERENCES
8.1 Joint Commillee o11 Structural Safety, CEB-CECM-CJB-F/P-IABSE, "First Order
Reliability Concepts for Design Codes", CEB Bulletin No. 112, July 1976 .
8.2 CJRIA, "Rationalisation of Safety and Serviceability Factors in Structural
Codes", Construction Industry Research and Information Association, Report No.
63, London, 1977.
 223

K..l IS: 456--1978, ''Code of Practice for- Plain and Reinforced Concrete", lndimt
Standards Institution, New Delhi, 1980.
H4 Cornell, C. A., "A Probability Based Structurnl Code", Joumal of ACI, Vol. 60.
Dec. 1969, pp. 975-985 .
~.~ Galambos, T.V. and M.K. Ravindra, "LoJd and Resistance Factor Design",
Journal of Struct. Div., ASCE, Vol. 104, ST--9, Sept. 1978, pp. 1325- 1336.
HJ> Hasofer, A.M . and N.C. Lind, "An Exact and Invariant First Order Reliability
Format", Jotmwl of Engg. Mech., ASCE, Vol. 100, EM-I, Feb. 1974, pp. I I i-121.
H.7 Shinozuka, M ., "Basic Analysis of Structural Safety", Journal of Struct. Dil•.,
ASCE, Vol. 109, ST-3, March 1983, pp. 721-740.
~.K Paloheimo, E. and H. Hannus, "Structural Design Based on Weighted Fractiles",
Journal ofStruct. Div., ASCE, Vol. 100, ST-7, July 1974, pp. 1367- J378.
R.() Rnckwitz, N. and B. Fiessler, "Struolllral Reliability Under Combined Random
Load Sequences", Computers and Stmctures, Vol. 9, 1978, pp. 489-494.
R.IO Ellingwood, B.R., T.V. Galambos, J.G. MacGregor, and C.A. Cornell, "Deve·
lopment of a Probability Based Load Criterion for American National Standards
ASS'', National Bureau of Stamlnrtls, Special Publication 577, Washington, D.C.,
June ·1980.
H.ll IS: 875-1982, "Code of Practice for Structural Safety of Buildings, Loading
Standards Part Jl-lmposcd (Live) Loads", India11 Standardr Institution ,
New Delhi, 1986.
~ . 12 Ranganathan, R., "Reliability Analysis and Design of RCC Slabs, Beams,
Columns and Frames--Code Calibration", Report No. DS and T 4/li83-STP ·
llfl5, Civil Engg, Dept., liT, llombay, Sept. 1987.

EXERCISE
H. I For the problem in Example 8.6, what is the reliability of the beam if the coeffi-
cient of variation of the load is 20%.- All other data remain the same.
(Ans. fJ = 3.558)
8.2 For the problem in Example 8.8, what is the reliability of the beam if the mean
value and standard deviation of the strength of concrete are 30.28 N / mm• and
4.54 N/mm• respectively. All other data are the same.
(AJtS. fJ = 3.293)
R.3 For the same problem in Example 8.6, what is the reliability of the beam if P
follows the Type I extremal largest distribution with mean, 100 kN and standard
1lcviation, 30 kN.
(Ans. fJ = 2.608)
f!.·l For the sam.: problem in Example 8.6, whnt is the reliability of the beam if P
follows the Type 2 extremal hll'gest distribution with parameters 11 -.~ 89 3 kmph and
k · 6.42. The corresponding mean = IOOand standard deviation = 23 kmph.
(Ans. fJ = 2.7)
8.5 Forth(! same problem in Example 8.11, determine the reliability index if
(i) the correlation coefficient between the variables X, and X 2 is 0 .5
(Ans. fJ = 5.169)
(ii) p between X, and X 2 is 0.5 and p between X 2 nnd X a is 0.8.
(Atlf. {J~4.90Il
8.6 (a) The shear sfrength. R. of a RCC beam is given by the following model
cqur~tion

R~ B[ 1.1 Asvfy.!{_s + 1.8566hd (feu~ !!...)''

bd a
8
]

where 11 is the model error, A,. is the area of stirrups, b is the breadth, d is
the effective depth, sis the spacing of stirrups, A, is the area of tension steel
and a is the shear span.
224

It is given :
fy : JL = 469 N/mm• a = 46.9 N/mm'
fe 11 : p. = 17.58 N/mm' a= 3.16 N/mm'
b : p. = 310.3 mm a= 9.47 mm
d : !' = 556 .3 mm a= 3.79 mm
s: 11 = 150 mm rr=13.5mm
B: p. = 1.2 (J ~ 0.13
VD : I' = 73430 N n = 7343 N
VL:~.t=57830N a=- 16400 N

where VD and VL are I he shear forces due to dead load and live load, respec-
tively, feu and VL follow the lognormal and Type 1 exlremal (largest) distribu-
tions, respectively. All other variables are normally di;;tributed. Determine
the reliability indtx of the beam at the limit state of collapse in shear if
Asv = 100.5 mm 2 , A/bd:'0.008 and a!d=4
(Ans. fJ = 7.68)
(ii) If the shear strength of the beam is predicted by the following model

R = B[!.y 1 ISi'
!!___
s
-1-
·
hd. ~--{vi+58
6 V O.Bfcu 0
- I}]
what is I he reliabilily or the beum if 8 is 2.175
(Am. f3 = 4.45)
§ (i) The safety checking format ol' a steel column sttbjectcd to axial load P and
bending moment M is as follows.

where kiP is the plastic moment capacity of the column when thcte is no axial
load and Puis the ultimate axiulload carrying capacity of tht: column under
pure axial load case.
Area of the cross-scclion is 6496 mm' and plastic section mouulus of the
section is 678700 mm•. It is given:
For fv : l-' 262.5 N, mm• a -- 26.25 Njrnrn'
PD: 11 ~-, 0.398·: 10' N a o,- 0.398x 10• N
l' L : II- 0 3 I08 - : l 0° N " ,- 0.870:< ro• N
flfv : JL - 0.1785 1o• N nun a-- 0.1785:.; 107 N llllll

M1.: JL "0.1394 :c to• N mm a ~ 0.3945 :< 10 7 N mrn

where Pn and PL are axial loads due dead load and live loud respectively.
MD and filL are moments due to dead load and live load respectively.
Determine the reliability of the column.
(Am. fl ~" 3.463)
(ii) If ihc safdy checking format uses •1 nonlinear model giwn by the following
equation

( ~)
.\fp
·I (!.-)' ~
f'u

what is the reliability of the column'!

(Ans. fl = 4.22)
 9
Reliability Based Design .

9.1 INTRODUCTION
In the last chapter, we studied the Level 2 (including advanced Level 2)
methods in detail. Using the same methods, the evaluation of the reliability
of structural elements was illustrated. Now the problem is reverse. One
wants to produce a structural design which will ensure a certain level of
reliability. That is to say, to provide a design for a specified level of risk/
reliability. This was demonstrated in Example 8.5 also, where the depth of
the girder was calculated to be.safe against the limit state of collapse in
shear, ensuring the required reliability level.
Consider the fundamental case: a structural element/system with a
resistance R subjected to an actionS. If R and S are independent normal
variates,
fl = /!R - JLS
(9.1)
V a~+ al
Therefore, the mean resistance (representing the design) required to ensure
the specified reliability or target reliability, flo, is
(9.2)
If one uses the other safety format, assuming R and S are independent
lognormal variates, the median value of the required resistance of the
design is

(9. ~ }-

But in practice, R is represented in terms of several resistance variables a11d

design constants, and S in terms of load variables and design constants.
For safety,
gR(X1,X2, .• . , Xm, C1, C2 ... ) ~ g 8 (Xm, 1, .:\'"m,2, ... ,
X,, C;, Cj; I ... ) (9.4)
where, X1, ... , Xm are the resisting variables, Xm+l, .. . , X, arc the loading
variables, and C1, C2, ... , Ci . .. are design constants. gR and gs are
resistance and load effect functions respectively.

.. .
If x~ are the design values of variables, then the design equation is
gR(XJ, X2, •.. , Xm, CJ, c2 •.. )
. .
~ gs(Xm+l, Xml-2, •.• , cj, cjrl . . . )
226

The partial safety factor or the safety factor is defined with respect to a
particular value of the variable. If it is defined with respect to the mean
value, as given below,
.
'Yci = -X; (9.6)
/Li
It is called the central safety factor. J.L; is the mean value of X;. If the partial
safety factor is specified with respect to the specified characteristic value,
Xk; of X; (corresponding to five per cent fractile in the case of resistance

'Yki = -
X;
.
vari3ble and 95 per cent fractiJe in the case of load variable), then

(9.7)
Xki

The partial safety factor, 'Y;, defined with respect to the nominal value, x,.;,
of the variable X; is given by

'Y; =
X;
. (9.8)
Xni
Jn this text, whenever 'Y; is used, it refers to the partial safety factor with
respect to the nominal value. Using these y;, the design equation (Eq. 9.5)
becomes
gR('Y1Xn1, 'Y2Xn2, . . . , 'YkXnk, lt, C2, . . . ) ?:
"
gs(Yk+l xn,k+t.Yk+l x,.,h2• ..... ,C1 ,CJ+I .... ) (9.9)
.~ ,. '~'Jv;w-.,

Presently. the reliability based desigli means arriving at these values of

pnrtial safety factors for a given target reliability for a particular failure
criteria. Once safety factors are calculated, the design values are known
and hence the design is proposed for the specified reliability. The computa-
tion of partial safety factors and the process of reliability based code
calibrati0n arc de,11t with in this chapter.

9.2 DETERMINATION OF PARTIAL SAFETY FACTORS

The reliability based design criteria is developed using the first-order second-
moment approach. In the last chapter, the reliability analysis was introduced
and illustrated using the Level 2 method. The probability of failure or
reliahility (in terms of {3) was calculated for given safety factors for a given
limit state. Now the process is reverse: partial safety factors are to be
evaluated for the given target (3. The same Level 2 reliability method can
he used. In the normalized coordinate system, for a given failure surface,
the shortest distance from the origin 0 to the failure surface defines the
safety of the design . Different levels of safety (i.e. ,8) will yield different
failure surfaces, as shown in Fig. 9.1, amounting to different designs.
Hence, in the reliability based design, the problem is to
determine the
de ign values of the variables that will result in designs having failure sur-
faces that comply with a required safety index {3. If x~ is the design value
of the original variable X;, the failure surface equation is
(9.1 0)
 227

FIG. 9.1 Design corresponding to different reliability indices

If the partial safety factors are attached to the nominal values of variables,
the above equation becomes
g(YIXnl, Y2Xn:: •... , YnXnn) = 0 (9.11)
The design point should be the most probable failure point. Now the
problem is to determine the most probable failure point. In the normalized
coordinate system, tbe most probable failure point is given by [Sec . 8.3.1 :
Eqs. (8.45) and (8.46)]

(9.12)

where (9.13)

The original variates are given by

Xi
•= + o;z •1
p.;

= p.; + CJj':J.~~ (9.14)

This equation can also be written as

X~ = p.;(l + 8;-x ~~) (9.15)

where 8; is the coefficient of variation of X;.
Hence the partial safety factor required for the given {3 is

)I;= X: = 1'/1 + 8,-x;{j) (9.16)

Xqt Xqt
 228

If the partial safety factors are specified with respect to the mean values, i.e.
XI
'>'cl =-
l-'1

then 'Yc; = 1 + 8; 11.~{3 (9.17)

If the partial safety factors are referred to the specified characteristic values,
then the nominal values are replaced with the characteristic values in
Eq. (9.16). The procedure of computation of the partial safety factors is
illustrated in the following examples.
~AMPLE 9. 1 A simply supported steel beam (RSJ) of span 8 m is designed
"· for th e f !lowing data:

Mean Nominal
Variable 8 Distribution
Nominal value

X1: Yield strength 1.10 250 N/mm 2 0.10 normal

of steel
Xz: Dead load 1.05 11.0 N/mm 0.10 normal
XJ: Live load 0.70 12.0N/mm 0.40 normal

Determine the partial safety factors for the design variables X; if the target
reliability is 4.0.
Solution The limit state equation in the original space, g(X) = 0, is

X1Zp- X2 (':) - XJ ( ~) = 0 (9.18)

where I is the span and Zr is the plastic section modulus of the section.
Normalizing the variables by using the equation ,
X; - p.;
Z ; = -- -
a,
the limit state equation in the z space is
'I J2 J2
~
I
g,(z) = Zp(aiZJ + P.l) --g(a2Z2 + fl-2),- (aJZJ + /-'l)g = 0

Let
Zp
A = 8-[2-

Then
g,(z) = A(aJZJ + p.J) - (a2z2 + P.2) - (aJZJ + P,J) = 0 (9. 19)
Using Eq. (8.46),
 229

where

K = ( E( ag,)z]r'z
1-t az, •
~~ = - i (utA)

1
.otz = - - (-uz)
K
1
~3=-x(-u3)

Since the limit state equation is linear, and all variables follow the normal
distribution, the reliabiJity index is given by
Af'r - 1-12 - f'l
+ u~ + u~] 1 12
(9.20)
fJ = [(Aut) 2
ln this design problem, the value of fJ is already given as 4. Hence
A14r - f'?. - f'J
(A 2ur
2
+ <722 + 2
UJ )I/1
= fJ

(AP.t - 1'2 - 1'3)2 = (J?.(A2a~ + a~ + ui}

The quadratic equation in A becomes
b1A 2 - h?.A + bJ = 0
where
bt = p.f - {J.2 u~
h1 = 2p.t(l'2 + f'J)
b3 = I'~ + 14~ + 21421'3 - /J2( 0'~ + 0'~)
Substituting the given values of ,.,, u; and {J, and solving the quadratic
equation, we have
bJ = 63525, b2 = 10972.5, bJ = 196
A= 0.153
Using the computed value of A, the di~ectional co'sines rx, can be calculated.
1
Ott=- K (27.53 X 0.153)

1.15 3.36
. 0t2 = - -K OtJ =K
Using l: rx7 = 1 and K = 5.507, we have
rxt = ·- 0. 764, ocz = 0.21, 0t3 = 0.61
230

Design points in the original space are

X~ = /LI + OCt {Ju1
= 275- 0.764X4X27.5 = 190.96

X; =; 11.55 + 0.21 X 4X 1.155 = 12.52

x; = 8.4 + 0.6Ix4x3.36 = 16.6

Hence, the partial safety factors with respect to the nominal values are
"IJ 190.96 = 0.764
'I = 250

Yz = -12.52
--
11
1.138

,..J =
16 ·6 . = 1.383
12
Here ">'t is the partial safety factor (multiplying factor) for the yield strength
of steel. (Note: In IS and British codes, !/">'1 is taken as the partial safety
factor for ~materials. That is, 1/0.764 = 1.309). Hence the design equation
[2
0.764 JynZp ~ g (1.138 Do+ l.383Ln)

will ensure a reliability level of f3 equal to atleast 4 for the given data. jy,.
is the nominal value of jy.
For example, if a beam is to be designed for a span of 6 m and for the
same nominal loads, the section modulus required is given by the condition

0.764X250 Zr ?-
36
~ lOr• (1.138 X 11 + 1.383X !2)
Zp required is 685930 mm 3 • If this Z 11 is provided, the reliability analysis
can be performed and it will be found that {3 = 4 for the same mean values
and standard deviations of jy, D and L.
EXAMPLE 9.2 For the same problem in Example 9.1, what are the values
of the partial safety factors with respect to (i) the mean values and (ii) the
characteristic values.
Solution Case (i)
From Example 9.1, the design points in the original space are
X; = 190.96, x; = 12.52,
•
XJ = 16.6
Hence, the partial safety factors with respect to the mean values are
Y; = x;'
f-1-i

Y, = ~~~·: 6 = 0.694
 231

12.52
"
2
= l 1.55
= 1.084

,3 = 16.60 = 1.976
8.4
Hence the dt>sign equation is

0.694 P-Jy2p ~ g12 (1.084 ftD + 1.976 P.L)

to ensure a reliability level of~ = 4.
Case (ii)
The partial safety factors with respect to the characteristic values are

X;•
'Y; = -
Xk;

where Xk; is the characteristic value of xi. For the yield strength of steel,
(5% fractile)
Xkl = P.l - 1.64 17J

= 275 - 1.64 X 27.5 = 229.9

For dead load (95tYo fractile)
Xk2 = !J-2 + I .64 172
= 11.55 -J- 1.64 X 1.155 = 13.44
For live load (95% fractile)
Xk3 = !J-3 + 1.64 17J
= 8:4 + 1.64 X 3.36 = 13.9
Hence, the partial safety factors with resl-'ect to the characteristic values
are
190.96
'Ykl = 229 .9
=0.831

12.52
Yk; = = 0.932
' 13.44
16.60
,k3 = = 1.194
13.90
Hence the design equation for f1 = 4 is
f2
0.831 }~kZp ~ 8 (0.932 Dk + 1.194 Lk)
where fyk, Dk and Lk are the characteristic values of jy, D and L respectively.
EXAMPLE 9.3 For the same problem in Example 9. 1, what is the value of
the combined load factor ?
 232

Solution From Example 9.1, the design equation is

0.764 hoZp ~ gf2 (1.138 Dn -\- 1.383 Ln)

We want to propose the design equation

I 0.764 fyn Zp ~ g12 [YT(Dn + Ln)l

such that it will ensure~ equal to alteast 4. 'YT is the combined load factor
on total load. This is computed as follows:
1.138 Dn + 1.383 Ln = 'YT(Do + Ln)
YT = 1.138 Dn +- 1.383 Ln
Dn + Ln
\.138 X 11 -\- 1.383 x 12
II+ 12
1.266
Hence the design equation becomes

~ ~
2
0.764 fynZp [1.266 Wn -1- / ..,)]

In Example 9.1, no iteration is involved as the failure surface equation is

a linear function of the normal variables. If the failure function is nonlinear
and/or the variables are nonnormal, the problem is to be solved iteratively.
This is illustrated in the following example.
~MPLE 9.4 onsider the :.ame problem in Exa mple 9.1. Determine the
Vp~·~~ial safet y factors for~ = 4, if th e yield strength of steel (XJ) and
live
load (XJ) follow the lognormal and Type 1 extremal (largest) distribution
respectively.
Solution The failure surface equation is
f2
ZpXi -lfX2

Let the design constant A he

(9.21)

If x;, x; and xi arc the design points, then

Ax; - x; - xi = 0 (9.22)
Since Xi and XJ are nonnormal, the equivalent means and standard devia-
tions of nonnormal variables (Xi, X,) are to be used.
Hence the failure surface equation in normalized variables hecomes
rl(a;zl + f.L;) -- hz2 + f.L2) - (a.;.n + J.L;) = 0
 233

Using Eq. (8.46), the directional cosines are computed:

I
--(A or)
IXt = ----
.K
(9.23)

oz
0(2 =- (9.24)
K

(9:25)

The procedure of computation of the partial safety factors is as follows:

(i) Start with any x;. x;
and x;. .
Us ing Eq . (8.67) and (8.69), compute and 1-4; and a;
and ~; of the a;
non-normal variable Xr and X1 at the design point x•.
(iii) Compute A u ing Eq. 9.:>.2).
iv) Determine the dircct i nal cosines oc; usitig Eqs. (9.23) to (9.25.)
(v) Determine the new design point
X~=":
I r1
+cr:I A
CX. ·I 1-'

(vi) Go to step (ii) and repeat the procedure till the required convergcll<{•
is achieved.
For example, stop if

(916
and/or
Xjl - X j- l,r] ~ 0.005 i = I, 2, 3 (9 17
[ Xjl

where j stands for thejth iteralion.

For calculating the equivalent a, and
I I
/li of X;, parameters a1nXI an
Xt are computed using qs. (3.91) and (3.92):
ara.\'1 = [In (8~ + 1)]1/2 = 0. J

Xt = l-'1 exp (- T afnxt) = 273.6

Using .Eqs. (3.115) and (3.116), the parameters of X1 following the Type
extremal distribution are calculated:
7T
IX = - - - - = 0.382
'V6 °3
0 57722
u = 8. 4 - •
0.382
= 6.839
Start with
XI
•= P.t = 275
•
X2 = l-42 = 11.55 •=
X3 P.l::: 8.4
 234

At x; = 275, using Eqs. (8.76) and (8.77),

I •
GJ = Xt GJnXI = 27.43
p.; = x~(l - In x; + In Xt) = 273.6

At x; = 8.4 using Eqs. (3.113) and (3.114),

F(x;) = exp [ --exp { -0.382(8.4 - 6.839)}]

= 0.5704
/(x;) = 0.382 exp [ -0.382(8.4 --- 6.389)
-exp{-0.382(8.4- 6.839)}]
= 0.1222
Using Eqs. (8.67) and (8.69),
, <fo [(/rl( .5704)1
a3 = ~--:0-.:-::
12:-::-2-::-
2~

= 3.213
!-'; = 8.4 - a;{ci>-1(0.5704)}
= 7.831
Using Eq. (9.22) compute A:
A = (11.55 + 8.4)/275
= 0.0725
The directional cosines are
CXJ = -0.504 ()(2 = 0.292 CXJ = 0.813
New design points in the original space are

i X;= 273.6- 0.504 X27.43 X4 = 218.4

x; = 11.55 + 0.292 X4X 1.155 = 12.9
f x3 = 7.831 + 0.813 >~ 4 x 3.213 = 18.28
I
With these new values of x:,
the whole process is repeated till the required
convergence is achieved. The results are summarized in Table 9.1.
At the end of the fourth iteration,
A= 0.1812

X~ = 235.1 x; = 30.57
The partial safety factors with respect to the nominal values are

Y1 = ,.,,y = 235.
_ 1 = o.94
50
 23!)

TARLE 9.1 Computation of partial safety factors-Example 9.4

Iteration
VArlllble Start
2 3 4

•
A' t 275.0 218.4 231.8 234.8 235.1

xl• 11.55 12.9 12.26 12.07 12.03

X•
) 8.40 18.28 27.54 30.38 30.57
A 0.0725 0.1428 0.1717 0.1807 0.1812
I
27.43 21.76 23.1 J 23.42
"•,., I
273.6 267.6 270.2 270.7
I
aJ 3.21 6.79 9.50 10.21
I
I'J 7.!!3 3.12 -4.47 -6 .97
-0.504 -0.411 -0.383 -0.3!!1
••
Ill 0.292 0.153 0.112 0.104
l!ll 0.813 0.899 0.917 0.919

l ;i~g = 1.093
3
~:} 7 = 2.548
'./

Hence the design equation is y\,

·'1
1'1
---
I
k : ~-
0.94 Zpfy ~ gJ2 [1.093 D + 2.548 LJ .\ k J_ \,U'
-~
;z lSure a reliability level of f3
MPLB 9.5
=

r
4.
The ultimate strength of a RCC beam is given by ,,, r
] A 1 ... b Y1 ·\1 A.,
n = f>A,d [ l - o.z~fcuA. ~ ~ s (), ----<9.28) bfu-.
Let the beam be subjected to a bending moment M due to the dead load
and live load. Then the failure surface equation is
R- M= 0 (9.29)
The main basic variables in this case are jy, feu and M. However, if we
compute the partial safety factors for Jy, /cu and M, we may end up with a
value of Y for concrete > 1.0 and sometimes with high values more than
1.5. This can be quite misleading. This happens because the compressive
strength of concrete does not play a significant role in determining the
flexural strength of the RCC beam. Hence what is done is, the partial safety
factor for concrete is prefixed or selected to account for the various uncer·
tainties . .The concrete strength may play a significant role in columns.
Let the partial safety· factor for concrete strength be 0.667 (given in the
present code as 1/Ymc = 1/1.5). Therefore, the design strength of M 15
concrete is 10 N/mm 2 •
236

It is given:
Variable _{y: (normal)
Mean = 320 N/mm 2 a = 32 N/mm 2
Nominal value = 250 N/mrn 2

Variable M: (Normal)
Mean = 0.82 / lOg N mm a c c.; 0.12 X 1Or: N mm
Nominal value= 0.8;< 108 N mm
Compute the partial safety factors for steel strength and bending moment
for a reliability index f3 = 4, b = 240 mm and d = 480 mm.
Solution Let
Xz = M
Using Eqs. (9.28) and (9.29), the failure surface equation becomes

10 As d Xt - ( 0 ·~ 7 ) A; XT -- 10 Xz = 0 (9.30)

Start with
Xt
• :.c 320 X;~= 0.82 X J08
Substituting the above values, and given values of b and d in Eq. (9.30)
and solving the same, we have
As = 614.7 mm 2
The directional cosines are

CXf = -- KI [ ( 10As d- r·
I. 44 As2 Xt• ) 1
a, (9.31)

(9.32)

Using the calculated value of A$ = 614.7 and other data, the directional
cosines can be evaluated. They are
C(l =-= -0.502 otz = 0.865
The hew values of design points, using
x: = IJ-i + cxi f3a1
are given by
X~ = 255.8 xi = 0. 124 x 109
With these new values of x;,
the whole process is repeated till the required
convergence is achieved. Results of iterations are given in Table 9.2.
The design points are
X~ = 229.3
 237

TABLE 9.2 Summary of calculations-Example 9.5

Iteration
Variable Sta rt
2 3 4

.\'~ 320.0 255.8 233.2 229.8 229.3

,.
•2 8.2 X 107 12.4 X 107 !J.7xJO? 11.6 x 10 7
11.6 x t0•
As 614.7 1291 1318.8 1319.6 1319.7
Gil -0.502 -0.678 -0.705 -0.709
(X2 0.865 0.735 0.709 0.706

The partial safety factors are

229.3
'Y, = 'Yfy . = 250 = 0.917

11.6X107
')'2 = 'YM = 0.80 :.< 10 8
= 1.45
VC.LE 9.6 The shear strength of a RCC beam is given by (9.1)

R = 1.1 Aavh sd + 1.8566 bd ( /cu bda

As d)l/l (9.33)

where Aav is the area of the stirrups, s is the spacing of the stirrups, A. is th·e
area of the tension steel and (a/d) is the shear span ratio. For the limit
state of collapse in shear, the failure surface equation is ·
R- VD- VL = 0
where VD and VL are the shear forces due to dead load and live load res-
pectively. 'It is given:
b = 300 mm d = 580 mm s = I00 mm

!!. = 4 Aa
d
bd o= 0,008
Variablefy: (Nominal value= 250 N/mm2)
p. = 320 N/rnm2 a = 32 N/mm2
Variable feu: (Nominal value = 20 N/mm2)
p. = 26.8 N/rnm 2 u = 4.02 N/mrn 2
Variable VD (Nominal value = 70.0 kN)
p. = 73.5 kN a = 7.35 kN
Variable VL: (Nominal value = 50 kN)
p. = 41.35 kN a = 11.70
Determine the partial safety factors for j,, /cu, VI) and VL for Po = 5,
assuming all variables arc normally distributed.
 238

So/ut ion Let

al = 1. I !!__
s = I . I "' S&O = 6), 38
A. 100

As d)l/3
02 = 1.8566 hd ( h;,-;
= 1.8566X300x580 c·~O&r/J = 40701

Then the failure surface equation can be written as

a1 AsvXi + 02 X~. .1 - XJ - X4 = 0 (9.34)
Start with
x; = 220
x; = 8opoo x: = 70000
I

Using these values, and 01 and 02 in Eq. (9.34),

Asv = 30.87
The d irectiona I cosines are

Substituting the computed value'of Asv and other given 'data 111 the above
equations, the computed directional cosines are
Cl:i = -0.368 IX2 = -0.463
•f
i
IXJ = 0.429 ~r.4 = 0.6Rl
The new values of x; using
X~ = f.ti -t- IX;{3a;

are given by
X~ == 261 X2
•= 17.5
X~ = 89260 •
X4 = 81300
With these new values, the whole process is repeated till the required con-
vergence is achieved. At the end of the second iteration, the final values of
x~ are (Table 9.3):
 239

x: = 248.8 xi=--= 17.69

xi = 88600 x4 .= 79610
TABLE 9.3 Results of iterat,io11s-Examp/e 9.6

Heration
Vnriable Start
2

x; 220.0 261.0 248.8

x•2 18.00 17.50 17.69
.\'; 80000 89260 88600
x4• 70000 81300 79610
Aav 30.87 38.96 39.16
lXI -0.368 -0.445 -0.449
IX2 -0.463 -0.453 -0.451
IX] 0.429 0.411 0.413
IX4 0.683 0.654 0.657

The partial faCtors are

248.8 .
,.., = ,..,y =- 250 = 0.995

,. =,..
2 fc
= 17.69 == 0 885
20 •
.
Y3 == 'Yv0 =. 88600
70000
= 1.266
.. · . . 79760
i'4 = YvL = 5 0000 = 1.595

Note: R~aders' attention is drawn to the point that in. the text all Yt are
multiplying factors.
As stated in Example 9.1, in the IS and British codes, 1/Yt is taken as
the partial safety factor for materials, and are collectively called as the
material reduction factors. As per this, the parWii material reduction
factors are
I
Yme = 0. 995 = 1.005

1
i'mc = 0.885 = 1.131

~FETY CHECKING FORMATS

The safety checking format for a code is defined as the number of partial
safety factors and the w.aY in which they are introduced into the design
equations. For the safety of the structure,
Factored resistance ;;;,. effect of factore~ loads
 240

ln th e de velopment of probability based limit state design criteria, different

national codes usc different format s.
8 - ( Cauada) Format
The ati onal Building Code of Canada (9 .2) uses the following probability
factor format

(9 .35)

where gs refers to the functi on that converts the loads to load effects in
brackets, and Yn, YL , . . . are the corresponding partial safety factor s or
load fact ors for the loads. lJI is a load combination probability factor
depending on one, two, or three loads included in the brackets. The value of
tp is less than or equal to I. This factor takes c_':lre of the reduced probabi-
lity or the ·imullaneo us <:CUtTe n e or I ads. The value gi ven are 1.0, 0.7 ,
and 0.6, respectively fo r one or two or three I adings acting si mu llane-
0 11 ly. The terms Yo, YL . .. lake care o r ari ati ns in the load itself plus
vuria ti ons in the load efrccts du e to un ce rtai nti e · in the load model an d the
structural analysis.
T he facto r Yu represents th e overall res istance l'tclor, based on chara ter-
islic trcnglhs, ma terial properti es, dimens io n, etc. This fa tor is intended
to ren ec t th e probability th at the member a u wh It; is und en;Lren gth .
CEB Format
CEB committee (9.3) recommends the following format

(9 36)

where g n nnd gs urc the resislanc · and load clfe L ru nctr ns wh ich co nvert
th e t ·rms iu the bracket t re i · t ~ n ce and It ad eJfec ts respective ly. ji. and
QA. arc th e chara tcr istic strength a nd J, ad · respective ly. Y,, is the mater ial
red uctio n !'actor. It is to be noted th at 'Ym1 =-- I. 'YJt is th e multipl ica tive
factor on the I au.
The material reduction factor Ym; is intended to take into account (9.3)
(i) the material strengths occasionally falling below the specified charac-
teristic value
(ii) the· possible difference between the strength' of the material in the
structure obtained from control test specimens
(iii) the possible weakness in the structural material or element structure
resulting from th e c n ·tru ·tion pr es ·
(iv) th e po iblc inaccura te a essment of the resistance of a structural
clement result ing fr m modelling errors (say, models derived from the
elementary strength of materials)
(v) the effects of poor dimensional accuracy in the finished structure on
the resistance of a section
The partial factors for loads, 'Yr;, are iniroduced to account for the follow
ing factors:

I
.l__
 241

(I) ')'r1: for the possibility of loads occasionally exceeding their character-
t•tic values
ti} ,r2: .multiplicative load combination factor for the reduced probability
f' nil loads exceeding their characteristic values simultaneously
ii) ')lrJ: multiplicative factor on load effects for possible errors in pre-
llctlng load etfects as a result of inaccurate structural analysis and as a
. re ult of neglecting dimensional inaccuracies.
In addition, either Ym or Y£ may be modified to take care of the nature
or the structure and the seriousness of attaining the limit state.
The European Concrete Committee Model Code (9.4) recommends the
rnnowing equation:

(9.37)

. where Qtk represents the characteristic value of the main time varying load
Q•, and Q2k . .. , Qnk are the characteristic values of other less dominant
time varying loads Q2, ... , Qn. 'Po1 is considered as the ratio of the
arbitrary point-in-time value of the jth load to the characteristic value of
that load. Ya is the load factor on the combination of time v~rying loads. It
consists of 'YJi'YfJ . While determining the maximum factored load effect for a
cnsc involving several time varying load s, it may be necessary to consider
11cveral combinations with each of the loads considered as the most domi-
nant load (i .e. QJk) in turn. Hence, in the above format, when a structure
hns to resist a number of stochastically independent time varying loads, a
number of load combinations are to be considered. For a situation with
dead, live, wind and snow loading, the CEB format requires a checking of
32 load combinations. If the NBC format [Eq. (9.35)] is selected for loads,
viz. dead load, live load, wind load and snow load, a total of 14 load com-
binations are to be considered .
However, the Load Resistance Factor Design (LRFD) checking format,
discussed below, requires only four load combinations to be considered.
LRFD Format
The load and resistant factor design checking format, proposed by Ravindra
Galambos, Ellingwood, et al. (9.5, 9.6) recommends only four load combi-
nations to be considered . They are
'Y RRn ~ 'YDf'D + 'Y Lf'Lm (9.3,8)

'YRRn ~ YnP.D + 'Yapt/LLapt + Ywp.wm (9.39)

'YRRn ;;?: 'Ynp.n + 'Yaptf'Lapt + Ysp.sm (9 .40)

'YRR" ;;?: 'Ywp.wm - "'n/LD (9.41)
where P.D is the load effect due to the mean dead load, 1-'Lm, p.wm and P.sm
are the load effects due to means of the maximum lifetime live load, maxi-
mum lifetime wind load, and maximum lifetime snow load respectively.
Here the term Yapt P.Lapt in Eq. (9.39) is equivalent to
242

gs('Yrt"'f3lJ'oJQJk = 'YalJ'oJQJk) in Eq. (9.37), the major difference being that

the load is given as a multiple of the maximum load (lJ'oJQj!,) in Eq. (9.37)
but as a separate loading case with its own load factors in Eq. (9.39). The
load factors, in general, should be applied to the loads before performing
the analysis which transforms loads to load effects. If the relation between
load and load effect is linear, load factors can be applied directly to load
effects.

9.4 DEVELOPMENT OF RELIABILITY BASED DESIGN CRITERIA

Before starting the procedure for the development of design criteria (evalu-
ation of partial safety factors), the precise scope of the work should be
defined. That is, the types of structures for which it is applicable, the types
of materials that will be used, and the range of parameters that will be
covered. The proposed work should be compatible with the present code.
It should also specify the range of application of the code and the different
limit states (ultimate and serviceability) considered in the,work.
One must specify the safety chec'king format selected. By format is meant
the number of partial factors and the way in which they are introduced in
the design equations (i.e. on loads, load effects, material strengths, resist-
ances, etc.).
It must also specify the basis on which the loads have been developed.
That is to say, whether the loads have been developed for a 50-year design
period or a 25-year design period. It means specifying the selection of the
period for a risk assessment for the class of structures being considered.
The development of a reliability based design criteria involves the follow-
ing steps:
(i) collection and statistical analysis of the data on basic variaoles.
Defining of the probability distribution of each variable-at least in terms
of mean values, standard deviations and probability distribution type
(ii) statistical study of the strengths (resistances) of members and establi-
shing their statistics
(iii) reliability analysis and determination of the reliability index f3 for the
members designed as per the present code for each load combination
(iv) selection of the target reliability index, f3o, (i.e. accepted or specified
level of reliability)
(v) determination of the partial safety factors for the desired uniform
reliability f3o under all design situations within the scope of the work.
For illustration, let us assume that the scope of our work is to determine
the partial safety factors for RCC members (slabs, beams, columns) for
Indian conditions. The limit states considered are the limit states of collapse
in flexure, shear, combined axial load and bending moment in columns.
The lifetime of structures is selected as 50 years. The safety checking format
used is as per the LRFD method.
An extensive data on the basic variaole::, viz. the mechanical properties
 243

hat different grades of steel and concrete, structural dimension s of RCC

37) mbcrs slabs, beams, and columns, including the position of steel for
'he lndl n conditions, has been collected and statistically analysed, and the
ing ulta of the same (9. 7-9 .I 0) are given in Chapter 4. The consolidated list of
!en 1 tl tic of the basic va riables is given in Table 9.4. The statistical analysis
>ad the data on live loads on office buildings, and wind loads based on wind
Mpeeds observed at various stations in India is given in Table 9.4 (9. 11, 9.12,
,13). The statistics of the lifetime maximum live load Lm and the lifetime
maximum wind load W m, given in Table 9.4, are based on the selected
deeign period of 50 years.
lu-
be TABLE 9.4 Statistics of basic variables
pes
Variable X 1'-X{Xn Probability
be
distributi on
de.
ent fou
Nominal Mix M 15 1.51 0.24 Lognormal
ant M 20 1.46 0.21 Norma I
lin Design Mix M 15 1.17 O.IH Lognormal
ist- M 20 1.34 0. 15 Normal
M 25 1.21 0 . 15 Normal

·ed.
Fe 250 1.28 0.10 Normal
ign
Fe 415 1.13 0. 10 NlHillal
the
Slabs
t/ ( mm) 1.87* 4 . 17** Normal
)W-
llcams
b (mm) 10.29* 9 ,47** Normal
d (Jlllll) 6 .25* 3. 79** Normal
les. s (mm) 0 .00 13 so·"* '\formal
·ms
Columns
b (mm) -0.25* 5.69** Normal
bli- D(mm) 0.113* 9.H9** Normal
Bar placement (mm) 0.640* IUJO** 1\:ormal
the Loads
D 1.05 0. 10 ~01'111:1 1
0 .620 02H I:Xr ,t
fled
0. 1'9 0.55 I,01gllOJ 111 <!1

)fm O.R04 0.334 r.,-,_,-"

0.045 0. 743 rx,,,
1ine *Deviation from mean (mm) ; **Standa rd dev inti on (mml
for -I EX1, 1 denotes Type I cx lr~nw I (larges t).
tpse
lnS. Statistical Study of Stre11gtfl of Memhl!l'.\'
mat
The concrete members, considered here, arc sla bs, beams, and columm . The
limit states considered are the limit states (lf (i) collapse due to flc :-. llll: <tnd
·ties
 244

shear in beams, (ii) collapse due to flexure in slabs, and (iii) collapse of
columns subjected to axial load and uniaxial bending. Hence, the partial
factors presented apply to only these cases.
The strength of RCC members vary from the calculated nominal strength
due to variations in the material strengths and dimensions of members, as
well as due to uncertainties inherent in the theoretical model chosen to
compute the member strength. The Monte Carlo technique, dealt with in
detail in Chapter 7, is used to establish the statistics of the strengths of
members in flexure, shear, etc. The procedure involves the following steps:
(i) Selection of a theoretical model to calculate the member strength for a
particular limit state and the model error associated with the same. The
model error, say for the flexural strength of beams, is to be obtained by
collecting data on the experimental results of beams tested for the ultimate
strength in flexure and comparing these values with values obtained by
using the theoretical model equation for predicting the ultimate strength of
beam . The collected data can be statistically analysed and the mean and
standard deviation of the model error can be fixed. (ii) Choosing a series of
representative cross sections or members (different sizes, different boundary
conditions, different spans, different percentages of steel, different grades of
concrete and steel, etc.), each defined by a set of nominal strengths and
dimensions. (iii) Establishing the statistics of the resistance of each selected
member is carried out as follows: For the selected member nominal resist-
ance, R", is computed based on the nominal material strengths and dimen-
sions substituted in the theoretical model with the resistance factor .ts unity.
This valu e of R, corresponds to the failure mode expected when nominai
strengths exist in the members. The design resi stance, Rn, is computed from
the model equation given by the present code using nominal values with
partial factors or material reduction factors (!'or concrete Ymc c~ 1.5, for
~ tee! Ym, -- 1. 15). The resistance reduction factor Yn is evaluated using
Yn -·- Rn/ Rn. A set of material strengths and dimensions is generated
randomly from the statistical distributions of each variable and are used to
calculate the theoretical resistance, R, along with the randomly generated
value for the model error. Then strength ratio R/ Rn is determined. This
procedure is repeated and a large number of samples of R!Rn is generated .
A probability model is fitted to the generated data. A normal distribution
is filled to the lower tail of the data and the statistics of R/ Rn are establish-
t:d . B) repeat1ng step~ (ii) and (iii), the statistics of the strength ratio of
dilrcrent members arc established. The procedure of the Monte Carlo
lltLlhod " ~" dealt with in det ail in Chapter 7. A few typical v alue~ of the
established resistance statistics a!ld the range of Yn values observed for
RCC members are given in Table 9.5.
Using the established statistics of resistance ratios and loads for Indian
conditions, the reliabili ty analysis of RCC members designed according to
tht: present code (9.1 4) procedures is carried out using Level 2 methods
described in Chapter 8. The reliability levels of the present desi gns are
found out using the Level 2 method for various load combinat i on~·

L I
~·
 245

:of TABLE 9.5 Typical resistance statistics of RCC members

tial
tUber Steel Concrete
grade grade
gth
as
to l"
'' way (SS) Fe 250 M 15 1.433 0.124 Range
lin Fe 415 M 15 I .275 0.124 0.835 -0 .865
of . 'l'wo way (SS) Fe 415 M 15 1.281 0.124 Average
:ps: Hrr wuy ( ) Fe 415 M 15 1.263 0 .136 0.85
or a I wny ( ) Fe 415 M 15 1.286 0.12'.1
tl Ill\ (no ure)
rhe 11111>' rein forced Fe 250 M 15 1.288 0.104
by Fe 415 M 15 1.170 0 . 104 Range
ate Fe 415 M 20 1.179 0.103 0.835 - 0-845
by Fe 415 M 25 1.16'.1 0.101 Average
. of Fe 415 M 15* 1.197 0.105 0 .84
l>llubly reinforced Fe 415 M 15 1.151 0. 103
tnd tll!ams (s hear)
; of Range
ary Fe 250 M 15 1.355 0.166 0.855- O. K65
of Avc n1g<.:
tnd Fe 415 M 15 l 277 0.165 0.86
ted Range;
ist- ( 'uru prc~s ion Fe 415 M 20 1.29 0.152 ().(,~ 0.7~
en- i\ vc ragc
ity. Fe 415 M 20* 1.38 0.224 0. 725
nai Range
I ens ion Fe 415 M 20 1.19 0. 13 O.fitHl.S9
om A vet age
'ith Fe 415 M 20* 1.22 0.15 o.g
for
ing N,t,•: SS ·=simply supported; C = continuous
ted * = indicates nominal mix.
to
ted (I) D + Lm (ii) D + Wm, and (iii) D + Lm + Wm. A summary of the
'his r~sults of the same is given in Table 9.6. Based on the above study, a proper
ed. lurget reliability is selected. For the selected target reliability, partial safety
1011 l'uctors are evaluated for different load combinations for each member/limit
sh- Htute. The evaluation of the partial safety factors is illustrated below.
ExAMPLE 9.7 (Load Combination : D + Lon) Consider a RCC beam. After
. of
rio considering all the possible combinations of the grades of concrete and steel
the l!latistics of the flexural strength of RCC beams have been taken, as given
ror helow, for the study ol' the partial safety factors at the limit state o f col-
lnpse in flexure (9.12) .
ran
to Grade of steel /1-R / RII a Distribution
xis Fe 415 1.17 0.122 Normal
are Fe 250 1.289 0.1289 Normal
246

TABLE 9.6 Range of reliability indexes for RCC members

Load Range of Average

Member Remark
combination fl fl
D -1- Lm
Slabs 4.2 to 4.8 4.5
Beams (fkxure) 4.3 to 5.5 4.9 Range of Ln/D 0
Beams (shear) 3.3 to 3.8 3.6 0.25 to 2.0
Columns 3.3 to 4.6 3.9
D +W 111
Beams (lkxu re) 3.5 to 5.1 4.3 Range of W 11 /D 11
Beams (shear) 3.2 to 3.5 3.33 0.25 (O 2.0
Columns 3.2 to 4.2 3.50
D + Lm + W,..,
l:l<.:ams (flexure) 2.9 to 4.6 3. 75 Range of L 11 / D11
Beams (shear) 2.9 to 3.4 3.15 0.5, 1.0, 1.5
Columns 2.8 to 4.1 35 Rang~:_ of W,fD,
0.25 to 2.0

lhe load statistics (from Table 9.4) and resistance statistics of' the beam
at the limit stale of collapse in flexure (for the steel grade Fe 415) arc
Variable !~/ Ro: (Normal)
/J. = 1.17 a ~= 0 122
Variable D/ D.,: (Normal)
,~ = 1.05 a=O.IOS
Variable Ln,/Lu: Type I extremal (largest)
0.62 a~- 0.1755
Para meters u - lUIS 'l. I.S95
1f Fe 415 steel grade is used l'or reinforcing bars, d~:termine tl1e partial
safety factors for the limit state of collapse in flexure under the load combi-
nation D -i- L if Ln/Dn "~ 1.0 and f3u =- 4.5.
5:iulution The safety checking format (LRFDJ under dead load D and live
load Lis

The limit state equation is

R- D - L ~" 0
The equation can he rewritten ns

(/)) Dn /Jn (!:.)L

Ln
. ·-· 0 (9.42)

Let
L, ---
Do -- Cl.l
Rll
 c247

X2= p X3= L
Dn Lo
(
I h n q. (9.42) becomes
RnXa - XzDn - aJX3.Pn = 0
II be remembered that for L, the statistics of Lm must be used for the
n1Do uti male limit stale. The reliability index is given by

{3 =I-'M
aM

_ Rni-J.t - J.t.lDn - aw.3Dn

(9.43)
- [(l?not) 2 + (azD,)2 + (aJ(T)D11 )2]l/l
\\ h r wl and a~ are the mean and standard deviation of the equivalent
1111rmul X3 of the nonnormal variable X3 at the design point. Let the starting
lr ·I n point be
WID
.
n 11

Xt
.= /1-1
.
xz = /J.2 X3
.= /1-3

eam
AI x; = /1-3 = 0.62, the parameters a:i and p.3 for the Type I extremal
(I tr •e·t) distribution are calculated as illustrated in Example 9.4. They are
a3 :;: 0.1678 p.) = 0.5903

Sll tituting the values of f3 = 4.5, m = I, o~, p.~, and othe~ as aml p., values
n l.q. (9.43), we have
S 1.17 R, - 1.05Dn - 0. 5\> Dn
4
. = [(0.122Rn) 2 + (0 . 105Dn)2 + (0 . 1680, )21''2
lving the above quadratic equation in Rn, we get
R,, =:= 3.004bn
I he directiohal cosines are
trtial 1
mbi- ·Y.t = K(Rnod
I · I
·live = - K(3.004 X 0.122) D,-= -y(0.366Dn)
I I
IX2 = K(o~~o) = K (O.l05Dn)

OC3 = ~ (OJ03Dn) = 1 (0.1680,)

U ing E oc} = I and K= 0.417D,;, we have

, OCt = -0.878. IX~ = 0.252 0':3 = 0.403
9.42)
The new design point x; is given by
x; = 11.a + ,,~~a3
= 0.5903 + -0.40JX4.5X0.168 = 0.'894
248

At this new design point, new values of a:J and 1-'J are calculated and the
whole process is repeated till the required convergence is achieved. Results
of subsequent iterations are given in Table 9.7.

TABLE9.7 Summary of computations-Example 9. 7

Iteration
Variable Start
2 3 4

x,.
. 1.17 0.687 0.730 0.772 0.~02

x2
XJ
. 1.05
0.62
1.169
0.!!94
t.l52
1.177
t.l39
l.t41
1.131
t.553
cr'.l O.tli8 0.270 0.368 0.440 0...179
li] 0.590 0.501 0.3t3 O.t26 0.0056
R 11 'D 11 3.004 3.193 3.304 3.348 3.359t
o:, -0 87H -0.802 -0.725 -0 671
CX:~ 0.252 0.216 O.t89 0.172
o:, 0.403 0.556 0.662 0 722

At the end of the sixth iteration

Rn 3.361
Dn =
\2
.= 1.1.' XJ = 1.62
The partial safety factors with respect to the nominal values at('

Y, =
.
~ = x;
X in

'>incc the \'ariablcs X; have been initially normalised with re~pect to their
corre'>ponding nominal values. Hence
Y1 = YR = 0.82
YJ = YL -= 1.62
The dc~ign equation is
YuRn ;;:: 'YvD, + Y1.L 11

0.82Rn ~ 1.13/Jn + 1.62Ln

The same problem has been solved for various \cdues of Ln/LJn equal to
0.25, 0.:\ 1.0, 1.5 and 2.0, and the variation of the partial safety factors
with Ln/Du is shown in Fig. 9.2. If the steel grade Fe 250 is used, f'R!Rn
- 1.289, aR R, = 0.1 lll9. Fot this case also, the variations of "/X, Yv and i'L
. with L,/Dn arc shown in the same: Fig. 9.2. It is observed that yu increases
slightly with an increase in the L,f Dn ratio~. This is due to the usc of the
higher \~lluc~ ofY1. at higher ln/Dn ratios. The de~1d load factor YD shows a
slight f~1ll with an increase in the /.n/ {)" ratio; hut can b(' treated to be a
fairly enn~t:1111 \'aluc. The variation in YJJ is very small because the vari:1tion
 249

the In d ud l~itd is situdJ compared to other load vai·iable&-: 'YL iricreascs with
suits uwr •a c in the Ln/ Dn ratio as its higher variability becomes incr'e~isingly
!llll'c dominant in determining -the total load effect.

~ :,·5
2·0 f~ ,,s
4

).802 .,; 1·5

L
1.131
.553 To
).479 '(R 1·0 ---------------fo
).0056 -----=--- YR
U591
0 ·5

0
0 05 1•0 1·5 2·0 2·5
2·0
~ ::4·5

'fL 1·5

lo
r---------'fo
"fp 1·0 YR

their
0·5

0 0 ·5 1•0 1·5 2·0 Z5

Ln/Dn

FIG. 9.2 Variation of pan ial safet y f<tc tn rs for BCC be<Jm m
flexure under load : D + Lm
utl to
:~.ctors Load Comhinulion: D + Wrn
J.'R/fiB he procedure f computation of the afety !'actors for the loau cmnbina-
nd 'YL Lion D + Wm is same as used for the load ca . e D + Lr,., expl11 incd a \HI
reases illustrated in Example 9.7. The only difference is that the correspomling ,
Jf the ·tatistics of Wm are to be used instead of lhose of Lm . Typical .:urve ~
ows a bowing variation f the wind load factor i'w, i'o, anti i'11 with respect t o
he a the Wn/Dn ratio are shown in Fig. 9. 3. Here al. o, similnr observation s arc
iation made about 'YR, Yn, and rw as in lhe previous case D + l.m, i.e. Y11 increases
 250

.shghtly with illL'ICasc in Wn.' Dn. YI> rcma1ns fairly ~:unstant, and 'Ywincrcascs
a~ Wn/Dn iucrt:<I~C'

2·5
r =3·5
Fe 415

2·0
'lw
lo
'fR 1·5

,..____
10 -------------------'0
"fR

o-5o! ---...J.,.-----,-,J,.o---:,*=
.s- --:;2:";·0;:--''
_ _ _ _ _'Yw

,0

0·5ol____o:S---',-=-o---'"7,.~.:.s,------='2.-=o-'
Wn/Dn
FIG 9.3 Variation of partial safety factors for RCC
beam in flexure under load: v~+ wn,

fhe determination of f3 for the load combination D -1- Lapt + Wm is

1 I illustrated below.
I
I ExAMPLE 9.8 (Load combination: D + Lapt + Wm) Consider the same
problem in Example 9.7. The beam is subjected to wind load along with the
graYity loads. From Table 9.4, the following load statistics are taken.
Variable: D/Dn: (normal)
f< = 1.05 a = 0. 105 a= 0.10
Variable: LaptfLn: (lognormal)
f< = 0.179 a= 0.098 a= o.ss
Variable Wm/ Wn: (Type I extremal (largest))
JL = 0.804 a = 0.269 a = 0.334
\SC I h t'c I ·tnnce statistics are the same .a~ given in Exat\1ple 9.7. That is, the
Itt 1111 nnu standard deviation of R/ Rn arc 1.17 and 0.122 respectively.
U I rtninc the partial safety factors for ~o = 4. It is a:Jso given that

~:
Jt'n
= 1.0 Dn = 1.0

\ ulution The failure surface equation is

R -- D - Lapt -- Wm -' 0 W44l
I~ \ riting the equation, we have

R ) Rn ·- ( Dn
( Rn D ) Dn L"P')
· ( Ln Ln - ( W~-
Wm) Wn = 0 . (9.45)
I

L•t
R
x, :.-:- D
X2 = -·-
Rn Dn
1-Vm
X4=
Wn

-Ln = {/J -Wn · =

Dn
(/4
Dn
l'hcn the failure surface equation becomes
RnXt - X2Dn - XJaJDn · .. X4a4Du = 0 (9.46)
l'he rel iabil~iW)ndex is giv~~ by
~ =- /.I..J R, --· P.2 Dn - /-'JUJlJn - 1'4a4Dn (9.47)
[(o, Ra)2 + (o2D,)2 + {o.lmD,,)2 + (rJ~a~ /) 11 )2]1 /2
rhc directional cosines ~re , ,,,

I
oq = -- K(atRnl

0t3 -= ~( a~aJDn)
, is Start with

tme
the
X~
.= :

X~= /-'3
Jl.t
:.
== L17
=
,·
0.~79
.
xi=·/J-2 =
X~= IJ-4
.

=
L05
.. , ·. .
0.804
The procedure of computation is the same as explained in Example 9.5.
Summary of the results is given in Table 9Jt After the fifth iteration;
•· xi = o;908 x2 = 1.1
x; = 0.199 x: = 2.29
The partial safety factors are
'Yt = 'YR = 0.908 'Y2 = = 'YD 1.1
,3 = 'h = 0.199 ""..;,. Yw = 2.29
252

!'ABLE 9.11 .'>tllllllll/1'_1' of l'l'lllitr-L.IItlllflic Y X

IIera I i<>ll
\ ariablr Star I
2 .1

I 1711 0 K~l II ·' -; ~ II ~'Jt1

I~ I 11~11 I I~,
i lit> 1.10.1
II 17'1 0 24'! II 2 'I II 211'1
II .~Il-l I -IK'J 2.1!1)11 ~.21)

I I (I'}~ ,; : ~ ..... ll.l I •J o Ill'

() l:il> II 1.14 0 I ~2 II 1·1'1
t) ~ ..;;~ II 511~ 0. 7:!8
0 M>'l
() 75'1 II 45- ll ,llhll -0,120
I .73X 3.2'H ' - ~ 11/ .1.935
IJ.5K7 --1),1>0 I - 0, 51\1 -- () 543
11.2'!1 () 157 Cl.l2~ 0. 11'1
11 ,25-1 11.1'!1 11.143 0.121
(J. l 0 71 I 0.760 II X05 (Ul23

Hence the design equation is

(l.901:>Rn ~ I.IDn !- 0.199Ln + 2.29W,
Similarly, for various values of ~V,/ Dn, the values of 'YR, 'YD, and Yw can be
determined. The variation of the partial safety factors for various values of
Wo/Dn is shown in Fig. 9.4. It can be observed that Yo remains fairly con-
stant. '>'L decreases with increase in Wn/Dn up to Wn/ Dn " I, and for 0

W.,/Dn ? I, YI. remains fairly constant. Yw increases with in·creasc in Wn/Dn,

increase is more up to Wn/ Dn == 1.0. The region of interest in design is up
to L./ Dn :( I and Wn/ Dn :( I. ln this region, the variation of load factors
is observed to be high (Figs. 9.2, 9.3 and 9.4). The process of code calibra-
tion involves proposing one set of partial factors for Level I code, irrespec-
tive of the load ratios (e.g. Ln/ Dn or Wn/ Dn), and probably other design
situations -different limit states, ensuring uniform reliability. For this, the
simple optimisation technique proposed by Ellingwood, et al. (9.6) or the
method used by Baker (9.15), can be used after assigning weighting factors
to load occurrence. They are explained in the following section.

9.5 OPTIMAL SAFETY FACTORS

As seen in the previous section, the partial safety factors are not constant
for a given safety checking format and a given target {3. For convenience,
the partial safety factors in the code checking format are to be constant
atleast over a large group of design situations.
As said earlier, the aim of code calibration is to determine a set of safety
factors which will ensure the . best approximate uniform reliability over
 253

:~-u..----------------------,

.li\16 2·5
. 103
::?.ti'J
.21.l 2·0
107
iR
. 149
.728 1o
1· 5
. 1::?.0
"\
.935
.543 1'w
.II'!
1· 0
·.121
t. R23

0·5

be
0~------~------~~------~------~~
>of 0 0·5 1•0 1·5 2·0
on- WniOn
for FIG. 9.4 Variation of partial safety factors for RCC beam in flexure
'Dn, under load: D +
Lapl + Wm
up
tors diiTerent design situations. If a constant set of factors are prescribed, the
xa- nssociated reliabilities will deviate from the target reliability f3o. To select
:)eC- une set (optimal set) of load factors, a function, S('Y;), which measures the
>ign "closeness" between the target reliability and reliability associated with the
the proposed partial factors set. ic; <iefined and this function is minimised to get
the the optimal safety factors.
tors For a given set of partial factors with an associated f3o, there is some
corresponding nominal resistance. Let it be called R~\ obtained using the
Level 2 method. This is a function of the load ratio and load combination.
Let the nominal resistance corresponding to a design equation, which pres-
cri bes a set of partial factor that are constant for all loa I ratios be R,~
tan t
which may differ from R~r · R~ corresponds to Level 1 code. The problem is
nee,
therefore, to find "1;, minimizing the function, S, defined by (9.6)
tant
.E (R~
1
S("';) = - R~)2w; (9.48)
I
Jety
:wer over a predefined set of combinations of dead, live and wind loads wherein
Wt is the relative weight assigned to the ith load ratio. The function selected
 254

is the square or thl: dlilcrcncc bci\\CCJl R, , 311d N,\ sn th:ll the deviations
11

!'rom f3n on either side c;~n be cqll;dly penalized. The determination of the
optimal safd) f<tt:l<ll~ i\ illu~trated hclnw.
EX\MPLF 9.9 Let the s:ti'ct) <.:hcding l'nrmal he
(9.49)
f'nr the loud combitlalion /) : /.111.
T:tking f), I.
('Yu • 'Yt.n;)
R~. (9.50)
'Yu

where o; -, (Ln/Dn);. The S function, given hy Eq. (9.4R), becomes

S(Yn, Yn, Y1.l""" f 11

R, -
Yn +
· ·-YR____
a;YL ] 2
II'; (9.51)
[

To find the minimum value of S, the partial derivatives of S with respect to

YR. Yn,and YL arc taken and made equal to zero. This leads to the following
two equations:

I; IV;a;R~/YR -- J: W;a;'Yo -- I; 1\';af'YL =-= 0 (9.52)

I

1
I; I\';R~ 'YR - - I; W;'Yn - L~ w;a;YL = 0 (9.53)
; ; I

The equations corresponding to oSfoYR and oSjo'Yo are the same.

The computed values of YR, Yo, and 'YL for various values of Lo/Do for
RCC beam in flexure are given in Table 9.9. The weights to be assigned
should be based on the likelihood of different load situations in practice.
The assumed weighting factors w; in percentage (9.6) are also given in the
same table.

TABLE9.9 Values of partial safety factors for beam in flexure-Load

combination D + Lm
~
1 Ln
Partial Dn
Remark
factor 0.25 0.5 1.0 1.5 2.0

YR 0.802 0.860 0.937 0.964 0.977 Po= 3.5

YD 1.200 1.154 1.105 1.087 1.078 JL Lmf Ln = 0.827
1.057 1.505 1.859 1.954 1.995 8L 01 = 0.283
Y/.
L 0 = 3 kNim'

Weights For RCC

10 45 30 10 5 beams
WI
(per cent)
 255

1
>nS h values of R~ for i = 1 to 5 are calculated using the expression
.he
(9.54)

•.nnd the values of 'Y; given in Table 9.9. For example, for i = I, Ln/ Dn
~9)
- 0.25.
Rn _ L!_.20 + _!:957 X 0.!~)
n - - 0.802
= 1.82
50) Using weighting factors given in Table 9.9, we get
5 II
E WJ(IiRnt = 2.6369
j-J

51) S II
E w;Rn; = 2.8049
i-1
to s s .
ing E w;a, = 0.8 2' \Vi= 1.0
1-1 i~l

Using these values, Eqs. (9.52) and (9.53) become

52)
2.6369YR - 0.8yn - 0.844YL = 0 (9.55)

53) 2.8049YR - Yo - 0.800YL =0 (9.56)

From the study of the results (Table 9.9), it is observed that Yn remains fairly
con~tant around 1.1. This has been observed for various load combinations
for an~ failure states {9. I 2). Since the value I. I is low and may not be accept-
ted ab!J by the profession, the value of Yn is fixed as 1.2. Using this value,
.ce. Eqs. (9.55) and (9.56) become
the 2.6369YR - 0.844YL = 0.96 (9.57)
2.8049YR - 0.800i'L = 1.2 (9.58)
Solving the above equations
YR = 0.9495
If it is desired that YR must be around 0.85, as existing in the present designs
corresponding to the material reduction factors Ymc = 1.5, ,.ms = 1.15 and
other material specifications, so that the partial safety factors for material
strengths and other material specifications on the resistance side are not
changed, then keeping the present value of YR = 0.85, and using the same
).827 in Eqs. (9.57) and (9.58), two values of YL are obtained. They are 1.52 and
13 1.48. Taking the average, Y1. is fixed as 1.50. Hence, the optimal safety
·;m• factors for this case in this example are
YR = 0.85
Similar studie~ can be done for other combinations of variables and other
limit states.
 25&

The approach used was suggested by Ellingwood, et al. (9.6). The approach
used by Baker (9.15) is given hclow.
The function used ·ror Sis
S =--~ E (log 10 pr - log 10 {'rc);ll '; (9.59)
I

where
(f!r), = i' the failure probability for the case (say ith load ratio Ln! Dn)
(pft), ,-= corresronding target failure probability
To deter mine the partial factors for the new code checking format, trial
values of partial factors are used in the new code format and~~ values and
corresponding (pr); are computed. These values are substituted in Eq. (9.59)
and the value of S is calculated. The process is repeated for different trial
values of Y;. Finally, the set of partial factors corresponding to the mini-
mum value of Sis taken for the new code checking format. This method is
illustrl" :eel below.
EXAMPLE 9.10 For the same problem in Example 9.9, determine the opti-
mal partial safety factors using Baker's approach (9.1 5). ~o = 3.5.
Solution As in the previous case, let us fix
"/R = 0.85 Yo =co 1.2

The problem is to find the optimal value of Yc

First select a trial value for 'h. se~y 1.3.
Using YR = 0.85, YD =-= 1.2, and YL = 1.3, determine ~ (as ex!)lained in
Sec. 8.3.3 and Example 8.9) for ee~ch value of a;. Find the corresponding
value of (pr); = !JJ( -{3),. Using Eq. (9.59), calculate S. A summary of the
calculations for YL = 1.3 are shown in Table 9.10. Repeat the process for
difrerent trial values of i'L and calculate the corresponding valt:'!S of S. YL
corresponding to the lowest value of Sis the optimal value of i'L. The opti-
mum value can be obtained by plotting YL versus S. The optimum value of
Sis 1.45.

TABLE 9.10 Stiiii!IUII'Y of calct•larions-E.ramp/e 9./0

f'r X 10- 1 (log Pr - log Pn)'ll'i Remark

'IL = 1.3
0.25 3.41 3.305 0.10 0.00233 Prt = <11( -3.5)
0.50 3.42 3.133 0.45 0.00753 = 2.326 X I o-•
1.00 3.19 7.094 0.30 0.07036
1.50 3.04 11.665 0. 10 0.04904
2.00 2.95 15.6j5 0.05 0.03429
E , ' 1.00 So= L' = O.J6J5

For 'IL ,,, 1.4 s = 0.0531

'IL = 1.5 s = 0.0547
'IL = 1.6 s '= 0.16108

I\ -
 257
:h he procedure of calculation of optimal safety factors for the load combi-
nation D + Lapt + Wm is same as explained in the previous load case:
+
D Lm. This is illustrated in the following example for the same member
RCC beam in flexure.
EXAMPLE 9. II Consider RCC beams in the limit state of collapse in flexure
under load combination D +
Lapt + Wm, as considered in Example 9.8.
The safety checking format is
(9.60)
al Taking Do = I,
td
9) Rl = ('Yv + a; YL + a, Yw) (9.61)
n YR
al
!l- La
where a; =
is Do
The function S defined by Eq. (9.48) becomes

S(Yn, 'Yo, 'YL, Yw) = f f {[. R"11 - 'Y o + a; i'L'Yn + OJ Yw] 2 IV;} WJ

(9.62)
The partial derivatives of S with respect to Yn, YD, YL, and Yw result in the
following equations:
~ 1: R011 w;lVJi'R -- 1: 2' ll'iW/Yn - .E 1: a;W;It'Ji'L - .E .E lV;WJOJYw = 0
ij lj i j " ij
in (9.63)
1g
.E .E R 11 w;w;a;Yn - E 1: W;IVJO;Yn --E .E OfWIWJ'YL - "'' ; 1: w;w;a1Yw
0
= 0
rle ij lj ij ij

or (9 .64)
'>'£ E E Rllw;W;a;'Yn - 1: E w;WJOJYD - E E a;OJll'iW;YL - .E ~ w;~v;a1 'Yw = 0 2
0
ti- I} I} I} ij

of (9.65)
It is to be noted that oSjoYn and oSjoyv will yield the same equation.
Here four variables are to be determined with three equations. Hence, the
value of one of the partial factor, generally Yv, is assigned or selected and
the other three factors are evaluated. The procedure is similar to that of the
gravity load case, D + L.
The computed values of Y; for the various values of Wn/ Do and Ln/ Dn arc
given in Table 9.11 for f3o = 3.5. The assumed weighting factors w; and II'J
J0-4 nre given in Table 9.12.
Since Yv is fairly constant as can be seen in Table 9.11 , Yv can be fixed .
The value of Yn is also fixed. Selecting Yv = 1.2 and Yn = 0.85, and using
values given in Tables 9.11 and 9.12 in Eqs. (9.63)-(9.65), the following
three equations are obtained:
0.725 'YL + 0.713 'Yw = 1.3748
0.7125 'YL + 0.5218 Yw = 1.1356
0.5218 'YL + 0.7127 Yw = 1.32
268

TABLE 9.11 Values of partial safety factors for RCC beams in flexure-Load
combination: D + Lapt +
Wm-Example 9.11

wn
L ~
n Particnl Dn
Dn fnctor --- - -- --- Remark
0.25 0.50 1.0 1.5 2.0

0.5 '>'a 0.791 0.836 0 .94 0.984 1.006 fto = 3.5

Yo 1.163 1.137 1.098 1.082 1.074 ~'Lopt/Ln
~ 0.239
'YJ. 0.3:!4 0.281 0 .240 0.229 0.224 L0 ~ 3 kN/m'
Yw 0 !)')(, 1.436 1.998 2.157 2.224 8I.aptfL 0 = 0.55
10 YR O.R54 0 .827 0.900 0.954 0.984
Yn I 123 1.121 1.077 1.0R2 1.074
yl. 0.114 7 0.456 0.290 0.255 0.241
Yw 0.884 1. 137 1.802 2.052 2.157
I .5 YR 0.932 0.900 O.R7 0.927 0.962

"n 1.095 1.097 1.095 1.082 1.074

Y~_ 0.911 0.772 0.397 0.292 0 2113

"w O.R2R 0.937 1. 527 1.921 2.076

T ABLE 9. 12 We ighting fact ors in fi<'I'C('I/tage fo r loacl co111hination D + Lnp 1+ W111-

Example 9. 11

Weighting facl(lr Wn
1.,
for~
L D,;
Dn Dn 0 .25 0 .5 1.0 1.5 2.0
(II'J) (w;)

tl.5 0.5 5 10 45 30 10 5
1.0 0.35 30 45 15 7 3
1.5 0.10 45 30 15 7 3

Using any two equations, three sets of y~_ and 'Yw can be obtained:
(i)i'L = 0.712 i'w ~= 1.2
(ii)i'J. = 0.512 Yw = 1.477
(iii) 'YL = 0.267 Yw = 1.656
Any one set or taking the average of the three values, i'L = 0.496 a:1d
Yw = 1.44 may be selected with i'R = 0.85 and i'D = 1.2.
This exercise of establishing the optimal safety factors for the various
values of target f3o can be done for various cases. A typical variation of the
upiimal values of partial safety factors for RCC beams in flexure for
\':lrious load combinations are shown in Figs. 9.5 , 9.6 and 9.7 (9.16). These
are applicable to Indian conditions. There are three values of Ln given in
the figures. The data used for live load is the one hased on the load survey
of office buildings. The lndian Standard Code (9.17) suggests nominal live
 211

1: Ln "'4kN/m2
J·O 2:Ln:lkNtm2
J: Ln :2·SkN/.m2 2
2-5

2·0

'fl 1·5
,,.
1.55
1·0 'fR :0·85
1'0:1·20
0· F"t 415

z.s
2

2·0
,-
YLt·S

1·0 YR :0-65
2.0
'Yo=1·20

5 0·5 Ff' 250

3
3 0
2·5 3·0 3·5 4·0 4·5 5·0 S·S
~
FIG. 9.5 Optimal values of partial safety factors for RCC beams in
flexure under load.: D + Lm

1 ad of 2.5 to 4 kN/m 2 for office building depending on the separate

torage facilities available. Office buildings are generally desig ned for a
nominal live load of 4 kN/m 2 assuming no separate storage facilities . The
)liS nnalysis of Jive load on office buildings indicates the mean value of Lm as
the 2.48 kN/m 2• However, the whole study has been carried out assuming
for !.her values of Ln equal to 3.0 and 2.5 kN/m 2 (i.e. P.Lm fLa = 0.827 and
ese 1.00), with a view whether it is possible to reduce the design loads for
in office buildings in India.
ICY
ive
 210

3·0
Y'0 =1·20
l"R:0·85
2·5 1: Fto 415 2
2: Fe 250

2·0
lw

l 5

1·0

0·5

I '
3·0 3·5 4·5 5·0 5·5

FIG. 9.6 Optimal values of partial safety factors for RCC beams in
flexure under load: D + Wm

9.6 SUMMARY OF RESULTS OF STUDY FOR INDIAN

STANDARDS - RCC DESIGN (9.16)
The development of reliability based design was illustrated for RCC beams
for limit state of collapse in flexure. A similar study (9.12) for Indian con-
ditions has been made for RCC slabs, RCC beams for limit state of collapse
in shear, and RCC columns for limit state of collapse under combined axial
load and uniaxial bending moment. A summary of typical resistance
statistics and results of reliability analyses of the RCC members mentioned
above have been given in Tables 9.5 and 9.6. While proposing a set of
pa rtia l sa fe ly fac t rs ~ r In d ian c nd iti ns, the sa me muterial facto rs given
in th e prese n~ cod e (9 . 14) have bee n re ta ined, ond hence th e co rresponding
re istance fac tor (yR), eva luo lcu by using nomina l va lues during the ta ti ti -
cal study of RCC members, has been kept constant. The variation in Yn
has been found to be sm all in all cases and can be considered almost con-
stant around 1.1. However, this va lue being very small a nd that the pro-
fession may not accept this, a value of 1.2 has been selected for 'Yn. The
 1: Lnz: 4·01cN/m2
2: Ln:3·01cNJm2
a : F•41S
3·.0 b: Ft! 250
YR" 0 ·8S
Yo='·2o 1o
2·S

2· 0

1·0 .

0·5

2·5 3·0 3·5 4·0 4·5 5·0 5·5

~
FIG. 9. 7 Optimal values of partial safety factors for RCC beams in
ns flexure under load : D + Lapl + Wm
n-
se ptimum values of 'Yt and "whave been fixed based on the above con'di-
al tioos for all the cases for the target reliability /3o. A set of curves for slab's
c:e nd beams in shear are given in Figs. BP·B5 (given in Appendix), connect-
!d ing optimal values of 'Yt. and Yw with /3o. table Bl is al so given in the
of Appendix for the opthnal values of '>'L and Yw for columns. Some of the
!n bservations and conclusions on safety factors for concrete design in Indian
1g conditions are given below:
(i) The yi~ld strength of steel has a significant effect on the statistics of
the strength ratio R/ Rn.for all RCC members.
Jn ~he ~~se of columns, the concrete grade also has a sig~ificant effect on
the statistics of the strength ratio in the region of compression faihne. In
262

the case of slabs, the effective depth also has a significant effect on the
statistics of the strength ratio.
(ii) The members designed as per the present code (IS: 456-1978 limit
state approach) have different safety levels under different design situations
and vary widely. For slabs, ~ varies from 4.2 to 4.8, for beams in flexure
from 3.2 to 4.7, for beams in shear from 3 to 3.8, and for columns from
2.9 to 4.6 . The safety levels of slabs and beams in flexure are higher than
that of beams in shear and columns.
(iii) Results of reliability based designs for slabs and beams clearly
indicate that the nominal live load of Ln = 4 kN/m2, used for the design
of office buildings is high. With this value of nominal live load, the load
factors obtained are low and may not be accepted by the profession. Hence,
it is proposed to use Lo = 3 kN/m 2 for the design of office buildings.
Although in column design it is not neces~ary to take the lower value of
Ln but, for the sake of uniformity, the value of Ln '= 3 kN/m 2 is suggested
for office buildings.
(iv) In all the cases of the reliability study of members and for all load
combinations, it is observed that the dead load factor, Yn, remains fairly
constant around a value of 1.1. This value being very low and that the
profession may not accept this , a value of Yv = 1.2 is suggested for all
load combinations.
(v) The values of resistance factor YR are taken as obtained by using the
nominal values of basic variables with Ymc = 1.5 and Yms = 1.15. This is
done so that the same material factors, suggested by the present code, can
be used.
(vi) A reliability level of 3.5 is suggested for the component failure.
(vii) The reliability based design for the load combination D+ Lm + Wapt
indicates that Yw has a very low value(< 0.1) and hence, this case tends to
the load combination, D + Lm case. The load combination, D+ Lm+ Wart,
is therefore not considered for the selection of partial safety factor s for the
gravity load plus wind load combination.
(viii) For Yn = 1.2, the target reliability f3o = 3.5, and for the resistance
factor YR corresponding to the material safety factors, 'Ymc = I .5 and
'Yms == 1.15 of the present code, tbe values of the live load factor and the
wind load factor to be used for different load combinations are given in
Table 9.13.
(ix) For slabs, beams in shear, and beams in flexure and columns, curves
or tables are also presented in Appendix B to choose the load factors YL
and Yw corresponding to the different reliability levels as desired by the
designer.
(x) In the case of columns, the quality of concrete (design mix or nominal
mix) significantly affects the partial safety factors for live and wind loads.
(xi) For columns, now-a-days at least M 20 concrete is used and the con-
crete is prepared based on the design mix proportions in major construc-
tions . Hence a partial safety factor for loads ranging from 1.4 to 1.8 for
different load combinations, as given in Table 9. 13, is suggested. In the
 the 'fABLE 9.13 Partial safety /actors /or dijfef'tttt components and load combinations
at ultimate limit states~.= 3.5, Yo= 1.2 and L 11 = j kN{m•
nit
)OS S. No. Load combination Component YR YL Yw
ure
om l. D + Lm Slab 0.85 1.4
1an Beam in flexure 0.85 1.5
Beam in shear 0.85 2.0
.rly Column
Comp* failure/Design mix 0.725 1.4
ign Tens+ failure/Design mix 0.80 1.!l
>ad 2. D + Wm Beam in flexure 0.85 1.6
iCC, Beam in shear 0.85 2.0
tgs. Column
of Comp failure/Design mix 0.725 J.5
Tens failure/Design mix 0.80 2.0
ted 3 D + Lat>t -i· wm Beam in flexure 0.85 0.45 1.4
Beam in shear 0.85 0.90 1.5
>ad Column
irly Comp failure/Design mix 0.725 0.27 1.5
the Tens failure/Design mix 0.8 0.24 1.8
all ---- -- ----- - ----
.Vure: *Comp ~ Compression
+Tens -- 'I ens ion
the
s is case of minor works where nominal mix is used for columns, higher safety
:an factors for live and wind load are to be used as suggested in Table B I in
Appendix B.
(xii n1e suggested values of' 'I L ami y w are fM the case when steel grade
Fe 41 i used . If steel grade Fe 250 i uset.l , thee values of loud factors
, \ ill ensure a slightly higher reliability than that conceived ~ ith the u c of
apt, t..el grade Fe 41 . If tbe • me reliability is to be achie cd, irrc pcctive of
the tee! grade, then lighlly I wer V<1lues of 'Y L an I 'Yo may be u cd when tee I
radc Fe 250 is used. H wever, lhe uifferencc i · very mnrginal. Hence, the
nee afety factors bnscd on steel gra.de Fe 415 arc finully s ugge ~ ted to be on
1nd lh safer side.
the (xiii) The proposed partial safety factors for loads, given in Table 9.13,
in will lead to more economical designs compared to the present vHiues given
in the code (9.14).
·ves (xiv) Even though the live load <latu on otlice buildings has been used in
;'1£ the study, the curve. or tables arc presen ted for va ri otiS ratio of /-1-Ln\ t Ln so
the that they could be used for any case of known f.LC.m t.n as ·uming the coeffi-
cient of variation of Ln does not change significantly.
ina! (xv) The Indian standard coder r RC Jc ign has not bee n yet calibrat-
;,
ed by the Indian Standard lnslituLion . IL is expe ted that they may usc the
on-
•B checking J'( rmat as foll owed by th · Briti h Standards. The LRFD
·uc-
r rmat has been u. cd in arriving at the results given in Table 9.13. However
for it is felt th at there ult will be (ln ly marginally atrectecl, and negligible,
the since, while arri,·ing at these values of Yv, 'YL, and Yw, the value of 'YR for
264

each case corresponding to the ·same material reduction factors and mate-
rial specifications as per present IS: 456-1978 (9.14) is used. Hence, the
present code format with new optimal values of Yn, Y1. and 'Yw may be
used.
The study has revealed many things for Indian conuitions. It has given
insight into (i) the present level of reliability available in RCC members,
(ii) how the safety factors vary with target {3, (iii) what is the reasonable
value of f3o, and (iv) how optimum safety factors could be fixed and how
these change for different failure criteria.

REFERENCES
9.1 CP 110 : Partl-1972. "Code of Practice for the Structur·al Use of Concrete, Pan I;
Design, Materials and Workmanship", British' Staudards lustitutiou, London,
t972.
9.2 NBCC, "National Building Code of Canada", Narioual Re.>carch Council 4
Canada, Ottawa, 1977.
9.3 CEB, "Common Unified Rules for Different Types of Construction and Materiul"
(3rd draft), Bulletin D' Information, No. /16-£, Comite European Du Beton,
Paris, 1976.
9.4 "Common Unified Rules for Different Types of Con sti'Uction and M~Jter·ial'' ,
Bulle/ill D'JnformatiOir No . 1]./ E, Comitc Euro-lntcrnational Du Beton (CEH) .
Paris. April 1978.
9.5 Galambos, T .V. and M.K. Ravindra, "Load and Resistance h1ctor !Jc-; ign tor
Steel", Joumal of Structural Div., ASC£, Vol. 104, ST9, Sept. 197R, pp. t: _i7-135J .
9.6 Ellingwood, B., J.G. MacGregor, TV GalambCJs, and C.A. Cornell, "Probability
Based Load Criteria : Load Factors and Load Combinatio n,", Journal of
Structural Dil·. , ASC£, Vol. 108, ST5, Ma y 1982, pp. 978-997.
9.7 Dayat·atnam, 1'. and R . Ranganathan, Statistical Analysis of Strl.'llift/r uj Cc>!lcre/e,
Building and Environment, Vol. II, Pergamon Press, 1976, pp. 145-152
9.8 Ranganathan , R . and C.l' . Joshi, "Variation s in Dimensions ot' RCC l\kmbcr\ ...
Journal of Bridgl' a11d Stmctuml Euginecr, Vol. 16, Sept. I \l86, pp . I· i 0.
9.9 David Arulraj and R. Ranganathan, "Reliability Design Critc1ia l'o r Slab< ',
llllemational Journal of Stmctures , Vol. 7, July-Dec. 1987, pp , 155-1 74.
9.10 Joshi, C.P. and R. Ranganathan, ··variations in Strength of Reinforcing Steel
Bars" Journal of rile bl.l'titutioll of Engi11eers (India), Civil Engg . Div., Vol. 68, May
1988, pp. 309-312.
9.1 J Ranganathan, R., "Statistical Analysis of Floor Loads in Office Building>", DS
a11d T Report No. 4/ 1/83/STP-1!1/3, Civi I Engineering Dept .. liT, Bombay . Oct.
1985.
9. 12 Ranganathan, R., "Reliability Analysis and Design of RCC Slabs, Beams and
Columns and Frames- Code Calibration", DS aud T Report No . .f/l/83iSTP-lff:'5 ,
Civil Engg . Dept. , liT Bombay, Sept. 1987.
9.13 Ranganathan, R., "Wind Speed and Wind Load Statistic' for Probabilistic
Design", Joumal of tire Justitution of Eug ineen (ludia), Civil [ngg. Div., Vol. 68,
May 1988, pp . 303-308.
9.14 IS: 456-1978, " Code of Practice f'nr Plain and Reinforced Conc rct~". lndiu11
Staru/ard> Institution, New Delhi, 1980.
9.15 CIRlA , R:ttionalisation of Safety and Serviceability Factors in Sttuctural Co .. ·~ "
Cnll5trucliou Industry Research and l11formarion Associutiou. Re{>ort No. '• •
London, 1977.
 265

9.16 Padmini Chikkodi and R . Ranganathan, "Partial Safety Factors for RCC Design"
te- lntemational Joumal of Stmcture1· , Vol. 8, July-Dec. 1988, pp . 127-149.
the 9. 17 IS: 875-1982, "Code of J>ractice for Structural Safety of Buildings, Loadings,
be Standards, J>art H- Imposed (live) Loads", Jan. 1982.

1en EXERCISE
:rs,
9.1 Determine the partial safety factors for the variables, the yield strength of steel,
ble
OW
fy Zr - D
,,- - M,.
dead load, and live load for the given limit state equation

L = 0
8
where I is the span, fy is the yield strength of steel and ZP is the plastic section
modulus.
It is given:
t I; Variable fy: I£ = 275 N /mm' cr ~ - 27.5 N (mm'
.on, Variable D: I£ = 11.55 N/mm• a = 1.155 N/mm'

1
Variable L: r.t ·•• 8.4 N/mm' a = 3.36 N/mm'
of
The nominal values of fy. D, and L are 250 N/mm•, 11 N/mm', and 12 N/mm'
ial" respectively.
~on,
(i) If the span is 8 m, determine the partial safety factors for flo = 3 assuming
[y and D are normal and L is Type I extremal (largest) .
.tl'',
(Ans: 0.955, 1.091, l.787)
:B) .
(ii) Determine the combined load factor. (Ans: 1.454)
(iii) Determine the partial safety factors with respect to mean values.
for (Ans. 0.868, 1.039, 2.553)
353. 9.2 For the problt'm in Exercise 9. 1, if the standard deviation of ZP is 60000 rnrn•
lity
and the mean deviation zero, determine the mean and partial safety factor for ZP
of'
and the combined resistance factor for flo~ 3. (All.\'. Yz = 0.968
p
Ycomb = 0.928, mean of Z P ~ 1.139;< 10' mm',
Y[y = 0.956, YD = 1.091, YL = 1.75)
rs''. 9.3 The limit state equation for the shear strength of steel beam is taken as
fyr.,d- VD- VL = 0
It is given:
.tee I
Variable !y: I£ = 275 N/mm• o = 27 .5 N/mm•
\'Jay
Nominal = 250 Njmm•
DS Variable V0 : r.t = 270.9 kN o = 27.09 kN
Oct. Nominal = 258 kN
Variable VL: 1.1 = 224 kN cr = 63.4 kN
and
Nominal = 361 kN
rt/5,
tw = 8.9 mm. If the standard deviation of dis 20 mm and the mean deviation 0,
is tic determine tbc partial safety factors of /y• V0 , VL• and d for 80 = 5. assuming
68. /y, V0 , VL and d follow lognormal, normal, Type I extremal (largest) and normal
respectively. What is Lhe combined resis tance factor? (lllls. j "( d 0.973 =
dian Yfy = 0.878, YD = l.lOS, YL = 2.229, Ycomb = 0 .855)
9.4 The shear strength of RCC beam is given by

R =fA .!!_ t bd
y sv s 6
yO.IIf. {vcu
J +58'-
,,
1}
266

I he limit slate equation is

R- In - I'L ·- 0

where Vv and I'L a1e shears due to dead and live load. It is given: {J' = 2.9,
b ~ 300 mm. d = 580 mm, s ·~- 100 111111.
Variable J;.: 11- · 320 N/mm• a ~c32N'mm'

Variablefeu: 1£ c 26.7 N/mm' a = 4.02 N/mm'

Variable VD: [L "c 94.5 k N a '~ 9.45 kN
Variable VL : fJ. ~~ 75.0 kN a ~- 21.~2 kN
The nominal vall1cs of/y• feu• VD and VL arc 250 N mm', 20 N mm', 90 kN and
90.69 kN respectively.
Determine the partial safety factors of /y. feu · VD ami I'L for Po= 5 if all
variable~
arc normally distributed.
(ilns. Yfy = 0.898, Yfcu " 0.97, YvD = 1.202 and YvL •= 1.588)
9.5 The ultimate strength of a RCC beam in shear is given by Eq. 9.32. Consider the
problem in Example 9.6.
(i) The statistics of shear force due to dead and live load a1e as follows:
Variahle Vn: (Nominal value = 79.5 kN)
:.t = 83.5 kN a = 8.35 kN
Variable JIL: (Nominal value = 54.41 kN)
1£ -~ 45.0 kN a = 12.73 kN
The statistics of feu and/y, and all other data are the same as given in Example
9.6. Determine the partial safdy factors for feu• fy· VD, and VL for {J 0 = 5.
(Ails. "IJy =- 0.9438; 'Yjcu = 0.9566; Yv "~ 1.269 and YL = 1.571)
(ii) If the code committee fixes the material reduction factor for concrete as 1.5,
determine the partial safety factors for /y, Vv, and VL for flo= 5.
(Ails. Yjy = 0.859; "lv = 1.267; YL '= 1.563)
(iii) If the code committee fixes the material reducation factor for steel as 1.15,
determine the partial safety factors for feu• VD, and VL.
(Ans. "lfcu = 0.817; Yv == 1.288 and YL ~" 1.673)
9.6 A column is subjected to combined axial load and bending moment. Under this
combined action, let the equivalent strength of column be R. The column is
subjected to dead, live and wind load. It is given:
Variable R/R 0 : (normal)
!' = 1.22 8 = 0.14
Variable D/Dn: (normal)
" = 1.05 8 = 0.1
Variable Lap1/Ln: (lognormal)
p. = 0.179 8 = 0.55
Variable Wml W 0 : [Type 1 extremal (largest)]
1£ = 0.804 8 = 0.334
Study the variation of the partial safety factors for various values of W0 /D0 = 0.5
0. 75, 1.0, 1.5 and 2 for flo = 3 and plot the same. Assume L 0 /D 0 = 1.0.
9. 7 The limit state equation of a structural component subjected to dead and wind
load is given as
 267

, The statistics of the variables arc given below.

Vuriable R! Rn: 1.1 = 1.355 tJ = 0.225 (normal)
2.9, Variable D/Dn: 1.1 = 1.05 tJ = 0.105 (normal)
Variable W/Wn: 1.1 = 0.804 tJ = 0.269 (Type 1 extremal (largest)]

l,lot the variation of..the ~artlal safety factors 'YR· '~'D· and Yw with W" nn ranging
from 0.5, 0.75 . 1.0,I.S and 2 for IJo = 3.5 . .Determine the optimal values of '~'R• "'D
and Yw using the method adopted by Baker (9.15) assuming suitable weighting
factors for the occurrence of each Wn!Da ratio.
1,1 , ~ The limit slate equation for an axially loaded short column is assumed as:
and
0.67 feu Ac + fy A8 - D- L = 0
r all The area of concrete, Ac• is 113000 mm•. It is given:
Variablt;4u :(normal)
588)
"= 26.8 Ninim• 6 = 4;02 Ntmrni
:the
Variable/y: (normal)
1.1 = 4.69 Niimi1 2 a = 46,9 N/mm•
.Dead load D: (Normal)
420 kN
I''"'" o = 42 kN
Live load L: [Type I extremal (largest)]
1.1 = i66.8 kN tr = 47.2 kN
If the nominal values of feu• !y• D, and L are 20 N/mm 1 , 415 N/mm•, 400 kN ancl
269 kN respectively, determine the partial safety factors for variables for Po = .5 •
.571) (1111'. >'feu = 0.458; 'Yjy = 1.044; Yo = I. IS; >'L '~ 1.073)
1.5,

.563)
1.15,

.673)
this
on is

~ 0.5

wind
 10
Reliability of Structural Systems

10.1 GENERAL
We have so far studied the reliability analysis and design of structural
components. The code calibration based on component reliability was
also introduced and illustrated in Chapter 9. But a structure or a structural
system, viz. building, bridge, offshore platform, water tank, etc. is built
up of many components (elements). The capacity of a structural system
will depend on the capacities of its components. The behaviour of the
system is probabilistic as it depends on the performance of its componenrs
whose behaviour is random . Civil engineering structures are invariably
a kind of system. Information is available only on the statistical per-
formance of components. With this information, the reliability of the
structural system must be determined. A structural system may have
(, several failure modes. These failure modes are to be identified, modelled,
and combined to determi11e the system reliability. Hence, the reliability of
structures/structural systems of multiple components and with multiple
failure modes is to be considered from the system point of view.

10.2 SYSTEM RELIABILITY

One of the important applications of probability theory is the evaluation
of the reliability of a system which is made up of components with known
reliabilities. The reliability of a component is the probability of its satis-
factory performance against the purpose for which it has been designed.
Block diagrams are used to demonstrate the computation of the reliability
of a system. Systems are classified basically into three groups as given
below:
(i) series system
(ii) parallel redundant system
(iii) mixed system

10.2.1 Series System

The term, commonly used in the field of electrical engineering, is easily
understood by everyone. In this system, even if one component fails to
function satisfactorily, the whole system will fail. Therefore, a series system
performs satisfactorily only when every component works satisfactorily,
 269

The block diagram for this system is as shown in Fig. 10.1 and the reliabi-

0 lity of the system is calculated as explained below:

Let
A1 = the event that component i works satisfactorily
tS Pss = probability of survival of the system
pr. = probability of failure of the system
Pn = l - Prs

2 --------8-
FIG. 10.1 Series system

ual As every component should func;Jion satisfactorily for the system to be

.vas reliable,
ual
uilt Pu = P(A1nA2n . . . nAn) (I 0. I)
:em If the events Ai are independent, the above equation simplifies to
the Pss = P(A1)P(A2) . .. P(An)
:nts II
.bly II (l -- pr;) ( 10.2)
)er- i-1

the where Pfl = the probability of failure of the component i, and " = the
ave number of components.
led , The model is also called the "weakest link model".
of In the case of structural systems in civil engineering, the values of Pr1 are
iple very small. If Pfi <( 1, Eq. (10.2) can be rewritten as
II

Pu~l-Eprt (10.3)
1-l
II

and pr. ~ E Prl (10.4)

:ion 1-1
)WO
10.1.2 Parallel Redundant System
1tis·
1ed. In this case, the system survives even if one component has failed. The
ility system fails to function satisfactorily only when every component of the
iven system has failed to function satisfactorily. The block model diagram for
the computation of reliability is shown in Fig. 10.2. The reliability of the
system is given by
Pu = 1 -Pta
~ 1 - P(AfnAin .. . nA!) (10.5)
where A~ = the event that component i docs not function satisfactorily. If
events A~ are independent, Eq. (lO.S) simplifies to
lSily p.. = 1 - [P(Ai)P(Az) ••• P(A:)]
; to
item
rily,
,. , /Iff
- 1-A no.6>
 270

FIG. 10 .2 Parallel redundant system

In structural engineering, this system may be referred to as a parallel

system with n perfectly ductile elements.

10.2.3 Mixed System

This is a combination of series and parallel redundant systems. The block
model diagram for the computation of the reliability of a mixed system is
shown in Fig. 10.3. This is visualised to consist of subsystems St and S2 as
''·· shown in Fig. 10.3. St is a series system and S2 a parallel redundant system,
and subsystems St and S2 are connected in series. For this mixed system
to survive, each subsystem should survive under the given conditions . Hence
the reliability of the system is given by
p,. --=- P(El n £2)
r---

-+EJ I

.f L _ _ _ ____ _ J

FIG. 10.3 Mixed system

where £1 = the event that subsystem I functions satisfactorily and £2 =the

event that subsystem 2 functions satisfactorily. Knowing how to compute
the system reliability of the series and parallel redundant systems, the
probability of the survival ..>f this mixed system, shown in Fig. 10.3, is
given by
Pss = P(Et)P(E2)
= (I - Pfat)(l - Pral) (1 0. 7)
 271

where pr,; is the probability of failure of the 5ubsystem i. It has been

assumed that events A; are statistically independent.
• AMPLE 10.1 Calculate the reliabilities of the systems shown in Figs. I 0.3
nnd I0.4 assuming the performance of components is statistically indepen-
dent. Compare the results. Given:
Pri = 0·1

FIG. 10.4 Block model-Example 1 0 1

(i) The reliability of the mixed sy tcm (Fig. 10.3):

The reliability of the subsystem I, using Eq. (10.2), is
P(E,) ""' (I ·- 0. 1)(1 - - 0.2)
!lei = 0.72
Using Eq. (10.6), the reliability of the subsystem 2 is
P(Ez) = I -- - PnPr4
ICk ""~ [I - (0.3){0.2)]
I IS ~-' 0.94
as
lienee the reliability of the mixed system is (Eq. 10.7)
~m,

em p., = P(E,)P(E2)
1ce = (0.72)(0.94) = 0.6768

(ii) The reliability of the series system, shown in Fig. 10.4, is

p,. =(I - 0. 1)(1 - 0.2)(1 - 0.3)
= 0.504
When the reliabilit-ies of the two systems are compared, it can be seen
that at the cost or an additional redundant component 4, the mixed system
is more reliable than the one shown in Fig. 10.4.
~AMPUl 10.2 Con ider a nuclear power plant designed for a tevel of
earthquake intensity. At this particular level of the earthquake intensity,
the controlled shutdown of the reactor depends on the functioning of the
control s stems; the cooling systems and the primary containment vessel.
• here are two redundant control systems, two redundant cooling systems,
the and a single primary containment vessel with two components A and B in
•ute series.
the (a) Draw the block model for the computation of the reliability of the
•, is plant with respect to shutdown at the given earthquake level.
(b) If it is assumed that there will be no major accident if either the
shutdown is controlled or the reinforced COJlcrete secondary containment
vessel C performs properly, model the total system with respect to the
).7)
major accident reliability.
 272

Solution Let
pr; = probability of failure of the component i
(a) The block model diagram for the computation of reliability is shown
in Fig. l0.5(a). The reliability of the plant with respect to shutdown at the
given earthquake level is
Pss =(I - Pf1Pf2)(1 - PnPr4)[(1 - {JfA)(I -- pro)]
Control system

Primary contaimeont
Vf'SSf'l

Cooling system Pr~mary conlaimeont

'·· Vf'SSt'l
r - - - - - - - - -,

I
L --------J

R c c Vt'SSt'l
c
(b)
FIG. 10.5 (a) Block model for case a and (b) block model for case b-
Example 10.2

(b) In this case, there will be no major accident if either shutdown is

controlled or the secondary concrete containment vessel C performs satis·
factorily. The block diagram is shown in Fig. 10.5(b). The reliability of the
system with respect to no major accident is
Paa = 1 - [1 - {(I - prtpn)(l - prJPr4)(1 - PrA)(l - prs)}](prc)
 273

MODELLING OF STRUCTURAL SYSTEMS

In
hilt Ill can be considered as a system and can be modelled into any-one
he lh tllrec basic systems, depending on the physical behaviour for comput-
11. reliability. The modelling of a few structural systems for the com-
1 t on of reliability is explained in the following sections.

mply supported beam is subjected to a load as shown in Fig. 10.6, the

of the beam occurs when the strength of the critical section (the
1 111 subjected to the maximum moment) is less than the external load.
I It beam can be considered as a system with one component (critical
II n) only. It is similar to a case of a tension member subjected to a load
hu\ n in Fig. 10.6(c).
L

(al Beam (b) Model (c )Tension Member

FIG. 10.6 Single member single load condition

If the beam shown in Fig. l0.7(a) or the tension member, shown in

I t~ I 0. 7(b) is subjected to several independent load conditions, L1, L~ . ... , Ln
,1 ndependent repetitions of a single load. the reliability model is a series
since for the beam to be reliable, the critical section should survive
t 111,
, 1 be reliable under each load. The block model is shown in Fig. I 0. 7(c),
h ·rc the Section A under each load is imagined as a component and is
1l n ted as A;.
'A 'IPLE 10.3 Consider a steel tension member, shown in Fig. !0.7(b), sub-
1 n ll to m independent repetitions of load L. It is given that the means
111\ l standard deviations of the resistance of the msmher R and l. nrc
/'II -· 50 kN "R - .'i kN
25 kN
/LL """" "'· ' , 12 k N
m -= 5
is ( ' mpute the reliability of the member if Rand L are normally distributed.
is- ,\olution Consider the member as a system subjected to m independent
he r petitions of L. The block model for the system to comrute the reliahility
will be a series system. The reliability of the system is
p,. = I(I --· pr l I'"
 274

j
Lm

Lz
r·J
Lt

I L2

j Lt
A
Lm

lal Beam lbiTension Member

-----&-{3- -{A~--------- -a-

lc l Block ModPI
FIG. 10.7 Single member m load conditions
'··
where Pr is the probability of failure of the member under l.. The value of
is comruted as given below:
{Jr

Pr = P[R < Ll
= P[(R - L) < Ol
Since R and r are normal, ming Eq. (6.16), we get
pr ~~ 1'>( -- {3)

---, 1'> [ ..... 2f4f. - -· /l-R

2 ]

(aR + aLJ1;2
Suhstituting the given data, we have

25 - 50 ]
!'t = c/J [ (52+ J22)1/2

= c/J(- I. 92) = 0.02743

The probability of survival of the system under repetition of L foe five
times is
Pss = (J -· 0.02743)5
=.-, 0.87
It is to he noted that the reliability of the system decreases as m Increases.
 27&

structural system consists of girders, piers, and abutments, as

in Fig. I 0.8(a). For the reliability of the system of piers, each pier
function satisfactorily, Hence, the reliability model for the system of
\ Ill be a series system. Similarly, for the reliability of the system of
nts each abutment should be reliable under the given loading con-
Hence, the reliability model for the satisfactory performance of the
of abutments is a series system. Likewise, for the reliability of the
.;b•""'m of girders, each girder should be reliable under the given conditions.
i~l oftiJn~;c the reliability model for the system of girders is a series system.

er

(a) Bridge System

G1rder system Pier system Abutment system

1e of
--- -- ---------l r - - - - - - ---, - --- -- --,
I I I I
r I I

f I I I
I I I I
I L _ __ _ _ _ __ J L ___ _ _ _ j
-- - -- -- - -- _ ......)

\ (b) Block Model

3
FIG. 10.8 Modelling of a bridge system

r the whole bridge system to survive under the given loading condi-
li n, each subsystem, i.e. system of girders, system of piers, and system of
t)11tmen ts, should survive. Hence, all the three subsystems are to be con-
no ted in series to compute the reliability of the system. The block model
five tl gr m is shown in Fig. 10.8(b).
AMPLE 10.4 Compute the reliability of the bridge system, shown in
fig. 10.8(a), having four piers, five girders and two abutments. The probabi-
1ty of failure of each pier, girder and abutment is Io-4, 10-s and Io--r,
r pectively. Compute the reliability of the bridge system .
LSeS.
 276

Solution The probability of failure of the pier system is (Eq. (10.4)]

.
(pr)pier = lJ Prt = 4 X I0- 4
,_,
The probability of failure of the girder system is [Eq. (1 0.4)]
5
(prkir~er =
t~l
1: Pri = 5X JQ-S

The probability of failure of the abutment system is [Eq. (10.4)]

2
(pr)Abut = lJ Prt = 2 X J0- 6
t~t

Hence the probability of failure of the bridge system is

(pr)Bridge = (4XI0- 4) + (5x J0- 5) + (2X J0- 6)
= 4.52X J0-4
The reliability or probability of survival of the system is
{lss = J - 4.52 X I0-4

10.3.4 Truss System

Consider the truss shown in Fig. I 0. 9(a). It is subjected tu a load Las shown
in the figure. This is adeterminate structurewhich is considered ~1s a system
having six members (components). For the truss to perform satisfa<.:torily,
i.e. to be reliable under the given load L, every member of the truss should
\ perform satisfactorily, i.e. should carry its load safely. Hence, this truss
can be modelled :1s a series system with Jive components, as shown in
Fig. \0.9(b), to compute the reliability.
Jh

JJL
2
L

(a) Truss System

(b) Block Model

FIG. 10.9 Modelling of a truss ~ystem
 117

10.5 The truss shown in Fig. 10.9 is subjected to a random load

parameters
~L = 25 kN U£ = 12 kN
pnrnmeters of resistances of members are given as
~RI = 50 kN aRI = 5 kN i = 1, 2, ... ; 6
/-LR1 = 60 kN aR1 = 6 kN
te the reliability of the truss system assuming resistances of the
are independent, and also that resistance and load are independent
Rt and L are normal.
h block model for the computation of reliability is shown in Fig. 10.9(b).
the given load, the forces developed in the members are given in
I .9(a) .
calculating the system reliability, the reliability of individual com-
nt · must be computed. Since members l, 3, 4 and 5 carry the same
d and their resistances are the same, the probability of failure of these
111'1mi:>Crs is the same:

Pfl = rp [ ~L - ~R l ]
(a L + <1Rt)l f l
hown 25 - 50 ]
rstem = rp [ (52 + 122)1 /2
orily, = 4>( -1.92)
lOU!d
truss
= 0.02743
m in pn = PC4 = prs = Ptt = 0.02743
he force in tqe members 2 and 6 is Lv'T/2. Hence, the mean value and
lundurd deviation of the force in member 2 are (VT/2)11-L and (v3/2)aL
peclively. Hence,

pn
·

--~L- ~Rl
v'3
~ pro ~ <P [ {('/3 )' + '}"'
- - C1L
2
C1R2
l
21.65 - 50 ]
= cp [ {(I 0.392)2 + 52}1' 2
= <I>( -2.458) = 0.006986
Simi larly ,

}--
278

= c <1J (~)
21.63 = f/J( - 0 77 1 )
° -

= 0.224338
Using Eq. ( 10.2), the probability of survival of the system is
i
flss = 11 (I -- [1r;)
i=t

= (I - 0.02743)(l -- 0.006986)(1 - 0.02743) X

(I - 0.02743)(1 - 0.02743)(1 - 0.006986) X
(I - 0.22438) J
= 0.68625
· nsider the tru e silO\ n in Figs. IO.lO(a) and lO. IO(b) in which the
number of members arc 3 and II respectively. Reliability of these trusses
can be computed as above and arc 0.948 and 0.148 respectively. Including
the result for Lhe same type of trus · with sevl!n member~ . it can be observ-
ed that as the number f members increases, the reliability of the
system de rca es when the performance f the members are statistically
independent .

(a) (b)

FIG. 10.10 Trusses

ln the ~t sc or tru s systems, for a given number of members, if the resis·

t :~ncc!sof members are c rrelat ed, the reliab ility of the system increases
willt th e increase in the c rrelation c em ient between member performances
resistanC!!S or memberS).

10.3.5 Indeterminate Beam

Consider a fixed steel beam shown in Fig. 10. I I. This is a redundant, per-
fectly ductile structure. In this case, the failure of the structure does not
occur if one section yields; failure occurs only when a sufficient number of
sections have yielded to form a collapse mechanism. In the case of the fixed
beam shown in Fig. 10.11, the beam fail s only when the critical sections I,
2 and 3 (positions of maximum moments) have yielded. Hence, the given
beam can be consi lered as a parallel redundant system, the block model
diagra m of which is sh wn in Fig. IO.Il(b).
In the case of redundaut structures, the reliability of the system increases
as the number or redundant components increases if R; are statistically

_]
 279

J,
I
T !
2
21
3
J 2
~----- 1

lbl Block Model

FIG. 10.11 Modelling of a fixed beam

:h the I 1l p •ndent. It can be proved also that in the case of a redundant parallel
russes I 111 , if the resistances of members are correlated, the reliability of the
uding It Ill decreases with the increase in correlation.
Jserv-
. the 11.,\.6 J.'rame Structural Systems
ically I 1 IIII C structures are highly redundant structures. In this case, the failure
•I single section (component) does not result in the failure of the frame
11
I tem). Assuming a perfect ductile structure, the frame fails only when a
ullt icnt number of plastic hinges are developed to cause a collapse
111 hanism . Again, there may be a number of possible collapse mechanisms
n u frame structure. These possible collapse mechanisms are to be synthe-
lml and the system failure probability is to be computed .
A failure mode, i.e. a collapse mechanism is composed of component
( '(•tion) failure events that are in parallel. For a failure moue to be formed,
1 ry critical section in that mode must have failed. Hence, to compute the
1 hubility of the frame under a particular failure mode, the critical sections

resis- In 1hat mode are to be connected in parallel. For the frame to be reliable,
~~~ hus to survive under all the possible failure modes. Hence, to compute
reases
.ances · . the . system reliability, these parallel subsystems are to be combined as a
-"rics system. The block model for the computation of the reliability of a
frume structure is shown in Fig. I 0.12(e). It is clear that this is a mixed
•yatem. The individual failure modes may be correlated because of common
load and resistance variables. There may be correlations between single
' per- elements in the same failure mode. The system reliability depends on
!S not
(I) topology, (ii) post failure behaviour of components, and (iii) correlation
er of
chural:lcristics of different variables and different failure modes .
fixed In the case of ideal plastic structures, each collapse mode, called collapse
>ns 1,
mc~.:hanism, (i.e. limit state) can be represented directly by an equation in
given lt:rms of the plastic moments of hinged sections in the mechanism and the
nodel length factors multiplied by loads (10.1), using the mechanism method of
nnalysis ( 10.~). Hence, the safety margin equations for failure modes can be
reases Llirc~· tly written and the reliability of a frame under each failurl! mode can
ically
 280

w w
5
H 6
6

l r----- a -----!
7

(al Frame lbl Mode 1

(c) Mode 2 (dl Mode 3

\I
I
I

~· Mode 1 Mode 2 Mode 3

(elBiock Model
FIG. 10.12 Modelling of a frame system

be calculated. Then the structure is modelled as a series system with the

failure mode as components, and the system reliability is determined. Tbis
i illustrated in the following example.
'·/~AMfllE 10.6 Consider the rigid steel frame shown in Fig. 10.13(a). It is
gi ven th at
 281

Ia I Frdme (b) Mode 1

., w L .s
~

(dl Mode 3

lei Mode 4

1-------10~-
Mode 1 Mode 2 Mode 3 Mode 4

(f)Biock Model
FIG. 10.13 Frame, failure modes and block model-Example 10.6

vith the P.Ml = = P.M6 = P.M1 = 490 kN Ill

P.Ml

U!lfl = CM2 = aM6 =~ U!1f7 = 73.5 kN m

d. This
{LMJ = #lM4 = fi-MS =653 kN Ill
.
a). It is aM3 = aM4 = OMs = 97.95 kN m

P.w = 446 kN aw = 69.9 kN

 282

\1 ill'IC A/; is the plastic ll10111ent capacity or section i. Compute the system
•cliability or the fr:1me assuming all M; arc independent, M; and W inde-
pendent and ;ill 1·ariablcs M; and W normally distributed.
Be:::ause of •.he random behaviour or the resistances of \'arious critical
sections and load, the frame may fall under failure modes having hinges at
2, 4 and 6, or 3, 4 <IJH.l 5 or 3, 4 and 6, or 2, 4 and 5 which arc shown in
Fi~s 10.13(b)-10.13(d.
The virtual wm 1-: method of plastic analysis (10.2) is used to determine the
n:,iqa!1cc or tlie r, am~. :u~d action at collapse for each mode. For the safety
ol' the fr:1111e under failure mode I [Fig. 10.13(b)l:
M2tJ + 21\148 + Mc.O > 3 JYO
11lll:rc 0 is the vi1tual rotation at section 2. Hence, the safety margin is
Z = ilh + 2M 4 + M 6 - · 3 IV
·1 he probability or survival of the frame under a mode is P• ~ P(Z ~ 0).
/\s 7 is a line:11· function of the variables M; :tnd W, we have

c.= 490 -: 2(653) + 490- (3)(446) = 948 kN 111

1
az =
~
l(lfa;ir2 -l- + (1)2aM6
1
+ (3)2o..iv]lll
(2) 2ai;r4
~

=- L73.5 + (4)(97.95)2 + 73.5 + (9)(69.9)2]1 12

2 2

= 305.2 kN m
~ '
/\s Z is a linear function of the independent, normally distributed variables,
'· ~is also a normal variable. The probability of failure of the frame under
the mode I is

{Jfl = P(Z < 0) = f1> (- P.z)

az

--- (/> ( - 948 )

-- 305.2
== <!J(--3.106) = 9.35 x I0-4
Similarly, for other failure modes shown in Figs. 10.13(c), IO.l3(d), and
I0.13(e), the probability of failure of the frame under each failure mode can
be calculated; the calculations are shown in Table 10.1. To compute the
probability of survival of the system under all failure modes, a block model
is drawn connecting all modes in series as shown in Fig. l0.13(f). Assuming
all failure modes are statistically independent (Note: this is not true as all
Z are correlated as seen in Table 10.1) and using Eq. (I 0.2), we have
4
p,. = II (I - pr;)
i-l

= (I - 9.35 X I0-4)(1 - 3.97 X to- 5)

'~ (I - 1.85 :/ w- 4)(1 J.85x w-4)
"' 0.9986553
 283

ystem T:\llLE 10.1 Rc/iahility allal,l'.li.•· of tl:c portal /miii('-E.'.'olllplc 10.6

in de- p
P.ZI Gz;
(kN m) (kN. m)

,ges at (Fill. IO.IJ(bl] !If,- 2.11, -:- M,- jw 948 305 ..! 9.35 >: to ·• 3.106
wn in If'ig. I 0.13(cl] 11-!3 +2M, '- Ms- 3W 1274 318.7 3'.17x to··• 3.997
[rig. to.t3(d>l !lr,..!..2.H,-, Me- 3W II II 312.0 t.ssxto-• 3.561
ne the [fig. IO.IJ(d] M, .!. 2/lf, -' M,- 3W ]]]] 312.0 t.ss x to-• 3.561
safety lllille hounds: 9.35xto-• ~Prs ~ 13.37;:Jo--•

rh probability or failure of the structural system is

is pr, "--= I - 0.9986553
1.3447 :< IO-J
;:.;:: 0). 4
cN.u: flfs ~ :E Pr;)
i=l

1ft • BOUNDS ON SYSTEM RELIABILITY

IO,.C.l Introduction
In the previous problem, it has been assumed that Z; are statistically
iables, tulopendent during the computation of the reliability of the system. It is
under nb i us when the equations of Z; are examined. (Table 10.1), that Z; are
1 lll'related, ns the same random variables appear in the equations. For
llllple, if z, and Z2 are considered. /114 a11d W appear in both the equa-
litn . Hence, Z1 and Z2 are correlated .
I
II

Z; = l' a;X; (I 0. 8)
i-1

ml
1), and
de can " (10.9)
Zk "'" 1: b;X,
ute the i-l

model (u1 and bk are constants)

;uming h• ~ovariance between Z; and Zk is given by
~ as all
• II

Cov(Z;, Zk) = E a;b;aL { 10.10)

i-1

l'hc correlation coefficient bet\veen Z; and Zk is given by

Pz; ~k = Cov(Z;, Zk) ( 10.11 a)

' a 7.; az;:

• rrelation coefficients can also bt: calculated using directional cosines r1.,
\
 284

uireclly obtained during the computation of rcli~tbility illdcx using Level 2

method. The wrrdation roetricient between Z and Zk is given by

f'L;, zk = "
l-' IJ."'l.kt
.J-1
'II
·:he abore equation and Eq. (IO.lla) arc same. This can be easily verified
for linear Eqs. (10.8) and (10.9) for Z; and Zk. For nonlinear equations,
directional cosines evaluated at the design point, Cl.~, on the failure surface
arc used. That is, in gener:1l,
n
Pz;, Zk = 1;· (10.11 b)
t-1

The probability of survival of the system is given by Eq. (10.1). That is,
p,. = P[All Z; > 0]

=
f <h
0
... J"'
()
lzl, z2, ... , z, (zt, z2, ... , z,) dzt, dz;;, ... , dz"
u~rotd

Where /zp z 2 , • • . , Zn (zr, Z2, . . • , Zn) is the n-dit!ICOSiona! joint probabi-

lity density function of Zt, Z2, ... , Z,. The joint probabilities are gene-
rally not known and the computation of n-fold integration is very difficult,
and may not be possible. Therefore, the above equation is simplified by
certain assumptions to derive bounds on the probabiiity of failure. It is
generally not possible to compute the unique value of the reliability of the
system and therefore, the reliability of the system is specified by its bounds.

10.4.2 Simple Bounds

\ Cornell (10.3) has established simple bounds on the reliability of structural
systems subjected to n failure modes and m load conditions. The assump-
tion that all fatlure modes, i.e. Z;, are perfectly correlated yields the upper
bound as
fJss = I - maxP(Z; ~ 0) = I - maxpr;
! i

The assumption of all failure modes to be statistically independent yields

the lower bound as
n
Pss = fl (I - Pfi)

Hence, the reliability of the system is bounded by

n
11 (I - pc;) ~ Pos ~ I - max pr; (10.12)
;~t I

If pr; ~ I, the bounds on P•• become

I - " pc; ~ pss ~ I - max pr.

}; ( 10.13)
i•l i

The bounds on the probability of failure of the system can be written as

JJ

max pr; ::::;; pr, ::::;;; }; pr; (10.14)

j i:oal
 286

~vel 2 II the system is subjected to several m load conditions, the bounds on p11
1111 p(, are given by (10 .3)
n 11

I - I: I: Pru ~ Pss ~ 1 - max Pfii ( 10."15)

i-1 ,_, lj

:rifled
tions,
"'" n m
uface max PfiJ ~ pr. ~ I: I: pr;i ( 10.16)
I) ~ 1~1 J=l

l.llb)
here PriJ is the probability of failure of the frame under mode i and load
'tiiiJ IIion j. These bounds are very wide for practical pmposes.
is,
10.4. Narrow Bounds
Th assumption of perfect correlation or, no correlation between failure
, dz. m d s, is not proper. The modes are usually positively correlated. The
·c rr lation coefficients between modes can be calculated using Eqs.- (To.-10)
·babi- nd (I 0.11 a). Ditlevsen (I 0.4) has developed narrow bounds for the structural
gene- tem failure probability through indicator function algebra . The lower
ficult, und on pr. is
:d by n

. It is prs ~ P(Zt < 0) + l-2

I: max {P(Z, < 0)
·f the ,_,
•unds. - ,_,I: P[(Z; < 0) n (ZJ < 0)], 0} (10. 17)

11 I the upper bound is

ctural n n
mmp- Pr• ~ I: P(Z;
J~l
< 0) - I:
1-2 , } <. 1
max P[(ZJ < 0) n (Z, < 0)1
upper ( 10.18)
Let E1 = (Z; < 0)
E, = (ZJ < O)
yields · Then the above Eqs. (I 0.17) and (10.18) become
n 1-l
Prs ~ prt + I: max [pfl -
1~2 ,_,
I: P(EI n EJ), 0] (10.19)
and
n "
10.12) Pt• ~ I: PrJ -
1-1
I:
1-2, )<I
max P(Et n E1) (10.10)

The joint probability, P(E; n EJ), may be approximated as follows. For

lower bound (Eq. 10.19)
10.13)
P(E1 n EJ) = P(A) + P(B) (10.21)
as upper bound (Eq. 10.20)

10. 14) P(E1nE1) -max [P(A), P(B)] (10.22)

 286

where (10.23)

( 10.24)

where

EXAMPt.r 10.7 For the same portal lrame in Example 10.(,, compute the
simple ~md narrow bounds on the prob<tbility of failure of the frame.
Simple bounds:
For !he possible four failure modes, the probability of failure of each mode
has already been calculated and given in Table 10.1. The bounds on the
probability of failure of the system, using Eq. (I 0.14), are
4
Upper hound = Z pr;
i=J

= (9.35 ~< JQ- 4 ) +· (3.17 X 10- +· 5) (1.85 X JQ- 4 )

+ (1.85 ~< J0- 4)
13.37 >~ JQ- 4
Lower bound = max. pr; = pn
i

= 9J5 X IQ-4
Hence, the bounds on [Jrs are
9.35 X I0- 4 ~ Pfs ~ 13.37 X J0- 4
\ Narrow bounds:
!
The failure modes are first renumbered, or ordered, in the descending order
of Pfi values. Hence, from Table to. I,
Mode 1 : Z1 = Ah + 2M4 + AfG- 3W ~ = 3.106
Mode 2 : Z2 = MJ + 2M4 + M6 - 3 W ~ = 3.561
Mode 3: ZJ = M2 + 2M4 + Ms- 3 W ~ = 3.56i
Mode 4 : Z4 = MJ + 2M4 + Ms - 3 W {3 = 3.997
The correlations among failure modes (that is safety margins Z; and Z1) are
next computed.
Using Eq. (10.10),
Cov(ZJ, Z2) = (2)(2)oM~ + (l)(l)a,J6 + (-3)(-3)afy
= 2
4 X 97.95 -j- 73.5 2
+ 9 X 69.9 2
= 87753
Using Eq. (IO.lla),

87753 --- =
----- 0 922
305.2 X 312 '
 287

th e ~.:orrcl~ltion bct,\ccn z, ami ZJ is

flz z = ( l)(l)a.lt~ + (2)(2)u.H~ + ( 3l{ - 3)r,f,.

I' 1 0"71 a/.J

= ?3 .52 + 4 X 97.95~ -1- 9 '< 69.9 2 = O ,

05 ._X3)2 .9- 2

\ lmi larl y the correlation between othe r pairs of z,z; ~:a n be computed. They
II.'

Pz 1 • z 4 = 0.847 P72 , z 3 = 0.846

P7 2 , z 4 = 0.925 Pz3 • / 4 = 0. 925

l·or the calculation of bounds, bounds on joint probabilities, P(E,Ej), are

(1, be comruted first.
Bounds o;J P(E,h):
tJc;in~ Eq. ( 10.23),

P(A) = cf>( -f3rl (h [ - - fl2.

(I

= f[J(-3 106) cf>[- 3.561- 0.922>~3.10() ]

· c1 - o:922 2)'' 2
= cf>(- 3. I 06) ([)( - J.R02)

= 0.334 /~ w- 4
. '·' . . . [ [j , - P12f12]
Cf.(-:-~2) ([), ·7"" (1-=. p~ )172
P( 8) =
2
= ([)( _
3.56 !) rt>[ _ 3. '106 - 0.922 X 3.56 1]
( I --· 0. 922 2) 112

= f1l( - 3.561) f1l(0.458)

= 1.25 X tQ-4
\Lower bound on P(Et E2) = P(A) + P(B)
= (0.334 + 1.25) >
no-4
= 1.584 X J0-4
Upper bound on P(E1E2) = max [P(A); P(B~]

= 1.25X 10"'4

Since Pl2 = ~r and fh = fJ2,

P(EtE3) = P(E1E2)
Bounds on P(E1E4):
 288

P(A) = t!>( __ 3 I0 6)
.
tt>[- 3.997(I -- .847 X 3.106]
0.847 2) 112
= t!>(- 3. I 06) <P(- 2. 569) = 0.048 X I o- 4
P(B) = <P(-- 3 997 ) <P[- 3.106 - 0.847 X 3.997)]
. (I - 0.847 2)1 ' 2
= t!>(- 3.997) <P(0.5258)
= 0.222 X JQ-4
P(A) -t- P(B) = 0.27 X J0-4
max [P(A); P(B)] = 0.222 >( JQ-4
Joint probability: P(E2EJ)

P(A) =·' t!>(-- · 3 561 ) <P[-3.561 - 0.846 X 3.561]

. . ( I - 0.8462)112
= t!>(---3.561) <P(-1.029) = 0.281 >< I0-4
P(B) = P(A) ..' /33 = f3z
P(A) + P(B) = 0.562 X IQ-4
max [P(A); P(B)l = 0.281 X I o-4
Joint probability: P(Ez£4)

P(A) = <P(- 3 _56 1) tt>[- .997 - 0.925 .<3 .561]

I - 0.925 2) 1' 2
\ '
= t!>(-3.561) t!>(-1.85) = 0.059 x JQ- 4
I
I
P(B) = !J>(- 3.997) t1>(0.358) =' 0.203 X JQ-4

P(A) + P(B) = 0.262 X I0-4

max [P(A); P(B)l = 0.203 X J0-4
Joint probability: P(EJ £4)

P(A) = <P( _ 3 561 ) <P[- 3.997 - 0.925 x 3.561]

. (I - 0.9252)1/2
= 0.059 X I0-4
P(B) = 0.203 X I0-4
P(A) + P(B) = 0.262X IQ-4
max [P(A); P(B)] = 0.203 X l.Q-4
Bounds on the probability of failure of the system are calculated using
Eqs. (10.19) and (10.20).
Lower bound:
ft 1- 1
pr. ;;::: Prt + 1-2
£ max [{pfl - E P(EtEj)}; 0)
J-1
 289

;;;:: Pfl + max [{pr2 - P(E2E1)}; O]

+ max [{pr3 - P(E3EJ) - P(£3£2}}; 0)
+ max [{pr4 - P(E4EJ) - P(E4E2) - P(£4£!)}; 0]
;;;:: {9.35 + max [(1.85 - 1.584), 0]
+ max [{1.85 - 1.584 - 0.562); 0)
+max [(0.317- 0.27-0.262- 0.262); O}x I0-4
;;;:: 9.616X J0- 4
lJ pper bounct.
n
E max [P(E1E1)]
1-2,}<1
,
~ E Pfl - max[P(E2E1)] - max[P(E3EJ); P(£3£2)]
i-1 .

- max[P(E4EJ); P(E4E2); P(E4EJ)]

~ [(9.35 + ·1.85 + 1.85 + 0.317) - 1.25 - max (1.25; 0.281)
- max (0.22; 0.203; 0.203)] x J0-4
~ 10.648 X 10-4
•renee, bounds on Pr• of the system are
9.616X I0- 4 ~ Prs ~ 10.648X 10- 4
XAMPLB 10.8 An under-reinforced concrete beam of breadth (b) 240 mm
nd effective depth (d) 480 mm is reinforced with steel bars of area (A.)
lo400 mm 2• The span of the beam (/) is 6 m. The beam is subjected to a
I tat uniformly distributed load Q over the entire span and torsional a
moment T at a distance of 1 m from one end. It is given: ·
Variablejy: p. = 320 N/mm 2; a = 32 N/mm 2
(Pe 250)
Variable /cu : . p. = 22.67 N/mm2 ; a = 5.44 N/mm2
(Mix M 15)
Variable Q : p. = 16 N/mm; a= 5 N/mm
Variable T:
,., p. = 5 x 106 N mm;
a= 1.5 X 106 N mm
The beam is reinforced with shear stirrups of area, Asv = 56.57 mm 2.
Spacing of stirrups, s = 300 mm. Three limit states of collapse (i) in flexure,
(fi) in shear and (iii) in combined bending and shear are considered. Deter-
mine the probability of failure of the beam considering all the three failure
sing
mOdes. Assume all variables are normally distributed.
Solution Collapse in flexure The ultimate resisting moment of the beam is

R =fyAsd[ 1- 0.77 ,{yA.]

. bd/cu
 290

The failure surface equation is

J2
Z=R-Q-=0
8
Substituting the given data, and designating
X1 = feu ; Xz = fy ; XJ =Q
the failure surface equation becomes
Z === 672000 X1X2 - 6288 Xi - 45 >< 105 X1X3 = 0
Using Level 2 method (Sec. 8.3.1), the problem is solved and the following
results are obtained.
~ =-= 3.305 pr = 47.42 x J0- 4

C<; == --- 0.987 8; Y.; = : 0.0325; ~; == 0.152

Collapse in ~hear The shear strength of the beam is given by
; ---
R = fyA., ~: -i /~/ yO.~ fc,~ [ .Y___I_-f}-~:::-_!]

(} = ~ 0 . ~ (c,
where 1: I
6 l\9 p,
I 00 A,
hd
0 is an emplrical~:oefficicnt depenuing on /~u and fi t- In this problem, (} 1s
assumed deterministic constant. For /~u == 15, A, =: 1400, h = 240, d ~=
480, 8 -==; 1.439. The failure surface is given by
\
I I
I Z '"" R Q 2 ''" 0
ll~ing the giYen data, the above equation becomes,

90 .5 x~
1
7 -= 22222 .v-: ' : -- 3ooo Q "" o
l.' sing level 2 method explained inCh . 8, following results are obtained .
f3 =-== 3.814 PI =--= fi.R47 10 4

:z; =~ -0.8753; :x; -~= -0.0917; :z; =' 0.4749

Collapse in combined shear and torsi oil For checking under combined shear
and torsion, IS: 456- 1978 gives the following equation to calculate the
equivalent shear (Ve).

Ve = V + 1.6 r
where Vis the shear at the section due to load Q. At the section (I m from
end) where torsional moment is acting,

Ve = Q (_!__
2
- 1000) +- I
.
(i _I_
240
= 2000 Q +- 0.00667 T
 211

1n the resistance part in the failure surface equation derived in the pre-
~ u failure case, the failure surface equation under combined shear and
t rt on becomes
Z = 22222 x?·' + 90.5 X:z - 2000 X3 - 0.00667 x..
here x. = T. Using Level 2 method, following results are obtained for
lhl failure case.
fJ = 3.262; Pt = S5.3X Jo-4
owing
•
~. = -0.8223; atz• = -0.1142

IX; = 0.3942; at4 = 0.3942

Considering the beam as a system under the three failure modes, simple
h unds on Pr• of the beam [E.q. (1 0.14)], are
55.3 X I0- 4 ~ Pf• ~ (47.42 + 6.847 + 55.3)X J0-4
5.53 X IO-l ~ Pra ~ 10.96 X JO-l
urrow bounds The failure modes are numbered in the descending order
tftheir pr values.
Mode 1 : z, = 22222 x~·' + 90.5 X2 - 2000 XJ -- 0.00667 X4
8 is Mode 2: Z2 = 672000 X1X2- 6288 X~- 45 x 10' XtX3
•d = Mode 3 : z3· = 22222 x?·' + 90.5 X2 - 3000 XJ
rretation between model and mode 2: Using Eq . (IO.IIb),
4
:E IX;,~;,
Pz" z 2 = ,_,
= ( -0.8223)( -0.9878) + (-0.1142)(0.0325)
d. + (0.3942)(0.152)
= 0.8685
· !oint probability : P( £1 £2)
Using Eqs. (10.23) and (10.24),
shear
;: the P(A) = ~(- 3 . 262 )~[ - 3,305- 0.8685 x 3.262]
VI - 0.8685 2
= 9.346>~ 10- 4
P(B) = ~(- 3 . 305 )~ [ - 3.262 - 0.8685 /. 3.305 ]
from VI - 0.~6852
= I0. I 84 ;.-: 10- 4
P( A) -1- P(B) = 19.53 >~ J0- 4
max[P(A); P(B)l = 10.184v JO-•
 U2

Correlation between mode 2 aDd mode 3: Using Eq. (lO.llb),

Pz 2, z 3 = (-0.9878)( -0.8753) + (0.0325)( -0.09168)
+ (0. I 52)(0.4749)
= 0 .9338
Joint probability: P(E2 £3)
Using Eqs. (10.23) and (10.24)

P(A) = tl>(- 3 . 305 )1P[- 3.814- 0.9338X3.305]

v1 - o.9338 2
= 1.004 x to-- 4

P(B) = tl>(- 3 .814) cp[_ 3.305 - 0.9338 X 3.8141

'\/ r -2 0.9338
= 5.219 X I0- 4
P(A) + P(B) = 6.223 X 1o-4
max [P(A); P(B)] = 5.2l9X 10- 4
Correlation between modes 3 and I:
Pz 3 , 7
1
= ( -0.8753)( - 0.8223) + (- 0.09168)( -0.1142)
+ (0.4749)(0.3942) + (0)(0.3942)
I ~
= 0 .9174
'··
For failure modes 3 and I, we have

= 4.958 X J0- 4
P(B) = tl>(- 3 . 262 ) cp [- 3.8 14 - 0 .9 174 x 3.262]
'\/ I - 0.91742
= 1.073 >< JO• 4
P(A) + P(B) = 6.031 X 10- 4
max[P(A); P(B)] = 4.958 X 10- 4

Lower bound on pr. is (Eq. 10. I 9)

Pr ~ 55.3 X J0- 4 + max [{pr2 -- P(£1£2)}; 0]
+ max[{pn - P(EJ£1) - P(£3£2)}; 0]
;;:;-: 55.3 x w- 4 +max [(47.42 - l9.53)x w- 4 ; 0]
+ max[(6.847 -- 6.031 - 6.223) x w-4; Ol
~ (55.3 + 27.89 + OJx w- 4 = 83 .19 x w- 4
 pper boand on,. is [Eq. (10.20)]
J
,_,
pr lEt: E pr1 - max [.P(£2E1)]

-max [P(EJEJ); P(E3E1>l

~ [(55.3 + 47.42 + 6.847)
-10.184- max (4.958; 5.2l9)]X lo-- 4
~ 94.164 X lQ--4
tl nee bou.nds on Pr• of the beam arc
8.289 x to-3 ~ pr. ~ 9.416 x to- 3

0.5 AUTOMATIC GENERATION OF A MECHANISM

he problem of the reliability analysis of a frame structure becomes formid-
blo if one uses sophisticated probability models for the basic random
v ria blcs and safety margins, as well as nonlinear analysis of structures. In
order to obtain tractable analytical models, the methodology of the reliabi-
l ty analysis of plane frame structures is developed using the stiffness matrix
nt thod, the linear elastic and piecewise linear elastic-plastic (PWLEP)
lructural analysis, and the first-order second-moment method of reliability.
I ng with the usual assumptions in the conventional plastic analysis of
1 uctures, it is also assumed that (i) applied loads are concentrated forces,
(l ) the PWLEP analysis is based on the mean values of basic random
ria bles, (iii) plastic moment capacities of sections, M,, and applied loads,
{h are the only random variables, and (iv) plastic moment capacities of
ec tion s are statistically independent of applied loads.

JC),5.1 Failure Models

At any &tage of the structural analysis, the failure of a section (member end)
I assumed to take place when the plastic moment capacity of the section is
reached. This failure is called the formation of the ~plastic hinge at the
ection. .
, In a redundant structure, a collapse mode forms only when a sufficient
number of hinges have developed. The fa ilure mode of a structure i defined
as the formation of a collapse mecha nism . When the PWLEP a nalysis is
carried out by moving from one hinge to a nother, the criter ion o f determin-
ing the formation of a mechanism is given by the singularity of the sti ffness
matrix, [K], i.e. I [K] I =0. I [K] I is read as the determinant of matrix [K ].
The finally formed hinge converts the structure into a collapse mechanism
and the failure model of the finally hinged section corresponds to the
collapse mechanism.

10.5.2 Safety Margin Equation -

The results of PWLEP analysis enable one to write the safety margin equa-
294

tions of a potential hinge section (the section selected for forming a hinge)
at any stage of a failure path in terms of M; and Qj. The safety margin is
the difference between the plastic moment capacity of the section and the
bending moment at the section, just before forming the hinge due to applied
loads and plastic moment capacity of earlier sections. For example, if the
frame shown in Fig. 10. 14 is considered and, if the sequence of hinges
formed in a failure path are at member ends 2, and 4,. and at 6 (the poten-
tial hinge), the hinge is going to be formed, the safety margin, Z6, of the
section 6 at this stage can be written as

Z6 = G62M2 + (/64M4 -1· G66M6- 1:" b6jQ; (10.25)

j = l

O.z

0 0.1
GJ 8
OJ 6
2 7 4

Fy
5m
[I] QJ

1
M
F~

J
rn.
I------ J rTI -+--Jm -.J

[.!] I Fl E'prE'se n ts n1emb€'r numbiH

-f I RP.presents member tond number

FIG. 10.14 One-bay one storey frame-Example 1 0 9

where /1[; is the plastic moment capacity of section i, (/6; is the moment at
section 6 due to unit M;, b6j is the bending moment at section 6 due to unit
load Qj, and 11 is the number of loads. It is to be noted that a66 is unity.
P(Z6 < 0) gives the probability of failure of the section 6, given that
sections 2 and 4 have already failed. As the analysis progresses, at every
stage of the progressive failure tree the safety margin equation for the hinge
to be formed can be written. As PWLEP analysis is carried out by moving
from one hinge to another. when the ~tiffness matrix of the structure
becomes ~ingular, the linall~ formed hinge convert;; the structure into a
mechanism and the failure model of the finally hinged section corresponds
to the coll:1psc mechanislll Th~ safet~ margin of the lin ally formed hinge
 295

inge) ll!'l' llllcs the safe ty margi n cq uat i n of the mechanism. This sa ti~ty margin
;in is l'llli:Hion c in cides \\' ith the safe ty rll'lrgin qu ation obtained from th e con-
I the ' ntional mechanism method of plastic analysis. When a mechanism is
plied l o~rrucd. P(Z; < O) gi1·cs the pro bability of occurrence of the ·failure
f the ITICld • i.
.nges The sa:'ety margin, in general, for a potential hinge section i or, the safety
>ten- Ill IPi n of a mechanism having the last hinge at section i, is expressed as
f the
Ill "
Z; = 2.' OijMi - 1J b;kQk (10.26)
)= ! k~l

).25) whore m is the number of critical sections (member ends) in the given frame.
In u particular failure path, if there is no hinge at the member end j, the
t'l)!l'espond ing coefficient a;i = 0. For i = j, a,, = 1.
If random variables M and Q are grouped in X, Eq. (10.26) can be
' ' rtlc n in the generalized matrix form as
Z; = [A]{X} (10.27)
'I he mean value and standard deviation of Z; are
1-'z; = [A]{!-'x} (l0.28)
al1 = [A][Cx][A]t (10.29)
\•here [A] is a row matrix of coefficients a;; and b;k for all variables X1, [A]'
the transpose of matrix [A] , {J.Lx} is a column matrix of the means of all
11111dom variable Xh and [Cx] is a covariance matrix of all random
,, l'iablc XJ. The reliability index {J; for the safety margin Z1 is given by
J.Lz;
{J; = -
az;

As Z; is a linear function of the number of variables XJ. the dist~ibution of

z, tends to normal, [based on the central limit theorem (10.5)] irrespective
of the individual distributions of the variables. Hence, assuming normal
distribution for Z1, the probability of a structure under a collapse mechan-
Ism i can be computed.
The methodology of the reliability analysis of ductile structural systems
nt at involves the following steps:
unit
nity. (i) Data
that (a) structural data
:very (b) probability description of the random variables
inge
X1 in terms of fL Xi' q Xi and PXj,Xk
ving
:ture (ii) Linear clastic analysis
tto a (a) determination of coefficients OiJ, b;k
:~nds
tnge (iii) For any potential hinge location
(a) writing the safety margin z, from Eq. (10.27)
 296

(b) compulalil'll of !l·:t.i and a:t.i using Eqs. (10 .28) and (10.29)
(c) computation of fi;
(iv) Selection of the next hinge location
(v) Formation of the plastic hinge at the selected member end
(vi) Modification of the member stiffness matri.\ having plastic hinges at
the ends, as shown in Figs. I0.15, I0. I 6 and I 0.17
(vii) Application of a plastic moment at the hinge in the form of equi-
,·alcnt forces, as shown in Figs. 10.15, 10.16 and 10. 17.
(viii) Determination of rhe structure stitrnes> matrix [K]
(ix) Linear elastic-plastic anaiysis and determinat io n of coefficients a;j.
(x) Repetition of steps (iii) to (ix) until the formation of a mechanism.
The above procedure is illustrat.::d with an example .

!a) He mber Loading lnd•Jced by H1

'I·'· '~> I
\ ' EA 0 0
EA
I, I I
0 0

J EI JEI J El
-,-J- 0 0 -,-) -,2-

0 0 0 0

•
f: 5YM
EA
I
0 0

JEI J El
7 -7
JEJ

(b) Member Sttffness Matm:

FIG. 10.15 Effect of hinge: left e•·d of member hinged

 297

Ia I Member loading Induced by H2

'ij.

m.
EA 0 0 EA
0 0
I I

JEI JEI JEI

7 7
0
-7 0

JEI JEI
0
-IT 0

EA
SY M -~- 0 0

J EI
0
7
0

(b) Member Stiffness Matrix

FIG. 10.16 Effect of hinge : right end of member hinged

EXAMPLE 10.9 A simple 011e-storey, one-bay porttd frame is .; ub~.:clcu w

vertical and horizontal loads, is shown in Fig. 10.14. The data for the
problem is given in Table 10.2.
For this frame the degr~e of redundancy is three, anu the maximum num-
ber of hinges required for a mechanism is four. The stepwise procedure of
generating a mechanism is illustrated below.
(i) The linear elastic analysis of the structure is performed to compute the
bending moments at the member ends, expressed in terms of the coefficients
au and b;k. The structure at this stage is considered intact and this stage is
called the first stage. The number of critical sections (potential hinge sec-
tions), m, is equal to eight. TJ1ey are marked in Fig. 10.14. The number of
loads, n, is equa·l to three. - _
(ii) At this stage, for all ·critical sections the safety margin equations in
terms of au, MJ, b;k and Q;k are generated and reliability indices, {1,, are
com outed.
 298

• Mz
~·-
El
l (MJt~)
- -1 - ~

(a J Member loading Induced by M1 and H 2

-,EA- 0 0 . -,-
EA 0 0

0 0 0 0 0

0 0 0 0

SVM -,-
EA
0 0

~ 0 0

(b) Member Stiftnus Matrix

FIG. 10.17 Effect of hinge: both ends of member hinged

p
t

TABLE 10.2 Data for frame in Fig. 10.14-Examp/e 10.9

Section or EA EJ
(J
variable (kN) (kN m 2 )

Seclion
I, 2, 3, 4 0.367 X J0? 0.92 X 10 1
5, 6, 7, 8 o.%5x 107 0.406X 10'
Variable
M 1 , 1112 , M 3 , M, 121.57 kN m 9.969 kN m
,'vf., M 8 135.04 12.424
Me, M, 344.65 32.74
Ql 105.0 kN 10.5 kN
1Q. 36.0 14.4
Q. 2.4 1.032
Note: All variables are statistically independenL
 299

hen the external acting moml:nt, S, at the section due to loads is positive,
U1 sufcty margin is written as
Safety margin = resistance - action
= R - - S = R - - (+S)

II , is negative,
Safety margin =-= R -1- S = R + (-S)
I I r example, if the po tential hinge section 7 is considered, the safely margin
1111 section7 at this stage I, using Eq. (10.26), is
s )
Z1 = L' ll7jMj - L' b1kQk
J= l k-1

I r m the clastic analysis of the frame,

"71 = -- 1.24 hn = - 1.24 lm ~ 0
·1he negative sign shows that the direction of the bending moment is clock-
wi c. The sign conventions for forc(.s are shown in Fig. 10. 14. Since this is
Htgc I, there is no hinge at any member end j. All au coefficients are zero
cept (177 = I. The capacity of the section is M1. The action .is - · (1.24 Q1
I 1.24 Q2) . Hence, the safety margin equation for section 7 is
Z1 = 1.0 M1 - 1.24 Q1 - 1.24 Q2
·rh mean value and standard deviation of z, are
I~Z7= 344.65 - 1.24X 105- 1.24X 36
= 169.81 kN m
Oz7 = [(32.74) 2 + (1,24 X 10.5)2 -t (J.24X }4,4)2]1 12
= 39.4 kN m

#7 = ~~~:!_!_ = 4.31

Similarly, for all other potential hinge sections, safety margin equations
~can be written and fl values found out at this stage.
'· (iii) Let us now assume that the first hinge is formed at section 7. Now
! ~he stiffness matrix of the member 4 having plastic hinge at the left end is
modified as given in Fig. 10.15.
(iv) A clockwise moment of 344.65 kN m, (i.e. negative moment), equal
to the plastic moment capacity of section 7 is applied at the left end.
m Corresponding to this moment, a self-equilibriating force system, shown in
Fig. 10.15, is considered as an additional load case for further analysis.
(v) The structure stiffness matrix is assembled and the determinant I [K] !
is computed and found to be greater than zero.
(vi) The linear elastic-plastic analysis at this stage 2 is carried out to
determine au and b;k for all the potential hinge sections. Knowing a,, and
 300

btk, safety margin equations for all potential hinge sections can be written .
For example, if section 4 is considered,
{/.17 = -- 1.0 b41 = 1.5 b~z=I.S

a44 = I .0

The capacity of the section is M4. G44 = 1.0. Hence, the safety margin
equation for section 4, given the hinge at section 7 has formed, is
7~ == I .0 M4 - (- -1.0 M1 + 1.5 Q1 + 1.5 Q2 + 1.2 QJ)
= M4 + M1 ·-- 1.5 Q, - I .5 Qz - 1.2 QJ
Using the given data in Table 10.2,
/LZ4 = 121.57 + 344.65- f.5-..; 105- J.5XJ6- 1.2X2.4
=c.=. 251.84
oz4 = l(9 .969) 2 + (32. 74)2 + (1.5 X 10.5) 2 + (1.5 X 14.4) 2
+ (1.2 X 1.032)2]11 2
= 43.44

f34 = 25 1.84 "''"' 5. 79

43.44
Similarly, for all potential hinge sections (excluding section 7 where the
l,, .~
hinge is already formeu) at this stage, the safety margin equations can be
I written and fJ found out given that the hinge at section 7 is already formed .
0
(vii) Now another hinge (second hinge) location, say section 4, is selected
). and the hinge is formed at that section, and the whole process is repeated
from steps (iii) to (v). Now the stiffness matrix of the member 2 having
plastic hinge at the right hand side member end 4 is modified as given in
Fig. I0.16. At section 4 an anti-clockwise moment of 12 I .57 kN 111, equal to
the plastic moment capacity of section 4, is applied. Corresponding to this
moment, a self-equilibriating force system, shown in Fig. 10.16, is consider-
ed as an additional load case for further analysis.
The stiffness matrix of the structure is assembled and I [K] I is round to
be greater than zero. The linear elastic-plastic analysis is c~mied out at this
stage 3 to determine the coefficients aij. If section 2 is considered, from
analysis
027 = 2.0 024 = 1.0
b21 ~~ --3 b22 = - . 3.0 b23 = 0
G22 ==

Hence, tbe safety margin equation for section 2, given hinges at sections 7
and 4 have formed, is
(I 0.30)
 301

tten. l l~lnJ:! mean values and stand ard deviat io ns of M2, Af?, M~. Q1, and Q2,
/L7.2 = 509.4
/32 = 5.94
II' \CCI ion 2 is selected for the next hinge location, a hinge is formed at the
r tlon. A clockwise moment of 121.57 kN m, equal to the moment capa-
lly of section 2 is applied at the member end 2. The stiffness matrix of
11 mb r I with a hinge at the right member end 2 is modified as shown in
l lJ I0.15. The stiffness matrix of the structure is now assembled and I [K II
fou nd to be ~ zero. This shows that when hinges are formed at sections
7, 4 nnd 2, a mechanism is formed. This is a beam mechanism. Using the
IIIC('hanism method of plastic analysis (10.2), one can directly write the
Jnfety margin equation for this failure mode:
Z = M2 + 2/117 ' M4 - - Q1 . 3 -- Q2 . · J
II can be observed that this equation, i.e. the safety margin equation for the
mechanism, coincides with the safety margin equation (Eq. 10.30) for the
r<Hential hinge section 2, written just before the hinge is formed there.
It has been shown in the example how a mechanism can be generated,
· unu how the safety margin equations are written at every stage of analysis
11nd {3; values for potential hinge sections computed, and how the failure
•urface equation or the failure m del of the fi nally hinged section corres-
:the ronds to the collapse mechani sm. {3; of the Jast hinged section becomes f3
n be f)]' the mechanism .
ned.
cted 10.6 GENERATION OF DOMINANT MECHANISMS
a ted
ving
In the last example, only one mechanism was generated out of 15 possible
failure mechanisms. A plane frame structure may fail in different collapse
n in
mechanisms, called failure modes. The reliability analysis of frames mainly
11 to
involves identification, modelling and synthesis of all possible failure modes
this
to estimate the system reliability. In the case of a frame structure of a high
der-
degree of indeterminancy, the number of possible collapse mechanisms is
quite large. To illustrate, for a one-bay two-storey rectangular frame with
d to
fixed bases, the number of elementary mechanisms, Ne, is equal to 8. The
this
number of possible collapse mechanisms is given by 2N' - 1 = 255. Due
'rom
to uncertainties of load and resistance variables, it is likely that the structure
may fail under any of the possible collapse mechanisms. Hence, the relia-
bility of frames of multiple components and with multiple failure modes is
considered from the system point of view. Out of the innumerable possible
collapse mechanisms, generally only a few mechanisms, having compara-
tively large failure probabilities, contribute significantly to the system
failure probability, pr•• These collapse mechanisms are called stochastically
ns 7
dominant failure modes. The identification and combination of these domi·
nant collapse mechanisms are necessary in the reliability analysis of a
1.30) frame structure to estimate its system reliability. It is practicany difticult
 302

and rather impossible to identify these dominant l'ailure modes. There is

no mt>:hod which assures, and mathematically proves . that all stochastically
domi1~ :111t modes arc genera ted . However, the methods, n;~ mcly ( i) exhaus-
tive enumeration, (ii) simulati on, and (iii) heuristic se arch, are generally
used for this purpose. The earlier studies ( 10.6, 10.7) concentrated on the
reliability analysis of knO\\ n failure modes. The f'o1 cmost essential step or
the identification or dominant failure modes in a frame structure has been
the subject of research during the past eight years. Ma anJ Ang (10.8) have
suggested a method 0f dererm in ing the most probable modes by using a
mathematical programming technique, based on independent failure modes,
obtained deterministically by Watwood's ( 10.9) method . Murotsu (10.10 ,
10. 11) has proposed a complex method, based on the joint probabilities of
hinged sections, for the automatic generation of stochastically dominant
failure modes. Moses (I 0.1 2) has proposed a strategy, using the incremental
load approach, to identify and enumerate the significant failure modes of
trusses. Tang and Melchers (I 0.13) have proposed a trurlt'rtted enumeration
method to search for stochastically dortlinant failure modes . Ranganathan
and Deshpande (10 . 14) have proposed a hePristic search technique to
generate dominant modes in frames. This is explained below.

10.6.1 Heuristic Technique (1 0.14)

The strategy developed for a sequential search of' plastic hinge locations
\ \ leading to clominan L mechanisms is exrlaincd he low .
I,
Search fm· Plastic Hinge Locations
It is quite logical to select the potential hinge section with the lowest
reliability index, for the plastic hinge at any stage of the ann lysis, to get
stochastically dominant mechanisms. However, it has heen observed that
this logic fails in certain situations. Dominant failure paths of equal likeli-
hood of occurrence may intermix, resitlting in a nondominant mechanism.
Therefore, the following strategy fnr the selection of plastic hinge locations
is suggested.
Selection of First Hinge and First Dominant Alechanism After performing
the intact analysis and computing f3 for all potential hinge sections, the
potential hinge section having the lowest value of f3 is selected as the
location for the first plastic hinge. This f3 is called the first damage
reliability index and denoted as f3o. The arithmetic mean of the reliability
indices of all potential hinge sections at the initial stage is termed as the
average reliability index ~av . The reliability of a mechanism is always higher
than the reliability of its hinges at each stage, and the reliability index of a
mechanism corresponds to f3 for the member end hinged at the stage of
mechanism. Hence, after selecting the ii.rst hinge, it is logical to select the
subsequent plastic hinge locations such that the reliability index of the sec-
tion is the lowest at that stage and is also greater than ~o. Following the
above strategy, hinges are selected and the first probabilistically dominant
mechanism is generated.
 303

Partial Full
coflaps~
l
collapsj
Seocond mech mech
First stage stage stage stage
~-------+-

rlo = 4 =''

n
~av z 7 28 4 ·31 5 79 5 94

1g a F'trst failurE!'

n
tdes, tree
), 10,
:s of

n
nant
Intact
Structure

tion
:han
~ to

11
ions

n
.vest
get
n
n
that
~eli­
sm.
ons
5 94 5 94 Reprt>sents reliability tndPx for section 2
ling
the
Q) 2 Reprl!'sPnls hmgt>d SPCt1on number
Rpprest>nls •dent1fit>d mt>chanism number
the
age FIG. 10.18 Failure tree diagram for frame in Fig. 10.14-Example 10.10
lity
the
Branching Strategy After generating the first dominant mechanism, it is
;her
obvious that if this mechanism is branched at all stages .with alternative
•f a
p tential hinge sections in succession, it may be possible to identify all the
:of
possible mechanisms. This procedure is computationai'Jy prohibitive as there
the
oan be a "Very brge number of reanalyses to perform. Moreover, during this
;ec-
process the same mechanisms may be repeatedly generated and insignificant
the
mechanisms identified, which are of no interest from the viewpoint of the
ant
system failure probability. In the light of this, primarily, dominant mecha·
 304

nisms are branched only at the mechanism stage (primary branching) to

develop the first failure tree and secondly, the first mechanism is branched
at the initial stage with alternative potential hinge sections to develop other
failure trees with the help of primary branching. This is secondary
branching (Fig. 10.18).
(i) Primary branching at the mechanism stage: A mechanism is branched
at its final stage with the potential hinge sections having~ greater than f3o
in succession. Two cases arise for this branching (Fig. I0.18).
(a) Partial collapse mode stage
Partial collapse mode is the mechanism having a number of hinges less than
(r + I), where r is the order of indeterminacy of the structure. For this
case, the branching may result in a mechanism or the extension of the failure
path (Fig. I 0.18).
(b) Full collapse mode stage
Full collapse mode is the one having (r -i· I) hinges. The branching at this
stage results in a mechanism and this is the terminating stage (Fig. 10.18).
(ii) Secondary branching at the initial stage: Secondary branching is
nothing but an alternative selection of the first hinge. A mechanism is
independent of the order of the hinges involved in it. Therefore, while
making an alternative choice of the flrst hinge, all the hinges of the first
mechanism except the last are discarded. Out of the remaining possible
locations, it is again logical that there is no propriety to start with the
I \ hinges having f3 of a higher order. Hence, the first dominant mechanism is
'·. branched at the first stage by selecting the potential hinge sections having
{3 < f3av in succession. ln the context of this heuristic technique based on
the logical strategies to identify dominant mechanisms, the various termino-
logy used is indicated in Fig. 10.18.
S)•stem Reliability After generating all the dominant mechanisms and cor-
responding Z;, the probability of failure of the frame under each mode, pp,
is calculated. The correlation coefficients between pairs of the generated
mechanisms arc computed. The failure modes are ordered as per the decreas-
ing values of PJ;, and simple bounds and Ditlevsen's narrow bounds (10.4)
are established for the system failure probability. The method is illustrated
with the following examples:
ExAMPLE 10.10 The same frame, considered in Example 10.9 and shown in
Fig. I 0.14, is taken here to illustrate the generation of dominant mechani-
sms. The data required for the reliability analysis of the frame is given in
Table 10.2. For this frame, the degree of redundancy is 3, the maximum
number of hinges required for a mechanism is 4, and the number of ele-
mentary mechanism~ is 4, whereas the number of possible mechanisms is 15.
The stepwise procedure of generating dominant mechanisms, and reliability
analysis, is explained below:
(i) The linear elastic analysis of the strw.~ture is performed to compute the
hending moments at the member ends, exp,·essed in terms of the coefficients
 305

) to I) 'rhc signs of the bending moments are noted . The structure at this stage
:bed I considered as intact and this stage is called the first stage.
ther ( ) At this stage, for all potential hinge sections, the safety margin
lary 111 t1 ns in terms of Gij, MJ, b,k, and, Qk are generated and reliability indices
, 1 computed (explained in Example 10.9) as given in Table 10.3. The
1 t nt ial hinge sections are ordered with increasing reliability index. It is
hed
1 fJo und that Po = 4.31 and f3av = 7.28.
iii) From Table 10.3, it is noted that the sections 6 and 7 have the same
~ . Therefore, comparing the reliability indices at both ends ofthe members
ond 4 (sections 5 and 8), section 7 with the lowest reliability index of 4.31
han lclected as the first hinge.
this (lv) The first plastic hinge is formed at section 7. The safety margin of
lure Ihe structure at this stage, with the hinge at section 7, is given in Table 10.3.
(v) The stiffness matrix of member 4 having a plastic hinge at the left end
modified as given in Fig. 10.15.
this (vi) The moment of 344.65 kN m, equal to the plastic moment capacity
). 1 f section 7 is applied at the member end, in the direction of the bending
: is me ment developed at the member end in the elastic analysis. Correspond-
1 is n to this moment, a self-equilibria ted force system, as shown in Fig. 10.15,
hile I considered as an additional load case for further analysis.
first (vii) The structure stiffness matrix [K] is assembled and the determinant
ible I [K] I is computed and found to be greater than zero.
the (viii) The linear elastic-plastic analysis at this stage is carried out to
1 is I terrnine au and b;k for the potential hinge sections.
·ing (ix) Steps (ii) to (viii) are repeated as explained below. At the second
on I age, the reliability indices for potential hinge sections are computed, which
no- ro given in Table 10.3. The second plastic hinge, having the lowest reliabi-
lity index of 5.79 and greater than f3o, is formed at section 4, as shown in
;or- Fig. l 0.18. Corresponding to the first and second plastic hinges at sections
PJI, 7 and 4 respectively, the modified member stiffness matrices for members 4
tted 11nd 2, and additional load cases equivalent to plastic moment capacities of
~as- &ections 7 and 4, as shown in Figs. 10.15 and 10.16, are considered for
).4) further analysis. The safety margin of the structure having the second hinge
1ted at section 4 is given in Table 10.3. The determinant of the structure stiffness
matrix is found to be greater than zero.
nin At the third stage, according to the selection strategy explained earlier,
lni- the plastic hinge 1 formed at section 2, as shown in Fig. 10.18. As I [KJI
lin ~ 0, the first dominant mechanism is generated.
.urn (x) This mechanism consists of three hinges at 7, 4, and 2. It is therefore
ele- n partial collapse mechanism. The safety margin equation of this mechanism
15. is same as the safety margin equation of the hinge at section 2 (Table I 0.3).
lity This mechanism is the most dominant mechanism, having a reliability index
5.94 and a probability of failure 0.145 X lo-s.
the (xi) As per the branching strategy, this mechanism is branched as shown
:nts in Fig. I 0.18. To do this, the last hinge of this mechanism at section 2 is
 ----
'-J.- .-?

w
0
Gt

TABLE 10.3 Details of development of first failure tree in Fig /0.14-Exanzp/e 10.10

Stages of Reliability index of sections Selected

analysis Safety margin equation
1 2 3 4 5 6 7 8 Hinge
p
section

First 10.3 7.87 9.67 7.35 7.43 4.31 4.31 7.01 Z7 = l.OM7 - 1.24Q1 - 1.24Q0 7 4.31
Second 8.17 5.92 7.90 5.79 6.13 0.0 0.0 6 01 z, = I.OM, + I.OM7 - 1.5Q1 4 5.79
-1.5Q 2 - 1.20Q0
Third 11.9 5.94 6.71 0.0 6.07 0.0 0.0 00 Z, = I.OM2 + I.OM, + 2.0M7 2 5.94
(Mecha- -3.0Q,- 3.0Q,
nism I)
Third 11.9 5.94 6.71 0.0 6.07 0.0 0.0 0.0 Z5 = l.OM, + l.OM5 + 2.0M7 5 6.07
(Mecha- -3.0Q, - 3.0Q,
nism 2)
fhird 11.9 5.94 6.71 0.0 6.07 0.0 0.0 0.0 Zs = I.OM3 + l.OM, + 1.5M7 3 6.71
-2.25Q 1 - 2.25Q, - 2.50Q,
Fourth 8.39 5.94 0.0 0.0 6.07 0.0 0.0 0.0 Z1 = I.OM 1 + I.OM 3 + 2.0M, 1 8.39
(Mecha- + 2.0M7 - 3.0Q1 - 3.0Q,
nism 3) -5.0Q0
 307

IIIWtrcssca and replaced by sections, in succession, except 7, 4 aDd 2 and

having a reliability index at this stage greater than Po.
i As shown in Fig. 10.18, the selection of section S with P as 6.07
in a dominant mechanism. The failure path continues with the aelcc-
of the section 3 with f3 as 6. 71, as shown in Fig. 10.18. Again, accord·
I the selection strategy, the next hinge, i.e. the fourth plastic hinge is
mcd at section 1 having f3 as 8.39. Since I [K] I ~ 0, the full collapse
mism is formed at this stage. The branching of this mechanism does
result in any new mechanism. As this is the terminating stage, the
~•IVCIIODtTleJlt of the first failure tree is completed. '
i i) To initiate other failure trees, according to the strategy of the selec·
h n of alternative first hinges, only one alternative is possible in this case,
h wn in Table 10.4.

TABLE 10.4 Selection of first hinge for failure trees other than the
first in Fig. 10.!4-Examp/e 10.10

fJ at Possibility of
first selection of Remarks
stage first hinge

I 10.3 Not possible f3 > Pav

2 7.87 Not possible .B > Pav
3 9.67 Not possible fJ > fiav
'4 7.35 Not possible Involved in the
first mechanism
5 7.43 Not possible fJ > Pav
6 4.31 Not possible Equinoda I to 7
7 4.31 ~ot pos~ibk Involved in the
first 111cchanism
R 7.01 Possible B< 11.,.

(xiv) The first hinge of the second failure tree is selected at section 8, and
thi failure tree is developed using the procedure similar to the first tree.
I he failure tree diagram for this example, including the dominant
m chanisms generated, is shown in Fig. I0.18.
(xv) The identified mechanisms are arranged in the increasing order of the
reliability index, as given in Table I 0.5.
(xvi) The correlation coefficients of mechanisms, calculated using
Eqs. (10.10) and (IO.IIa), are presented in Table 10.6.
(xvii) Simple bounds are computed using Eq. (I 0.14), and narrO\\ hounds
u ing Eqs. (10.19) and (10.20). They are also shown in Table 10.5.
From the rusults it can be seen that for this example, the first dominant
mechanism, having a probability of failure of 0. 145 >< IQ- 8, is the most
significant mechanism. Therefore, the lower t lUnd on the system collapse
probability of 0.159 x w-s, is close to the failure probability of the first
dominant mechanism.
.....
~ - ~ .... ---- ~-==-~
_..y

TABLE IO.S Details of generated dominant mechanisms and results of reliability analysis of frame ill Fjz. /0./4-Example 10.10

SI. Hinged Fail me

No. sections
z p Pr tree
- -
I. 2, 4, 7 l.OM, + l.OM, + 2.0M 3.0Q 1 - 1 - 3.0Q2 .:i.94 O.t45x to-•
2. 4, 5, 7 1.0M, + l.OM + 2.0M 3.0Q
5 7 - 1 - 3.0Q1 6.07 0.642xt~ 1
3. 2, 7, 8 l.OM + 2.0M1 + l.OM
1 3.0Q 8 - 1 - 3.0Q1 6.07 0.642xlo-t 2
4. 5, 7, 8 l.OM + 2.0M + l.OM
5 13.0Q 1 - 1 - 3.0Q2 6.20 0.282xJo-• 2
5. ], 3, 4, 7 l.OM' + l.OM~ + 2.0M, + 2.0M
1 7 - 3.0Q1 - 3.0Q 2 - 5.0Q. 8.39 0.239X to-11 I
6. t, 3, 7, 8 l.OM1 + !.OM,+ 2.0M + 2.0M 7 8 - 3.0Q1 - 3.0Q,- 5.0Q1 8.51 0.542 X I o-17 2

System failure probability

Simple bounds 0.145 x to-a ~ Prs ~ 0.302 x 10-•
Narrow bounds 0.159x w-• ~ Prs ~ 0.214x to-•
 Correlations betwe~n geflerated mechanisms-Example 10.10

Correlations Pu
2 3 4 s 6

1.0 0.982 0.982 0.965 0.972 0.932

1.0 0.965 0.963 0.968 0.928
1.0 0.983 0.942 0.969
1.0 0.938 0.965
1.0 0.935
Symmetrical 1.0

An unsymmetrical two-storey two-bay frame, carrying

horizontal loads, is shown in Fig. 10.19. The data for the

a2

9 ..:> 11 12
4

Gt
. 16

GJ
T
J·6m

J 15
5 6 7 8 17 20
2 14 18 19 22

J 6m
21 _l
f- --3Om -+-3-0m -4----J Om---~3 Om----j
FIG. 10.19 Two-storey two-bay unsymmetrical frame-Example 10.11

umple is given in Table 10.7. For this structure, the number of elementary
me banisms is 10 and the number of possible mechanisms is 1023. The
tlentified dominant mechanisms for this example are indicated in Fig. 10.20.
he results of the identified dominant mechanisms and of the reliability
nalysis are given in Table 10.8. The correlation coefficient matrix, repre-
cnting the correlation between pairs of mechanisms, is shown in Table 10.9.
I' r this example, the dominant mechanisms are very close to each other.
M st of the dominant mechanisms are identified in the· first tree only.
The same problem has been solved by Ma and Ang (10.8) and Murotsu
(10.10), and the 'results of the generation of dominant mechanisms are
mpared with their results (Table 10.8).
310

TABLE 10.7 DaJa/Qr frame in Figure 10./9-Examp/e 10.11

EA El
Section /Variable
(kN') (kN m•) v. ll p

------
Section
1, 2, 3, 4, 21, 22 0.105 X ]07 0.105xl01
5, 6, 7, 8 0.132 x 107 0.277 X 101
9, 10, 11.12 0.101 X 107 0.154xlQ6
13, 14, 15, 16 0.101 X 101 0.758 X 101
17, 18, 19,20 0.116x 10 1 0.207x 105
Variable
lvf" M 2 • M,, } 0. 15
95.0 kN m
Ma. M,., M 22
M,, M,, J.fa, lv1 10 95.0 0.15
10\
1.0 Other-
M 5 , Me, M 7 , h1s 204.0 0.15 1.0 wi se un-
!vf, , M 1o. M 11 , M., 122.0 0. I 5 1.0 co rrelated
1\1 1?. M 1s. M 11 , M, 0 163.0 0.15 I. OJ
Q, 169.0 0.15 Loads arc
Q. 89.0 0.25 independent
Q, I 16.0 0.25 except
Q, 62.0 0.25 PQ4.QS ~- I
Q, 31 .0 0.25

Di.WIS.I'ion and Concl11sion (L0.14) Tracing of the critical failure path is c: ru-
ia l. ll is ob erved in Fig. 10. 18 that wh ile tracin g lhe critical path, if the
reli ability indi ces of the sequential hinges are monoton ically increasing, th e
failure path is efficient, and it lead to the most domin a nt failure m de and
also that bran ' hing this path at the mechanism stage results in many of the
dom ina nt mcchnnisms in the first failure tree.
The number of branchings and the number of failure trees vary with the
type of problem, depending on the structural topology, load distribution,
etc. The more the parallel failure paths, the more the branching operations
will be. If parallel paths get mixed, the randomness increases. In some
circumstances, inadmissible mechanisms are generated . In certain situations
there can be a very large number of cycles to perform.
IIi. bse rved that in Exa mple 10.11, all the dominant modes generated
by uth er rc ·ea rch w rkers have also been obt ined using the proposed
meth d. Howe er, more modes, including a few insignificant mode are
generated in the process. It is found from Tables I0.5 and 10.8 that the
most dominant fai lure mode i. btainetl in the lit t tree for both the prob-
blems. In Example 10.11 (Table 10.8), all the modes identified by Ma and
Ang (1 0.8) and Mur tsu (10. 10), except one have bee n generated in the f-ir l
tree itself.
It is observed that the accuracy of estimating prs may be improved margi-
nally by generating more failure trees, but is quite expensive. For all practi-
cal purposes, the generation of the first failure tree and the system failure
probabilit y calculated based on that appears to be adequate.
It is concluded that the proroseJ method used simple logical strategies
 311

11 11

~r
16

7 • 1S
n 14 22

1l 21

MECH 21(1:1· 9~ MECH J(fl=2 05) MECH 4(f}=2 06)

9 11

MECH
7 •
14

n
6W =2·09)
22

21 ~ ....
~ 7

MECH 7((3:2 14)

•
""" ~'T Fh
MECH B( (3=2 21)
~

= 1
11
.--
4 16

1 & 19
?
I I

L J.J3 _.,.1
... 91fl=2 21) MECH 101[1=2 23) ME.CH 111(L2 28) MECH 12(fl=2 JZ)

I ~ an'',." n 1-- ~
II

:rated
1
1, 't.!n
. J~-r--.
l,__;_ ~~ 1
'J:....
f--,
.. ~'
I
. ~~· :
posed M CH 1Jirl=2 41) MECH 1t.(fl=2 t.t.) MECH 1Sii1=-f 48; rv'~(H 16([1 o7 /t.l
;, are
t the 10 .10.20 Location of hinges in identified dominant mechanisms-Example 10.11
prob-
t · r the selection of hinges and the branching of failure paths to identify the
l and
J)f babilistically dominant mechanisms. It is simple, fairly efficient, and is
~ first
ilpable of generating the dominant mechanisms for a practical complex
tructure. However, being a heuristic technique, it is not possible to prove
largi-
theoretica ll y whether all dominant mechanisms can be generated by using
racti-
the proposed technique. For practical problems, it is felt that it is enough
1ilure
I the first failure tree is generated and the system failure probability calcu-
lated based on the generated mechanisms in the first failure tree .
.egies
 -- _.,.

TABLE 10.8 Details of generated dominant mechanisms and results of reliablity analysis of frame in Fig. /0.1?-Exaltlple JO .ll
...w
N

Sl . Hinged Failure Whether identified by

No. sections
z /3 Pr tree Ma and Ang (10.8) Murotsu
Fi F2 (IO.IOJ
-
(1) (2) (3) (4) (5) (6) (7) (8) (9)
I. 1,2,13,14 l.OM, + l.OM2 + 1.0M11 + !.OM,, + J.OM11 1.97 0.247x J0- 1 I Yes Yes Yes
21 , 22 + l.OM11 - 3.60Q, - 3.60Q,
2. I , 7, 8, 11. J.OM, + 2.0M1 + I.OM 8 + 2.0M11 + l.OM,. 1.99 0231 x iD- 1 I Yes Yes Yes
13, 14, 15. + l.OMu + J.OM,, + 2.0M e + I 0:\1, 1

16, 21, 22 + l.OM 3.0Q

22 - 3.0Q
1 - 3.6Q, - 7.'2Q,
2 -
3. •4,11,16 I.OM, + 2.0M11 + 1.0M18 - 3.0Q, 2.05 0.200X J0- 1 1 Yes Yes Yes
4. 17, 19,22 l.OM + 2.0M,. + l.OMn - 3.0Q,
17 2.06 0.198xJ0- 1 3 Yes Yes Yes
s. 1,7,8,11, O.SM1 + J.OM, + l.OM 8 + I.OM11 -1- 0.5M,. 2.06 o.I97 x w-• I Yes No Yes
13, 16, 18, + l.OM,. + l.OM + O.SM + l.OMu
11 21

21,22 -1.5Q, - I .SQ. - 1.5Q. - 1.8Q, - 3.6Q,

6. I, 3, 7, 8 I.OM, + l.OM 3 + 2.0M, + I.OMo + 1.011113 2.09 0.183 X ]Q- 1 I Yes No Yes
13, 14, 21 I.OM" + I.OM11 + l.OM22 - 3.0Q 1
22 - 3.6Q, - 3.6Q,
7. 5, 7, 8 J.OM, + 2.0M1 + l.OM8 - 3.0Q, 2.14 0.160 x 10-• I Yes Yes No
8. 9, 11, 16 l.OM0 + 2.0M11 + I.OM,. - 3.0Q 2 2.21 O.J34x Io-• 1 Yes No No
9. 4, Jl, 12 l.OM, + 2.0M11 + l.OMu - 3.0Q2 2.21 0.134x J0- 1 4 No No No
10. 9, 11, 12 l.OM, + 2.0M11 + l.OM,. - 3.0Q 2 2.23 0.124x 10-1 4 No No No
JJ. 1, 4, 7, 8, I.OM1 + l.OM, + 2.0M, + 2.0M1 + l.OMu 2.28 0.112 X IQ-l 3 No No No
13, 16, 19, + l.OMu + 2.0M1o + l.OM,, + 2.0Mu
21,22 - 3.0Q, - 3.0Q, - 3.6Q, - 7.2Q,
 13, 16, 19,
+ I.OMu + 2.0M1, + I.OM11 + 2.0M1 t
21,22 - 3.0Q1 - 3.0Q1 - 3.6Q, - 7.2Q 1

12. 2, 7, I, II, l .OM1 + l .OM, + lAM, + :Z.OMu + 1.........

15, 16 + 2.0M11 - 3.0Q1 - 3.0Q1 - 3.6(1.
13. 1, 7, 8, 9, J.OM1 + 2.0M1 + 2.0M, + I.OM, + J.OM11 2.41 o.mxro-s 3 No No No
13, 16, 18, + l.OM1, + 2.0M1, + 1.0M,. + 2.0M11
21,22 -3.0Qa - 3.0Q, - 3.6Q, - 7.2Q&
14. 2, 3, 7, 8 J.OM1 + J.OM1 + 2.0M1 + J.OM,- 3.0Q 1 2.44 0.734x10" 1 4 No No No
u. 3, 11, IS, t.OM, + 2.0M11 + I.OM,1 + 2.0M11 2.48 0.664xl0" 1 2 No No No
16 - J.oo. - 3.6Q,
16. I, 7, 8, 11, I.OM1 + 2.0M, + 2.0M1 + 2.0M 11 + l.OM10 2.74 0.307 ~' 1o-• 1 No No No
13, 16, 17 +2.0M11 + I.OM1, + I.OM11 + I.OM:s
21,22 - 3.0Q, - 3.0Qt - 3.6Ql - 7.2Q.

System failure probability

Simple bounds 0.247X J0- 1 ~ Pr1 ~ 0.20S
<
Nnrrow bounds 0.702 :-:10 1 p 1i ~ 0.147
UPrs = 0.116 given by Ma and Ang ( IO.K) using Monte Curio simulation with sumplc ~izc SOOO and 0.74$ >{ to- 1 < Pr1 ~ 0.907 ;, to-a given by
Murotsu (10~ 10)]

......
io"
""'- .41"

....w
""
TABLE 10.9 Corre!aticm> between generated mec hani.wt~-E.Hmtplc· 10. I 1

Mecha- Corn.:1ations P;;

nism
No. l 2 3 4 5 6 7 8 9 JO 11 I~ 13 14 15 16

1.0 0.65 o.o 0.09 0.551 0.672 0.0 0.0 0.0 0.0 0 58~ 0.197 0.584 0.084 0.197 0 .568
2 1.0 0.435 0.045 0.894 0.905 0.595 0.407 0.407 0.363 0.814 0.866 0.818 0.660 0.577 0.973
3 1.0 0.0 0.347 O.Q3l 0.0 0.965 0 .%5 0 .866 0.045 0.571 0.06 0.041 0.921 0.389
4 1.0 0.453 0.054 0.0 1).0 00 \JO 0.479 0.012 0.48 0.015 0.0 0.109
5 1.0 0.835 0.615 0.335 0.335 () 307 0.946 0.79() 0.953 0.651 0.454 0.921
6 1.0 0.726 0.014 0 014 0.0 0.88:1 0.739 0.88 0.784 0. 195 0.9{}(;
7 1.0 0.0 0.0 0 .0 0.65! 0.7lC 0.652 0.977 0.00 0 .690
8 1.0 1.0 0.977 0 021 0.535 0.062 0.019 0.838 0.375
9 1.0 0977 0.02! () 535 0.062 0.019 0.838 0.375
10 Symmetrical 1.0 00 0.477 0.061 0.0 0.727 0.345
11 1.0 0.66S 0.095 0.689 0.216 0.849
12 1.0 0.672 0.823 0.602 0.886
13 1.0 0 .683 0.210 0 .857
14 1.0 0.068 0.722
15 1.0 0 .509
16 1.0
 315

I• the safety margin for the potential hinge scl:tion i , let the C\' Cilt

E; o·= P(Z; < ())

flustratcd method, when the hinges were selected at any stage along
hi• Opath before a mechanism is formed, the hinge location/~ (over the
1 II eligible locations) with the greatest probability of failure (i.e . min .
, elected. A better, or more rational logic may be to select the next
1 cation such that the joint probability of occurrence of the hinge
including the selected hinge location, is maximum. That i ~ . at the
t ~e, select jth hinge location such that
P(ErnEln . .. n£,) =max [P;(Ern£!() . . . nE}J] (HUll
i

the maximum is over the set of all possible hinge locations at that
It II n stage, that is, all hi ngc locations other than the sections where
have already formed All events t'; arc correlated because of the
unmon load variables in all equations for Zt. The evaluation of the joint
b bility is complicated and time consuming. In the proposed method.
hi joint probability was never used. Murotsu ( 10.10) used these approxi-
ti n , given below, which might be the upper bounds f'or Ey . (10.31).
P(E• n £2 . .. n 1:}) < max [Pll:j)J i? (I 0.32)
i

II P(E• n £2 n ... n t}) ~ max I_P;(£; n EJ) I i? 2 ( 10.33)

ll ng these, he proposed a branching and bounding ;dgorithm for the

ilneration of dominant modes (I 0.1 0, I 0.11 ).

10.7 RELIABILITY .ANALYSIS OF RCC FRAMES

10.7.1 Introduction
'fh e failure of a frame structure by the formation of a collapse mechanism
requires a large rotational capacity of plastic hinges. Steel structures satisfy
these requirements. Nonlinear and inelastic deformation characteristics of
R C structures do not allow to use the available resistance of sections to
maximize the structural reliability. Moment-rotation relationships and the
limited rotation capacity of RCC sections pose difficulties and limitations
in the reliability analysis of plane frame structures.
The reliability analysis of RCC frame structures was initiated by Ticky
and Vorlicek (10.15). They formulated the reliability of RCC structures
ubjected to loads from one or several sources based on the ultima te load,
and it was shown how the deformability (ductility) of critical sections could
be taken into account in studying RCC frames. Webster (1 0.16) presented
a probabilistic procedure to forecast the performance of RCC fram es sub-
jected to an arbitrary number of sequential loads. Chou , Mcln to h and
Corotis (10.17) had investigated the correlation between resistance and
 316

reliability fo1 a simple RCC frame with known collapse mechanisms.

Ranganathan and Deshpande (10.18) p1csented a method of the reliability
an:llysis of RCC frames, con sidering the limited rotation capacity of con-
crete sections. A reliability model compatible with the collapse mechanism
wa s proposed for the rotation failure mode and the method was illustrated
with examples. Th e sam e thing is presented in this ~ection .
In the case of RCC frames, any critical section, hinged earlier, may fail
due to an insuflicient plastic rotation capacity before a collapse mechanism
is formed This mode of failure will be called a rotation failure mode.
Considering the limited plastic rotation capacity in RCC frames, a method
is suggested lo verify and analyse the identified dominant mechanisms and
to generate rotation failure modes, if necessary. A reliability model com-
patible with the collapse mechanism is proposed for a rotation failure mode
on the basis of partial utilization of the plastic moment capacity of an
incipient hinge section at the failure stage . The rotation failure modes are
then combined with other possible mechanisms to assess the system
reliability of a RCC frame. The method is illustrated with examples .

10.7.2 Strength and Stiffnl'SS of RCC Sections in Flexure

ldealisations
The nonlinear stress-strain and moment-rotation relationships of RCC
sections pose some difficulty in the assessment oftheir strength- stiffness
properties, required to carry out a reliability analysis for RCC frames.
Using the idealized stress-strain curves for concrete and steel, shown in
Fig. 10.21, and the bilinear moment rotation diagram of RCC plastic
hinges, shown in Fig. 10.22(a), computational methods are developed to
., determine the moment capacity, flexural rigidity, and rotational capacity of
RCC plastic hinges. Furthermore, nonlinear behaviour is approximated as
linear elastic and piecewise linear elastic-plastic (PWLEP) to simplify the
structural analysis. Also, first-order second-moment (FOSM) method is used
1 ~I to formulate the reliability analysis.
The limits It and h, shown in Fig. 10.22(a), are considered as idealised
,) elastic and plastic limits respectively (I 0.19). The elastic limit II corresponds
to either a maximum compressive strain in the concrete, eel, equal to 0.002
[Fig. 10.2l(a),] or the yielding of steel by attaining the yield strain esy in
mild steel bars as shown in Fig. 10.2l(b), or offset strain of 0.001 in high
yield strength deformed bars as shown in Fig. 10.2l(c), according to which-
ever condition is attained first. The resisting moment of the section corres-
ponding to l1 in Fig. 10.22(a) is termed as the yield moment, My. The plastic
limit h is attainecl when either the concrete or steel fails corresponding to
the maximum strain fc2 (taken as 0.0035) for concrete as shown in
Fig. l0.21(a), or strain ofO.Ol in steel as shown in Fig. 10.2l(b)or 10.2l(c).
The resisting moment of the section at h is taken as the plastic moment, M.
 317

•ct •c2
0 002 0 0035
Strain
(a)
:tem

,, '2
0 975 fy

0 Bfy
tee
ness Ill I
nes.
n in
-
Ill
Cl>

.ii I
I I
I
tstic I I
_______ _!_
i to 0 •sy •s 2 0 t·ftoy -- - ___l__
•s'Z
ybf 0 01 0 001 0 01
I as Stram Stram
the
( bI (c)
Jsed
FIG. 10.21 Idealized stress-strain curves for (a) concrete (b) mild steel and (c) cold-
worked steel
ised
mds Re$isting Moment and Flexural Rigidity
.002
The resisting moment of a rectangular or tee section at limits It or 12 can
IY in
be computed by satisfying force equilibrium and strain compatibility.
ligh
Referring to the bilinear moment-rotation diagram, shown in Fig. 10.22(a),
ich-
the flexural rigidity, El, of a RCC member is assumed to be constant in the
·res-
range 0 to II. Also, EI is assumed constant between critical sections of the
~stic
member. The value of a uniform EJ obtained from sh:ess and strain con-
to
ditions at limit It is given by (10.19)
m
(c). EI = Myct = My(d - Ct) (10.34)
M. ect e,y
where ct is the depth of neutral axis at limit It and d is the effective depth.
318

rotat10n,e

0 Rotat1on

I al

Plast1c h1nge

I
rll
----:::-:"' ~
-~J r:l\
I.V
2 ,L~J ___ ..........

~Z = Rotal•on of node 7
=Slope at end 7
of member 1
4>J =Slope at h1nged e-nd J
of member 7

8 = Piast.c rola!IOI'I at J
=¢J-(b

(b)

FIG . 10.22 (a) Moment rotation diagram for a RCC

plastic hinge and (b) plastic rotation at a
hinged section

Plastic Rotation Capacity

The rotation capacity ofa RCC plastic hinge, 8, shown in Fig. 10.22(a), is
the angular rotation which the section can sustain under the constant
plastic moment without the local failure of the section due to limiting strain
conditions at the plastic limit h defined earlier. The plastic rotation capa-
city of a section depends on the (i) material properties, (ii) amount of rein-
forcement, (iii) confinement percentage, and (iv) axial load. The plastic
rotation capacity can be derived (10.19) as a function of the (i) ultimate
strain of concrete, (ii) strain variation in concrete from II to h, (iii) spread
of the plastic zone, i.e., the length of the plastic hinge, and (iv) position of
the neutral axis.
 319

I tensive experimental investigations show large variations in the plastic

'' tnlion capacity of RCC regions. Empirical formulae have been repurtcd
v rious research workers. Baker and Amarakone (10.20) have proposed
t of curves representing the permissible plastic rotation capacity, 811 , as
function of the depth of the neutral axis at limit h for different percent-
r of confinement in the member. Hence, one can use these curves to get
rrnissible rotation.

10.7.3 Statistics of Plastic Moment Capacity

1rlations in geometric parameters of a section, generally being small, are
n l(lected . Hence, for establishing statistics of M, the random variations of
/,u nnd fy only are considered. Using the developed prediction equations
h r M, and using a first order approximation, the mean value and standard
d viation of Mare calculated from the known statistics of t:·u and (y.

W.7.4 Reliability Analysis of RCC Frames (10.18)

.lutamutic Generation of Dominant Mechanisms
After c.,tahlishing the strength and stiffness properties of RCC members
ltHI the statistics of M of various critical sections, stochastically dominant
m· hanisms are generated using the stiffness method of linear clastic and
I'W LEP analysis of the structure, and FOSM method of reliability analysis,
11 uming. initially an unlimited rotational capacity available for all plastic
hinges to form a collapse mechanism. To simplify the analysis, the axial
1 idity. T:A, and flexural rigidity, £/, are assumed as deterministic. The
~ •q ucntial selection o!' the most probable hinge locations, to determine the
\'I of plastic hinges which converts the structure to a mecha11ism having a
11rgc probability of failure, is the key consideration in the process of gene-
rul ing the dominant mechanisms. Methods suggested by Murotsu (10.10)
nr Ma and Ang (10.8) or Tang and Melchers (10.13) can be used to gene-
rate stochastically dominant mechanisms; however, using the technique
(10.14) explained and illustrated in the previous section. stochastically
dominant mechanisms are generated.
Checking of Plastic Rotations
The technique for generating stochastically dominant mechanisms selects
is
the plastic hinges on the basis of~ and determines the set of hinges which
tnt
converts the structure into a mechanism without verifying the plastic rota-
1in
)a-
tions of the hinged regions. It is observed that this set of hinges may consist
of inactive hinges; moreover, the sequence of hinges may be random with
in-
respect to load factors. When the actual plastic rotation of plastic hinges is
tic
to be checked against the permissible plastic rotation, the physical process
lte
of the sequential occurrence of plastic hinges due to load increments has to
ad
be considered. Therefore, the sequential analysis of dominant mechanisms
of
based on the load factor is employed to check the plastic rotation of hinges
 320

at each stage of the 'c4ucntial an•li~ ~is, anu Lo fl,rmulate the failure prob-
ability if the rotation check fails.
The plastic rotation is assumed Ill be concentrated at the critical section.
Therefore, the relative slope at the node of the plastic hinge of the section
is considered as the plastic· rnt:uion of the section, i.e. the angle of di:-.-
continuity as represented in Fig. 10.:22(h). The rotations of nodes. obtained
from the analysis, correspond to the slopes at the intact ends of the
members meeting at the node, and not to the hinged ends. The slope at the
hinged end of a member is obtained by slope deflection equations of the
corresponding member. Then the plastic hinge rotation, as shown in
Fig. 10.2~(h), is given by the difTerence between the slope at the hinged end
and the rotation ofthe corresponding node.
An identified mechanism with known active hinge sections is regenerated
sequentially for checking the rotations ofhingc ~ections on the basi~ of load
factors. The load factor. 'I• for :1 potentirll hinge secti,,n 1 at ;1nv stage i<;
given by

1); (lfU5l

At any stage, let the selected potential hinge section be i ha\'ing the
lowest load factor ,,,_and the earlier hinged sections be i and k. The actual
\\ plastic rotation r1l j (or k), fl, i\
Ill
"
0; -· l' O;,tvf, !__ l/i l ~- llj,Q,] flfU6)
s-1 1-1

where Af, is the plastic moment capacity Pi' the critical sections, Q1 is the
lth applied load on the structure. e;, and O;t ;He the plastic rotations at the
hinged section i due to the unit plastic moment M, and unit load Qt
respectively. (Jj, corresponding to 1\f, of the nonhinged section is zero. It is
possihle that (Jj and/or e, may exceed permissible plastic rotation capacities
(Jpj and epA respecti\'ely. In such a cusc it is not possible for a hinge to be
formed at section i as indicated in Fig. 10.23. If fl ; > Op;. then, considering
~ . -,
' f1 I ''
I
L -- J

_; _ _j ------
1 \
' I
)'R Rotat•on 1atlurf' modf'

AI 1"), ,e 1 >ep 1 afld /or

eM> epk

FIG . 10.23 Checking of plastic rotation of hinged sections during regeneration

of a mechanism
 321

ro b· rotation capacity, the loact factor >.j at which 01 = 0"1 is glvon b~

tion , m
:tion [Opj- E 8;,M,l
di - Aj = n •··I (10.37)
[ E 8"Q,J
t~l

un ilarly, if 8k > 8pk, the expression for the load factor >..k, at which
.., fJ,k, can also be obtained. Aj and >.k are now compared and the lowe.st
I ted and denoted by >.R. At this value of >.R, the RCC frame is assumed
111 fail under the rotation failure mode, prior to the formation of the
111 ·hunism. As the rotation check fails, a full strength of section i is not
ated Jlilized. Whereas a plastic hinge at section i forms at the load factor.>.R,
I lld lit coefficient a;;, instead of being unity, is modified as (I 0.18)
~e is
a;; = [ j}' OijMj]- >.R[
J-1
E b;kQk] (Ml;)
k=l
(10.38)
Ni

.35) , ub tituting this value of au in Eq. (10.26), the safety margin Z; of the
1 1tation failure mode is formulated and the reliability index {3; ·and prob-
ubility of failure pu are calculated as usual. The process of regeneration of
the the mechanism is terminated at this stage.
tun I Likewise, all dominant mechanisms identified earlier, assuming full
redistribution, are regenerated and analysed, in addition, a plastic rotation
heck is performed for hinged sections at every stage. All failure modes are
.36) · mbined for establishing bounds on the system failure probability. Hence,
the proposed formulation of the re!:ability analysis of RCC frames involves
the following steps: (i) analysis of 'RCC cross se.ctions of beam& and coiumns
the
the
nd establishing statistics of M;, (ii) generation of dominant mechanisms
11 suming unlimited rotation capacity of scct.ions, and a reliability analysis,
j Q,
iii) regeneration of failure modes with checking of plastic rotations of
t is
hinged sections, and (iv) synthesis of all failure modes and assessment of
ities
pr.. A flowchart for the reliability analysis of RCC frames is given in
• be
ig. 10.24. The proposed method is illustrated in the following examples
ring
(10.18).
EXAMPLE 10.12 The simple one-bay one-storey RCC frame, shown in
"Fig. 10.25, has been designed as per ISS (10.21) with the following data:
(i) Characteristic loads:
Live load : 4 kN/m 2
Wind load : 1.5 kN/m 2
(ii) Load combinations with partial safety factors:
(a) 1.S(D + L)
n
(b) I.5(D + W)
(c) 1.2(D +L + W)
322

8 r
- - - - _ _j__ - - - -- ---.,
DATA. cross sectional properties;
statistics. of basic variables;
l __gp~~ it~~eoctions _ _ _ ___.

Computr: )J. an d cr of M of I
. . ._ - - - - - ·- -------r- -
I cnt1c:OI sections

r·G P"flration of dom1~~-~l ,_;;~-hanis,.;Sj

Lassuming full rt>distribution J
-- --- -t==: -=:~-
Choou an identif ied mechanism
-
and h1ng es 1nvolved

r------o-4~;; ~ ne- 1[K J fl

]_-- _...J Formation of
- -- -- mechanism failureo
----1.:._1[ KJ I S o;=_ >---'V-=-t~!>- - . -""""""mode

1~. .-;;-
IN_o_d_a_l _diiplacemf.nt and ·-·]
' Formation of

1
- ---r·-----'
m ~mber ~nd forces _

- ~--,
Load f actors a~dSfl l r:c li ~ :--e- 1-i a-b-i -1it y
!
~ f hinge from 91 V l'~ tw~f's__j
1.-:R
analysis of
\ failurt> m o~
Plastic rotations of hlng~td
sections

~----L-------~
.___-! Formation

FIG. 10.24 Flowchart for reliability analysis of RCC frames

(iii) Char:1cteristic strength of materials:

Concrete (M 20) : 20 N/mm 2
Steel (Fe 415) : 415 N/mm 2
 us
D,L

T
lo()()()

, ®
3 1
~ 3000 -+--- 3000 -f

Ast :670mm
2

~·
I

400
T
JSO l • I
Js
l • 1 l j_
A•t :720mm 2
1-XK> -I 1-Joo~
Section 1,2,3,4 Section 5,8

t- 1-- - -- 1700

T I.___ I~
0 I I. I
L
r-300 --I
1 Ast :1660mm2
T1
Section 6, 7
FIG. 10.25 One-bay one-storey RCC frame and detai Is of cross ·
sections-Example 10.12

(iv) Partial safety factors for material strengths:

Concrete : Ymc = 1.5
Steel : Yms = 1.15
(v) Young's modulus of elasticity:
Concrete (M 20) : 25.5 kN/mm 1
Steel : 200 kN/mm 2
 324

Using the strain compatibility ondition, and force and moment equili-
brium equations, expression for ultimate resisting moments of rectangular
and tee beam RCC sections can be written in terms of design parameters
and basic random variables ..fcu and jy ( 10.19). For the developed expression
for M, using the Monte Carlo technique or first order approximation
[Eqs. (3.82) and (3.84)], the mean value and standard deviation of M can be
computed using statistics of /cu and /y. Let us assume that this has been
done and the computed values of the mean and standard deviation (or
coefficient of variation) of moment capacities of critical sections are known.
They are given in Table 10.10 along with other data (including Bp) required
for the reli ahility analysis. The reliability analysis is carried out for two load
combinations, viz. (i) D + Lrn + Wapi and (ii) D + Lapt + Wm .
TARLE 10.10 Properties of cross sectio11s and statist in of l'ariables for
RCC frame · Examp/e 10.12

Sec tion or EA El Ill,

variable (kN) (kN m 2 ) (radian) I'

Section
1' 2, 3, 4 0 .356 10' 0.915 IO' 0 .017
5, 8 0.949 : 10' 0.3 92 10' 0 .018
0, 7 0.949/JO' 0,392 . 10 5 0.019
Variable
feu 26. 8 1 N/mm' 0. 150
fy 469 0.100
,\/,, 1\1 2 , M,, ,' f , 122 .3 4 kN 11 1 ().()82
.·H ,, /\I 8 1Jo .37 0.093
1\f, , M , 332 97 0 ,099
Mcan/nominnl
D 1.05 0 . 100
l (L 111 ) 0.55 8 0.334
L(Lapo .l 0.319 0.397
W(W 111 ) 0 6<) 3 0.236
IV(Wapt) 0.200 0 420

l Remark : All variables are statistically independent .

I
Case (i) D + Lm .. !. Wapt
Assuming full rotation capacity at all critical sections for the formation of
mechanisms, and using the method (10. 14) explained in Sec. I 0.6, stochasti-
cally dominant modes are llrst generated, as shown in Fig. 10.26, for the
load combination D + Lm + Wnrt · The identified mechanisms are ordered
and the system reliability is assessed from the synthesi s or theses mechanisms ,
which are represented by their hinges as shown in Fig. 10.27, and safety
margins as given in Table 10.11. The correlations between dominant failure
modes are computed (Table I0.12). Details of the identilled mechanisms
and results of the reliability analysis of the frame, assuming full redistri-
bution (i .e. without limiting the pla stic rotations of hinge s), are presented
in Table 10.11.
 32&

·j..
lr
4·33 S·7S S·89
ra
m
•n Fir•t
>e t,.. 6·03
:n
>r
1.
Intact
:d structure
.d 6·72 5·89
J =1

8·47

iO
lO
l2
n
)9

)0
!4
l7 6·87 6 OJ
16
!0

,f
J•
e FIG. 10.26 Failure tree diagram for RCC frame in Fig. 10.25 under
D + Lm + Wap 1-Example 10.12
d
;,
y Each identified mechanism, indicated in Fig. 10.27, is regenerated, as
e shown in Fig. 10.28, according to the procedure outlined in the flowchart,
.S given in Fig. 10.24, for checking the plastic rotations of hinged sections
against their permissible plastic rotations, Bp. The final failure modes,
corresponding to the possible mechanisms or rotation failures are generated
as explained in Sec. 10.7.3, and are given in Table 10.13. Correlations
 326

4 4 5
2 7

MECH 1( fl: 5·89) MECH 2(f"l=603)

7 ~ 7 8

MECH J( fl =6 OJ)
[ l
MECH4(fJ=6·17)

\.~
I
'I
, I
I
'
r r MECH 5( (1 :B 47)
FIG. 10.27
MECH 6(["l: 8 66)
Location of hinges in dominant mechanisms of RCC
frame in Fig. 1 0.25 under D ·I· Lm 1 W0 p 1-Example
10.12

TABLE 10.11 Identified Meclwnim1s and Re1111t.l oj Reliahility Analysi.1· of RCC

Frame in Fig. 10.25, Assuming Fn/1 Redistril)(ltioll Undl.'r

)'t
D + Lm ; Wap 1-Example 10.12

Sl. Hinged Failure

Safety margin fJ Pr
No. sections tree

I. 2, 4, 7 1.0M 2 -1- !.OM• + 2.0M, - 3.0D - 3.0L 5.89 0.197 x JO-• I

2. 4. 5, 7 1.0/114 + !.OM, + 2.0M7 -3.0D - 3.0L 6 .03 0.821 x JO-" I
3. 2, 7,8 !.OM 2 + 2.0/lf, + I.OM, - 3.0D - 3.0L 6.03 0.821 /~ JO-D 2
4. 5, 7,8 1.0!1/ 5 + 2.0M, + I.OM. - 3.0D - 3.0L 6.17 o .338 ·...: 1o-• 2
5. ' · 3. 4 , 7 I.OM, + 1.0,'\!f, + 2.0M 4 + 2.0M 7 8.47 O.l24 x Jo- n
- 3.0D - 3.0L - 4.0W
li. I, 3, 7, 8 I.OM, + 1.0M, + 2.0/11, .:.. 2 .0M 8 8.66 0.239 :.-: 1o- 17 2
- 3.0D - 3.0L - 4.0W

Bounds on system probability of failure

Simple 0. 197 ·• 10- • ~ f'rs ~ 0 )95 v 10-•
Narrow 0 215 >~ 10- • .;;;; Prs ~ 0 .284 ; Io-s

dl. \..
 327

Correlations betw~cif mechanisms given in Tllbi•IO.IJ- ' •mp/1 10.1~

Pq
2 3 4

I 1.0 0.981 0.981 0.962

'
0.971 0.927
6

2 1.0 0.962 0.981 0.966 0.923

3 ;.0 0.981 &.938 0.968
4 1.0 0.934 0.963
Symmetrical
1.0 0.929
1.0

I' OLE 10.13 Regenerated failure modes and results of reliability a11alysis of
RCCframe in Fig. 10.25 under D + Lm + Wap 1-Example 10.12
Hinged
. sections
Safety margin fl Pr Remarks

2, 4, 7 l.OM2 + !.OM, + 2.0M 7 - 3.0D 5.89 0.197 X IQ-B Mechanism

- 3.0£ failure
4, 5, 7 ! .OM, + 0.939M t 6 2.0M7 5.94 0.1 44 xto-s Rotation
- 3.0D- 3.0L failure
2, 7, 8 l.OM2 + 2.0M + t .OM
7 8 6.03 0. 821 X 10-o Mechanism
3.0D- 3.0£ failure
4' 5, 7, 8 0.99JM5 + 2.0M + t.OM 7 8 6.16 0. 368x w-t Rotation
- 3.0D- 3.0£ failure
3, 4, 7 0.620M 8 l.OM, + J.SM.
t 6.02 o .861 '< 10- 9 Rotation
- 0.22SD - 0.225£ - 2.0W failure
6 3, 7, 8 0.636M I + l.SM. + l.OM, 6.23 0.235x w-o Rotation
- 0.225D - 0.22SL- 2.0W failure

Bounds on sys lem proba bility o f fa ilure:

Si mpl e 0.197 x 10- s ~ '' rs ~ 0.57 x 10- 8
Narrow 0.245 X 10-e ~ Pr, ~ 0.37 1 >: 10- a

TABLE 10.14 Correlations between failure modes given in Table 10.13-Exampfe 10.12
e Failure Pu
mode No. 2 3 4 5 6

l. 1.0 0.982 0.962 0.986 0.981 0.959

2. 1.0 0.981 0.983 0.983 0.956
3. 1.0 0.959 0.981 0.982
4. Symmetrical 1.0 0.963 0.967
s. 1.0 0.986
6. 1.0

between failure modes are computed (Table 10.14) and the bounds on
system failure are established. Results of the same are given in Table\!0. 13.
 328

MECH 1 Mtchanism failur• modt

[3 = 5 ·8!'

r·- -,
2 ·19 I
t.----·
I

MECH 2 D----0--0 /- 1 98 2 ·1J

-- -1. ... 5.,~
2·16 Rotation fa•lureo mode
~ =5 ·94

2 ·19
/- ,
MECH 3
,_8 / I Me-r.han ism failure mode
fl =6-0J

r---,
I2 ·22
L __ _,I

MECH 4 D----0--0 -- ~_rs)

/-
1 98 2·20

2 21 Ro\at1on failure mode-

f3=61Ei
r-- -~
I 2 31 I

MECHS D----0--0 /-' 1 98 2 1J ~-- ~

---1 , 3_1
~
I
I
2 17 Rotat•on failure mode
n ~6 o2
r - ... ,
I 2 )5 I
1 98 2 20 L- . - J
/-,
MECH 6
D----0--0- - -i J I
' -
2 22 Rotation I ai lure mode
fl =6 23

2 19 2 19 Indi cates toad factor of h1nge section 5

0
2 16
5 Indicates h•nge sect1on
2 16 Indicates the load factor for rotation
fa i lure mode
FIG. 10.28 Regeneration of individual mechanisms in Fig. 10.2 7 for checking
plastic rotations under load caseD Lm + +
Wap 1-Example 10.12

Case (ii) D + Lapt + Wm

The procedure is repeated for this example to assess the system reliability
under the second load combination D + Lart --:· Wm. The genera ted domi·
nant mechanisms, assuming full redistribution , and their safety margins and
resulh or the reliability analysis h:tscd on these are given in Table 10.15 .
The mechanisms arc rc:'generated as sbown i11 Fig. 10 29. Table 10.16 gives
 329

'!'ABLE 10.15 Identified mechani.\'111.1 and remits of reliability analysis of

RCC frame in Fig. 10.2j, assuming full redistribution under
D + Lapt + Wm-Example 10./1

Hinged Failure
Safety margin Pr
sections tree

2. 4, 7 I.OM, -i- 1.0.11 4 '- 2.0.\f, - 3.0/J 6.79 0.567 X I0- 11

- 3.0£
4, 5. 7 I.OM 4 + 1.0,\/ 5 + 2.0:\f, - 3.0D 6.93 o.2o6 >~ 1o-u
- 3.0£
2, 7, 8 I.OM~ +2.0,\-£7 1.0;\/, - 3.0/J 6.93 0.206 .< l O-'~ 2
- 3.0£
5, 7, 8 t.Ot\1 5 + 2.0M, + l.OM,- 3.0/J 7.0H o. 736 ;; 1o-•• 2
- 3.0£
I, J, 4, 7 I.OM1 -!· I.OM, :- 2.0,'\o/ 4 2.0M, 9. 14 < IQ-18
3.0D - 3.0£ - 4.0W
I, 3, 7, 8 I.O.M1 ~- 1.0.\/, -' 2.0.\1, 2.0M 8 9.32 < Io-u ' 2
- 3.0/J - 3.0L - 4.0W

Bounds on system probability of failure :

Simple 0.5&7 x J0- 11 ~ Prs ~ 0.105 :< 10-"
Narrow 0.626 A to-n~ Prs,; 0 . 812 :~ I0- 11

OLE 10.16 Regenerated failure mode.f amlresulls vf reliability analysis of RCC

frame in Fig. 10.25 under D + Lapr , IVm -Example 10.12

Sl. Hinged
Safety margin f3 pr Remarks
sections
- - - - ---- - - - -- -
2, 4, 7 0.877M, -;- l.OM 4 2.0M, 6.61 0.193 · t0- 10 Rotation failure
-3.0/J - 3.0L
4, 5, 7 I. OM I 0. 786M 5 -' 2,0(1•/, 6.60 0.211 ., I0- 10 Rotation failure
-3.0/J- 3.0L
2, 7, 8 0 .931 t1f, + 2.0.-H, -!- I. OM, 6.83 0.414 :· 10 · II Rotation failure
-3.0/J- 3.0L
4 5, 7, 8 0.835M. ' 2.0M, ! I. OM 8 6.82 0.457 ;< to - u Rotation failure
-3.0/J - 3.0£
3, 4, 7 0.729Al, + I.OM, · - l.5M7 6.93 0.218 ;.: to-u Rotation failure
-2.25D - 2.25L - 2.0W
3, 7, 8 0.749M1 !- 1.5M, + l.OM, 7.14 0.469 >: to -a Rotation failure
-2 .25D - 2.25£ - 2.0W

Bounds on system probability of failure:

Simple 0.211 ~< I0- 10 ~ Prs ~ 0.518 :< IQ- 10
Narrow 0.289 x I0- 10 ~ Prs ~ 0.378 ;.: 10-1o

ity the final failure modes obtained after checking the plastic rotations of the
ni- hinges. The correlations are computed. The estimated system reliability is
nd given in the same Table 10.16.
15. EXAMPLE 10.13 The two-bay two-storey RCC frame, shown in Fig. I 0.30,
1es bas been designed as per the ISS (10.21) with the same data given in
330 s

r-- 1
I 2 41 I
222 2·)1 L-- ..J

MECH 1
G---0---0-- - --,I 2 ' '
2 37 Rot at1on

r-- 1
I 2 45 I
2 22 2· 31 L. - - .J

ME CH 2 G---0---0- --- ~,s~'l

2 J7 Rotat1on fa1lure mode
fl = 6·60
r- -,
I 2 '45 I
2·22 2·J8 L--J

MECH J
n {.;\ ._____!;;\__ _ - - -i ~
'_ I
2·\
2 42 'Rotat1on failure mode
fl = 6 BJ

r.--,
12' 48 I
2·22 2·38 L.--...J

MECH 4 G---0---0---- -(,~~

2·42 Rotation failure mode
n=6·82
; 2-48;
{. 22 2 ·31 L- - ..J

MECH 5 []---0~ -----<~;

2· 37 Rotation failure mode
[l ~ 6·93
,-- ,
2· 22 2 )8 :...2 ~: 3 J
~-

MECH 6 - --- ~ 3 'I

'- '
2·42 Rotat1on failure mode
n= 7 ll,

FIG. 10.29 Regeneration of individual mechanisms in Table 10.15 for check·

ing plastic rotations under load caseD Lapt + W,-+
Example 10.12

Example I 0.12. Details of cross-sections of the frame are given in

Fig. 10.30. Flexural rigidities, plastic moment capacities, and permissible
plastic rotation capacities of sections are calculated and given in Table 10.17.
The results of the reliability analysis for the two load combinations
(i) D + Lm + Wapt and (ii) D + Lapt + Wm are given in Tables 10. I 8,
10.19, 10.20, and 10.21.
 331

j Dz ,Lz 04,L4
Q)g
4
10
0)
11 12®21
16
22
@
23 24
@128

4000

0 3 .L3 I
)

2
5
@
6
® 14 ®
19 20
26 ®--+
I
4000

e
l CD JOOO --4.-- 3000
1)

® 3000 - l --3000
25

~ ®
_l
I

·w/
Jo
Asc :480mm 2

·T
350
-r/
s~
Ast :900mm2

"T
I I
Ast =1650mm2

4~5
I

1_ . . 1 I ~, 500
I
2 ..L ..L L~.-.-
· ~J
L
Ast=480mm
300 ~ j-300 -----.j 1---)00 --4
Asc;JJO mm]

Section 1, 2, 3, 4, Section 5.9.20.24 Sect10n 8,12.17. 21

13,14,15,16,
25. 26, 27,28
~--------- 1700
,-----------~ _l__
-.-
II
I
r------__j ~ T
500 465

l ____ •
..__ 300 ~ Ast :1)40mm2
Section. 6, 7,10,11,18,19,22,23
in All th• d1m•ns1ons ar• 1n mm
>le
7. FIG. 10.30 Two-bay two-storey RCC frame and details of cross-sections-
Example 10;23
ns
8,
 332

TABLE 10.17 /'roperries of cross sec/ions unci statistics of 1•ariables for RCC frame-
E,ample 10./3

Section or EA El 8p
1L 8
variable (kN) (kN m') (radian)

Section
1to4,13 to 16. 0 356 ~< 10 7 0. 786 ;< 10 1 0.014
25 to 28
5, 9. 20, 24 0.453 ~< I0' 0.294 x to• 0.017
8, 12, 17. 21 0.453 ~<I 0' 0.294x 105 0.013
6, 7, 10. II. IH, I<J. 0.453 ·:I 0 7 o.294 >· to• 0.017
22, 23
Variable
M;. (i , I to 4. 114.78 kN m 0.058
13 to 16, 25 to 2HJ
!vf;, (j c~ 5, 9, 20, 24) 178.49 0.092
Mk, (k = 8, 12. 17, 21> 315.80 0.090
M,, (I , 6, 7, 10, II. 274.11 0.098
18, 19, 22, 23)

Remarks: Statistics of variables feu• fy• D, Lapt• Lm, Wapt and Wm are the same as
giwn in Table 10.10
All variables are statistically independent.

TABLE 10.18 Identified meclwnisms and results of reliability analysiJ of RCC frame
ill Fig. 10.30 assuming full redistribution 1111der D t Lrn + w.P,-
Example 10.13
t\ Sl. Hinged
Safety margin
Failure
f3 Pr
l No. sections tree

4, 11' 12 !.OM, + 2.0Mu -1- I.OM12 7.17 0.38 ;< JO-ll

-3.00 2 - 3.0L,
2 21,22,28 1.0Mu + 2.0Mn + I.OM2s 7.17 0.38 x to-u 2
-3.001 - 3.0L,
3 9, 11, 12 I.OM 0 + 2.0Mu -1- I.OM11 7.85 0.216X JQ-U

\l
)
4 17, 18, 20
-3.00 1 - 3.0L,
l.OM17 + 2.0/vlu + l.OM1o 7.85 0.216x JQ-u
-3.003 - 3.0£3
5 21, 22, 24 l.OM11 + 2.0Mu + !.OM.. 7.85 0.216X JQ-U 2
-3.00, - 3.0L 1
6 5, 6, 8 !.OM. -1- 2.0M 0 + J.OM 0 7.85 0.216x w-a
-3.00 1 - 3.0L,
7 2, 3, 6, 8 l.OM1 + !.OM,+ 2.0M 0 8.62 0.347 X to-n 2
-1- I.OM 8 - 3.001 - 3.0L 1
8 17, t8, 26,27 I.OM17 + 2.0Mu + I.OM, 1 8.62 0.347 x to-n
+I.OM., - 3.001 - 3.0L1
9 4, 10, 16,21 !.OM, + 2.0M10 + l.OM11 8.62 0.347 x to-n 2
+ l.O!v/11 - 3.001 - 3.0L 1
Bounds:
Simple 0.380x 10-11 .,;; Prs.,;; 0.769x J0- 1•
Narrow 0.765 ;..c to-•• ~ Prs .,;; o. 765 x to-u
 133
Rcgenemud failure modes and results of rtllablllty Q/IQ/)1111 tf/ ltCC
frame in Fig. /0.30 under D + Lm + Wap 1- E .I'lii/I(JI1! 10.1.1

Hinged p
Safety margin Pr Rom arks
sections

4, II, 12 I. OM t -i- 2.0M11 + I.OMu 7.17 0.38 X JO-ll Mechanism

- 3.0D2 - 3.01,• failure
21,22,28 I.OMu + 2.0Ms• 7- J.OMu 7.17 0.38 x J0- 11 Mechanism
- 3.0D, - 3.0£, failure
9, II, 12 0.671M, -1- 2.0Mu + t:&Mu 7.19 0.330 X JQ- 12 Rotation
- 3.0D 1 - 3.0£ 2 failure
17, 18,20 I.OM17 + 2.0Mu ·~- 1.0Mu 7.85 0.216x w-u Mechanism
~3.0D 1 - 3.0£ 3 failure
21, 22,24 I.OM11 + 2.0M 2 ~ -!- 0.714M 1, 7.28 0.172 :~ I 0- 1• Rotation
-3.0D, - 3.0L, failure
s, 6, 8 l.OM5 -1- 2.0Ms -!- I.OM 8 7.85 0.216 x w-a Mechanism
- 3.0D1 - 3.0£ 1 failure
3, 6, 8 0.9SSM, + 0;967M 0 8.27 0.605 ;< JO-ll Rotation
-t-0.484Ms - 1.4SD, failure
-0.32SW,
17, 18, I .OMn + 2.0M11 + 0.792M2s 8.32 o.438 x to-••
Rotation
26, 27 -t-I.OM 27 - 3.0D1 - 3.0L1 failure
4, 10, 16 l.OM, -1- 2.0M10 + 0.374M" 7.72 0.593x Jo-u Rotation
21 + l.OMn - 3.0D1 - 3.0L 1 failure

II \IOd. :
Simple 0.380x I0- 11 ~ Pr, ~ 0.127x to- 11
Narrow 0.952 X 10- 11 ~ Pr1 ~ 0.108 X 10-u

BLE 10.20 Identified mechanisms and results of reliability a11alysis of RCC frame
in Fig, 10.30 assuming full redistribution under D + Lapt + Wm-
Example 10.13

Sl. Hinged p Failure

Safety margin Pr
2 No. sections tree

4, I I, 12 I.OM1 + 2.0Mu + J.OM11 - 3.0D1 8.25 0.809 X J0-11

-3.0L 1
2 21, 23,28 I.OM11 + 2.0M21 + l.OM.. - 3.0D, 8.25 0.809x J0-1.4
~3.0L 1
2 3 17,18,20 l.OM17 -1- 2.0M" -1- l.OMso - 3.0D, 8.94 0.217 X J0- 11
-3.0L,
4 5, 6, 8 O.SM. + l.OM, + O.SMs- I.SD1 8.94 0.217 X }()-' 11
-1.SL1
.s 9, lJ, 12 l.OM1 + 2.0M11 + l.OM11 - 3.0D1 8.94 0.217 X 10-11
-3.0L1
6 17, 18, I.OM17 -1- 2.0M1a -1- l.OM~t -1- l.OM17 9.78 < ttr- 11
26,27 - 3.0D, - 3.0£1
2 7 12, 16, l.OM11 -1- l.OMu -1- 2.0M11 -1- I.OM11 9.78 < IQ-11
23,28 -J.OD, - 3.0£,

Bounds on system probability of failure:

Si111ple 0.809xJo-u ~P,, ~ 0.162xlo-11
Narrow O.l62x w-u <Pta<
0.162x Jo-11
 334

TABLE 10 21 R;·gen<'mtcd joilure modn 1111d l<'.l llft, of 1rliability Aualrsis of RCC
fwnl<' in Fig. 10.30 under D -' Lart -1 Wm--Fwn1pl<' 10 13

Sl. Hinged
S<Lfety 111argin 8 l'r Remarks
No . sect ion

4 . II , 12 0 .952M 4 - ~ 01/11 I .tJ-1 f" X IS 0.143 10 ,. Rotation

-3.0D,- J .OJ., failure
2 21,2J.2R 1.0,11,. 1- ~ . 0 . \f, I.OM, 8 X.25 O.R09 IO·IR Mechanism
-- J 0/J, - .1.0L, failure
3 17. 18, 20 I OM, . :!.OM" I 0,\ I , 0 ~.94 0.217 . 1()-18 Mechani-;m
-- 3.0D, -- J.OL, failure
4 5. 6, s O.R02M, :. 2 .0 If' 1.0 . 1/~ 8 54 0.694 10'" Rotation
-- 3.0D, -- J.OL 1 failure
5 9. II. 12 o.n12"-'• :. 2.0!1fll I !.OM, R. l J 0.217 'J()-16 Rot<llion
-3.0D,- 3 OJ., failure
6 17, 18. 10.1117 2.Q,\JIA . I I O.lf,. 9.78 < JO-IR Mechanism
21), 27 + I .0:11, - J.OD, - 3 OL, failure
7 12, IIi . I.OM" 0 123.1/" 2 0.1/" ~.44 0.158 10 " Rotation
23 28 -' 1.0 If, -- 3.0D 4 -- 3.0L 4 failut c

Hounds on syo;tem prob;~bility of failure:

Simple 0.217 J0- 15 ~firs~ 0.468 10-"
Narrow 0 379 .· JO-" ~ f'l\ ~ 0.416 10· "

10.7.5 Discussion
A simple and practical method of the reliability analysis of RCC frames,
considering the limited rotation capacity of RCC sections, had been deve-
loped and illustrated. The probability of failure of a rotation failure mode,
generated from the mechanism through a check for plastic hinge rotation ,
is found to be higher than that for mechanisms with unlimited rotational
capacity, which is expected . This increase in pr is observed to be considerable
in the case of the least dominant mechanism.
A comparison of results for limited ductility and full redistribution shows
that the bounds on {Irs are generally higher and wider for limited ductility.
For the two case studies, it is noted that the probability of f<tilure of the
frame under the load combination D + Lm + Wapt is more than that of
under D + Lapt + Wm. The effect of limited rotational capacity on {Irs is
found to be more critical under D + Lapt + Wm than under D + Lm + Wapt
for these two case studies.
For the two case studies of RCC frames (design according to ISS), the
system failure probability is found to be of the order of w- 9 for the one-
bay one-storey frame and 10- 12 for the two-bay two-storey frame. These
values of failure probability are very small. This is due to high design loads
and low material design strengths specified by the IS code.
The checking of plastic rotations of hinges and remodelling of the failure
modes improves the accuracy of the system reliability of RCC frames. How-
ever, the improvement in the present case studies, where the load combina-
tion D + Lm + Wapt is more dominant than D + Lapt + Wm, is not
significant in the context of computer effort.

1
 335

a, Oz

r
OJ
J

~ I.
16m

l.
f----1 6 m ---1
11
FIG. 10.31 Five member truss
n

ism

T
I
10 2Om

o, Oz
f--------2 Om 2 Om---+---2 ·Om- - -!
es,
·e- FIG. 10.32 Indeterminate truss
le,
n,
tal 16
>le 12
I
J-66m
NS
:y. 7 8 11
he
of
is
J
+
J 66 m
I
IPI I. J_

he
e- FIG. 10.33 Two-storey one-bay frame
se
:is 10.8 STRUCTURAL SAFETY IN OTHER FIELDS

re Reliability analysis is a tool in the design process. It can be applied to any

y- field. The importance of making reliability assessments, especially for the
a- purpose of making comparative design judgements, has received recognition
:>t
in the last decade. Even though the reliability analysis and design of steel
and RCC building structures are mainly treated in tlie examples of this
 336

book, methods can be applied to other types of strucures and other fields
of engineering, as well, viz. aeronautical, mechanical and nuclear. The
reliability theory has been used extensively in the analysis and design of
bridges, buildings, transmission towers, etfshore structures, ship structures,
nuclear power plants, and in the development of general purpose structural
design codes. Jn this hook, mainly failure criteria based on strength have
been considered. The reliability methods given can be applied to other
criteria, such as serviceability limit states, viz. deflection, cracking, corrosion;
etc. Fatigue and fracture behaviour is an important consideration in the
design of hridges, offshore structures, aircraft structures, pressure-vessels,
cranes, etc . Hence, reliability predictions against fatigue crack initiation,
growth, and fracture is important. A considerable research has been done
and is going on in developing analytical techniques for fatigue reliability.
1n the case of dynamically sensitive structures subjected to dynamic loads,
the reliability analysis of such structures is more involved. This is so in the
case of deep offshore platforms. Reliability analysis with respect to such a
type of structure is briefly explained below.
Off~hol'e Structures
The safety of an offshore structure depends on predicting the environmental
phenomena. such as wind, current, wave, seismic loading, accurate caku-
lation or the response of the structure to these loads, and determining the
strength of the structure. Level 2 reliability methods have been used in the
\ evaluation of component reliabilities in jacket structures. The various steps

,)
' ,,
I
that are involved for such an analysis are (I 0.22):
(i) defining the basic random variables for the structural resistance and
loading, viz. extreme wind speed, drag coefficient, inertial coefficient,
current speed, marine growth, deck load, yield strength of steeL tube
thickness, leg diameter, damping coefficient, strength model uncertainty,
etc.
(ii) selecting the appropriate failure criterion and the associated model
uncertainty for the component under consideration
(iii) developing an appropriate idealisation of the structure for the
purposes of evaluating combined wave and current forces
(iv) developing an appropriate mathematical model relating the natural
frequency of the structure in its dominant mode of vibration to the basic
random variables which affect it, such as the soil and structure stiffness,
superimposed deck loads, thickness of marine growth, and the coefficient
of the added mass
(v) developing an efficient algorithm to determine the stochastic response
of the structure under dynamic loads
(vi) obtaining the relationship between the displaced shape of the struc-
ture and the loads and moments in the individual components of the
structure, by an appropriate structural analysis
(vii) combining the mathematical mode:s given by steps (ii) to (vii) above
to obtain the safety margin eauation and
 337

determining ~ for the failure criteria.

(10.22) has done the reliability analysis of jacket platforms in the
Sea. The Level 2 r~'iability methods have been applied for taking
s for the safety of offshore structures against fatigue (I 0.23).

REFERENCES
Stevensen, J. and F. Moses, "Reliability Analysis of Frame Structures", Journal
of Struct. Div., ASCE, Vol. 96, ST-11, Nov. 1970, pp. 2409-2427.
Neal, B.G., The Plastic Methods of Structural Analysis, Halsted Press, Third
Edition, 1977.
Cornell, C.A., "Bounds on the Reliability of Structural Systems'', Joumal of
Struct. Div. ASCE, Vol. 93, ST-1, Feb. 1967, pp. 171-200.
Dltlevsen, 0., "Narrow Reliability Bounds for Structural Systems", Joumal of
Structural Mech,, Vol. I, No.4, 1979, pp. 453-472.
s, Benjamin, J.R. and C.A. Cornell, Probability, Statistics and Decision for Civil
te Engineers, McGraw-Hill, New York, 1970.
a Ang, A.H.S. and M. Am in, "Reliability or Structures and Structural Systems".
Journal of Engg. Meclr. Div., ASCE, Vol. 94, EM4. April 1968, pp. 671-691.
Stevenson, J. and F. Moses, "Reliability Analysis of Frame Structures",
Journal of Struct. Div., ASCE, Vol. 96, STII, November 1970, pp. 2400-2427.
Ma, H.F. and A.H.S. Ang, "Reliability Analysis of Redundant Ductile
,.II Structural Systems", Report UILEENG-81.2013, University of Illinois, Aug . 1981.
Watwood, V.B., "Mechanism Generation for Limit Analysis of Frames",
e Joumal of Struct. Div., ASCE, Vol. 105, STI. Jan . 1979, pp. l-15 .
·C . 10 Murotsu, K ., "Reliability Analysis of Frame Structure through Automatic
IS Generation of Failure Modes", Reliability Theory and its Structural and Soil
Mechanics, Ed. by P. Thoft Christensen, NATO, ASI Series, !983, pp. 525·-540.
Thoft Christensen , P. and Y. Murotsu, Application of Stmctural System>
d Reliability Tlreory, Springer-Verlag, Berlin, 1986.
t, Moses, F., "System Reliability Developments in Structural Engineering" ,
e Strucwral Safety, No. I, 1982, pp. 3- 13.
I
Tang, L.K. and R .E. Mclchcrs, "Reliability of Large Structural Systems", Proc.
' of the ln.ffitutfon of Engineers, Australia, Civil fngg. Transactions, 1985. pp. 136-
143.
:I .14 Ranganathan, R. and A.G. Beshpan'dc, "Generation or ,Vominant Modes and
Reliability Analysis of Frames", Structural Saf~IY, No.4, 191!7, pp. 217- 228 .
e IO.IS Ticky, M. and M. Vorlicek , "Safety of Reinforced Co,ncrete Framed Structures",
Proc . lmemational Symp. 011 Fle.mml Mtsclumic.r of Rehl/orced Cot~cretc.• , Miami,
ASCE, Nov. 1964, pp. 53-84.
!tt
J}).l6 ,lVebstetr ".P&,Obabilisf.i,c~nalysis of a Simple Portal Structure", Joumaf of
§' ~merictl 4jll~e~Jnsl~t, JS7{gNo, 9, Sept. 1973, pp, 649-651.
::10.17 ~hou, CEg ~l!)l.oi}l »3.torotis, "Observations on Structural S stem
~ :1c:liabil ~'8!1'@ ' ~1~ ~di} ~rrelalions", Structural Safety, No. I, 1983,
~ t];p. 189· Bg '7 ~ ~ &' n ::r g_ ~
a_o.I8t'ltangao3Jh~l,'<R~a~~d~3J.a~lil'!ande, "Reliability Analysis of Reinf reed
S, ~oncret~ F.Ja'&e~, ~0Wrrrii1 !{if 'f$t#tct . Div., ASCE, Vol. 113, No.6, June 1987,
n !;BP· 131 ~:\is.§= ~ n ':? ff ~ '0 g
~0. 19i:§aker, ~.lfu_~ ~iiH~s$ t~· Ql:s'&n of Reinforced Concrete", Cc·menr a11d
:;· ~oncret~a,octa@~nd'b~l?joa
§-0.20 aker,ts,L::J:..~·n]tgM .~A@~I$·n e , "Inelastic Hyperstatic Frame Anal sis",
g roc . ln~~tigvtG.SJimp. /flf.lff.e':rit/lfj.Mechanics of Rei11/orced Concrete , iami,
· e. SCE, lOot 196\fopg.' 85-SI~ P. ~
'on ro-n
:;! s· p 5- ~
0
::1 ::;· ~ < . n n
3< i:J'5' n 0"
I:l ::J g
o ~ § n Y' 5 ' ::r < ::J
338

10.21 JS : 456- 1978, "Code of Practice for Plain and Reinforced Concrete", ltrdian
Standards Institutions, New Delhi , 1980.
10.22 Baker, M.J. and T.A . Wyatt, "Methods of Reliability Analysis for Jacket
Plat f rms", ecotrd International C01rference on Behaviour of Offshore Structures,
Lon don . August, 1979, pp. 499- 520.
10.23 Manners, W. and M.J. Baker, "Reliability Analysis in Fatigue", Second Inter-
national Symposium 011 the Integrity of Offshore Structures, Scotland, July 1981.

EXERCISE
10.1 Consider the structural system (5 member truss) shown in Fig. 10.31. It is given:
A,= 4.5 cm 2 A, = A, = 1.67 em•
A, = 1.2 em• A, = 4.5 em•
Variable f1 : !J. = 276 kN/m 1 a = 27.6 kN/m'
Q, Q. : 14 = 30 kN a = 6 kN
Qa: p. =50 kN a= 15 kN
Assume all variables are independent and normal.
(i) Compute simple bounds on the reliability of the system (pr1 ).
(Ans . 0.0602 ~ Pro ~ 0.0896)
(ii) Compute Ditlevsen 's narrow bounds on Prs·
(Ans . 0.0732 ~ Pr, ~ 0.0773)
10.2 For t he same problem given above, determine the narrow bounds on Pr. i~ Q,.
Q1 , and Q, follo w the Type I extremal (largest) distri bu tio n and fy follo ws the
I gn o rmal di~t ribu t io n .
(Ans. 0.0223 .;; Pro ~ 0.0236)
10.3 Consider the indeterminate tru ss shown in Fig. 10.32. II is given that for
Variable :
R1 • Ru. Ru p. = 77.6 kN a= 7.76 kN
R, !.L = 88.6 kN a= 8.86 kN
R, p. =50 kN a= 5 kN
R, p. = 67.6 kN a= 6.76kN
R. 14 = 78 .6 kN a= 7 86 kN
Ra p. = 40 kN a= 4kN
R1 , R1 , R,, R 10 p. = 75 kN a= 7.5 kN
R1 (i = 11 to 14) p. =50 kN a= 5kN
Q,, Q. p. = 50 kN a= 10kN
Q. 14 = 20 kN a= 6kN

Assuming all variables are normally distributed and statistically independent,

determine simple bounds on the Prs of the system.
(Am . 0 .00866 ~ Prs ~ 0.019)
10.4 o ns ider the RCC frame, shown in Fig . 10.25, and given in Example 10.12. All
da ta are the same as given in Table 10.10 except that for L 0 " Mean / Nominal
= 1.38 and B = 0.25. Generate dominant modes for the load combina tion
D + Lm + Wapi and determine
(i) The bounds on Pr1 assuming full redistribution.
(Ans. 0.726x Jo-• ~ Pr1 ~ 0.978x J0- 1 )
(ii) The bounds on Pr1 assuming limited ductility.
(A11s. 0.724 x 1o-a ~· Pr, ~ 0.818 x J0- 3 J
 331

n JO~S , The steel frame shown in Fia. 10.33 is taken from Reference 10.8, The data for
tho frame is aiven in Table -E 10.5. Generate dominant modes and determine
:t Ditlevsen's narrow bounds on the probability or failure of the system. R.eeultt
r, are available in Reference 10.14.

r-

02
1J 14 15 16
04 -
!1:
10 12
T
3-66m
01
7 8 II
,J
+
366m
· I.
l
t- l-Om--4- 3 0 m--1
6)
FIG. 10.& Two-storey one bay frame
3) TABLE E 10.5 Dutafor frame in Fig. E 10.5

It Section/ EA El a P
variable (kN) (kN m1)
6)
~tion
1, 2. 3, 4, 9, 0.105xJ07 0.84 X 10'
10, 11, 12
s, 6, 7, 8, 13, O.i68XI07 o.336x 101
14, JS, 16
Variable
Ma. -M 1 ,M,,Mc 110.0 kNm 0.15 1 0 l
M,,M11,Mu,
Mn I
~
Independent
M 1 , M,,Jivt, 275.0 0.15 1.0
Me,Mn,Mu.
Mu. M,, J
Q, O.lS
Q,
180.0 kN
90.0 0.25 l >-
Loads are in-
dependent
Q, 32.0 0,25 except
lt, Qc 16.0 0.25 J PQa. Qc "" 1

9)
~II
1al
Jn
 11
Advanced Reliability Methods

11.1 INTRODUCTION
In Chapter 7, Level 2 reliability method has been explained and illustrated in
detail. The method can be applied to linear or non-linear limit state functions of
correlated or uncorrelated normal or nonnormal variables. In this method, the
failure surface is linearized at the design point and reliability index is
calculated. The method is also called as First Order Reliability Method
(FORM). Here, probability offailure is taken as

(11.1)

given by Eq. 8.3 1. Only in the case of linear function of normal variables, the
value of probability of failure estimated by the above equation gives the exact
value. In other cases, it gives only approximate value called as notional value of
probability of failure. In general, Ute probability of failure estimated by Eq. 11.1
is sufficiently accurate and holds good for the majority of comp lex engineering
problems wiUt number of variables as long as the probabilty of failure is not too
small and the distributions of the variables do not deviate too far from the
normal distribution. This estimate of probability of failure is enough and quite
adequate for decision making problems in the field viz. fixing partial safety
factors, calibrating codes, development of inspection strategy and maintenance
schedule etc. The estimated Pr by Eq. 11.1 gives significant error when the
failure surface has large curvature and highly nonlinear and the function is in
terms of correlated nonnormal variables. In such cases, when one is interested

+
)
in estimating more accurate value of pr , he may have to use Second Order
Reliability Methods (SORM). Basic Monte Carlo technique explained in
Chapter 7 gives true value of Pr ; however, it takes more time and large number
of samples are to be generated to estimate Pr with a certain minimum
confidence level in the estimated Pr· Better sampling methods, which are called
here as advanced simulation methods, are available to estimate Pr without much
statistical error. In this chapter, the principle behind second order reliability
method is just introduced and advanced simulation methods are explained in
detail and illustrated with examples.

11.2 SECOND ORDER RELIABILITY METHOD

The first order reliability methods are easy and simple to apply but
approximation used to linearise the failure surface at design point does not
always hold good. When the failure surface is very non-linear, the estimated
reliability index shows an erroneous Pr value. The figure 11.1 brings out the
 M1

drawback of FORM. In Fig. 11.1 two failure surfaces arc shown. Surfice :8 :te
more non-linear than the surface A. It can be easily seen that the probabtUt.y

FIG.11.1 Drawback of FORM

of tailure of B is less than that of the surface A. But using the Hasofer-Lind
method, the values of reliability index p evaluated for both surfaces for
linearization at design point D are the same. This shows that not only the
distance of a design point D from the origin in the independent standardized
co-ordinate system but also the nature of the failure surface affects the failure
probability. Thus it becomes essential to take into account the nature of the
failure surface while evaluating the probability of failure in problems involving
non-linear surfaces. It is drawn to the attention of the readers that if the
original distributions of the variables significantly deviate from the normal

~.,.....,.nc
a"rlaee

FIG. 11.2 Parabolic approximation to failure domair.

J42

distribuiton, original smooth surface ( even if the original equation is linear)

can become distinctively curved in the normalized space. But it is particularly
difticult to find the exact nature of the surface every time. The second order
approach in the space of normalized variables will yield results close to the
exact value. In the second order reliability method the failure surface in the
standard normal space is approximated by a parabolic surface (Fig. 11.2) at the
design point, the axis of the parabola being the direction of z* (the design point
in the independent standard normal space) The corresponding probability
content is determined by asymptotic formula and by approximate formulae
(11.1 , 11.2). Tvedt (11 .3) has presented a method calculating from the full
second order Taylor series expansion of the failure function at the design point
z*.

11.3 IMPORTANCE SAMPLING METHOD

In Monte Carlo simulation, as probability of failure for any stmcture is
generally very low, large number of samples will have to be generated to get
sufficient number of points in the failure domain. This will require evaluation
of stmctural response for large number of times which affects the efficiency of
the method. This drawback is overcome by replacing the joint density function
fx(x) by new sampling function hx(x) which ensures the sampling in the region
which contribute most to the probability of failure. The probability of failure is
given by,

II { g(x)SO} fx(x;)
i=l lrx(xt)
Pr = (11.2)

where
N, is the number of simulations and
1 { } is an indicator function given by

{
I { } = 1 for g(x)::;; 0
0 otherwise

The main purpose of the new density function i.e. weighting function is to
centre the simulation in the most important region i.e. around design point. It is
possible theoretically that variance of the results from Eq. 11.2 can be reduced
to as low as zero, if the values of the weighting function are equal to values of
the actual probability density in the failure domain. This assumes that the
information of the design point is exact along with the right choice of the
weighting function. Because of the finite number of weighted simulation, there
will be always some statistical uncertainty, apart from due to choice of
weighting function. Applying statistical analysis, this uncertainty can be
estimated. The variance (s2) of the calculated probability of failure can be
estimated as,

j
 343

(liJ)

The so called standard error (or statistical error) in the estimate of Pr is given
by,
.[;2
t;>r = - - (11.4)
P[

From the above equation it is clear that statistical error is not only dependent
upon the number of simulations bot also type of weighting function hx(x).

Choice of weighting function

It is obvious from Eq. 11.4 that, whole suecess of the importance sampli ng
approach depends upon the choice of the weighting function hx(x). Several
suggestions have been made for the choice of hx(x) in the importance sampling.
Harbitz (11.4) suggested the weighting function as the same original joint
density function but onJy shifted at the design point, which is calculated by
Level 2 method. But tlte question comes, as once the design point by Level 2 is
known, why to go for further artalysis, unless some improvement in accuracy is
needed. Also original distribution may be complex when variables arc
correlated, which causes difficulty in sampling process.
Another choice for weighting function hx(x) is to use independent standard
mnltinonnal density function centred at the design point and standard
deviation equal to or greater than the original standard deviation (11.5, 11.6,
11 .7). Design point can be calculated based on the assumption of uncorrelatcd
Gaussian variables or uncorrelated with original distributions. Generally hx(x)
is taken as independent n-dimensional multinormal density function, centred at
the design point calculated on the assumption of oncorrelated Gaussian
variables. Tlte standard deviation is taken as, one to three times of tbe original
standard deviation. As this choice for hx(x) will produce the sample points
unbiased with respect to all variables, it will cover the wide region around the
design point. Due to this advantag exact form of limH state g(x) is not
necessary while evaluating the probability of failure . Due to ~implicity of the
hx-(x) generation of the sample points can be done very efficiently. As hx(x) is
the independent multinorrnal density function, unless the failure surface is
highly nonlinear, there will be 50 % probability that sample point falls in the
failure domain . .Random deviates for the nonnal distribution are generated
using Box and Muller technique explained in Chapter 7.

Correlated nonno.,.m variables

In Level 2 method explained in Chapter 8, the treatment of correlated
nonnormal variables has been explained when the covariance matrix. [Cx] is
known. If the conclation matrix. [Px) is given. the procedure is slightly
344

modified and is explained below. It can be easily proved that the correlation
matrix of original variates becomes covariance matrix of reduced variates. Here
reduced variate Z; means,

X.-!J·
Z=-'-'
' U;
(11.5)

where 1-4 and cri are mean value and standard deviation of Xi. In the case of
correlated non-normal variables, the original probability density fx(x) is found
at the sample point in consideration by transforming them into equivalent
independent Gaussian components. This is done by first transforming them
(nonnormal variables) into equivalent normal at the sample point by using the
procedure explained in Chapter 8. The Gaussian components obtained are then
transformed into independent components by orthogonal transfonnation. For
the correlation matrix lPx] the eigen values are evaluated from which eigen
vectors are found out for each eigen value. Then the transformation matrix [T]
will be the matrix with each column as eigenvector for respective eigenvalue.
The independent standard normal variates Y 1, Y 2, ...... Y n will be given by,

Y = [T]t Z (11.6)
E[Y] = [T]t E[Z] (11. 7)
[Cv] = [T]t [px] [T] (11.8)

That is eigen values of [px] are the variances of the respective variates Yi.
Though this transformation is approximate, it can be applied very efficiently
and gives results within good approximation.
Following steps are involved in the computation of Pr using ISM when
statistics of all variables, the correlation matrix and the limit state function are
given:

1. An eigen value analysis of the correlation matrix is carried out to find the
transformation matrix [T] . Each column of the transformation matrix is an
eigen vetor corresponding to the respective eigenvalue (Refer Chapter 7).
2. Find the design point x* using Level 2 method. For simplicity, assuming all
variables as uncorrelated normal variables, x* can be found out and this
may be used as a sampling point.
3. Two uniform random numbers v1 and v2 are generated between 0 and 1 for
each variable.
4. A standard normal variate u tor each variable is obtained as
u = [2 ln(ll v 1)] 112 cos(27tV2)
5. Select a value for standard deviation multiplier, Sdm, from 1 to 3. A sample
point xis obtained as
X = X* + Sdm U 0'
6. The value of limit state function g(x) is evaluated.
7. If g(x) < 0 proceed; otherwise go to Step 3.
8. The equivalent m.ean. and standard deviation at poiDI lound ouc ••
explained in Chapter 8. They are given by Eqs. 8.67 IDd 1.69.

P'xI =-<i'xI «»- 1 J:l'

r~x . I (x,)]+x, ( 11. 9)

(11.10)

9. The equivalent normal variables Z are found at the point X as

Xt-PX,
ZI= (11.11)
ax,
10. The independent variables Y are found at point X as

Y=[Ttz (11.12)

11. The probability density and the sampling density at X are found out as

lx = I n I
n- exp~ - ~
n( I y;2) (11.13)
{Jtif I= I O'J'j i=l 2 c:r,

(11.14)

Here hx is independent multinormal density function at X

12. Calculate fx I hx. Go to Step 3.

The whole process is repeated from Step 3 to Step 12 for number of
required simulations N•.

13. Compute prusing Eq. 11.2 and Cpr using Eq. 11.4.
The procedure explained above is shown in the flow chart given in Fig. 11.3.
The importance sampling method is illustrated with the following examples.

EXAMPLE 11.1 The limit state function is given by

Here the number of variables is 3. The statistics of the variables are given in
Table 11.1. The correlation matrix is follows:
 346

1.0 0.5
0.0]
[p,J 0.5 1.0 0.0
[ 0.0 0.0 1.0

A•cul ~o . of varlobl••. their diatrlbwtlon

and correlation mat.rix R•ad al•o the
limit stat• function and no. of almulatlOfta
to b• don•

F'lnd th• d••lgn point by caesumpt.l o

of uncorr•lated gaua-">n Yarlabl••

Oo

G•n•rat• th• aompl• poi"t uel

ind•pend•nt multlnormal d•nalty
c•nt•red at d-lgn point

FIG. 11.3 Flow Chart for importance sampling

•
I
TABLE 11.1 Statistics of variables- Example 11.1
Variable Mean Standard deviation Distribution
5. 0 Type I extremal
(largest)
x2 5o.o 2.5 Normal
x3 15oo.o 100.0 Lognormal

Compute Pr by using ISM taking SDM equal to 1.

Detailed stepwise calculation for the computation ofpr is given below.

Step 1: For the given correlation matrix [px] the eigen value analysis is
carried out.
 347

The eigenvalues of the oo~tion matrix are A.1 = 1.5 ; ~ • 0.5 ; A., • 1.0. The
transformation matrix is obtai.oed as

0.707 - 0. 707 0.000]

[T] = 0.707 0.707 0.000
[
0.000 0.000 1.000

Step 2: Using Level2 method, the design point x* and corresponding p are
obtained.

x* = 13:;~::} p = 2.0868
1623.47
Step 3: The two uniform random numbers VI and v 2 generated for first
variable, are
VI =0.8704 ; V2 =0.3995

Step 4: The normal variate is

. + :J]y,~(2..,)
=- 0.4254
m(
Step 5: The sample point XI is given by

xi= xi• + S.n ui ox1

= 33.029 + (1.0)(-0.4254) 5
= 30.901
Similarly random numbers are generated for random variables X 2 and X 3 and
the values for the other two variables are evaluated as
X2 = 49.6598; X3 = 1582.928

Step 6: The value of the limit state function is

g(X) =XI X2 - X3
= (30.9018) (49.6598)- 1582.928
=- 48.350

Step 7: Check g(x). Since g(x) is negative, it is proceeded further.

Step 8: The equivalent mean and standard deviation of variables at the sample
point are calculated. The parameters a and u of XI following Type 1 extremal
(largest) distribution, are determined as follows using Eqs. 3.115 and 3.116.
s
348

- I " = 0.2565
- J65 .0
0.5772
u = J.lx - - -
1 a
0 5772
= 40.0- · = 37.75
0.2565

The probability density function of random variable X~, following Type I

extremal (largest) distribution is given by Eq. 3.113 .

/x1 = aexp(-a (x1-u)-exp{-a {x1-u)}]

The value of fx 1 at x = x1 is
fx 1 = 0.2565 exp [-0.2565(30.9018-37.75)-exp{-0.2565(30.9018-37.75)}]
= 0.004532
The cumulative distribution function ofX1 is given by Eq. 3.114.

Fx 1 = exp[-exp{-a (xt -u)}]

The value of Fx 1 at x = x1 is
Fx = exp [-exp{-0.2565 (30.9018-37.75)}]
1

= 0.00305
Using Eqs. 11.9 and 11.10, mean and standard deviation of equivalent normal
at x1 are obtained. They are calculated as

= 2.0012
1
J.l'x1 =-ax1 <Il- [Fx1 0]+xJ
= ( -2.0012) <1>"1 [0.00305 J+ 30.9018 = 36.3942

Similar procedure is followed to evaluate the equivalent mean and standard

deviation of variables X2 and X 3 . They are given as

J.l'x2 = 50.0 a'x2 = 2.5

J.l'x3 = 1494.239 a'x3 = 105.412
Step 9: The equivalent nonnal variables at the sample point in the nonnalised
co-ordinate system are obtained using Eq. 11.11
 349

z1 = x, - JJx, =
30.9018-36.3942
= -2.7445
cr'x, 2.0012

z2 = X2- J.IX2 = 49.6598- 50.0 -0.1361

uxl 2.5

z3 = xJ~ J.Ix~ = 1582.928-1494.239. 0.8414

uxl 105.412

Step 10: Using the transfonnation matrix [T] the variables are converted into
independent variables using Eq. 11.12.

0.707 0.707 0.000]!- 2.7445]!-2.0369]

y = - 0.707 0.707 0.000 - 0.1369 = 1.8444
[
0.000 0.000 1.000 0.8414 0.8414

Variance ofYi is given by eigenvalue A.i . Hence

OlJ = Jl.5 = 1.225 O'y2 = .JD.S = 0.707
O'fl = .Jt:O = 1.0
Step 11: Using Eqs. 11.13 and 11.14 the probability density fx and the
sampling density hx are computed as follows:
1 1
h-
- (&1 (2.001xl .225) (2. 5.t0.707) (t05.412xl) X

cxp{-I[( -~~;:·r ·c~s:)' ·(o~~·r]

= 8.14 X 10"7
1 1
hx = (&r ~.o)(2.s) (wo.o) x
2 2 2
e.J_1_[(30.901 - 33.029) +(49.6958 - 47.64) +(1582.928-1623.47) ]}
""1 2 5.0 2.5 100.0
= 3.084 X 10"5

The ratio of fx and hx is calculated and stored. The process is repeated for the
specified number of simulations. The value of Pc is computed using Eq. 11.2
and the statistical error using Eq. 11.4. The results obtained for different values
of specified number of simulations are given below.

Sl. No. N. Pr epc(%)

1. 500 0.0291 10.567

2. 1000 0.032 7.178
3. 1.500 0.0328 .5.950
 350

The exact value ofpr, by Monte Carlo method is 0.032. It can be observed that
as number of simulation increases, the accuracy of Pr also increases aud
percentage error decreases. It should be noted that the set of random numbers
obtained for different starting points will be different Hence for the same
number of simulations, the value of Pr obtained will not be exactly same.

f ~XAM PLE 11.4 The limit state function is given as

g(X) = __!_
8
~?+Xi+x:f)-x 4 + 4.o
Variables X; arc normally distributed with mean and standard deviation of each
variable are 0 and I respectively. That is, they arc standard nonnal variables
The variables are uncorrelated. Determine the probability of failure by using
ISM.
The starting point is the design point obtained by Level 2 method. That is

x* = ~~~!~~
0.250
3.997
The procedure of computation of Pr is same as explained in the previous
example. It is to be noted that since the given variables are uncorrelatcd
standard normal variables,

y=z=x

where x is the sampling point obtained by generating random numbers and

using standard deviation multiplier. All the intennediatc steps in the
computation for SDM = 1 for tlte first simulation are given in Table 11 .2. The
'I whole process is repeated for number of simulations and the values of Pr and Cpr
are computed using Eqs. 11.2 and 1 J .4 respective! .

TABLE 11.2 Results of the analysis in the first simulation using ISM-
Example 11.2
t~
Initial point Random X g(x) y=z=x Fx hx
numbers

X =
['"]
0.25
0.25
3.99?
r5134}
0.3206

{0.6311}
0.2595 r""J0.1930
-0.80 12
4.6012
-0.6996
r~"J
0.1930
-0.8012
4.6012
0.431 X
10-<i
0.1056 X
to-•

r52?8}
0.4401

{0.?832}
0.9289
 351

The problem is solved for various values of number of simulations and SDM.
The results are given in Table 11.3. It can be observed from the table, that the
statistical error decreases for SDM equal to 2 at which the results are very
consistent in successive runs. The corresponding value of Pr agrees with the
exact value of 0.423 x 10"3

TABLE 11.3 Results obtained by ISM- Example 11.4

Number of simulations
1000 1500 2000
SDM Pr c;,r Pr Cpr Pr epr
~% 2 (%} {% }
I.O · 0.326 X 10"3 23.77 0.315 X 10"3 17.54 0.303 X 10-3 14.76
1.5 0.412 X 10·3 10.90 0.422 X 10"3 8.74 0.435 X 10"3 7.68
2.0 0.423 x 10·3 9.59 0.417 X 10"3 7.87 0.409 x 10·3 6.97

11.4 ADAPTIVE SAMPLING METHOD

The main limitation to the importance sampling method is the difficulty in
selecting a good sampling density. To choose such a density one needs to know
which part of the failure domain has a relatively high probability density. This
knowledge is not usually available priori and hence it is difficult to choose a
good sampling density. Adaptive sampling method (ASM) can be used to
overcome this difficulty. Th.is technique utilises the fact that even with a poor
initial choice of importance sampling density, the knowledge about the failure
domain increases with the sampling process. Hence after each sample the
importance sari'lpting density can be modified for this increased knowledge and
finally a good sampling density can be obtained.
It is already said that the much prior knowledge about the important region or
the region where probability density is relatively high is not available. This may
result into the poor initial choice for the sampling density. If such a poor
density is used, the sample points generated may lie in the region where
probability density is relatively low. Thus the sample points are clustered
around an unimportant region. However, while sampling with such a poor
density some sampling points may have relatively more probability density than
that of the chosen point. Thus while sampling, the knowledge about the
important .region increases i.e. the region of relatively more probabiHty density
is known. Adaptive sampling technique makes use of this knowledge to move
towards the more useful density. For this, the sample point having more
probability density is chosen as the new centre of the sampling density. Thus
the sampling density is moved towards the more important region. Figures 11.4
and 11.5 show the poor and improved choice of sampling density respectively.
The steps involved in the procedure of computation of Pc using ASM are almost
same as given for ISM except the following changes.
In Step 2, any point x., can be chosen as the starting point for mean of
sampling density. At this point, the original variables are converted into
Step 13: Compute Prusing Eq. 11.2 and~cbyusing Eq. 11.4,' ~
The whole procedure is shown in the flowchart given fn Pi , ll.
procedure of computation of Pr using ASM is illustrated ·witfl·tho fOUOWilrt.l
examples. ·

Rtod No. of variablu, their

distributions and correlation
limit state function, No. of
simulations to bt done

Find design point using FORM

and take it as stllrting point lie

Do for I = No of simulatlonr.INa)

Generatt two random numbtrs

for each vartablt and obtain a
normal variatt for edch variable

•tntra1t a sample point X ~King

lndeptndtnt multlnormal density
ctntrtd at point lie

FIG. 11.6 Flow chart tor ASM

354

EXAMPLE 11.3 The same problem given in Example 11.1 is considered here.
g(X) = x1 x2 -x3
Statistics of the variables are same as given in Example 11.1. Detennine Pr
usingASM.

Step 1:
For the given correlation matrix (Px], the transfonnation matrix is obtained. it
is same as obtained in ISM.

0.707 -0.707
0.000]
[T] = 0.7u7 0.707 0.0000
[
0.000 0.000 1.000

Step 2:
Any starting point can be selected. However, here the design point obtained by
FORM is taken as the starting point.

!
33.029)
x., =x• = 47.64
1623.47

At this point, original correlated nonnorrnal variables are converted into

independent nonnal variables. Mean and standard deviation of equivalent
nonnal at ;;,* are

p'x = 37.665
1
P'x2 = so.o P'x 3 =1491.45
a'x1 = 2.554 a'x2 = 2.50 u'x3 =108.112
The equivalent nonnal variables at the sample point in the nonnalized
coordinate system are obtained as

3.029 - 37.665 = -1.815

2.554

X2 - ;.lx1 = 47.64-50.0 = _ _
0 944
a'x1 2.5

XJ - PX, 1623.47 - 1491.45 1.2 !

2
ax 3 108.112

Using the transfonnation matrix, the variables are converted into independent
variables using Eq. 11.12.
 355

0.707 0.707
y ::: - 0.707 0.707
00.000
.000] {-1.815)
- 0.944
[
0.000 0.000 1.000 1.221

-1.951)
=
{0.616
1.221

Standard deviation of Y are square roots of eigen values of (Px]. They are
(7'11 = 1.225 or2 = o.707
The probability density is comJ?uted at point x* as independent multinonnal
density and is taken as

1 I
/max = (&f (2.554x1.225) (2.5.x0.707) (108.112xl.O) x
2 2 2

A1
ej_![(-1.951)
2 1.225
+(0.0.707
616) +(1.221)
1.0 ]}

= 0.9698 X 10"5

Step 3 to Step 11 are same as in the previous case solving by ISM in Example
11.1. (They are not repeated here).
At the end oftStep 11,
fx = 8.146 xl0"7 ; hx"" 3.084 X 10"5
The ratio of fx and hx is calculated and stored. Since fx is less than fmax , the
starting point is not shifted and one simulation is over. The procedure is
repeated from Step 3 for specified number of simulations. The probability of
failure and percentage error are calculated using Eqs. 11.2 and 11.4. At the end
of 500 simulations using standard deviation multiplier equal to one, Ute
probability of failure is found to be O.IJ39 with t:pr =- J 1.4%.
The problem is solved using different standard deviation multipliers. The
simulation is carried out for 500, 1000 and 2000 number of simulations with
different "SEED" i.e. different starting points for generating random numbers.
(Note: In all available programmes for generating random numbers, starting
point, called SEED, is to be given). The results ob.tained by ASM are given in
Table 11.4. The exact probability of
356

TABLE 11.4 Results obtained by ASM- Example 11.3

s s
D E Nwnber of simulations
M E
D
500 1000 2000
Pr s. Pr s. Pr s.
( %) ( %) ( %)
0.0390 11.40 0.0355 7.59 0.0340 5.23
1.0 2 0.0335 12.58 0.0330 7.76 0.0328 5.01
3 0.0300 12.69 0.0320 7.88 0.0350 4.99
1.0373 14.19 0.0345 9.30 0.0326 6.38
1.25 2 0.0310 15.42 O.DJ16 9.72 0.0320 6.39
3 0.0280 14.43 0.0312 9.15 0.0340 6.06
Note: The exact probability of failure is 0.032 as per Monte Carlo Method

failure is found to be 0.032. From Table 11.4, it is obse1ved that the statistical
error is found to be decreasing with the increasing number of simulations. For
2000 simulations the statistical error is very low and the results are very close
to the exact value.

EXAMPLE 11.4 Consider the same limit state function given below (11.4).

g(X)
= X X X _ XsX} X1
2 3 4 v \'
X
I
,, ll ' 7
All the variables are normally distributed and uncorrclatcd. The statistics of the
variables are given in Table 11.5. Compute Pr using ASM.

TABLE 11.5 Statistics ofvnriab/es- Example 11.4

Variable Mean Standard deviation Distribution
XI 0.01 0.003 Normal
x2 0.30 0.015 Normal
XJ 360.0 36.0 Normal
x4 2.26 x w- 4 1.13 x w- 5 Normal
Xs 0.50 0.05 Normal
x6 0.12 0.006 Normal
x7 40.0 6.0 Normal

The results obtained by using ASM for different starting points, different seeds,
and for diiTerent number of simulations are given in Tables 11.6 and 11.7.
From the tables, it can be seen that certain minimum number of simulations are
required to get probability of failure close to the exact value. In general, as
number of simulations increases, statistical error in estimated Pr decreases. It
should be noted also, that the value of standard deviation multiplier plays a
role. For this problem, it appears that percentage error is very less for SDM =
 357

1.25. It can also be noted that the selection of st.a.rting point affects the result in
this case. The choice of seed is not affecting the result significantly when
large

number of simulations is used. The probability of failure is found to be very

close to the exact value 0.34 x 10'3 .

TABLE 11.6 Results by ASM with starting point I -Example ll. 4

s s Starting point 2: x.= (0.015, 0.25, 300.0, 2.26 X 104 , 0.5, 0.12, 40.0i
D E Nwnber of simulations
M E
D
500 1000 1500 2000
Pr e,t Pr e,r Pr e,c Pr E,c
X 10"' (%} x 10"' (%} x 10"' (%) xiO_. (%)
1 3.92 41.84 4.27 25.75 3.74 21.06 3.60 17.00
1.25 2 8.00 30.38 .5.21 23.81 4.58 18.35 4.10 1.5.57
3 4.04 30.74 3.28 19.88 3.38 13.8.5 3.30 11.43
1 2.35 38.16 4.13 27.18 3.91 20.21 3.37 16..59
l.S 2 3.33 S7.08 2.50 23 ..53 3.35 20.65 3.31 16.72
3 2.77 42.72 3.80 18.60 3.85 15.16 3.60 13.56
1.9.S 78.76 1.44 34.08 2.66 31.63 2.70 26.83
2.0 2 2.20 40.46 3.12 3.5.00 3.21 26.48 2.65 24.20
3 \.30 39.73 2 ..56 28.73 2.57 25 ..53 2.60 23.45

General points
• The value of SDM is generally found to vary between 1 and 2.
• While simulating using ASM or ISM, a suitable value of SDM is chosen by
performing a few number of simulations wit11 different SDM values
between 1 and 2 and the best one is to be selected and it is the one with the
least statistical error. The same value is to be used for calculating Pr by
conducting enough number of simulations.
• Generally, while solving problems witJ1 ASM and ISM, it is suggested that
the termination criterion to stop the simulation, may be where statistical
error in computing Pr is less than 20 percent.
• The starting point affects the convergence of Pr value. For a good starting
point, the value of Pr converges to the exact value with less number of
simulations and less statistical error, as compared to the poo.r starting
point. The design point obtained by Level 2 method is a good starting
point.
• Generally ASM requires less number of simulations to evaluate Pr value by
maintaining the same statistical error as tllat of the ISM.
• For any arbitrary starting point, ASM is preferable as it is requires less
number of simulations.
358

TABLEll.7 Results by ASM with starting point 2- Example 11.4

s s Sla.rting point 2: x.:= (0.015, 0.25, 300.0, 2.26 x 10-4, 0.5,
D E 0 .12, 40.0)t
M E Number of simulations
D
500 1000 1500 2000
Pr t;,r Pr Cpr Pr Cpr Pr epl
x w- 4
{%2 X 10' 4 . {%) x Jo- 4 ( %} X10-4 (%)
1 3.86 20.53 3.91 14.28 3.82 11.06 3.75 9.11
1.25 2 3.09 15.74 3.20 13.00 2.91 10.51 3.03 10.09
3 2.27 18.46 2.54 12.27 2.80 10.53 3.02 10.S4
I 2.42 47.61 1.23 24.13 3.55 17.44 3.47 16.95
1.5 2 2.09 32.31 2.87 24.56 2.67 18.75 2.55 15.38
3 1.36 26.84 2.90 24.22 3.45 17.77 3.29 14.84
1 l.ll 58.96 1.85 26.23 l.SO 21.77 2.25 25 .46
2.0 2 8.80 45.99 7.27 31.93 6.00 28.31 5.63 24.45
3 4.40 67.06 3.75 43.08 4.18 35.20 3.96 29.49

Jt.S RESPONSE SURFA ~METHOD

Advan cd Monte arlo imulation methods arc exact and compul.ationall

clTicicnl fr m probabilistic point of VIC\ • While ~ nluating the struc!ural
rcliabili , the maximwn time is spent for eva luau ng ·(ructuraJ response on! .
Since simulation methods arc numerical c ·p rimcnls carried out randomly,
they require the full analysis of th structural i>Tiem f, reach generated set of
load, r sistancc and systen random ariablcs. This rna result in large
computational efforts to an unacceptabl level. Hence it is desirable to simplify
the whole mechanical proce s by a ne' 111echanical model for evaluatmg the
structure/system response. While de clo h g the new model, it is important that
it will allow an easy and efficient computation of failure function response
under loading/ system condition but still preset es the es ·entia! features of tl1e
structure/ system. This new mcchani al model representing U1e original limit
stale function is called res onse surface.
The representation of the li1rut state 'function by response surface should be
independent of properties of the basic variables involved. However for
improving effictency and accuracy of Ute method including subsequ nl
reliability analysis, ome prior knowledge of the stochastic properties of the
variables i to be used. In most of U1e cases, mean alue and standard deviation
of variables are known. Use of such information wtll produce response surface
suitable fo~ wlde range of stochastic properties of basic variables.
The aim of Ute response surface is to replace Ute original fflilure function
g(X) by an equivalent function R(X) oy which computational procedure can be
simplified maintaining the accuracy. The limit state sorface can be repres!ilitoo
in polynomial form (11.8).
 359

n n
R(X) = a+ L b,X; + Lc;X;2 (11.15)
i=l i=l

where X1 , i == 1, 2, .... .. , n are basic variables and parameters a, b; , c; , i = I, 2,

.... , n are constants to be detennined. Here Eq. 11.15 docs not contain mixed
tenns XXi , h¢nce tlte function R(X) basically represents the original flmction
g(X) along the coordinate axes X; only. As number of free parru,neters in Eq.
11.15 are less i.e. 2n+ 1, only few numerical experiments are required to obtain
unique R(X). However this implies that, in general. sample between the axes
wiU not be covered sufficiently. Tllis is improved by using information on mean
and standard deviation of basic variables while updating the R(X).
Bucher and Bourgand (11.8) suggested the way of obtaining R(X) by
interpolation using points along X;. The starting central point chosen is the
mean vector. (see Fig. 11 .7). Around tltis starting point 2n points are generated
as X; =J.I; +t0'1 , i : 1 2, ..., n, in which tis tl1e arbitrary factor varying from I

xl .

~ -- r.lx(llhO
it
l ''
"'
'X,
P.

FIG. 11.7 Starting approximation for response surface

to 3. Using function values of original surface at 2n+ I points, the parameters a,

bit Ci are obtained by solving the set of 2n+ 1 linear simultaneous equations.
Thus first approximation to RQ{) is obtained.
In the next stage, function R(X) is used along with the infonnation on mean
and standard deviations of basic variables to obtain the estimate of the desigu
point. The estimate is based on tile assumption of uncorrelated Gaussian
variables. This design point obtained b Level 2 me od Js used or
inte1p0lation Le new centre J?-Qint on origi;ru failure .surface which is in
-- of interest Le area from which maximum contribUtion to_the
pr~>ih of failure is made (see Fig. 11.8). So tile 'nQW centre point for
i tion of R(X) can be obtained as,
 360

X N = II + (x D -- Jl )----'1-'-'('--
!. p "'-)- (ll.lG)
g(p ) -1( Xo )

\
I
x, .

Fig.11.8 Upating the interpolation point for response surface

where XN is the new centre point and X 0 is the design point obtained for first
approximation to R(X). This interpolation guarantees that the new centre point
is sufficiently close to the exact limit state g(X) = 0. The response surface is
updated by evaluating the coetlicients a, b" ci, i = I, 2, ...., n at the new centre
point XN. So the total number of evaluation of the original limit state equation

~,,!
required is 4n+3.
The update of the polynomial ensures that the critical domain is sufficiently
i~ covered by the numerical experiments from the full mechanical model. Once
I the R(X) is found, the reliability analysis can be proceeded in any suitable way,
preferably using advanced Monte Carlo technique - Importance sampling
method or adaptive sampling method. R(X) need not produce the exact limit
state surface in entire space but, only sign of original limit state near the design
point (i.e. in the region which contributes most to the failure probability) is
important (ll.8) . A simple computer program can be easily written combining
response surface method with ISM or ASM. A flow chart for RSM is given in
Fig. 11.9.
In some problems, the response surface obtained by using Eq. 11 .15 may not
give sufficiently accurate mechanical model. To improve the accuracy, mixed
terms may be added to Eq. 11.15 as given below:
n n
g(X) = a+ l,b;X; + l,c;X;2 + l. l,dijX;X 1 (11.17)
i=I i=I i'liJ

Various numerical and structural engineering problems solved using the

response surface method with ISM are given below. Probability of failure is
calculated using Level 2 and Importance sampling technique for both, original
failure surface and response surface and results are compared.
 First approximation is the
mean vector IJ

C)•nerate the 2n pOints oround

t.his, atild evaluate ctPef!fi ci e:.nts
ot the:t·e:sponse surface. ·- ..

Find tge desig, p~nt Xo for

the res,lons• surface assuming
uncorre1ated gawssion wsdobles.

Evaluate 9x (X) at Xo

Generate 2n points around XN

Evaluate 2n + 1 ·-t6efficrcnts at
new cenhPr point and ,update
response surface

FIQ. 11.9 Flow chart for response surface method

362

EXAMPLE 1I.5 Reliability analysis of the three bay five storey RC.C. frame
shown in Fig. 11.10 is carried out in this example. The structure data and tho

JJ JS A

11 ,. ,
~~
30 31 32

13
" ~~ . lS 16 320.() trn

27 ll 1t

g 10 1l
,~ '2 30&·6 tm

l4 2S M

5 6 7
•
'21 22. 23

1 2 J 4 201.9 Cln

m7r TrJrr '"'~ .,~

~ 323·1 em + + 240.QC/!\
I
32J.8t.m ~

FIG. 11.10 Three bay five storeyed portal frame Example 11.5
I
I
l_
statistical data of the random variables involved are Jivcn in T11bl011 11.1
11.9 ~ly. The limit state criterion is the displacement It( top thtor
frame (i.e. at point A) should not exceed h/350, where his the total hoiaht of
the structure. So limit state function can be written as,

g(X) = 4.10- ~(A) (11.18)

where ~(A) is the function of the loads acting on the structure and material
and geometrical properties of the structure. Here all these param~ters i.e. loads,
material properties and geometrical properties of the structw:e are random
variables. ln addition to the horizontal and vertical random loads, each beam is
carrying a .constant dead ioad of24.50 kg/em.

TABLE 11.8 Structure data- Example 11.5

Element No. Moment of inertia Cross section
I, 4, 5, 8, 24, 25, 26, 28, 31, 34
2, 3, 6, 7, 10, 11, 14, 15, 18, 19
9, 12, 13, 16, 17, 20, 21, 22, 23
27,29,30 32 33,35

TABLE 11.9 Statistics of the random variables- Example 11.5

Variable Mean Standard deviation Distribution
PI (kg) 3000.0 1200.0 EX.. L •
P2 (kg) 4284.0 1235.20 EXr,L
E (kg/cm2 ) 0.225E+06 0.225E+05 Lognonnal
At {cm2 ) 1045.15 52.25 Nonnal
A2 (cm2 ) 2264.51 113.22 Nonnal
A3 (cm2 ) 870.96 43.55 Nonnal
~ (cm2 ) 1393.54 69.67 Nonnal
lt(Cm4 ) 0.182E+06 0.182E+05 Nonnal
h(.cm4 ) 0.986E+05 0.986E+04 Nonnal
h(cm4 ) 0.105E+06 0.105E+05 Nonnal
14(cm4 ) 0.43IE+06 0.431E+05 NonnaJ
Correlation .coefficients are PA. A = Pt. A = Pt. r = 0.30
• EX,. L denotes Type 1 extremal (largest)

The response surface is generated and probability of failure is found out for the
generated surface using Level 2 and importance sampling methods. The results
are compared with that of the results obtained by Level 2 analysis using
original failure surface. All the results are presented in Tab1e 11.10. It can be
seen from the table that the relip.hility analysis using response surface is
showing considerable computational advantage over that of the use of original
failure surface. Also, results with response surface are very close to that using
original surface:
364

TABLE 11.10 Results of the reliability analysis- Example 11.5

Using Rx (X) Using g... (X)
1. Level 2 method
Beta 3.255 3.274
Pr 0.567E-03 0.528E-03
Computer time (CP sec) 24.90 (For RS) 239.88
0.65 (for Pr)
2. Importance sampling
Number of simulations 5000
SDM 1.0
Pr 0.520E-03
epr ( %) 11.28
Computer time (CP sec) 25.16 (for RS)
14.18 (for pr)

EXAMPLE 11.6 Reliability analysis for the 25 bar transmission tower shown
in Fig. 11.11 is carried out. The tower is considered as a space truss. The
structure data and the statistical data for the random variables involved are
given in Table 11 .11. Failure criterion is the displacement at top (i.e. at point P
in Fig. 11.11) should not exceed h/250, where h is the total height of the
structure. For this, failure function is given by,
g(X) = 0.02- ~(P)
Here 8h(P) is the function of loads acting on the truss and geometrical and
material properties of the structure which are random variables. Reliability
analysis is carried out for the response surface, using Level 2 and importance
sampling method. Results can been compared with Level 2 analysis using
original failure surface. All the results are given in Table 11.12.

TABLE 11.11 Statistics of the random variables- Example 11.6

Element Variable Mean Standard Distribution
No. deviation
1 A1 (m2 ) 6.45E-05 9.675E-06 Normal
2 to 3 A2 (m2 ) 2.43E-04 3.635E-05 Normal
6 to 9 A3 (m2 ) 3.04E-04 4.560E-05 Nonnal
10 to 13 ~(m2) 6.45E-05 1.290E-05 Normal
14 to 21 As (m2 ) 1.79E-04 2.680E-05 Normal
22 to 25 ~(m2) 2.45E-04 3.678E-05 Normal
E (K.N/m2 ) 2.04E+08 1.860E+07 Normal
P, (K.N) 10.0 3.5 EXLL
p 2 (K.N) 15.0 4.0 EXl.L
Correlation coefficients are p Ai, Aj = 0.25 p Pl ,P2 = 0.5
2·54m

~OS m ---.
1-905m

24 V .:
---- ~

----- ~
w
Q
FIG:. 11.11 TwentyfiV8.1Jar.:hnsmission tower- Example 11.6 (,"'
Here also. reliability analysis using response surface is showing consider ahlt-
computational advantage over the use of original surface with reasonahk
accuracy. In Level 2 with response surface analysis, it can be seen I h:rl
maximum time is taken for the evaluation of response surface. while Level J
analysis is taking negligible time.
Response surface method is not to be used for the cases where explicit linrrl
state functions are directly available. It is advocated only in those cases where
repetition of structural analysis is to be carried out number of times to generate
the limit state function at every time .

TABLE 11.12 Results of the reliability analysis- lc.xample 11.6

Using Rx (X) Using g. (X)
Pr 0.934E-05 0.621E-05
Computer time (CP sec) 10. 16 (For RS) 84.21
0.42 (for Pr)
2. Importance Sampling
Number of simulations 5000
SD Multiplier 1.3
Pr 0.810E-05
o/o Se 12.20
Computer Time (CP sec) 10.38 (for RS)
10.41 (forpr)

11.6 ASM and ISM in SYSTEM RELIABILITY

System reliability has been introduced in Chapter 10. If the system probability
of failure is formulated as union of component failure events, i.e.

Pr. = P [(Zt < 0) u (~ < 0) u ...... .... u (ln < 0)] (11.19)

then the adaptive sampling method or importance sampling method can be

applied to evaluate system probability of failure also. The method is so
developed that it evaluates the probability of failure for each component and the
system simultaneously. As a sampling density is used for a component, it is
required to use the same sampling density for system also. The sampling
density for a system is taken as the combination of all these component
sampling densities with weights Wi attached to every component sampling
density (11.6). This is shown in Fig. 11.12. The sampling density for a system
is expressed as
h.Y' = Wtht + w2h2 + ..................+ Wnh, (11.20)

where w, + Wz + ...... ...................+ Wn = 1

 367

0 x,
(0,0)

FIG. 11.12 System sampling density

Karamchandani (11.6) suggests equal weights i.e.

W1 =w2 = ............... =wn

The procedure for evaluating the system probability of failure is very much
si milar to tlte procedure used for component except tltat the sample generated is
checked for failure for not only the component (of which sampling density is
used to generate a sample) but also for all other components. If the sample
point is observed to be failed wit11 reference to one or more components, the
sampling density for tltese components is updated if required. Also the
sampling density for a system is updated. The probability of failure of the
system is given by

(I 1.21)

The procedure of computation of probability of failure of a structural system

using ASM is illustrated with the following examples.

EXAMPLE 11.7 A structure can collapse under any one of the three limit
states whose equations are given below.

gt(X) = Zt =X,+ X2 + X. + Xs - 5 . 0~ (11.22)

g2(X) = ~ =X, + 2.0X3 + 2.0X. + Xs - 5.0~ - 5.0X7 (11.23)
~(X)=~= x2 + 2.ox3+ x.- s.ox1 (11.24)
368

The individual variables are uncorrelmcd and lognormally dis!ributcd. Tho

statistics of variables are given in Table 11.13. Compute sy tem failure
probability using ASM.

TABLE 11.13 Statistics of Variables- Example 11.7

Variable Mean Standard deviation Distribution
X1 to x~ 134.9 13.49 Lognormal
Xt 50.0 15.0 Lognormal
X: 40.0 12.0 Lognormal

Here the variables are uncorrelated: hence the evaluation of transformation

matrix is not required. For ASM, any point on the failure surface can be
selected as startinP, point. However to get a good starting point, FORM is
carried out and design point for each failure criterion is found out. They are as
follows:

I 1.281 \31.616 134.900

131.281 134.90 129.878

•
It =
134.90
131.281
•
12 =
128.473
128.473
.=
IJ
125.231
129.878
131.281 131.616 134.900
90.484 84.22 50.000
40.0 59.528 81.053

A sample point is found out from the first starting point x1' using the generated
random numbers as explained in Example 11.3 . The sample point (for SDM =
2) is given by
130.0149
109.3791
156.8981
x= 122.4430
114.6321
115.6861
26.9407
The values of limit state functions are calculated by substituting the value of
sa mple poinlin Eqs. 11.22 - ll.24. They are as follows:
gl () =-101.94 ;g2 () = 90.2152 ; g3 () = 410.9148
Si nce only g1 < 0. the sample point fails under first failure criterion only. The
equivalent mean and standard deviation of variables are calculated as explained
in Example ll.l. They are as follows:

J
 369

134.1642 12.9711
131.7732 10.9107
132.4160 15.6508
Jix = 133.6971 ax= 12.2138
132.7642 11.4347
13.6572 33.9608
36.4279 7.9087
The independent equivalent nonnals are calculated as explained in Example
11.1. They are given as:

n -0.3183
10 -2.0525
is
1.5643
IS
Y, = -0.9214
-1.5822
3.0043
-1.1996

The probability density fx and sampling density hx which are multivariable

normal densities are calculated as explained in Example 11.3; As there are
three failure criteria, there are three different sampling densities namely hx 1 ,
hXl and hXJ .Their values are given below:

fx = 0.4092 X 10"12
t hx1 "' 0.23767 X 10"10 ; bXl = 0.20109 X 10"12 ; bXJ = 0.32618 X 10"10
The prob~1bility density evaluated at x1• is taken as fmax . This is given as
fmax = 0.69226 X 10"9

Since the sample point Y 1 fails under first failure criterion only, for evaluation
of probability of failure under first fail11re criterion, the term fx/hx. 1 is used in
Eq. 11.2.

Here the system is expressed as union of three failure criteria. Hence failure
of any criterion causes the faHure of the system. Thus here the generated
sample point which fails under first failure criterion causes the failure of the
system. Fc.t valuation of system failure probability, the tenn fxji'Xsys is used
in Eq. 11.21 where hx >ys is given by Eq. 11 .20. Attaching equal weights i.e. W1
= w2 == W3 = l/3 ,
I I 1
hxoy, -hx +-hx +-hx
3 I 3 2 3 3
= o.I862 x w-to
370

As the probability density fx at sample point is less than fmax , the sampllna
density is not improved and is kept the same. Tha means, the point x 1• is not
shifted and remains same for the next simulation. Similarly using points ll•
and x3·,sample points Y2 and Y3 are generated as explained earlier. The pro·
cedure of calculation for h x for points Y2 and Y3 is as explained for Y 1.
"1•

All the intermediate values are given in Table 11.14. With this the thnlC
simulations are completed.

TABLE 11.14 Intermediate values in computing system probability of

failure using ASM- Example 11. 7

Initial point Independent fx hx h,y, Next point

Nonmd point x ro- 12 X JO· II X 10-10

131.281 -0.3183 0.41 2.4 0.18 131.281

131.281 - 2.0525 O.Q2 131.281
134.90 3.2 134.90
x,• = 131.281
1.5643 X,*=
131.281
y I = -0.9214
131.281 131.281
-1.5822
90.484 90.484
3.0043
40.0 40.0
-1.1996
Not improved

"·"" "]
0.75 2.0 0.16 131.616
131616!
134.90 -I 6~01 0.018 134.90
128.473 - 02101 2.7 128.473
Xl* = X*=
2
128.4 73 128.473
Y,- r i41JH
IJ 1.616 131.616
-·I JI)IJ7
84.22 84.22
.\ 21 'IX
59.52!1
I 1.2blJI':
59.528

Not improved

134.90 I I J, )() 1:\ 24 2.9 0.21 134 90

129 878 I 'J :;95(1 0.0\7 \29 .878
125.231 125 .231
X)* = J UT\<1\ 3.2 X/=
129.878
--IU(,Q.l
]"''78
134.90
50 0
Y2'·'

I J.4 752
134.90
50.0
2.1::43
81.053 81 .053
2.2274
Not improved
 371

The same procedlll'C is repeated for a number of simuhitions and the

probability of failure under each failure criterion and the probablJity of rallurc
of the system is evaluated using Eq. 11.21. After 500 simulations and usi rP
standard deviation multiplier as 2.0, the probability of failure under the tlrrec
failure criteria and the probability of failure of the system are found to be,
p fi = 2.68 x 10~ Pr
3
2
= 2.98 x 10-
3
=
Pr3 2.65 x 10""
and PrI
= 5.05 x 10"3
where Pr1 is probability offailure mtder i111 criterion.

In the above illustration, equal weight has been attached to each sampling
density. One may try by attaching different weights to each sampling densities
probably according to Pr:I values. But it is found that this technique doesn't
give good results and also gives large statistical error. In general, attaching
equal weight to each sampling density is found to give better results with less
statistical error.
The procedure for evaluating system probability of failure using ISM is very
much similar to the procedure of ASM. In ISM starting point is found out
using FORM and the simulation is carried out. The difference between ISM
and ASM is that the sampling density in ISM once selected is. not improved
during ft1rther simulations . That means once the starting point is taken it is
not changed tluoughout the simulation process. The remaining procedure for
ISM is similar to the ASM procedure.
The method of computing system reliability using the method explained in
Example 11.7 has been applied to roof trusses and frames (11.10).

11.7 APPLICATION·OF ASM TO STRUCTURAL SYSTEMS

Application of ASM to compute reliability of a steel truss is illustrated.
Reliability analysis or a steel tn1ss, shown in Fig. ll.l3 is to be carried out and
the system reliability is to be found out. The truss is located in Mumbai and the
height of the building is assumed to be less than IOm. The ffi!ss has been
designed as per Indian Standard specifications (11.12, 11.13). The loading
cases consideJed for design are as follows:
I. Dead load + Live load
2. Dead load+ Wind load

Formulation of safety margin equation

The safety margin equation is basically formulated as
M=R-S
where R is a resistance and S is an action. The resistance and action are further
modelled as explained below.
.I i2

1l
6.75Qm

FIG. 11.13 Roof truss

The action S here is developed force in member due to dead load, live load,
wind load. Thus S is expressed as
S B f(W,D,L) (11.24)
where
B Uncertainty due to assumption in analysis
f(W, D L) = Force in member due to dead load live load, wind load
The resistance R of a member is a resistance in tension or compression. R is
expressed as
R = A Y" f(M, F, P) (11.25)
where
A = Cross sectional area of a member
Yn =Nominal value ofyield strength
M = Material variability
F = Fabrication variability
P = Professional Factor
Statistics of strength variables
The length of each member is assumed to be statistically independent of
each other. The yield strength of a member is expresses as
Y = Yn M F P A (11.26)
where A is assumed to be deterministic. The statistics of variables M, F, P are
given in Table 11.15. For compression member buckling is considered by
taking into account the effective slenderness ratio.
,,, __ KL I
"'
I' T(
[f'
E
(11.27)
where
-KL = Effiecllvc
. sIen dcrness ratto
.
r
E = Modulus of elasticity
Ctl = Slenderness constant
 .J73

The critical strength in buckling, y ar is obtained from th.o followinl equations

depending upon value of a>.
Yar = (I • 0.25 0>2 ) Y if()) < 1.414 (11.28)
y
Yar= 2 ifm ~ 1.414 (11.29)
Cl)

Statistics of load varlabks

i) Basic wind velocity
The lruss is located in Mumbai for which the mean, standard deviation and
coefficient of variation of SO year life time maximum wind speed are
JJv = 32.239 m/s av = 3.417 m/s 6v = 0.106
The model for wind load can be expressed as (Refer Eq. 5.49)
W=ApKCEGV2 (11.30)
where A = Projected surface area, p = Air density, K =. Uncertainty in
modelling of load, C = Pressure coefficient depending on geometry of a
structure, E = Exposure coefficient and G = Gust factor. The nominal design
wind load is given by
2
Wn =An Pn K,. C,. E., Gn Vn (11.31)
The variable is considered as ratio of probabilistic wind load tol
the nominal
wind load and is following Type 1 extremal (largest) distribution.
W ApKCEGV 2
(11.32)
-W-n = -A-n p- ,-'-
, K-n-C-
nE
...._n_G_n-Vn-=-
2

(11 .33)

(11.34)

The combined mean of (K E C G) is taken as unity and the coefficient of

variation of C, E and G are given as
8c=0.12 8s=0.16 8a=O.ll

The nominal wind speed for Mumbai is 40.04 m/s. Using the same in Eqs.
11.33 and 11.34, •
32.239 )2
(Vn.)2 :: (-
J.lV
JJ{wfWn) = - - = 0.648
40.040

8(w;wn) = ~0.12) 2 +(0.16)2 +(0.11)2 +4(0.106)2~ = 0.312

ii) Live load

For 50 year life time maximum live load, the following data are taken.
374

JJL = 0.65 Or. = 0.3

4r
The live load is following Type 1 extremal (largest) distribution

iii) Dead load

The mean and coefficient of variation for dead load D are given as follows.
JJD = 1.05 8o = 0.1
D,

iv) Uncertainty due to assumption in analysis

From analysis point of view, members of a structure are assumed to be
connected by pin jointed frictionless hinges; but each joint has some rigidity
which actually decreases the force in a member. The statistical data for the
variable B is taken as
I!B = 0.909 8a = 0.1
The statistical data of all the variables are given in Table 11.15. In the table,
Dn, Lru and Wn, are the nominal values ofD, Land W respectively. Failure of
a member in direct tension or compression is called as a failure mode in a
determinate roof truss. The analysis of truss is carried out and the forces in
members under the load combination of (a) Dead load+ Live load (b) Dead
load + Wind load are determined. The truss, being determinate, fails even if
one member fails. After analysing the truss and knowing the force in each
member, the safety margin equation for each member is written. Using the

TABLE 11.15 Statistics ofvariab/es (Roof truss - Fig. 11.13)

Variable Mean Standard deviation Distribution
D/Dn 1.05 0.105 Normal
L/Ln 0.65 0 .195 Gumbel
W/Wn 0.648 0.312 Gumbel
y 305.29 N/mm2 22.77N/mm2 Normal
B 0.909 0.0909 Normal
In tension
M 1.0 0.0898 Normal
F 1.0 0.05 Normal
p 1.0 0.001 Normal
In compression
M 1.0 0.0925 Normal
F 1.0 0.05 Normal
p 1.0 0.016 Normal

statislics of variabl es given in Table 11.1 5 and FORM, tlte reliability index is
calculated for each mer11bcr. fl is fow1d that the value of p for members 9, 12,
22 and 23 a re very smalJ compared to tl1e values of ~ for the remaining
members. Hence, tlle members 9, 12, 22, 23 will only contribute signillcanlly to
 375

the system probability of failure. Safety margin equations arc given in Table
11.16 only for these dominant members. Probability of failure of these membc.•·
is evaluated using ASM. System reliability is calculated usi'ng ASM with values
of standard deviation multiplier and number of simulations 1.0 and 1000
respectively as explained in Example 11.2. The same problem is solved using
ISM also and the system reliability is evaluated. Using FOPM, the value of ~
for each dominant member is delermined and using these results bounds on
system probability of failure are established as explained in Chapter 10. These
results are also given in Table 11.17. From the table it is seen that the
probability of failure of the truss is about 0.006 and the corresponding value of
pis 2.522.

TABLE 11.16 Safety Margin Equations (Roof truss - Fig. 11.13)

Failure Member Safety Margin Equation Failure in
Mode No.
9 497.76 M F P Y+B{68276.25 D-131245W) Compression
2 23 347.4 M F P Y+B(29261.25 D-74524.77W) Compression
3 12 497.76 M F P Y+B(58522.5 D-106404.7W) Compression
4 22 347.4 M F P Y+B{19507.5 D-49683.18Wl Com2ression

TABLE 11.17 Results by Adaptive Sampling Method

(Roof truss - Fig. 11.13)
Failure Reliability
mode Index (~) Probability of failure
FORM Importance Adaptive
Sampling Sampling
R ~ & ~ ~
{%) (%)
2.537 0.00562 0.00596 12.22 0.00584 11.34
2 2.722 0.00325 0.00407 17.41 0.00412 15.36
3 2.962 0.00154 0.00208 23.23 0.00188 19.96
4 3.587 0.000168 0.000629 21.16 0.000618 19.54
System failure probability bounds are 0.00578 < p6 < 0.00629
System failure probability obtain~ by ISM Pr. = 0.00605
System failure probability obtained by ASM p6 = 0.00594

In this cbapter, advanced reliability methods bave been explained aad

illustrated. It must be remembered that in general, when explicit functions for
limit states are available, response surface .method is not to be used. For
decision making problems, application of FORM is sufficient. Only in cases
where more accurate values of probability of failure are to be estimated, SORM,
ISM aDd ASM are to be used.
376

REFERENCES

II. I Fiessler, B., Neuman, H. J. and Rackwitz, R., "Quadratic Limit State
in Structural Reliability", Journal of Engg. Mechanics, ASCE, Vol
100, 1979, pp66l-676.
II 2 Breitung, K., "Asymptotic Approximations for Multinonnal
Integrals", Journal of Engg. Mechanics, ASCE, Vol. 110, 1984, pp
357-366.
11.3 Tvedt, L., "Distribution of Quadratic Foro\ in Normal Space
Application to Structural Reliability", Journal of Engg. Mechanics.
ASCE, Vol.ll6, l987,pp 1183-1197.
11 4 Harbiz, A., "An Efficient Mct110d for Probability Failure Calculation".
Structural Safety, Vol. 3, 1986, pp 109-116.
11.5 Melchers, R. E., "Importance Sampling in Structural Systems",
Structural Safety, Vol. 6, 1989, pp 3-10.
11.6 Karamchandani, A., "New Met110ds in System Reliability", Ph.D.
Thesis, Department of Civil Engg., Stanford University, 1990.
11.7 Melchers, R. E., "Search Based Importance Sampling", Structural
Safety, Vol. 9, 1990, pp 117-128.
11.8 Bucher, C. G. and Bourgand, U., "A Fast and Efficient Response
Surface Approach for Structural Reliability", Structural Safety, Vol. 7,
1990, pp 57-66.
11.9 Ang, A. and Tang, W. H., "Probability Concepts in Engineering
Planning and Design", Vol. 2, John Wiley, Canada, 1984.
11.10 Kulkarni, R. R., "Structural Reliability using Response Smface with
Importance Sampling", M Tech. Thesis, Department of Civil Engg. ,
liT, Bombay, 1993.
11.11 Himanshu, P., "Simulation Based Reliability Assessment of Structures
using Adaptive Sampling", M. Tech. Thesis, Department of Civil Engg.
, liT, Bombay, 1999.
I 1.12 IS:S00-1994, "Indian Standard Code of Practice for General
Construction in Steel", Indian Standards lnslitulion, New Delhi, 1994.
11.13 IS:875-1992(Part 3), "Indian Standard Code of Practice for Design
Loads (other than Earthquake) for Buildings and Structures", Indian
Standards Institution, New Delhi, 1994.

EXERCISE

11.1 For the problem 8.6(a) under Exercise in Chapter 8, detemune the
probability of failure of the RCC beam in shear using (a) ISM, (b)
ASM and (c) response surface with ISM
5
(Ans. Pr = 4.1 X 10' )
11.2 For the problem 8.7(ii) under Exercise given in Chapter 8, det tn J
probability of failure of the steel colwnn under combined bendinM and
axial load using (a) ISM. (b) ASM and (c) importance, swfacc WiUt
ASM.
(Ans. pr• 1.22 x to·')
 12
Fatigue Reliability

12.1 INTRODUCTION
The word "fatigue" refers to the behaviour of materials under the action of
repeated stresses or strains as distinguished from the behaviour under
monotonic or static stresses. Fatigue is defined as follows (12.1)

1<- "Fatigue is defined as the proce s of progre sivc localized pem1aucnt

slructuraJ change occurring ln a maleriaJ su~jected to condition,
which produce Ouctuating stresses and strains at some point or points
and which may culminate in cracks or complete fracture after a
sufficient number of fluctuations".

This definition implies that fatigue process occurs over a period of time or
usage and operates at local areas rather than throughout the entire component
or structure. The ultimate cause of all fatigue failures is that a crack has grown
to a point at which the remaining material can no longer resist the stresses or
strains and sudden fracture (i.e. the separation of the component into two or
more parts) occurs.
The fatigue life of a structure is determined by the sum of the elapsed cycles
required to (i) initiate a fatigue crack and (ii) to propagate the crack from sub-
critical dimensions to the critical size. The size of the crack at the transition
from initiation to propagation is usually unknown and often depends on the
point of view of the analyst and the size of the component being analyzed. For a
research worker using microscope to measure crack size, it may be on the order
of crystal imperfection or location of a 0.1 mrn crack while to the engineer on
the field, it may be the smallest crack that can be detected with the available
equipment for nondestructive tests. Depending on the nature of the structure
and the service loads applied to it, either crack initiation or crack propagation
or both phases may be important in assessing structural performance.
The need to consider fatigue damage in the design of structural components
arises when the service loading conditions involve cyclic or pulsating
variations. Fatigue can be classified into two categories; low cycle fatigue and
high cycle fatigue. For low cycle fatigue, plastic strain predominates and
ductility controls performance. For high cycle fatigue, elastic strain dominates
and strength controls perfonna:nce. The dividing line between low and high
cycle fatigue depends on the material being considered; but usually falls
between 10 and 1o.s cycles. In the case of traJwniuion towers, oftlboro
structures and bridges, their vibration amplitudes nrc within tht'
range. They come under high cycle fatigue (their life span excess of Ill' :.
For many components in high cycle fatigue, the fatigue life is dominated b)'

Ill

t..
.....=
Ill

Cy,ltS to toilurt, N

FIG. 12.1 S-N Curve obtained from constant amplitude test results

crack initiation. On the other hand, when stress fluctuations are high or cracks,
notches and other stress risers are present, fatigue craok initiates quite early and
a signifiqmt life portion of the setvicc life may be spent propagating the crack
to critical size.
The classical approach to fatigue has focussed on the S-N diagram (Fig.l2.1)
which relates fatigue life (cycles to failure, N) to cyclic stress, S, which may be
specified in terms of s~ amplitude or cyclic stress range. Common LemlS
used with S-N diagram are fatigue life, fatigue strength and fatigue limit. The

RDngt

Sm
Cyclt

0~------------------~------~

FIG. 12.2 Nomencllture for constant amplitude loading

380

fat1guc life. N. is the number of cycles of stress or strain of a specified characlc1

that a g1ven specimen sustains before fatlure occurs. There are four possible
ba ·ic parameters. which can be used in the definition of stress cycle to whi ch a
fatigue test specuncn is su~Jected Refernng to Fig. 12 .2. they are

(i) The minimum stress in the cycle Smin

(i i) The maximum stress in the cycle S'"""
(iii ) The mean stress : Sm =_!_ (Smin + Sma.J (12 . 1)
2
( iv ) The stress range : S, = Smax - Smin (12.2)

Graphical representntion of above is shown in Fig. 12.2. The cycle is fully

defined when any two of these four quantities are known. Following definitions
are also used when discussing mean and alternating stresses.
Stress amplitudeS.= _!_(Smax- Smin) (12.3)
2
Stress ratio R (12.4)

Amplitude ratio A = ~ (12 .5)

Sm
Most design engineers find it convenient to think in terms of minimum stress
and maximum stress in the cycle, which in many cases corresponding to dead
load stress and dead load plus live load stress respecti ve!. . Some times the
cycle is referred to by the stress ratio R which is defined as the algebraic ratio
of minimum stress to the maximum stress. Tensile stress is being taken as
positive and compressive stress as negative. Baseline fati gue data usually are
obtained by cycli ng testing specimens at constant amplitude stress (or strain)
until the specimen fails. Such tests are repeated several times at different stress
levels to establish the S-N curve. Generally, in an S-N curve, both S and N are
plotted on logarithmic scales and the resulting curve is a linear representing the
mean of the data (Fig.l2.3). The results do not lie on a single line but are
scattered on each side of it. Hence the line represents the mean of the data. This
scatter is inherent reatnre of the fatigue tests. ht general, the degree of scancr
tends to increa e as the applied stress is decreased and also tends to be greHter
as the stress concentration effect decreases. Ccrtai11 materials have an
endurance or fatigue limit which is a stress level below which the material has
an infinite life. For engineering purposes, the infinite life is usually considered
to be one million cycles. Knowing the endurance stress and the ultimate or
yield stress of the material, available data may be converted into a Goodman
diagram (12 .2) to account for the effect of mean stress. As value of me:m stress
increases, life (in terms of number of cycles) of the specimen will decrease. The
fatigue base line S-N data are from the case of polished smooth specimens
loaded under fully reversed stress. The endurance limit Sen obtained from tlus
 181
r · test is to be modified for design taking into account effect of various facton
:viz. size, type of loading, swface finish, surface treatment (notches, residual

log N: log K• " tog S

logS

togN
FIG. 12.3 S-N Curve on log-log plot

(notches, residual stress) temperatute and environment. The effects of these

factors are quantified experimentally through modification factors which are
applied to S~n which is obtained from the baseline S-N data.
(lU))
where K.m. is the modification factor for the size effect. The modification
factors are supposed to be applied to determine endurance limit and the
modification for the reminder 6f the S-N curve is not clearly defined. However
a conservative approach is to use these modification factors on the entire S-N
curve.
The fatigue strength at any particular life is defined as the stress at which the
S-N curve cuts the particular value ofN. Further the curve being a straight llne
in log-log plot, a linear equation can be formulated to predict the value of S for
any given value of N and vice versa. Fatigue is one of the principaL modes of
failure in bridges, offshore structures pressure vessels etc. However, most of the
civil engineers in India may not know how to check the safety and evaluate a
given bridge under fatigue. Presently there is no Indian standard code for
fatigue design and evaluation of a bridge. Even though considerable
development has taken place in reliability analysis and design, most fatigue
assessment procedures, currently used, do not take advantage of such
developments. Instc~ad, typical fatigue assessment guidelines for structural
elements require that engineers refer to stress range cycle Life curves (S-N
c:Urves). Fatigue strength is determined from S-N curve drawn approximately
setting at two standard deviations below mean curve obtained from laboratory
testing. This approach does not consider the inherent variation in loading
382

models. The evaluation of safety also does not consider the interaction between
resistance and action. Hence this way of checking does not provide consistent
evaluation or safety of joints. This has been observed by Albrecht and Moses
( 12.3, 12.4) while checking reliability of structural joints of steel bridges
designed as per AASHfO (12.5) specifications.
Performance of a structure in fatigue generally depends on number of cycles
of load, stress-str~n history, operating environment, physical properties of
materials, geometry at the crack initiation locations and other factors. In
practice, informations on these input variables are never precise, certain and
complete. Most of the parameters are subjected to significant random
variations. The fatigue process is clouded with uncertainties arising from errors
in idealization and incomplete information. Engineering decisions can be
improved if efforts are made to identify the sources of uncertainty and quantify
them. In view of these uncertainties, achievement of absolute prevention
of some fatigue damage is impossible. Therefore, risk must be considered in
stmctural design against fatigue and fracture. Since many of the parameters
involved in fatigue analysis and design, as said earlier, are random in nature,
the relevant measure of structural performance is the reliability which is taken
as
Reliability = 1 - Pr
where Pr is the probability of failure. The application of structural reliability
theory to design has several advantages (i) The use of reliability (or probability
of failure) is the most meaningful index of structural performance (ii) It
provides a systematic method of treatment of uncertainties (iii) Provides a tool
for making rational decisions (iv) All components can be designed to a
balanced reliability level thereby producing an efficient system (v) The
technique permits the sensitive studies of uncertainties with the greatest impact
on the solution to be evaluated (vi) It is a tool for establishing partial safety
factors to result designs with uniform reliability under different design
situations (vii) It is a tool for updating standards (viii) It is a tool to develop an
inspection criteria or remedial measures on existing structures.
Evaluation of fatigue reliability of joints in bridges appears to have started in
1981 (12.1, 12.3). The problem has been initially formulated on S-N curve. In
1982, the ASCE Committee on Fatigue and Fracture Reliability (12.1)
presented a series of papers dealing with the state of art on fatigue reliability
aspects and introducing fatigue reliability models for reliability analysis and
development of criteria for assuring integrity against fatigue and fracture using
principk::. of structural reliability. Afterwards, the attention of research workers
was diverted to evaluation of fatigue reliability using system approach based on
S-N curve. Fracture mechanics approach is essential for the development of
inspection and maintenance strategy. Research was carried out in applying
fracture mechanics approach for the evaluation of fatigue reliability of bridge
structures. The formulation of fatigue reliability analysis and design based on
S-N curve and fracture mechanics approaches are presented in this chapter.
 11.l S-N CURVE APPROACH
The most commonly used ~model for fatigue behaviour w1dcr co nc
amplitude loading is of the form
N sm = K (12.7)
In which m and K are empirical constants denoting slope of S-N line and
intercept on S axis respectively. N is number of cycles to failure and S is the
applied stress range. When Eq. 12.7 is plotted on log-log · scale, the S-N
relationship has a linear form (Fig. 12.3) as given below.
log N = log K - m logS (12.8)

12.2.1 Equivalent Stress Range , /

In practice, the loading on structures does not take the form of a cyclic
constant amplitude stress. Rather the loading is a sequence of variable
amplitudes and frequencies, which do not repeat themselves. For variable·
amplitude loading the concept of equivalent stress range based on Palmgren-
Miner's (P-M) cwnulative damage hypoth~ is generally used. It states that
*failure occurs when the totill sffilin energy due to n cycles of variable
amplitude loading is equal to the total strain energy from N cycles of constant
amplitude loading. That is the cwnulative damage, D, is written as
B B
D = L Dj ='E DjiNj (12 .9)
• i i
where D. is the damage incurred at stress t.·vel Sl , n; is the nwnber of stress
cycles at stress range level S1 and N; is the number of cycles at constant stress
range level S; (from S-N curve) to cause failure. B is t11e number of stress range
blocks. D is generally taken as I at failure. Equivalent stress range is calculated
as given below (12.6)
If Nr is tile total number of cycles in the life of the structure, then nwnber of
cycles, n;, in the stress range block i is given by (Fig. 12.4)

"......r:::
~
"....0
0
.... ' I
u
r::: ' I
"
"
~
::J
IT
I
I
'
I Ps(st)

FIG. 12.4 Histogram and probability density function for induced stress range
384

ni = N[p 8 (sj)L'>s]. (12 . 10)

where p8 (s; )is the probability density function for induced stress. Using Eq.
12.10 in Eq. 12.9.

(12.11)

From Eq. 12.7,

N; = K/sm
I

Substituting the same in Eq. 12.11

B
D = I
i=!

N E(Sm) (12.12)
K

E(Sm) is read as expected value of sm. For continuous random variable S, as

L'>s--+ 0, Eq. 12.11 becomes

D = J NSm Ps(s) ds
o K

(12.13)

If Sc is the equivalent constant amplitude stress range for random variable

amplitude, then
N
-:::- (12.14)
K

If D is assumed to be equal to one,

S. = [E(Sm)]l/m (12.15)
If variable amplitude stress range history is available in the form of histogram,
then
B
s. = [ I Pi sj 11m (12.16)
i=l

Where Pi is the frequency of occurrence of the ith stress range. If

si = '~-'i sn~ (12.17)
 385

where 'Pi is the ratio of the mean value of the stress range block ito the design
stress range and S,d is the allowable design stress based on design load, then
lim

s. = [
.r
1'=1
Pi '¥,m
]
s,d (12.18)

In 1983. Albrecht (12. 3) presented a lognormal format method of calculating

reliability of a structural detail of a highway bridge, p1aintaining the concept,
the resistance is given by number of cycles to failure and the load by the
applied stress range history. Load specya in t11e form of stress histograms is
replaced by a lognormal distribution of equivalent stress ranges. The fa~gue
properties of a detail are represented by an S-N curve. At any point S on the
mean regression line, the fatigue life considered as resistance is found to be
lognorrnally distributed about the point with mean ~R = log N and standard
deviation aR = O"Jog N . This defines the resisrance. This is shown in Fig.
12.5.

FIG. 12.5 Transformation of resistance

12.2.2 Load Curve

Field measurements of actual stress ranges by the application of live load or
actual load from loadometer surveys are generally available in the form of a
histogram of stress range (o truck weight) versus frequency of occurrence. For
development of load curve, stress range histograms recorded on several bridges
are required. For each stress range histogram. s. is calculated. This replaces
that histogram and provides a point for the load curve. Calculation of values of
s. for all histograms and plotting them on a vertical line through Nd reSUlts the
load curve. N4 is the total number of cycles estimated to occur in the design life
3X6

of the structure. Mean value of S, and coefficient of variation of s. arc

computed. On S-N curve (Fig.l2.5) plotted on log-log scale, point d represents
mean value of log s. along a vertical line through log Nd. For lognormally
distributed s•. standard deviation of log s.. G'tog s, , is given by

G'(og s. = [0.4343 log (1 + o}eJ 112 (12.19)

This is called as standard deviation of load, (action or load effect) G'Q. Hence
Jse represents coefficient of variation of s.. The prime added to Q represents
that it is measured along vertical line.

12.2.3 Transformation of Resistance

Calculation of reliability requires that load and resistance are expressed in
tenns of the same basic quantities i.e. either cycles to failure or stress range.
That is both load and resistance curves are to be plotted on the same ax is.
Hence. in the present case, one of the ctuves is to be lransfonncd. Transforming
t.he rcsi tance when di stribution of resista nce is plotted along the vertical line
throug h point b (Fi g . .12.5). the point with the sa me survival probability mus t
li e on the a me line parall el to the mea n resi stance. From geometry it is clea r
that

• G'R G'tog N
G'R=-=-- (12.20)
m m

erR indicates the standard deviation of resistance measured along the vertical
line. The prime added to any symbol indicates a quantity measured along a
vertical line in Fig. 12.5 (Note : If load curve is transformed, G'Q = mG'Q) . The
distance between the mean resistance and mean load, measured along the
ert ica lline d-b, in Fig. 12.5, is given by

(12.21)

Reliability index is given by

p
 = (log N -log Nd)/ ut ( 12.21)

where

(12.23)

If one is interested in evaluating design life for specified reliability index Po,
it can be calculated as follows. Using Eq. 12.22,

log Nd = log N • Po Ut (12.24)

Using Eq. 12.8, and substituting for log N,

Log Nd = (log K - Po Ut) - m log F,.. (12.25)

Here F,.. is the allowable equivalent stress range. The above equation can be
rewritten as

(12.26)

For th~ known value of Nd from the actual field data, the value of F,.. can be
calculated for a given Po. The method developed has been applied to designs
meeting AASHTO specifications (12.3). Computation of p against fatigue
failure criterion based on the· above method is illustrated in the following
examples.

EXAMPLE 12.1 For a particular joint or detail in a highway bridge, the

value of K and m from test results are

K = 0.37 x 1012 and m = 3.0

The coefficient of variation of N is 0.24. From the field data, mean value and
coefficient of variation of equivalent stress range s. calculated from 100
histograms are 36.5 N/mm2 and 0.114 respectively. The actual nwnber of
7
cycles for a 50 year design life is estimated to be 4.56 x 10 cycles. Determine
the probability of failure of the joint against fatigue.
The resistance mean S-N curve plotted on log-log scale is shown in Fig. 12.6.
The position of the actual design point d and load curve are also shown in the
same figure. Using Eq. 12.19, standard deviation of log s. is calculated.
388

AttiiiOIIU c-
{tMan S+1 clltW l

log N
Numbtr ot cycln

FIG. 12.6 S-N curve for example 12.1

2 1/2
O"(og s. = [0.4343 log (I+ 0.14 )]

= 0.0494

Here logs. is considered as load (action). Hence

O"Q =O"(og s, = 0.0494

Similarly, using Eq. 12.19,

O"(og N =[0.4343 log (1 +0.24 2 )] 112

= 0.1028
Here log N is considered as resistance. Hence

O"R =O"Jog N =0.1028

Using Eq. 12.23,

= ((3 X 0.0494) 2 + (0.1028) 2 ] 112 = 0.1804

At this given value of s. '= 36.5 N/nun2, the actual nu
joint can withstand is obtained using the resistance mean
by,

log N = log K - 3 log s.

= log (0.37 x 1012) - 3 log 36.5
= 6.881
N = 7.603 x 10 6 cycles .
The reliability index p is obtained using Eq. 12.22 for given Nd = 4.56 x 107•
P·= log N -log Nd
Ut
6.881-6.676 0.205
=
O't
= O't .
The nwnerator of the above exp~on ~,qq~ tQ 2,: O'tog N . That is
p = 0.205 = 20'Jog N
· Q.l8<M OJ·.804
= 1.136
The fatigue specifications for bridges give the allowable stress range as a
function of type of detail and nwuber of loading cycles Nd. They speci.JY
allowable S-N line which is sel at two standard deviations, 2 uR , to the left of
the resistance. The specifications do not make any allowance for load
variability. . ,.,,:. -' · .. · o 1· . , · _, •

EXAMPLE 12.2 Consider tbe same problem in Example 12.1. The allowable
design S-N curve is given by (12.7),
log N = 0.2306 x 1012 -3 log Sn~
The value of K for design S-N curve, ~ , can be obtained from the mean curve
by using the following equation (12.7).
~ = (K.)m-. lld
where d = 2, when design curve is drawn al two standard deviations from l11e
mean curve and L\ = 0.7893
~· = (0.37 X 10 12) (0.7893)2 = 0.2306 X J0 12
The mean S-N curve and allowable (design) S-N curve are shown in Fig. 12.7.
lfthe detail is to be designed for 2 x 10 cycles the design stress is given by the
pointe in Fig. 12.7.
log Srd = .!.3 [log 0.2306 x 1012
•
- log 2.0 x 10 6)

Srd = 48.672 N/nun2

This is the hypoilietical design point. To locate the actual design poinl, d, one
must find equivalent stress range, So, and lhe actual number of loading cycles,
Nd estimated from the data. Let us assume that the gross vehicle weigh!
distribution based on !odometer survey yielded.
390

1
. ,u1
z .
• 36·SO
r
D
~

.!
Ill
...

l•I06 4.5h 107 7.6Q3xi06

Nutnbtr ol cycln
FIG. 12.7 Mean and design S-N curves on log-log-plot- Example 12.2

p = L: Pi 'lfim = 0.35
Then s. = (0.35) 113
S,d
= o.7o5 s,d
= (0.705) (48.672) = 36.5
N/mm2
Corresponding value of Nd (from design S-N curve) is

Log Nd = log 0.2306 x 1012 - 3 log 36.5

Nd = 4.56 x 10 6 cycles.
Generally fatigue design specifications do not reflect the actual fatigue
conditions that occur. High stress range is specified with low number of stress
cycles to produce a reasonable design. But in actual field conditions, fatigue
stresses are well below this value (equivalent stress is very much lower) ; but a
much higher number of cycles. For the above value of Nd, point on the design
curve is given by the point din Fig. 12.7. This is the actual design point. The
actual number of cycles that the joint can withstand at s. = 36.5 N/mm2 is
6
7.603 x 10 cycles (Refer Example 12.1).
The reliability index fl is given by

6 6
fl = log (7 .603 x 10 ) - log ( 4.744 x 10 )
at

fi .88 I- 6.676 0.205

If the coefficient of variation of load is ~f9\ i.e. D'Q • Q, or,ff. ~ ·.~~. not
considered,

then D't =D'tog N. Hence

20'tog N
P= = 2
O'tog N

The value of probability of failure corresponding to p = 2, is

p= <I>" 1(-2) = 0.0227 E: 2.3%

In the conventional fatigue design. the uncertainty in load is not taken into
account. Because of this, for different values of o0 = 5togSe the values of P
will differ significantly. For example,

For OtogSe = 0.114,

O't = 0.1804. p = 1.136

Similarly for 5tog se =0.25,

O't = 0.337. p = 0.2716
Hen~ the conventional design will not give consistent level of safety in
different design situations.
In the regular design, one would have selected the value of design stress
range 48.67 N/mm2 for the desired life 2 x I06 cycles. Tllis is given by t11e point
e in Fig. 12.7. The detail would have been designed for this stress. But the
actual strength of joint is given by point e, for which the number of cycles that
the joint can withstand is 3.207 x 106 cycles. The distance e-e 1, is equal to d-d 1
and is equal to 2 u 108 N . Value of p is equal to 2 when uncertainty in load is
not taken in to account.

EXAtvtPLE 12.3 The mean resistance S-N curve and the allowable resistance
curve, shown in Fig.l2.8, for a detail are given by

log N ,;, log(0.37 x W 1 ~) - 3 logS

log N = log(0.2306 x 1012) - 3 logS

392

1..
z ~twtSrd
~45-11 '":".? - - - -

e 3"·"1 ~«?~~~- --- --- -- ­

:~
-
I f)

.,
6·6 6 .,sa
( lo94X 10~ (log I.Oh ~)
N~Mn~r ot (yclts

FIG. 12.8 Determination of allowable design stress range - Example 12.3

From the load history, equivalenl conslanl amplitude stress range is equal to
0.75 S,d a11d the number of loading cycles is 4 x I06 cycles. It is given:
oN = 0.2; ; ose = 0 .12

Determine the allowable stress range for design based on equivalent truck
weight and for design based on design load for the desired level /} 0 = 2.
Using Eq. 12.19,
112
CTJog se = [0.4343 log (1 + 0.12 7)] = 0.052
CTJog N = (0.4343 log (1 + 0.22)] 112 = 0.086
Using Eq. 12.23,
cr 1 = [(3 x 0.052)2 + (0 .086)2] 112 = 0.178
Reliability index is given by Eq. 12.22
p= log N - log Nd
CTt

For the desired reliability level f3 = /3 0 = 2,

log N -log Nd = 2 x 0.178 = 0.356
The design life Nd is
log Nd = log N - f3 cr1
Using the mean resistance S-N curve,
log Nd = (log K- m log S.) - f3 cr1
= log (0.37 x 10 12) - 3 log s.- 0.356
= ll . 212-3log~.
 .•.18:1

Using the above equation, s. can be calculated for given Nc1 • 4 lt 10•. Heaoe,
s. = .!.3 [11.212 - log (4 x 10~)
S. = 34.408 N/nun2
This is the aliowable equivalent stress range for Po = 2. Knowing s. = 0.75 x
S,d, the allowable stress based on design load
Sd = 34.408 = 45.88 N/mm2
' 0.75
For s. = 34.408 N/mm2, the corresponding value ofN from resistance curve is
log N = log (0.37 x 1012) - 3 log 34.408
= 11.568 - 4.61 = 6.958
N = 9.078 x 106 cycles

J2.3 LRFD FORMAT

In LRFD format( Refer Chapter 9) uncertainty in random loading can be taken
care of ex'PliciUy. Adoption of the format makes the designer to detennine
partial safety factors to resistance, 'YR and partial safety factor to Load 'Ys , for
the desired reliability level. Smith and llirt (12.8) proposed a safety format
similar to LRFD format for calibrating European convention for constructional
steel works (ECCS) 1985 standards. For safety
SRI 'YR ~ 'Ys S. (12.27)
The fatigue strength SR is defined by the S-N curve corresponding to the
detaiVjoint which is evaluated. The equivalent constant amplitude stress range
s. is calcul~ted from U\e resulting stress ttistories due to Ute application of
design load spectra and applying Ute reservoir or rain flow method of cycle
counting. The safety fat.:h.)r 'YR reflects the uncertainty quantified by
• variations in effects of fabrication, workn1anship, size, shape, local stress
concentration and fatigue crack shapes
• size of detail, residual stresses, metallurgical effects.
Total uncertainty in fatigue strength is represented by b'R.
The partial safety factor 'Ys reflects the uncertainty
• in estimating the effects of stress analysis
• due to errors in fatigue model and use of Miner's rule
• in developing stress histories due to loads and detennining stress ranges
and counting numbel of cycles using rain flow or reservoir method
• in estimating the equivalent constant amplitude effects of the design
spectrum.
 J94
.,
Total uncertamty in load (action) is represented by 8se. In the lognonnal
safet) format all variables are assumed to be lognormally distributed. When
limit state equation is written in terms of stress ranges, Eq. 12.22 becomes
fl= logSR-logSe
[(a~)2 +(a~)2)\ / 2

If the resistance curve is defined in terms of design S-N curve drawn at

2 CTJog
N from mean S-N curve, then the above equation becomes

jJ = log S R + 2 a R - log S e
(12.28)
(a 1 im)

Taking loganthm on both sides of Eq. 12.27

log SR = log S. + log Ys + log YR
Substituting the same in Eq. 12.29, reliability index expressed in tenns of
partial safety factors becomes,

jJ = log r s + log YR + 2 a~ (12.29)

(a 1 1 m)

For given YR, Ys and m, one can compute f3 if JR and Jse are known from the
field data .
Since in fatigue design, design S-N curves are drawn at mean minus two
standard deviations to take care of variation in R, YR is taken as one.
Considering the same Smith and Hirt (12.8) have found that f3 varies from 2 to
3.5 at the end of service life for fatigue designs of details designed as per
ECCS. With the above format it is possible to establish pa rtial safety factors YR
and Ys directly for the specified reliability index Po. This approach has been
used in updating fatigue provisions of Swiss code for steel
design. This method of calculating f3 for given r s and r R and (ii) calculating
YR and ys for desired /3 0 is illustrated below.
EXAMPLE 12. For a given detail used in a bridge, it is found from the field
data that the values of JR and Jse are 0.36 and 0.2 respectively. For the
particular detail, m = 3, the code has specified YR = I and Ys = 1.8.
Determine f3 .
Using Eq. 12.19, standard deviation of log s. and log SR can be computed.
O'logSe = [0.4343log(l +0.2 2 )] 112 = 0.086
112
O'R =atogN =[0.4343 log(l +0.36 2 )] =0.15

crR = aR.!m
= 0.15/3 = 0.05
Using Eq. 12.23.
u1 = [(3 X 0.086) 2 + (0.0647)2)
= 0.298
The value of p is calculated using Eq. 12.29.
p = log 1.8 + log 1.0 + 2 x 0.05
(0.298/3)
= 3.3
EXAMPLE 12.4 A detail is to be designed for a reliability level of flo = 2.5.
Detem1ine Ys fixing YR = 1. It is given
ose = 0.2 ; oR= 0.36 ; m = 3
From the previous example; forlhe above values of 6 se and 6R .
• dtog sI ~ '''o.OS6~ utogR ~O.l5;u1 · ::o:298
•

Using the above values and given values of /Jo ·and min Eq. 12.29,
. = logrs +log(I.0}+2(0.15/3)
25
(0.298/3)
log
..
rs ~ o:l48 ::
. '"-. ~ ' .; :.

' ·,; }•~ Y:& · "" ~1'~97 ; .. \;; "

Hence the partial safety factor for stress range is 1.487. Similarly for different
valus of 6 se and ~, corresponding values of Ys can be calculated. Variations Ys
with. 8 se and p are shown in Fig. 12. 9. [t can be noted that as Bs increases,
Ys increases for given P. Again for given Bs. , as Pincreases Ys decreases.
2·2
OR=O·OS
'2.
396

So far failure function has been formulated based on number of cycles or

stress range. In general limit state function for evaluation of fatigue reliability
can be of any one of these when S-N curve approach is used.
i) Pr = P[ Tr < T,) (12.30)
ii) Pr =P(Dr < D, ) (12.31)
iii) Pr =P(N <Nr) (12.32)
iv) Pr = P( S. < SR ) (12.33)

Here Tr denotes actual time to fatigue failure and T, is the service life (desired
life) of the structure with which is deterministic. T is a function of seveml
random vari.ables. Dr is the cumulative damage at failure and D, is tlle specified
damage. N is the actual number of cycles that the detaiUjoint can witb stand
and N,. is the total number of cycles in time T, (desired number of cycles).
Moses el al (12.9), in 1985 have dealt with modelling of bridge loads and its
application to fatigue design of bridges in accordance with AASHTO
specifications using damage based failure criterion.

12.4 APPLICATIONS IN BRIDGES

Ravi and Ranganathan (12.10) started the formulation for fatigue reliability
assessment from Eq. 12.30. For a particular bridge in service, general
formulation of limit state equation for a bridge is explained below.
Let the limit state equation under fatigue loading is defined by
Z = Yc - Y. (12.34)
Where Yris the life at failure andY, is the specified life. Both Yrand Y, are in
tenus of years. Limit state is reached when Z is equal to zero. The damage
accumulated per year, Dy, using Miner's law (Eq. 12.9) is written as
J
Dy = 1:n;IN;
1•1

where j is the number of distinct stress ranges. Alternatively, Dy can also be

written as

all s 1
Dy = r --
,_~ N(Sj)
(12.35)

by taking each stress range into summation. Here, N(S;) is the number of cycles
to failure at a constant amplitude stress range Si. From the S-N curve,
n.:presented by Eq. 12.7, it can be wri~ as

N = _£ (12.36)
sm
Substituting this value ofN in Eq. 12.35,
 387

I all S
Dy = Is~ (12.37)
K i=l I

The true stress range for any truck crossing of a bridge depends on several
variables and may be written as
S; = Wi(l+lfi)(iei)(g)(h) (12.38)
Zx

Where W1 = i th truck crossing gross vehicle weight,

ifi = impact factor
g = lateral girder distribution (expressed as percentage of gross
span moment or force carried by single member)
h = factor to account for closely spaced or multilane presence of
vehicles which amplify the load effect
Zx = the actual section modulus or cross sectional area

i, = the influence factor which converts the load to load effect.

Influence factor is defined as
absolute m.aximwn load effect
ir = (12.39)
total load on span
Representing the volume in total number of equivalent stress cycles in a year by
V, Dy is written as

D =
Y
V [(l+ifi)(ili)(g)(h)]3
K Zx
L wtV (12.40)

The term within the sununation in the above equation can be represented by
equivalent fatigue truck weight, W"'l• which is given by Eq. 12.16
1/ m
Woo,= [
~fi wi ] (12.41)
t=l

where 11t. is the number of load categori~. f1 is the relative frequency of the load
category I and Wi is that part of the load acting on the structure corresponding
to maximum load effect for category load i. Here, maximum load effect can be
bending moment or shear force etc. Hence Eq. 12.40 can be written as

D--
_ v [Woq (1 +ifi) (iti) (g) (h)r (12.42)
Y K Zx
.
The equivalent number of cycles per year, V, can be expressed as
V = ( Nt )( Noq) (12.43)
398

in which NT .is annual traffic (lmck traffic in vehicles per day x 65 or train
traffic) and N.., is equivalent number of Ires range cycles per passage of train
or truck crossing. Thus Eq. 12.42 is rcwri tlen a

3
D = Nr Neq [ Weq0+in)Citi ) (g)(h)]
(12.44)
Y K Zx

Yc represents the life at failure when the cumulative damage of Miner's model
is equal to one. However, this cumulative damage is seen to be a random
variable, its value lying anywhere between 0.84 to 2.06 (12.1). Hence
cumulative damage at failure, X, is treated as a random variable. Knowing the
damage accumulated per year as Dy , Yr can be written as
3
y _ X K Zx
(12.45)
r- Nr Neq [ W cq (I +i n )(iu)(g)(h) ]

Hence the limit state equation 12.34 becomes

3
z= X K [ Zx ] _ y (12.46)
Nr Neq W cq I +i n)(ill)(g){h) s

The S - N curve intercept, K , is expressed as

K = N. S3
where N. is the desired life in cycles. It is calculated as
N. = Nr Neq Y8 (12.47)

where Nr and Ncq are the mean values ofNr and Noq. Hence

(12.48)

Let A Nr/Nr (12.49)

Neq
B (12.50)
Neq
and Ill l+ifi (12.5 1)
where A and B represent the volume ratio and equivalent cycle ratio
respectively. 16 is the combined impact factor which takes care of live load and
imp~ct effects. Hence Eq. 12.48 becomes
 3
Z = X Y8 [ Zx S ] _Y (1 2..52)
AB Weq l fi i.u g h s

The above equation represents the limit state equation in tenus of actual values.
This equation is nonnalized as follows. Defining

p = ~ (12.53)
Zd
in which z.t is the section modulus as per design. This is given by
Wd(l+ird)i.td hd gd
Zd = ___::...:..._-=.;;.....::..::~...:::.=.... (12.54)
Srd

Using Eqs. 12.53 and 12.54, the expression iOr Z (Eq. 12.52) becomes
3
Z= X Y,[PWd(l+ird)i.tdKdhdS ] -Ys (12.55)
AB Weq(l+ifi)i.tiSrdgh

Let
w = Woq/Wd (12.56)
1 +ifi
}F (12.57)
l+ird
i.ti
IL = (12.58)
i.td

G ...!. (12.59)
gd

Sn = S/Sro (12.60)
H = hlh.. (12.61)
Using the same Eq. 12.55 becomes

z _XY
- AB
1

[ r P Sn
W.lpiLGH

The failure surface equation becomes (ie Z = 0),

_ Y.
I
(12.62)

z- ....!.._
[
PSn
- AB W.lp IL GH r -1 = 0 (12.63)

The above equation represents the failure surface in normalized format. The
random variables included in the above fatigue criterion contains material
terms X, P and So. truck variables, W, A. B, IL and H and analysis uncertainties
I, and G. Once the probability distribution and parameters of all random
variables are knoW~\ probability of failure can be eval~ using any reliability
method. This is demonstrated with examples.
400

EXAMPLE 12.5 The fatigue reliability of a riveted railway plate girder bridge
of span (L) 32 m is to be evaluated. Here reliability for a joint in tension flange
at mid span is computed. The joint detail comes under category class D as per
British standards (12.7). Statistics of variables are given in Table 12.1. In the
case of railway bridges, the factors G and Hare not considered in Eq. 12.63.

TABLE 12.1 Statistics of variables- Example 12.5

Sr. Variable Mean 0 Median aln

No
1. X~ Model uncertainty 1.04 0.300 0.999971 0.293560
2. 0.855 0.100 0.851055 0.099751
P - Sec. Mod. ratio
3. S- Stress range ratio 1.380 0.142 1.365799 0.141292
4. A- Volume ratio 1.000 0.100 0.995037 0.099751
5. B - Equivalent cycle ratio 1.000 0.011 0.999940 0.011000
6. W- Weight ratio 0.536 0.100 0.532842 0.099751
7. IF -Impact factor ratio 1.000 0.150 0.988936 0.149166
8. h - Influence factor ratio 0.986 0.111 0.979981 0.110660

The mean value of sectional modular ratio is first computed as follows.

Assuming that the live load, given by IRS bridge rules (12.11), holds good
for fatigue design also, the design value of section modulus is calculated from
Eq. 12.54 deleting factors G and H.

(12.64)

But
Snt = (K/N) 11m

From British standards (12.7), for designS- N cwve of class D detail,

K = 1.52 x 1012 and m = 3.0
For a desired life of 2 x 106 cycles,

Sro =[t.52xtol2]113 = 91.258 N/mm2

2xt0 6
For plate girder of span 32 m, design values of Wd and lilt obtained from IRS
bridge rules (12.11) are,
wd = 1.437 x 1o6 N;
 401

.F.or.single track spans,

8
ird = 0.15 + { --}
6 +L
For L =32m, ifd = 0.361
For simply supJ.oorted uniform distributed beam,
. span 32
•ed= - =- = 4m
8 8
Hence
1.437xl06 (1+0.361)(4xl0 3 )
zd-------~----~--~
91.258
= 8.569 x 107 mm3
But the section modulus provided by Railways is
z,. = 7:33 x 10
7
mm3
Mean value of z,. is taken as the proVided section modulus. Hence the mean
value ofP is
P=7.330x l0 7 'f 0.855
8.569 X 10
7

Considering Eq. 12.63, let

R = X (P S)3 (12.65)
and Q = A. B (W lr..Id '
(12.66)
in the problem all variables are logrtorrnally distributed. Their parameters are
given in· blc 12.1. Using titem, parameters of lognormally distributed Rand
Q can be calculated as follows.
R: = x ci>s)3 (12.67)
3
= ( 0. 999971) (0.851 X 1.366) = 1.57
- · -----3
lr IL)
Q = A B (W (12.68)
3
= 0.995 X 0.999.94 (0.533 X 0.989 X 0.98) = 0.137
Using the given values of coefficie11ts of variations of variables, values of
cren R and CT£n Q are calculated as follows.

crfn R =in[(l+6i)(l+6~) 9 (1+6,h 9 ] (12.69)

2 9
= £n[(l+0.3 2
) (1+0.1 ) + (1+0.1422 ) 9]
CTen R = 0.596
2
crin =ln[(1+8 2 )(l+8 2 )(l+6 2 ) 9 (1+6 2 )9 (1+6 2 )9 ] (12.70)
Q A B W Ir IL
= en [(1+0.1 2 ) (1 +0.0 11 2) ( 1+0.1 2) 9 (1 +0.15 2) 9 (1+0.111 2) 9]
CTin Q = 0.64
402

Since R and S are lognonnally distributed, (R/S) is also lognormally

distributed. Hence reliability index is given by (Refer chapter 6).

= 2.788

This is the fatigue reliability index of the joint in the tension flange at mid span
of the bri=dge.

EXAMPLE 12.6 Fatigue reliability of the lower chord member ~ L 3 of the

riveted truss 'bridge of span 36m, shown in Fig. 12.10, is to be evaluated. The
statistics of the variables are given in Table 12.2.
The mean value of section area ratio is computed as follows. The design value
of sectional area Z<J is

TABLE 12.2 Statistics of variables- Example 12.6

Sr. Variable Mean g Median (jln

No.
1. X - Model uncertainty 1.040 0.3000 0.999971 0.293560
2. 0.985 0.1000 0.985 0.099751
P- Cross sectional (Jrea
ratio
3. S - Stress range ratio 1.380 0.1420 1.366 0.141292
4. A - Volume ratio 1.000 0.1000 0.995 0.099751
5. B - Equivalent cycle ratio 1.000 0.0065 0.99994 o:oo647
6. W - Weight ratio 0.513 0.1000 0.510 0.10
7. /p- Impact factor ratio 1.000 0.1500 0.989 0.150
8. h -Influence factor ratio 0.005 0.0990 0.990 0.099
From British standards (12.7), for designS- N curve of class D detail,
K = 1.52 x 10 12 and m = 3.0
For desired life of 2 x 106 cycles,
 403

6• 6m PODtiJ

(1) TNA bridge conftguatlon- SpM-38M.

~I:,
~_r
l
\ .. j ; " .
(b) tnnuence line for force In member ~L,

Fig. 12.10 T-:ua bridge ':" Example 12.10

s, = 1.52 x LO
12 ]l/3 = 91.258 N/mm2
.
d [ 2x 106

For truss bridge of 36 m, design values of Wd and lfd are obtained using IRS
bridge. rules (12.11 ). They- are
, ·
wd .='· 1595 ~~. . kN ; . . ..

ifd = 0.34
The influence line diagram for force in member~~ is shown in Fig. 12.10.
Using this

itd =(~ x L;d ) w~ L

=0.643
tS9's:s xto3 x(1 .34)(0.643)
Hence zd-
91.258
404

= 1.507 x 10 4 mm2
But the area provided by Railways is
Zx = 1.484 x 104 mm2
Mean value of z, is taken as the provided sectional area. Hence the mean value
ofP is
4
p= 1.484 10
= 0.9848
1.507 X !0 4
For the known or assumed 8p, median of P and standard deviation of f. n P
can be calculated. They are given in Table 12.2. The procedure of further
calculations is same as given in the previous example. The median values of R
and Q are
R = (0.999971) ( (0.985) (1.366)] 3
= 2.397
Q = 0.995 X 0.99994 (0.51 X 0.989 X 0.990) 3
= 0.124
Value of aln R is the same as calculated in the previous example i.e.

O'tn R = 0.596
Using Eq. 12.70,
CTtn Q =t n[(l+<l.099751 2) (1+<1.00647 2) (1+<1.12) 9 (1+<1.15 2) 9 x
(1 +{).099 2) 9]
CTtn Q = 0.623
Hence the reliability index is

l n 2.397)
p= 0.124 = 3.435
Jo.S96 2 + o.623 2
This is the value of fatigue reliability index for the member ~ L3 of the riveted
railway truss bridge.

12.5 APPLICATIONS IN OFFSHORE AND SHIP STRJ,JCTURES

Lognormal format
Wirsching (12.12) has formulated the fatigue reliability problem of welded
joints in offshore structures and given a closed form expression to compute Pr
assuming lognonnal format. If fo is defined as the average frequency of the
cycle, that is
NT (12.71)
T
 405

then cumulative damageD, using Eq. 12.13, can be written as

D • (f~T) E(Sj (12.72)

where Nr is the total number of cycles in time T. If spectral approach is used

for analysis of random process and if it is assumed that the process is
stationary, Gaussian aad narrow band, then (12.13),

f 0 E(Sm) =A.(m)(2./Z)"' (m2 +t)'1 G fj u."'

i•J I
(12. 73)

where fi is the frequency of wave loading in i th sea-state and ui is the root

mean Square (RMS) stress process in the i th sea-state. fi and ui can be
calculated from Ule given spectral density function wi (f) for tlle fatigue stress
range. ri is percent of time in the i th sea-state, and .t1 is a correction factor to
be used for the narrow~ JlSSUIDPtion. It is computed by calculating Di from
rain flow analysis and comparing it to the narrow band assumption. Wirsching
(12.13) bas found that ;, (m) 1t1 0.86 form= 3 and A. (m) ~'~~ 0.76 form= 4.38.
Instead of spectral approach, if Wcibull model is assumed for long tetm
distribution of stress range S, then

s :<!: 0 (12.74)

where u arul k are parameters ofthe distribution. The weibull shape parameter
k varies from O.S to 1.4 for offshore platforms and is equal to one for ship
structures (12.12). lfNr is the total number of cycles in service life T, long tenn
design stress range, Sro , is defined as
1
P[ s > src1 1 = - - (12.75)
Ny
This is the stress Sn1 that is exceeded, on the average, once every NT cycles. Sed
is also called as "once in a life time ' stress .. Hence using Eq.l2. 74,

(12.76)

Using Eq. 12. 75, it can be written as

Snt
·-exp
=u [ tn NT] Ilk
r
[-(·~ l=·- ~T
406

Or u = s,d( enNTr' ilc

If S follows Type 3 extremal (smallest) distribution(Weibull). S"' also follow.~
the same distribution with mean,

E( Srn) = um f ( ~ +I) (l2 .7X)

Using Eq. 12.77 in the above equation,

E(S
111
) = (S,d)"' [en NT r"Vk r (: +I J

E(S"') = A(m) (S,d)'" [in NT rmik r(~ +I) (12.79)

M u ~e r ' smlc states that ht ilurc u11der vanable stress range occurs when D ~ I
But random fatigue experimental results show that th critical value of the
o,; umu lattve damage at fai lure. Dr, is not <llways close to 1.0 ; but in fact vane
vuJcl~ herefore, Dr is taken as a random variable which quantifies modelling
error associated with Miner's rule. Failure can be defined as the event D >Dr.
If T denotes time to fatigue failure and letting D = Dr , the basic damage
expression Eq. 12.72 can be rewritten as
Dr K
T = (12.80)
Bm fo E(Sm)

where B is model error in estimated stress range. That is, if S is the estimated
tres range, actual stress range = 8 S Since Dr , k and 8 are random
variables, T is also a random variable. If T, is the service life of structure,
fatigue failure of a joint occurs when T < T,. Then

Pr = P (T < T,) (12.81)

Failure function is
g( ) =T - T, (12.82)
Here. T, is deterministic. If statistics of random variables Dr , K and 8 are
known, ~ can be calculated using Level 2 reliability method. If De, K and 8 are
1ognormally distributed, then

P= (12.83)
a ln T

where

0' ln T (12.84)
 407

<12.H5)

Usmg the above approach, Wirsching (12 .12) demonstrated the computation of
fatigue reliability of welded joints in offshore structures.
The model parameter B which is a random variable can be split into several
factors, as given below. which contribute to the overall variation (uncertainty)
in B Let
(12 .86)
where
BF = uncertainty due to fabrication and workmanship
Bs = uncertainty due to sea state description
Bw = uncertainty due to wave load prediction
BN = uncertainty in predicting nominal loads
BH :;: uncertainty in estimation of hot spot stress
concentration factor
The above factors are the sources which contribute to the overall uncertainty in
the estimation of fatigue stress. Any other factor can be included. If the
coeffocient of variation of each variable is known, the overall variation in B can
be computed.

(12.87)

If the variables are assumed lognormal1y distributed, the parameters B and

atn 8 of the lognorrnally distributed B can be found out as follows .

B = Bp Bs Bw BN BH . (12.88)

2 2 2 2 2 ] 1/ 2 (12.89)
a ln B = [ a ln Bp + a ln B8 + a ln Bw + a ln BN + a ln BH

Or (12.90)

Bi isthe median of Bi . Wirsching (12.12) has suggested B about 0.7 and

6a about 0.5 in evaluating fatigue reliability of joints in offshore platfonn.
Values of Dr and c5or equal to 1.0 and 0.3 respectively have been recommen-
ded. The procedure of computation of fatigue reliability of a joint in offshore
structure is illustrated with an example.

EXAMPLE 12.7 ~the reliability of a welded joint in an offshore

platform using Wirsching's approach assuming all variables arc lognonnally
408

distributed. It is given that for a 20 year life, long tenn stress range is 3Xl .l
NITtmt2 (That is S,d = 383.3 N/nun 2) and long term stress range follows Weibull
distribution. Following data are also given.
T, 20 yr k = 0.69
m 3 fa = 0.25 hertz
1.9365 X 1013 oK = 0.73
1.0 r5nr = 0.30

0.7 r5B = 0.50

0.86
Mean value of sm is first calculated using Eq. 12.79

E(Sm)= A(rn)(S,d)m [fnNTJ"mlk r(~+l)

Substituting the given values, each tenn in the above equation, is calculated as
follows .

r(~+I) = r( 0 .~ 9 +I)
= 41

NT= fo Ts
= 0.25 X 20 X 365 X 24 X 36()()
= 1.575 x 108 cycles in 20 years

[tnNr tmtk = ~n~.s?sxws)r'o69

= 2.836 X 10-6
(Std)m = (383.3) 3 = 56314010
Hence mean value of sm is
E(Sm) = (0.86) (56314010) (2.836 X 10-6) X 41
= 5629.3
fo E(Sm) = 0.25 X 5629.3 = 1407.33
Since all variables are lognonnal1y distributed, using Eq. 12.85,
B1 i
f- - ---''----
jjm fo E(Sm)

13
= (1.0)(';9365xl0 ) = 129 1. 39 )"'
(0.7) 0407.33)
 409

Using Eq. 12.84

a,,r +((1 + 6~f) ~ +6l) ~ + •§ )"' l]''

~n ~ + 0.3 2) ~ + 0. 73 2) ~ + 0.5 2)9f
2
u tnT =
= 1.588
Using Eq. 12.83, the reliability index is calculated.

= 2.62

Using the lognonnal fonnat explained above, it is also possible to detennine the
allowable (design) stress range for required service life of the structure and
target reliability level. This is illustrated in the following example.

EXAMPLE 12.8 Detennine the minimum allowable stress range for 20 year
life, for the design of a welded joint in an offshore platfonn for a reliability level
of ~o = 3 against fatigue. All th~ variables are lognormally distributed. Long
tenn stress range follows Weibull distribution. Following data are given.

Ts = .20yr. k = 0.69
m 3
Fo = 0.25 hertz A.(m) 0.86
K = 1.9 X 1013 ~ 0.7
Dt = 1.0 0Dt 0.3
B 0.7 Sa 0.5

Using Eq. 12.83

f = Ts exp Lno Utnd
Using the same in Eq. 12.85 ·

E~m) = DtK
jjm Vo Ts )exp{po UtnT)
For long tenn stress range following Weibull distribution. Eq. 12.79 gives
E(S"'). Using the same, the expression for design (allowable) stress mnge for
given ~o becomes
. 410

1
I

= [ln(j0 Ts) Jfk [ Dr K Ifm (12.91)

VoTs)A.mB 111 exp(po <7tnT )r(; +I)
The various terms in the above equation are first calculated. The value of
o- tnT , using Eq. 12.84, is _

=~n {t + 0.3 2) ~ + 0.7 2) ~ + 0.5 2)9 Jl

2
<7 {nT
= 1.579
r(mk +l)=r(-
3
0.69
+I)=4I
expl,B [(3) (1 .579)] =I 14.09
0 <7 tnT)=

fo T s = 0.25 X 20 X 365 X 24 X 3600

8
= 1.575 X 10

Substituting the above values and other given data in Eq. 12.91,

s d'- ~n (I 575 x lOs) ]vo.69 [ Ixl.9xiOI3 ]1/3

' - · ~ .575x!0 8 ) (0.86}(0.7)3 (23.52)(41)
(70.592) (4.437)
= 3 13. 2 Nlrrun ~
This is the design stress range or allowable stress range for 20 year service
period for the required reliability level 0o = 3.

Weihull format
Here N and long tem1 stress range are assumed to follow Weibull distribution. If
N 1s a random variable denoting the number of cycles to failure in variable
amplitude fatigue loading and if it is assumed that N rollows Weibull
distribution (Type 3 extremal smallest distribution - refer Chapter 3) with
parameters u and kN. tlten (12 .14)

k;-J ::: (~)-I~ (12.92)

Jl,v =ur(f+1) (.12.93)

It is to be noted that Eq. 3.133 is approximated by Eq. 12.92 and Eq. 3.131 and
Eq. 12 93 are same. Cumulative distribution ofN is given by Eq. 3.130.

n 2 0 (12.94)

If n = N1 • failure occurs when N < N1 . Hence

Pr = P[N < NT]

 4.11

. 1-eop[-(~)'"l
Ifpr < < 1, the al:ove equation can be approximated. For Pr « 1,

(N:t:Pf
Or u=~
t.PrJ''"'N
Using the above equation in Eq. 12.9.3, ~mean value of N is given by

Nr.r(- 1- +1
kN
liN= f l/kN (12.94)

Assuming Minei"'inule is applicable and D = 1 at failure, for safety

E(S)m s 1!K.. (12.95)
PN
where parameters K and N in S·N curve, are random variables with mean J.LK
and J.LN respectively. Using Eq. 12.94 in Eq. 12.95

(12.96)

(12.97)

Expression for E(Sm) is given by Eq. 12.79 assuming WcibuJI. distribution for S.
Ln the aoove treatment, kN is a function of ~ . To compute ~ . let the fati!,>ue
model be
N = f(K s·m) (12.98)

where the parameter f accounts for the scatter in the constant amplitude S-N
data. Using Taylor's series expansion, approximate value for ~ is given by

(12.99)
412

Uncertainties in workmanship and fabrication are also included in 8r . Once

overall variation in N is detennined, probability of failure can be evaluated
using Eq. 12.97. MlUlSe et al (12.15) analysed fatigue reliability of ship details
using WeibuJJ format.
Using Weibull format an expression for design stress range Snl can also be
written. If Pr. which is equaJ to P [N < NT ], is specified, then Eq. 12.96 can be
rewritten as.

Nr E(Sm) r(I 1-)

+-
_ _ _ _.....:..__
k _N..;.. = (pf)ifkN
J.lK
Assuming long term stress range follows Weibull distribution, the expression for
E(Sm) given by Eq. 12.79, can be used in the above equation. Hence

Rewriting the same equation for S,d ,

(12.100)

But from the mean S-N curve,

J.lK 1-)m
-=,s (12.101)
Nr
Here § is the value of stress obtained from the constant amplitude mean S-N
curve (frQm the test results). Using A.(m) = 1, Eq. 12.100, can be rewritten as
 413

Let

(12.102)

(12.103)

Then
(12.104)
Munse (12.15) calls Rc as reliability factor and ~ as random load factor. Hence
to get the design stress range, the stress range obtained from mean S-N curve is
to be multiplied by Rr and ~ . Here, the idea is to reduce the equivalent stress
range by reliability factor. The equivalent stress range is found by using the
mean value of the fatigue life for calculating stress range from the S-N curve.
The reliability factor contains the tenn B.~ <kN is related to B.~ ) which covers
the uncertainty of all the factors in resistance and the tenn Pr wbicJt contains the
desired level of ihe exceedance of design life. The random load factor connects
the constant amplitude equivalent stress range for tlte loading to the once in a
lifetime design stress. For ship structures, k is generally found to be l. If the
same value is used,
~ = (inNr )r{m +It11m

Using the same, White and Ayyub (12.67) have detennined the design stress
rnnges for details of ship structures.

EXAMPLE 12.9 The design stress rnnge is to be suggested for the fatigue
design of a welded structural detail in a shlp. Determine the design stress range
using Munse's approach based on Weibull format for the desired reliability level
of 0.999 for a design life, N.,. , of 108 cycles. It is given:

~ = l.l37 m = 7.0
K = 7.4 x 1021 (Mpa units) for mean S-N curve.
Since design life is given as 108 cycles Nt = Nd = 108 in Eq. 12.103.
Required reliability level = 0.999. Hence Pc = I- 0.999 = 0.001. It is known
N sm = K
Using the given values ofNt and K,

1/7
21
s= 7.4 xto
[ 10 8 ]

= 95.79 N/mm2
414

The value of kN is calculated using Eq. 12.92

kN = c~ rl ~
_l_ = (1.137) 1 OR= l.l487
kN
Using Eq. 12.102, the value of reliability factor is calculated.

li m 17
(pf)l ! kN
=
(0.00 1)1 1487] = () 3187

r(t + ks) rr(t+ 1 --1~-87 )

Taking the value ofk as I in Eq. 12 103 , the random load factor is
.; = f'n (NT )[r(m + t)j 1 ' m
7
l'n (t o8 ) {r [(7) + 1] :- 1'
= (18.42) (0.2959) = 5.45

Hence the design stress range for reliability level of0 .999 for a life of lOx cycles
is
S,d = ~~j(Rr )(;')
(95 .79) (0.3187) (5 .45)
166 38 N/mrn 2

12.(i FRACTURE MECHANICS APPROACH

Application of Fracture Mechanics for modehng fatigue crack growth'
propagation is well established (12.2. 12.16, 12.17). Fracture mechanics
provides the meU1ods by which techniques of applied mechanics can be apphed
to stmctures in the presence of a crack. In majority of fatigue situations, the
crack will occur under elastic conditions. Hence U1e size of the plastic zone at
ilie crack tip would be small compared to the crack size, thus making way for
using Linear Elastic Fracture Mechanics (LEFM) concept. Inherent assumptions
are small displacements and general linearity between stresses and strains. The
behaviour of a cracked component is characterized by stress, cmck size and
stmctural dimensions. The effect of these parameters is modelled by defining
Stress Intensity Factor (SIF), which is detennined as

k = Y(a) S .;-;;-;; (12.105)

in which a is the crack size. S is the stress acting on the component and Y(a) is
a geometric function depending on the shape of the specimen and crack
geometry.
 41S

There are. generally three models of loading which involve different crack.
surface <lliplac,ements (12.2) in fracture mechanics study. They are
• Mode I : Opening or tearing mode,
• Mode II : Sliding or in-plane shear, and
• Mode III : Tearing or out-of-plane shear.

Mode I is the predominant loading mode in most of the structures (12.2). For an
infinite plate subjected to unifonn tensile stress (Mode I), SIF is given by

(12.106)
At the moment of failure, the value of SIF reaches a critical value known as
fracture toughness which is a material parameter. Fracture toughness represents
the ultimate ability of a material to resist progressive crack extension. This
property of a material has to be detennined experimentally. It is seen that
fracture toughness decreases with increase in specimen thkk.ness upto a certain
limit beyond which it almost becomes a constant.
One of the important parameters required for application of fracture mechanics
is the crack size which can be suitably assumed or obtained by field
measurements. The parameters involved in fracture mechanics studies, like
fracture toughness, stress range, crack size, cannot be quantified exactly. There
is always a certain amount of uncertainty in U1ese parameters. Hence tile
principles of structural reliability can be made use of for estimating the
probability of failure of a structute. Here a method for finding fatigue Life is
explained using principles of LEFM as applied to fatigue.
It is well known that fracture mechanics gives a better picture of fatigue crack
growth than empirical S-N curve approach. In FM ~pproaoh, Paris law (12.18) is
used for modelling crack growth. The concept of equivalent stress range for
representing tile variable amplitude stress history is used.
Fatigue crack propagaUon is modelled using the concepts ofLEFM. The crack
growth rate is a function of stress intensity factor range which is given by

(12.107)

where kmax is the maximum SIF and kmin is the minimum SIF. The rate of
fatigue crack propagation follows Paris crack growth law (12.18) given by,

(12.108)

in which a is the crack size, N is the number of cycles, C and n are crack
growth parclllleters. C and n have to be determin,ed experimentally. Figure
12.11 represents the typical crack growtl1 rate curve. ~e curve has three distinct
regions. Region I begins wiili a threshold value of Sl.F ra11ge. 1:!.. klh , below
which crack does not propagate. Region n is the zone in which the plot is linear
where Paris Law holds good. Region III has a steep slope and the curve
approaches the maximum stress intensity factor range which is equal to the
416

fracture toughness of the material. The steep gradient indicates unstable crack
extension.

I
I I
I
I
...
z I
I~-.-Ill
~•-•__, ~ion 11

b.Kth
Str4'•• inl4'-\ly rangoo ,log b.K

Fig. 12.11 Regions of fatigue crack growth

The general expression for stress intensity factor range is

1:1 k =Y(a)S~ (12.109)

in which S is the far field stress range from applied load. In actual situations, the
stress range is not of constant amplitude, but of variable amplitude and
frequency. For such a case equivalent static stress range, s. is determined, and
the same is used in Eq. 12.109. Hence stress intensity factor range, 1:1 k, becomes

1:1 k=Y(a)S 6 ~ (12.110)

Y(a) depends on the dimensions of the component. For various shapes and crack
configurations, equations for determination of SIF are available (12.2, 12.19,
12.20). Once the expression for SIF is known, fatigue propagation life can be
determined from Eq. 12.108 by separation of variables and adopting numerical
integration. The fatigue life

N (12.111)
 417

where aria the initial crack size and It is tbc final crack size. N is the tnJmber of
cyclea req~ for the crack to grow from a, to 8( . Using Eq. 12.109 in the
above equation.

(12.112)

For constant stress rangeS, and Y(a) constant (that is Y(a) = Y) during crack'
growth from Iii to ar over N cycles, the above equation simplifies to

(12.113)
~-1
- aJ
This correspond to an S-N curve N sm = K and suggests that the constant K
can be expressed as a function of more basic quantities. Final crack size using
the above equation becomes,

(12.114)

For stress cycles of varying amplitude, Eq. 12.113 may be used as S-N curve
equation and .reliability analysis can be carried out as explained earlier lUlder S-
N curve approach.

For reliability analysis two separate types of failure criteria can be used.

i) Failure occurs when the crack developed exceeds the predetermined or

specified critical size a., . The limit state function is written as

Z=a.,-a (12.115)

This criterion is based on the concept Lt-aat when the crack has developed to the
size a., , it becomes unstabk and the component is assumed to fail. This comes
under serviceability limit state.

ii) Failure occurs when the stress intensity factor K at the leading edge of
the crack exceeds the fracture toughness K., . -ryte limit state function is

Z=K.,-K
= K., - Y(a) S ~ (12.116)
418

The resistance is characterized by the material parameter K., . The criterion

comes under ultimate timit state. If small variance approximation is used.
reliability mdex is given by (12.1),

en Kc
0.637 S (rriiJ 12 (12.117)
fJ

Hence p can be calculated if statistics of K,. , S and a are known. Here K"
means the sample mean value of K.: .

For the development of inspection strategy and maintenance, it is necessary to

know the number of cycles required to propagate the cr.tck from a to a crack
size ac . The general expression for stress intensity factor range Me is given by
Eq. 12.109. Y(a) depends on crack shape, size and other factors. Sometimes Ilk
is generally written as

(12.118)

where ki are correction factors for crack shape, free surface effect, fmite width
effecL stress gradient effect etc. Equations for stress intensity factors are
available for a variety of problems (12.2. 12.19, 12.20). The expression for SIF
being known, the fatigue propagation life can be determined from Eq. 12.108 by
separation of variables and adopting numerical integration. Hence fatigue life,
N, is given by
a!
N = f da (12.119)
a1 C(M)"
The final crack size is calculated using Eq. 12.110

(12.120)

where K, is the fracture toughness. Equation 12.120 is to be numerically solved

since Y(a) is a function of a . Newton Raph.soo method can be used. The scheme
of oomputation for ac is as follows.

i) For the problem on hand, appropriate expess:ion for SIF is selected.

 419

ii) Knowing the initial cracks size and the fracture touglmess, the final
crack size is computed using Eq. 12.120. In the expression for SIF, a1
is substituted for a.

Monte Carlo Simulmion

The variables involved iii the scheme of computation for N are random variables
in nature. Hence the number of cycles to fatigue crack propagation will also be a
random variable which brings the concept of probability of failure. Monte Carlo
technique is generally used for computing probability of failure for various
desired number of cycles. The scheme of computation is as follows:

i) Knowing the distribution and parameters of random variables

considered (say ai , m, K., , model parameter attached to the calculated
stress range etc.), random values are generated for each of the
variables.

ii) The final crack size, ac , is computed using the generated values at the
given stress range level and using Eq. 12.120.

iii) Knowing a;, and ar, number of cycles elapsed for the crack propagating
from a;, to ar is determined from Eq. 12.119.

iv) The desired life in terms of cycles, N, , being given, the limit state
function is

Z = N -N.

in which N is the number of cycles computed in step (iii).

v) Steps (i) to (iv) are repeated for a number of times say, n. , to get an
ensemble of realizations for Z.

vi) The probability of failure is then calculated as

Pr=-
,
nr (12.121)
n,

where 11c is the nwnber of times Z < 0 during simulation Reliability index is
taken as
P=-~- 1 (pr)

The number of simuJations, n. , is fixed on the Schooman's error criterion (Refer

Eq. 7.37 in Chapter 7).
Considerable wo.rk bas been done on the fatigue reliability evaluation of
riveted railway steel bridges in India (12.23, 12.24), welded steel bridges in
U.S.A (12.22) and marine structures (12.21, 12.25, 12.26) and application of
fatigue reliability to offshore platform inspection (12.27, 12.28).
420

REFERENCES

12.1 ASCE Committee on Fatigue and Fracture Reliability,''(i) Fdigue

Reliability: lntroductjon, (ii) Fatigue Reliability : Qualily
Assurance nnd Maintainability, (iii) Fatigue Reliability : Variable
Amplirudc Loading and (iv) Fatigue Reliability : Development of
Criteria for Design", Journal of Structural Engineering, ASCE.
Vol. 108, 1982, pp.3-88.

12.2 Bannantine, J.A. Comer, J.J. and Handrock, J.L., Fundamentals

of Metal Fatigue Analysis, Prentice Hall, Englewood Cliffs, New
Jersey, 1990

12.3 Albrecht, P.," F<ttigue Reliabilicy Analysis of Highway Bridges",

Probabilistic Fracture Mechanics and Fatigue Metho'l'· .
Applications for Structural Design and Maintenance, ASTM -
STP 79 , J.M Bloom and J. C.Ekvall, Eds. , ASTM 1983, pp. 184-
204.

12.4 Moses, F., Schillinq, C.G. and Raju, K.S., "Fatigue Evaluation
Procedures for Steel Bridges', NCHRP 299, TRB, Washington
D.C., 1987

12.5 "Standard Specifications for Highway Bridges". 12th edition,

American Association of State Highway and Transportation
Officials (AASHTO), Washington, D.C., 1977.

12.6 White, G.J. and Ayyub, B.M., "Reliability Based Fatigue

Design", Naval Engineers Journal, ASNE, Vol. 99, pp.l35 - 149.

12.7 BS : 5400 : PART 10 : 1980, "Steel, Concrete and Composite

Bridges-Part 10: Code of Practice for Fatigue", British Standards
Institution, 1980.

12.8 I.F.C. Smith and M:A.Hirt, "Fatigue Reliability: ECCS Safety

Factors", Journal of Structural Engineering, ASCE, Vol. 113,
ST-3, PP.623-628.

12.9 Nyman W.E. and Moses, F., "Calibration of a Bridge Fatigue

Design Model", Journal of Struchlral Engineering, ASCE,
Vol. 111, ST 6, JlDlC 1985, pp. 1251-1266.

12.10 Raoganatban, R and Ravi G., ..Fatigue Reliability Analysis of

Riveted Stccl Bridges", Proceedings of National Conference on
Civil Engineering. Materials and Slnlchlres ( NC-CFMS),
Hydcrabad, Jan. 1995, pp. 424-431.
 421

12.11 Bridge Rules (in SI units) incorporating correction slips 1 -16,

Revised 1964, Ministry ofRailways, Government ofIndia

12.12 Wirsching, P.H., 'Fatigue Reliability of Offshore Structures',

Journal of Structure Engineering, ASCE, Vol. 110, ST 8, October
1984, pp. 2340 -2356.

12.13 Wirsching, P.H., "Fatigue Reliability in Welded Joints of

Offshore Structures", Proceedings of Offshore Technology
Conference, Paper 3380, Houston, Texas, 1979, pp. 197-206.

12.14 Wirsching, P.H., "A Review of Modern approaches to Fatigue

Reliability Analysis and Design". Proceedings of fourth National
Congress on Pressure Vessels and piping Technology on Random
Fatigue Prediction, Portland Oregon, June 1983, Ed. Y.S Sllin
and M.K. Au Yang, Published by ASME, New York, 1983, pp.
107- 120.

12.15 Munse, W.H., Wilbur, T.W., Tellalian, M.L., Nicol~ K and

Wilson K., "Fatigue Characterization of Fabricated Sllip Details
for Design", Ship Structuress Committee Report- 318, 1983.

12.16 Rolfe S.T. and Barsom, J.M., Fracture and Fatigue Control in
Structures - Applications of Fracture Mechanics, Prentice Hall,
Englewood Cliffs, 1977.

12.17 Parker, A.P., The Mechanics of Fracture and Fatigue, E and F.N.
Spon Ltd. New York, 1981.

12.18 Paris, P. and Erdogan, F., "A Critical Analysis of Crack Growth
Propagation Laws", Journal of Basic Engineering, Vol. 85, 1945,
pp. 528 - 534.

12.19 Rooke, D.P. and Cartwright, D.J., "Compendium of Stress

Intensity Factors", Her Majesty's Stationery Office, London
1976.

12.20 Stress Intensity Factors HtJI?dbook, Vol. I, Ed. Y. Murakami,

Pergamon Press, Oxford, 1988.

12.21 Rackwitz, R, "Probabilistic Deterioration Models and

()ptim.izlltion of Inspection". Bridge Evaluat1on, Repair and
Rehabilitation Pro~edings of US EIII'Opean Workshop, Eds. A.S.
Nowak and E. Absi, 1987, pp. 359 - 364_.

12.22 Yazdani. N. and Albrecht, P. "Risk Analysis of Fatigue Failure of

Highway Steel Bridges" Journal of Structural Engineeri11g.
ASCE, Vol 113, ST 3, March 1987, pp. 483 • 500.
422

12.23 Ravi, G. and Ranganathan, R., "Fatigue Crack Growth Reliability

of Riveted Bridges", International Journal of Structures, Vol. 14,
No.2, 1994, pp. 103-104.

12.24 Ravi G.; 'Fatique Reliability Analysis and Design Approach to

Riveted Steel Railway Bridges', Ph.D. Thesis, Indian Institute of
Technology, Bombay, 1993

12.25 Wirsching, P.H. and Torng, T.Y., " Fatigue Reliability and
Maintainability of Marine Structures", Marine Structures, No.3,
1990, pp. 265 -284.

12.26 Ayyub, B.M. White, G.J. and Purcell, E.S., "Estimation of

Structural Service Life of Ships", Naval Engineers Journal, Vol.
101, May 1989, pp. 156- 166.

12.27 Ashok Kumar and Karsen, D.I. "Fatigue Reliability of Parallel

Systems", Journal of Structural Engineering, ASCE, Vol. 116, ST
3, March 1990, pp. 719 -729.

12.28 Karsan, D.I. and Ashok Kumar, "Fatigue Failure Paths for
Offshore Platfom1 Inspection", Journal of Structural
Engineering, ASCE, Vol. 116, ST 6, June 1990, pp. 1679- 1695.

EXERCISE

12.1 Determine the fatigue reliability of a detail tn a bridge designed as per

LRFD fonnai. It is given:
~ = 0.36 8s. = 0.15 m =3
YR = 1.0; Ys = 1.536
( Ans. p = 3.5)2

12.2 A detail is to be designed for a reliability level of Po= 3.5. Detennine

Ys fixing YR = 1.0. It is given :
8s. =0.25 ~ =0.36 m =3
( Ans.ys = 2.056)

12.3 Determine the fatigue reliability of a welded joint in an offshore

platform using Wirsching's approach for a 20 year life and long term
~design stress range S,d = 383.3 N/mm2 • It is given:
Ts 20 yr k 0.69 f..(m) = 0.79
m 4.42 /o 0.25 hertz
K 9. 22 x10 15 8K 1.35
B 0.7 8a 0.5
-·
Dr 1.0 r5D = 0.3
1
 42S

Long term stress range follows Woibull distribution and all other
t~> variables arc lognonnally distributed.
( Ans. p = 2.09)

12.4 Detmninc the design stress range of a welded detail in a ship using
Munse's approach based on WeibtJll format for the d~ fatigue
reliability level of0.999 and for a design life of 101 cycles. It is given:
~ • 0.78 ; m• 3.71
K - 2.53 X 1014 (Mpa units)
(Au. 109,?Nimm')
 APPENDIX A
Standard Normal Tables

TABLE A I Cumulative probability of staudard normal variate = 'll(u)

u ~(u) u 'll(u) II 'll(u)

0 .50000

-.01 .49601 -.37 .35569 -.73 .23270

-.02 .49202 -.38 .35197 -.74 .22965
- .03 .48803 - .39 .34827 -.75 .22663
-.04 .48405 -.40 .34458 - .i6 .22363
·-.05 .48006 -.41 .34090 -.77 .22065
- .06 .47608 - .42 .33724 -.78 .21770
- .07 .47210 -.43 .33360 - .79 .21476
- .08 .46812 - .44 .32997 -.80 .21186
-.09 .46414 -.45 .32636 -.81 .2089i
-.10 .46017 - .46 .32276 -.82 .20611
-.11 .45620 -.47 .31918 -.83 .20327
- .12 .45224 - .48 .31561 -.84 .20045
-.13 .44828 - .49 .31207 -.85 .19766
-.14 .44433 ·-.50 .30854 -.86 .19489
-.15 .44038 -.51 .30503 -.87 .19215
- 16 .43644 - .52 .30153 - .88 .18943
-.17 .43251 -.53 .29806 -.89 .18673
·-.18 .42858 - .54 .29460 -.90 .18406
- .19 .42465 - .55 .29116 - .91 . \8141
-.20 .42074 - .56 .28774 -.92 .17879
-.21 .41683 -.57 .28434 - .93 .17619
- .22 .41294 -.58 .28096 - .94 .17361
-.23 .40905 -.59 .27760 -.95 .17106
-.24 .40517 -.60 .27425 -.96 .16853
-.25 .40129 -.61 .27093 -.97 .16602
-.26 .39743 -.62 .26763 -.98 .16354
-.27 .39358 -.63 .26435 -.99 .16109
-.28 .38974 -.64 .26109 -1.00 .15866
- .29 .38591 -.65 .25785 -1.01 .15625
-.30 .38209 -.66 .25463 -1.02 .15386
-.31 .37828 - .67 .25143 -1 .03 .15151
-.32 .37448 -.68 .24825 -1 .04 .14917
-.33 .37070 -.69 .24510 -1.05 .14686
-.34 .36693 -.70 .24196 -1.06 .14457
-.35 .36317 -.71 .238.85 -1 .07 .14231
-.36 .35942 -.72 .23576 -1.08 .14007
 425

II •<u) u •(u) u fl(u)

-1,09 .13786 -1.60 ..OS480 -2.11 .01743

-1.10 .13!167 -1.61 .0Sl70 -2;12 .01700
-1.11 .13350 -1.62 .05262 -2.13 .01659
-hll .i ~l36 .-·1.63 .05155 --'2.f4 .01618
--1.13 .12924 -1.64 .ososo_ -2.15 .01578
-1.14 .12714 ;-'1.65 .04947,.- -2.16 .01539
~1.15 .12S07 - 1.66 .04846 -'2.17 .01500
-Ll6 .12302 - 1.67 .04746 -2.18 .01463
-1.17 .12100 - 1.68 .04648 -2.19 .01426
-l.l8 .11900 - 1.69 .04551 -2.l0 .01390
-Ll9 .11702 - 1.70 .04451 -2;21 .oms
-1.20 .11507 - 1.71 .04363 -2.22 \.01321
-1.21 .11314 - 1.72 .04272 --,2.23 .01287
-1.22 .11123 -1. 73 .04182 -2.24 .01255
-1.23 .10935 - 1.74 .04093 -2.25 .01222
-1.24 .10749 -1.75 .04006 -2.26 .OJ191
-1.25 .10565 - 1.76 .03920 -2.27 .01160
-1.26 .10383, - .1.77 .03836 -2.28 .01J30
-1.27 .10204 - 1.78 .03754 :-2.29 .01101
-1.28 .10027,' - 1.79 .03673 -2.30 .01072
-1.29 .09853 - 1.80 .03593. -2.31 .01044
-1.30 .09680 - 1.81 .03515 -2.32 .01017 .

=i~
-1.31 .09510 - 1.82 .03438 .00990
-1.32 .09342 - 1.83 .03362 .00964
-1.33 .09176 -1.84 .03288 -2:35 .00939
..,.,..J-.34 .()9012 -us .03216 -2.36 .00914
-L35 .08851 -1.86 .03144 -2.37 .00889
-1.3~ .08691 -1.87 .03074 -2.38 .00866
-1.37 .08534 -US .03005 -2.39 .00842
-1.38 .08379 "'"'1.89 .01938 -2.40 .00820
...,.L39 .o8226 -L90 ,02872 -2.41 .00798
-.1.40 .08076 -1.91 .02807 -2.42 .00776
.,...J.41 .07927 -1.92 .02743 -2.43 .00755
-L42 .07780 -1.93 .02680 -2.44 .00734
. -1.43 .076:36 -1.94 .02619 -2.45 .00714
-1.44 .07493 -1;95 .02559 -2.46 .00695
-1.45 .07~53 -1.96 .02500 -2.47 .00676
-1.46 .07215 -1.97 .02442 -2.48 .00657
-1.47 .07078 -1.98 .02385 -2.49 .00639
-1.48 .06944 -1.99 .02330 -2.50 .00621
-1.49 .06811 -2.00 .02275 -2.51 .00604
-l:SO .06681 -2.01 .02222 -2.52 .00587
-1.51 .06552 -2.02 .02169 -2.53 .00570
-1.52 .06426 -2.03 .02118 -2.54 .00554
-1.53 .06301 -2.04 .02068 -'-2.55 .00!139
-1.54 .06178 -2.05 .02018 -2.56 .00!123
-1 55 .06057 -2.06 .01970 -2.57 .00508
.,..J.56 .05938 -2.07 .01923 -2.58 .00494
-1.57 .05821 -2.08 .01876 -2.59 .00480
-1.58 .05705 -2.09 .ol831 -2.60 .00466
-1.59 .05592 -2.10 .01786 -2.61 .00453
(ConJd.)
426

TABLE A 1 (Co111d.)

u t/l(u ) II t/l(u) II <ll(u)

-2.62 .00440 -3.13 .87403£-03 -3.64 . 13632£--0.l

-2.63 .00427 -3.14 .84474E-03 -3.65 . 131 12E-OJ
-2.64 .00415 -3.15 .81635£-03 -3.66 . 12611E--OJ
-2.65 .00402 -3.16 .78885£-03 -3.6 7 . 12128E--O.l
-2.66 .00391 -3.17 . 76219£--- OJ -3.68 . I 16621::- OJ
-2.67 .00379 -3.18 .73638£-03 -3.69 .I 12131:-0J
-2.68 .00368 -3.19 .71136E- 03 -3.70 .10780E-- OJ
-2 .69 .00357 -3 .20 .68714£-03 -3.71 .10363E--OJ
-2.70 .00347 -3.21 .66367£-03 --3.72 .99611 E- -04
-2.71 .00336 -3.22 .64095£-03 -3.73 .95740£--()4
-2 .72 .00326 -3.23 .61895E-03 -3 74 .920JOE-U4
-2 .73 .00317 -3.24 .59765E-03 -3.75 .884171::-04
-2 .74 .00307 -3 .25 .57703[-03 -3. 76 .849571::-04
-2.75 00298 -3 .26 .55706E-03 -3 .77 .81624[--04
-2.76 .00289 -3.27 .53774E-03 -3.78 .78414£-04
-2.77 .00280 -- 3.2!l .51904£-03 -3 .79 .75324E-04
-2.78 .00272 -3.2'J .50094E-03 -3.80 .72348£-04
-2 .79 .00264 -3 .30 .48342£-03 -HI .69483E-04
-vm .00256 -3.31 .46648£-03 -3.82 .66726£- 04
-2.81 .00248 -3.32 .45009E-03 - 3.H3 .64072£- 04
-2.82 .00240 -- 3.33 .43423£-03 --3.84 .61517E-04
-2.83 .00233 -3.34 .418891::-03 -3.85 .59059£-()4
-2 .84 .00226 - 3.35 .40406£-03 --3.86 .56(:94£--{)4
-2.85 .00219 -3.36 .3897JE-03 -3.87 .54418£-04
-2 .86 .00212 -3.37 .37584£-03 -3.88 .52228£-04
-2.87 .00205 -3.38 .36243£-03 -H9 .50122£-04
-2.88 .00199 -3.39 .34946E-03 -3.90 .48096£-04
-2.89 .00193 -3 .40 .33693£-03 -3.91 .46148£--()4
-2.90 .00187 -3.41 .32481£-03 -3 .92 .44274£-04
--2.91 .00181 - -3.42 3131JE-03 -3.93 .42473£-{)4
-2 .92 .00175 - 3.43 .30179E-03 --3.94 .40741£-04
-2.93 .00169 -3.44 .29086£-03 -3.95 .39076£-04
-2 .94 .00164 -3.45 .28029£-03 -3 .96 .37475£- 04
--2.95 .00159 - -3.46 .27009E-03 -3.97 .35936E-04
-2.96 .00154 -3.47 .26023£-03 -3.98 .34458E-04
-2 .97 .00149 -3.4~ .25071E-03 -3.99 .33037E-04
-2.98 .00144 -3.49 .24151£-03 -4.00 .31671E-04
-2 .99 .00139 -3 .50 .23263E-03 -4.01 .30359E-04
-3.00 .00135 -3.51 .22405£-03 -4.02 .29099E-04
-3.01 .13062E-02 -3 .52 .21577E-03 -4.03 .27888E-04
-3.02 .12639E-02 -3.53 .20778E-03 -4.04 .26726E-04
-3.03 .12228E-02 -3.54 .20006E-03 -4.05 .25609£--()4
-3.04 .11829E-02 -3 .55 .19262E-Q3 -4.06 .24536E-D4
-3.05 .11442E-Q2 -3 .56 .18543£---{)3 -4.07 .23507E-D4
-3.06 .11067E-02 -3 .57 .17849£-03 -4.0ll .22518E-04
-3 .07 . 10703£-02 -3 .51) .17180E-03 -4.09 .21569£-04
-3.08 .10350E-Q2 -3 .59 .16534E-03 -4 .10 .20658E-04
-3 .09 .10008E-02 -3 .60 . 15911 E-03 -4. 11 .19783E-04
-3 .10 .96760[--03 - 3.61 .153JOE-03 -4.12 .18944[-04
-3.11 .935441:::-03 -3.62 .14730E-03 -4.13 .18138[--04
-3 .12 .904261:::-03 -3.63 . 14171E-03 -4.14 .17365E-04
 427

u •<u> u tl>(u) u tl>(u)

-4 . 1~ .16624E-o4 -4.82 .71779E-06 -5.84 .26100E-08

-4.16 .159l2E-04 -4.84 .64920E-06 -5.86 .2314JE-o8
-4.17 .15230E-o4 -4.86 ..58693E-06 -5.88 .205IJE-os
-4.18 . l457.S~ -4.88 .53043E-06 -5.90 .18175E-o8
-4.19 . l~948E-o4 -4.90 .47918E-06 -5.92 .t6097E-o8
-4.20 . l~46E-04 -4.92 .43272E-06 -5.94 .14251E-08
-4.21 :12769E-o4 -4.94 .39061E-06 -5.96 .J2612E-o8
-4.22 .122l5E-o4 -4.96 .35247E-o6 -5 .98 .tllS7E-os
-4.23 .1168SE-o4 -4.98 .31792E-06 -6.00 .98659E-09
-4.24 .11176E-o4 -5.00 .28665E-06 -6.02 .87209E-o9
-4.25 .10689E-o4 -5.02 .25836E-06 -6.04 .77057E-09
-4.26 .10221E-{)4 -5.04 .23277E-06 -6.06 .68061E-09
-4.27 .97736E-o5 -5.06 .20963E-06 -6.08 .60091E-09
-4.28 .93447E-oS -5.08 .18872E-06 -6.10 .53034E--Q9
-4.29 .89337E-o5 -5.10 .16983E-o6 -6.12 .46788E-09
-4.30 .85399E-{)5 -5 .12 .15277E-06 -6.14 .41261£-09
-4.31 .81627E-05 -5.14 .13737E-06 -6.16 .36372£--09
-4.32 .78015E-05 -5.16 .12347E-o6 -6.18 .32051£-09
-4.33 .74555£-{)5 -5.18 .11094E-06 -6.20 .28232E-09
-4.34 .71241E-05 -5.20 .99644E-07 -6.22 .24858£-09
-4.35 .68069£-05 -5.22 .89462E-07 -6.24 .21879E-09
-4.36 .65031E--05 -5.24 .80288E-07 -6.26 .19249E-09
-4.37 .62123E-05 -5.26 .72028E-07 -6.28 .16929£-09
-4.38 .59340£-{)5 -5.28 .64592E-{)7 -7.30 .14882£--{)9
-4.39 .56675E-05 -5.30 .57901E-o7 -6.32 .13078E-09
-4.40 .54125E-{)5 -5.32 .51884E--07 -6.34 .11488E-09
-4.41 .5168SE-05 -5.34 .46473E-07 -6.36 .10088E-09
-4.42 .49350£--05 -5.36 .4161JE-07 -6.38 .88544E-IO
-4.43 .47117E-05 -5.38 .37243E-07 -6.40 .77688E-IO
-4.44 .44979E-05 -5.40 .33320E-07 -6.42 .68137£-10
-4.4.5 .4293.5E--05 -5.42 .29800E-07 -6.44 .59737£-10
-4.46 .40980E-05 -5 .44 .26640E-07 -6.46 .52351E-10
-4.47 .39JJOE-05 -5 .46 .23807E-07 -6.48 .45861E-IO
-4.48 .37322£--05 -5.48 .21266E-07 -6.50 .40160E-l0
- 4.49 .35612E--05 -5.so .l8990E--07 -6.52 .35154E-10
-4.50 .33977E-05 -5 . .52 .16950E-07 -6.54 .30759£-10
-4.52 .30920£-05 -5.54 .15124E-07 -6.56 .26904E-IO
-4.54 .28127£-05 -5.56 .13489E-07 -·6.58 .23522E-10
-4.56 .25577E-05 -5 ..58 .12026E-07 -6.60 .20558E-IO
-4.58 .23249E-0.5 -5.60 .l0718E-07 -6.62 .l7960E-IO
-4.60 .21125E-o.5 -5.62 .95479E-08 -6.64 .15684E-IO
-4.62 .t9187E-o5 -5 .64 .85025E-o8 -6.66 .13691E-JO
-4.64 .17420E-05 -5 .66 .75686E-o8 -6.68 .ll947E-10
-4.66 .15810E-{)5 -5.68 .67347E-08 -6.70 .l0421E-10
-4.68 .14344E--05 -5.70 .59904E-08 -6.72 .90862E-ll
-4.70 .13008E-o5 -5.72 .53262E-08 -6.74 .79193E-ll
-4.72 .11792E-o5 -.5 .74 .47338E-08 -6.76 .68996E-Il
-4.74 .l0686E-o5 -.5.76 .42057E-08 -6.78 .60088E-ll
-4.76 .96796E-06 -5.78 .3735 OE--()8 -6.80 ..52310E-l I
-4.78 .87648E-o6 -5.80 .33157E-08 -6.82 .45520E-11
-4.80 .79333E-06 -5.82 .29424E-08 -6.84 .39597E-11
(Contd)
428

TABLE A 1 (Corrtd)

II 'IJ(u) II 'IJ(u) II •(u)

-6.86 .34430E-11 -7.88 .16369E-14 -8.90 .27923£-18

-6.88 .29926£- 11 -7.90 .13945£-14 -8.92 .23314E-J8
-6.90 .26001£-ll -7.92 .11876£-14 -8.94 .19459£-18
-6.92 .22582£-ll -7.94 .10109E-14 --8.96 .16234£-18
-6.94 .19605E-Il -7.96 .86020E-15 -8.98 .13538E-18
-6.96 .17014£-ll -7 .98 .73167E-15 -9.00 .11286E-18
-6.98 .14759E-11 -8.00 .62210£-15 -9 .02 .94045£-19
·-7.00 .12798E-11 -8.02 .52873£-15 -9.04 .78336£-19
-7 .02 .11093E-11 -8.04 .44919£-15 -9.06 .65225E-19
-7.04 .96120E-12 -8 .06 .38147E-15 -9.08 .54287E-19
--7.06 .83251£-12 -8 .08 .32383E-15 -9.10 .45166E-19
-7 .08 .72077E-12 -8.10 .27480E-15 -9.12 .37562E-19
-7.10 .62378E-12 -8.12 .23309E-15 -9.14 .3J226E-19
-7.12 .53964E-12 -8.14 .19764E-15 -9.16 .25949£-19
-7.14 .46665E-12 -8 .16 .16751E-15 -9.18 .21555E-19
-7.16 .40339£-12 -8.18 .14192E-15 -9.20 .17897£-19
-7 . 18 .34856E-12 -8.20 .12019£-15 -9.22 .14855E-19
-7 .20 .30106E-12 -8.22 . 10175E-15 -9.24 .12325E-19
-7 .22 .25994E-12 -8.24 .86105£-16 -9.26 .J0222E-19
-7.24 .22434[-12 -8 26 .72836E-16 -9.28 .84739E-20
-7 .26 .19355£-12 -8.28 .61588£-16 -9.30 .70223E-20
-7 .28 .16691£-12 -8.30 .52056£-16 -9 .32 .58170E-20
-7.30 .14388£-12 -8 .32 43982£-16 -9.34 .48J97E- 20
-7.32 .12399[-12 -8 .34 .37145E-16 -9.36 .39M68E-20
-7 .34 .10680E-12 -8 .36 .31359E-16 -9.38 .32986[-20
-7.36 .91955E-J3 -8.38 .26464E- 16 -9.40 .27282E-20
-7 .38 .79145£--13 - ·8.40 .22324£-16 -9.42 .22554E--20
-7.40 .68092E-13 -8.42 .18824£-16 -9.44 .1 8639E--20
-7.42 .58560E-J 3 -8.44 .15867£-16 -9.46 .15397E--21J
-7.44 .50343E-13 -8.46 .13369£-16 -9 .48 .12714£-20
-7.46 .43261E-13 --8 48 .11260£--16 -9.50 .10495[-20
-7.48 .37161£-13 ---8.50 .94795£-17 --9.52 .86590£-- 2l
-7 .50 .31909E-13 -8.52 .79777E-17 -9.54 .71416E-21
-7.52 .27388£-13 -8.54 .67LIIE-17 -9.56 .58878E-21
-7.54 .23499£-13 - · 8.56 .56434E-17 -9.58 .48522E-21
-7.56 .20153[-13 --8.58 .47437E-17 -9.60 .39972E-21
-7.58 .17278E-· U -· 8.60 .39858£-17 -9.62 .32916E-21
-7.60 . 148071 --13 -8.62 .33477E-17 -9.64 .27094£-21
-7 .62 .12684E-13 -8.64 .28107£-17 -9.66 .22293£-21
-7.64 .1086IE--13 -8 .66 .23588E-!7 -9.M .18336E---21
-7.66 .92967E-14 -8.68 .19788E-17 -9.70 .15075E-21
-7 .68 .79544E-14 -8.70 .16594E-17 -9.72 .12389E-2.i
-7.70 .68033E-14 -8.72 .13910E-17 -9.74 .IOliSP.-21
-7;72 .58165E-14 -8.74 .11656E-17 -9.76 .83578E-22
-7.74 .49708E-14 -8.76 .97625£-18 -9.78 .68605E-22
-7 .76 .42465E-14 ·-8.78 .81737£-18 -9.80 .56293E-22
-7.78 .36262E-14 -8 .80 .68408E-18 -9.82 .46!72E-22
-7.80 .30954E-14 -8.82 .57230E-18 -9.84 .37855E-22
-7.82 .26412£-14 -8.84 .47859E-18 -9.86 .31025E-22
-7.84 .22527E--14 -8.86 .40U07E-18 -9.88 .25416E-22
-7.86 ,19207E-14 -8.88 .33'130E-18 -9.90 .20814E-22
 429

u fl(uJ u <l>(u} II fl(u)

-9.92 .17038£- 22 ... 2.32 IQ-1 - 5.20 J0-7

-9.94 .J3941E- 22 -3.09 JQ- • -5.61 to-•
-9.96 .11403£- 22 -3.70 J0-4 - 6.00 to-•
-9:98 .93233E-23 -4.26 I0-1 - 6.36 J0-10
; 10.00 ' .76199E- 2) '. -4.75 JQ-• - 6.71 to-u
. .• ;_
~1.28 zo-•
 APPENDIX B
Partial Safety Factors for RCC
Members

4·0
F• 2SO
-r0 ,. 1·20
3·5 /L
- - Ln:4·0kN/m2
---Ln:J·OkN/m2
/ ,rL
I ,'\
J 0 - - Ln= 2· 5kN/m2 I ,'
I
/ I
I
~ 2 s I I
...
0
I
I

-v I
0 I
,... I ,
I

....'t 2 0 I I
I
0
"' I I
I

0 I I

-e I· 5 I

& I I
I

I I

1 0
rR

0 5

00 2 J 4 5 6
n
FIG. 81 Optimal values of partial safer.y factors for: RCC slabs in
flexure under load D + Lm for steel grade Fe 260
 35 F• 415
r0 .1 zo
-Lna40kN/m2
J 0 - - - - Lj, :J ·OkN/m2
-·-·-Ln:2· SkN/m2

.
11125
0
:vc
....

1 1Q

05

l
' 5 6

FIG. 12 Optimal values of partial safety factors tor ·RCC slabs In

flexure' under load D + L".a for steel grade Fe .416

1.ln:2·5kNlm2
J 2. l n :3·0kN/m2
3. Ln : 4·0kN/m 2

- - - F• 415
---•.Ft>250
0.~--~~----~--~~----~----~--~
2·0 2·~ J·O 3:~ 4o 4·5 s·o
(J . '·
Fl(l. 13 Qptlmal v~luea of pa.rtial safetr facrore ~r, AC~ belllns in
lfl..r under load D + Lm • · ·
 433
TABLE 81 Optimal partial safety factors for loads for columns
YD = 1.2 For Comp. YR = 0.725
For Tens. Yx = 0.80

Load Combination D + Lm D -1- Wm D + Lapt + Wm

Mix Failure Po YL Yw 'IL Yw

Case (i)
L0 = 3 Design Comp. 3.0 1.2 1.2 0.30 1.1
kN/m 1 3.5 1.4 I.S 0.27 1.5
4.0 1.9 2.0 0.24 2.0
Tension 3.0 1.3 I.S 0.25 1.4
3.5 1.8 2.0 0.24 1.8
4.0 2.4 2 ..5 0.23 2.3
Nominal Comp. 3.0 1.4 1.6 0.20 1.6
3.5 2.5 2.6 0.17 2.6
4.0 4.0 3.6 0.15 3.6
Tension 3.0 1.4 1.5 0.22 l.S
3.5 2.3 2.3 0.20 2.3
4.0 3.5 3.2 0.18 3.2
Case (ii)
L0 = 4 Design Comp. 3.0 0.8 1.2 0.20 I. I
kN/rn1 3.5 1.1 I.S 0.1~ l.S
4.0 1.5 2.0 0.1 2.0
Tension 3.0 1.0 1.5 0.18 1.4
3.5 1.4 2.0 0.17 l.S
4.0 1.8 2.5 0.16 2 ..3
Nominal Comp. 3.0 1.1 1.6 0.15 1.6
35 2.1 2.6 0, 13 2.5
4.0 3.9 3.6 0.17 3.5
Tension 3.0 1.0 1.S 0.17 1.3
3.5 1.7 2.3 0.15 2.3
4.0 2.7 3.2 0.12 3.2
436

genes:ations of 301, 309 strength 381,393

stochastically 302 Fatigue reliability of details I joints in
bridges 382, 396
Eigenvalues 216-221 highway bridges 387
Elastic offshore structuia 404
analysis 3 railway bridges 400, 407
behaviour 2 ship structures 404, 413
Equival~nonnal 211 First order reliability method (FORM)
Erlang distribution 76 340
Events First order second-moment (FOSM)
certain 24 method 182, 188, 316
collectively exhaustive 27,41 Fracture mechanics 414
complement of 24 Fracture toughness 418
compound 23 Frames
disjoint 26 reliability analysis of 337, 362
intez:section 26 Frechet distribution 134
mutually exclusive 26, 29,41 Frequency distnoution
null 24 relative 14, 15,29
random relationships among 26 cumulative 14, 16
simple 23, 43
Wlion 26 Gamma distribution 75-77, 87, 109, 125,
Expectation 60 128, 130, 159, 164
algebra of 62 Gamma filnction
conditional 62 incomplete 75, 89
of a function 62 Gaussian distribution, (see normal
Expected value 175, 193,217,221 distribution)
Exponential distribution 88, 130, 150, Goodness-of-fit tests 93
161 Gumbel distribution 80, 134
Extremal distributions 79
Rayleigh 88 Hasofer Lind method 190
Type 1 (smallest) 81, 28, 87 Hazard filnction 144
Type 1 (largest) 55, 80, 81, 87, 128, High yield strength
132,134,136,139,162,209,212- defonned burs 98, 108
216,246 Histogram 14
Type 2 (largest) 83, 84, 87, 88, 134,
150,162,204,206 Inelastic 2
Type 3 (smallest) 84, 85, 86, 162 Importance sampling method (ISM) 342
Influence area 122, 123, 125
Factor of safety 1, 107, 108 (.feesafety Influence surface 122
factor) Inverse transfOIIWltion technique 159,
Failure 161
filnction 144, 180, 191
modes 172,293,301 Joint probability distribution
correlated 172 cumulative distribution function 157
probability, (see probability of failure) probability density function 157
rate 145
surface 180, 181, 183, 189 Kolmogorov-Smirnov test 89, 93
path 294 Kurtosis
point 199 coefficient of 61
Fatigue 378
 437
Lagrange multiplier method 192 Median 12
Lifetime 118, 125 Mode
design wind speed 137 of a random variable 12
maximum live load 117 Momarta
maximum wind speed 139 mr:thods of 109
Limit state ofjointly distributed variables 64
aerviceabilit,. 4 of nudom variables 60
ultimate 4, 13? Monte Carlo
Limit state design method 158, 164-167, 175 (HI also
probability based 240 tedmique)
Live load sample aize 165-167
~y 116,127,129,131 aimulati'on 139 ·
Load ~ 139; 158, 159, 167, 170,
arbitrary point in-time 113, 125, 127, 116, 244, 324 •
132
dead 1-3,111-114 NBC (Canada) format 240
earthquake 2 Normal diatribution 70, 87, 93, 109, 149,
factored 4 162,221
floor 10,112,132 equivalent 199,247
equivalent unifonnaly distributed standard 70
117, 119, 123, 125, 129
extraordinary 113-115, 132 Palmgrcn-Miner's cumulative daJna&e
lifetime maximum 113,117,119, 383
126,130,210,241,243 Paris crack growth law 415
live 1,2, 10,111-119,163,199 Pearson Type 3 distribution 76
modeling Piecewise linear elastic plastic {Pwi.Ep)
maximum total 114, 130 293,294i316
susUdned 113-118,120,126,128 Poisson distribution 88
service 2 Poi990n p-ocess 125, 126
transient 129, 130 Probability
lifetime maximum 130 axioms 42, 43
ultimate 2, 3 conditional 32, 38
~d 1-3, 132, 138 joint 35,46,287,288
Load factor 3, 111,241,319,321 mass function 43
combined 232 notional 179,
Lognonnal distribution 72, 87, 93, 109, tree diagram 36, 37
125,127,139,149,163,203 Probability density function
fonnat 394, 404 conditional 51, 57
variate 74 joint 46, 47, 50
LRFD (load and resistance factor design) ~ 48,50,51
223,241,242,246,263,264,393 Probability offallilre 7, 144, 146, 148,
157,167,269,270
Material reduction factor 239, 240,244 conditiooaJ 157, 158
MaXimum likelihood ofmaterial lOS
method of 109 Probabili~y of lllllVjval 7, 143, 269
Mean
functions of variables 65 Quality control 91, 93, 143
of a random variable 60
sample 17, 18 Random number generation 159
value method 185 composition method 160
438

from beta distribution 163 central 5, 226

from exponential distribution 161 charac~stic 226,231
from gamma distribution 164 partial 226, -229, 239, 245, 257-263
from lognonnal distribution I63 optimal 252,254,257,258,259,260,
from nonnal distribution 162 346-349
from Type I extremal (largest) stress range 395
distribution 161, 162 Safety margin I48, I72, I75, I83, 189,
from Type I extremal (smallest) 2Q2,282
distribution I61 Sample size 14, 165-167
from Type 2 extremal (largest) Sample space 23
distribution I6I, I62 conditional 24
from Type 3 extremal (smallest) continuous 24
distribution I61, 162 discrete 24
from uniform distribution 161 reduced 33
from Weibull distribution 16I two dimensional 30
inverse transformation tecbniquc 159 Second order reliability method (SORM)
pseudo random numbers 159 340
Random process 22 Series system 268-270
Random variable 43 Simulation 302 (.Jee also Monte Carlo
continuous 44,45,5I,60 simulation
discrete 43,48 Skewness
functiollll of S1 coeffiCient of I6, 61
independent 51 S-N cmve approach 382
jointly distributed 46, 49 Standard bete variate n
Range 13 Standard deviation 13, 61
Rayleigh distribution 89 Standard nonnal
Reliability 7, 143, I44, I48 density function 70
analysis I46, I77, 213,226,268 tables 340-345
analysis ofRCC frames 315, 3I9, variable 70
322,334 Standard normal variate 70
analysis of trusses 371 Standard unifonn distribution 69
based design 226, 242, 262 S~cal independence 36
factor 413 Steel properties
index I49, 183 elasticity, modulus of 97, 99, 104
targd 225,245,253,256,262 statistics of 97
Resistance factor 240 ultimate strength 97
Response surface method 358 yield strength 94-97, 104
Retumperiod I33-135 Yowtg's modulus 99
Risk 137, 138 Stochastic process 23
Rotation failure mode 316, 334 · Stocllastic variable 23
Rule of multiplication 35 Stress
. permissible I, 2
Saferegion I81 ultimate I, 2
Safety checking format 224, 239, 246, intensity factor
256 Stress range 379
Safety checking methods constant amplitude 383, 393
Levell I79, 180, 2S3 equivalent constant amplitude 384
Levell 179, 180,225,226,244,2S3 design 385
Levei3 I79 long term design 405
Safety factor S1ructural design 223, 225