A Textbook On Structural Reliability: Arunasis Chakraborty

Arunasis Chakraborty
A Textbook on Structural
Reliability
Analysis and Design
February 10, 2019
Springer
Dedicated to
Foreword
The foreword covers introductory remarks preceding the text of a book that are
written by a person other than the author or editor of the book. If applicable, the
foreword precedes the preface which is written by the author or editor of the book.
Place, month year Firstname Surname
vii
Preface
A preface is a book’s preliminary statement, usually written by the author or editor

of a work, which states its origin, scope, purpose, plan, and intended audience, and
which sometimes includes afterthoughts and acknowledgments of assistance.
When written by a person other than the author, it is called a foreword. The
preface or foreword is distinct from the introduction, which deals with the subject
of the work.
Customarily acknowledgments are included as last part of the preface.
Place(s),
month year Arunasis Chakraborty
ix
Acknowledgements
If you prefer to set your acknowledgement section as a separate chapter instead of

including it as last part of your preface.
xi
Contents
1 Introduction to Reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Section Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Section Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2.1 Subsection Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Section Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2 Theory of Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1 Section Algebra of Variance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Section Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Section Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Level 2 Reliability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3.1 Lecture 01: Concept of limit state and Reliability . . . . . . . . . . . . . . . . 13
3.2 Lecture 02: First Order Second Moment Method (FOSM) . . . . . . . . . 15
3.3 Lecture 03: First Order Reliability Method (FORM) . . . . . . . . . . . . . 19
3.4 Lecture 04: FORM for Correlated Non-Normal Random Variables . 22
3.5 Lecture 05: FORM Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.6 Lecture 06: Modified Rackwitz Algorithm and Nataf Model . . . . . . . 32
3.7 Lecture 07: Second Order Reliability Method (SORM) . . . . . . . . . . . 36
4 Simulation Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
5 Meta-Modeling Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
6 Stochastic Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
xiii
xiv Contents
7 Application on Design Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

7.1 Reliability Based Design Optimization . . . . . . . . . . . . . . . . . . . . . . . . . 45
A App 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
A.1 Section Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
A.1.1 Subsection Heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
Acronyms
Lists of abbreviations, symbols and the like are easily formatted with the help of the
Springer-enhanced description environment.
ABC Spelled-out abbreviation and definition
BABI Spelled-out abbreviation and definition
CABR Spelled-out abbreviation and definition
xv
Chapter 1
Introduction to Reliability
Abstract Each chapter should be preceded by an abstract (10–15 lines long) that
summarizes the content. The abstract will appear online at www.SpringerLink.
com and be available with unrestricted access. This allows unregistered users to read
the abstract as a teaser for the complete chapter. As a general rule the abstracts will
not appear in the printed version of your book unless it is the style of your particular
book or that of the series to which your book belongs.
Please use the ’starred’ version of the new Springer abstract command for
typesetting the text of the online abstracts (cf. source file of this chapter template
abstract) and include them with the source files of your manuscript. Use the
plain abstract command if the abstract is also to appear in the printed version of
the book.
1.1 Section Heading
Use the template chapter.tex together with the Springer document class SVMono
(monograph-type books) or SVMult (edited books) to style the various elements of
your chapter content in the Springer layout.
1.2 Section Heading
Instead of simply listing headings of different levels we recommend to let every

heading be followed by at least a short passage of text. Furtheron please use the
LATEX automatism for all your cross-references and citations.
Please note that the first line of text that follows a heading is not indented,
whereas the first lines of all subsequent paragraphs are.
Use the standard equation environment to typeset your equations, e.g.
1
2 1 Introduction to Reliability
a×b = c , (1.1)
however, for multiline equations we recommend to use the eqnarray environ-

ment1 .
a×b = c
a·b = c (1.2)
1.2.1 Subsection Heading

LATEX automatism for all your cross-references and citations as has already been
described in Sect. 3.2.
Please do not use quotation marks when quoting texts! Simply use the quotation envi-
ronment – it will automatically render Springer’s preferred layout.
1.2.1.1 Subsubsection Heading

described in Sect. 2.2.1, see also Fig. 2.12
Paragraph Heading

1 In physics texts please activate the class option vecphys to depict your vectors in boldface-italic
type - as is customary for a wide range of physical subjects.
2 If you copy text passages, figures, or tables from other works, you must obtain permission from
the copyright holder (usually the original publisher). Please enclose the signed permission with the
manucript. The sources must be acknowledged either in the captions, as footnotes or in a separate
section of the book.
1.2 Section Heading 3
For typesetting numbered lists we recommend to use the enumerate environ-

ment – it will automatically render Springer’s preferred layout.
1. Livelihood and survival mobility are oftentimes coutcomes of uneven socioeco-
nomic development.
a. Livelihood and survival mobility are oftentimes coutcomes of uneven socioe-
conomic development.
b. Livelihood and survival mobility are oftentimes coutcomes of uneven socioe-
nomic development.
Subparagraph Heading
In order to avoid simply listing headings of different levels we recommend to let
every heading be followed by at least a short passage of text. Use the LATEX au-
tomatism for all your cross-references and citations as has already been described
in Sect. 3.2, see also Fig. 2.2.
For unnumbered list we recommend to use the itemize environment – it will
automatically render Springer’s preferred layout.
• Livelihood and survival mobility are oftentimes coutcomes of uneven socioeco-
nomic development, cf. Table 2.1.
– Livelihood and survival mobility are oftentimes coutcomes of uneven socioe-
nomic development.
Fig. 1.1 If the width of the

figure is less than 7.8 cm use
the sidecapion command
to flush the caption on the left
side of the page. If the figure
is positioned at the top of the
page, align the sidecaption
with the top of the figure – to
achieve this you simply need
to use the optional argument
[t] with the sidecaption
command
Fig. 1.2 Please write your

figure caption here
Run-in Heading Boldface Version Use the LATEX automatism for all your cross-
references and citations as has already been described in Sect. 3.2.
Run-in Heading Italic Version Use the LATEX automatism for all your cross-refer-
ences and citations as has already been described in Sect. 3.2.
Table 1.1 Please write your table caption here

Classes Subclass Length Action Mechanism
Translation mRNAa 22 (19–25) Translation repression, mRNA cleavage

Translation mRNA cleavage 21 mRNA cleavage
Translation mRNA 21–22 mRNA cleavage
Translation mRNA 24–26 Histone and DNA Modification
a Table foot note (with superscript)
1.3 Section Heading

If you want to list definitions or the like we recommend to use the Springer-
enhanced description environment – it will automatically render Springer’s
preferred layout.
Type 1 That addresses central themes pertainng to migration, health, and disease.
In Sect. 3.1, Wilson discusses the role of human migration in infectious
disease distributions and patterns.
In Sect. 2.2.1, Wilson discusses the role of human migration in infectious

tomatism for all your cross-references and citations citations as has already been
If you want to emphasize complete paragraphs of texts we recommend to use

the newly defined Springer class option graybox and the newly defined envi-
ronment svgraybox. This will produce a 15 percent screened box ’behind’
your text.
If you want to emphasize complete paragraphs of texts we recommend to
use the newly defined Springer class option and environment svgraybox.
This will produce a 15 percent screened box ’behind’ your text.

Theorem 1.1. Theorem text goes here.
Definition 1.1. Definition text goes here.
Proof. Proof text goes here. t

u
Paragraph Heading

Note that the first line of text that follows a heading is not indented, whereas the
first lines of all subsequent paragraphs are.

u
Acknowledgements If you want to include acknowledgments of assistance and the like at the end
of an individual chapter please use the acknowledgement environment – it will automatically
render Springer’s preferred layout.
Appendix
When placed at the end of a chapter or contribution (as opposed to at the end of
the book), the numbering of tables, figures, and equations in the appendix section
continues on from that in the main text. Hence please do not use the appendix
command when writing an appendix at the end of your chapter or contribution.
If there is only one the appendix is designated “Appendix”, or “Appendix 1”, or
“Appendix 2”, etc. if there is more than one.
a×b = c (1.3)
Problems
1.1. A given problem or Excercise is described here. The problem is described here.
The problem is described here.
1.2. Problem Heading

(a) The first part of the problem is described here.
(b) The second part of the problem is described here.
Chapter 2
Theory of Probability
2.1 Section Algebra of Variance
Use the template chapter.tex together with the Springer document class SVMono
(monograph-type books) or SVMult (edited books) to style the various elements of
your chapter content in the Springer layout.
2.2 Section Heading

Use the standard equation environment to typeset your equations, e.g.
a×b = c , (2.1)
however, for multiline equations we recommend to use the eqnarray environ-

ment1 .
a×b = c
a·b = c (2.2)
1 In physics texts please activate the class option vecphys to depict your vectors in boldface-italic
type - as is customary for a wide range of physical subjects.
7
8 2 Theory of Probability

Please do not use quotation marks when quoting texts! Simply use the quotation envi-
ronment – it will automatically render Springer’s preferred layout.

described in Sect. 2.2.1, see also Fig. 2.12
Paragraph Heading

Fig. 2.1 If the width of the

figure is less than 7.8 cm use
the sidecapion command
to flush the caption on the left
side of the page. If the figure
is positioned at the top of the
page, align the sidecaption
with the top of the figure – to
achieve this you simply need
to use the optional argument
[t] with the sidecaption
command
2 If you copy text passages, figures, or tables from other works, you must obtain permission from
the copyright holder (usually the original publisher). Please enclose the signed permission with the
manucript. The sources must be acknowledged either in the captions, as footnotes or in a separate
section of the book.
For typesetting numbered lists we recommend to use the enumerate environ-

ment – it will automatically render Springer’s preferred layout.
nomic development.
a. Livelihood and survival mobility are oftentimes coutcomes of uneven socioe-
b. Livelihood and survival mobility are oftentimes coutcomes of uneven socioe-
nomic development.
Subparagraph Heading
tomatism for all your cross-references and citations as has already been described
in Sect. 3.2, see also Fig. 2.2.
For unnumbered list we recommend to use the itemize environment – it will
automatically render Springer’s preferred layout.
nomic development, cf. Table 2.1.
nomic development.
Run-in Heading Boldface Version Use the LATEX automatism for all your cross-
references and citations as has already been described in Sect. 3.2.
Run-in Heading Italic Version Use the LATEX automatism for all your cross-refer-
ences and citations as has already been described in Sect. 3.2.
2.3 Section Heading

Fig. 2.2 Please write your

figure caption here
Table 2.1 Please write your table caption here

If you want to list definitions or the like we recommend to use the Springer-
enhanced description environment – it will automatically render Springer’s
preferred layout.
In Sect. 3.1, Wilson discusses the role of human migration in infectious
In Sect. 2.2.1, Wilson discusses the role of human migration in infectious

tomatism for all your cross-references and citations citations as has already been
If you want to emphasize complete paragraphs of texts we recommend to use

the newly defined Springer class option graybox and the newly defined envi-
ronment svgraybox. This will produce a 15 percent screened box ’behind’
your text.
If you want to emphasize complete paragraphs of texts we recommend to
use the newly defined Springer class option and environment svgraybox.
This will produce a 15 percent screened box ’behind’ your text.


u
Paragraph Heading

Note that the first line of text that follows a heading is not indented, whereas the
first lines of all subsequent paragraphs are.

u
Problems
2.1. A given problem or Excercise is described here. The problem is described here.
The problem is described here.
2.2. Problem Heading

(a) The first part of the problem is described here.
(b) The second part of the problem is described here.
Chapter 3
Level 2 Reliability Analysis
Abstract Each chapter should be preceded by an abstract (10–15 lines long) that
summarizes the content. The abstract will appear online at www.SpringerLink.
com and be available with unrestricted access. This allows unregistered users to read
the abstract as a teaser for the complete 14 chapter. As a general rule the abstracts
will not appear in the printed version of your book unless it is the style of your
particular book or that of the series to which your book belongs.
Please use the ’starred’ version of the new Springer abstract command for
typesetting the text of the online abstracts (cf. source file of this chapter template
abstract) and include them with the source files of 14 your manuscript. Use the
plain abstract command if the abstract is also to appear in the printed version of
the book.
3.1 Lecture 01: Concept of limit state and Reliability
In this section, let us first review the design of a cantilever beam as shown in figure
3.1.1. The basic idea behind this design is to ensure safety, serviceability and econ-
omy. These are achieved in an iterative procedure by selecting material and cross
section of the beam so that it can withstand a particular value of the applied force
called design load. In this design process, material strength is factored down while
the load is increased by a factor proposed in standard code of practice.
Figure 3.1.1 shows the cantilever beam subjected to a point load at free end. The
beam is made of steel and is supposed to be designed against serviceability limit
state. The governing equation of this limit state is given below
L PL3
LS = − (3.1)
325 3EI
It can be noticed that allowable defection of the free end is L/325 and is termed
as ’Capacity’ of the structure. The second term in Eq. 3.1 represents the applied
deflection due to the point load P at the free end i.e. ’Demand’. Therefore, any limit
13
14 3 Level 2 Reliability Analysis
state equation can be represented in the form
g(x) = Capacity − Demand = C − D = 0 (3.2)
In the above equation X = {C D} represents the variable to be considered in the

design. In general, there can be 0 n0 such variables in the design. Out of them, p0 may
be associated with Capacity i.e. C = {X1 X2 X3 ....X p } and 0 q0 may be associated with
Demand D = {X p+1 X p+2 X p+3 ....X p+q } leading to p + q = n. If the variables are
deterministic, the design may be iteratively carried out to reach an optimal value of
them which, in turn, will ensure all three fundamental requirements stated earlier.
The problem becomes more complex when the design variables are random in na-
ture i.e. only defined by their pdf. Let us consider only two random variables i.e.
C and D which are independent. Figure 3.1.2 shows the limit state that divides the
design space into safe and unsafe regions.
From this Figure 3.1.2 one can conclude that there may be infinite combinations of
C and D that falls on the limit state i.e. satisfy the design. So, unlike deterministic
design, it is rather tough job for the designer to choose the optimal value. An intu-
itive solution to this problem is to select the point that has least probability of failure.
This gives rise to an important question, how to assess the point on the limit state in
the first place? To answer this question let us consider Figure 3.1.3 that shows the
pdf of these random variables.
From this Figure 3.1.3, one can easily conclude that the common area under these
two pdfs is the probability of failure. The reason behind this claim is that c is always
less than d in this region and hence, ensuring failure. Let us evaluate the common
area in this case i.e. p f .
Figure 3.1.4 shows the enlarged version of this area. Note that the probability of 0 D0
assuming a value of 0 d 0 is

dc dc
P c− <C < c+ = fC (c)dc = A1 (3.3)
2 2
while, probability of c > d is given by

Z ∞
P(D > c) = fD (d)dd = A2 (3.4)
c
Therefore, the reliability of design when 0C0 attains a value of 0 c0

Z ∞
dR = A1 × A2 = fC (c)dc fD (d)dd (3.5)
c
It may be noticed that Eq. 3.5 is applicable when C and D are independent which is
true in most of the cases. On integrating Eq. 3.5, one can evaluate the reliability of
the design
Z Z ∞ Z ∞
R = dR = fC (c) fD (d)dd dc −∞ < x < ∞ (3.6)
−∞ c
3.2 Lecture 02: First Order Second Moment Method (FOSM) 15
Since, p f and R are complementary to each other, the probability of failure can be
easily evaluated as
Z ∞ Z ∞
pf = 1−R = fC (c)FD (d)dc = 1 − fD (d)FC (c)dc (3.7)
−∞ −∞
Case 1: C and D are independent normal
Let us consider a special case when C and D are both normal and independent.
The probability of failure is given by
p f = P[(C − D) < 0] = P(G < 0) = FG (0) (3.8)
As, C and D are normal, g is also normal. So, by using algebra of variance as dis-
cussed in Chapter 2, one can evaluate the mean and standard deviation of 0 g0 .
Figure 3.1.5 shows the pdf of g. From this figure one can easily evaluate the proba-
bility of failure (i.e. area on the left hand side of zero) which is given by

0 − µg µg
pf = Φ =Φ − (3.9)
σg σg
where, µg = µc − µD and σg = (σC2 + σD2 )1/2 .
Case 2: C and D are independent Lognormal
Another illustrative case is considered where both, C and D are independent and
following lognormal distribution. The p f is evaluated by modifying the limit state
which is given by
C
pf = P <1 (3.10)
D
As, C and D are lognormal, g is also lognormal distributed. Thus, following the
definition of p f given in Eq. 3.1.10, one can show

ln(C̃/D̃)
pf = Φ 2 + σ 2 )1/2
(3.11)
(σlnC lnD
3.2 Lecture 02: First Order Second Moment Method (FOSM)
In the previous section, we derived an expression for p f when C and D are both
normal and independent which is given by

µg
pf = Φ − = Φ[−β ] (3.12)
σg
In the above expression 0 β 0 is measure of the reliability for the limit state g(X). It
was proposed by Cornell and hence, is termed as Cornell’s Reliability Index. Figure
3.2.1 below shows the definition of Cornell’s Reliability Index graphically.
It can be observed that p f corresponds to the area of the left side of origin i.e. P[g ≤
0] while the mean (µg ) of the limit state lies at a distance of β σg from origin. On
further investigation of Eq. 3.12, one can easily conclude that the random variables
are characterized by their first two moments only i.e. µg and σg . This is achieved
by first order approximation of the non-linear function g(X) and hence, the method
is termed as First Order Second Moment (FOSM) method. In the above definition
of Cornell’s Reliability Index, the limit state is assumed to be linear combination of
the random variables. Let us extend this model to consider any non-linear function
of basic random variable i.e. g(X) = 0. The first two moments for this function
may be estimated approximately by series expression of g(X) about any point X =
{X1 X2 X3 ....X p }. With this in view, let us expand g(X) = 0 using Taylor’s series
expression at around X ∗ which is given by
(Xi − Xi∗ )2
2
∗ ∗ ∗ n ∂ g ∗ n ∂ g
LS = g(X1 , X2 , ..., Xn ) = Σi=1 (Xi −Xi ) +Σ i=1 +...
∂ Xi X ∗ ∂ Xi2 X ∗ 2
(3.13)
In the above expression, .|X ∗ denotes the value of the respective function (.) at X ∗ .
Let us assume,X ∗ = µ and evaluate the first two moments. Considering first two
terms in the Taylor’s series expression and taking expectation on both sides one can
prove that
µg = E[g(X)] ' g(µ1 , µ2 , ..., µn ) + 0 (3.14)

∂ g ∂ g
Var(g) = σg2 ' Σi=1 n
Σ nj=1 Cov(Xi , X j ) (3.15)
∂ Xi X ∗
∂ X j X ∗
Using Eq. 3.14 and Eq. 3.15 in Eq. 3.12, the probability of failure for any general-
ized g(X) can be evaluated. In this context, one can notice that the first two moments
of g(X) are evaluated at around mean value. Hence, the method is often referred as
Mean Value First Order Second Moment (MVFOSM) method.
Although, MVFOSM provides the mathematical framework to consider the non-
linearity of G(X), it has major drawbacks. These are listed as below
• Mean is considered to be the reference point. However, in most of the cases µ

does not necessarily lie over the limit state and hence, evaluation at this reference
point introduces significant error.
• Only first two terms are considered in Taylor’s series expression i.e. the limit
state is linearized at about mean of the random variables. However, in all practical
cases, limit states are highly non-linear. Hence, its linearization at around mean
point of basic random variable contributes huge error.
• This method provides completely different results for the same limit state but in
different formulation. Let us consider the limit states below
g(X) = C − D (3.16)
3.2 Lecture 02: First Order Second Moment Method (FOSM) 17
D
= 1− (3.17)
C
Although, both of them are different representation of the same limit state, the
probabilities of failure as per MVFOSM are completely different. This is due
’Lack of Invariance’.
Examples:
Ex 01. A steel cantilever beam of a rectangular cross section subjected to point load
at its free end is considered. The limit state of serviceability for maximum deflection
at the free end is defined by Eq. 3.1 where P, L, E and I are representing point load,
length, modulus of elasticity and cross sectional moment of inertia of the beam re-
spectively. p, L and E are random variables following normal distribution whereas I
is deterministic and equal to 3.4133108 mm4 . Two cases are considered where limit
state is defined by Eq. 3.16 and 3.17. The statistical properties are given as
S.No. RANDOM VARIABLE(X) µX σX

1 P(N) 6000 9000
2 L(mm) 2500 125
3 E(N/mm2 ) 2 × 105 1 × 104
Solu.: Case 1
L PL3 L PL3
g(X) = − 8
= −
325 3 × 3.4133 × 10 E 325 1.024 × 109 E
2500 60000 × 2500 3
g(µ) = − = 3.1146
325 3 × 3.4133 × 108 × 2 × 105
3 × 60000 × 25002

∂ g 1
= − = −2.4163 × 10−3
∂ L µL 325 3 × 3.4133 × 108 × 2 × 105
25003

∂ g
=− = −2.2888 × 10−5
∂ P µP
3 × 3.4133 × 108 × 2 × 105
3 × 60000 × 25002

∂ g
= = −7.6294 × 10−5
∂ E µE 3 × 3.4133 × 108 × 22 × 1010

σg = [(−7.6294 × 10−5 × 9000)2 + (−2.4163 × 10−3 × 125)2

+ (2.2888 × 10−5 × 1 × 104 )2 ]0 .5 = 0.7843
3.1146
β= = 3.9712
0.7843
p f = Φ(−3.9712) = 3.5756 × 10−5
Case 2
PL3
3×3.4133×108 E PL2
g(X) = 1 − L
= 1−
325
3.1508 × 106 E
60000 × 25002
g(µ) = 1 − = 0.4049
3.1508 × 106 × 2 × 105
25002

∂ g
= − = −9.9181 × 10−6
∂ P µP 3.1508 × 106 × 2 × 105

∂ g 2 × 60000 × 2500
=− = −4.7607 × 10−4
∂ L µL 3.1508 × 106 × 2 × 105
60000 × 25002

∂ g
= − = 2.9754 × 10−6
∂ E µE 3.1508 × 106 × 22 × 1010
σg = [(−9.9181 × 10−6 × 9000)2 + (−4.7607 × 10−4 × 125)2
+ (2.9754 × 10−6 × 1 × 104 )2 ]0 .5 = 0.1113
0.4049
β= = 3.6379
0.1113
p f = Φ(−3.6379) = 1.3744 × 10−4
In this lecture, we have learned the definition of Reliability Index as proposed by

Cornell for linear combination of random variables. This is further extended to MV-
FOSM to consider any non-linear limit state. However, before we conclude this lec-
ture, let us revisit Cornell’s definition of Reliability Index for the limit state given
by Eq. 3.12. Let us consider the following transformation
(C − µC )
u1 = (3.18)
σC
(D − µD )
u2 = (3.19)
σD
Using these two transformation in Eq. 3.12 one can show that the limit state becomes
u1 σC − u2 σD + µC − µD = 0 (3.20)
In this context, it may be noted that the random variables C and D in Eq. 3.1 are in
their original space while u1 and u2 are in Standard Normal Space. These different
versions of the same limit state in two different spaces are shown in Figure 3.2.2.
Note that P and R are the intercepts of the limit state on two orthogonal axes u1 and
u2 respectively. Using trigonometry,one can find out the normal distance of the limit
state [i.e.OQ] from the origin in standard normal space which is given by
3.3 Lecture 03: First Order Reliability Method (FORM) 19
(µC − µD )
OQ = =β (3.21)
(σC2 + σD2 )1/2
Above expression for the normal distance is the Reliability Index as proposed by
Cornell. Hence, we get a more robust definition of Reliability Index.
Definition: Reliability Index is the shortest distance of the linearized limit state
from the origin in the Standard Normal Space.
The above definition of reliability index is commonly known as Hasofer-Lind Reli-
ability Index who developed this model.
3.3 Lecture 03: First Order Reliability Method (FORM)
In the last section, we derived expression for perpendicular distance of the limit
state from the origin in Standard Normal Space i.e. β . However, the assumption
behind this derivation is that the limit state has linear combination of basic random
variables. So, the question remains what happens when the limit state is non-linear
function of basic random variables. Figure 3.3.1 shows the non-linear limit state
g(X) = 0 that divides the entire standard normal space into safe and unsafe do-
mains. Let u∗ be the failure point that lies on the actual limit state. Now imagine
another hypothetical linear limit state tangential to g(X) at u∗ . Since, u∗ is common
in both the cases (i.e. optimal distance for both linear and non-linear limit state), the
reliability obtained from both Cornell’s model and Hasofer-Lind’s model are same.
This analogy tells us a more general definition of Hasofer-Lind Reliability Index for
any non-linear limit state function.
Definition: Reliability Index is the optimal distance of the limit state from the
origin in the Standard Normal Space
With this definition in view, let us derive an expression for reliability index for any
non-linear limit state function g(X) = 0. The objective is to find the optimal distance
of the transformed limit state in the Standard Normal Space. For this purpose, let us
apply the following transformation
X = [σ ]{U} + {µ} (3.22)
where, [σ ] is the diagonal matrix whose leading diagonal consists of the standard
deviation of basic random variables and {µ} is the mean vector. Using variance
algebra one can show that u has zero mean and unit standard deviation. Using Eq.
3.22 in Eq. 3.2, one can transform the original limit state g(X) into Standard Normal
Space i.e.
g(U) = 0 (3.23)
The objective is to optimize dn and find u∗ so that the constraint is satisfied i.e.
g(u∗ ) = 0. In this context, ||.|| represents Euclidean Norm of the vector. To solve this
optimization problem, Lagrange Multiplier technique is adopted. The Lagrangian in

this case is given by
L = dn + λ g(U) (3.24)
In the above equation, λ is the Lagrange multiplier. From the above equation one
can notice that the Lagrangian has (n + 1) unknowns. To solve these unknowns,
following conditions are satisfied
∂L ui ∂g
= +λ =0 i = 1, 2, ..., n (3.25)
∂ ui kUk ∂ ui
∂L
= g(u! , u2 , ..., un ) = 0 (3.26)
∂λ
The solution of first n equations provides optimal point while the solution of (n +
1)th automatically satisfies the constraint condition. Using matrix notation, the first
n equations can be written as
U
+λG = 0 (3.27)
kUk
In the above equation, G represents the gradient vector { ∂∂ugi }. Using these equations,
the optimal solution can be obtained as
U ∗ = −λ dn G∗ (3.28)
Where, λ = kG∗ k−1 . Finally, pre-multiplying both sides of Eq. 3.12 by Gt and sub-
stituting λ , one can evaluate the optimal distance as
U ∗ G∗t
dn = β = − (3.29)
kG∗ k
Using Eq. 3.57 into Eq. 3.56 one can find MPP as
β ∗ G∗
U∗ = − = αi∗ β (3.30)
kG∗ k
Where, αi∗ is the direction cosine of the tangent on the limit state g(u) = 0 at u∗ and
is given by
∂g
∂ u∗i
αi∗ =− ∗ (3.31)
kG k
From the above model, one can notice that optimal distance in Eq. 3.57 needs prior
information about MPP i.e. u∗ and needs to evaluate β first. To solve this, an iterative
algorithm proposed by Rackwitz is generally employed. The algorithm is as follows
• Transform the limit state function g(X) into standard normal space i.e. g(U) = 0.
• Evaluate the derivative of the transformed limit state with respect to standard
normal variables (i.e. G).
3.3 Lecture 03: First Order Reliability Method (FORM) 21
• Assume initial β value (generally, considered in range of 35) and choose direc-
tion cosine in such a way that the summation of square of direction cosine is one
2
(Σ αi∗ = 1).
• Write the limit state g(U)n terms of β and αi∗ , substituting the values using Eq.
3.58.
• Solve the expression for new β value. It can be shown as
g(βk , α1∗k , α2∗k , ..., αn∗k , ) = 0 (3.32)
• New β is used to evaluate kG∗ k by

s 2
n
∗ ∂g
kG k = ∑ (3.33)
i=1 ∂ u∗i
• New direction cosines αi∗k+1 are calculated employing Eq. 3.59.

• Now, check the difference in two successive iteration values of β . Stop the itera-
tion if it is less than stopping criteria otherwise go back to Step 5.
• Design point or MPP is evaluated by transforming the U ∗ value as per Eq. 3.58
back to original space.
Alternate Proof of FORM
An alternative proof of the optimal distance β can be obtained using Taylor’s series
expansion also. Let us expand the limit state in the standard normal space which is
given by
1 n ∂ g1 k
∞
∗ k
g(U) = ∑ ∑ ∗ (u i − u i ) (3.34)
k=0 k! i=1 ∂ ui
Considering only first order terms, it can be shown that the mean and standard de-
viation of the limit state function g(u) are
k
∂g t
µg = − ∑ u∗i = −U ∗ G∗ (3.35)
i=1 ∂ u∗i
k 2 1/2
∂g
σg = ∑ = kG∗ k (3.36)
i=1 ∂ u∗i
As only first order terms are used in Taylor’s series [i.e. linearization of g(u) = 0]
the reliability index β may be obtained using Cornell’s definition which is given by
µg U ∗ G∗t
β= =− (3.37)
σg kG∗ k
3.4 Lecture 04: FORM for Correlated Non-Normal Random

Variables
In this section, we will revisit Hasofer-Lind’s Reliability Index for more general
cases. Earlier we derived for optimal distance in Standard Normal Space (i.e. β )
for a limit state whose basic random variables are uncorrelated and normal. The
design point is obtained using Lagrange Multiplier technique. In this context, it
should be noted that any other optimization algorithm can be used to find out the
global minimum distance in Standard Normal Space. Rackwitz algorithm is just an
efficient gradient based technique to find global minimum in an iterative manner.
However, the model presented in previous section does not answer the following
questions
• How to find out global minima when the basic random variables are non-normal?
• How to find out global minima when the basic random variables are correlated?
Let us consider the first question and solve for global minima when the random vari-
ables follow any arbitrary pdf. To evaluate the optimization problem, an equivalent
normal distribution is used to replace the arbitrary pdf in each successive design
point. Let X ∗ be the design point in Rackwitz’s algorithm. The random variable X
follows an arbitrary pdf fX (x) whose CDF is denoted by FX (x). To fit an equivalent
normal distribution at x∗ , following points are kept in mind
• pdf at x∗ obtained from the original distribution and the equivalent normal distri-
bution are same
• CDF of the two distributions at x∗ are equal.
Using first condition, one gets
∗
Xi − µieq

1
fxi (Xi∗ ) = φ (3.38)
σieq σieq
Xi∗ − µieq

Fxi (Xi∗ ) = Φ (3.39)
σieq
In the above equation, φ and Φ represents standard normal pdf and CDF. Also, µieq
and σieq in Eq. 3.67 and 3.68 is the mean and standard deviation of the equivalent
normal distribution at Xi∗ . In this context, it may be noted that the reason behind
satisfying pdf and CDF are to from two equations for unknown parameters µieq
and σieq of the equivalent normal distribution. Moreover, as the design point X ∗
changes in successive iterations, the values of the equivalent normal distribution
also changes. Using second condition, one gets
µieq = −σieq Φ −1 [FXi (xi∗ )] + Xi∗ (3.40)
φ {Φ −1 [FXi (xi∗ )]}

σieq = (3.41)
fXi (xi∗ )
3.4 Lecture 04: FORM for Correlated Non-Normal Random Variables 23
Once the equivalent mean and standard deviation are evaluated at the beginning of
each iteration, one can follow the Rackwitz algorithm to find optimal distance. So
the modified Rackwitz algorithm for uncorrelated non-normal case will be
• Evaluate equivalent mean and standard deviation of random variables as per the
defined distribution scheme.
• Transform the limit state function g(X) into standard normal space (i.e., U space)
using equivalent mean and standard deviation, as per Eq. 3.69 and 3.70. Now, the
limit state is in terms of equivalent standard normal variables, i.e. g(U).
• Evaluate the derivative of the transformed limit state with respect to standard
normal variables (i.e., G) as explained for Eq. 3.55.
• Assume initial β value (generally, considered in range of 35) and choose direc-
tion cosine in such a way that the summation of square of direction cosine is one
2
(Σ αi∗ = 1).
• Write the limit state g(U) in terms of β and αi∗ substituting the values using Eq.
3.58.
• Solve the expression for new β value. It can be shown as
g(βk , α1∗k , α2∗k , ..., αn∗k , ) = 0 (3.42)
where k represents the kth iteration step

• New β is used to evaluate kG∗ k by
s 2
n
∗ ∂g
kG k = ∑ (3.43)
i=1 ∂ u∗i
• New direction cosines αi∗k+1 are calculated employing Eq. 3.59

• The design point is calculated by
Xi∗ = µieq + αi β σieq (3.44)
• New equivalent mean and standard deviation are calculated based on the new
design point.
• Now, check the difference in two successive iteration values of β . Stop the itera-
tion if it is less than stopping criteria otherwise go back to Step 6.
Examples:
Ex 01. Find equivalent normal parameters (i.e. mean and standard deviation) of
a random variable x following lognormal distribution with mean and standard devi-
ation as 1 and 0.5, respectively at x∗ = 1.5. Also, check whether the area under the
pdf up to x∗ is equal for original and equivalent parameters.
Sol. From previous module, we know the basic statistical parameters and pdf distri-
bution of lognormal distribution, they are
σlnx = [ln(δx2 + 1)]2

1 2
x̃ = µx e− 2 σlnx

ln x∗
x̃

1
fX (x∗ ) = φ
x∗ σlnx σlnx
Further, one to one mapping of the CDF and pdf as specified above (specifically,
Eq. 3.43), one can get equivalent standard deviation σxeq , shown as

−1 x∗
φ Φ Φ{ln( x̃ )/σlnx }
σxeq =
fX (x∗ )

x∗
Φ ln x̃ /σlnx
σxeq =
fX (x∗ )
σxeq = x∗ σlnx
Moreover, equivalent mean µxeq can also be evaluated via same mapping.
∗
∗ ∗ −1 x
µxeq = x − x σlnx Φ φ {ln /σlnx }
x̃
= x∗ (1 − lnx∗ + lnx̃)
The calculation for equivalent normal mean and standard deviation is shown below
σlnx = [ln(0.52 + 1)]1/2 = 0.4724
1
x̃ = 1 × e( − × 0.47242 ) = 0.8944
2
σxeq = 1.5 × 0.4724 = 0.7086
µxeq = 1.5 × (1 − ln(1.5) + ln(0.8944)) = 0.7244
Now, one can evaluate the pdf at the point x∗ and subsequent area under the pdf
(i.e., CDF) as shown on the Figure 3.4.
1.5
∗ 1 ln( 0.8944 )
fX (x ) = φ = 3.0928 × 10−1
1.5 × 0.4724 0.4724
∗
ln( x ) ln( 1.5 )

1 1
FX (x∗ ) = 1 + er f √ x̃ = 1 + er f √ 0.8944 = 8.6314 × 10−1
2 2σlnx 2 2 × 0.4724
(x∗ −µxeq )2 2
(1.5+0.7244)
1 2σx2eq 1
φ (x∗ , µxeq , σxeq ) = √ e =√ e 2×(0.7086)2 = 3.0928×10−1
2πσxeq 2π × 0.7086
3.4 Lecture 04: FORM for Correlated Non-Normal Random Variables 25
∗
x − µxeq

1 1 3.8908 + 0.7244
Φ(x∗ , µxeq , σxeq ) = 1 + er f √ = 1 + er f √
2 2σxeq 2 2 × 0.7086
= 8.6314 × 10−1
Both, pdf and CDF of the original and equivalent normal matches with each other.
Once the first question is addressed, let us focus on the second question i.e. what
will be the optimal distance if the random variables are correlated? So let us con-
sider a g(X) here, X follows correlated normal distribution. In this context, it may
be noted that the most general case is the one where X is correlated non-normal.
However, once the problem of correlated normal is solved, one can easily combine
that the previous case of equivalent normal to address the most general case. Let
X = {X1 , X2 , ..., Xn } are correlated normal whose correlation matrix is given by
 
Var(X1 ) Cov(X1 , X2 ) Cov(X1 , X3 ) . . . Cov(X1 , Xn )
Cov(X2 , X1 ) Var(X2 ) Cov(X2 , X3 ) . . . Cov(X2 , Xn )
[CX ] =  (3.45)
 
.. .. .. .. 
 . . . . 
Cov(Xn , X1 ) Cov(Xn , X2 ) Cov(Xn , X3 ) . . . Var(Xn )
To decompose the correlation in X an orthogonal transformation is used which is

given by
X = VZ
where, [V ] is the eigen vector of CX which is given by
 
ρ11 ρ12 . . . ρ1 j . . . ρ1n
ρ21 ρ22 . . . ρ2 j . . . ρ2n 
[V ] =  . . (3.46)
 
.. .. 
 .. .. . . 
ρn1 ρn2 . . . ρn j . . . ρnn
using this orthogonal transformation the limit state g(X) is converted to g(Y ) =
0. However, the question remains to be addressed as what are the remains to be
addressed is what are the statistical properties (i.e., mean and variance) of the new
random variable Z? With this in view, let us take the expectation on both sides of
Eq. 3.48. and use the orthogonal properties of V (i.e., V −1 = V T )) which yields
Y = [V ]T X (3.47)
X = [V ]Y (3.48)
T
E[Y ] = [V ] E[X] (3.49)
Further, using the orthogonal properties of the eigen vector one can show that
 
λ1 0 ... 0
0 λ2 ... 0
[CY ] = [V ]T [CX ][V ] = [λ ] =  . (3.50)
 
.. .. .. 
 .. . . .
0 0 . . . λn
where, λ is the diagonal matrix whose leading diagonal represents the variance of
Z (i.e.,λi = σZ2i )
(Yi − µYi )
Ui = (3.51)
σYi
Once the covariance string is decoupled, the limit state in transformed coordinate
g(Z) = 0 may be further transformed into standard normal space to find out β using
following transformation

X = [V ] [σY ]U + µY
= [V ]σY U + [V ]µY (3.52)
Using above equation in Eq. 3.51, one can show that following transformation con-
verts the original limit state in standard normal space
 
σY1 0 . . . 0
 0 σY . . . 0 
2
[σY ] =  . . . .  (3.53)
 
 .. .. .. .. 
0 0 . . . σYn
µY = (µY1 , µY2 , ..., µYn )t

U = (U1 , U2 , ..., Un )t
Examples:
Ex 02: Two correlated random variables with statistical parameters given in table
below follow normal distribution. The coefficient of correlation between the random
variables is 0.35. Evaluate uncorrelated statistical parameters and prove the random
variables are uncoupled.
Random Variables (X) µX σX

x1 34 3.4
x2 50 10
Solu. Following the Eq. 3.67, one can find the covariance matrix as

3.4 × 3.4 0.35 × 3.4 × 10 11.56 11.9
[CX ] = =
0.35 × 3.4 × 10 10 × 10 11.9 100
3.5 Lecture 05: FORM Examples 27
Decomposing the correlated matrix via orthogonal transformation, one needs to find
the transformation matrix and for doing that eigen values and vectors are evaluated
as shown below
(11.56 - λ ) 11.9
=0
11.9 (100−λ )
(11.56 − λ )(100 − λ ) − 11.902 = 0
λ = 9.9868, 101.5732

−0.9914 0.1311
v1 = , v2 =
0.1311 0.9914
Orthogonal transformation matrix and decoupled random variables are

−0.9914 0.1311
V = [v1 v2] =
0.1311 0.9914

t −0.9914 0.1311 34 −27.1536
Y =V X = × =
0.1311 0.9914 50 54.0248
The X matrix contains the mean values of correlated random variables. So, the Y
matrix gives the uncorrelated mean values whereas standard deviation of the uncor-
related random variables, i.e. σY are evaluated by the eigen values.
√
√ 9.9868 3.1602
σY = =
101.5732 10.0784
Now, evaluate variance matrix for Y

t 9.9868 0
CY = V CX V =
0 101.5732
We get a diagonal matrix with eigen values which, indeed, are variance of the cor-
related random variables. In other words, this proves the decoupling of correlated
random variables by using the orthogonal properties of eigen vectors.
3.5 Lecture 05: FORM Examples
Ex 01. A steel cantilever beam of a rectangular cross section subjected to point

load at its free end is considered. The limit state of section failure at the fixed end is
defined by equation below
g(X) = X1 X2 − X3 L
where X1 , X2 , X3 and L are representing yield strength, section modulus, point load
and length of the beam, respectively. Out of all design variables,1X, X2 and X3 are
random variables whereas L is deterministic and equal to 2m. Three cases are con-
sidered where random variables are either following normal or lognormal distribu-
tion. The statistical properties are given as
Random Variables (X) µX σX

x1 (kN/m2 ) 2.5 × 105 2.5 × 104
x2 (m3 ) 1.2 × 10−3 6 × 10−5
x3 (kN) 100 15
Solu.
Case 1: All random variables are following normal distribution
Transforming the original function into standard normal space, individual transfor-
mation of random variables as shown below
X1 = µ1 + u1 σ1
X2 = µ2 + u2 σ2
X3 = µ1 + u3 σ3
are substituted in the original equation. This in turn modifies the limit state function
as
g(U) = (µ1 + u1 σ1 )(µ2 + u2 σ2 ) − (µ1 + u3 σ3 )L

= µ2 σ1 u1 + µ1 σ2 u2 + σ1 σ2 u1 u2 − Lσ3 u3 + µ1 µ2 − Lµ3
Equating the above expression to zero, substituting the direction cosines αi and β ,
one can get
g(U) = µ2 σ1 α1 β + µ1 σ2 α2 β + σ1 σ2 α1 α2 β 2 − Lσ3 α3 β + µ1 µ2 − Lµ3 = 0
Now, iteratively solve the above polynomial equation for unknown β . Considering
stopping criteria as |βk+1 − βk | ≤ ε, where ε is allowable error (taken as 0.001). The
intermediate iteration calculation values are shown in the table below Results from
Elements Iteration
1 2 3 4
β 0 2.2601 2.2576 2.2576
α1 0.5774 -0.6667 -0.6631 -0.6648
α2 0.5774 0.3333 -0.2926 -0.2923
α3 0.5774 0.6667 0.6890 0.6875
u1 0 -1.5067 -1.4969 -1.5008
u2 0 -0.7534 -0.6606 -0.6599
u3 0 1.5067 1.5555 1.5521
X1 2.5×105 2.1233×10−3 2.1258×10−3 2.1248×10−3
X2 1.2×10−3 1.1548×10−3 1.1604×10−3 1.1604×10−3
X3 100 122.60 123.33 123.28
4th iteration gives β = 2.2576 and now, p f is calculated as Φ(2.2576) = 1.1984102
Case 2: X2 is following lognormal distribution and rest are following normal

distribution
In case of non-normal distributed random variables, transformation of the origi-

nal function into standard normal space is done by equivalent normal mean and
standard deviation, as specified in the algorithm for non-normal variables. Now, the
individual transformation of random variables can be shown as per Ex 01 in Section
3.4.
X1 = µ1 + u1 σ1
X2 = µ2eq + u1 σ2eq
X3 = µ3 + u3 σ3
where, µ2eq and σ2eq are equivalent mean and standard deviation for random variable
X2 . After substituting these transformation, direction cosines αi and β in to the
original limit state function, one gets
g(U) = (µ1 + u1 σ1 )(µ2eq + u2 σ2eq ) − (µ1 + u3 σ3 )L

= µ2eq σ1 u1 + µ1 σ2eq u2 + σ1 σ2eq u1 u2 − Lσ3 u3 + µ1 µ2eq − Lµ3
Before going to iterations, one must evaluate the parameters of non-normal random
variables. In this case, parameters of lognormal distributed random variable X3 are
evaluated from a set of equations.
Once the equivalent parameters are evaluated one can resume the iteration proce-
dure. As note, one should remember to evaluate the equivalent parameters at each
iteration step. Finally, the polynomial equation is iteratively solved for β with stop-
ping criteria same as the previous case. The intermediate iteration calculation values
are shown in the table below Results converges at 3rd iteration giving β = 2.2537
Elements Iteration
1 2 3
β 0 2.2529 2.2537
α1 0.5774 -0.6663 -0.6650
α2 0.5774 -0.3333 -0.2826
α3 0.5774 0.6671 0.6913
µ2eq 1.1985 1.1977 1.1979
σ2eq 5.9963×10−5 5.7711×10−5 5.8051×10−5
u1 0 -1.5010 -1.4987
u2 0 -0.7510 -0.6368
u3 0 1.5028 1.5580
X1 2.5×105 2.1248×105 2.1253×105
X2 1.2×10−3 1.1549×10−3 1.1618×10−3
X3 100 122.54 123.37
and now, p f is calculated as Φ(2.2537) = 1.2107102.
Case 3: X3 is following lognormal distribution and rest are following normal

distribution
For brevity, direct calculations of each iterations are shown for this case
Results converges at 4th iteration giving β = 2.1887 and now, p f is calculated as
Elements Iteration
1 2 3 4
β 0 2.3174 2.1904 2.1887
α1 0.5774 -0.6683 -0.5960 -5.9375
α2 0.5774 0.3342 -0.2620 -2.6575
α3 0.5774 0.6646 0.7590 7.5950
µ3eq 98.887 96.147 95.731 95.732
σ3eq 14.917 18.363 18.637 18.636
u1 0 -1.5487 -1.3055 -1.2996
u2 0 -7.7436 -5.7388 -5.8164
u3 0 1.5401 1.6625 1.6623
X1 2.5×105 2.1128 2.1736 2.1751
X2 1.2×10−3 1.1535 1.1656 1.1651
X3 100 123.10 1.2494 1.2493
Φ(2.1887) = 1.4308102 .
Ex 02. In addition of previous example another point load X4 at mid span is con-
sidered. The new limit state of failure for maximum deflection at the fixed end is
defined by equation below
g(X) = X1 X2 − X3 L − 0.5X4 L
where notations follow same meaning as in Ex 01. Additionally, correlated ran-

dom variables XX and X4 has correlation coefficient 0.6. The statistical properties
are given as
S.No. Random Variables (X) µX σX Distribution

1 X1 (kN/m2 ) 2.5 × 105 2.5 × 104 Normal
2 X2 (m3 ) 1.2 × 10−3 6 × 10−5 Normal
3 X3 (kN) 60 90 Normal
3 X4 (kN) 70 10.5 Normal
Solu.
The Covariance matrix is evaluated as

(2.5 × 104 )2
 
0 0 0
 0 (6 × 10−5 )2 0 0 
CX =  2

 0 0 (9) 0.6 × 9 × 10.5
0 0 0.6 × 9 × 10.5 (10.5)2
6.25 × 108
 
0 0 0
 0 3.6 × 10 −9 0 0 
= 
 0 0 81 56.7 
0 0 56.7 110.25
Now, eigen value problem is solved for finding the orthogonal transformation ma-
trix. Following are the steps for finding the matrix.

6.25 × 108 − λ 0 0 0
−9

0 3.6 × 10 − λ 0 0
=0

0 0 81 − λ 56.7
0 0 56.7 110.25 − λ
The roots of the above expression are
λ = 3.6 × 10−9 , 37.069, 154.18, 6.25 × 108
and the transformation matrix is

 
10 0 0
0 1 0 0 
V = 
0 0 −0.7905 0.6125
0 0 0.6125 0.7905
Where every column vector is normalized eigen vector. Now, uncoupled random
variables are evaluated as per Eq. 3.67. and substituted to limit state in place of
original random variables with corresponding mean and standard deviation (shown
in table below)
Uncoupled Random µY σY
Variable(Y)
Y1 (kN/m2 ) 2.5 × 105 2.5 × 104
Y2 (m3 ) 1.2 × 10−3 6 × 10−5
Y3 (kN) −4.5568 6.0884
Y4 (kN) 92.0830 12.417
The intermediate iteration calculation values are shown in the table below
Results converges at 4th iteration giving β = 2.6615 and now, p f is calculated
Elements Iteration
1 2 3 41
β 0 2.6664 2.6615 2.6615
α1 0.5 -0.7099 -0.7088 -0.7114
α2 0.5 -0.3549 -0.3016 -0.3007
α3 0.5 -0.1395 -0.1463 -0.1457
α4 0.5 0.5922 0.6207 0.6183
u1 0 -1.8928 -1.8865 -1.8934
u2 0 -0.9464 -0.8027 -0.8002
u3 0 -0.3720 -0.3892 -0.3877
u4 0 1.5789 1.6519 1.6455
Y1 2.5×105 2.0268×105 2.0284×105 2.0266×105
Y2 1.2×10−3 1.1432×10−3 1.1518×10−3 1.1520×10−3
Y3 -4.5568 -6.8219 -6.9266 -6.9175
Y4 92.0830 111.69 112.59 112.52
X1 2.0266×105
X2 1.1520×10−3
X3 74.380
X4 84.706
as Φ(2.6615) = 3.889810−3 .
3.6 Lecture 06: Modified Rackwitz Algorithm and Nataf Model
In this lecture, we will explore the difficulties encountered in reliability analysis

by Rackwitz algorithm and possible solution. From Eq. 3.3.11 in Lecture 3 in this
module one can easily notice that Rackwitz algorithm needs to solve that polynomial
equation of β to find optimal distance in each iterations. The precondition to this
step is that g(X) must have algebraic function and roots are stable. However, in real
cases, limit state often contains non-algebraic terms and hence, Rackwitz algorithm
is not applicable. To overcome this problem, Rackwitz and Fiessler proposed an
iterative solution strategy based on NewtonRaphson algorithm.
3.6 Lecture 06: Modified Rackwitz Algorithm and Nataf Model 33
1
u∗k+1 = [5g(u∗k )t u∗k − g(u∗k )] 5 g(u∗k ) (3.54)
| 5 g(u∗k )|2
|u∗k − u∗k−1 | ≤ δ (3.55)

|g(u∗k )| ≤∈ (3.56)
Let Xk+1 be the solution at (k + 1)th step. In this context, one should note that Xk+1
is generic in this case. Using Taylor’s series expression one gets
g(u∗k+1 ) = g(u∗k ) + 5g(u∗k )(u∗k+1 − u∗k ) (3.57)
If Xk+1 is a possible solution of g(X) then the value of g(X) is zero. Therefore, one
gets
g(u∗k ) + 5g(u∗k )(u∗k+1 − u∗k ) = 0 (3.58)
Using above equation in iterative scheme one can solve/identify the root of the equa-
tion. This is further depicted in figure below
Limitations of Rackwitz Fiessler’s Algorithm
As NewtonRaphson based RackwitzFiessler algorithm solves roots based on ini-

tial guess and gradient vector, it may oscillate or diverge in some special cases as
shown in figure below
Moreover, it may diverge in some cases also. Besides this, it depends on initial
guess and hence may converge to a different root if the initial guess is close to some
other root than the actual one.
This algorithm might also fail in some special case as shown by Der Kiureghian
(1986). Imagine the limit state of the form g(Z) = Z1 Z2 − C If the initial guess is
now at (p, q) which is on the ellipse Z12 + Z22 ±C = 0 then the design point oscillates
between (p, q) and (q, p) as shown in figure below (see Figure 3.6.2).
However, it should be noted that the global minimum distance β can be found by
any optimization algorithm. Therefore, if any convergence problem is noticed, al-
ternative algorithm may be employed to find β .
Nataf Model
This transformation model was first proposed by Liu and Der Kiureghian (1986)
based on Nataf (1962) mathematical explanation about distribution of variables. The
model describes the joint pdf of random variables based on their individual distri-
butions and correlation
Before starting the explanation of this mathematical model, let us first look into
assumptions considered in the formulation
• Mapping of the transformation from original to standard normal (as per Eq. 3.6.5)
must be one to one. In other words, CDF of random variables must be continuous
and strictly increasing.
0
• The correlation matrix C must be positive definite which is true by default.
According to this model, joint pdf of the correlated random variables is evaluated
by
0
ϕn (u,C )
fx (x) = fx1 (x1 ) fx2 (x2 ).... fxn (xn ) (3.59)
ϕ(u1 )ϕ(u2 )...ϕ(un )
where, fxi (xi ) is the pdf of random variable (xi ), ϕ(.) represents normal probability
distribution function. Transforming random variables from original space to stan-
dard normal space via one to one CDF mapping, one can use ui = ϕ −1 [ fxi (xi )]
where fxi (xi ) is CDF of random variable and ϕ(.) is CDF of normal distributed
0
random variable C is the modified correlation matrix as shown below
0
C = matrix (3.60)
(xi − µi ) (x j − µ j ) ϕ2 (ui , u j , ρi j )
Z ∞Z ∞
ρi j = fXi (xi ) fX j (x j ) dxi dx j (3.61)
−∞ −∞ σi σj ϕ(ui )ϕ(u j )
(xi − µi ) (x j − µ j )
Z ∞Z ∞
ρi j = ϕ2 (ui , u j , ρi j )dui du j (3.62)
−∞ −∞ σi σj
0 0
where, the elements of correlation matrix C , i.e. ρi j can be calculated from known
correlation coefficient ρi j Eq. 3.6.7 gives a relation between both defined and mod-
ified correlation coefficients (ρ 0i j and ρi j , respectively) considering two correlated
random variables. The unknown correlation coefficient ρ 0i j can be evaluated from
ρ 0i j = ρi j F, where the unique factor F is calculated from a series of empirical for-
mulas suggested by Liu and DeKiureghian (1986). Application of this model can
be justified to a wider range of correlation coefficient as compared to its competent
Morgenstern Model [by Liu and Der Kiureghian (1986)].
Examples :
Ex 02 : Consider two correlated random variables, x1 and x2 with correlation coef-

ficient 0.3. Individual properties of both the random variables are stated in the table
below
Random Variable Distribution CDF pdf

x1 Exponential 1 − e−x1 e−x1
−x −x
x2 Type ILargest e−e 2 e−x2 −e 2
3.6 Lecture 06: Modified Rackwitz Algorithm and Nataf Model 35
Evaluate the joint pdf by using Nataf model
Solu. Using the empirical expressions given by Liu and Der Kiureghian (1986) for
exponential and type Ilargest distribution one can calculate the unique factor F as
expressed below
F = 1.142 − 0.154ρ + 0.031ρ 2
where, ρ is the known correlation coefficient (= 0.3, in this case). Eventually, F

comes out to be 1.0986 and modified correlation coefficient ρ is calculated as shown
below
0
ρ = 0.3 × 1.0986 = 0.3296
Now, Nataf’s estimate of the joint pdf is evaluated as per Eq. 3.6.8.
0
ϕ2 (u1 , u2 , ρ )
fx1 ,x2 (x1 , x2 ) = fx1 (x1 ) fX2 (x2 )
ϕ(u1 )ϕ(u2 )
where, the joint normal probability distribution function ϕ2 (.) for two random vari-
ables is expressed as
0
u2 −2ρ u1 u2 +u2
0 1 − 1 0
2
ϕ2 (u1 , u2 , ρ ) = e 2(1−ρ 2 )
0
p
2π 1−ρ 2
Now, all pdf elements as per the joint pdf equation stated above can be shown as
fx1 (x1 ) = e−x1

−x2
fx2 (x2 ) = e−x2 −e
u2 2
1 −2×0.3296×u1 u2 +u2
0 1 −
2(1−0.32962 )
ϕ2 (u1 , u2 , ρ ) = √ e
2π 1 − 0.32962
1 u2
1
ϕ(u1 ) = √ e− 2
2π
1 u2
2
ϕ(u2 ) = √ e− 2
2π
Substituting all these, one can easily evaluate the expression for joint pdf as per
Nataf model. It can be expressed as
−x2 2 2
fx1 ,x2 (x1 , x2 ) = 1.0592e−x1 −x2 −e e0.0609u1 −0.3698u1 u2 +0.0609u2
−x1
where, u1 can be evaluated by one to one CDF mapping given by u1 = ϕ −1 [e−e ]
−x
Similarly, for second standard normal variable, i.e. u2 = ϕ −1 [e−e 2 ] .
3.7 Lecture 07: Second Order Reliability Method (SORM)
In previous lectures, a first order approximation for g(X) are discussed for various
combinations of the type of random variables to evaluate reliability index. But in
reality, the limit states are highly non-linear in standard normal space and hence a
first order approximation may contribute significant error in reliability index evalu-
ation. Thus, a better approximation by the second order terms is required for highly
non-linear limit states. This is often termed as Second Order Reliability Method
(SORM). Revisiting the Taylor’s series,
∂g
g(X1 , X2 , .., Xn ) = g(x1∗ , x2∗ , .., xn∗ ) + Σi=1
n
(xi − xi∗ ) (3.63)
∂ Xi
1 n n ∂ 2g
+ Σi=1 Σ j=1 (xi − xi∗ )(x j − x∗j ) + ... (3.64)
2 ∂ Xi ∂ X j
One can notice that unlike FORM the expression in Eq. 3.7.1 is considered till
second order terms. Using this higher order expansion, Breitung (1984) suggested
an asymptotic approximation as shown in Eq. 3.7.2 for estimating probability of
failure based on the β estimated in FORM,
n−1 1
p fSORM ≈ ϕ(−β ) ∏ (1 + β ki )− 2 (3.65)
i=1
where ki is the main curvatures of the limit state surface at design point. It is com-
puted through a series of steps as explained below
• Initial orthogonal matrix To is evaluated from the direction cosines evaluated as
explained in FORM under Rackwitz algorithm (see Eq. 3.3.12)
 
1 0 ... 0
 0 1 ... 0 
To =  . . . .  (3.66)
 
. .
. . . . . .
α1 α2 . . . αn
• Further,To = [t01 ,t02 , ..t0n ]t is modified using Gram-Schmidt orthogonal proce-

dure as
t
titok
n
tk = t0k − Σi=k+1 ti (3.67)
tittt
3.7 Lecture 07: Second Order Reliability Method (SORM) 37
where, tk is row vectors of modified orthogonal matrix T = [t1 ,t2 , ..tn ]t and k ranges
from n, n − 1, n − 2, ..., 2, 1. note suffix t, i.e.(.)t denotes transpose of corresponding
vector or matrix. The rotation matrix in produced by normalizing these vectors to
unit.
• Orthogonal transformation of random variables into as shown in Eq. 3.7.3 is
evaluated using orthogonal matrix (also known as rotation matrix)
Y = TX (3.68)
• Again using orthogonal matrix T , another matrix A is evaluated as
(T HT t )i j
A = [ai j ] = i, j = 1, 2, ...n − 1 (3.69)
kG∗ k
where, H represents a double derivative matrix (or Hessian matrix) of the limit state
in standard normal space at design point
• Further, the last row and last column in the A matrix and last row in the Y matrix
are eliminated to consider a factor that last variable yn coincides with β computed
in FORM.Thus, limit state is expressed as
1
yn = β + yt Ay (3.70)
2
• Now, the size of coefficient matrix A is reduced to (n − 1) × (n − 1) and main
curvatures ki are given by computing eigen values of matrix A.
Examples :
Ex 01. Consider a limit state g(X) = x1 x2 − 1140. Result shows the reliability in-
dex estimate by FORM is 5.0957. Now, using SORM, estimate the reliability index.
With equivalent normal mean and standard deviation as 33.8083, 54 and 2.1751, 2.7
, respectively and the design point in standard normal space,
z1 = −4.2614 z2 = −2.7940
and direction cosines are α1 = −0.8363 and α2 = −0.5483 (evaluated in the last
iteration).Also, kG∗ k is 120.8318.
Solu. As per the algorithm discussed above, one have to first evaluate the rotation or
orthogonal matrix with the help of direction cosines which are α1 = −0.8363 and
α2 = −0.5483 Initially, To is evaluate as

1 0
−0.8363 −0.5483
Using Eq. 3.7.4, GramSchmidt orthogonal expression, one can evaluate the rotation
matrix as
t2 = {−0.8363 −0.5483} = {−0.8363 −0.5483}

(−0.8363)
t1 = {1 0} - (0.8363)2 +(0.5483)2
{−0.8363 − 0.5483} = {0.3006 0.4585}
Both the row vectors of the rotation matrix are normalized to form a unit vector
and finally, the rotation matrix T can be given as

0.5483 0.8363
T=
−0.8363 −0.5483
Next, the Hessian matrix or the double derivative matrix is formed as

 2
∂ 2g

∂ g
∂z 2 ∂ z1 ∂ z2 0 σ1eq σ2 0 5.8728
H =  ∂ 21g ∂ 2g
 = =
σ1eq σ2 0 5.8728 0
∂ z1 ∂ z2 ∂ z22
Also, kG∗ k is directly evaluated from FORM as 120.8318. Now, coefficient ma-
trix A is evaluated as per Eq. 3.7.6.

1 0.5483 0.8363 0 0.58728 0.5483 0.8363 −0.0446 0.0194
A = 120.8318 =
−0.8363 −0.5483 0.58728 0 −0.8363 −0.5483 0.0194 0.0446
As explained above, the last row and column of matrix A are removed keeping in
view that that the rotation coordinates makes the last variable coincide with β vec-
tor. Now, only a11 = −0.0446 element of the matrix A is left. Finally, calculating
the eigen values as specified in Step 6, above, one gets main curvature k1 = −0.0446
Substituting the required values in Breitung’s model

1
p fSORM ≈ ϕ(−5.0957)[1 + 5.0957(−0.0446)]− 2
βSORM = −ϕ −1 (1.9763 × 10−7 ) = 5.0712

Note that for this example, SORM gives lower reliability index (thus, higher failure
probability) in comparison to FORM results. This is due to concave nature of failure
surface in standard normal space
Breitung’s model didn’t consider the mixed terms in the second order term of Tay-
lor’s series expansion as shown in Eq. 3.7.1. Additionally, his model is less accurate
for lower β values as it employs asymptotic approximation. Thus, a more general-
ized model was suggested by Tvedt (1990) without using asymptotic approximation.
He considered the complete second order term for estimating accurate probability
of failure structure. According to this model, the limit state is approximated using
complete Taylor’s series till second order. SORM estimations are done based on
diagonal formation at approximated failure surface in standard normal space using
eigen value problem. Der Kiureghian et al. (1987) developed an asymptotic approx-
imation which didn’t calculated the complete Hessian matrix but used curvatures
same as Breitung’s Model and approximated the limit state by two semi-parabolas.
The details of these methods may be obtained from the following references.
Chapter 4
Simulation Techniques
39
Chapter 5
Meta-Modeling Methods
41
Chapter 6
Stochastic Finite Element Method
43
Chapter 7
Application on Design Problems
7.1 Reliability Based Design Optimization
45
Appendix A
App 1
Use the template appendix.tex together with the Springer document class SVMono
(monograph-type books) or SVMult (edited books) to style appendix of your book
in the Springer layout.
A.1 Section Heading

heading be followed by at least a short passage of text. Further on please use the
A.1.1 Subsection Heading

described in Sect. A.1.
For multiline equations we recommend to use the eqnarray environment.
a×b = c
a×b = c (A.1)
A.1.1.1 Subsubsection Heading

47
48 A App 1
Fig. A.1 Please write your

figure caption here
described in Sect. A.1.1.
Table A.1 Please write your table caption here

References
1. Broy, M.: Software engineering — from auxiliary to key technologies. In: Broy, M., Dener,
E. (eds.) Software Pioneers, pp. 10-13. Springer, Heidelberg (2002)
2. Dod, J.: Effective substances. In: The Dictionary of Substances and Their Effects. Royal
Society of Chemistry (1999) Available via DIALOG.
http://www.rsc.org/dose/titleofsubordinatedocument.
Cited15Jan1999
3. Geddes, K.O., Czapor, S.R., Labahn, G.: Algorithms for Computer Algebra. Kluwer, Boston
(1992)
4. Hamburger, C.: Quasimonotonicity, regularity and duality for nonlinear systems of partial
differential equations. Ann. Mat. Pura. Appl. 169, 321–354 (1995)
5. Slifka, M.K., Whitton, J.L.: Clinical implications of dysregulated cytokine production. J. Mol.
Med. (2000) doi: 10.1007/s001090000086
6. J. Dod, in The Dictionary of Substances and Their Effects, Royal Society of

Chemistry. (Available via DIALOG, 1999), http://www.rsc.org/dose/
titleofsubordinatedocument.Cited15Jan1999
7. H. Ibach, H. Lüth, Solid-State Physics, 2nd edn. (Springer, New York, 1996), pp. 45-56
8. S. Preuss, A. Demchuk Jr., M. Stuke, Appl. Phys. A 61
9. M.K. Slifka, J.L. Whitton, J. Mol. Med., doi: 10.1007/s001090000086
10. S.E. Smith, in Neuromuscular Junction, ed. by E. Zaimis. Handbook of Experimental Phar-
macology, vol 42 (Springer, Heidelberg, 1976), p. 593
11. Calfee, R. C., & Valencia, R. R. (1991). APA guide to preparing manuscripts for journal
publication. Washington, DC: American Psychological Association.
12. Dod, J. (1999). Effective substances. In: The dictionary of substances and their effects. Royal
Society of Chemistry. Available via DIALOG.
http://www.rsc.org/dose/Effectivesubstances. Cited 15 Jan 1999.
13. Harris, M., Karper, E., Stacks, G., Hoffman, D., DeNiro, R., Cruz, P., et al. (2001). Writing
labs and the Hollywood connection. J Film Writing, 44(3), 213–245.
14. O’Neil, J. M., & Egan, J. (1992). Men’s and women’s gender role journeys: Metaphor for
healing, transition, and transformation. In B. R. Wainrig (Ed.), Gender issues across the life
cycle (pp. 107–123). New York: Springer.
15. Kreger, M., Brindis, C.D., Manuel, D.M., Sassoubre, L. (2007). Lessons learned in systems
change initiatives: benchmarks and indicators. American Journal of Community Psychology,
doi: 10.1007/s10464-007-9108-14.
49
50 References
16. Alber John, Daniel C. O’Connell, and Sabine Kowal. 2002. Personal perspective in TV inter-
views. Pragmatics 12:257–271
17. Cameron, Deborah. 1997. Theoretical debates in feminist linguistics: Questions of sex and
gender. In Gender and discourse, ed. Ruth Wodak, 99–119. London: Sage Publications.
18. Cameron, Deborah. 1985. Feminism and linguistic theory. New York: St. Martin’s Press.
19. Dod, Jake. 1999. Effective substances. In: The dictionary of substances and their effects.
Royal Society of Chemistry. Available via DIALOG.
http://www.rsc.org/dose/title of subordinate document. Cited 15 Jan 1999
20. Suleiman, Camelia, Daniel C. OConnell, and Sabine Kowal. 2002. ‘If you and I, if we, in this
later day, lose that sacred fire...’: Perspective in political interviews. Journal of Psycholin-
guistic Research. doi: 10.1023/A:1015592129296.
21. Brown B, Aaron M (2001) The politics of nature. In: Smith J (ed) The rise of modern ge-
nomics, 3rd edn. Wiley, New York
22. Dod J (1999) Effective Substances. In: The dictionary of substances and their effects. Royal
Society of Chemistry. Available via DIALOG.
http://www.rsc.org/dose/titleofsubordinatedocument.
Cited15Jan1999
23. Slifka MK, Whitton JL (2000) Clinical implications of dysregulated cytokine production. J
Mol Med, doi: 10.1007/s001090000086
24. Smith J, Jones M Jr, Houghton L et al (1999) Future of health insurance. N Engl J Med
965:325–329
25. South J, Blass B (2001) The future of modern genomics. Blackwell, London
Glossary
Glossary in the Springer layout.

glossary term Write here the description of the glossary term. Write here the de-
scription of the glossary term. Write here the description of the glossary term.
51
Solutions
Problems of Chapter ??
2.1 The solution is revealed here.

2.2 Problem Heading
(a) The solution of first part is revealed here.
(b) The solution of second part is revealed here.
53
Index
A P
acronyms, list of xv paragraph 4, 9

permission to print 2, 8
C preface ix
problems 53
citations 2, 8
cross-references 2, 8
S
G
solutions 53
glossary 51 symbols, list of xv
55

A Textbook On Structural Reliability: Arunasis Chakraborty

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Textbook On Structural Reliability: Arunasis Chakraborty

Uploaded by

Copyright:

Available Formats

Arunasis Chakraborty

February 10, 2019

Place, month year Firstname Surname

A preface is a book’s preliminary statement, usually written by the author or editor

If you prefer to set your acknowledgement section as a separate chapter instead of

3 Level 2 Reliability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

6 Stochastic Finite Element Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

7 Application on Design Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

1.1 Section Heading

1.2 Section Heading

Instead of simply listing headings of different levels we recommend to let every

however, for multiline equations we recommend to use the eqnarray environ-

1.2.1 Subsection Heading

Instead of simply listing headings of different levels we recommend to let every

1.2.1.1 Subsubsection Heading

Instead of simply listing headings of different levels we recommend to let every

Instead of simply listing headings of different levels we recommend to let every

For typesetting numbered lists we recommend to use the enumerate environ-

Fig. 1.1 If the width of the

Fig. 1.2 Please write your

Table 1.1 Please write your table caption here

Translation mRNAa 22 (19–25) Translation repression, mRNA cleavage

1.3 Section Heading

Instead of simply listing headings of different levels we recommend to let every

1.3.1 Subsection Heading

In order to avoid simply listing headings of different levels we recommend to let

If you want to emphasize complete paragraphs of texts we recommend to use

1.3.1.1 Subsubsection Heading

Instead of simply listing headings of different levels we recommend to let every

Theorem 1.1. Theorem text goes here.

Definition 1.1. Definition text goes here.

Proof. Proof text goes here. t

Instead of simply listing headings of different levels we recommend to let every

Definition 1.2. Definition text goes here.

Proof. Proof text goes here. t

1.2. Problem Heading

2.1 Section Algebra of Variance

2.2 Section Heading

Instead of simply listing headings of different levels we recommend to let every

however, for multiline equations we recommend to use the eqnarray environ-

2.2.1 Subsection Heading

Instead of simply listing headings of different levels we recommend to let every

2.2.1.1 Subsubsection Heading

Instead of simply listing headings of different levels we recommend to let every

Instead of simply listing headings of different levels we recommend to let every

Fig. 2.1 If the width of the

For typesetting numbered lists we recommend to use the enumerate environ-

2.3 Section Heading

Instead of simply listing headings of different levels we recommend to let every

Fig. 2.2 Please write your

Table 2.1 Please write your table caption here

2.3.1 Subsection Heading

In order to avoid simply listing headings of different levels we recommend to let

If you want to emphasize complete paragraphs of texts we recommend to use

2.3.1.1 Subsubsection Heading

Instead of simply listing headings of different levels we recommend to let every

Theorem 2.1. Theorem text goes here.

Definition 2.1. Definition text goes here.

Proof. Proof text goes here. t

Instead of simply listing headings of different levels we recommend to let every

Definition 2.2. Definition text goes here.

Proof. Proof text goes here. t

2.2. Problem Heading