Professional Documents
Culture Documents
5.1 General
In the previous two chapters, the safety analysis of hybrid uncertain system (systems
characterized by both the probabilistic and the possibilistic variables) are presented in
probabilistic and possiblistic format. In those studies, the probabilistic variables are described by
the associated pdf and the possibilistic variables are described by the associated fuzzy mf.
However, in cases, for example in preliminary design phases, even though some experimental
data is available, it is not enough to construct reliable pdfs or mfs. The available data can be used,
uncertainties. One can only obtain the range or the lower bound and upper bound of the
structural parameters. In such cases, the uncertain design variables can be well modeled using
ellipse or any convex sets. This kind of mathematical model is called uncertain but bounded
model. Ben Haim and Elishakoff (1990), Ben-Haim (1996) proposed a version of worst-case
design based on convex models for design problems where there is scarce information about the
uncertain variables. A design should survive when the values of the uncertain variables vary in a
convex set, which is specified by the designer based on experience. In recent studies convex
models and interval analysis methods of uncertainty were developed in which bounds on the
encloses all possible combinations of the uncertain variables i.e. consist of all the possible
probabilities that are consistent with available information. It is basically a worst scenario
method since the convex models have to cover the uncertainties that cause the worst
when the design parameters needed to create a probabilistic models cannot be precisely
determined due to lack of data. But these works are restricted to single type of uncertain input
variables i.e. bounded variables and not compatible to tackle mixed uncertain input variables in
existing reliability analysis formats. The primary objective of the present chapter is to present a
second moment safety evaluation format for structural system characterized by hybrid
uncertainties i.e. both probabilistic and interval type uncertain parameters. The available
literatures on the application of interval analysis approach in the framework of set theoretical
description is limited to deal with response evaluation and optimization. But it is appeared to be
potential to apply in solving structural reliability analysis problem. The approach is extended
involving both random and bounded variables. The bounds of performance function (limit state
function) of a typical structural reliability analysis problem with UBB parameters are presented
in this section by making use of interval mathematics. Subsequently, the next section elaborates
An interval scalar consists of a single continuous domain in the domain of real numbers . The
lower and the upper bound of an interval scalar x I are denoted by xl and xu. A real closed interval
x {x ( x )( xl x xu )} . (5.1)
If xi is the nominal value of the i-th UBB variable viewed as the mean value, then the UBB
parameter value deviates from the nominal value can be further expressed as,
xiI [ xil , xiu ] [ xi xi , xi xi ]
function generally obtained by numerical evaluation of the related structural analysis problem.
Thus, it is extremely difficult to calculate the exact interval solution. The interval analysis
problems can be approximated to equivalent deterministic one through Taylor series expansion
to yield conservative response bounds (Qiu and Wang 2003). Chen and Wu (2004) have further
The limit state function of a reliability analysis problem involving both random and UBB
variables can be expanded with respect to `m` numbers of UBB variables in Taylor series about
Assuming monotonic responses and by making use of interval extension in interval mathematics,
m
g
g( X ) g( X )
I
xi e .. (5.6)
i 1 xi
Thus the interval region of the limit state function involving the UBB variables can be separated
m
g
g u ( X ) g ( X) xi (5.7)
i 1 xi
and
m
g
g l ( X ) g ( X) xi (5.8)
i 1 xi
It can be noted from above approximation that the limit state functions now involve the random
variables only. Thus, the rational approximation of the performance function is a linear function
of the bounded uncertainties and each such variable appear once in the expression and, the
Once the lower and the upper bounds on the performance function are obtained from Eqn.
(5.7) and (5.8), the bounds of reliability for UBB variables in combination with the remaining
random variables can be easily computed relying on the conventional second moment reliability
analysis format.
5.3 Numerical Study
The effectiveness of proposed safety evaluation approach is investigated by considering the same
two example problems of chapter three, viz : i) A reinforced concrete beam (Fig. 3.5) and ii) A
R.C. pile embedded in clay (Fig. 3.25). However in present case, the information available about
the uncertain variables are not sufficient to describe those probabilistically and can be modeled
A simply supported reinforced concrete beam of section 3.5.1 of chapter 3 (refer Fig. 3.5) has
been re considered to elucidate the proposed algorithm of reliability analysis of structures under
hybrid uncertainties. Specifically, the purpose of the example problem is to demonstrate the
capability of the present approach to tackle mixed uncertain parameters in existing safety
assessment format. In the example problem, for the demonstration purpose, some of the variables
are considered to be random and some are UBB type. . The characteristic strength of concrete
(fck) and yield stress of steel (Fy) are assumed as random variables. On the contrary, the load (P),
width (b) and depth (D) of the beam, the cover of reinforcements (t) are modeled as UBB type
variables.
The bounds on reliability indices are evaluated following section 5.2 and are depicted in Fig. 5.1
assuming P, B, D and t as UBB variables and Fy and fck as normal random variables. The radiuses
of bounds are expressed as the percentage of the nominal property values and the coefficient of
variation (COV) of random variables are taken as 0.2. The nominal value of P is considered as
10 KN. The reliability is also computed assuming that the interval variables are uniformly
distributed. These are now transformed to equivalent normal distribution using Eqn. (3.8 & 3.9
of chapter 3). It can be noted that the reliability index corresponds to the uniform distribution i.e.
the most conservative assumption is within the width of the reliability bounds. Thus, the
conservative bounds obtained for structural responses as reported in Qiu and Wang (2003) is also
4.90
Upper Bound
4.85 Lower Bound
Uniform
Reliability Index
4.80
4.75
4.70
4.65
4.60
5 10 15 20 25
x(%)
Fig. 5.1: The variation of reliability index with varying radius of bounded variables.
Now keeping interval length expressed as percentage of mean value to be same i.e. 10%, the
variation of reliability index with increasing COV (more uncertainty) is shown in Fig. 5.2. The
Reliability Index 7
2
10 15 20 25 30 35
COV (%)
Fig. 5.2: The variation of reliability index with increasing COV for P=10KN and Δx = 10%
Again keeping COV of random and UBB variables to be 20%, the bounds of reliability index as
well as reliability index based on uniform assumption are shown in Fig. 5.3 with increasing value
of nominal load. The value of interval length expressed as percentage of mean is taken as 0.1 to
5.0
Upper Bound
Lower Bound
4.5
Uniform
Reliability Index
4.0
3.5
3.0
2.5
2.0
0 10 20 30 40 50 60 70 80 90
Load (KN)
Fig. 5.3: The variation of reliability index with increasing value of nominal load for
Δx =10% and COV=20%.
A reinforced concrete cast in situ bored pile embedded in clay, subjected to horizontal and
vertical load as well as moment at tip of the pile as shown in Fig. 3.25 of chapter 3 is taken up to
elucidate the proposed approach of safety evaluation of hybrid uncertain system. The statistical
properties of the random and UBB variables are described in table 3.3 in chapter 3. The pile
diameter (X1), modulus of elasticity(X2), applied horizontal load (X11), vertical load (X13) and
moment (X12) at tip of the pile are considered to be random variables. The length of pile in
various soil layers (X3 to X6) and the corresponding values of modulus of sub-grade reaction (X 7
The bounds of reliability indices are evaluated following section 5.2 considering the limit state
function for maximum horizontal displacement at tip as described by Eqn. (3.42) of chapter 3.
The radiuses of bounds are expressed as the percentage of the nominal property values and COV
of random variables are taken as 0.1. The nominal value of load is considered to be 30 KN to
develop this plot. The reliability is also computed assuming that the interval variables are
uniformly distributed. The reliability results are presented in Fig. 5.4 with increasing value of
5.70
5.65
5.60
Fig. 5.4: The variation of reliability index with increasing value of interval length.
Keeping value of interval length (Δx) and COV of random and UBB variables to be 0.1, the
bounds of reliability index as well as reliability index based on uniform assumption are shown in
6.0
Upper Bound
5.5
Lower Bound
Uniform
5.0
R eliab ility In d ex
4.5
4.0
3.5
3.0
2.5
2.0
20 30 40 50 60 70 80 90 100 110
Load (KN)
Fig. 5.5: The variation of reliability index with increasing horizontal load for Δx =10% and
COV=10%.
The variation of reliability index with increasing COV is shown in Fig. 5.6. Here interval length
and nominal value of load is kept same i.e. 0.1 and 30KN respectively to develop this plot.
6.0
Upper Bound
5.5 Lower Bound
Uniform
R e lia b ility In d e x
5.0
4.5
4.0
3.5
3.0
2.5
Fig. 5.6: The variation of reliability index with increasing COV for P=30KN and Δx = 10%
The reliability analysis is performed by considering the limit state function for maximum
bending moment as described by Eqn. (3.43). The bounds on reliability indices with increasing
value of interval length are evaluated following section 5.2 and are depicted in Fig. 5.7. Here
COV of random and UBB variables are assumed to be 0.1. The nominal value of load is
considered as 30 KN to develop this figure. The reliability is also computed assuming that the
interval variables are uniformly distributed and is shown in the same plot.
7.1
Upper Bound
Lower Bound
7.0 Uniform
6.8
6.7
6.6
5% 10% 15% 20% 25%
x(%)
Fig. 5.7: The variation of reliability index with increasing value of interval length.
Keeping value of interval length (Δx) and COV of random and UBB variables to be 0.1, the
bounds of reliability index as well as reliability index based on uniform assumption are shown in
Again, keeping interval length and nominal value of horizontal load to be same i.e. 0.1 and
30KN respectively, the variation of reliability index with increasing COV is shown in Fig. 5.9.
7.0 Upper Bound
Lower Bound
6.5 Uniform
6.0
5.0
4.5
4.0
3.5
20 30 40 50 60 70 80 90 100 110
Load (KN)
Fig. 5.8 The variation of reliability index with increasing horizontal load for Δx =10% and
COV=10%.
5.25
4.50
3.75
3.00
Fig. 5.9: The variation of reliability index with increasing COV for P=30KN and Δx = 10%
5.4 Reliability Analysis of System Characterized by Uncertain Fuzzy and Random
Variables
The concept of various transformation methods are applied in this regard. However, there are
always some chances to lose some of the information at transformation stage. The reliability
analysis format as proposed here does not require any such transformation of data. The
convenient way of representing the fuzzy variables by the α-discretization method of its
possibility distribution i.e. the membership function is used. To circumvent the disadvantages of
the multilevel α-discretization method, the interval analysis approach is used. The fuzzy solution
optimization problem of finding the least favourable and the most favourable performance
response obtained by using the bounds of all fuzzy input parameters at the particular α-
discretization. Subsequently, the lower and the upper bounds of the reliability at particular α-
discretization level are computed using the conventional second moment reliability analysis
format. The procedure of reliability evaluation is performed for each α-cut level to obtain the
membership distribution of fuzzy reliability. The defuzzified value of the fuzzy reliability is
finally achieved by identifying its centre of gravity according to the real axis. The computational
procedure of the possibilistic distribution of the reliability defined by the membership function
under hybrid uncertainty is elucidated numerically to study the consistency of the results
The evaluation of reliability bounds as discussed in the previous section can be applied
for computation of the distribution of the membership function of the reliability in the framework
of α-discretization approach. For the sake of completeness, the fundamentals of fuzzy description
L U
3.3). For ith α-cut level, the upper and lower bounds f and f as depicted in Fig. 3.3 (a) of
fL min f : f F
(5.9)
fU min f : f F
Now, for a given fuzzy variable, if n sets of α- cuts are established for all the fuzzy quantities
of the variable, a 2 by n array can be used to represent all of the quantities as follows:
F ( f L , f U )1, ( f L , f U )2, , ( f L , f U ) n (5.10)
For given support description of a fuzzy variable F, one can thus obtain the two interval
values at any specified α-cut level, μ( F ) , as shown in Fig. 3.3(b),
F ( fL , fU ) [ c (1 )a, c (1 )b] (5.11)
In this regard it is of worth mentioning that the possibility distributions as described by the
respective membership function are resulted from the overlapping interval observations. The
possibilistic mathematics generalizes the interval analysis (Cliff 1997). Once the fuzzy quantities
are expressed in interval form, the fuzzy arithmetic can be carried out using interval operations at
each of the ‘n’ α-cut levels independently. Thus, the core problem of the calculation of the
results of a function of fuzzy variables can be envisaged as the calculation of the result of that
The presents study is intended to perform the reliability analysis of structure when some
variables are random, defined by their pdf, and some are fuzzy defined by the associated
membership functions. It is apparent that to perform the safety evaluation of structures by the
conventional second moment format all the parameters must be random. Any fuzzy variable
involved in the limit state functions of the reliability analysis problem is converted to a bounded
variable at ith α-cut level following the procedure as described previously in section 5.2. Once a
fuzzy variable is converted to a bounded variable at i th α-cut level, the bounds on the reliability
index and the associated probability of failure can be readily obtained following second moment
reliability analysis format. If some of the random variables are not normal, those can be further
transformed to equivalent normal variables, using Eqn. (3.8 & 3.9 of chapter 3). However it
should be noted that the procedure of reliability evaluation has to be performed for each α-cut
level to obtain the distribution of upper and lower bound reliability in the entire range. Thus, the
proposed algorithm can be applied to evaluate the membership distribution of fuzzy reliability of
5.6 Defuzzification
The defuzzification is a process by which the collection of membership values are reduced to a
single scalar quantity-presumably to be the most typical or representative value. There are
various forms to convert the fuzzy sets to crisp sets (Ross 1997). In present study, the defuzzified
value of the reliability index is computed from the following (Lallemand et al. 1999):
n
AC i ( ALi ARi )
i 1
AD n (5.12)
1 2 i
i 1
Where AD is the defuzzified value of A and A C is the crisp value of A. ALi and ARi are lower and
upper bound value of A at αi, where, αi, is the i-th α-level cut and n is the number of α-level cut.
A simply supported reinforced concrete beam of section 3.5.1 of chapter 3 (Ref. Fig. 3.5 of
chapter 3) has been considered again to elucidate the proposed algorithm of reliability analysis of
structures under hybrid uncertainties. The present study is intended for the reliability evaluation
of systems which are characterized by both the probabilistic and the possibilistic (fuzzy type)
uncertain variables. The uncertain fuzzy variables of the numerical example considered here are
P, B, D and t which are described by the triangular fuzzy distribution as shown in Fig. 3.6. The
two random variables (Fy and fck) are assumed to be normal. From Fig. 3.6 for i-th fuzzy variable,
a xi w xi , c xi , b xi w xi
where, xi , xi is the mean and standard deviation of the ith variable, respectively, and w defines
Each fuzzy variable is converted to a bounded variable at ith α-cut level following section 5.2.
Once fuzzy variables are converted to a bounded variable at ith α-cut level, the bounds on the
reliability index and the associated probability of failure is readily obtained following standard
second moment reliability analysis format. The reliability indices description (upper bound and
lower bound) with the increasing membership values (α) are depicted in Fig. 5.10 and 5.11 for
nominal value of load (P) as 10 KN and 50 KN, respectively. The COV required to define the
support width of each fuzzy variable is taken as 10%. The COV of normal random variables are
also taken as 10%. Again Fig. 5.12 and 5.13 show the fuzzy membership distribution of
reliability for load P=10 KN and 50 KN, respectively, but for 15% cov of the support width of
0.8
M em bersh ip 0.6
0.4
0.2
0.0
9.0 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8 9.9 10.0
Reliability Index
0.6
0.4
0.2
0.0
Membership 0.6
0.4
0.2
0.0
Fig. 5.12: The distribution of reliability index (COV =15%, P=10 KN)
0.6
0.4
0.2
0.0
1 2 3 4 5 6 7
Reliability Index
Fig. 5.13: The distribution of reliability index (COV =15%, P=50 KN)
The variation of deffuzzified reliability index obtained from Eqn. (5.12) with increasing
uncertainty level and different load is shown in Fig. 5.14. The support width as defined by the
3
5 10 15 20 25
COV(%)
Fig. 5.14: The variation of defuzzified reliability index with increasing COV for different loads
The distribution of reliability index for different support widths having same COV is depicted in
Fig. 5.15. Fig.5.16 further shows the variation of defuzzified reliability index for varying support
0.8
0.6
0.4
0.2
0.0
Fig. 5.15: The distribution of reliability index for different support width.
14
COV=0.1 COV=0.15
12 COV=0.20 COV=0.25
10
Defuzzified 8
0
1 2 3 4 5 6
Support width (w)
Fig. 5.16: The variation of defuzzified reliability index for different support width and
COV.
The same example as taken in section 5.3.2 is re considered here to elucidate the proposed
approach of safety evaluation of hybrid uncertain system. The statistical properties of the random
and fuzzy variables are described in table 3.3 of chapter 3. The pile diameter (X 1), modulus of
elasticity(X2), applied horizontal load (X11), vertical load (X13) and moment (X12) at tip of the pile
are considered to be random variables. The length of pile in various soil layers (X 3 to X6) and the
The variation of deffuzzified reliability index obtained from Eqn. (5.12) with increasing
uncertainty level and different load is shown in Fig. 5.17 considering the limit state function for
maximum horizontal displacement at tip as described by Eqn. (3.42). The support width as
defined by the weight factor w is taken as 3.0. The value of interval length expressed as
6.0 P=30KN
5.5
P=20KN
P=50KN
D efu zzified B eta
5.0
4.5
4.0
3.5
3.0
2.5
2.0
Fig. 5.17: The variation of defuzzified reliability index with increasing COV for different loads
for Δx =10%
The reliability indices description (upper bound and lower bound) with the increasing
membership values (α) are depicted in Fig. 5.18 and 5.19 for nominal value of load (P) as 80
KN. The COV required to define the support width of each fuzzy variable is taken as 10% and
15% respectively. The COV of normal random variables are also taken same as fuzzy variable
The distribution of reliability index for different support widths having same COV is depicted in
Fig. 5.20. The COV of random and fuzzy variables are assumed to be 10% to develop this plot.
1.0 Lower Bound
Upper Bound
0.8
M em b ersh ip 0.6
0.4
0.2
0.0
Fig. 5.18: The distribution of reliability index (COV =10%, P=80 KN).
0.8
M em bership
0.6
0.4
0.2
0.0
1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00
Reliability Index
Fig. 5.19: The distribution of reliability index (COV =15%, P=80 KN).
Support width =
1.0
Support width =
Support width=
0.8
M embership 0.6
0.4
0.2
0.0
Fig. 5.20: The distribution of reliability index for different support width for COV =10%.
The reliability analysis is performed by considering the limit state function for maximum
bending moment as described by Eqn. (3.43) of chapter 3. The reliability results i.e. the variation
of deffuzzified reliability index obtained from Eqn. (5.12) with increasing uncertainty level for
different load is shown in Fig. 5.21. The support width as defined by the weight factor w is taken
as 3.0. The value of interval length expressed as percentage of mean value is assumed to be 10%
Deffuzified Beta
5.5
5.0
4.5
4.0
3.5
3.0
2.5
Fig. 5.21: The variation of defuzzified reliability index with increasing COV for different loads
for Δx =10%
The reliability indices (upper and lower bound) with the increasing membership values (α)
are depicted in Fig. 4.22 and 4.23 for nominal value of load (P) as 80 KN. The COV required to
define the support width of each fuzzy and random variable is taken as 10% and 15%
respectively.
The distribution of reliability index for different support widths having same COV is depicted in
Fig. 4.24. Here COV for both random and fuzzy variables are assumed to be 10%.
1.0 Lower Bound Lower Bound
Upper Bound 1.0
Upper Bound
0.8
0.8
Membership
0.6
Membership
0.6
0.4
0.4
0.2
0.2
0.0
3.50 3.85 4.20 4.55 4.90 5.25
0.0
Reliability Index 1.5 1.8 2.1 2.4 2.7 3.0 3.3 3.6
Reliability Index
0.6
0.4
0.2
0.0
Fig. 5.24: The distribution of reliability index for different support width for COV =10%.
5.8 Summary
A second moment safety evaluation format for structural system characterized by hybrid
uncertainties i.e. both probabilistic and interval type uncertain parameters has been presented in
this chapter. The available literatures on the application of interval analysis approach in the
framework of set theoretical description is limited to deal with response evaluation and
problem. The approach is extended here to reliability analysis problem with an objective of
evaluation of reliability of structures involving both random and bounded variables. The bounds
of performance function (limit state function) of a typical structural reliability analysis problem
with UBB parameters are presented first by making use of interval mathematics. Subsequently,
the application of the approach to fuzzy variables has been elaborated next.
The example of a reinforced concrete beam and a R.C. pile subjected to horizontal, vertical load
and moment at tip are considered to validate the efficiency of proposed safety evaluation
approach. Parametric studies are performed for different settings of COV of input and increasing
numerical examples it has been observed that the lower bounds of reliability obtained by the
interval analysis approach are smaller than those predicted by assuming that the UBB variables
are uniformly distributed probabilistic variables. Similarly, the upper bounds yielded by the
present interval method are larger than those obtained by the probabilistic assumption. Thereby,
the reliability of structures with the UBB parameters obtained by the interval analysis method
contains those obtained based on the most conservative uniform pdf assumption. This kind of
results is in consistent with the membership distribution of fuzzy reliability. In other words, it
ensures that the possibility distribution is always greater than the probability distribution. The
fuzzy reliability distribution and the defuzzified reliability index for different load levels or
uncertainty levels as obtained by the proposed approach are also found to be consistent.
However, the present analysis is based on the assumption of monotonic responses i.e. response
lies at the end points of the UBB parameters. If the uncertain parameters are not so, alternative
method should be applied. It may be noted that the proposed approach is based on the first order
Taylor series expansion. Consequently, as the uncertainty level of the parameters becomes
larger, the predicted fuzzy membership distribution of reliability will be increasingly inaccurate.
A second order perturbation analysis may be explored to improve the accuracy. However, it