You are on page 1of 18

1 23

Acta Mechanica

ISSN 0001-5970
Volume 223
Number 1

Acta Mech (2012) 223:109-124
DOI 10.1007/s00707-011-0546-3
System reliability analysis of spatial
variance frames based on random field
and stochastic elastic modulus reduction
method
Lu Feng Yang, Bo Yu & J.Woody Ju
1 23
Your article is protected by copyright and
all rights are held exclusively by Springer-
Verlag. This e-offprint is for personal use only
and shall not be self-archived in electronic
repositories. If you wish to self-archive your
work, please use the accepted authors
version for posting to your own website or
your institutions repository. You may further
deposit the accepted authors version on a
funders repository at a funders request,
provided it is not made publicly available until
12 months after publication.
Acta Mech 223, 109124 (2012)
DOI 10.1007/s00707-011-0546-3
Lu Feng Yang Bo Yu J. Woody Ju
System reliability analysis of spatial variance frames
based on random eld and stochastic elastic modulus
reduction method
Received: 5 April 2011 / Published online: 4 October 2011
Springer-Verlag 2011
Abstract This paper presents the stochastic elastic modulus reduction method for system reliability analysis
of spatial variance frames based on the perturbation stochastic nite element method (PSFEM) and the local
average of a random eld. The stochastic responses and reliability index of each element of a structural frame
are characterized by the PSFEM and the rst-order second-moment method, to properly handle the correlation
structures and scale of uctuation of randomelds. Astrategy of elastic modulus adjustment for the estimation
of systemreliability is developed to determine the range and magnitude of elastic modulus reduction, by taking
the element reliability index as a governing parameter. The collapse mechanism and system reliability index
of a stochastic framed structure are determined through iterative computations of the PSFEM. Compared with
the failure mode approaches in traditional system reliability analysis, the proposed method avoids two major
difculties, namely the identication of significant failure modes and estimation of the joint probability of
failure modes. The inuences of the correlation structure and scale of uctuation of the random eld upon
system reliability are investigated to demonstrate the accuracy and computational efciency of the proposed
methodology in system reliability analysis of spatial variance frames.
1 Introduction
It has long been recognized that a fully satisfactory estimation of the reliability of a structure must be based
on a system approach. However, the system reliability analysis of the spatial variance structure presents two
major difculties [1]. One is the identication of significant failure modes of a redundant structure; the other
is the estimation of joint failure probability contributed from the significant failure modes. It is very difcult
to enumerate all collapse modes of a statically indeterminate structure that has many possible modes or paths
to cause failure. However, among these possible failure modes or paths, only some contribute considerably
to the structural system failure probability while others have low probability of occurrence. To identify these
probabilistically dominant failure modes or paths, many methods have been developed, such as the incremental
L. F. Yang supported by the Guangxi Lab Center of Science and Technology (LGZX201002).
L. F. Yang (B) B. Yu
Key Laboratory of Disaster Prevention and Structural Safety of Ministry of Education,
School of Civil Engineering and Architecture, Guangxi University, Nanning 530004, China
E-mail: lfyang@gxu.edu.cn
Tel.: +86-771-3236827
J. W. Ju (B)
Department of Civil and Environmental Engineering, University of California, Los Angeles, CA 90095, USA
E-mail: juj@ucla.edu
J. W. Ju
Chang-Jiang Scholar Chair Professor, Guangxi University, Nanning, China
Author's personal copy
110 L. F. Yang et al.
load approach [2], the beta-unzipping method [3], the truncated enumeration method [4], the mathematical
programming technique together with the Monte Carlo simulation [5], etc. On the other hand, the evaluation of
joint failure probability is often intractable even if the potential failure modes can be identied. The numerical
integration methods [6, 7] and several approximate methods, such as the point estimation methods [8] and
various bounding techniques [9, 10], have been developed in the last three decades.
To date, the system reliability analysis has not been widely used in the safety evaluation of complex engi-
neering structures, as there still remain many pending problems to be solved. For example, nearly all failure
mode approaches in existing system reliability analysis describe the structural parameters and external loads
as random variables, ignoring the spatial variances of these parameters, thus introducing additional modeling
errors. Therefore, a random eld should be employed to model and describe random quantities embedded in
the stochastic structures.
Over the past three decades, several methods for the discretization of a random eld have been developed
[1113]. Among these, the local average method (LAM), which is easy to implement and insensitive to the
types of correlation structures, has been widely adopted in various stochastic structural analyses [13, 14].
Typically, the second-order perturbation technique and stochastic variational principle-based perturbation sto-
chastic nite element method (PSFEM) [15, 16] combined with the LAMprovide a powerful tool for stochastic
analysis and reliability evaluation of stochastic structures with small variations.
The plastic limit analysis approach is one of the important bases for structural system reliability analysis.
In recent years, the plastic limit analysis method based on various elastic modulus adjustment procedures has
made considerable advance [1722]. In particular, the elastic modulus reduction method (EMRM) for limit
analysis and safety evaluation of complex structures was developed in recent years [2022]. In the EMRM,
the element bearing ratio (EBR) and its degree of uniformity are dened to describe the loading state of each
element as well as the redistribution of internal forces in a structure. We also refer to the elastic-damage
stiffness reduction mechanisms and damage mechanics developed by Ju [23, 24], Ju and Lee [25] as well as Ju
et al. [26]. On the basis of linear elastic nite element analysis, numerous sets of statically admissible stress
distributions can be generated. This method was applied to stochastic structures for system reliability analysis
of frame structures while random quantities are still described as random variables [27].
In this paper, key randomquantities are modeled as randomelds which are discretized by the local average
method (LAM); the PSFEM is employed to compute stochastic responses while the element reliability index
of the frame structures is produced by the rst-order second-moment (FOSM) method. A strategy of elastic
modulus reduction is developed for stochastic structures by dening the degree of uniformity of reliability
index (DURI) as well as the reference reliability index (RRI), such that the stochastic elastic modulus reduction
method (SEMRM) is deployed for the estimation of system reliability of spatial variance structures through
iterative analysis. Compared with traditional failure mode approaches in existing system reliability analysis,
the proposed methodology avoids the difculties in the identication of significant failure modes and the
evaluation of joint probability of failure of those significant failure modes. Several numerical examples are
presented to demonstrate the accuracy and computational efciency of the proposed framework.
2 The local average method for discretization of a random eld
Let us start by considering a continuous wide-sense homogeneous isotropic random eld {X(x, )} with
multiple components, where is the outcome of the random parameter. Here, x denotes the coordinate vector.
The values of the mean and variance of {X(x, )} are specied as
_

X
_
=
_

X
1
,

X
2
, . . . ,

X
M
_
T
;
_

2
X
_
=
_

2
X
1
,
2
X
2
, . . . ,
2
X
M
_
T
, (1)
where

X
i
,
X
i
and
2
X
i
denote the mean, standard deviation and variance of the i th component of the random
eld {X(x, )}. There are M components in total. For a given spatial point x
0
, {X(x
0
, )} is the randomvector
containing M random variables. Conversely, for a given outcome
0
, {X(x,
0
)} represents a realization of
the random eld.
A frame component forms an angle of with the x-axis, as illustrated in Fig. 1. The random parameters
concerning the material, geometry and load can be modeled as a multicomponent randomeld with two dimen-
sions along the planar straight bar component. According to the relationship between the polar coordinate r
and the Cartesian coordinates x and y, the randomeld {X(x, )} can be denoted as {X(r, )} along the straight
component.
Author's personal copy
System reliability analysis of spatial variance frames 111
o
L
e
y=x tan
y
x

x=y/ tan
r
r

Fig. 1 A planar element of a frame structure


We assume that the random eld is discretized into N elements; the local average of {X(r, )} over the
element e with the length L
e
yields a set of random variables:
{X}
e
=
_
X
e
1
, X
e
2
, . . . , X
e
M
_
T
, (2)
where X
e
i
denotes the local average of the i th component of {X(r, )}, X
i
(r, ), over the element e, which
can be dened as:
X
e
i
=
1
L
e
_
L
e
X
i
(r, ) dr; e = 1, 2, . . . , N, i = 1, 2, . . . , M. (3)
The mean and variance of the local average X
e
i
are:
E
_
X
e
i
_
=

X
i
, (4)
Var
_
X
e
i
_
= E
_
_
X
e
i


X
i
_
2
_
=
i i
(L
e
)
2
X
i
, (5)
where
i i
(L
e
) represents the variance function of X
i
(r, ) which can be dened as [14]

i i
(L
e
) =
2
L
e
_
L
e
_
1
r
L
e
_

i i
(r)dr; (6)
here,
i i
(r) is the autocorrelation function of X
i
(r, ).
For random variables X
e
i
and X
e
j
evaluated by local averaging over the element e, the covariance of the
two variables is [14]
Cov
_
X
e
i
, X
e
j
_
=
X
i

X
j

i j
(L
e
), (7)
where
i j
(L
e
) is the variance function of X
i
(r, ) and X
j
(r, ):

i j
(L
e
) =
2
L
e
_
L
e
_
1
r
L
e
_

i j
(r)dr, (8)
where
i j
(r) represents the correlation function between X
i
(r, ) and X
j
(r, ).
For random variables X

i
and X

j
achieved by local averaging of the i
th
and j
th
components X
i
(r, ) and
X
j
(r, ) over two arbitrarily situated planar linear elements and , respectively, as exhibited in Fig. 2, the
covariance of X

i
and X

j
can be dened as follows [14]:
Cov
_
X

i
, X

j
_
=

X
i

X
j
L

_
L

_
L

i j
(r)dl ds. (9)
Author's personal copy
112 L. F. Yang et al.
y
x
y
k
l
s
0
0
o
y
p
y
p
y
k
x
p
x
k
x
p
x
k
A
L

L
B

Fig. 2 Two arbitrarily situated planar elements and


Here, L

and L

are the lengths of elements and ; r represents the distance from point A with the coor-
dinates (x

, y

) located in the element to point B with the coordinates


_
x

, y

_
located in the element .
If the elements and are not parallel to the y-axis, one has
y

= y
p
+
_
x

x
p
_
tan
0
; y

= y
p
+
_
x

x
p
_
tan
0
, (10a)
where
tan
0
=
y
k
y
p
x
k
x
p
; tan
0
=
y
k
y
p
x
k
x
p
. (10b)
Here, the coordinates
_
x
p
, y
p
_
, (x
k
, y
k
),
_
x
p
, y
p
_
and
_
x
k
, y
k
_
denote the ends of the elements
and , respectively. Accordingly, we arrive at
ds = dx

/cos
0
; dl = dx

/cos
0
. (11)
The distance r between points A and B is
r =
_
__
x

x
p
_
tan
0

_
x

x
p
_
tan
0
+ y
p
y
p
_
2
+
_
x

_
2
. (12)
Substitution of Eq. (12) into Eq. (9) leads to
Cov
_
X

i
, X

j
_
=

X
i

X
j
L

cos
0
cos
0
_
L

_
L

i j
(r) dx

dx

. (13)
If both the elements and are parallel to the y-axis, the distance r between points A and B is
r =
_
_
y

_
2
+
_
x
p
x
p
_
2
. (14)
Then, the covariance of X

i
and X

j
gives
Cov
_
X

i
, X

j
_
=

X
i

X
j
L

_
L

_
L

i j
(r)dy

dy

. (15)
Further, if the element is parallel to the y-axis, but the element is not, one expresses the distance r as
r =
_
_
y
p
+
_
x

x
p
_
tan
0
y

_
2
+
_
x

x
p
_
2
. (16)
The covariance of X

i
and X

j
then takes the form
Cov
_
X

i
, X

j
_
=

X
i

X
j
L

cos
0
_
L

_
L

i j
(r) dx

dy

. (17)
Author's personal copy
System reliability analysis of spatial variance frames 113
Similarly, if the element is parallel to the y-axis, while the element is not, the distance r becomes
r =
_
_
y


_
y
p
+
_
x

x
p
_
tan
0
__
2
+
_
x
p
x

_
2
. (18)
Accordingly, the covariance of X

i
and X

j
reads
Cov
_
X

i
, X

j
_
=

X
i

X
j
L

cos
0
_
L

_
L

i j
(r) dy

dx

. (19)
All the random variables X
e
i
(e = 1, 2, . . . , N; i = 1, 2, . . . , M), as displayed in Eq. (2), can be assembled
into a vector:
{X

} =
_
X

1
, X

2
, . . . , X

n
_
T
, (20)
where n = M N denotes the number of random variables resulting from discretization of the random elds.
3 The stochastic nite element method
The functional of potential energy of a linear structural system of continuum reads [15]:

=
1
2
_
V
{}
T
[D]{}dV
_
S

{u}
T
{F}dS, (21)
where {} is the strain vector, [D] is the elasticity matrix, {u} is the displacement vector, {F} denotes the bound-
ary traction vector, and V denotes the domain of interest. In addition, S signies the boundary of the domain V,
which can be simply divided into two parts: the traction boundary S

and the prescribed displacement boundary


S
u
.
Based on the second-order perturbation techniques [15], one can obtain the zeroth-, rst- and second-order
expanded expressions, respectively, of the functional of potential energy:

=
1
2
_
V
{ }
T
_

D
_
{ }dV
_
S

{ u}
T
_

F
_
dS, (22a)

i
=
1
2
_
V
_
{ }
T
[D]
i
{ } + 2{ }
T
_

D
_
{}
i
_
dV
_
S

_
{ u}
T
{F}
i
{u}
T
i
_

F
_
_
dS, (22b)

i j
=
1
2
_
V
_
{ }
T
[D]
i j
{ } + 2{}
T
i
_

D
_
{}
j
+ 2{ }
T
[D]
i
{}
j
+ 2{ }
T
[D]
j
{}
i
+2{ }
T
_

D
_
{}
i j
_
dV
_
S

_
{ u}
T
{F}
i j
+ {u}
T
i j
_

F
_
+ {u}
T
i
{F}
j
+ {u}
T
j
{F}
i
_
dS, (22c)
where i, j = 1, 2, . . . , n. Moreover, ( ) signies the value of () evaluated at the mean of the random eld;
()
i
denotes the rst-order partial derivatives of () with respect to the i th component of the random eld
X

i
, evaluated at the mean of the random eld; ()
i j
signies the second-order partial derivatives of () with
respect to X

i
and X

j
, evaluated at the mean of the random eld.
According to the stochastic variational principle [16], only the highest order expansion of the functional,
that is
i j
herein, is necessary and sufcient to derive all governing equations of the perturbation stochastic
nite element method (PSFEM):
_

K
_
{ a} = {

P};
_

K
_
{a}
i
= {P}
i
[K]
i
{ a};
_

K
_
{a}
II
= {P}
II
, (23)
Author's personal copy
114 L. F. Yang et al.
where
{a}
II
=
1
2
n

i =1
n

j =1
{a}
i j
Cov
_
X

i
, X

j
_
, (24a)
{P}
II
=
1
2
n

i =1
n

j =1
_
{P}
i j
2[K]
i
{a}
j
[K]
i j
{ a}
_
Cov
_
X

i
, X

j
_
. (24b)
Here, [K], {a} and {P} denote the structural stiffness matrix, nodal displacement vector and the equivalent
nodal force vector, respectively.
Once the nodal displacement vectors { a}, {a}
i
and {a}
II
are obtained from Eqs. (23) and (24), the mean
and variance of the internal force vector {P}
e
of the element e can be expressed as:
E
_
{P}
e
_
=
_

K
_
e
_
{ a}
e
+ {a}
e
II
_

_

P
d
_
e
{P}
e
II
+
1
2
n

i =1
n

j =1
{P}
e
i j
Cov(X

i
, X

j
), (25)
Var
_
{P}
e
_
=
n

i =1
n

j =1
Cov
_
X

i
, X

j
_ _
_

K
_
e
{a}
e
j
+ [K]
e
j
{ a}
e
{P
d
}
e
j
_
T
_
_

K
_
e
{a}
e
i
+ [K]
e
i
{ a}
e
{P
d
}
e
i
_
,
(26)
where
_

P
d
_
e
and {P
d
}
e
i
are the equivalent nodal force vector and its rst-order partial derivatives with respect
to X

i
, evaluated at the mean {

X

}.
4 The SEMRM for system reliability analysis
In this section, the stochastic elastic modulus reduction method (SEMRM) is adopted for system reliability
evaluation of structural frames. According to probabilistic structural analysis [1, 3, 27], it is not necessarily
true that only the element with the lowest reliability index would fail at each stage during the failure process
for a stochastic structural system. However, it is inevitable that the elements with reliability indices lower
than a certain threshold will be damaged, and one or more of those elements would fail. Consequently, the
distribution of internal forces keeps changing such that the redistribution of internal forces often occurs in the
process from damage to failure for a ductile structural system.
Here, the elastic moduli of elements with reliability index lower than the reference value will be reduced in
order to simulate the structural damage and the redistribution of internal forces in a stochastic structural system
through an iterative analysis. In each iterative step, the reliability index of each element can be approximately
estimated by the rst-order second-moment (FOSM) method as follows:

i
=

R
i

S
i
_

2
R
i
+
2
S
i
, (27)
where
S
i
and
2
S
i
are the mean and variance of load effects of the element i achieved by the PSFEM and the
LAM. Moreover,
R
i
and
2
R
i
are the mean and variance of resistance of the element i , which can be evaluated
according to the material and geometric parameters of the element i .
The reference reliability index (RRI),
0
k
, for the elastic modulus reduction in the kth iteration can be
dened as

0
k
=
min
k
+
_

max
k

min
k
_
d
k
, (28)
where d
k
signies the degree of uniformity of reliability index (DURI) in the kth iteration:
d
k
=

k
+
min
k

k
+
max
k
(29)
in which
max
k
,
min
k
and

k
denote the maximum, minimum and the average of the elements reliability
indices in the kth iteration, respectively.
Author's personal copy
System reliability analysis of spatial variance frames 115
The elastic modulus of the element with reliability index lower than the RRI should be reduced in the kth
iteration according to the following expression:
E
e
k+1
=

E
e
k
_

e
k

0
k
_
q
for
e
k
<
0
k
E
e
k
for
e
k

0
k
(30)
where
e
k
denotes the reliability index of the element e in the kth iteration; E
e
k
and E
e
k+1
signify the elastic
moduli of the element e in the kth and the next iteration, respectively. Further, q represents the adjustment
factor for the reduction of elastic modulus and may take an arbitrary value in the closed interval [0.1, 3.0], as
demonstrated later in our numerical examples. The iterative procedure will stop when d
k
meets the following
criterion for convergence:
|d
k
d
k1
|
d
k
; k 2, (31)
where represents the prescribed admissible error. In what follows, we adopt 1.0 10
6
for this value.
If the criterion for numerical convergence described in Eq. (31) is satised after N
f
iterations, it indicates
that the entire structural system is at the plastic limit state and may fail as soon as a single element fails. Since
the potential failure element is limited to the elements with reliability indices lower than the RRI, the failure
interaction of the frame with n elements can be modeled by a series system with M
1
elements, where M
1
denotes the numbers of the potential failure elements. Therefore, the probability of failure of the structural
system can be dened as:
P
f s
= P
_
U
1
U
2
U
M
1
_
, (32)
where U
i
(i = 1, 2, . . . , M
1
) denes the event that the element i would fail. The system reliability index can
now be derived from P
f s
as follows:

S
=
1
(P
f s
), (33)
where
1
() is the inverse standard normal cumulative distribution function.
Many methods were developed for estimating the joint probability of failure in Eq. (32), including the
rst-order bounding method (FOBM) [28], the second-order bounding method (SOBM) [29], the third-order
bounding method (TOBM) [30], the equivalent correlation coefcient method (ECCM) [8] and the probabilis-
tic network evaluation technique (PNET) [31]. Each one of these methods can be employed to characterize the
joint probability of failure of the structural frame; then, the system reliability index can be derived by Eq. (33)
accordingly.
5 Numerical examples
5.1 Portal frame
The geometry and loading conditions of a portal frame are illustrated in Fig. 3, with the elastic modulus of
2.0 10
5
MPa and carrying the stochastic external loads F
1
and F
2
. The black dots denote the locations of
potential plastic hinges, and the numbers in the circles signify the serial numbers of elements. The cross-
sectional area and moment of inertia of beam and column elements are modeled as random variables and
represented by A and I , respectively. The plastic resistances of elements are described as random variables
and denoted by M
b
and M
c
, respectively. The statistical characteristics of all the random variables are listed
in Table 1.
The proposed SEMRM is employed to calculate the system reliability of the stochastic frame structures.
The serial numbers of the potential failure elements and their probabilities of failure as well as the element
reliability indices are listed in Table 2, in which the adjustment factor q takes the value of 0.6.
In the process of the iterative computation of the SEMRM, the redistribution of reliability indices is sim-
ulated by adjusting the elastic modulus of each element, as exhibited in Fig. 4. It is apparent that the initial
reliability indices of element 1 and element 2 are higher than those of the others and keep decreasing during
the iterations. By contrast, the initial reliability indices of element 3 and element 4 are lower than those of the
Author's personal copy
116 L. F. Yang et al.
1
2 3
4
F1
F2
3
0
0
0
m
m
3000mm 3000mm
x
y
o
Fig. 3 The geometry and loading conditions of a portal frame
Table 1 Statistical characteristics of random variables
Variable Mean value Coefcient Distribution
of variation type
A 1.3500 10
5
mm
2
0.05 Normal
I 2.2781 10
9
mm
4
0.05 Normal
M
b
3.7969 10
9
MPa 0.05 Normal
M
c
4.3664 10
9
MPa 0.05 Normal
F
1
3.3000 10
6
N 0.10 Normal
F
2
2.5000 10
6
N 0.10 Normal
Table 2 Reliability index and failure probability of potential failure element
Element Reliability Failure probability
index
1 3.091851 9.945633 10
4
2 3.091984 9.941178 10
4
3 3.091835 9.946169 10
4
4 3.091892 9.944260 10
4
0 10 20 30 40
1
2
3
4
5
6
7
Iteration step
E
l
e
m
e
n
t

r
e
l
i
a
b
i
l
i
t
y

i
n
d
e
x
element 1
element 2
element 3
element 4
Fig. 4 The iterative process of element reliability index
others but keep increasing during the iterations. The reliability indices of all elements are adjusted to the same
level in about 20 steps of iterations; this behavior shows that the system reliability of the frame depends on the
reliability level of all elements. Moreover, it implies that the collapse mechanism of the frame is an integral
failure mode.
In the process of iterations, the DURI increases gradually as displayed in Fig. 5 and converges to approx-
imately 1.0 in about 20 steps of iterations. This implies that the reliability indices of all elements in the frame
Author's personal copy
System reliability analysis of spatial variance frames 117
0 10 20 30 40
0.45
0.65
0.85
1.05
Iteration step
D
U
R
I
Fig. 5 The iterative process of the DURI
0 10 20 30 40
3.0
3.2
3.4
3.6
3.8
Iteration step
R
R
I
Fig. 6 The iterative process of the RRI
Table 3 System reliability index and failure probability of the portal frame by the SEMRM
Method Reliability index Failure probability
FOBM [2.654456, 3.091835] [9.946169 10
4
, 3.971815 10
3
]
SOBM [3.091835, 3.091851] [9.945633 10
4
, 9.946169 10
4
]
TOBM [3.091835, 3.091851] [9.945633 10
4
, 9.946169 10
4
]
ECCM 3.091851 9.945631 10
4
are adjusted to the same level when the structural failure occurs, as demonstrated in Fig. 4. The RRI decreases
along with the iterations, as shown in Fig. 6, and converges to approximately 3.09 in about 20 steps of iterations.
Based on Eq. (32) and the reliability of the potential failure elements listed in Table 2, the systemreliability
index as well as the joint probability of failure of the frame system can be estimated by means of the FOBM,
the SOBM, the TOBM and the ECCM. The corresponding results are listed in Table 3. We observe that the
system reliability index predicted by the SEMRM is approximately 3.09.
In order to validate the precision and computational efciency of the SEMRM, the traditional failure mode
approach (FMA) is employed to evaluate the system reliability of the frame. Eight significant failure modes of
the frame structure are identied according to the FMA as well as the limit state functions, reliability indices
and corresponding failure probabilities, as listed in Table 4. The system reliability index of the frame can
be obtained from the joint probability of the eight failure modes, which can be estimated by the FOBM, the
SOBM, the TOBM, the ECCM and the PNET. For comparison, the Monte Carlo simulation (MCS) technique
is also employed with 5million cycles. These systemreliability results are summarized in Table 5, which shows
Author's personal copy
118 L. F. Yang et al.
Table 4 Significant failure modes of the portal frame
Limit state function Reliability Failure probability
index
Z
1
= 4M
b
+ 2M
c
3000F
1
3000F
2
= 0 3.088169 1.006970 10
3
Z
2
= 2M
b
+ 4M
c
3000F
1
3000F
2
= 0 3.572171 1.770171 10
4
Z
3
= 2M
b
+ 2M
c
3000F
1
= 0 5.008424 2.743877 10
7
Z
4
= M
b
+ 3M
c
3000F
1
= 0 5.375582 3.816785 10
8
Z
5
= 4M
c
3000F
1
= 0 5.730992 4.992249 10
9
Z
6
= 4M
b
3000F
2
= 0 7.202758 2.950332 10
13
Z
7
= 3M
b
+ M
c
3000F
2
= 0 7.590942 1.587940 10
14
Z
8
= 2M
b
+ 2M
c
3000F
2
= 0 7.962255 8.446594 10
16
Table 5 System reliability index and failure probability of the portal frame by the FMA
Method Reliability index Failure probability
FOBM [3.039676, 3.088169] [1.006969 10
3
, 1.184165 10
3
]
SOBM [3.074761, 3.082958] [1.024771 10
3
, 1.053355 10
3
]
TOBM [3.077451, 3.077467] [1.043839 10
3
, 1.043897 10
3
]
ECCM 3.051625 1.138031 10
3
PNET 3.088084 1.007259 10
3
MCS 3.089402 1.002800 10
3
L2
L
1
L
1
F1
F2
F3
2F3
1
2
3 4
5
6
7 8
x
y
o
Fig. 7 The geometry and load conditions of a one-bay two-story frame
that the system reliability indices rendered by the TOBM and the ECCM are somewhat lower than those by
the SEMRM, while the PNET and the MCS achieve results agreeing well with those by the SEMRM. As the
MCS can be regarded as a benchmark, the accuracy of the proposed SEMRM is demonstrated by this portal
frame example.
5.2 One-bay two-story frame
Example 2 considers a one-bay two-story frame exhibited in Fig. 7, loaded by four stochastic external loads with
amplitudes F
1
, F
2
, F
3
and 2F
3
; their mean values and coefcients of variation are tabulated in Table 6. The
elastic modulus E, story height L
1
and span length L
2
are all deterministic: E = 2.110
5
MPa, L
1
=3,000mm
and L
2
=4,800mm. The beams cross-sectional area, moment of inertia and plastic resistance are all random
variables and represented by A
b
, I
b
and M
b
, respectively. Similarly, the columns cross-sectional area, moment
of inertia and plastic resistance are also randomvariables and represented by A
c
, I
c
and M
c
, respectively. Their
statistical characteristics are displayed in Table 6. The black dots in Fig. 7 denote the locations of potential
plastic hinges, and the numbers in the circles signify the serial numbers of elements.
In order to provide a benchmark to verify the computational effectiveness of the proposed SEMRM,
the traditional failure mode approach (FMA) is employed to estimate the system reliability of the frame.
Eight significant failure modes of the frame structure are identied according to the FMA. The limit state
Author's personal copy
System reliability analysis of spatial variance frames 119
Table 6 Statistical characteristics of random variables
Variable Mean value Coefcient Distribution
of variation type
A
b
1.2000 10
5
mm
2
0.10 Normal
A
c
1.2250 10
5
mm
2
0.10 Normal
I
b
1.6000 10
9
mm
4
0.10 Normal
I
c
1.2505 10
9
mm
4
0.10 Normal
M
b
3.6000 10
9
MPa 0.10 Normal
M
c
3.2156 10
9
MPa 0.10 Normal
F
1
1.6000 10
6
N 0.15 Normal
F
2
3.2000 10
6
N 0.15 Normal
F
3
1.3000 10
5
N 0.15 Normal
Table 7 Significant failure modes of the one-bay two-story frame
Limit state function Reliability Failure probability
index
Z
1
= 4M
b
0.5F
2
L
2
= 0 3.644054 1.341890 10
4
Z
2
= 4M
c
+ 2M
b
0.5F
2
L
2
= 0 5.352326 4.341540 10
8
Z
3
= 4M
c
+ 6M
b
0.5F
1
L
2
0.5F
2
L
2
4F
3
L
1
= 0 5.485902 2.056830 10
8
Z
4
= 4M
c
+ 3M
b
0.5F
2
L
2
3F
3
L
1
= 0 5.497299 1.928260 10
8
Z
5
= 6M
c
+ 2M
b
0.5F
2
L
2
3F
3
L
1
= 0 5.988532 1.058720 10
9
Z
6
= 8M
c
+ 2M
b
0.5F
2
L
2
4F
3
L
1
= 0 6.665395 1.319770 10
11
Z
7
= 4M
c
+ 4M
b
0.5F
1
L
2
4F
3
L
1
= 0 7.719356 5.845940 10
15
Z
8
= 4M
c
3F
3
L
1
= 0 9.006916 1.059660 10
19
Table 8 System reliability index and failure probability of the one-bay two-story frame by the FMA
Method System reliability index System failure probability
FOBM [3.643975, 3.644054] [1.341886 10
4
, 1.342297 10
4
]
SOBM [3.644005, 3.644025] [1.342038 10
4
, 1.342140 10
4
]
TOBM [3.643996, 3.644011] [1.342111 10
4
, 1.342191 10
4
]
ECCM 3.644089 1.341701 10
4
PNET 3.643975 1.342297 10
4
Table 9 Element reliability index with different adjustment factors
q Number Reliability Failure probability
index
0.5 7 3.644055 1.341880 10
4
8 3.644041 1.341953 10
4
1.5 7 3.644049 1.341911 10
4
8 3.644050 1.341906 10
4
2.5 7 3.644048 1.341917 10
4
8 3.644051 1.341901 10
4
function, reliability index and the corresponding probability of failure for each failure mode are summarized
in Table 7. Accordingly, the system reliability index of the frame and the probability of system failure can
be estimated by the FOBM, the SOBM, the TOBM, the ECCM and the PNET; these results are presented in
Table 8. It is observed that the system reliability index predicted by the traditional system reliability methods
is approximately 3.644.
It should be pointed out that the random parameters of the frame must be modeled as random variables in
the FMA. To facilitate the comparison of the system reliability results between the FMA and the SEMRM, the
assumption of random variables for all the random parameters remains effective here when adopting the SEM-
RM to evaluate the system reliability of the stochastic two-story frame structure example. The serial numbers
of potential failure elements and the probabilities of failure as well as the corresponding reliability indices are
presented in Table 9. Here, the adjustment factor q in Eq. (30) takes the values of 0.5, 1.5 and 2.5, respectively.
Author's personal copy
120 L. F. Yang et al.
0 5 10 15 20
2.5
4.5
6.5
8.5
Iteration step
E
l
e
m
e
n
t

r
e
l
i
a
b
i
l
i
t
y

i
n
d
e
x
element 1
element 2
element 3
element 4
element 5
element 6
element 7
element 8
Fig. 8 The iterative process of element reliability index
0 5 10 15 20
0.60
0.65
0.70
0.75
0.80
Iteration step
D
U
R
I
Fig. 9 The iterative process for the DURI
We observe that the potential failure elements are always element 7 and element 8 in three different numerical
cases; the respective reliability indices of the corresponding elements are approximately 3.644. It is clear that
the value of the adjustment factor q has little effect on the accuracy of the SEMRM.
In the process of numerical iterations for the SEMRM, the redistribution of element reliability indices of
the frame is simulated by adjusting the elastic modulus of each element, as illustrated in Fig. 8. Here, the
adjustment factor q takes the value of 2.0 without loss of generality. We observe that the initial reliability
indices of element 5 and element 6 are higher than those of other elements and continue decreasing during
subsequent iterations. By contrast, the initial reliability indices of element 7 and element 8 are less than those
of other elements and continue increasing during subsequent iterations though they remain significantly lower
than others during the entire process of iterations. Therefore, the system reliability of the two-story frame
is largely dependent upon the reliability level of element 7 and element 8; the collapse mechanism of frame
system features a local failure mode. In the process of iterations, the DURI gradually increases and converges
to approximately 0.746 in about 17 steps of iterations using q = 2.0; see Fig. 9. Moreover, the RRI decreases
during the iterations as exhibited in Fig. 10 and converges to approximately 5.994 in about 17 steps of iterations.
Based on Eq. (32) and the reliability of the potential failure elements listed in Table 9, the systemreliability
index as well as the joint probability of failure of the frame can be estimated by means of the FOBM, the
SOBM, the TOBM and the ECCM. The corresponding numerical results are rendered in Table 10, which
shows that the system reliability index produced by the SEMRM is approximately 3.644 in all three cases of
q = 0.5, q = 1.5, and q = 2.5. Comparison of the numerical results by the SEMRM in Table 10 with those by
the FMA in Table 8 demonstrates the accuracy and computational effectiveness of the proposed methodology.
In order to investigate the inuence of the adjustment factor q on the numerical results by the SEMRM, the
system reliability index
s
and corresponding iterative steps are estimated with q varying from 0.1 to 3.0. The
corresponding numerical results are summarized in Table 11. It is apparent that
s
remains almost constant,
Author's personal copy
System reliability analysis of spatial variance frames 121
0 5 10 15 20
5.9
6.0
6.1
6.2
6.3
Iteration step
R
R
I
Fig. 10 The iterative process for the RRI
Table 10 System reliability index and failure probability of the one-bay two-story frame by the SEMRM
q Method System reliability index System failure probability
0.5 FOBM [3.461748, 3.644041] [1.341953 10
4
, 2.683401 10
4
]
SOBM [3.538277, 3.642846] [1.348199 10
4
, 2.013736 10
4
]
TOBM [3.642585, 3.642585] [1.349568 10
4
, 1.349568 10
4
]
ECCM 3.644089 1.341701 10
4
1.5 FOBM [3.461748, 3.644049] [1.341912 10
4
, 2.683401 10
4
]
SOBM [3.537899, 3.642564] [1.349677 10
4
, 2.016621 10
4
]
TOBM [3.442242, 3.642242] [1.351368 10
4
, 1.351368 10
4
]
ECCM 3.644089 1.341701 10
4
2.5 FOBM [3.461748, 3.644048] [1.341917 10
4
, 2.683401 10
4
]
SOBM [3.537937, 3.642565] [1.349676 10
4
, 2.016334 10
4
]
TOBM [3.442242, 3.642242] [1.351367 10
4
, 1.351367 10
4
]
ECCM 3.644089 1.341701 10
4
while the number of iterative steps decreases as q varies from 0.1 to 3.0. Therefore, q has negligible effect
on the accuracy of the SEMRM, but much on its computational efciency. The higher the adjustment factor
q is, the fewer the number of iterations is for the SEMRM. Furthermore, the numerical results converge in
13 iterative steps (or less) when q 2.5, thus demonstrating the computational efciency of the proposed
SEMRM methodology.
5.3 Spatial variance of random eld
As random parameters always vary with respect to time or space, they should be modeled as stochastic pro-
cesses or randomelds rather than randomvariables. In order to investigate the inuence of spatial variances of
structural random parameters in Example 2 upon the system reliability, all cross-sectional areas and moments
of inertia listed in Table 6 are described as independent Gaussian random elds with wide-sense homogeneous
and isotropic properties. According to the principle of stochastic process, both the scale of uctuation and
correlation structure of the random elds might affect the outcomes of stochastic analysis. Therefore, four
types of correlation structures of random elds are considered in the following:
1. The triangular correlation function (TCF)
i j
() and its corresponding variance function
i j
() take the
form:

i j
(r) =
_
1
|r|

i j
, |r|
i j
0, |r| >
i j
;
i j
(L
e
) =

1
L
e
3
i j
, L
e

i j

i j
L
e
_
1

i j
3L
e
_
, L
e
>
i j
(34)
in which
i j
represents the scale of uctuation.
Author's personal copy
122 L. F. Yang et al.
Table 11 System reliability index with different adjustment factors
q
s
Iterations
0.1 3.651855 274
0.2 3.644089 145
0.3 3.644089 101
0.4 3.644089 78
0.5 3.644089 64
0.6 3.644089 53
0.7 3.644089 46
0.8 3.644089 41
0.9 3.644089 37
1.0 3.644089 33
1.1 3.644089 30
1.2 3.644089 28
1.3 3.644089 25
1.4 3.644089 24
1.5 3.644089 22
1.6 3.644089 21
1.7 3.644089 20
1.8 3.644089 18
1.9 3.644089 17
2.0 3.644089 17
2.1 3.644089 16
2.2 3.644089 15
2.3 3.644089 14
2.4 3.644089 14
2.5 3.644089 13
2.6 3.644089 13
2.7 3.644089 12
2.8 3.644089 12
2.9 3.644089 12
3.0 3.644089 11
2. The exponential correlation function (ECF) and its corresponding variance function are:

i j
(r) = exp
_

2|r|

i j
_
;
i j
(L
e
) =

2
i j
2L
2
e
_
2L
e

i j
1 + exp
_

2L
e

i j
__
. (35)
3. The second-order autoregressive correlation function (SACF) and its corresponding variance function read:

i j
(r) =
_
1 +
4|r|

i j
_
exp
_

4|r|

i j
_
, (36a)

i j
(L
e
) =

i j
2L
e
_
2 + exp
_

4L
e

i j
_

3
i j
4L
e
_
1 exp
_

4L
e

i j
___
. (36b)
4. The Gaussian correlation function (GCF) and its corresponding variance function are:

i j
(r) = exp
_

r
2

2
i j
_
;
i j
(L
e
) =

2
i j
L
2
e
_
L
e

i j
er f
_
L
e

i j
_
+ exp
_

L
2
e

2
i j
_
1
_
, (37)
where erf() denotes error function and can be expressed as
erf(x) =
2

x
_
0
exp(u
2
)du. (38)
For simplicity, we assume that all cross-sectional areas and moments of inertia have the same types of corre-
lation structure and the same scale of uctuation . The results produced by the SEMRM corresponding to
above four types of correlation structures, including the TCF, ECF, SACF and GCF, are presented in Table 12,
Author's personal copy
System reliability analysis of spatial variance frames 123
Table 12 System reliability index via the scale of uctuation and correlation structures
( mm) System reliability index
TCF ECF GCF SACF
1,500 3.581844 3.592637 3.580751 3.584780
3,000 3.584596 3.595593 3.578759 3.585149
4,500 3.607544 3.605251 3.597614 3.598872
6,000 3.624785 3.614230 3.617531 3.613104
7,500 3.635650 3.622135 3.633223 3.625318
9,000 3.643291 3.628000 3.644891 3.635096
with varying from 1,500 to 9,000mm. It is apparent that the system reliability indices in accordance with
four different types of correlation structures agree very well with one another. That is, the system reliability
achieved by the SEMRMbased on the PSFEMand the LAMis essentially insensitive to the type of correlation
structure of the random eld. Moreover, it can be readily seen that the system reliability index is smaller when
the structural parameters are modeled as random elds than that as random variables. Evidently, the system
reliability index increases along with the scale of uctuation of the random elds and approaches gradually to
the value when the structural parameters are modeled as random variables. This implies that the assumption
of random variables for all random parameters of the frame structure overestimates the system reliability of
the frame and might lead to unsafe design of structures.
6 Conclusions
Based on the local averaging of random elds, perturbation stochastic nite element method and the elastic
modulus reduction method for limit analysis of structures, this paper presents the stochastic elastic modu-
lus reduction method (SEMRM) for system reliability analysis of spatial variance frame. Several numerical
examples demonstrate the computational effectiveness and accuracy of the proposed SEMRM methodology.
When the randomeld approach is employed to describe the randomparameters of the stochastic structure,
the system reliability index predicted by the SEMRM is smaller than that when the random variable approach
is utilized. The system reliability index increases along with the scale of uctuation of the random elds and
approaches gradually to the value when the random variable model is used.
Comparing the SEMRM with the FMA and other traditional system reliability approaches, the proposed
methodology manifests considerable advantages as it does not require the identication of significant failure
modes. Moreover, the estimation of joint probability of failure elements in the SEMRM is much simpler than
the estimation of joint probability of failure modes in the traditional system reliability methods. Therefore,
the SEMRM is superior in computational efciency and simplicity in comparison with the existing system
reliability methods.
Acknowledgments This research was supported by the National Natural Science Foundation of China (50768001), the China
Scholarship Council Postgraduate Scholarship Program (2008666001) and the Guangxi Natural Science Foundation (0991020Z,
0992028-7).
References
1. Chen, J.B., Li, J.: Development process of nonlinearity based reliability evaluation of structures. Probab. Eng.
Mech. 22(3), 267275 (2007)
2. Moses, F.: System reliability developments in structural engineering. Struct. Saf. 1(1), 313 (1982)
3. Thoft-Christensen, P.: Consequence modied -unzipping of plastic structures. Struct. Saf. 7(24), 191198 (1990)
4. Melchers, R.E., Tang, L.K.: Dominant failure modes in stochastic structural systems. Struct. Saf. 2(2), 127143 (1984)
5. Corotis, R.B., Nafday, A.M.: Application of mathematical programming to system reliability. Struct. Saf. 7(24),
149154 (1990)
6. Yuan, X.X., Pandey, M.D.: Analysis of approximations for multinormal integration in systemreliability computation. Struct.
Saf. 28(4), 361377 (2006)
7. Nadarajah, S.: On the approximations for multinormal integration. Comput. Ind. Eng. 54(3), 705708 (2008)
8. Thoft-Christensen, P., Dalsgrd Srensen, J.: Reliability of structural systems with correlated elements. Appl. Math.
Model 6(3), 171178 (1982)
9. Zhang, Y.C.: High order reliability bounds for series systems and application to structural systems. Comput.
Struct. 46(2), 381386 (1993)
Author's personal copy
124 L. F. Yang et al.
10. Greig, G.L.: An assessment of high-order bounds for structural reliability. Struct. Saf. 11(34), 213225 (1992)
11. Stefanou, G.: The stochastic nite element method: past, present and future. Comput. Method. Appl. M. 198(912),
10311051 (2009)
12. Spanos, P.D., Ghanem, R.: Stochastic nite element expansion for random media. J. Eng. Mech. ASCE 115(5),
10351053 (1989)
13. Vanmarcke, E., Shinozuka, M., Nakagiri, S.: Random elds and stochastic nite elements. Struct. Saf. 3, 143166 (1986)
14. Knabe, W., Przewlcki, J., Rzynski, G.: Spatial averages for linear elements for two-parameter random eld. Probab. Eng.
Mech. 13(3), 147167 (1998)
15. Yang, L.F., Li, Q.S., Leung, A.Y.T., Zhao, Y.L., Li, Q.G.: Fuzzy variational principle and its applications. Eur. J. Mech. A
Solid 21, 9991018 (2002)
16. Yang, L.F., Leung, A.Y.T., Yan, L.B., Wong, C.W.Y.: Stochastic spline Ritz method based on stochastic variational princi-
ple. Eng. Struct. 27, 455462 (2005)
17. Hamilton, R., Boyle, J.T.: Simplied lower bound limit analysis of transversely loaded thin plates using generalised yield
criteria. Thin Wall. Struct. 40(6), 503522 (2002)
18. Adibi-Asl, R., Fanous, I.F.Z., Seshadri, R.: Elastic modulus adjustment proceduresimproved convergence schemes. Int.
J. Pres. Ves. Pip. 83(2), 154160 (2006)
19. Chen, L.J., Liu, Y.H., Yang, P., Cen, Z.Z.: Limit analysis of structures containing aws based on a modied elastic compen-
sation method. Eur. J. Mech. A Solid 27(2), 195209 (2008)
20. Yang, L.F., Yu, B., Qiao, Y.P.: Elastic modulus reduction method for limit load evaluation of frame structures. Acta Mech.
Solida Sin. 22(2), 109115 (2009)
21. Yu, B., Yang, L.F.: Elastic modulus reduction method for limit analysis of thin plate and shell structures. Thin Wall.
Struct. 48, 291298 (2010)
22. Yu, B., Yang, L.F.: Elastic modulus reduction method for limit analysis considering initial constant and proportional
loadings. Finite Elem. Anal. Des. 46(12), 10861092 (2010)
23. Ju, J.W.: On energy-based coupled elastoplastic damage theories: constitutive modeling and computational aspects. Int.
J. Solids Struct. 25(7), 803833 (1989)
24. Ju, J.W.: Isotropic and anisotropic damage variables in continuum damage mechanics. J. Eng. Mech., ASCE 116(12),
27642770 (1990)
25. Ju, J.W., Lee, H.K.: A micromechanical damage model for effective elastoplastic behavior of partially debonded ductile
matrix composites. Int. J. Solids Struct. 38(3637), 63076332 (2001)
26. Ju, J.W., Ko, Y.F., Ruan, H.N.: Effective elastoplastic damage mechanics for ber reinforced composites with evolutionary
partial ber debonding. Int. J. Damage Mech. 17(6), 493537 (2008)
27. Yang, L.F., Yu, B., Mo, Y.C.: Structural systemreliability analysis based on the elastic modulus reduction method. J. Guangxi
Univ. 33(3), 350353 (2008) (in Chinese)
28. Cornell, C.A.: Bounds on the reliability of structural systems. J. Struct. Div., ASCE 93(1), 171200 (1967)
29. Ditlevsen, O.: Narrow reliability bounds for structural systems. J. Struct. Mech. 7(4), 453472 (1979)
30. Feng, Y.: A method for computing structural system reliability with high accuracy. Comput. Struct. 33(1), 15 (1989)
31. Ang, A.H.S., Ma, H.F.: On the reliability analysis of framed structures. In: Proceedings of 4th ASCE Speciality Conference
on Probabilistic Mechanics and Structural Reliability, Tucson, pp. 106111 (1979)
Author's personal copy

You might also like