International Journal of Solids and Structures 41 (2004) 5035–5050

www.elsevier.com/locate/ijsolstr

Efficient method for calculation of system

reliability of a complex structure
S. Park a, S. Choi b,*
, C. Sikorsky c, N. Stubbs d

a
School of Architecture, Youngsan University, Junam-Ri, Ungsang-Up, Yangsan-Si, Kyungnam 626-847, South Korea
b
Structural Systems and Site Evaluation Department, Korea Institute of Nuclear Safety, Guseong-Dong,
P.O. Box 114, Yuseong-Gu, Daejeon 305-600, South Korea
c
State of California Department of Transportation, Office of Earthquake Engineering, CA 94274, USA
d
Department of Civil Engineering, Texas A&M University, College Station, TX 77843, USA
Received 30 January 2004; received in revised form 8 April 2004
Available online 18 May 2004

Abstract
It is quite difficult to calculate the collapse probability of a system such as a statically indeterminate structure that
has many possible modes or paths to complete failure and the problem has remained essentially unsolved. Generally, a
structure is synthesized by several components or elements and its capacity to resist the given load(s) is a function of the
capacity of the individual element. Thus it is reasonable to assess the probability of failure of the system based upon its
element-wise failure probability information. This paper proposes an efficient technique to directly assess the reliability
of a complex structural system from the reliabilities of its components or elements. The theory for the calculation of the
probability of a structural system is presented along with the target requirements and the fundamental assumptions
governing the method. A portal frame and two truss structures are selected to demonstrate the efficiency of the method
by comparing the results obtained from the proposed method to those from the existing methods in the literature.

2004 Elsevier Ltd. All rights reserved.

Keywords: System reliability; Probability of failure; Statically indeterminate structure

1. Introduction

The modern trend in structural mechanics has been to take into consideration the random nature of the
loads to which structures are subjected, and the variation in the material properties of the structural
constituents. In other words, the loads impacting a structure and the resistance of the structure are con-
sidered to be random variables. The safety margin provided by the structure is the amount by which the
random resistance of the structure exceeds the random load applied to the structure. Failure is said to occur

*
Corresponding author. Tel.: +82-42-868-0666; fax: +82-42-868-0523.
E-mail addresses: [email protected] (S. Park), [email protected] (S. Choi), [email protected] (C. Sikorsky), n-stubbs@
tamu.edu (N. Stubbs).

0020-7683/$ - see front matter
2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijsolstr.2004.04.028
5036 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

when the safety margin is less than, or equal to, zero. The relative safety of a structure is now expressed in
terms of a probability of failure, which is the complement of the reliability of the structure. Many attempts
have been made on the evaluation of reliability of structures. In general, real structures which are highly
indeterminate, in other words ‘‘complex’’, may fail in one or combination of several modes. The approaches
to evaluating structural safety of such complex structures can be categorized into three methods: (a)
numerical integration method; (b) failure path method; and (c) simulation.
In the numerical integration method, a failure mode is defined by a limit state surface which divides the
space of basic variables into two sets, the safety set and the failure set (Ditlevsen, 1979). The probability of
failure can be calculated by integration of the probability density functions over the failure set. Some
researchers have attempted to evaluate the probability of failure by integrating numerically the joint
probability density function of the applied loads over the failure set (Kam et al., 1983; Lin and Corotis,
1985). However, these approaches were limited to two-dimensional spaces since direct integration is
impractical in multi-dimensional space.
In the failure path method, reliability analysis of a structural system can be divided into two steps: (a)
identification of failure modes; and (b) estimation of failure probabilities of individual modes and the
overall system. The failure modes can be identified using the standard structural analysis and may be
generated automatically as demonstrated by Watwood (1979) and Gorman (1981). However, real structures
can have many different failure modes and it is infeasible to enumerate all the possible failure modes. Many
studies have focused on the possibility of using dominant failure modes based on the assumption that safety
of a structural system can be estimated efficiently using these dominant failure modes (Moses and Stahl,
1978; Ang and Ma, 1981; Murotsu et al., 1984; Thoft-Christensen and Murotsu, 1986; Ranganathan and
Deshpande, 1987; Tung and Kiremidjian, 1992). The last step of reliability analysis is calculation of failure
probability of a system using combination of failure modes and bound techniques.
Monte Carlo simulation has been often used to estimate the probability of failure and to verify the
results of other reliability analysis methods (Ang and Tang, 1984; Grimmelt and Schueller, 1982; Schueller
and Stix, 1987). In this technique, the random loads and random resistance of a structure are simulated and
these simulated data are used to find out if the structure fails or not, according to pre-determined limit
states. The probability of failure is the relative ratio between the number of failure occurrence and the total
number of simulations. This technique is easy to apply but when encountered low failure probability which
is typical in real structural systems, the number of simulation is larger so the method is generally not
practical for most realistic problems.
Although these previous works represent great strides in structural safety evaluation, many problems
remain to be solved. One of obstacles in the area of reliability analysis is difficulty to calculate failure
probability of complex structures, which are, for example, highly indeterminate and have many elements,
using currently developed methods. The objective of this paper is to present a practical methodology that
can be used to estimate the system reliability of arbitrary structures and to provide evidence to support the
validity of the methodology.

2. Theory

The reliability or safety of a structure is its ability to withstand the design loads for a specified period of
time. If the probability of failure of the system is given by Psystem , the reliability of the system, R, is defined as
R ¼ 1
Psystem ð1Þ
A practical methodology is presented below for the evaluation of the reliability of arbitrary structural
systems. To meet this objective, the following four tasks will be accomplished: (a) describe the background
of the methodology; (b) discuss a set of realistic target requirements for the methodology; (c) postulate a set
 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050 5037

of assumptions to be used as a basis for the methodology and describe the basic components of the
methodology that are consistent with the target requirements and the basic assumptions; and (d) select a
model to estimate the failure probabilities of components/elements.

2.1. Background

A mechanical system is any combination of individual elements that are synthesized to perform a
dedicated mechanical function. For the purposes of this study, any mechanical system may be assigned to
one of the following three categories: series systems (see Fig. 1), parallel systems (see Fig. 2), or hybrid
systems (see Fig. 3). Series systems are those systems in which the failure of any element leads to the failure
of the system. Parallel systems are those systems in which the combined failure of each and every element of
the system results in the failure of the system. If a system does not satisfy these strict definitions of ‘‘series’’
or ‘‘parallel’’ systems, the system is classified as a hybrid system. Hybrid systems are those systems in which
‘‘few,’’ ‘‘some,’’ or ‘‘most’’ (but not ‘‘anyone’’ or ‘‘all’’) of the elements must fail in order for the system to
fail. Any combination of elements whose failure results in the failure of the system is known as a failure
mode. In the language of structural mechanics, hybrid systems are statically indeterminate (redundant)
structures with many possible failure modes.
A hybrid system may be described as a series system of parallel modes of failure. For a single mode of
failure in a system, the possibility exists for there to be statistical correlation between the demand on the
elements in that mode or the capacity of the elements to resist the demand. Furthermore, since the same
element may participate in more than one mode of failure, statistical correlation must also exist among the

Fig. 1. Series system.

Fig. 2. Parallel system.

Fig. 3. Hybrid system.

5038 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

various failure modes of a hybrid (or series or parallel) system. In this study, hybrid systems in which
statistical correlation exists among elements and failure modes are designated as complex systems. A major
problem in reliability theory is the efficient and accurate determination of the failure probabilities of highly
redundant structural systems with correlated modes of failure.

2.2. Target requirements of the methodology

The target requirements for the proposed methodology are divided into two categories: primary
requirements and functional requirements. Primary requirements are those characteristics and features that
the methodology should possess. Functional requirements are specific tasks that must be performed by
components of the methodology such that the primary requirements are satisfied. The desired primary
requirements associated with the methodology can be stated as follows: (a) the reliability predictions should
be accurate; (b) the rationale for the method should be logical; (c) the methodology should be internally
consistent; (d) the methodology should be relevant; (e) the methodology should be versatile; (f) the
methodology should not be overly complex; and (g) the methodology should be computationally com-
petitive. These primary characteristics will be satisfied if the following functional requirements are per-
formed:

1. The predicted probabilities of system failure should be close to values predicted by other theories;
2. The prediction should agree with Monte Carlo simulations of structural system failure;
3. How the proposed method is related to other methods of system reliability evaluations should be clearly
indicated;
4. The method should be at least as computationally efficient as the other leading methods;
5. The methodology should apply to any structural type (e.g., frames, trusses, etc.. . .); and
6. The basic assumptions underlying the methodology should be clearly stated.

2.3. Fundamental assumptions governing the methodology

The proposed methodology builds on five fundamental assumptions that involve the following: (a) the
definition of failure modes; (b) the extent of statistical correlation among safety margins of basic elements;
(c) the sequence of component failure; (d) the definition of system failure; and (e) a rule for the rapid
estimation of the failure probability of the system and the identification of the components of the most
likely mode of failure. Theses basic assumptions are discussed below.

2.3.1. Types of failure modes

Two types of failure modes considered in this methodology can be discussed with the truss shown in Fig.
4. The first type of failure mode is designated the ‘‘global failure mode’’. This type of mode exists when the
structure as a whole becomes geometrically unstable. In Fig. 4(a) for example, the degrees of static inde-
terminacy (redundancy) of the structure are three. Therefore, the structure will become kinematically
unstable if any combination of four of its 16 elements fail. In general, if s is the redundancy of a structure,
the structure fails if s þ 1 components fail.
Note that the structure may also be considered to have failed if anyone of the three stories fails. Fig. 4(b)
depicts three logical substructures of the complete system. Note that each substructure has a redundancy of
one. Therefore, if any two elements of one of the substructures fail, a floor of the structure will fail. The
latter failure modes are deemed local failure modes. Thus the number of failure modes for a structure is the
sum of the global modes and the local modes. This discussion can be extended to other types of structures.
All that one has to do is to determine the redundancy of the global structure, isolate the basic substructure
units, then determine the redundancy of the substructures.
 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050 5039

Fig. 4. A 16-member truss structure: (a) whole system (b) three substructures of the system.

2.3.2. Justification of the correlation assumption

The basic unit in the reliability of a hybrid system is the failure mode. Note that each failure mode is a
parallel system. The computation of the failure probability of a parallel system can be simplified signifi-
cantly if it is assumed that the safety margins of the basic elements are completely correlated. This
assumption holds true for many practical structures. Let a structure have M failure elements (e.g., the
failure in tension of each and every one of the truss members of the truss shown in Fig. 4). The random
safety margin, gi , of the ith element is given by
X
NL
g i ¼ Ri
eij Sj ð2Þ
j¼1

where Ri is the random resistance of element i, Sj is the applied jth load on the structure, eij is the influence
coefficient which depends upon the geometry, redundancy, and material properties of the structure, and NL
is the number of applied loads.
Let elements k, l, m, n, and o be involved in failure mode j. Then the probability of failure of that mode,
Pj , is given by
Pj ¼ P f½gk 6 0 \ ½gl 6 0 \ ½gm 6 0 \ ½gn 6 0 \ ½go 6 0g ð3Þ

For most structures made of the same material, the resistance, Ri , are correlated. Also, because of the
common existence of the loads Sj in (2), the random safety margins are also correlated. Thus in most
structures, one expects a significant degree of correlation among the safety margins. Therefore, if it is
assumed that the safety margins of each and every mode are complete correlated, an estimate of the
probability of failure of the mode given by (3) is
Pj  Minfpk ; pl ; pm ; pn ; po g ð4Þ
where, for example,
pk ¼ P ½gk 6 0 ð5Þ
5040 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

Note that the estimate of the failure probability in (4) represents an upper bound to the failure probability
of a parallel system. This estimate is, therefore, a conservative estimate of the failure probability of the
mode.

2.3.3. The consequence of ignoring the sequence of component failure

Here it is asserted that the estimation of the failure probability of a highly redundant system can be
significantly simplified if the sequence of the failure of the components of a given mode is ignored. If the
stated assumption is adhered to, at least two major consequences follow. First, the number of failure modes
to be considered is reduced significantly. For example, if there are a total of n failure components and each
mode contains r elements, in the beta-unzipping method (Thoft-Christensen and Murotsu, 1986), there are
P ðn; rÞ possible modes; whereas, if the sequence of the failure of the components is ignored, there are Cðn; rÞ
possible modes. Note that P ðn; rÞ ¼ r!Cðn; rÞ where P denotes permutation and C denotes combination. For
the truss example shown in Fig. 4, for the case of global modes, r ¼ 4 and n ¼ 16. Accordingly,
P ðn; rÞ ¼ 43; 680 and Cðn; rÞ ¼ 1820.
The second consequence of the sequence-relaxation assumption is the reduction in the number of
structural analyses that have to be performed to make a determination of the probability of failure of the
system. In the beta-unzipping method, for example, if one wants to determine the most likely mode of
failure for the truss in Fig. 4, for s ¼ 3, three structural analyses and re-analyses are required to determine
the most ‘‘probable’’ sequence of failure. If the sequence of failure is ignored, only a single structural
analysis is required.

2.3.4. The probability of system failure

Having introduced the simplifying assumptions of complete correlation among modes and the unim-
portance of the sequence of component failure in a given mode, the safety of a structure with many modes
must now be defined. As noted previously, any hybrid system may be modeled as a series system of parallel
modes such as the one depicted in Fig. 3. Note that the failure probability of mode j is assigned to the
structure according to the rule stipulated in (4). If Fi is denoted to be the event that failure of mode i occurs
and Pi is the probability of occurrence of that event, then the failure probability of the system is given by
" #
[
NM
Psystem ¼ P Fi ð6Þ
i¼1

where NM is the number of possible failure modes.

Since complete correlation among the modes has been assumed, it follows that
Psystem ¼ Max½P1 ; P2 ; P3 ; . . . ; PNM  ð7Þ

2.3.5. Rapid estimation of system reliability

In order to execute (7), one needs to know the probability of failure of each mode. Even though the
sequence of component failure has been ignored, the task of identifying all NM possible failure modes
could be formidable. For example, in the structure provided in Fig. 4, the number of failure modes in the
structure with a redundancy of as little as 3 and 16 possible failure elements is 1820. Thus the objective of
this section is to develop a simple rule that engenders the solution of (7) without having to identify the
numerous individual failure modes. This objective can be accomplished via mathematical induction.
Let the system have n failure components. For example, in the truss depicted in Fig. 4, the number of
components is 16. Assuming that the component failure probabilities, ai , are known, reorder the proba-
bilities in descending order of magnitude such that a1 > a2 > a3 >    > an . Note that the magnitude of the
 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050 5041

probability a1 is not necessarily the probability assigned to component 1. With this reordering, the authors
can make the following inferences:

1. If the failure model of the structure is a series system,

Psystem ¼ Max½P1 ; P2 ; P3 ; . . . ; PNM  ¼ a1 ð8Þ
2. If the redundancy of the structure is one, the mode of maximum probability of failure contains compo-
nents with failure probabilities a1 and a2 . Consequently,

Psystem ¼ a2 ð9Þ
3. If the redundancy of the structure is two, the mode with the maximum probability of failure contains the
components with failure probabilities a1 , a2 , and a3 . Consequently,
Psystem ¼ a3 ð10Þ
4. By mathematical induction, if the redundancy of the structure is s, the mode with the maximum
probability of failure contains the components with failure probabilities a1 ; a2 ; a3 ; . . . ; as ; asþ1 . Conse-
quently,
Psystem ¼ asþ1 ð11Þ

On the basis of Statement 4 above, the authors propose the following rule to estimate the probability of
failure of an arbitrary structure without knowledge of the sequence of failure of the components in a mode
or the elements of the mode.
If the failure probabilities of the elements are rearranged in order of descending magnitude
a1 > a2 > a3 >    > aN and if the redundancy of the structure (or substructure) is s, an estimate of the
failure probability of the system is given by Psystem ¼ asþ1 .
Furthermore, the s þ 1 elements participating in the failure mode are those elements with failure
probabilities a1 ; a2 ; a3 ; . . . ; as ; asþ1 . Note that the estimate given by (11) may be thought of as a lower bound
of the upper bound of failure mode probabilities.

2.4. Estimation of failure probabilities of components

Suppose a structural element has only one load effect S resisted by one resistance R. The load effect S may
be obtained from the applied loading Q through a structural analysis. It is important to note that R and S
should be expressed in the same units. If its resistance R is less than, or equal to, the stress resultant S acting
on it, that element is considered to have failed. The probability of failure, pf , of the structural element can
be stated in any of the following ways:

pf ¼ P ðR 6 SÞ ¼ P ðR
S 6 0Þ ¼ P ðR=S 6 1Þ ð12Þ
or
pf ¼ P ½GðR; SÞ 6 0 ð13Þ
where GðÞ is termed the ‘‘limit state function’’ or the ‘‘failure function’’.
The most notable example is when R and S are normal random variables with means lR and lS and
variances r2R and r2S , respectively. The safety margin M ¼ R
S then has a mean and variance
lM ¼ lR
lS ð14Þ
5042 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

and
r2M ¼ r2R þ r2S ð15Þ
using well-known rules for addition (subtraction) of normal random variables. Eq. (12) then becomes


0
lM
pf ¼ P ðR
S 6 0Þ ¼ P ðM 6 0Þ ¼ U ð16Þ
rM
where U is the standard normal distribution function (zero mean and unit variance). Using (14) and (15)
it follows that
!

ðlR
lS Þ
pf ¼ U 1=2
¼ Uð
bÞ ð17Þ
ðr2R þ r2S Þ

where b is defined as the safety index (Cornell, 1969)

b ¼ lM =rM ð18Þ
Evidently, (17) yields the exact probability of failure when both R and S are normally distributed and hence
b is a direct measure of the safety of the structural element, i.e., greater b represents greater safety.
In (17), only two random variables, i.e., load and resistance, were considered. If the limit state function
GðXÞ consists of more than two basic random variables, a useful approach is to calculate approximate
moments by expanding GðX1 ; X2 ; . . . ; Xn Þ as a first-order Taylor series about the means lX1 ; lX2 ; . . . ; lXn
(Ang and Tang, 1975)
lM  GðlX1 ; lX2 ; . . . ; lXn Þ ð19Þ

and
n

X 2 
oG 
r2M  2
lXi rXi ð20Þ
i¼1
oX i

This is the well-known first-order second-moment (FOSM) method. The safety index b defined by (18) is
still effective in this case and the probability of failure can also be formulated, provided the basic variables
Xi (i ¼ 1; 2; . . . ; n) are normally distributed
pf ¼ Uð
bÞ ð21Þ
In this study, (19)–(21) will be used as a mathematical model to estimate the failure probabilities of
structural elements.

3. Numerical examples

The validity of the proposed methodology which can estimate the system reliability of arbitrary structure
is demonstrated via numerical examples. It is assumed that the yield stresses of the members and the applied
loads to the structure are statistically independent normal random variables. The objective of this section is
to compare the exact reliability of a structure obtained using existing reliability theory with the reliability
resulted from the proposed methodology. Three structures are selected for the purpose of this section.
These structures include: (a) a redundant portal frame; (b) 6-member truss with one degree of redundancy;
and (c) 16-member truss with three degrees of redundancy.
 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050 5043

3.1. Portal frame

3.1.1. Reliability using the existing method

A simple portal frame is shown in Fig. 5 (Thoft-Christensen and Murotsu, 1986). In this example, all
joints are rigid moment-resisting joints. The height of the column is 5 m and the span of the beam is 10 m.
Material properties are estimated such that all structural elements have the same area A ¼ 0:459  10
2 m2
and the same second moment of area I ¼ 0:579  10
4 m4 . All members are assumed to be made of the
same material with modulus of elasticity E ¼ 0:21  109 kN/m2 , density q ¼ 7850 kg/m3 , and Poisson’s
ratio m ¼ 0:3. The plastic bending moment capacities of the beam (Rb ) and columns (Rc ) are given such that
lRb ¼ 135 kN m; rRb ¼ 13:5 kN m
lRc ¼ 135 kN m; rRc ¼ 13:5 kN m
where l denotes the mean and r represents the standard deviation.
As shown in Fig. 5, this portal frame is subjected to a random vertical load V and a random horizontal
load H. The concentrated vertical load acts at the middle of the beam and the horizontal load is applied at
the top of the left column. All loads are taken to be static. The vertical load V and horizontal load H are
given by
lV ¼ 45 kN; rV ¼ 4:5 kN
lH ¼ 55 kN; rH ¼ 5:5 kN:
The collapse of the example structure may be caused by the formation of plastic hinges at the joints. The
potential locations of plastic hinges are indicated by ‘‘·’’ in Fig. 6. Ten possible collapse mechanisms

Fig. 5. Geometry of portal frame.

Fig. 6. Potential locations of plastic hinges.

5044 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

Fig. 7. Possible collapse mechanisms.

resulting from the combination of the vertical load V and the horizontal load H are shown in Fig. 7. The
corresponding limit state functions (gi ) using the virtual work theory (McCormac, 1975) are also presented
in Fig. 7. In each function, the work performed by the external loads during the displacement is subtracted
from the internal work absorbed by the hinges. Failure is said to occur when this function is less than, or
equal to, zero.
To compute the failure probabilities of the possible failure modes, the standard procedure from the first-
order second-moment (FOSM) reliability method summarized in the previous section is utilized. Table 1
lists the safety indices and the corresponding failure probabilities for the 10 failure modes.
A simple bounds technique (Cornell, 1967) is adopted to estimate the probability of failure of the system
and then the lower and upper bounds are
Y
10 X
10
maxðPi Þ 6 Psystem 6 1
ð1
Pi Þ  Pi ð22Þ
i¼1 i¼1

Table 1
Failure probabilities for possible failure modes
Mode no. Failure function Safety index (b) Failure probability (Pi )
1 g1 ¼ 4Rb
5V 5.385 0.363 · 10
7
2 g2 ¼ 2Rb þ 2Rc
5V 7.107 0.591 · 10
12
3 g3 ¼ 3Rb þ Rc
5V 6.528 0.334 · 10
10
4 g4 ¼ 3Rb þ Rc
5V 6.528 0.334 · 10
10
5 g5 ¼ 4Rc
5H 4.373 0.612 · 10
5
6 g6 ¼ 2Rb þ 2Rc
5H 5.632 0.892 · 10
8
7 g7 ¼ Rb þ 3Rc
5H 5.218 0.902 · 10
7
8 g8 ¼ Rb þ 3Rc
5H 5.218 0.902 · 10
7
9 g9 ¼ 4Rb þ 2Rc
5H
5V 4.425 0.482 · 10
5
10 g10 ¼ 2Rb þ 4Rc
5H
5V 4.425 0.482 · 10
5
 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050 5045

where Pi is the probability of occurrence of mode i. Note that the lower bound assumes that all failure
modes are fully dependent and upper bounds assumes that all failure modes are independent. From Table
1, the system failure probability is bounded by 0:612  10
5 6 Psystem 6 0:160  10
4 . Clearly, the failure
modes 1, 2, 3, 4, 6, 7, and 8 have negligible effects on the failure probability of the system.

3.1.2. Reliability using the proposed method

The critical failure elements 1–7 are indicated by the symbol ‘‘·’’ in Fig. 6. In this example, a joint fails
when the resisting plastic moment capacity of the joint is exceeded. These critical elements are taken to be
those locations where plastic hinges are most likely to occur in the structure. Note that the structure is
statically indeterminate to the third degree. If the sequence of failure is ignored, failure of any four of the
seven joints leads to the failure (kinematic instability) of the system. Therefore, there are 35 possible
combinations of seven element taken four at a time. The general failure model which describes this simple
structure is therefore a system with 35, four-arm, parallel modes in series.
To compute the failure probabilities of the joints, the standard procedure from the first-order second-
moment (FOSM) reliability method is utilized. First, failure functions are generated for each joint.
Beginning with the structure in Fig. 5, one can first set H ¼ 1 and V ¼ 0. Next, a numerical structural
analysis on the system can be performed and the resultant moment at each joint due to unit horizontal load
can be set equal to eHi . Next, one can set H ¼ 0 and V ¼ 1, then perform a similar numerical structural
analysis, and set the resultant moment at each joint due to unit vertical load to be eVi . With Ri as the
resisting moment of joint i, the safety margin, gi , for the joint becomes:

gi ¼ Ri
jeHi H þ eVi V j ð23Þ

where jeHi H þ eVi V j represents the absolute value of the resultant moment. These functions are provided in
the second column of Table 2. Next, using the loading and resistance statistics with (19)–(21), safety indices
and the corresponding failure probabilities are listed in the third and fourth column of Table 2.
According to the rule proposed in this study, the system failure probability is the fourth largest value
among the seven component failure probabilities. For example, the value of 0.584 · 10
5 , which is the
fourth largest value in Table 2, is taken as the system failure probability.

3.2. 6-Member truss structure

3.2.1. Reliability using the existing method

Consider a statically indeterminate 6-member truss with one degree of redundancy shown in Fig. 8
(Murotsu et al., 1980). The truss height is 91.44 cm and the width is 121.9 cm. Table 3 lists the material
properties of truss members. As shown in Fig. 8, one static concentrated force, P , is acting at the top of the
left vertical member and the statistics of P are given such that lP ¼ 44:45 kN and rP ¼ 4:445 kN.

Table 2
Component failure probabilities using proposed methodology (portal frame)
Joint no. Failure function Safety index (b) Failure probability (Pi )
1 g1 ¼ Rc
j
1:5681H þ 0:4919V j 4.384 0.584 · 10
5
2 g2 ¼ Rc
j0:9379H
0:9973V j 8.478 0.115 · 10
16
3 g3 ¼ Rb
j0:9379H
0:9973V j 8.478 0.115 · 10
16
4 g4 ¼ Rb
j0:0015H þ 1:5027V j 4.457 0.415 · 10
5
5 g5 ¼ Rb
j
0:9349H
0:9973V j 2.558 0.526 · 10
2
6 g6 ¼ Rc
j
0:9349H
0:9973V j 2.558 0.526 · 10
2
7 g7 ¼ Rc
j1:5590H þ 0:4919V j 1.680 0.465 · 10
1
5046 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

Fig. 8. Statically indeterminate 6-member truss structure with one degree of redundancy.

Table 3
Material properties for truss members (6-member truss structure)
Member no. Cross-sectional area A (cm2 ) Mean value of yield stress, ry (Pa) Elastic modulus, E (Pa)
8
1, 2, 5, 6 1.33 2.76 · 10 2.06 · 1011
3, 4 1.49 2.76 · 108 2.06 · 1011

Murotsu et al. (1980) defined that the complete failure of the structure is determined by investigating
singularity of the total structure stiffness matrix K formed with the members in survival. That is, the cri-
terion for complete failure is given by
jKj ¼ 0 ð24Þ
where jKj represents the determinant of matrix K.
In case of a statically indeterminate truss, there exist many possible failure modes or paths to complete
failure of the structure and it is impossible in practice to generate all of them. Hence, the failure probability
is estimated by evaluating its lower and upper bounds. The lower bound is evaluated by selecting the
dominant modes of failure and calculating their failure probabilities. The upper bound is evaluated by
assuming that the redundant truss behaves itself like a statically determinate truss, i.e., the structure fails if
any one member is subject to failure. Table 4 presents the lower bound and upper bound (Murotsu et al.,
1980) of the system failure probabilities along with the Monte Carlo simulation result performed by Kathir
(1991) on the same structure. In this example, failure due to buckling is not considered.

3.2.2. Reliability using the proposed method

The failure element can be any member of 6 truss elements. In this example, a member fails when the
resisting axial load capacity in tension or compression of the member is exceeded. Failure via buckling is
not considered. Note that the structure is statically indeterminate to the one degree. If the sequence of
failure is ignored, failure of any two of the six truss elements leads to the failure (kinematic instability) of
the system. Therefore, there are fifteen possible combinations of six taken two at a time.

Table 4
Failure probabilities from Murotsu et al. (1980) and Kathir (1991)
COV Failure probability
P ry Lower bound Upper bound Simulation
0.1 0.1 1.96 · 10
4 6.40 · 10
3 3.60 · 10
3
 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050 5047

To compute the failure probabilities of the failure elements, the standard procedure from the first-order
second-moment (FOSM) reliability method is utilized. First, failure functions are generated for each ele-
ment. Beginning with the structure in Fig. 8, set P ¼ 1 and perform a numerical structural analysis on the
system then set the resulting internal axial force at each member due to the unit load to be ei . With Vi as the
resisting axial capacity of element i, the safety margin, gi , for the element becomes:
gi ¼ Vi
jei P j ð25Þ
where jei P j is the absolute value of the resultant axial force in tension or compression. These functions are
provided in the second column of Table 5. Next, using the loading and resistance statistics with (19)–(21),
safety indices and the corresponding failure probabilities are listed in the third and fourth column of Table
5. According to the rule proposed in this study, the system failure probability is the second largest value
among the six component failure probabilities. For example, the value of 3.59 · 10
3 , which is the second
largest value in Table 5, is taken as the system failure probability.

3.3. Member truss structure

3.3.1. Reliability using the existing method

Consider a statically indeterminate 16-member truss with three degree of redundancy shown in Fig. 9
(Murotsu et al., 1980). The truss width is 121.9 cm and consists of three stories in which the height of a

Table 5
Component failure probabilities using proposed methodology (6-member truss structure)
Element no. Failure function Safety index (b) Failure probability (Pi )
1 g1 ¼ V1
j0:50P j 3.375 3.69 · 10
4
2 g2 ¼ V2
j0:3751P j 4.969 3.36 · 10
7
3 g3 ¼ V3
j0:6250P j 2.689 3.59 · 10
3
4 g4 ¼ V4
j
0:6250P j 2.689 3.59 · 10
3
5 g5 ¼ V5
j
0:3751P j 4.969 3.36 · 10
7
6 g6 ¼ V6
j
0:50P j 3.375 3.69 · 10
4

Fig. 9. Statically indeterminate 16-member truss structure with three degrees of redundancy.
5048 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

Table 6
Material properties for truss members (16-member truss structure)
Member no. Cross-sectional area A (cm2 ) Mean value of yield stress, ry (Pa) Elastic modulus, E (Pa)
1 3.35
2, 5 8.64
3, 4, 14 5.76
6 2.29 2.76 · 108 2.06 · 1011
7, 8, 10 4.03
9 7.35
11, 12, 15 1.58
13, 16 2.29

Table 7
Failure probabilities from Murotsu et al. (1980)
COV Failure probability
P ry Lower bound Upper bound

6
0.1 0.1 2.73 · 10 3.84 · 10
3

story is 91.44 cm. Table 6 lists the material properties of truss members. As shown in Fig. 9, three static
concentrated forces denoted P are acting at the top of each story where lP ¼ 44:45 kN and rP ¼ 4:445 kN.
The failure modes are the same as described in the example of a 6-member truss. Table 7 presents the
lower bound and upper bound (Murotsu et al., 1980) of the system failure probabilities. In this example,
failure due to buckling is not considered.

3.3.2. Reliability using the proposed method

The failure element can be any member of 16 truss elements. In this example, a member fails when the
resisting axial load capacity in tension or compression of the member is exceeded. Failure via buckling is

Table 8
Component failure probabilities using proposed methodology (16-member truss structure)
Element no. Failure function Safety index (b) Failure probability (Pi )
1 g1 ¼ V1
j1:2430P j 3.455 2.76 · 10
4
2 g2 ¼ V2
j3:1828P j 3.498 2.34 · 10
4
3 g3 ¼ V3
j2:1964P j 3.288 5.04 · 10
4
4 g4 ¼ V4
j
1:5538P j 5.187 1.07 · 10
7
5 g5 ¼ V5
j
3:5684P j 2.788 2.65 · 10
3
6 g6 ¼ V6
j
0:3929P j 6.975 1.52 · 10
12
7 g7 ¼ V7
j1:7734P j 2.377 8.73 · 10
3
8 g8 ¼ V8
j0:7949P j 6.503 3.93 · 10
11
9 g9 ¼ V9
j
1:7052P j 5.867 2.21 · 10
9
10 g10 ¼ V10
j
1:2271P j 4.576 2.37 · 10
6
11 g11 ¼ V11
j0:0648P j 9.319 5.86 · 10
21
12 g12 ¼ V12
j0:5257P j 4.091 2.15 · 10
5
13 g13 ¼ V13
j0:3741P j 7.127 5.14 · 10
13
14 g14 ¼ V14
j
0:8760P j 7.334 1.12 · 10
13
15 g15 ¼ V15
j
0:2245P j 7.517 2.79 · 10
14
16 g16 ¼ V16
j
0:2992P j 7.727 5.52 · 10
15
 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050 5049

not considered. The system may fail as a result of global instability or local instability. First, the global
instability is examined. Note that the structure is statically indeterminate to the third degree. If the sequence
of failure is ignored, failure of any four of the 16 truss elements leads to the failure (kinematic instability)
of the system. Therefore, there are 1820 possible combinations of 16 taken four at a time. As illustrated
in Fig. 4, one may define local failure if anyone of the three stories fails. Therefore, if any two elements of
one of the substructures fail, a floor of the structure will fail.
The failure functions for each element are generated using (25). These functions are provided in the
second column of Table 8. Next, using the loading and resistance statistics with (19)–(21), safety indices and
the corresponding failure probabilities are listed in the third and fourth column of Table 8. According to
the rule proposed in this study, the system failure probability is the fourth largest value among the 16
component failure probabilities for the global failure modes. For example, the value of 2.76 · 10
4 , which is
the fourth largest value in Table 8, is taken to be the system failure probability. For the local modes, the
failure probability for each floor is given by the second largest failure probability of the elements. For
example, the failure probability for the first floor is 5.04 · 10
4 , the failure probability for the second floor is
2.37 · 10
6 , and the failure probability for the third floor is 5.14 · 10
13 . Therefore, if the local and global
modes are considered together, the failure probability of the system is 5.04 · 10
4 .

4. Discussion of the results

In Table 9, the estimated failure probabilities generated using the proposed methodology are compared
to those calculated from existing reliability techniques. Note that the basic assumptions behind the
methodology were: (a) the safety margins of the elements and failure modes are fully correlated; (b) the
kinematic instability of a structural system defines the failure or collapse of the structure; and (c) the order
of the failure of components in a given mode is disregarded. Certainly, investigations and comparisons on
other structures to confirm the assumptions must be produced in the future. From the results, the following
observations are made:

1. The predicted probabilities of system failure using the proposed technique close to or fall within bounds
proposed in the literature;
2. In the case where data were available, the predicted failure probability agrees well with the result ob-
tained via Monte Carlo simulation;
3. The proposed technique is computationally efficient when compared to other methods, especially for the
complex structures that are highly indeterminate and have many elements;
4. The proposed technique automatically seeks and identifies the most probable failure mode; and
5. The proposed technique can be applied to any structural type.

Table 9
Comparisons of reliability techniques
Type of structure Probability of failure of system
Bounds technique Monte Carlo Proposed
simulation technique
Lower Upper
Portal frame 6.12 · 10
6 1.60 · 10
5 – 5.84 · 10
6
6-Member truss 1.96 · 10
4 6.40 · 10
3 3.60 · 10
3 3.59 · 10
3
16-Member truss 2.73 · 10
6 3.84 · 10
3 – 5.04 · 10
4
5050 S. Park et al. / International Journal of Solids and Structures 41 (2004) 5035–5050

5. Summary and conclusions

In this paper, a general methodology for the estimation of systems reliability for arbitrary structural
systems was proposed. First, a theory to estimate the system failure probability was developed under the
basic assumptions that the failure modes are completely correlated and that the sequence of component
failure in a given mode is not important. According to the consequence of these assumptions, a simple rule
has been proposed for direct estimation of system failure probability from the element or component failure
probabilities. A model to estimate the element or component failure probabilities was formulated by the
first-order second-moment (FOSM) reliability method. Finally, the proposed technique was compared to
the current reliability methods (i.e., bound technique and Monte Carlo simulation) through three examples:
(1) a simple portal frame; (2) a 6-member truss; and (3) a 16-member truss.
From this study, two conclusions are made. First, the results obtained for the example structures used
here indicate that the predicted system reliability values agree closely with those values predicted by other
approaches. Second, the proposed technique is computationally efficient when compared to other methods,
especially for the complex structures that are highly indeterminate and have many elements.

References

Ang, A.H-S., Ma, H-F., 1981. On the reliability of structural systems. In: Proceedings of 3rd International Conference on Structural
Safety and Reliability, Trondheim, Norway. Elsevier, Amsterdam, pp. 295–314.
Ang, A.H-S., Tang, W.H., 1975. Probability Concepts in Engineering Planning and Design, vol. I. John Wiley and Sons, USA.
Ang, A.H-S., Tang, W.H., 1984. Probability Concepts in Engineering Planning and Design, vol. II. John Wiley and Sons, USA.
Cornell, C.A., 1967. Bounds on the reliability of structural system. J. Struct. Div., ASCE 93 (ST1), 171–200.
Cornell, C.A., 1969. A probability-based structural code. J. ACI 66 (12), 974–985.
Ditlevsen, O., 1979. Narrow reliability bounds for structural systems. J. Struct. Mech. 7 (4), 453–472.
Gorman, M.R., 1981. Automatic generation of collapse mode equations. J. Struct. Div., ASCE 107 (ST7), 1350–1354.
Grimmelt, M., Schueller, G.I., 1982. Benchmark study on methods to determine collapse failure probabilities of redundant structures.
Struct. Safety 1, 93–106.
Kam, T-Y., Corotis, R.B., Rossow, E.C., 1983. Reliability of nonlinear framed structures. J. Struct. Eng., ASCE 109 (7), 1585–1601.
Kathir, N.M., 1991. System consequence factors for offshore oil production platforms. Ph.D. Thesis, Texas A&M University, USA.
Lin, T.S., Corotis, R.B., 1985. Reliability of ductile systems with random strengths. J. Struct. Eng., ASCE 111 (6), 1306–1325.
McCormac, J.C., 1975. Structural Analysis, third ed. Intext Educational Publishers, USA.
Moses, F., Stahl, B., 1978. Reliability analysis format for offshore structures. In: Proceedings of 10th Annual Offshore Technology
Conference, Houston, Texas, USA, Paper 3046.
Murotsu, Y., Okada, H., Niwa, K., Miwa, S., 1980. Reliability analysis of truss structures by using matrix method. J. Mech. Des. 102,
749–756.
Murotsu, Y., Okada, H., Taguchi, K., Grimmelt, M., Yonezaw, M., 1984. Automatic generation of stochastically dominant failure
modes of frame structures. Struct. Safety 2, 17–25.
Ranganathan, R., Deshpande, A.G., 1987. Generation of dominant modes and reliability analysis of frames. Struct. Safety 4, 217–228.
Schueller, G.I., Stix, R., 1987. A critical appraisal of methods to determine failure probabilities. Struct. Safety 4, 293–309.
Thoft-Christensen, P., Murotsu, Y., 1986. Application of Structural System Reliability Theory. Springer-Verlag, USA.
Tung, A.T.Y., Kiremidjian, A.S., 1992. Application of system reliability theory in the seismic analysis of structures. Earthquake
Spectra 8 (3), 471–494.
Watwood, V.B., 1979. Mechanism generation for limit analysis of frames. J. Struct. Div., ASCE 109 (ST1), 1–15.