See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/334596220

GLOBAL SEISMIC RELIABILITY ANALYSIS OF BUILDING STRUCTURES BASED

ON SYSTEM-LEVEL LIMIT STATES

Conference Paper · August 2008

CITATIONS READS

17 318

4 authors, including:

Da-Gang Lu Xiaohui Yu
Harbin Institute of Technology Guilin University of Technology
276 PUBLICATIONS 2,378 CITATIONS 109 PUBLICATIONS 1,061 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Da-Gang Lu on 21 July 2019.

The user has requested enhancement of the downloaded file.

The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

GLOBAL SEISMIC RELIABILITY ANALYSIS OF BUILDING STRUCTURES

BASED ON SYSTEM-LEVEL LIMIT STATES
1 2 3 4
Da-Gang Lu , Peng-Yan Song , Xiao-Hui Yu and Guang-Yuan Wang
1
Professor, School of Civil Engineering, Harbin Institute of Technology, Harbin. China
2
Graduate Student, School of Civil Engineering, Harbin Institute of Technology, Harbin. China
3
Graduate Student, School of Civil Engineering, Harbin Institute of Technology, Harbin. China
4
Professor, School of Civil Engineering, Harbin Institute of Technology, Harbin. China
Email: ludagang@hit.edu.cn, songpengyan@sina.com, yuxiaohui2830582@163.com, wanggy@hit.edu.cn

ABSTRACT

In this paper, a global limit state function for load-carrying capacity of structural system is firstly set up, in
which the margin of safety is the difference between the limit base shear of structural system and the total
horizontal seismic action. To probabilistically assess the global seismic capacity of structure, a new point
estimation method (PEM) for analyzing statistical moments of complex random function is put forward, and
then it is combined with deterministic finite element analysis to produce the so-called “random pushover
analysis (RPA)”. On the basis of the above new methodologies, a semi-analytical approach which integrates the
improved point estimation method, pushover analysis and first order reliability method (FORM) is developed to
analyze the nonlinear seismic reliability of structure as a global system. By applying the proposed approach in
R.C. frame structure, the changing rules of the global seismic reliability of the structure with the coefficient of
variation of the total seismic action and correlation coefficient of storey-level seismic forces are derived. It is
demonstrated by a numerical example that the newly developed method in this paper is simple, practical and
efficient compared with MCS, and that it has the same accuracy as MCS.

KEYWORDS: Global Reliability, Seismic Reliability, Random Pushover Analysis (RPA), Point Estimation
Method (PEM), First Order Reliability Method (FORM), Semi-Analytical Approach

1. INTODUCTION

As a conventional method in structural system reliability theory (Moses, 1982), the failure mode approach
(FMA) is difficult to apply in civil engineering practice, since it has many disadvantages: First, the constitutive
relations of materials are assumed to be perfect rigid-plastic, however, this is not the real case of many civil
engineering materials, such as concrete, structural steel, soil, etc.; Second, it is hardly to identify the significant
failure modes and determine their corresponding failure mode equations of large-scale and complex structures,
because the number of possible failure modes of realistic complexity can be extremely large; Third, the
correlation between failure modes is another important problem not easy to deal with; Forth, the overall failure
probability of structural systems cannot be evaluated accurately even though the dominant failure modes and
limit state equations are known a prior.
On the other hand, a new trend in which structural systems reliability is approximately calculated by using
global limit states based on nonlinear structural analysis techniques recently has been increasingly of interest in
many different communities. This novel approach encompass the following items: (1) the integrated nonlinear
analysis methods of structural systems considering real constitutive relations of materials, e.g., first order
inelastic analysis, second order inelastic analysis, etc., are utilized to search for the dominant modes of failure;
(2) the statistics of structural global load-carrying capacity are obtained by Monte Carlo simulations; (3) the
probability density function (PDF) of the global load-carrying capacity is fitted by its first few moments; (4) a
global limit state equation is set up, which comprises the global load-carrying capacity and structural load
effects; (5) the classic structural component reliability theory, such as first order reliability method (FORM), is
applied in the global limit state equation, the system reliability is then obtained approximately. This approach
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

has two advantages: first, it can directly bypass the difficulties in searching the significant modes of failure in
FMA; second, it can consider the real constitutive relations of structural materials. Therefore, it is a practical
and efficient approximate method to solve the systems reliability problems of structures. In this paper, we call
this approach as global reliability theory of structures.
To the author’s knowledge, Gorman and Moses (1979) perhaps are the first researchers who presented the idea
of directly computing the system reliability by structural system resistance. Grigoriu (1983) proposed a control
variable approach to approximate the reliability of complex problems from estimators developed for the
distribution of the control variable. Nowak and Zhou (1988, 1990) developed a numerical integration method to
calculate the first few moments of complex random function, and then applied the approach in system reliability
of highway bridges. Sigursdon et al. (1994) proposed a probabilistic collapse analysis method to assess the
system reliability of jacket platforms. Zhao and Ono (1998) developed a failure mode independent performance
function using load factor obtained by limit analysis, and then used response surface approach to approximate
the performance function and FORM to evaluate the system failure probability of ductile frames. Onoufriou and
Forbes (2001) reviewed and critically examined the recent developments in system reliability methods for fixed
steel offshore platforms. Moreover, they paid special attentions to pushover analysis, simplified models and
“component-based approach”. Ou et al. (2001) developed a probabilistic pushover analysis to approximately
evaluate the system reliability of buildings by randomizing both the capacity spectrum and demand spectrum.
Ou et al. (2003) also established a global limit state function considering the limit base shear of the whole
structure based on limit analysis to evaluate the system reliability of existing fixed jacket platforms. Li and
Cheng (2004) employed pushover analysis together with Monte Carlo simulation to check the probability
distribution types that the global resistance of steel and R.C. frames satisfies with K-S testing. Li et al. (2002,
2004, 2006) presented a reliability-based integrated design (RID) methodology of steel frames based on
nonlinear structural analysis, and gave a system-level RID format like as the LRFD formulations of structural
members to directly checks the structural system limit states and the corresponding system reliability. To
compute the system reliability, they proposed a semi-analytical simulation method to assess the reliability of
structural systems, which combines variance-reduction techniques including systematic sampling and antithetic
variates simulations to obtain the moments of system resistance, the procedure of fitting the PDF of system
resistance by exponential polynomial method (EPM), and first order reliability method (FORM). They have
succeeded in applying their methodology in advanced design of steel portal frames with tapered members in
industrial buildings and plane steel frames in high-rise buildings.
The theory of global reliability provides a practical and operational means of moving from member design level
toward system design level for reliability-based probability design of structures, and bridges the gap between the
two design levels. Moreover, the main ideas of this theory are consistent with those of performance-based
design theory now prevailing in the community of earthquake engineering. Therefore, it has a broader prospect
of applications. However, the research on global reliability theory of structures is still insufficient, and has not
been paid much attention to. Furthermore, there is a little study on the global seismic reliability of structural
systems, and nearly all the existing research employed Monte Carlo simulation to get the moments of system
resistance. Since the probability of failure of structures due to strong earthquakes is usually very small, the
number of nonlinear finite element analysis required by MCS is usually around 105-107, and so the
computational cost may be prohibitively large.
In this paper, a global limit state function for global seismic reliability of structures is provided, which is the
difference of the limit base shear of structural system minus the total horizontal seismic action. A new
semi-analytical approach combing point estimation method (PEM), pushover analysis with FORM is developed
for analyzing the global seismic reliability of structures. The developed methodology is applied in R.C. frame
structures considering the nonlinear effects. The applied method is also compared with MCS. A numerical
example demonstrates that the approach put forward by this paper can significantly reduce the number of finite
element simulations, and has the same accuracy as that of MCS.

2. GLOBAL LOAD-CARRYING CAPACITY LIMIT STATE FUNCTION OF STRUCTURAL

SYSTEM AND A NEW SEMI-ANALYTICAL METHOD FOR SEISMIC RELIABILITY ANALYSIS

2.1. Global Load-Carrying Capacity Limit State Function of Structural Systems

In Chinese seismic design code of buildings (GB50011-2001), the base shear method is a prevailing approach to
obtain the seismic action for low to medium-rise buildings. In this paper, we take the limit base shear of
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

structure as the global seismic capacity of structural system, and take the total horizontal seismic action as the
seismic demand of structures. Based on these considerations, we propose the following global seismic capacity
limit state function for structural systems:
g (VS , FE ) = VS − FE (2.1)
where VS = limit base shear of structures, FE = the total horizontal seismic action of structures in the base. They
are all random variables, so we can use static reliability theory to conduct the analysis of seismic reliability,
which is a dynamic reliability problem in nature.
The formulation of Eq. (2.1) is the same as the performance function of structural members in the format, in
other words, the limit base shear VS and the total horizontal seismic action FE correspond with the resistance and
load effect of structural members, respectively. Therefore, the classical component reliability methods such as
FORM and SORM (Ditlevsen and Madsen, 1996) can be used to approximately compute the system reliability
of structures.
There is another advantage in applying Eq. (2.1) in seismic reliability analysis: the real constitutive relations of
structural materials and nonlinear effects of structural systems can be considered through the limit base shear VS.
As such, although Eq. (2.1) is linear in the format, the approach based on this formulation is a nonlinear
reliability analysis method in nature.
There are two causes for Eq. (2.1) just to include one load effect, i.e. seismic action. One reason is that both the
mean value and the variability in the seismic intensity are much larger than those of the live loads and wind load,
so the randomness in the live loads and wind load should not be considered, and their characteristic values are
taken, when analyzing the seismic reliability of structures. Therefore, only one single random load, i.e., seismic
action, is included in Eq. (2.1). The second reason is that, when the total horizontal seismic action is computed,
the equivalent total gravity load has included the combination effects of dead loads, live loads and wind load, so
the total horizontal seismic action FE is equivalent to the comprehensive load effect.

2.2. The Limit Base Shear of Structures

The limit base shear of structures VS depends on not only the limit load-carrying capacity of structural members,
but also the constitutive relations of materials, the correlation relationships among structural members, the
correlation relationships between member resistance and loads, load path, system redundancy, structural types,
loading cases of structures, etc. Due to the complexity of the problem, the traditional limit load analysis method
is not suitable to the limit base shear analysis of structures. Instead, we take pushover analysis as the basic tool
to evaluate the limit base shear of structures. Pushover analysis is an incremental static elastoplastic analysis
method under the increasing monotonic load. The advantage of applying pushover analysis in system reliability
evaluation is that it can trace the developing sequences of plastic hinges and so identify the significant failure
modes. Actually, it is an extension of the incremental load approach proposed by Moses (1982) in system
reliability theory of structures.
Much research has proven that the global load-carrying capacity of structures can be approximately modeled by
log-normal distribution. Therefore, in this paper, the limit base shear of structures is also assumed to satisfy
log-normal distribution:
⎛ ln v − λVS ⎞
FVS (v) = Φ ⎜ ⎟ (2.2)
⎜ ζV ⎟
⎝ S ⎠
where λVS and ζ VS are logarithmic mean and logarithmic standard deviation of VS respectively, their
relationships with the mean value and coefficient of variation (COV) of VS are
⎛ μV ⎞
λV = ln ⎜ S ⎟ (2.3)
S
⎜ 1 + δV2 ⎟
⎝ S ⎠
ζ V = ln (1 + δV2
S S
) (2.4)
in which μVS and δVS are mean value and COV of VS respectively.

2.3. Equivalent Static Random Seismic Action

The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

Based on response spectrum of single-degree-of-freedom (SDOF) oscillator, the equivalent static random
seismic action can be described as
Am
FE = α GD = β (T , ξ )GD (2.5)
g
where, G is the equivalent total gravity load, D is an appending factor considering the modeling uncertainty
Am
from, α = β (T , ξ ) is the earthquake effect coefficient, g is the gravity acceleration, Am is the peak
g
ground acceleration (PGA), β (T , ξ ) is the dynamic amplification factor, in which T and ξ are the vibration
period and damping ratio of the oscillator respectively.
Ou et al. (1994) has proven that the random seismic action under deterministic earthquake intensity satisfies
type I extreme value distribution. When the seismic intensity I takes J, the CDF of FE is
FFE ( f | I = J ) = exp {− exp[−α ( f − u )]} (2.6)
where the distribution parameters take the forms of
π
α= (2.7)
6μ F VF
J J

u = μ FJ (1 − 0.45VFJ ) (2.8)
where μ FJ and VFJ are the mean value and COV of FE under the jth intensity respectively. Based on the
research of Ou et al. (1994, 1995), μ FJ = 0.75 FJK , in which FJK is the characteristic value of the horizontal
seismic action under the jth intensity, VFJ = 0.73 . Put the above results in Eqs. (2.7) and (2.8), we can obtain
the final results: α = 2.34 / FJK , u = 0.5 FJK .

2.4. A New Semi-Analytical Method for Global Seismic Reliability Analysis of Structures
From the above statements we can come to the conclusion that the central problem of applying Eq. (2.1) in
analysis of global seismic reliability is how to get the moment information of limit base shear VS, since the
probability model and distribution parameters of random seismic action FE have been certain. Most of the
available research generally makes use of Monte Carlo simulation combined with pushover analysis to obtain
the samples and then the estimators of statistical moments. To reduce the variance of simulation, many
techniques have been introduced, such as importance sampling, systematic sampling, antithetic variates, etc.
Unfortunately, the computation cost of these random simulation approaches based on MCS is still extremely
large. In next section, we will propose a random pushover analysis approach based on point estimation method
(PEM), which can significantly reduce the number of nonlinear finite element analysis while keeping the same
accuracy as MCS.
After obtaining the statistical moments of VS by numerical analysis techniques, the approximate analytical
method such as FORM/SORM can then be applied to solve Eq. (2.1). For this purpose, we suggest herein a new
semi-analytical method as follows:
(1) Obtaining the statistical moments of limit base shear VS by using random pushover analysis based on point
estimation method;
(2) Fitting the PDF of VS according to its statistical moments. If the probability model of VS can be decided a
prior, then this step can be omitted;
(3) Using FORM and/or SORM to solve Eq. (2.1).

3. RANDOM PUSHOVER ANALYSIS BASED ON POINT ESTIMATION METHOD AND ITS

APPLICATIONS IN PROBABILISTIC ANALYSIS OF STRUCTURAL LIMIT BASE SHEAR

3.1. Characteristics and Difficulties of Probabilistic Analysis of Structural Limit Base shear
Since there are randomness and uncertainties in many factors influencing structural limit base shear VS, we
should make use of probability theory to analyze and compute the statistical moments of VS. Put all factors
influencing VS together into a basic random vector X = [ X 1 , X 2 ," , X n ]T , then VS can be generally denoted as
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

an implicit and nonlinear function of X:

VS = h( X) = h( X 1 , X 2 ," , X n ) (3.1)
There exist the following characteristics and difficulties in probabilistic analysis of limit base shear of structures:
(1) Multiple scales. The factors influencing VS can be classified as five scales, i.e. material scale, section scale,
member scale, sub-structure scale and structural system scale. The random information propagates across the
above scales from bottom to top, so as to make the analysis of uncertainty propagation very difficult. (2)
Nonlinearity. The limit base shear of structures is a nonlinear function of the above influencing factors in nature,
especially in the case that the structure goes into the severe damage or even collapse. (3) Correlation. There may
be correlating relations to some degree between the factors in the same scale, or between the factors across the
different scales. (4) Highly implicitness. The limit base shear is usually determined by numerical analysis
techniques, such as FEM. Therefore, the limit base shear is a highly implicit function of basic random variables,
generally takes the form of a black box.
Due to the above difficulties, the conventional methods for analysis of uncertainty propagation, such as
mean-value first order second moment (MVFOSM) method and Monte Carlo simulation, have the
disadvantages of low accuracy or too much computational cost. Therefore, we should look for some approaches
whose accuracy and efficiency can all be accepted by engineering community, among which the point
estimation method is such an approach.

3.2. Point Estimation Method Based on Nataf transformation

Point estimation method (PEM) was proposed by Rosenblueth (1975) to approximate the lower-order moments
of functions of random variables. It is a special case of numerical quadrature based on orthogonal polynomials.
For normal variables, it corresponds to Gauss-Hermite quadrature. While the point estimate method is popular
in practice, it has many detractors. Numerous modifications or improvements have been made for the original
PEM. However, the early developments of PEM are all undertaken in the original space of random variables,
requiring the higher order moments of random variables without considering the distribution information. To
overcome these shortcomings, Zhao and Ono (2000) introduced a new point estimation method based on
Rosenblatt transformation in which the numerical quadrature is completed in standard normal space.
Unfortunately, Rosenblatt transformation cannot deal with the case of random variables with given marginal
distributions and correlation information. In this paper, we introduce Nataf transformation (Liu and Der
Kiureghian, 1986) into Zhao-Ono point estimation method.
The forward Nataf transformation TN can be denoted by
TN : u = L−01Φ −1[FX (x)] (3.2)
where, x and u are the realizations of n dependent non-normal random variables X and independent
standard normal random variables U , respectively; Φ −1 () represents the column vector composed of all
inverse functions of standard normal random variables; FX (x) is the column vector comprised of CDFs of
random variables X i (i = 1," , n ) ; L 0 is the lower triangle matrix of Choleski decomposition of correlation
coefficients matrix R 0 of dependent normal random vector Y = Φ −1[FX (x)] , i.e. R 0 = L 0 LT0 ; the
relationships between the elements ρ 0,ij of R 0 and the elements ρ ij of R , the correlation coefficients
matrix of X , are
ρ0,ij = Fij ρij (3.3)
where, the coefficient Fij is function of correlation coefficient ρ ij and marginal distributions FX i ( xi ) and
FX j ( x j ) of random variables X i and X j . In general, Fij ≥ 1 . Liu and Der Kiureghian (1986) gave the
practical formula for computing coefficient Fij corresponding to different probability distributions.
The inverse Nataf transformation TN−1 can be denoted by
TN−1 : x = FX−1[Φ(L 0u)] (3.4)
−1
where, F (⋅)
X represents the column vector composed of all inverse functions of random
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

variables X i (i = 1," , n ) ; Φ(⋅) denotes the column vector comprised of all CDFs of standard normal random
variables.
The computation of the first two moments of random function h( X) is undertaken in the standard normal
space by using the inverse Nataf transformation,
μh = ∫ h(x) f X (x)dx = ∫ h ⎡⎣TN−1 (u) ⎤⎦ ϕn (u)du (3.5a)

{
σ h2 = ∫ [ h(x) − μ Z ] f X (x)dx = ∫ h ⎡⎣TN−1 (u) ⎤⎦ − μ Z ϕ n (u)du }
2 2
(3.5b)
where, μh and σ h are mean value and standard deviation of random variable h respectively; f X (x) is the
joint PDF of random vector X ; ϕn (u) is the joint PDF of n-dimension standard normal random variables.
For single-variable function h( X ) , Nataf transformation reduces to iso-probability
−1
transformation x = F [Φ (u )] , and then Eq. (3.5) can be approximated by using Gauss-Hermite numerical
X
quadrature in standard normal space:
m
μh ≈ ∑ Pj h { FX−1[Φ (u j )]} (3.6a)
j =1
m
σ h2 ≈ ∑ Pj ⎡⎣ h { FX−1[Φ (u j )]} − μh ⎤⎦
2
(3.6b)
j =1

where, u j ( j = 1, " , m) are estimation points; Pj are corresponding weights; m is the number of
estimation points.
The abscissas x j and weights w j of Gauss-Hermite quadrature with weight function exp(−u 2 ) are listed in
Table 3.1.

Table 3.1 Abscissas and weights for Gauss-Hermite integration

Order ( m ) Abscissas ( x j ) Weights ( w j ) Order ( m ) Abscissas ( x j ) Weights ( w j )
1 0 1.7724538509 0 0.9453087205
2 ±0.707106781 0.8862269255 5 ±2.020182871 0.0199532421
0 1.1816359006 ±0.958572465 0.3936193232
3
±1.224744871 0.2954089752 ±2.350604974 0.0045300100
±1.650680124 0.0813128354 6 ±1.335849074 0.1570673203
4
±0.524647623 0.8049140900 ±0.436077412 0.7246295952
0 0.8102646176 ±2.930637420 0.0001996041
±2.651961357 0.0009717812 ±1.981656757 0.0170779830
7 8
±1.673551629 0.0545155828 ±1.157193712 0.2078023258
±0.816287883 0.4256072526 ±0.381186990 0.6611470126

The estimating points u j and weights Pj in Eq. (3.6) can be obtained according to Table 3.1:
wj
u j = 2 x j , Pj = (3.7)
π
For a function of random variables h( X) , it is approximated by a non-product function proposed by Zhao and
Ono (2000):
n
h(X) ≈ h′(X) = ∑ ( H i − H μ ) + H μ (3.8)
i =1
in which,
H μ = h(μ) = h( μ1 ," , μi ," , μn ) (3.9)
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

H i = h ⎡⎣TN−1 (ui ) ⎤⎦ = H (ui ) = H (uμ1 , uμ 2 ," , uμi −1 , ui , uμi +1 ," , uμ n ) (3.10)

where μ represents the vector in which all the random variables take their mean values; u i represents the
vector in which only ui is a random variable, while other variables take the corresponding transformed values
of their mean values in standard normal space; uμ j ( j ≠ i ) is the jth element of the transformed vector u μ
who corresponds the vector μ in standard normal space u ; H (u) = h[TN−1 (u)] is the formulation of random
function h(x) in standard normal space based on Nataf transformation.
Note that we have introduced Nataf transformation into Eq. (3.8), so it is different from that proposed by Zhao
and Ono, although their forms are the same.
Based on Eq. (3.8), the first two moment of multi-variable random function h( X) can be estimated by
n
μ h ≈ ∑ ( μi − H μ ) + H μ (3.11a)
i =1
n
σ h2 ≈ ∑ σ i2 (3.11b)
i =1

where μi and σ i are mean value and standard deviation of H i by using point-estimation of single-variable
function.

3.3. Application of Point-Estimation Based Random Pushover Analysis in Statistical Moments Computation
of Structural Limit Base shear
The nature of point estimation method is that it is a kind of deterministic sampling in standard normal space
according to Table 3.1 and Eq. (3.7), whose total sampling number is m × n , in which m is the order of
numerical quadrature, and n is the number of basic random variables. Compared with the huge sampling
number of Monte Carlo simulation, obviously the sampling number of point estimation method reduces
dramatically. On the other hand, from the viewpoints of experimental design, point estimation method belongs
to a kind of deterministic experimental design.
Since point estimation method makes use of deterministic sampling or experimental design techniques, we can
combine this method with deterministic finite element analysis, herein the pushover analysis, to compute the
statistical moments of limit base shear of structures. We call this combination of pushover analysis with point
estimation method as “random pushover analysis (RPA)”, the detailed implementation steps are as follows:
(1) Building the finite element model of structure;
(2) Determining the probability distribution types and their distribution parameters of basic random variables
X that influence the limit base shear VS of structures;
(3) Generating structural samples by sampling of the basic random variables X according to Table 3.1 and Eq.
(3.7);
(4) Conducting pushover analysis for each structural sample to derive its base shear-top displacement curve,
from which the limit base shear is obtained;
(5) Computing the statistical moments of limit base shear VS according to Eqs. (3.8) to (3.11).

4. NONLINEAR SEISMIC RELIABILITY ANALYSIS FOR GLOBAL LOAD-CARRING CAPACITY

OF STRUCTURES BASED ON FORM
After obtaining the statistical moments of structural limit base shear VS , we can solve for Eq. (2.1) to make the
analysis of nonlinear global load-carrying capacity-based seismic reliability of structures by using structural
component reliability methods. Notice that VS conforms to log-normal distribution, while FE satisfies Type-I
extreme value distribution, therefore the first order second moment (FOSM) method cannot be utilized which
only considers the first two moments of basic random variables. Instead, the full distribution methods, such as
FORM and SORM (Ditlevsen and Madsen, 1996), should be used.
Let’s group VS and FE into a vector Y = [VS , FE ]T , then Eq. (2.1) can be reformulated as
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

g (Y) = g (VS , FE ) (4.1)

The failure probability of structures is calculated by the integral,
p f = P[ g (Y) ≤ 0] = ∫ f Y ( y ) dy (4.2)
g ( y )≤0

where f Y (y ) represents the joint PDF of vector Y , y is the realization of Y .

Radom vector Y should be firstly transformed into standard normal space by forward Nataf transformation,
u = TN (y ) ; and then, performance function g (y ) in y space is transformed into G (u) in u space:
g (y ) = g ⎡⎣T −1 (u) ⎤⎦ = G (u) (4.3)
Then Eq. (4.2) is changed into
pf = ∫ ϕ n (u)du (4.4)
G ( u )≤ 0

where ϕn (u) is the joint PDF of n-dimension standard normal random variables.
After variable transformation, the linear performance function Eq. (2.1) in original random space is changed into
a nonlinear performance function (4.1) in standard normal space. In u space, the nonlinear function G (u) is
approximately expanded as one-order Taylor series:
G (u) ≈ G (u∗ ) + ∇G (u* )T (u − u∗ ) = ∇G (u* )T (u − u∗ ) (4.5)
Eq. (4.5) is a linear function. For linear functions and standard normal variables, it has been proven that
Hasofer-Lind reliability index β HL equals Cornell reliability index β C = μG / σ G , then
μG −∇G (u∗ )T u∗ −∇G (u∗ )T u∗
β HL = = = = α*T u∗ (4.6)
σG ∇G (u ) ∑ U ∇G (u )
∗
∗ T ∗ || ∇G (u ) ||
where ∑ U is the covariance matrix of standard normal variables, which is an identity matrix, i.e. ∑ U =I;
α* is the sensitivity vector at the design point u∗ :
α* = −∇G (u* ) / || ∇G (u* ) || (4.7)
∗
Since the design point u is unknown a prior, it needs numerical algorithms to solve by iterations. This paper
makes use of well-known HLRF algorithm:
⎛ G (ui ) ⎞
ui +1 = ⎜⎜ + α iT ui ⎟⎟ α i (4.8)
⎝ ∇G (ui ) ⎠
where || ⋅ || is l2-norm; α i = −∇G (u i ) / || ∇G (ui ) || is the sensitivity vector at the ith iteration; the gradient
∇G (ui ) is computed according to the chain rule of differential:
T
∇G (ui ) = ⎡⎣ J y ,u (y i , ui ) ⎤⎦ ∇g (y i ) (4.9)
where ∇g (y i ) is the gradient of performance function g (y ) in original random space at y i ; J y ,u is the
Jacobian matrix of inverse Nataf transformation, T represents transpose of matrix.
After obtaining the design point u∗ and reliability index β HL , the failure probability in Eq. (4.4) can be
approximated by
p f ≈ Φ ( − β HL ) (4.10)

5. Applications of the methodology in R.C. frame buildings

5.1. Basic Data of the Structure

The analyzed structure shown in Figure 1 is a three-bay and six-storey reinforced concrete frame building, the
sizes of beams and columns are listed in Table 5.1. The uniformly distributed load on the top floor is 16.6 KN/m,
the loads on other floors are all 20.06 KN/m.
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

16.6KN/m
F6
20.06KN/m 3.6m
F5
20.06KN/m 3.6m
F4
20.06KN/m 3.6m
F3
20.06KN/m 3.6m
F2
20.06KN/m 3.6m
F1
3.6m

6.6m
6.6m 2 7m
2.7m 6.6m
6.6m

Figure 1 Three-bay and six-storey R.C. frame

Table 5.1 Sizes of structural members

Strength Grade Steel Grade Steel Grade
Members Height (mm) Width (mm)
of Concrete of rebar of hoops
Interior Columns 500 500 C30 HRB335 HPB235
Exterior Columns 500 500 C30 HRB335 HPB235
Main Beams 600 300 C30 HRB335 HPB235
Side Beams 500 200 C30 HRB335 HPB235

5.2. Probability Models and Statistical Parameters of Basic Random Variables

According to the research of Ou et al. (1994, 1995), the random horizontal seismic forces acting on structural
floors computed by base shear method all satisfy Type-I extreme value distribution. The statistical parameters of
storey-level seismic forces according to Eqs. (2.7) and (2.8) are listed in Table 5.2. Ou et al. (1995) assumed that
the storey-level seismic forces were perfect correlation. In order to investigate the effects of the variation of total
horizontal seismic action and correlation relation of storey-level seismic forces on the limit base shear of
structures, this paper assumes the variation of total horizontal seismic action changes from 0.1 to 1.0; while the
correlation coefficient of storey-level seismic forces changes from 0 to 0.9. The randomness considered in
structural resistance includes yielding strength fc of concrete, yielding strength fy, elasticity modulus E and the
second stiffness factor α of steel, their probability models and distribution parameters are also listed in Table
5.2.

Table 5.2 Statistics and probability types of basic random variables

Correlation
RVs Mean value Std COV Types
coefficient
fc (N/mm2) 14.3 2.86 0.2
fy (N/mm2) 363 72.6 0.2 Log-normal
0.3
E (N/mm2) 2×105 0.4×105 0.2
α 0.05 0.001 0.2 normal
F1 (KN) 33.513 3.351~33.513 0.1~1.0
F2 (KN) 58.379 5.838~58.379 0.1~1.0
F3 (KN) 80.345 8.035~80.345 0.1~1.0 Type-I
0.0~0.9
F4 (KN) 98.292 9.829~98.292 0.1~1.0 largest
F5 (KN) 111.170 11.117~111.170 0.1~1.0
F6 (KN) 91.721 9.172~91.721 0.1~1.0

5.3. Probabilistic Analysis of Limit Base shear of the Structure

The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

By applying the random pushover analysis based on Nataf transformation proposed in this paper in probabilistic
analysis of limit base shear of the R.C. frame, the changing rules of mean value and standard deviation of limit
base shear with COV of total horizontal seismic action and correlation coefficient of storey-level seismic forces,
as shown in Figures 2 and 3.

820 100
Mean value of limit base shear (kN)

90

Std of limit base shear (kN)

800
80
780
70
760 60

740 50

40
720
30
700
20

680 10
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Coefficient of variation of seismic action Coefficient of variation of seismic action
(a) Mean value of limit base shear (b) Std of limit base shear
Figure 2 Changing of mean value and standard deviation of limit base shear with COV of seismic action

730 45
Mean value of limit base shear (kN)

720
Std of limit base shear (kN) 40
710

35
700

690
30

680
25
670

660 20
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Correlation coefficient of storey-level seismic forces Correlation coefficient of storey-level seismic forces

(a) Mean value of limit base shear (b) Std of limit base shear
Figure 3 Changing of mean value and standard deviation of limit base shear with correlation coefficient of
storey-level seismic forces

From Figure 2 it is evident that the mean value and standard deviation of limit base shear tend to be large with
the increasing of COV of total horizontal seismic action. This result is predictable since the pushover results of
structures depend on the loading cases and lateral load patterns. Obviously, the more is the randomness in the
total seismic action, the more are the statistics of limit base shear.
From Figure 3 we can see that the mean value and standard deviation of limit base shear do not necessarily
increase with the correlation coefficient of storey-level seismic forces. When correlation coefficient ρ > 0.6 , the
mean value of limit base shear becomes smaller; while when ρ > 0.7 , the standard deviation becomes smaller.
If we assume the storey-level seismic forces are all perfect correlated, then the conservative results will be
obtained. In other words, the storey-level seismic forces are not perfect correlated in nature, since the
combinations of dead load, live load and wind loads have been considered when calculating the characteristic
value of total seismic actions, thus it lead to the partial correlation between storey-level seismic forces.

5.4. Seismic Reliability of Global Load-Carrying Capacity of the Structure

We herein only analyze the seismic reliability of structures under major earthquakes. The first two moments of
the total horizontal seismic action according to base shear method and Eqs. (2.7) and (2.8) are listed in Table
5.3.
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

Table 5.3 Statistics of total horizontal seismic action

Characteristic value (KN) Mean value (KN) Std (KN) COV
631.224 473.418 236.709 0.5

By using the semi-analytical method proposed in this paper, the changing rules of the reliability index of global
load-carrying capacity of the structure with the COV of total seismic actions and the correlation coefficient of
storey-level seismic forces, as shown in Tables 5.4 and 5.5 as well as Figures 4 and 5, are obtained. To
investigate the accuracy and efficiency of the proposed method, Monte Carlo simulations are also conducted
with simulation number N = 106 . The results of MCS are also listed in the corresponding tables and figures.

Table 5.4 Results of global seismic reliability index considering the variations of total seismic action
δ Fi 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00
β FORM 2.8645 1.9306 1.5394 1.3209 1.1716 1.0667 0.9836 0.9248 0.8537 0.8220
β MCS 2.8699 1.9335 1.5380 1.3216 1.1729 1.0684 0.9874 0.9300 0.8572 0.8280

Table 5.5 Results of global seismic reliability index considering the correlation of storey-level seismic forces
ρ Fi 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90
β FORM 0.9032 0.9695 1.0161 1.0492 1.0704 1.0821 1.0838 1.0741 1.0447 0.9772
β MCS 0.9003 0.9706 1.0189 1.0493 1.0710 1.0844 1.0824 1.0756 1.0453 0.9774

3
FORM
MCS
2.5
seismic reliability index

1.5

0.5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Coefficient of variation of seismic action
Figure 4 Changing of global seismic reliability index with the COV of seismic action

1.15
FORM
MCS
1.1
seismic reliability index

1.05

0.95

0.9
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Correlation coefficient of storey-level seismic forces
Figure 5 Changing of global seismic reliability index with the correlation coefficient of seismic forces
The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

From Tables 5.4 and 5.5, it is evident that the results by using the method proposed in this paper has nearly the
same accuracy as that of MCS, while the number of nonlinear finite element analysis in our method is only 50.
It is shown from Table 5.4 and Figure 4 that the global reliability index tend to become smaller with the
increasing of COV of total seismic action, and that the speed of decreasing becomes more rapidly. This
illustrates that the more the randomness in total seismic action is, the smaller the reliability index is, and hence
the larger the failure probability is, which makes the structure more unsafe.
It is shown from Table 5.5 and Figure 5 that the global reliability index tends to become larger with the
increasing of correlation coefficient of storey-level seismic forces when ρ < 0.6 . When ρ = 0.6 , the global
reliability index attains the largest value. Otherwise, when ρ > 0.6 , the global reliability index becomes
smaller. This result shows that not considering the correlations between storey-level seismic forces or assuming
their perfect correlation will lead to conservative results in any case.

6. CONCLUSIONS

This paper built up a global load-carrying limit state function based on limit base shear, put forward a new
semi-analytical method to analyze the nonlinear global seismic reliability of structures, which comprises point
estimation method, pushover analysis and FORM. By applying the proposed methodology in reinforced
concrete frame buildings, some changing rules of global seismic reliability of the structure with COV of total
seismic action and correlation coefficient of storey-level seismic forces were obtained. Through the
comprehensive study in this paper, some conclusions are derived as follows:
(1) The method of system reliability analysis based on global load-carrying capacity is simple, practical and
efficient. On the one hand, this method can overcome many difficulties of conventional system reliability
theory; on the other hand, it can be linked with the current design codes so that the static reliability method can
solve the difficult dynamic seismic reliability problems.
(2) Semi-analytical method is a compound approach which combines numerical simulation or integration
methods, deterministic finite element analysis and approximate analytical reliability methods such as
FORM/SORM. This method is a practical and efficient approach to conduct the seismic reliability analysis of
large-scale and complex structures. The practice shows that this method has the same accuracy as MCS.
(3) Random pushover analysis is a good alternative for probabilistic seismic capacity analysis (PSCA) of
structures. Meanwhile, it is also an efficient tool of uncertainty propagation structural system.
(4) The variation of total seismic action and the correlation between the storey-level seismic forces have great
effect on the limit base shear and the global seismic reliability of structures.
(5) The methodology proposed in this paper can be extended to probabilistic seismic performance assessment
and design of structures based on global reliability.

7. ACKNOWLEDGEMENTS
The support of National Science Foundation of China through projects (Grant No. 90715021, 50678057,
50108005) is greatly appreciated. Any opinions, findings, and conclusions or recommendations expressed in this
material are those of the authors and do not necessarily reflect those of the National Science Foundation of
China.

REFERENCES

Ditlevsen, O. and Madsen, H.O. (1996). Structural reliability methods. John Wiley & Sons, New York, NY.
Gorman, M.R. and Moses, F. (1979). Direct estimates of structural system reliability. Proceeding of ASCE the
7th Conference on Electronic Computation, St. Louis, Missouri, Aug., 432-445.
Grigoriu, M. (1983). Approximate analysis of complex reliability problems. Structural Safety 1:4, 277-288.
Krawinkler, H. and Seneviratna, G.D.P.K. (1998). Pros and cons of a pushover analysis of seismic performance
evaluation. Engineering Structures 20:4-6: 452-464.
Li, G. and Cheng, G.-D. (2004). Performance-Based Seismic Design: Theory, Methods and Applications.
Science Press, Beijing.
Li, G.-Q. and Li, J.-J. (2002). A semi-analytical simulation method for reliability assessments of structural
systems. Reliability Engineering and System Safety 78:3, 275-281.
 The 14th World Conference on Earthquake Engineering
October 12-17, 2008, Beijing, China

Li, G.-Q., Liu, Y.-S. and Zhao, X. (2006). Advanced Analysis of Steel Frames and System Reliability Based
Design. Building Industry Press of China, Beijing.
Li, J.-J. and Li, G.-Q. (2004). Reliability-based integrated design of steel portal frames with tapered members.
Structural Safety 26:2, 221-239.
Liu, P.L. and Der Kiureghian, A. (1986). Multivariate distribution models with prescribed marginals and
covariances. Probabilistic Engineering Mechanics 1:2, 105-112.
Moses, F. (1982). System reliability developments in structural engineering. Structural Safety 1:1, 3-13.
National Standard of China P.R. (2001). Seismic Design Code of Buildings (GB50011-2001). Building Industry
Press of China, Beijing.
Nowak, A.S. and Zhou, J.H. (1988). Integration formulas to evaluate functions of random variables. Structural
Safety 5:4, 267-284.
Nowak, A.S. and Zhou, J.H. (1990). System reliability models for bridges. Structural Safety 7:2-4, 247-254.
Onoufriou, T. and Forbes, V.J. (2001). Developments in structural system reliability assessments of fixed steel
offshore platforms. Reliability Engineering and System Safety 71:2, 189-199.
Ou, J.-P., Duan, Y.-B. and Liu, H.-Y. (1994). Random seismic action and its statistics. Journal of Harbin
Building Engineering and Architecture 27:5, 1-10. (in Chinese).
Ou, J.-P. and Duan, Y.-B. (1995). Seismic reliability analysis and optimum design of high-rise building
structures. Earthquake Engineering and Engineering Vibration 15:1, 1-13. (in Chinese).
Ou, J.-P., Hou, G.-L. and Wu, B. (2001). Probabilistic pushover analysis and its applications in seismic
reliability evaluation of structural systems. China Building Structures Journal 22:6, 81-86. (in Chinese).
Ou, J.-P., Duan, Z.-D. and Xiao, Y.-Q. (2003). Safety Assessment of Platforms: Theory, Methods and
Applications. Science Press, Beijing. (in Chinese).
Rosenblueth, E. (1975). Point estimates for probability moments. Proceedings of the National Academy of
Science 72:10, 3812-3814.
Sigursdon, G., Skjong, R., Skallerud, B., et al. (1994). Probabilistic collapse analysis of jackets. International
Conference on Offshore Mechanics and Arctic and Engineering, Houston.
Zhao,Y.G. and Ono, T. (1998). System reliability evaluation of ductile frame structures. ASCE Journal of
Structural Engineering 124:6, 678-685.
Zhao, Y.G. and Ono, T. (2000). New point estimates for probability moments. ASCE Journal of Engineering
Mechanics 126:4, 433-436.

View publication stats