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

net/publication/262415412

Fatigue reliability analysis for structures with known loading trend

Article in Structural and Multidisciplinary Optimization · February 2014

DOI: 10.1007/s00158-013-1044-0

CITATIONS READS

15 813

5 authors, including:

Zhen hu Daniel Conrad

University of Michigan-Dearborn hussmann
183 PUBLICATIONS 3,505 CITATIONS 13 PUBLICATIONS 206 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Zhen hu on 20 May 2014.

The user has requested enhancement of the downloaded file.

Struct Multidisc Optim
DOI 10.1007/s00158-013-1044-0

RESEARCH PAPER

Fatigue reliability analysis for structures

with known loading trend
Zhen Hu · Xiaoping Du · Daniel Conrad · Ray Twohy ·
Michael Walmsley

Received: 3 September 2013 / Revised: 10 December 2013 / Accepted: 21 December 2013

© Springer-Verlag Berlin Heidelberg 2014

Abstract Variations, such as those in product operation It is desirable to assess the fatigue life probabilistically
environment and material properties, result in random rather than deterministically. The most commonly used
fatigue life. Variations in material fatigue properties depend probabilistic assessment method is the fatigue reliability
on stochastic stress responses due to their nonlinear relation- analysis, which provides the probability that the actual
ships with other random variables such as stochastic loading fatigue life is greater than a desired life.
and dimensions. In this work, an efficient fatigue reliability Fatigue reliability analysis methods are classified into the
analysis method is developed to accommodate those uncer- following three categories:
tainties for structures under cyclic loads with known loading • Strain-life based method (Correia et al. 2013; Zhang
trend. To reduce the computational cost, the method incor-
et al. 2013)
porates the fatigue life analysis model and the saddlepoint
The method predicts fatigue life according to the
approximation method with the fast integration method. The
strain response, which is usually related to the initial
new method is applied to the fatigue reliability analysis of a
crack.
cantilever beam and a door cam. The results show high accu- • Stress-life based method (Asi and Yeşil 2013; Sousa
racy and efficiency of the proposed method benchmarked
et al. 2013; Lee and Song 2012; Li and Low 2012;
with Monte Carlo Simulations.
Rathod et al. 2012)
Fatigue life is evaluated based on the material S-
Keywords Fatigue reliability · Stress-dependent · Monte
N curve. The initiation and propagation of the crack
Carlo simulation · Cyclic load
are not differentiated from each other in the stress-life
model. Only the total fatigue life is considered.
• Fracture mechanics method (Beck and Gomes 2013;
1 Introduction
Chan et al. 2012; Larsen et al. 2013)
Fracture mechanics methods are used to estimate if
Fatigue life assessment is a critical issue during the design
a crack grows to a critical size. This method usually
process for many products. Due to inherent uncertainties,
combines the strain-life method to estimate the crack
fatigue life always varies around the designed fatigue life.
initiation.
Z. Hu · X. Du () This work employs the stress-life based fatigue reliabil-
Department of Mechanical and Aerospace Engineering, ity analysis method, and the effects of uncertainties in the
Missouri University of Science and Technology,
400 West 13th Street, Toomey Hall 290D,
design variables and the S-N curve on the fatigue life are
Rolla, MO 65401, USA investigated. The relevant research is reviewed below.
e-mail: dux@mst.edu In addition to aforementioned methods (Asi and Yeşil
2013; Lee and Song 2012; Li and Low 2012; Rathod et al.
D. Conrad · R. Twohy · M. Walmsley
2012; Sousa et al. 2013), other methods have also been pro-
Division of Quality and Reliability,
Hussmann Corporation, posed. For instance, Guo and Chen (Guo and Chen 2013)
Bridgeton, MO 63044, USA developed a fatigue reliability analysis method for steel
 Z. Hu et al

bridges based on the long-term stress monitoring. Liu (Liu The new method can account for uncertainties in both
and Mahadevan 2009) proposed an efficient time-dependent design variables and stress-dependent uncertainties in mate-
fatigue reliability analysis method by using the moment rial fatigue properties. With the saddlepoint approximation
matching method and the First Order Reliability Method (SPA) (Huang and Du 2006) imbedded in the fast integra-
(FORM). A unimodal distribution characterized by four tion (Wen and Chen 1987), the method can produce a quick
parameters was introduced by Low (2013) in predicting the and accurate solution. The information required (inputs) and
uncertainty in fatigue damage. To account for the correla- the outcome of the method are summarized below.
tion effect of fatigue reliability, a fast reliability assessment Input:
approach was proposed based on the detail fatigue rating
• Distributions of random input variables (dimensions,
method (Huang et al. 2013a). The Kriging and radial basis
loading, etc.) for stress responses
functions were applied to the fatigue reliability analysis of a
• Distributions of random fatigue material properties
wire bond structure by Rajaguru et al. (2012). Baumert and
• Cyclic loading trend
Pierron (2012) studied the implication of fatigue properties
of batteries on the reliability of flexible electronics. To over- Outputs:
come the expensive computational effort of Monte Carlo
• The distribution of fatigue life
simulation (MCS), Norouzi and Nikolaidis (2012) presented
• Fatigue reliability
an efficient fatigue reliability analysis method for structures
subjected to a dynamic load. A review of the fatigue life analysis under known loading
Many probabilistic models have also been developed to trend is given in Section 2, followed by uncertainty anal-
model the statistical characteristics of the S-N curve (Ayala- ysis for fatigue life in Section 3. Section 4 discusses the
Uraga and Moan 2007; Bengtsson and Rychlik 2009; Hasan proposed method, whose numerical procedure is summa-
et al. 2012; Jha et al. 2013; Lee et al. 2013; Wei et al. 2011a, rized in Section 5. Two numerical examples are presented in
b, 2013; Xu et al. 2012; Gu and Moan 2002). Studies of Section 6, and conclusions are made in Section 7.
the S-N curve indicate that material fatigue properties are
uncertain with stress-dependent characteristics (Kam et al.
1998; Kamiński 2002; Le and Peterson 1999; Liu and 2 Fatigue life analysis with known loading trend
Mahadevan 2007; Liao et al. 1995; Ni and Zhang 2000;
Pascual and Meeker 1999; Rowatt and Spanos 1998). Since This work is for structures under cyclic load with known
stress responses are usually also uncertain, material fatigue loading trend. As shown in Fig. 1, the known loading trend
properties are uncertain factors whose stochastic nature means that the same trend of the load repeats cycle by
is governed by other uncertainties. The stress-dependent cycle and that each cycle of the load is identical. As stress
fatigue properties make the fatigue reliability analysis dif- responses in one load cycle is predictable with a mathemati-
ferent from and more difficult than regular reliability analy- cal model or computer aided engineering (CAE) simulation
sis problems. model, the trend of the stress responses is also known.
The stress-dependent uncertainty in fatigue properties As mentioned previously, this assumption is applicable for
has not been sufficiently considered in the majority of many problems.
fatigue reliability analysis methods. A few studies, such as Many fatigue life prediction methods (Fitzwater and
the two methods developed by Liu and Mahadevan (2009), Winterstein 2001; Huang and Moan 2007; Ko 2008; Kwon
have concentrated on the reliability analysis with the stress- and Kareem 2011; Li and Wang 2012) and fatigue damage
dependent properties, and their accuracy and efficiency accumulation models (Cruzado et al. 2013; El Aghoury and
can be further improved. For instance, the assumption of Galal 2013; Suyuthi et al. 2013) are available. We herein
known stress distribution in the methods can be released
by relating the fatigue reliability with basic random design
variables.
Load Cycle 1 Cycle 2
The objective of this work is to improve the accuracy
and efficiency of fatigue reliability analysis for special
problems where structures are under cyclic loads with
known loading trend. This kind of problem is common in
many applications, especially for mechanisms with cyclic
motions (Huang et al. 2013b; Petrescu and Petrescu 2013),
for example, the transmission shaft under periodic load-
t
ings (Cihan and Yuksel 2013; Hu and Du 2012), cams
with known motion trajectory, and linkage mechanisms. Fig. 1 Illustration of cyclic load with known loading trend
Fatigue reliability analysis for structures

briefly review the fatigue life analysis model for structures o

in which smax o
and smin are the maximum and minimum
with known loading trend. o
values of s , respectively.
Let x = [x1 , x2 , · · · , xn ] be a vector of input variables The Gerber’s correction is Kihl and Sarkani (1999)
to the nonlinear function or simulation model for stress  2
responses as follows: sa sm
+ =1 (6)
s su
so = g(x) (1)
It is usually recommended that the Goodman correction is
 o o g(x) iso the stress responses function, and s =
where o

s1 , s2 , · · · , sm are blocks of stress responses in one cycle used for brittle materials and that the Gerber’s correction is
of the cyclic load. It should be noted that the left-hand used for ductile materials. After the mean value correction
side of (1) is a vector because one cycle of the cyclic load is made, the number of cycles to failure at stress level s is
may contain multiple loading peaks as will be seen in the then computed by
numerical examples. N = h(s) (7)
When the stress response is available and deterministic,
the fatigue life analysis is straightforward. The most com- where h(s) is obtained from the S-N curve and is a function
monly used model is the Palmgren-Miner’s rule (Siddiqui of stress level s.
and Ahmad 2001), which is given by Siddiqui and Ahmad With the Palmgren-Miner’s rule, the fatigue life is esti-
(2001) mated by

m
ni
D= (2) 1 1
LF = = m (8)
Ni
i=1 1/N1 + 1/N2 + · · · + 1/Nm 
1/Nj
where D is the accumulative fatigue damage, m is the num- j =1
ber of stress blocks, ni is the number of stress cycles at stress
where LF is the number of load cycles or fatigue life, and
level si , and N i is the number of cycles to failure at stress m
level si . N i is obtained from the constant amplitude fatigue j =1 1/Nj is the fatigue damage in one cycle.

experiment.
In this work, we use the Palmgren-Miner’s rule for
3 Uncertainty analysis of fatigue life
fatigue damage analysis. However, other fatigue damage
analysis methods can also be used with the proposed
3.1 Uncertainties in stress responses
method. Since the fatigue experiments are conducted under
constant amplitude loadings, the mean value corrections are
The fatigue life analysis model in Section 2 is in a determin-
usually applied before evaluating the fatigue damage using
istic form. In reality stress responses from one product to
the Palmgren-Miner’s rule (Siddiqui and Ahmad 2001).
another vary inevitably even if the design is the same. The
Many empirical corrections were developed in the past
stress variations stem from variations in stress analysis input
decades. The most widely accepted corrections include the
variables, for instance, stochastic loading, manufacturing
Goodman’s and the Gerber’s corrections. The two correc-
imprecision, and other noises in the operating environment.
tions relate the alternating stress amplitude to the mean
We divide input variables into deterministic variables
stress response with the ultimate tensile strength (Aygül
d and random variables X. The stress response is then
et al. 2013).
presented by
For a general stress response so (a component of so ), the
Goodman’s correction is given by Wang and Sun (2005) So = g(X, d) (9)
sa sm
+ =1 (3)
s su Output variables So become random variables with distri-
where su is the ultimate tensile strength, s is the stress butions governed by the nonlinear function g(·) and the
response after correction, and sa and sm are the alternating distributions of X. The cumulative distribution function
stress amplitude and the mean stress, respectively, which are (CDF), or the probability that So , which is a component of
given by So , is less than a specific value s, is then computed by

o
smax − smin
o
sa = (4) Pr{S o ≤ s} = f (x)dx (10)
2
and S 0 ≤s
o
smax + smin
o
in which Pr{·} stands for a probability, and f (x) is the joint
sm = (5)
2 probability density function (PDF) of X.
 Z. Hu et al

3.2 Uncertainty in material fatigue properties can be transformed into independent ones using the Nataf
transformation (Goda 2010; Noh et al. 2009) or other meth-
Uncertainty in material fatigue properties also results in ods, such as the method proposed by Noh, Choi, and Du
uncertainty in fatigue life. The variations in material fatigue (Gupta et al. 2000; Noh et al. 2007). The developed method
properties have been extensively studied (Ayala-Uraga and is also applicable for the other two groups of S-N curves.
Moan 2007; Bengtsson and Rychlik 2009; Gu and Moan What is in common between the three groups is that
2002; Hasan et al. 2012; Jha et al. 2013; Lee et al. 2013; the number of cycles to failure under a stress level is a
Wei et al. 2011a, b, 2013; Xu et al. 2012). For instance, stress-dependent random variable. As a result, the mean and
the uncertainty of the fatigue crack growth model has been standard deviation of the number of cycles depend on stress
investigated (Hasan et al. 2012; Jha et al. 2013; Lee et al. levels (Pascual and Meeker 1999). For a specific stress level
2013; Wei et al. 2013), several probabilistic fatigue damage s, the number of cycle, N|s follows a Lognormal distribution
accumulation models have been developed (Bengtsson and or a Weibull distribution (Liu and Mahadevan 2007). For the
Rychlik 2009; Wei et al. 2011a, b; Xu et al. 2012), and mod- Lognormal distribution,
els for probabilistic S-N curves have also been developed
(Ayala-Uraga and Moan 2007; Gu and Moan 2002).
log( N | s) − μlog N
As the stress-life model is used in this work, we ∼ N (0, 12) (11)
σlog N
mainly consider variations in the S-N curve. In the past
decades, many models were developed for describing
the statistical nature of the S-N curve. The associated where μlogN and σlogN are respectively the mean and stan-
methods are classified into three groups - the statisti- dard deviation of log(N|s) and are given by
cal S-N curve (Kam et al. 1998; Kamiński 2002; Le
and Peterson 1999; Liao et al. 1995), the quantile S-N
μlog N = h1 (s) (12)
curve (Ni and Zhang 2000; Pascual and Meeker 1999;
Rowatt and Spanos 1998), and the stochastic S-N curve
(Liu and Mahadevan 2007). A detailed review about all and
the three groups can be found in Liu and Mahadevan
(2007).
σlog N = h2 (s) (13)
What distinguishes the three groups is the way of han-
dling correlations between stress levels. The statistical S-N
curve assumes that the distributions of the cycle number in which h1 (s) and h2 (s) are functions of mean and stan-
at stress levels are independent while the quantile S-N dard deviation. These two functions are obtained based on
curve assumes that they are dependent. The stochastic S- the experimental testing data under the constant amplitude
N curve developed by Liu and Mahadevan (2007) releases fatigue life testing. N(·,·) stands for a normal distribution
the assumptions by modeling the dependence between stress with the first parameter being the mean and the second
levels using the Karhunen-Loeve (KL) expansion method parameter being the variance.
(Phoon et al. 2002, 2005; Loeve 1977). We use the statistical In the subsequent sections, the effect of the uncertainties
S-N curve in this paper since the dependence between stress on the fatigue life is analyzed. Based on the analysis, the
levels is not our focus and the dependent random variables new fatigue reliability analysis method is developed.

Fig. 2 Flowchart of MPP

search
Fatigue reliability analysis for structures

cut the computational cost. To use the new method, we first

transform the probability in (14) into

pf = Pr{LF < l}
= Pr {1/( N1 | S1 ) + 1/( N2 | S2 ) + · · · + 1/( Nm | Sm ) > 1/ l}
(16)

The distribution of N i is dependent on Si , which is governed

Fig. 3 A cantilever beam subjected to cyclic load by g(X, d). The fatigue probability of failure pf depends on
X as shown below.
4 The proposed fatigue reliability analysis approach pf = Pr{LF (X, N| X) < l} (17)
where N| X = [ N1 | X, N2 | X, · · · , Nm | X] are random
4.1 Fatigue life reliability
numbers of cycles dependent on X.
In the following sections, we at first discuss the direct
Due to the uncertainties in the stress response and material
use of FORM and SORM for the fatigue reliability analy-
fatigue properties, the fatigue life given in (8) is random.
sis. As will be seen, this treatment may not be accurate and
The CDF of the fatigue life LF or the probability that LF is
efficient. We then present the new method, which improves
less than a specific value l is given by
both accuracy and efficiency. The comparison of the direct
FORM/SORM and improved FORM/SORM are shown in
1
pf = Pr{LF < l} = Pr <l the example section.
1/( N1 | S1 ) + 1/( N2 | S2 ) + · · · + 1/( Nm | Sm )
(14)
4.2 Direct FORM and SORM

where Ni | Si , i = 1, 2, · · · , m, are random numbers of One way of approximating the fatigue reliability is using
cycles dependent on random stresses Si given by FORM or SORM directly with the Rosenblatt transforma-
tion (Choi et al. 2007). Before applying FORM or SORM,
S = [S1 , S2 , · · · , Sm ] = gC (X, d) (15) the most probable point (MPP), at which the joint proba-
bility density of random variables is the highest, needs to
in which gC (·) is the stress response function after mean be identified. To determine the MPP, the dependent random
value correction on So = g(X, d). Equations (14) and (15) variables X and N|X are transformed into independent stan-
show that the fatigue life is a random variable and is a non- dard normal variables using the Rosenblatt transformation
linear function of random variables N i whose distributions as follows (Choi et al. 2007):
are dependent on Si . Liu and Mahadevan (2007) developed
UX = −1 (FX (X))
two methods for estimating the probability given in (14) (18)
UN = −1 F N|X ( N| X)
when the distribution of Si is known. The two methods
include the moment-based method and FORM. Even though where −1 (·) is the inverse CDF of a standard normal vari-
they can efficiently approximate the fatigue reliability given able, F x (·) is the CDF of random variable X i , F N|X (·) is
the stress distribution, there are still some limitations. The the CDF of random variable N i |X conditioned on X, and
major limitation is to know the stress distribution, but it is UX and UN are independent standard normal variables cor-
usually unknown in the design stage. To obtain the distribu- responding to random variables X and N = [N 1 , · · · , N m ],
tion of the stress, we need to call (15) many times. If (15) respectively.
involves CAE simulations, the computational cost will be After the transformation, (17) becomes
high. As will be seen, the method proposed in this work can pf = Pr{LF (UX , UN ) < l} = Pr {−1/LF (UX , UN ) < −1/ l} (19)

Fig. 4 Load trend over time

 Z. Hu et al

Table 1 Random variables but is more computationally expensive than FORM as sec-
ond derivatives are required. The Breitung’s formulation for
Variable Mean value Standard deviation Distribution type
SORM is given by Breitung (1984)
l (in) 9 0.01 Normal 
m+n−1
1
b (in) 0.2 0.005 Normal pf = (−β) (1 + βνi ) 2 (26)
h (in) 0.4 0.005 Normal i=1
Su (ksi) 221.7 5 Lognormal where νi (i = 1, 2, · · · , m + n − 1) are the principal curva-
F1 (lb) 80 3 Lognormal tures of −1/LF (UX , UN ) at the MPP. Details of SORM can
F2 (lb) 60 2 Lognormal be found in Choi et al. (2007).
F3 (lb) 70 2 Lognormal For n random variables in X and m stress responses in S,
F4 (lb) 65 2 Lognormal there are totally n + m variables in (20). Herein, the m stress
responses in S are m different peak stresses in the dynamic
stress responses. When m is large, the number of calling the
stress response function in (15) will be high, and the effi-
The MPP u∗ is then obtained by solving the following ciency will be low. In this work, we regard the situation that
optimization model given a group of x and getting the corresponding m stresses
⎧
⎪ min u as one function evaluation. The efficiency of direct use of
⎨ u=[uX ,uN ]
FORM and SORM for reliability analysis can be improved.
subject to (20)
⎪
⎩ As will be seen in the example, the accuracy of the direct
−1/LF (uX , uN ) ≤ −1/ l use of FORM may not be good either, and its accuracy also
in which || · || is the norm of a vector, and 1/LF (uX , uN ) is needs to be improved.
given by
4.3 Proposed method
1/LF (uX , uN ) = 1/(N1 ) + 1/(N2 ) + · · · + 1/(Nm ) (21)
where To overcome the drawbacks of the direct use of FORM
or SORM, we propose a new method that integrates the
−1
Ni = F N i |si
((uN )), i = 1, 2, · · · , m (22) fast integration method (Wen and Chen 1987) and SPA
and (Huang and Du 2006). The fatigue reliability introduced in
Section 4.1 is computed with two steps: calculating the con-
  ditional fatigue reliability and calculating the unconditional
s = [s1 , s2 , · · · , sm ] = gC FX−1 1
(uX1 ), FX−1
2
(uX2 ) , fatigue reliability.

. . . , FX−1
m
((uXm )), d 4.3.1 Conditional fatigue reliability analysis
(23)
−1 The conditional fatigue reliability is based on the condi-
in which F N i |si
(·) is the inverse CDF of N i |si conditional
tion that random variables X are fixed at specific values x,
on si , and FX−1
i
(·) is the inverse CDF of X i . which lead to specific (deterministic) stress responses s. The
Once the MPP u∗ is available from (20), pf is approxi- conditional probability of failure is then given by
mated using FORM as follows:
pf (x) = Pr{ LF < l| X = x} (27)
pf = (−β) (24)
or
  
where  1 
m
1 1
  pf (x) = Pr LN = = ≥ X = x (28)
β = u∗  (25) LF Ni | si l
i=1
When the accuracy of FORM is not good, SORM can be With the known values of s, computing the above proba-
employed. SORM is in general more accurate than FORM bility is just a traditional reliability analysis problem, and

Table 2 Results of fatigue

reliability analysis of a Method FORM Improved FORM SORM Improved SORM MCS
cantilever beam
pf 0.0056 0.0096 0.0085 0.0096 0.0095 [0.0094, 0.0097]
Error (%) 41.32 1.06 10.67 1.06 –
NOF 261 80 352 135 3 × 106
 Fatigue reliability analysis for structures

[2.44 × 10−5 , 3.69 × 10−5 ]

[1.02 × 10−4 , 1.26 × 10−4 ]
[7.38 × 10−4 , 8.01 × 10−4 ]
[1.58 × 10−3 , 1.68 × 10−3 ]
[5.58 × 10−3 , 5.75 × 10−3 ]
therefore existing methods, such as FORM, SORM, and
SPA, can be used. In this work, we use SPA (Huang and Du
2006) because of the following reasons: (1) The limit-state
m

[0.0150, 0.0153]
[0.0328, 0.0332]
[0.0613, 0.0618]
[0.1014, 0.1021]
[0.1530, 0.1538]
[0.2144, 0.2153]
[0.2824, 0.2835]
[0.3545, 0.3555]
function 1/( Ni | si )(i = 1, 2, . . . , m) in (28) is nonlinear
i=1
with respect to random variables Ni | si (i = 1, 2, . . . , m).
The first order and second order approximations of the
limit-state function may result in errors if FORM and
SORM are used. (2) SPA treats the limit-state func-

m
tion 1/( Ni | si )(i = 1, 2, . . . , m) as a function of ran-
i=1
dom variables1/( Ni | si )(i = 1, 2, . . . , m), and the limit-
state function becomes the sum of independent random
3.07 × 10−5
1.14 × 10−4
7.69 × 10−4
1.63 × 10−3
5.66 × 10−3

variables and is therefore linear. There will be no error from

0.0151
0.0330
0.0615
0.1018
0.1534
0.2148
0.2830
0.3550
the function approximation.
MCS

To use SPA, we first derive the Cumulant Generating

m
Function (CGF) of LN = 1/( Ni | si ), which is given by
i=1
 ∞ 
KLN (t) = ln et ln fLN (ln )dln (29)
2.95 × 10−5
1.11 × 10−4
7.75 × 10−4
1.68 × 10−3
5.83 × 10−3

−∞
Improved

where fLN (ln ) is the probability density function (PDF) of

SORM

0.0150
0.0339
0.0616
0.1005
0.1505
0.2092
0.2725
0.3385

the random response LN .

When N i |si , i = 1, 2, · · · , m are independent, we have

fLN (ln ) = f N1 |s1 (n1 )f N2 |s2 (n2 ) · · · f Nm |sm (nm ) (30)

in which f Ni |si (ni ) is the PDF of N i |si .

3.13 × 10−5
1.03 × 10−4
6.73 × 10−4
1.44 × 10−3
5.06 × 10−3

Substituting (30) into (29) yields

⎡
 ∞ t
SORM

m
0.0136
0.0298
0.0557
0.0923
0.1387
0.1928
0.2516
0.3120

1/( ni |si )
KLN (t) = ln ⎣ e i=1 f N1 |s1 (n1 )f N2 |s2 (n2 )
−∞
⎤

· · · f Nm |sm (nm )dn1 dn2 · · · dnm ⎦ (31)

2.78 × 10−5
1.04 × 10−4
7.75 × 10−4
1.68 × 10−3
5.83 × 10−3
Improved
FORM

0.0150
0.0315
0.0570
0.0925
0.1386
0.1948
0.2598
0.3306
Table 3 Probabilities of failure at different failure levels

2.13 × 10−5
6.89 × 10−5
4.44 × 10−4
9.43 × 10−4
3.30 × 10−3
8.93 × 10−3
FORM

0.0199
0.0385
0.0662
0.1037
0.1507
0.2058
0.2670
pf
Limit State

0.8 × 104
0.9 × 104
1.1 × 104
1.2 × 104
1.4 × 104
1.6 × 104
1.8 × 104
2.0 × 104
2.2 × 104
2.4 × 104
2.6 × 104
2.8 × 104
3.0 × 104

Fig. 5 Probability of failure under different failure levels

 Z. Hu et al

Table 4 Percentage of error under different failure levels Table 5 Number of function calls needed under different failure levels

Limit state Error (%) Limit state NOF

FORM Improved SORM Improved FORM Improved SORM Improved MCS

FORM SORM FORM SORM

0.8 × 104 30.42 9.32 2.23 3.70 0.8 × 104 313 100 404 155 3 × 106
0.9 × 104 39.72 8.80 10.29 2.87 0.9 × 104 287 80 378 135 3 × 106
1.1 × 104 42.30 0.72 12.48 0.72 1.1 × 104 287 80 378 135 3 × 106
1.2 × 104 42.14 3.29 11.69 3.29 1.2 × 104 261 80 352 135 3 × 106
1.4 × 104 41.74 2.97 10.69 2.97 1.4 × 104 261 80 352 135 3 × 106
1.6 × 104 41.01 0.77 10.28 0.76 1.6 × 104 235 80 326 135 3 × 106
1.8 × 104 39.62 4.74 9.95 2.57 1.8 × 104 235 80 326 135 3 × 106
2.0 × 104 37.43 7.44 9.40 0.09 2.0 × 104 209 80 300 135 3 × 106
2.2 × 104 34.98 9.11 9.28 1.26 2.2 × 104 183 100 274 155 3 × 106
2.4 × 104 32.42 9.68 9.58 1.93 2.4 × 104 183 100 274 155 3 × 106
2.6 × 104 29.87 9.31 10.26 2.74 2.6 × 104 157 100 248 155 3 × 106
2.8 × 104 27.28 8.19 11.07 3.81 2.8 × 104 157 80 248 135 3 × 106
3.0 × 104 24.78 6.89 12.11 4.72 3.0 × 104 131 60 222 115 3 × 106

Equation (31) is rewritten as If the first four cumulants are used, the cumulants κi,j , j
= 1, 2, 3, 4, are given in terms of moments as follows:
KLN (t) = K N1 |s1 (t) + K N2 |s2 (t) + · · · + K Nm |sm (t) (32) ⎧
⎪ κi,1 = mi,1
⎪
⎨ κ = m − m2
i,2 i,2 i,1
Directly evaluating (32) is very difficult. Herein, we use the
⎪
⎪ κ i,3 = 2m 3 − 3m m
i,1 i,2 + mi,3
power expansion of the CGF (Kendall and Stuart 1958). For ⎩ i,1
κi,4 = mi,4 − 4mi,1 mi,3 − 6m4i,1 + 12m2i,1 mi,2 − 3m2i,2
K Ni |si (t), the power expansion is given by
(34)
∞
 tj in which mi,j , j = 1, 2, 3, and 4, are the first four moments
K Ni |si (t) = κi,j (33)
j! about zero of Ni |si .
j =1
mi,j , j = 1, 2, 3, 4, are given by
where κi,j is the j-th cumulant of N i |si .  ∞  j
1
mi,j = f Ni |si (ni )dni , ∀j = 1,2,3,4 (35)
0 ni

Fig. 6 Percentage error under different failure levels Fig. 7 Function evaluations under different failure levels
Fatigue reliability analysis for structures

Fig. 8 A door cam

If higher order cumulants are used, the n-th order cumulant in which
is given by  !1/2
w = sign(η) 2 ηKLN (η) − KLN (η) (40)
n−1 
 
n−1
κi,n = mi,n − ki,j mi,n−j (36)  1/2
j −1
j =1 v = η KLN (η) (41)
Plugging (33) into (32), we have
 m   m 
∞  
4  ηj
tj KLN (η) = κi,j (42)
KLN (t) = κi,j (37) j!
j! j =1 i=1
j =1 i=1
⎧
Once the expressions of KLN (t) are available, the saddle- ⎨ 1, η > 0
point is obtained by solving the following equation: sign(η) = 0, η = 0 (43)
 m   m   m  ⎩
−1, η < 0
1   η  η2
= κi,1 + κi,2 + κi,3 where φ(·) is the PDF of a standard normal variable,
l 1! 2!
i=1 i=1 i=1 KLN (η) and KLN (η) are the first and second derivatives of
 m 
 η3 KLN (η), respectively.
+ κi,4 (38) The derivation of KLN (t) is based on the condition that
3!
i=1
N i |si , i = 1, 2, · · · , m, are independent. It is the assump-
With the saddlepoint η solved from (38), the conditional tion for the statistical S-N curve we use in this work.
probability of failure is then calculated by Huang and Du When N i |si , i = 1, 2, · · · , m, are dependent (i.e. stochas-
(2006) tic S-N curve), the dependent random variables should be
 transformed into independent random variables. Then, the
1
pf (x) = Pr LN ≥  X = x dimension reduction method (DRM) can be applied to esti-
l
  mate KLN (t) (Huang and Du 2006). Once the KLN (t) is
1 1
= 1 − (w) − φ(w) − (39) available, (37) through (43) are used to approximate the
w v conditional probability of failure.
Note that, the above analysis only calls the stress analysis
once.

Fig. 10 Working position of the shoulder and force analysis for the
Fig. 9 Door cam and door engagement
 Z. Hu et al

To approximate the probability given in (49), we define a

new limit-state function
 
gnew (Ue , X) = Ue − −1 pf (X) (50)

If the FORM or SORM is employed, the MPP search is

then given by

min β = u
u=[ue ,uX ] (51)
ue − −1 [pf (x)] ≤0

in which a general component x of x is x = Fx−1 [(uX )],

Fig. 11 Working position of the shoulder and force analysis for the where uX is a general component of uX .
disengagement After the MPP u∗ is found, pf is computed by FORM as
follows
4.3.2 Unconditional fatigue reliability analysis  
pf = 1 − (β) = 1 −  u∗  (52)
The conditional probability of failure obtained in the last
subsection is conditional on the stress or random variables If SORM is used to approximate (49), pf is obtained by
X. The unconditional probability of failure is given by plugging u∗ and the main curvatures of gnew (Ue , X) at the
 MPP into (26). We called the two methods the improved
pf = pf (x)fX (x)dx (44) FORM and improved SORM, respectively.
Directly calculating the integral above is costly, especially m + n random variables exist if FORM or SORM
when the dimension of X is high. To reduce the cost, fol- is directly used as indicated in (20). With the proposed
lowing the same principle in Wen and Chen (1987), we method, the number of random variables is reduced to n +
introduce a new random variable Ue ∼ N (0, 12) such that 1 as shown in (51). The dimension reduction means less
calls of the stress analysis, thereby less computational effort.
(upf ) = Pr{Ue ≤ upf } = pf (x) (45) As a result, the proposed method is more efficient than the
Then direct use of FORM or SORM. The accuracy of the pro-
posed method is also better than the direct use of FORM.
upf = −1 [pf (x)] (46)
The major reason is that the conditional probability obtained
Substituting (45) into (44) yields from SPA is accurate.
   Since we use MCS as a benchmark for methodology eval-
pf = Pr{Ue ≤ upf }fX (x)dx = φ(ue )due fX (x)dx uation, next, we briefly discuss how to use MCS for the
Ue ≤upf fatigue reliability analysis.
(47)
Equation (47) can be further written as 4.4 Monte Carlo Simulation for fatigue reliability analysis

pf = Pr{Ue ≤ upf (X)} = Pr{Ue − upf (X) ≤ 0} (48) For MCS, let the number of samples be nMCS . We first gen-
Combining (46) with (48), we have erate samples for the n independent variables X, we then
"   # generate samples for N i , i = 1, 2, · · · , m. The two steps are
pf = Pr Ue − −1 pf (X) ≤ 0 (49) used because N i , i = 1, 2, · · · , m depend on X. With the

Fig. 12 Stress trend of the

corner on the upper leg over
cycles
Fatigue reliability analysis for structures

samples of N i , i = 1, 2, · · · , m, we generate nMCS samples Since the corresponding valley of each peak of F = [F 1 ,
for LF . The probability of failure is then estimated by F 2 , F 3 , F 4 ] is zero, we have

pfMCS =
nf
(53) Somin = [0, 0, 0, 0] (55)
nMCS
Equation (54) implies that the stress response of the beam is
in which nf is the number of samples that satisfy LF < l. proportional to the load on the beam. With the known trend
of load over time, the trend of stress response is therefore
known. Due to the uncertainties in the geometrical param-
5 Numerical procedure eters, cyclic loading, and material fatigue properties, the
fatigue life of the beam is also uncertain.
Figure 2 shows the numerical procedure for identifying the Since the material is brittle, the Goodman mean value
MPP. The procedure is explained in details below. correction is applied (Wang and Sun 2005). The corrected
stress amplitude Si , i = 1, 2, 3, 4, are given by
Step 1: Initialization: Set initial point u = [uX , ue ] for the
MPP search. Sio Su
Si = (56)
Step 2: Stress analysis: For a given point uX perform 2Su − Sio
stress analysis using (15). where Su is the ultimate tensile strength of the material.
Step 3: Use the fatigue life model: Obtain the statistical According to (8), the fatigue life of the beam presented
parameters of the number of stress cycles, N i |si , i in cycles is given by
= 1, 2, · · · , m, with (12) and (13).
1
Step 4: Conditional reliability analysis: Perform the con- LF = (57)
ditional reliability analysis based on the informa- 4
1/(Ni |Si )
tion obtained in Step 3. i=1
Step 5: Limit-state function evaluation: Transform the in which N i |Si , i = 1, 2, 3, 4, are numbers of cycles to failure
conditional probability of failure into the equiv- under the stress level Si .
alent standard normal variable and evaluate the As discussed in Section 3.2, N i |Si is a stress-dependent
limit-state function in (50). random variable and follows a Log-normal distribution,
Step 6: Convergence check: If the reliability indexes β in defined by
two subsequent iterations are close enough, the  
MPP is identified and convergence is reached; log( Ni | Si ) ∼ N μlog(N) , σlog(N)
2
(58)
then compute the probability of failure using μlog(N) and σlog(N) are
FORM or SORM. Otherwise, generate a new " #
point for uX and ue , and go to Step 2. μlog(N) = log 10[c−d log10 (Si )] (59)
and

6 Numerical examples σlog(N) = 0.04μlog(N) (60)

where c = 12.2, and d = 3.68. The required fatigue life is l
Two numerical examples are presented to evaluate the pro- = 1.5 × 104 cycles.
posed method. Table 1 gives the distributions of the random variables.
There are eight random variables (i.e. l, b, h, Su , F 1 , F 2 ,
6.1 A cantilever beam F 3 , and F 4 ) in the stress response function, and four ran-
dom responses, Si , i = 1, 2, 3, 4, in the fatigue life function.
As shown in Fig. 3, a cantilever beam is subjected to a ran- The problem was solved by the direct FORM and SORM,
dom cyclic load F, which is plotted in Fig. 4. There are the improved FORM and SORM, and MCS. For MCS, the
four blocks of load in each cycle of F. The peak values of numbers of samples was 3 × 106 . The percentage error with
the four blocks are F 1 , F 2 , F 3 , and F 4 , respectively. The respect to MCS is defined by
corresponding valley value of each peak is zero.  
 
The maximum stresses Somax of the beam are given by pf − pfMCS 
ε= × 100 % (61)
  6Fl pfMCS
Somax = S1o , S2o , S3o , S4o = 2 (54)
bh where pfMCS is obtained from MCS while pf is obtained
where b, l, and h are the geometrical parameters as shown from other methods.
in Fig. 3 and F = [F 1 , F 2 , F 3 , F 4 ] is the vector of forces in Table 2 shows the results, including the MCS solution
one cycle. and the associated 95 % confidence interval in brackets, and
 Z. Hu et al

Fig. 13 One snapshot of stress distribution under engagement motion

the number of function calls (NOF) of the stress response and SORM, and MCS. The number of simulations of MCS
function, which is used as the measure of efficiency. is 3 × 106 . The failure thresholds vary from 0.9 × 104 to 3.0
The results show that the proposed method is more × 104 . The results are given in Table 3 and plotted in Fig. 5.
accurate and efficient than the direct FORM and SORM. Table 4 presents the percentage errors of the four methods
To study the robustness of the proposed method, we with respect to MCS. The percentage errors are also plotted
also performed reliability analyses at different failure levels in Fig. 6. The numbers of function calls are listed in Table 5
using the direct FORM and SORM, the improved FORM and plotted in Fig. 7.

Fig. 14 One snapshot of stress distribution under disengagement motion

Fatigue reliability analysis for structures

Table 6 Random variables of example 2 The robustness study indicates that the improved FORM
Variable Mean value Standard deviation Distribution type
and SORM significantly increase the accuracy and effi-
ciency of the direct FORM and SORM, respectively.
dgap (in) 0.107 0.009 Normal The cam is made of brittle material, and the Goodman
Su (ksi) 221.7 5 Lognormal correction was made as well. The corrected stress responses,
Si , i = 1, 2 are given by
6.2 A door cam

A door cam, as shown in Figs. 8 and 9, is used to hold the Sio (dopen )Su
Si = (65)
door open while stocking. The fatigue reliability of the cam 2Su − Sio (dopen )
is to be evaluated during the product development process.
For each cycle of the door opening and closing, the cam
experiences two kinds of motion, which are the engage- Due to manufacturing imprecision, the initial gap dgap and
ment and disengagement of the shoulder. During the motion the diameter of the shoulder dsh are random. But we treat
cycle, the upper and lower legs of the cam deflect until the dsh as deterministic because its randomness is negligible
shoulder passes the gap between the two legs. Figures 10 compared with that of dgap . Also considering variations in
and 11 show the working positions and force analysis for the the ultimate tensile strength of the material, we have two
engagement and disengagement of the cam, respectively. random variables dgap and Su in the stress response func-
Figure 12 shows the simplified stress history of the cor- tion. According to the stress response analysis given in
ner of the upper leg during cycles of engagement and Fig. 12, there are also two random variables in the fatigue
disengagement. Since the motion trend of the cam is known, life analysis model.
the stress response of the cam is also known. For every cycle The number of cycles to failure follows a Lognormal
of motion, we have Somax = S1o , S2o and Somin = [0, 0]. distributions with mean value of
Figure 12 indicates that the stress history of the cam is
characterized by the maximal stresses of engagement and
disengagement i.e. S1o and S2o . The force and stress anal- " #
μlog(N) = log 10[12.2−3.68 log10 (Si )] (66)
yses found that the stress response is dependent upon the
open distance dopen between the upper and lower legs. The
stress responses therefore can be expressed as functions of
dopen . and standard deviation of
dopen is a parameter related to the initial gap between two
legs and the diameter of the shoulder and is given by
dopen = dsh − dgap (62) σlog(N) = 0.03μlog(N) (67)

in which dsh is the diameter of the shoulder, and dgap is the

initial gap between the two legs.
In this example, dsh = 0.187 in, and the target fatigue life is
To explore the relationship between the stress responses
l = 2 × 104 cycles. Table 6 provides all the random variables
and dopen , we performed finite element analyses (FEA)
needed for the analysis.
based on the force analyses given in Figs. 10 and 11, which
The probability of fatigue failure of the cam was com-
result in the following stress responses:
puted by the direct FORM, SORM, the improved FORM,
S1o (dopen ) = 1.437 × 103 (dsh − dgap ) − 0.1021 (63) the improved SORM, and MCS. The numbers of samples of
3
= 1.2 × 10 (dsh − dgap ) − 0.5
S2o (dopen ) (64) MCS was 1 × 106 . Results are given in Table 7.
Two snapshots of the stress distribution under engagement The results also confirm that the proposed method is
and disengagement of the cam obtained from FEA are given more accurate and efficient than the direct use of FORM and
in Figs. 13 and 14. SORM.

Table 7 Results of reliability

analysis Method FORM Improved FORM SORM Improved SORM MCS

pf (×10−4 ) 6.53 7.70 7.55 7.82 8.16 [7.84, 8.48]

Error (%) 19.96 5.61 7.53 4.15 –
NOF 142 32 157 42 1 × 106
 Z. Hu et al

7 Conclusion Cruzado A, Leen SB, Urchegui MA, Gómez X (2013) Finite element
simulation of fretting wear and fatigue in thin steel wires. Int J
Fatigue 55:7–21
It is important to account for the stress-dependent charac- El Aghoury I, Galal K (2013) A fatigue stress-life damage accumula-
teristics of material fatigue properties for fatigue reliability tion model for variable amplitude fatigue loading based on virtual
analysis. Directly using the First Order Reliability Method target life. Eng Struct 52:621–628
(FORM) or Second Order Reliability Method (SORM) Fitzwater LM, Winterstein SR (2001) Predicting design wind turbine
loads from limited data: comparing random process and ran-
for the analysis may not be efficient and may produce dom peak models. J Sol Energy Eng Trans ASME 123(4):364–
large errors in the predicted fatigue reliability as shown 371
in the examples. The accuracy, as well as the efficiency, Goda K (2010) Statistical modeling of joint probability distribution
can be improved with the proposed method that integrates using copula: application to peak and permanent displacement
seismic demands. Struct Safe 32(2):112–123
the saddlepoint approximation and the conditional fatigue Gu XK, Moan T (2002) Long-term fatigue damage of ship struc-
reliability analysis. tures under nonlinear wave loads. Marine Technol 39(2):95–
The new method can predict the fatigue reliability or 104
Guo T, Chen YW (2013) Fatigue reliability analysis of steel bridge
the probability distribution of the fatigue life for structures
details based on field-monitored data and linear elastic fracture
under cyclic loadings with known trend. This assumption mechanics. Struct Infrastruct Eng 9(5):496–505
holds for many applications. The method accommodates Gupta AK, Móri TF, Székely GJ (2000) How to transform corre-
not only random variables with different distributions in lated random variables into uncorrelated ones. Appl Math Lett
13(6):31–33
the input variables to stress response functions, as well as Hasan SM, Khan F, Kenny S (2012) Probabilistic transgranular stress
uncertain parameters in the S-N curve. corrosion cracking analysis for oil and gas pipelines. J Press Vess
Technol Trans ASME 134(5):051701. (9 p)
Hu Z, Du X (2012) Reliability analysis for hydrokinetic turbine blades.
Renew Energy 48:251–262
Acknowledgments This material is based upon work supported by Huang B, Du X (2006) Uncertainty analysis by dimension reduction
the National Science Foundation through grant CMMI 1234855. The integration and saddlepoint approximations. J Mech Design Trans
support from the Intelligent Systems Center (ISC) at the Missouri ASME 128(1):26–33
University of Science and Technology is also acknowledged. Huang W, Moan T (2007) A practical formulation for evaluating
combined fatigue damage from high- and low-frequency loads. J
Offshore Mech Arctic Eng 129(1):1–8
Huang W, Wang TJ, Garbatov Y, Guedes Soares C (2013a) Dfr based
References fatigue reliability assessment of riveted lap joint accounting for
correlations. Int J Fatigue 47:106–114
Asi O, Yeşil T (2013) Failure analysis of an aircraft nose landing gear Huang CY, Hu CK, Yu CJ, Sung CK (2013b) Experimental investiga-
piston rod end. Eng Fail Anal 32:283–291 tion on the performance of a compressed-air driven piston engine.
Ayala-Uraga E, Moan T (2007) Fatigue reliability-based assessment Energies 6(3):1731–1745
of welded joints applying consistent fracture mechanics formula- Jha SK, John R, Larsen JM (2013) Incorporating small fatigue crack
tions. Int J Fatigue 29(3):444–456 growth in probabilistic life prediction: effect of stress ratio in Ti-
Aygül M, Bokesjö M, Heshmati M, Al-Emrani M (2013) A compara- 6al-2sn-4zr-6mo. Int J Fatigue 83–95
tive study of different fatigue failure assessments of welded bridge Kam TY, Chu KH, Tsai SY (1998) Fatigue reliability evaluation for
details. Int J Fatigue 49:62–72 composite laminates via a direct numerical integration technique.
Baumert EK, Pierron ON (2012) Fatigue properties of atomic- Int J Solids Struct 35(13):1411–1423
layer-deposited alumina ultra-barriers and their implications for Kamiński M (2002) On probabilistic fatigue models for composite
the reliability of flexible organic electronics. Appl Phys Lett materials. Int J Fatigue 24(2–4):477–495
101(25):251901 Kendall MG, Stuart A (1958) The advanced theory of statistics:
Beck AT, Gomes WJDS (2013) Stochastic fracture mechanics using distribution theory, vol 1. Charles Griffin & Company, London
polynomial chaos. Probabilist Eng Mech 34:26–39 Kihl DP, Sarkani S (1999) Mean stress effects in fatigue of welded
Bengtsson A, Rychlik I (2009) Uncertainty in fatigue life predic- steel joints. Probabilist Eng Mech 14(1–2):97–104
tion of structures subject to gaussian loads. Probabilist Eng Mech Ko NH (2008) Verification of correction factors for non-gaussian
24(2):224–235 effect on fatigue damage on the side face of tall buildings. Int J
Breitung K (1984) Asymptotic approximations for multinormal inte- Fatigue 30(5):779–792
grals. J Eng Mech 110(3):357–366 Kwon DK, Kareem A (2011) Peak factors for non-gaussian load
Chan KS, Enright MP, Moody JP, Hocking B, Fitch SHK (2012) effects revisited. J Struct Eng 137(12):1611–1619
Life prediction for turbopropulsion systems under dwell fatigue Larsen JM, Jha SK, Szczepanski CJ, Caton MJ, John R, Rosen-
conditions. J Eng Gas Turb Power 134(12):122501. (8 pp) berger AH, Buchanan DJ, Golden PJ, Jira JR (2013) Reducing
Choi SK, Grandhi RV, Canfield RA (2007) Reliability-based structural uncertainty in fatigue life limits of turbine engine alloys. Int J
design. Springer, New York Fatigue 57(1):103–112
Cihan K, Yuksel Y (2013) Deformation of breakwater armoured Le X, Peterson ML (1999) Method for fatigue based reliability when
artificial units under cyclic loading. Appl Ocean Res 42:79–86 the loading of a component is unknown. Int J Fatigue 21(6):603–
Correia JAFO, De Jesus AMP, Fernández-Canteli A (2013) Local 610
unified probabilistic model for fatigue crack initiation and prop- Lee YJ, Song J (2012) Finite-element-based system reliability analy-
agation: application to a notched geometry. Eng Struct 52:394– sis of fatigue-induced sequential failures. Reliab Eng Syst Safety
407 108:131–141
Fatigue reliability analysis for structures

Lee D, Kim S, Sung K, Park J, Lee T, Huh S (2013) A study on Phoon KK, Huang HW, Quek ST (2005) Simulation of strongly non-
the fatigue life prediction of tire belt-layers using probabilistic gaussian processes using Karhunen-Loeve expansion. Probabilist
method. J Mech Sci Technol 27(3):673–678 Eng Mech 20(2):188–198
Li FZ, Low YM (2012) Fatigue reliability analysis of a steel catenary Rajaguru P, Lu H, Bailey C (2012) Application of kriging and radial
riser at the touchdown point incorporating soil model uncertain- basis function in power electronic module wire bond structure
ties. Appl Ocean Res 38:100–110 reliability under various amplitude loading. Int J Fatigue 45:61–
Li J, Wang X (2012) An exponential model for fast simulation of multi- 70
variate non-gaussian processes with application to structural wind Rathod V, Yadav OP, Rathore A, Jain R (2012) Reliability-based
engineering. Probabilist Eng Mech 30:37–47 design optimization considering probabilistic degradation behav-
Liao M, Xu X, Yang Q-X (1995) Cumulative fatigue damage dynamic ior. Qual Reliab Eng Intern 28(8):911–923
interference statistical model. Int J Fatigue 17(8):559–566 Rowatt JD, Spanos PD (1998) Markov chain models for life prediction
Liu Y, Mahadevan S (2007) Stochastic fatigue damage modeling under of composite laminates. Struct Safety 20(2):117–135
variable amplitude loading. Int J Fatigue 29(6):1149–1161 Siddiqui NA, Ahmad S (2001) Fatigue and fracture reliability of Tlp
Liu Y, Mahadevan S (2009) Efficient methods for time-dependent tethers under random loading. Marine Struct 14(3):331–352
fatigue reliability analysis. AIAA J 47(3):494–504 Sousa C, Rocha JF, Calçada R, Serra Neves A (2013) Fatigue analy-
Loeve M (1977) Probability theory, 4th edn. Springer, New York sis of box-girder webs subjected to in-plane shear and transverse
Low YM (2013) A new distribution for fitting four moments and its bending induced by railway traffic. Eng Struct 54:248–261
applications to reliability analysis. Struct Safety 42:12–25 Suyuthi A, Leira BJ, Riska K (2013) Fatigue damage of ship hulls due
Ni K, Zhang S (2000) Fatigue reliability analysis under two-stage to local ice-induced stresses. Appl Ocean Res 42:87–104
loading. Reliab Eng Syst Safety 68(2):153–158 Wang X, Sun JQ (2005) Effect of skewness on fatigue life with mean
Noh Y, Choi KK, Du L (2007) New transformation of dependent stress correction. J Sound Vib 282(3–5):1231–1237
input variables using copula for RBDO. In: 7th world congresses Wei Z, Yang F, Cheng H, Nikbin K (2011a) Probabilistic predic-
of structural and multidisciplinary optimization, COEX Seoul, 21 tion of crack growth based on creep/fatigue damage accumulation
May–25. Korea mechanism. eds. 1539 STP, pp. 230–252
Noh Y, Choi KK, Du L (2009) Reliability-based design optimiza- Wei Z, Yang F, Cheng H, Nikbin K (2011b) Probabilistic predic-
tion of problems with correlated input variables using a gaussian tion of crack growth based on creep/fatigue damage accumulation
copula. Struct Multidisciplin Optim 38(1):1–16 mechanism. J ASTM Int 8(5):JAI103690. (15 pp)
Norouzi M, Nikolaidis E (2012) Efficient method for reliability assess- Wei Z, Yang F, Lin B, Luo L, Konson D, Nikbin K (2013) Deter-
ment under high-cycle fatigue. Int J Reliab Qual Safety Eng ministic and probabilistic creep-fatigue-oxidation crack growth
19(5):1250022. (27 pp) modeling. Probabilist Eng Mech 33:126–134
Pascual FG, Meeker WQ (1999) Estimating fatigue curves with the Wen YK, Chen HC (1987) On fast integration for time variant struc-
random fatigue-limit model. Technometrics 41(4):277–290 tural reliability. Probabilist Eng Mech 2(3):156–162
Petrescu FIT, Petrescu RVV (2013) Dynamic synthesis of the rotary Xu YL, Chen ZW, Xia Y (2012) Fatigue assessment of multi-loading
cam and translated tappet with roll. Intern Rev Modell Simul suspension bridges using continuum damage model. Int J Fatigue
6(2):600–607 40:27–35
Phoon KK, Huang SP, Quek ST (2002) Simulation of second- Zhang DK, Geng H, Zhang ZF, Wang DG, Wang SQ, Ge SR (2013)
order processes using Karhunen-Loeve expansion. Comput Struct Investigation on the fretting fatigue behaviors of steel wires under
80(12):1049–1060 different strain ratios. Wear 303(1–2):334–342

View publication stats