Special Issue Article

Advances in Mechanical Engineering

2016, Vol. 8(11) 1–9
Ó The Author(s) 2016
Fatigue life analysis based on six sigma DOI: 10.1177/1687814016679314
aime.sagepub.com
robust optimization for pantograph
collector head support

Yonghua Li1, Mingguang Hu1 and Feng Wang2

Abstract
In this article, a new fatigue life analysis method based on six sigma robust optimization is proposed, which considers the
random effects of material properties, external loads, and dimensions on the fatigue life of a pantograph collector head
support. Some main random factors are identified through fatigue reliability sensitivity analysis, which are used as input
variables during fatigue life analysis. The six sigma optimization model is derived using the second-order response surface
method. The response surface is fitted by the Monte Carlo method, the samples are obtained by the Latin hypercube
sampling technique, and the proposed model is optimized using the interior point algorithm. Through the optimization,
the collector head support weight is reduced, the mean and the standard deviation of fatigue life have been decreased,
and the effect of design parameter variation on the fatigue life is reduced greatly. The robustness of fatigue life prediction
of collector head support is improved. The proposed method may be extended to fatigue life analysis of other compo-
nents of electric multiple units.

Keywords
Fatigue, P-S-N curve, six sigma robust optimization, collector head support, Monte Carlo

Date received: 27 June 2016; accepted: 14 October 2016

Academic Editor: Yongming Liu

Introduction
The pantograph is one of the most important electric
equipments in the electric multiple units (EMU), which
collects the electric power from catenary for the EMU.
The pantograph is mainly composed of collector head
support (CHS) and carbon slipper, which is subjected
to shock loads between carbon slipper and contact
wire.1 The working conditions of pantograph are
becoming worse and worse with the increasing speed of
the EMU.
The pantograph has been subjected to variable load-
ings due to the uneven track, the impact force of panto-
graph and catenary, and the air flow force, which may
cause fatigue damage of the pantograph, shorten its ser-
vice life, and affect the safety and reliability during
EMU operation.2,3 When the pantograph worked for a
period, some cracks were found in the pantograph CHS
and the vane where fatigue failure may occur. To ensure
the normal usage of pantograph and reduce accident
occurrences, it is necessary to investigate the fatigue life
and reliability of the pantograph CHS.
Fatigue reliability analysis, which combines the fati-
gue life analysis and reliability-based design, is an effec-
tive method to improve the reliability of engineering
components.4 In this method, the dispersion problem of
1
School of Traffic and Transportation Engineering, Dalian Jiaotong
University, Dalian, China
2
College of Bullet Train Application and Maintenance Engineering, Dalian
Jiaotong University, Dalian, China
Corresponding author:
Yonghua Li, School of Traffic and Transportation Engineering, Dalian
Jiaotong University, Dalian 116028, China.
Email: yonghuali@163.com

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 3.0 License
(http://www.creativecommons.org/licenses/by/3.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 Advances in Mechanical Engineering
affecting the component fatigue life is fully considered.
Comparing the conventional fatigue analysis, however,
a large number of random factors make the fatigue
reliability analysis much more complex. Moreover, fati-
gue reliability analysis is also related to probability dis-
tribution of fatigue life.5 Generally, the probability
distribution needs to be statistically analyzed based on
a large number of experimental data, and these experi-
ments are often time consuming and costly. Thus, the
development of the investigation on structural fatigue
reliability is relatively slow. In recent years, many
researchers have made some achievements on structural
fatigue reliability analysis.6–8 The key to solve the issue
of fatigue reliability analysis is to establish an effective
fatigue reliability model, which should be consistent with
fatigue failure mechanism and also reflect the dispersion
of various factors during fatigue failure.9–18 At present,
the established models for fatigue reliability analysis
mainly include residual strength model,9 cumulative dam-
age model,10 strain energy model,11,12 ductility exhaustion
model,13 damage-strengthening model,14 and prob-
abilistic life prediction models.15,16 However, the
robust optimization is rarely considered during fati-
gue reliability analysis of pantograph CHS. In recent
years, the six sigma robust optimization has been
widely applied to engineering practice.18–22
In this article, fatigue life analysis based on six sigma
robust optimization for pantograph CHS is proposed,
where the fatigue reliability analysis is conducted to
find the uncertain factors which affected the fatigue life
of structure, and six sigma robust optimization analysis
is conducted to decrease the effects of uncertain factors
on fatigue life.
From the engineering application prospect, the com-
bination of six sigma and robust optimization in rail-
way industry can improve the robustness of mechanical
components and reduce the impact of random factors
on the component performance. Meanwhile, this
method lightens the component weight and contributes
to the light weight design.
Through the fatigue reliability analysis of CHS, its
fatigue reliability can be predicted considering many
factors, such as surface quality, stress concentration,
external loads, and plate thickness, and the main factors
and secondary factors that cause its fatigue failure can
be obtained. The dispersion problems of fatigue life are
analyzed using six sigma robust optimization design.
Accordingly, the random factors affecting the sensitiv-
ity of fatigue life are reduced greatly.
This article is organized as follows: section ‘‘Six
sigma robust optimization design method’’ provides six
sigma robust optimization design method. The static
strength analysis of CHS based on finite element analy-
sis is given in section ‘‘Static strength analysis of CHS
based on finite element method.’’ Section ‘‘Fatigue
reliability analysis of CHS’’ conducts fatigue reliability
analysis of CHS. Section ‘‘Fatigue life analysis based
on six sigma robust optimization for CHS’’ carries out
fatigue life analysis based on six sigma robust optimiza-
tion for CHS. A brief discussion and conclusion closes
the article.
Six sigma robust optimization design
method
Robust optimization design18 is an effective method to
reduce the effects of various uncertain factors on the
target response value. Namely, the mean square value
of the target response can be reduced. It can also
achieve the goal of lower sensitivity value for target
response under the random plenty of uncertainty fac-
tors. Six sigma robust optimization19,20 is an advanced
design method with combination of six sigma quality
management theory and robust optimization. It mini-
mizes the objective response value to meet the reliabil-
ity design requirement.
Considering the complex nonlinear relationship
between target response values and design parameters,
the Monte Carlo method is used for numerical calcula-
tion in this article. The used sampling method is Latin
hypercube in the ANSYS probabilistic design system
(PDS) module. The six sigma robust optimization
model is built through the response surface method.
Finally, the six sigma robust optimization design of
CHS is accomplished using the optimization toolbox of
MATLAB 2010b.
Static strength analysis of CHS based on
finite element method
Structural analysis of CHS
In order to obtain an accurate simulation result, the
integrated model of the pantograph was established
including CHS and the contact strip. The CHS stress
was calculated by analyzing the strength of the inte-
grated model. The geometry of the CHS is shown in
Figure 1.
Finite element model of CHS
To improve the model calculation accuracy, a finite ele-
ment model of CHS is meshed by hexahedral element.
The loads and constraints of CHS are defined based on
the force and boundary conditions of contact strip and
CHS. Figure 2 shows the integrated finite element
model of contact strip and CHS. In Figure 2, the grid
size of contact strip and CHS finite element model is
5 mm, and the total grid numbers are 92,646.
According to the actual loading conditions of CHS,
the force in the pantograph can be simplified as follows.
First, the friction between pantograph and contact wire
Li et al.
Figure 1. Geometry of the CHS.
Figure 2. The integrated finite element model of contact strip
and CHS.
Figure 3. Von Mises stress results of the CHS and the carbon
slipper.
can be simplified as the longitudinal force FX. The load
caused by the car body vibration and the impact of
the pantograph-catenary is simplified as the vertical
3
Figure 4. Von Mises stress of the CHS.
force FZ. Then, the air pressure can be simplified as
aerodynamic load PRES. The values of the above-
mentioned simplified forces are 450 N, 350 N, and
6000 Pa, respectively. The material of model for the
CHS and carbon slipper is the aluminum alloy, and its
yield strength is 435 MPa.
Static strength check of CHS
The load condition of static strength calculation
includes longitudinal force FX, vertical force FZ, and
aerodynamic load PRES.22 This analysis is conducted
in the ANSYS 14.0. When the calculation is completed,
the maximum Von Mises stress results are obtained.
Figure 3 shows the maximum Von Mises stress results
of the CHS and the carbon slipper. The Von Mises
stress distribution of the CHS can be obtained as
shown in Figure 4.
From Figure 4, it can be seen that the maximum
value is 93 MPa, which is smaller than the material yield
strength 435 MPa. Therefore, the static strength of the
CHS meets the design requirements.
The static strength calculation for the CHS and the
carbon slipper indicates the position of the maximum
Von Mises stress located in the installing hole of the
CHS. The simulation result is consistent with that of
actual test conditions. Thus, the simulation results pro-
vide a certain reference value for the primary design
work. It can also shorten the product development cycle
and reduce the cost to a certain extent.
Fatigue reliability analysis of CHS
Fatigue life assessment of CHS
Fatigue life of CHS under variable loading is evaluated
according to the P-S-N curve of material and the load
spectrum of CHS.23–26 However, the test data of
4 Advances in Mechanical Engineering
variable load spectrum are difficult to be obtained due
to the limitation of experimental conditions. In this
study, a simplified fatigue load spectrum according to
the actual situation is applied to calculate the CHS fati-
gue life together with the material P-S-N curve.
The parameter of P-S-N curve. For the CHS, the P-S-N
curve of the material is often used to calculate its fati-
gue life, and the formula can be expressed as follows
( lg N )P = AP + BP lg s1 ð1Þ
where N is the fatigue life, s1 is the equivalent stress,
and AP and BP are material constants.
Mean stress correction. During the operation of the
EMU, the pantograph will generate vibrations, which
makes the stress of the CHS fluctuant around a certain
average stress. Often such cyclic loadings with mean
stress considerably influence the component damage
accumulation process. The stress spectrum can be
expressed by average stress and stress amplitude. The
average stress is zero according to the material P-S-N
curve. Therefore, the simplified stress spectrum equa-
tion is modified based on Goodman diagram26 as
follows
sa
s1 = ð2Þ
1  sm =sb
where s1 is the equivalent stress, sa is the stress ampli-
tude, sm is the mean stress, and sb is the material ulti-
mate tensile strength.
The fatigue life of CHS can be calculated by equa-
tion (1), where the equivalent stress s1 is obtained by
equation (2) to consider mean stress corrections.
Fatigue reliability calculation of CHS
The calculation of fatigue reliability life is completed
using the ANSYS PDS module and HyperMesh 11.0
Table 1. Distribution characteristics and numerical values of each random variable parameter.
Random variables Sign
Elastic modulus (MPa) E Gaussian
Poisson ratio U Gaussian
Density DENS
Longitudinal load (N) FX
Vertical load (N) FZ
Aerodynamic load (MPa) PRES
Thickness of carbon slipper mount (mm) T1
Thickness of U shape mount (mm) T2
Thickness of spring mount (mm) T3
software. A parametric model of CHS is built using the
APDL language.27
The parameters of variable loads, structural dimen-
sions, and materials are defined as random inputs, and
the distribution characteristics and values of each para-
meter are listed in Table 1. A limit state equation can be
expressed as the difference between the calculated life
and the designed life28
G = N  N0 ð3Þ
where G is the limit state function, N is the fatigue life
calculated by equation (1), and N0 is designed life.
G . 0 represents the structural safety. G = 0 represents
the limit state. G \ 0 represents the structural failure.
Reliability calculation of CHS. The Monte Carlo method is
used to calculate the structural fatigue reliability.21,28 In
this article, the fatigue probability analysis for CHS is
completed using ANSYS PDS module. The sample
points of 500, which are obtained using Latin hyper-
cube sampling technique, are introduced to calculate
the structural state function of G. Figure 5 represents
the variation trend of the sample mean values. The
vertical coordinate is the difference between the calcu-
lated life and the designed life, and the horizontal
coordinate is the number of sample points. The mid-
dle line represents the mean value, and the other two
lines are the upper and lower bounds of the structural
state function. From Figure 5, it can be seen that the
variations of mean value of structural state function
G tends to be stable, which indicates the reliability
agrees well with the design requirement and the relia-
bility is 99.60%.
Sensitivity analysis of fatigue reliability. Through the sensi-
tivity analysis, the main factors that affect CHS fatigue
reliability can be obtained. Figure 6 shows the sensitiv-
ity of reliability results, in which the size of areas repre-
sents the important degree of different influencing
Distribution Mean Coefficient of variation
7.1 3 104 0.05
0.33 0.05
Uniform 2.7 3 1029
Log normal 450 0.1
Log normal 350 0.1
Log normal 6.0 3 1023 0.1
Gaussian 2.5 0.05
Gaussian 3.5 0.05
Gaussian 4.0 0.05
Li et al.
Figure 5. The variation trend of the mean value of structural state function samples.
Figure 6. The sensitivity of fatigue reliability results.
factors. The positive input parameters mean that the
parameters are positively correlated with the output,
while the negative input parameters mean that the
parameters are negatively correlated with the output.
From Figure 6, note that the main factors are the longi-
tudinal force FX and the structure size T1 and T2 of
CHS. The other factors have little impact on the fatigue
reliability of the CHS.
5
Fatigue life analysis based on six sigma
robust optimization for CHS
Establishment of approximate response surface
model
The longitudinal force accuracy is difficult to control
since such force is determined by many factors, such as
the CHS structure, air flow impact, and vibrations.
6 Advances in Mechanical Engineering

Table 2. Upper and lower bounds of variables and the distribution types.

Design variables Lower (Ti L ) Upper (Ti U ) Type of distribution Standard deviation

T1 (mm) 1.5 3.5 Gaussian 0.05

T2 (mm) 2.5 4.5 Gaussian 0.05
T3 (mm) 3.0 5.0 Gaussian 0.05

Figure 7. The variation trend of the mean value of fatigue life samples.

This article focuses on the influence of the size of CHS
structure on its fatigue reliability.
However, the relationship between fatigue life and
structural dimensions is the implicit nonlinear, it is dif-
ficult to be formulated in the analytic expression. The
response surface method may describe accurately impli-
cit nonlinear relationship of the fatigue life and struc-
tural dimensions. The robust optimization model of six
sigma is built using the response surface method to
improve the analysis accuracy.
In this research, a response surface–based model is
built considering the parameters of the plate thickness
and fatigue life values. In engineering application, the
second-order response surface model is used widely,
and the basic formula is
X
n X
n X
n
y= cii x2i + cij xi xj + c i xi + c0
i=1 i.j i=1
where n is the number of design variables, c0 is the con-
stant, and ci , cii , and cij are the polynomial coefficients.
The Monte Carlo method is used to simulate and fit
an accurate response surface model of CHS. The
response value of the CHS fatigue life is obtained by
500 times Latin hypercube sampling technique. The
distribution types and the design variables are shown in
Table 2.
Figure 7 shows the variation trend of the mean value
of fatigue life samples. The vertical coordinate is the
fatigue life of the CHS, and the horizontal coordinate
is the number of sample points. From Figure 7, the
mean values of fatigue life samples have stabilized by
500 times simulation to the CHS. Thus, the predicted
fatigue life is reasonable.
Figure 8 shows the sensitivity analysis for the CHS
fatigue life, in which the area size represents the impor-
tant degree of influencing factor. From Figure 8, it can
ð4Þ be seen that the effect degrees of the analyzed design
variables on the fatigue life are ranked as T1 . T2 . T3.
Based on the sensitivity analysis, the design variables
on the fatigue life, T1, T2, and T3, are used to fit the
response surface.
Li et al.
Figure 8. The sensitivity of fatigue life.
The response surface of the mean value of CHS fati-
gue life is as follows
uN =  0:1902T12  0:05965T22
 0:1444T32 + 0:06946T1 T3
ð5Þ
+ 0:1139T2 T3 + 1:0441T1
+ 0:1652T2 + 0:6551T3 + 3:5981
The response surface for standard deviation value of
fatigue life for CHS is as follows
sN = 0:03T12 + 0:0127T22
+ 0:0106T32 + 0:0235T1 T3
ð6Þ
+ 0:0175T2 T3  0:2159T1
 0:1523T2  0:1989T3 + 0:9032
where uN is the mean value of fatigue life and sN is the
standard deviation value of fatigue life. T1, T2, and T3
are dimensions of the CHS.
Evaluation of response surface model error
The fitting precision of response surface model is evalu-
ated by the determination coefficient. The calculation
method of the determination coefficient,29R2, is as
follows
X P
(^yi  yi )2
R2 = ð7Þ
i = 1 (yi   yi ) 2
7
where yi is the response value for the sample point of i,
^yi is predictive value, yi is the response mean value for
the sample point of i, and P are the number of sample
points. When the determination coefficient R2 is close
to 1, the prediction accuracy of the response surface is
higher. The value of R2 is 98.6%, which indicates that
the accuracy of the fitted response surface is higher.
Six sigma robust optimization design of fatigue life
The fatigue life of CHS is calculated using its para-
metric model considering the first principal stress.
Then, the response surface models of the fatigue life
mean and standard deviation are obtained using the
Monte Carlo method. Finally, the optimal solution is
obtained using the interior point algorithm method.
The specific flow is shown in Figure 9.
Fatigue life analysis based on six sigma robust opti-
mization model is as follows
min F½uN (Ti ), sN (Ti )
= w1 ½uN (Ti )  N0 2 + w2 s2N (Ti )
ð8Þ
s:t: uN (Ti )  6sN (Ti )  N0
TiL + 6sTi  uTi  TiU  6sTi
where T is dimension parameter, i is the number of con-
straints, and N0 = 1 3 107:0 is the design life.21 uN (Ti ) is
a function for fatigue life mean value response surface,
sN (Ti ) is a function for the standard deviation values
response surface of fatigue life, Ti L and TiU are the lower
8 Advances in Mechanical Engineering

Table 3. Six sigma robust optimization results of fatigue life.

Variables Pre-optimization Post-optimization Change rate (%)

Standard deviation, sN 2.75 3 1022 2.31 3 1022 16.0

Mean value, uN 1.0 3 107.47 1.0 3 107.06 61.1
Thickness, T1 (mm) 2.5 1.9 24.0
Thickness, T2 (mm) 3.5 3.0 14.3
Thickness, T3 (mm) 4.0 3.9 2.5
Weight, m (kg) 8.59 8.04 6.4

From Table 3, it can be concluded that the fatigue

life value of post-optimization is more close to designed
life value N0. The CHS weight is reduced with the
decrease in the sizes of design variables. Considering
the engineering practice, the optimized thickness values
of T1, T2, and T3 are round to 2.0, 3.0, and 4.0 mm,
respectively. Meantime, the mean and standard devia-
tion of fatigue life have been decreased by 16%, which
indicates that the effects of parameter variation on fati-
gue life are reduced greatly. Therefore, the robustness
of CHS design is improved.

Conclusion
This article analyzed the fatigue life based on six sigma
robust optimization method for pantograph CHS. A new
six sigma optimization model is built by the second-order
response surface. The response surface is fitted by Monte
Carlo method and the samples are obtained by the Latin
hypercube sampling technique. The design parameters
are optimized and the robustness of fatigue life of CHS is
improved. Some conclusions are as follows:

1. Through the sensitivity analysis of the CHS fati-

2.
Figure 9. Process of fatigue life analysis based on six sigma
robust optimization design.
and upper bounds of plate thickness, uTi is mean values
of the dimension parameter, sTi is the standard devia-
tion values of the dimension parameter, and w1 and w2
are weight coefficients.
gue life, the effects of its main design variables
affecting on the fatigue life are identified, which
are thickness of carbon slipper mount T1, thick-
ness of U shape mount T2, and thickness of
spring mount T3.
The sensitive design parameters of T1, T2, and
T3 are optimized by the six sigma robust optimi-
zation method. Through the optimization, the
CHS weight is reduced with the decrease in the
sizes of design variables, the mean and standard
deviation of fatigue life have been decreased by
16%, and the effect of parameter variation on
fatigue life is greatly reduced. Six sigma robust
optimization for pantograph CHS is realized.
The proposed method may be extended for
application on the other components of EMU.

Result analysis Declaration of conflicting interests

The interior point algorithm is performed by MATLAB The author(s) declared no potential conflicts of interest with
2010b software. The results of pre-optimization and respect to the research, authorship, and/or publication of this
post-optimization are shown in Table 3. article.
Li et al.
Funding
The author(s) disclosed receipt of the following financial sup-
port for the research, authorship, and/or publication of this
article: The authors would like to acknowledge the partial
supports provided by the program of Educational
Commission of Liaoning Province under contract number
JDL2016001, the program of National Natural Science
Foundation of Liaoning Province under contract number
2014028020, and the program of the Dalian Science and
Technology Project under contract number 2015A11GX026.
References
1. Li MG, Yang YQ and Su C. Research on anti-fatigue
performance of upper-arm cross beam of pantograph
structure of electric multiple units. J Dalian Jiaotong
Univ 2013; 34: 23–26 (in Chinese).
2. Ibrahim MH, Kharmanda G and Charki A. Reliability-
based design optimization for fatigue damage analysis.
Int J Adv Manuf Tech 2015; 76: 1021–1025.
3. Zhang HF, Tong LS, Chen ZB, et al. Improvement of
pantograph framed structures for metro vehicles based
on fatigue strength analysis. Electr Locomot Mass Transit
Veh 2010; 33: 30–33 (in Chinese).
4. Zhu SP, Huang HZ, Li YF, et al. Probabilistic modeling
of damage accumulation for time-dependent fatigue
reliability analysis of railway axle steels. Proc IMechE,
Part F: J Rail Rapid Transit 2015; 229: 23–33.
5. Schmidt D, Manuel L, Nguyen HH, et al. Fatigue relia-
bility analysis for brace–column connection details in a
semisubmersible hull. J Offshore Mech Arct 2015; 137:
061301-1–061301-7.
6. Zhu SP, Huang HZ, Peng W, et al. Probabilistic physics
of failure-based framework for fatigue life prediction of
aircraft gas turbine discs under uncertainty. Reliab Eng
Syst Safe 2016; 146: 1–12.
7. Jablonski M, Lucchini R, Bossuyt F, et al. Impact of
geometry on stretchable meandered interconnect uniaxial
tensile extension fatigue reliability. Microelectron Reliab
2014; 55: 143–154.
8. Rafsanjani HM, Sørensen JD and Sciubba E. Reliability
analysis of fatigue failure of cast components for wind
turbines. Energies 2015; 8: 2908–2923.
9. Jun W and Zhi PQ. Fatigue reliability based on residual
strength model with hybrid uncertain parameters. Acta
Mech Sinica 2012; 28: 112–117.
10. Park TG, Choi CH, Won JH, et al. An efficient method
for fatigue reliability analysis accounting for scatter of
fatigue test data. Int J Precis Eng Man 2010; 11: 429–437.
11. Zhu SP and Huang HZ. A generalized frequency
separation-strain energy damage function model for low
cycle fatigue-creep life prediction. Fatigue Fract Eng M
2010; 33: 227–237.
12. Fan YN, Shi HJ and Tokuda K. A generalized hysteresis
energy method for fatigue and creep-fatigue life predic-
tion of 316L (N). Mater Sci Eng A: Struct 2015; 625:
205–212.
13. Zhu SP, Huang HZ, Liu Y, et al. An efficient life predic-
tion methodology for low cycle fatigue-creep based on
9
ductility exhaustion theory. Int J Damage Mech 2013; 22:
556–571.
14. Zhao LH, Zheng SL and Feng JZ. Fatigue life prediction
under service load considering strengthening effect of
loads below fatigue limit. Chin J Mech Eng 2014; 27:
1178–1184.
15. Zhu SP, Huang HZ, Ontiveros V, et al. Probabilistic low
cycle fatigue life prediction using an energy-based dam-
age parameter and accounting for model uncertainty. Int
J Damage Mech 2012; 21: 1128–1153.
16. Hu D, Ma Q, Shang L, et al. Creep-fatigue behavior of
turbine disc of superalloy GH720Li at 650°C and prob-
abilistic creep-fatigue modeling. Mater Sci Eng A: Struct
2016; 670: 17–25.
17. Karadeniz H. Uncertainty modeling in the fatigue relia-
bility calculation of offshore structures. Reliab Eng Syst
Safe 2001; 74: 323–335.
18. Ke L, Zhang YJ, Qin S, et al. A new robust design for
imperfection sensitive stiffened cylinders used in aero-
space engineering. Sci China Technol Sci 2015; 58:
796–802.
19. Sun GY, Li GY, Chen T, et al. Sheet metal forming based
six sigma robust optimization design. J Mech Eng 2008;
44: 248–254 (in Chinese).
20. Asafuddoula M, Singh HK and Ray T. Six-sigma robust
design optimization using a many-objective decomposi-
tion-based evolutionary algorithm. IEEE T Evolut Com-
put 2014; 19: 490–507.
21. Li YH, Hu MG and Wang J. Six sigma robust optimiza-
tion of fatigue life for the passenger car battery hanging
device. J Donghua Univ 2016; 33: 108–111.
22. Kim JW and Yu SN. Design variable optimization for
pantograph system of high-speed train using robust
design technique. Int J Precis Eng Man 2013; 14: 37–43.
23. Shimizu S, Tosha K and Tsuchiya K. New data analysis
of probabilistic stress-life (P–S–N) curve and its applica-
tion for structural materials. Int J Fatigue 2010; 32:
565–575.
24. Gagnon M, Tahan A, Bocher P, et al. Influence of load
spectrum assumptions on the expected reliability of
hydroelectric turbines: a case study. Struct Saf 2014; 50:
1–8.
25. Torgeir M and Efren AU. Reliability-based assessment
of deteriorating ship structures operating in multiple sea
loading climates. Reliab Eng Syst Safe 2008; 93: 433–446.
26. Zhao YX, Yang B and Peng JC. Drawing and applica-
tion of Goodman-Smith diagram for the design of rail-
way vehicle fatigue reliability. China Railw Sci 2005; 26:
6–12 (in Chinese).
27. Rui FS, Jesus AMPD, Correia JA FO, et al. A probabilis-
tic fatigue approach for riveted joints using Monte Carlo
simulation. J Constr Steel Res 2015; 110: 149–162.
28. Lu H, He YH and Zhang YM. Reliability-based robust
design of mechanical components with correlated failure
modes based on moment method. Adv Mech Eng. Epub
ahead of print 9 June 2014. DOI: 10.1155/2014/568451.
29. Zhou LR and Ou JP. Finite element based on the model
updating of long-span cable-stayed bridge based on the
response surface method of radial basis function. China
Railw Sci 2012; 33: 8–15 (in Chinese).