You are on page 1of 20

Engineering Optimization

ISSN: 0305-215X (Print) 1029-0273 (Online) Journal homepage: http://www.tandfonline.com/loi/geno20

Lightweight and crashworthiness design of an


electric vehicle using a six-sigma robust design
optimization method

Z. Li, L. B. Duan, A. G. Cheng, Z. P. Yao, T. Chen & W. Yao

To cite this article: Z. Li, L. B. Duan, A. G. Cheng, Z. P. Yao, T. Chen & W. Yao (2018):
Lightweight and crashworthiness design of an electric vehicle using a six-sigma robust design
optimization method, Engineering Optimization, DOI: 10.1080/0305215X.2018.1521396

To link to this article: https://doi.org/10.1080/0305215X.2018.1521396

Published online: 15 Oct 2018.

Submit your article to this journal

Article views: 14

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=geno20
ENGINEERING OPTIMIZATION
https://doi.org/10.1080/0305215X.2018.1521396

Lightweight and crashworthiness design of an electric vehicle


using a six-sigma robust design optimization method
Z. Lia , L. B. Duanb , A. G. Chenga , Z. P. Yaoc , T. Chena and W. Yaoa
a State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body, Hunan University, Changsha,

People’s Republic of China; b School of Automotive and Traffic Engineering, Jiangsu University, Zhenjiang, People’s
Republic of China; c SAIC GM Wuling Automobile Co., Ltd, Liuzhou, People’s Republic of China

ABSTRACT ARTICLE HISTORY


Design optimization plays an important role in electric vehicle (EV) design. Received 21 August 2017
However, fluctuations in design variables and noise factors during the form- Accepted 28 August 2018
ing process affect the stability of optimization results. This study uses six- KEYWORDS
sigma robust design optimization to explore the lightweight design and Electric vehicle safety;
crashworthiness of EVs with uncertainty. A full-scale finite element model of lightweight design; robust
an EV is established. Then, multi-objective design optimization is performed design; adaptive importance
by integrating optimal Latin hypercube sampling, radial basis functions sampling (AIS); material
and non-dominated sorting genetic algorithm-II to achieve minimum peak design
acceleration and mass. Finally, six-sigma robust optimization designs are
applied to improve the reliability and sigma level. Robust optimization
using adaptive importance sampling is shown to be more efficient than
that using Monte Carlo sampling. Moreover, deformation of the battery
compartment and peak acceleration of the B-pillar are greatly decreased.
The EV’s safety performance is improved and the lightweight effect is
remarkable, validating the strong engineering practicability of the method.

1. Introduction
With the increase in the number of vehicles globally, the automobile industry is facing a series of
problems, such as energy shortages and environmental pollution. The development of new-energy
vehicles and lightweight designs has become an inevitable trend, and electric vehicles (EVs) have
the potential to reduce significantly the amount of petroleum used in the next 10–20 years (Pesaran,
Kim, and Keyser 2009). Nowadays, the major obstacles which restrict the development of EVs are
their range and safety. Lightweight design is one of the main ways to reduce energy consumption
and improve the limited range of EVs; it includes the application of lightweight materials, structural
optimization and advanced processing technology, and many studies have been conducted in this
field (Zhao, Liu, and Li 2013; He, Xiao, and Li 2017; Liu et al. 2018; Xiao, Zuo, and Zhou 2018; Wu
et al. 2018; Xiao et al. 2019). Liu et al. (2013) developed a lightweight carbon-fibre reinforced plas-
tic (CFRP) structure for EVs. Ni et al. (2016) studied the application of lightweight materials in the
lightweight design of EVs. Delogu et al. (2017) studied innovative composites and hybrid materials
for the lightweight design of EVs from a sustainability perspective, and Wu et al. (2017) studied the
crashworthiness characteristics of a bio-inspired sandwich structure composed of CFRP panels and
aluminium honeycomb. Overall, the research into lightweight EVs has achieved a certain amount of
progress.

CONTACT A. G. Cheng cheng_aiguo@163.com; Z. P. Yao li2220083084@163.com

© 2018 Informa UK Limited, trading as Taylor & Francis Group


2 Z. LI ET AL.

In addition to lightweight design, the crashworthiness of EVs has become an important research
topic. Improving the passive safety of EVs is an important task for designers (Fang et al. 2017). The
US National Highway Traffic Safety Administration (NHTSA) began to study hydrogen and fuel cell
vehicle safety in the 1990s and has proposed safety performance criteria. Systematic research into
EV collision safety regulation was carried out by the NHTSA (Hennessey and Nguyen 1990). Xie
et al. (2005) investigated EV safety and proposed a new type of vehicle body structure. Wang and Yu
(2010) studied the frontal crashworthiness of mini electric passenger cars through the load path and
structural design method, and found that EVs could achieve good crashworthiness in a frontal crash.
Xie et al. (2011) studied the frontal impact load path of an EV using topology optimization, and the
crashworthiness of the EV was improved considerably. Ge et al. (2013) investigated the crashworthi-
ness of EVs using a structural optimization method, which reduced the peak acceleration by 3.8%.
Boria and Pettinari (2014) studied EV crashworthiness by proposing a new energy-absorption struc-
ture, and compared the energy-absorption capabilities using metallic and composite materials. Lei,
Li, and Wang (2015) studied the frontal crash safety of an EV by adopting a topology optimization
method. Zhu et al. (2016) investigated the crashworthiness analysis of a composite body structure for
EVs under roof-crash and side-pole impacts. Arora, Shen, and Kapoor (2016) investigated the safety
of the EV’s battery in a crash. Hao et al. (2017) studied the crashworthiness of EVs by optimizing
the structure of vehicle battery packs, which provides a new direction to solve the issues faced by EV
designers. Holtz et al. (2017) investigated occupant safety of a subcompact urban EV using innovative
constructive solutions.
The above research works focused on the safety design of EVs with deterministic optimization.
However, uncertainties involved in engineering problems are inevitable; for example, design variables
and noise factors have certain fluctuations during the process of metal forming, which often violate
some constraints (Yang and Gu 2004; Sun, Li, and Chen 2008; Fang et al. 2015; Fang et al. 2017).
In the automotive engineering field, many publications are available regarding non-deterministic
optimization for vehicle crashworthiness design, and researchers have tried to consider robust design
during the process of vehicle design. Du and Chen (2004) proposed the sequential optimization and
reliability assessment method, which was used in reliability-based design for vehicle crashworthi-
ness in a side impact, while Koch, Yang, and Gu (2004) researched the side-impact robustness using a
reliability-based robust optimization design method. Hsu et al. (2007) developed a robustness method
using the six-sigma design concept for frontal impact occupant protection performance. Then, six-
sigma robust methods were used in the design of vehicle crashworthiness by many researchers,
improving the performance and robustness of crashworthiness. Sun et al. (2011) developed a multi-
objective robust optimization for vehicle crashworthiness by introducing the dual response model.
Gu et al. (2013) presented a comparative study on multi-objective reliable and robust optimization
for the crashworthiness design of vehicle structures. Sun et al. (2017) developed a multi-objective
and multi-case reliability-based design optimization (RBDO) to optimize the tailor-rolled blank hat-
shaped structure, and Gu et al. (2017) investigated the optimal crashworthiness design of vehicles
under multiple impact cases with the RBDO method.
In the process of robust design optimization, the sampling-based method using Monte Carlo
sampling (MCS) is generally applied to enable robust analyses, but a large number of sample
points is needed. It is computationally intensive for large-scale systems, especially for vehicle crash-
worthiness design (Gu et al. 2013; Zhu et al. 2015). Therefore, to improve the robustness and
optimization efficiency, researchers have proposed an adaptive importance sampling (AIS) method
for reliability analysis and come up with approximate limit state surface curvature update ideas
(Wu 1992). Shi, Yang, and Zhu (2013) presented an adaptive response surface selection strat-
egy in crashworthiness design optimization, while Zhu et al. (2015) proposed a new sampling-
based RBDO method using a score function with a reweighting scheme to improve the compu-
tational efficiency and accuracy. Duan et al. (2017a) developed an improved multi-objective sys-
tem reliability-based design optimization method to explore the lightweight and high-performance
design of a concept car body and improve the computational efficiency. Xiao, Zuo, and Zhou
ENGINEERING OPTIMIZATION 3

(2018) proposed a new adaptive sequential sampling method for efficient structural reliability
analysis.
To the authors’ knowledge, there are few reports available concerning multi-objective robust crash-
worthiness for EVs. Therefore, it is important to improve the computational efficiency and accuracy
during multi-objective robust optimization for the crashworthiness design of EVs. This article aims
to explore the lightweight and crashworthiness design of EVs by using six-sigma robust design opti-
mization methods and improving the computational efficiency. The article is organized as follows.
Section 2 describes the crash safety of EVs. In Section 3, robust design optimization methodology is
briefly introduced. In Section 4, six-sigma robust design optimization of EVs is performed, followed
by conclusions in Section 5.

2. Description of crash safety of electric vehicles


Compared with conventional vehicles, EVs have more prominent collision safety problems. The bat-
teries are located under the chassis frame; this makes the seating position of an EV somewhat higher
than in a conventional vehicle. The load capacity of an EV is less than that of a conventional vehicle
owing to the energy requirements. The front overhang of an EV is shorter than in conventional cars
because of the limits of length and range, so the front-end components must have a higher energy-
absorption capacity. To avoid fire or explosion after a collision, it is important that the battery packs
are well protected in case of an impact and the deformation of the battery compartment should
be minimized (Li 2013). In recent years, the development of EVs has been overshadowed by fre-
quent safety incidents around the world. Safety under crash conditions is a special concern for EVs
(Figure 1), as seen in cases involving the Chevrolet Volt, the Mitsubishi i-MiEV and other EVs (Smith
2012; Chen et al. 2015).
In addition, because of the special structure of EVs, it is not enough to gain top scores in
European New Car Assessment Programme (Euro NCAP) crash tests, especially for small EVs
(http://www.euroncap.com/en). For example, the Tazzari Zero achieved a score of only 4 in frontal
tests. When this small EV was inspected by workers, it was found that the structure of the vehicle had
reached its limit in the frontal test and would not be capable of withstanding a more severe impact.
The battery pack was pushed backwards into the footwell in the frontal crash test. Although the risk
of injury to the lower legs was not assessed in these tests, such a collapse of the footwell could be
extremely hazardous in a collision.
The battery pack is a core part of an EV, and lithium-ion batteries are considered sensitive to varia-
tions in several factors. It has been suggested that the battery temperature must be maintained below
50°C for safe operation, and the battery cell temperature also affects the reliability and cycle life of
lithium-ion battery packs (Rao et al. 2013). If the battery compartment is seriously deformed during
a collision, the battery cells will be compressed and the temperature of the battery pack will rise, which
will lead to serious safety issues (Arora, Shen, and Kapoor 2016). According to previous research, bat-
tery packs are usually placed in well-protected areas, away from the crumple zone (Xia et al. 2014).

Figure 1. An electric vehicle fire.


4 Z. LI ET AL.

However, evidence from the examples mentioned above indicates that the safety of EVs needs to be
further improved. Therefore, there is a need to study the collision safety of EVs and the safety of the
battery pack.

3. Robust design optimization methodology


3.1. Six-sigma robust optimization
To improve product quality, the Motorola Company put forward a method of design for six sigma (6σ ,
where σ is the standard deviation), which is used to evaluate the quality of products quantitatively
(Jiang and Fang 2012). Six-sigma design has been widely used in product development and in other
fields.
In deterministic optimization, the optimal solution is often located in the constrained boundary,
but random fluctuations in noise factors can easily cause design failures (point 1 in Figure 2). However,
RBDO can overcome the defect and move the design far away from the constraint boundary (point 2
in Figure 2). Robust design focuses on the average value of the objective function, and makes the least
mean square error to reduce its sensitivity to the noise factors simultaneously (point 3 in Figure 2).
The mathematical expression for the deterministic optimization problem is:

Min F(x)
s.t. gj (x) ≤ 0, j = 1, 2, . . . , n (1)
xL ≤ x ≤ xU

where F(x) is the objective function; gj (x) is the constraint function; n is the number of constraint
functions; and xL and xU are the lower limit and the upper limit of design variables, respectively.
The key points of robust design are to find the ‘flat area’ in the design space, reduce the fluctuations
in the output response caused by the uncertainty factors and meet the requirements of the reliability
probability (Koch, Yang, and Gu 2004; Sun et al. 2011; Xiao et al. 2014). The typical mathematical
formula of robust optimization is:

Min G(μF (x), σF (x))


s.t. P(gj (x) ≤ 0) ≥ Rj , j = 1, 2, . . . , n (2)
xL ≤ x ≤ xU

Figure 2. Diagram of deterministic optimization, reliability optimization and robust optimization.


ENGINEERING OPTIMIZATION 5

Figure 3. Adaptive importance sampling.

where μF (x) and σF (x) are the mean and standard deviation of the objective function, respectively;
Rj is the reliability probability meeting specified constraint functions; P(.) is the reliability probability
meeting constrained boundaries; and x = [x1 , . . . xk ]T is a k-dimensional random vector.

3.2. Review of adaptive importance sampling


Types of six-sigma robust optimization analysis include the reliability technique, MCS and design
of experiments (DOE). The reliability technique is a significant topic in structural engineering, and
importance sampling is a type of Monte Carlo reliability method; the central idea in the importance
sampling technique is to lower the sampling effort. The AIS method samples mainly in the failure
region to compute the probability of failure (Ang 2006; Tang and Chen 2009; Shayanfar et al. 2018).
As shown in Figure 3, a transformation is adopted to map the original random vector (in X-space)
to the standard normal vector (in U-space). It is assumed that the design point in the original space
corresponds to the origin in U-space and g(u) is a limit state function to the random vector (also
called the ‘constraint function’). g(u) = 0 is known as the limit state (also called the ‘constrained
boundary’), which divides the design space into the failure region (g(u) > 0) and safe region (g(u) <
0). The minimum distance from the origin of the standard normal space to a point on the limit state
surface is called the most probable point of failure (MPP). Let f (u) represent the joint probability
density function of the random vector, and the failure probability Pf can be written as:

Pf = P[g(u) ≥ 0] = f (u)du (3)
g(u)≥0

For highly nonlinear functions, the integral of Equation (3) is quite difficult, especially for large-scale
practical engineering system problems. However, the AIS method can overcome this difficulty by
generating samples according to a new sampling density function where more samples are likely to
fall into the failure domain. Equation (3) can be written as (Zhu, Zhang, and Chen 2009):

1 
N
f (u) f (uj )
Pf = h(u)du ≈ I(·) (4)
g(u)≥0 h(u) N j=1 h(uj )

where N is the total number of samples; h(u) is the new sampling density function; and I(.) is an
indicator function.

1 g(u) ≥ 0
I(.) = (5)
0 g(u) ≤ 0
6 Z. LI ET AL.

Assuming that the current design point of failure rate is Pf , the reliability for the design point is:

R = 1 − Pf (6)

4. Six-sigma robust optimization and results analysis


4.1. Finite element modelling of an electric vehicle
This article takes a small EV as the research object. The model description is shown in Figure 4. The
EV model includes the body-in-white (BIW), chassis, wheels, battery pack and other components.
First, the Hyperworks software was used to model the EV and numerical analyses were performed
using the nonlinear finite element (FE) code LS-DYNA. The basic size of elements is 10mm × 10mm
and the main element types is a quadrilateral shell. However, the basic size of the elements of some
key structures, such as the front longitudinal beams, sub-frame and crash boxes, is 5mm × 5mm. In
addition, to balance the computation time and model precision, the minimum element size of the FE
model should not be less than 3 mm.
The main contact algorithms used in the model were ‘automatic single surface’, which consid-
ers self-contact between shell elements, and ‘automatic surface to surface’, which considers contact
between the car and the wall. According to engineering experience, the analysis accuracy of FE models
can meet the demands of engineering well when the static and the dynamic coefficients of friction are
defined as 0.1 and 0.2, respectively; and the hourglass control is employed to avoid elements with spu-
rious energies caused by reduced integration (Gedikli 2013). To perform a full frontal crash, the EV
is subjected to a rigid wall impact with an initial velocity of 50 km/h, which is taken from the vehicle
frontal crash regulation of China (GB11551-2014). The structural crashworthiness test is performed
without considering details of the occupant restraint system.
Figure 5(a) shows the FE model of the small EV. There are 531,000 elements and 534,800 nodes
in total, and all the metal components in the BIW are fastened together by welding. The materials

Figure 4. Model description of an electric vehicle. BIW = body-in-white.


ENGINEERING OPTIMIZATION 7

Figure 5. Finite element model: (a) electric vehicle; (b) battery pack.

of the BIW are different kinds of aluminium alloy, including Al-6063_T6, Al-5182, Al-5052 and Al-
6060_T6. In the simulation, strain rate effects were taken into account in the analysis (Smerd et al.
2005; Zhu, Zhu, and Chen 2007). The effect is accounted for through the Cowper–Symonds model
(Halquist 2007), given as:
  1 
ε̇ p
σy = σ0 1 + (7)
C
where ε̇ is the strain rate; σ0 is yield strength; σy is the scaled yield strength; and C and P are the strain
parameters.
In Figure 5(b), the central dark gridlines represent the framework of the battery pack, and the
darker squares represent the eight battery modules. The battery cells were modelled using fully inte-
grated solid elements, and the crushable foam material was taken from the LS-DYNA library. The
shell casing of the cell was discretized using four-node fully integrated shell elements (Sahraei, Camp-
bell, and Wierzbicki 2012). The sizes of the shell and the solid elements were approximately the
same. Owing to the high cost and long production cycle of real EVs, experimental tests could not
be conducted within the scope of this study.

4.2. Definition of design variables and responses


The main methods that improve EV safety under full-width frontal impact are reducing the peak
acceleration, and minimizing the firewall intrusion and the deformation of the battery pack. In gen-
eral, the front end of a vehicle can be divided into three functional zones: the safety cage, transition
zone and crush zone (Figure 6). The safety cage is used to resist the crash load and maintain the
integrity of the passenger compartment and battery pack. The role of the transition zone is to transfer
crash loads from the front to the back end of the vehicle, while the main role of the crush zone is to
absorb as much crash energy as possible (Duan et al. 2016; Duan et al. 2017b).
The main methods used to achieve optimal safety performance during lightweight design of a vehi-
cle, are using new materials, new technology and new structures (Duan et al. 2016). The material and
thickness of the frame structure are adopted as the design domains in this study. In a frontal impact,
the peak acceleration at the left rocker in the B-pillar (αB ) is an important indicator in evaluating the
safety of vehicles, which will reflect the injury risk to occupants through the action of the restraint sys-
tem. As a result, the crashworthiness design aims to minimize the peak acceleration. However, in the
process of EV design, the goals of light weight and crashworthiness often conflict with each other. In
this article, minimization of weight and αB were chosen as the two objectives to be optimized within
a multi-objective framework (Gu et al. 2013).
Figure 6 demonstrates the force transfer paths of the EV. It can be concluded that the front lon-
gitudinal beams and sub-frame are the main components transmitting impact forces. According to
8 Z. LI ET AL.

Figure 6. Frontal impact force transfer path: (a) side view; (b) bottom view.

Figure 7. Design variables and key components: (a) design variables: 1 = front longitudinal beam B; 2 = front longitudinal beam;
3 = front anti-collision beam; 4 = sub-frame anti-collision beam; 5 = crash box; 6 = sub-frame front beam; 7 = sub-frame rear
beam; (b) crash box; (c) anti-collision beam; (d) section of front longitudinal beam.

engineering experience, the lightweight and crashworthiness potential of a frame structure can be
fully exploited by optimizing the parameters of the three functional zones, i.e. the material and thick-
ness of the key components in each functional zone. Therefore, the front longitudinal beam B, front
longitudinal beams, front anti-collision beams, sub-frame and crash boxes are defined as design vari-
ables (Figure 7). All the thickness design variables are considered continuous, which allows them to
be varied in the range from 1.5 to 3.0 mm. The deterministic design variables are x1 ∼ x7 and the
random design variables are x8 ∼ x14 . Design variables and probabilistic distribution attributes are
summarized in Table 1. All design variables are assumed to be normally distributed in this article (Gu
et al. 2013; Duan et al. 2017a).
In Table 1, x1 , x2 , . . . x14 are design variables, e.g. x1 represents the material of the front longitudinal
beam B and x8 presents the thickness of the front longitudinal beam B.
ENGINEERING OPTIMIZATION 9

Table 1. Initial value and variation range of design variables.

Symbol Design variable Type Initial Lower Upper CoV DT


x1 Material of front longitudinal beam B Discrete Al-6060_T6 Al-5052 Al-6063 – –
x2 Material of front longitudinal beam Discrete Al-6060_T6 Al-5052 Al-6063 – –
x3 Material of front anti-collision beam Discrete Al-6060_T6 Al-5052 Al-6063 – –
x4 Material of sub-frame anti-collision beam Discrete Al-5182 Al-5052 Al-6063 – –
x5 Material of crash box Discrete Al-5052 Al-5052 Al-6063 – –
x6 Material of sub-frame front beam Discrete Al-5182 Al-5052 Al-6063 – –
x7 Material of sub-frame rear beam Discrete Al-5182 Al-5052 Al-6063 – –
x8 Thickness of front longitudinal beam B Continuous 3 1.5 3.0 0.03 Normal
x9 Thickness of front longitudinal beam Continuous 2 1.5 3.0 0.03 Normal
x10 Thickness of front anti-collision beam Continuous 2 1.5 3.0 0.03 Normal
x11 Thickness of sub-frame anti-collision beam Continuous 2 1.5 3.0 0.03 Normal
x12 Thickness of crash box Continuous 2 1.5 3.0 0.03 Normal
x13 Thickness of sub-frame front beam Continuous 2 1.5 3.0 0.03 Normal
x14 Thickness of sub-frame rear beam Continuous 2.5 1.5 3.0 0.03 Normal
Note: Material types are Al-5052, Al-5182, Al-6060_T6 and Al-6063. CoV = coefficient of variation; DT = distribution type.

4.3. Parametric study


Since the crash performance of an EV is related to the design variables, it is necessary to study the
influence of design variables on the response. First, sampling points are selected through the DOE
method. The selection of sampling points will be critical for the parametric study as well as for the
construction of surrogate models for multi-objective optimization design.
There are many DOE methods, such as full factorial, orthogonal array and optimal Latin hyper-
cube designs. Optimal Latin hypercube sampling (OLHS) is an experimental design method which is
used to study multifactor problems and is also used for sampling a large design space. The method is
suitable for moderately nonlinear problems and has been used in many articles (Hou et al. 2011; Duan
et al. 2016; Duan et al. 2017c). OLHS can randomly generate sampling points and avoid repeated sam-
pling points. It also has high sampling efficiency. Therefore, OLHS is adopted in this article (Cui et al.
2013).
In the following study, according to the design variables in Table 1, sampling points are selected
using OLHS. First, 240 sampling points are selected, but the precision of the surrogate model cannot
meet the requirements. Thus, more sampling points are selected to update the surrogate models. A
total number of 360 sample points is generated to take into account uncertainties and noise, and
360 FE analyses are calculated using the commercial code LS-DYNA. The responses of the EV are
evaluated at these 360 sampling points.
Figure 8 shows the main effects of the design variables on the response of the EV under frontal
impact. In Figure 8, the slope of the line indicates the effects of design variables on the response:
the larger the slope, the greater the effect. In Figure 8(a), the slopes of x4 and x8 are positive, which
indicates that x4 and x8 have a positive effect on m (where m is the total mass of design variables).
However, in Figure 8(b), the slopes of x4 and x8 are negative, which indicates that x4 and x8 have a
negative effect on αB . It may be concluded that αB and m could conflict with each other and the opti-
mum must be imposed in a Pareto set. Therefore, multi-objective optimization design is conducted
to search for the optimal performance in this study.

4.4. Definition of design optimization problem


For the purpose of seeking a balance between lightweight design and safety performance during the
design process, the peak acceleration of the B-pillar and the mass of design variables are selected as
optimization objectives in this article. The firewall intrusion is another important indicator of vehi-
cle safety, and this should be controlled within the target range. Therefore, the firewall intrusion,
vehicle deformation and vehicle collision time are selected as design constraints. The multi-objective
10 Z. LI ET AL.

Figure 8. Main effects of design variables on the response: (a) main effects of plot m; (b) main effects of plot αB .

optimization model can be described as follows:

Min {αB , m}
s.t. 0 mm < D1 < 500 mm
0 mm < D2 < 130 mm (8)
35 ms < T < 60 ms
xiL ≤ xi ≤ xiU (i = 1, 2, . . . , 14)

where αB is the peak acceleration at the left rocker in the B-pillar; m is the total mass of design vari-
ables; T is the vehicle collision time; D1 is deformation of the vehicle; D2 is the firewall intrusion; and
xiL and xiU are the lower limit and upper limit of design variables, respectively. The multi-objective
design process is shown in Figure 9.

4.5. Establishing surrogate models


Many meta-modelling techniques have been proposed to reduce the computational cost of expensive
simulations of engineering problems, such as radial basis functions (RBFs), response surface method
and kriging models, and comparisons of these meta-modelling techniques can be found in several
references (Jin, Chen, and Simpson 2001; Song et al. 2013; Xiao et al. 2014; Duan et al. 2016).
RBF approximation is a type of neural network employing a hidden layer of radial units and an
output layer of linear units, which are characterized by reasonably fast training and reasonably com-
pact networks. RBF is useful in approximating a wide range of nonlinear spaces and has been used
in many articles for the optimization of engineering design (Kurtaran et al. 2002; Guo et al. 2009;
Kitayama, Arakawa, and Yamazaki 2011). Initialization of the RBF approximation requires at least
2n + 1 design points to be evaluated, where n is the number of inputs. According to the design vari-
ables in Table 1, OLHS is adopted to select sampling points in this article. Then, RBF is adopted to
construct surrogate models; the surrogate models are constructed on the basis of the analysis results
(Jiang and Fang 2012).
The accuracy of the surrogate models can be measured by several criteria. R-square (R2 ) denotes
the fitting degree of the surrogate model and is used to validate the accuracy of the surrogate model,
RAAE denotes the relative average absolute error and RMSE denotes the root mean square error. In
general, the value of R2 is no less than 0.95 in engineering applications (Jiang and Fang 2012). Twenty
random points are selected to analyse the error of the surrogate models and the results are listed in
Table 2.
As shown in Table 2, the precision of the surrogate models meets the requirements. Therefore,
RBF models are adopted in the following optimization design process.
ENGINEERING OPTIMIZATION 11

Figure 9. Flowchart of the optimization design process. EV = electric vehicle; DOE = design of experiments; AIS = adaptive
importance sampling.

Table 2. Error analysis of responses.


Response R2 RAAE RMSE
αB (g) 0.97894 0.0176 0.0556
m (kg) 0.95542 0.0252 0.0597
T (ms) 0.98256 0.0135 0.0426
D1 (mm) 0.99936 0.0023 0.0073
D2 (mm) 0.99042 0.0257 0.0305
Note: RAAE = relative average absolute error; RMSE = root mean
square error.

4.6. Optimization process


Depending on the surrogate models, deterministic optimization is first performed using the non-
dominated sorting genetic algorithm-II (NSGA-II). NSGA-II is a multi-objective genetic algorithm
based on a fast non-dominated sorting principle, which can find the Pareto front more efficiently
and can also avoid focusing on the local Pareto front. This algorithm has proven to be effective in
solving crashworthiness design problems (Liao et al. 2008a; Murugan, Kannan, and Baskar 2009).
The optimization parameters in this article are set as in Table 3 (Jiang and Fang 2012).
The optimization progress is run 1600 times in total and the number of feasible solutions is 1420.
The deterministic optimization results are listed in Table 4: αB is decreased from 52.9 g to 39.6 g, m
is decreased from 8.74 kg to 7.01 kg, and the firewall intrusion is less than 130 mm. The optimization
results show that the safety performance and the lightweight effect are acceptable simultaneously.
Then, reliability analysis is carried out to evaluate the sigma level of the optimization solutions.
As shown in Table 5, the sigma level of m is 5.5, which does not meet the design requirements. In
12 Z. LI ET AL.

Table 3. Optimization parameters of non-dominated


sorting genetic algorithm-II (NSGA-II).
Parameter Value
Population size 32
Number of generations 50
Crossover probability 0.9
Crossover distribution index 15
Mutation distribution index 20

Table 4. Deterministic optimization results.


Symbol Initial Optimization results
Design variable x1 Al-6060_T6 Al-6060_T6
x2 Al-6060_T6 Al-5182
x3 Al-6060_T6 Al-6060_T6
x4 Al-5182 Al-5182
x5 Al-5052 Al-5052
x6 Al-5182 Al-5182
x7 Al-5182 Al-6060_T6
x8 3 2.5
x9 2 1.8
x10 2 1.6
x11 2 2.4
x12 2 1.6
x13 2 1.5
x14 2.5 1.8
Constraint function T 39.8 44.3
D1 328.8 346.8
D2 126.1 129.6
Objective function αB 52.9 39.6
m 8.74 7.01

Table 5. Results of reliability analysis.


Symbol Initial Optimization results Reliability (%) Std deviation σ Level
Constraint function T 39.8 44.3 100 – 8
D1 328.8 346.8 100 – 8
D2 126.1 129.6 100 – 8
Objective function αB 52.9 39.6 100 0.438 8
m 8.74 7.01 100 0.036 5.5

addition, the standard deviation of αB is 0.438, which means that the optimization results will be
unstable (Sun, Li, and Chen 2008). It can be concluded that the reliability and sigma level should be
improved using six-sigma optimization methods.
To improve the reliability and sigma level, robust optimization is carried out in this study. The
mathematical model of six-sigma robust optimization is given by:

Min {μ(αB ) + 6σ (αB ), μ(m) + 6σ (m)}


s.t. 0 mm < μ(D1 ) + 6σ (D1 ) < 500 mm
0 mm < μ(D2 ) + 6σ (D2 ) < 130 mm (9)
35 ms < μ(T) + 6σ (T) < 60 ms
xiL ≤ μ(xi ) ± 6σ (xi ) ≤ xiU

where σ (x) is the standard deviation; μ(x) is the mean value; 6σ is the sigma level; and xiL and xiU are
the lower limit and upper limit of the design variables, respectively. Two different sampling methods
ENGINEERING OPTIMIZATION 13

Figure 10. Comparison of Pareto frontiers for different optimization. MCS = Monte Carlo sampling; AIS = adaptive importance
sampling.

are used during the six-sigma optimization progress: MCS and AIS reliability techniques. NSGA-II
is adopted to search for the optimal solution in this optimization process.

4.7. Result and discussion


The Pareto frontiers of the robust optimization and deterministic optimization are plotted in Figure
10. It can be seen that the Pareto frontier generated by the AIS method is very close to that of the MCS
method, which indicates that the robust optimization with the AIS method has sufficient accuracy to
solve the engineering problems. Moreover, Figure 10 shows that the variation in the Pareto frontier
obtained by the deterministic optimization is much larger than that obtained by robust optimization
when uncertainties are present. Therefore, multi-objective robust optimization is capable of reducing
the performance fluctuations of Pareto solutions induced by random variables. In general, the Pareto
frontiers can provide designers with a large number of design solutions. It is important to make a
decision about the most satisfactory solution from the Pareto frontiers. In this article, the minimum
distance selection method is adopted to obtain the knee point (Sun et al. 2011; Gu et al. 2013).
Table 6 shows the robust optimization results using the different sampling methods. The robust
optimization results obtained with MCS and AIS are close to each other. Compared with the deter-
ministic optimization results, the sigma level of m of robust optimization has improved from 5.5 to
8. In addition, the standard deviations of αB and m are smaller. The reliability of the robust optimiza-
tion results has improved significantly and the values of the design variables are no longer close to
the constraint boundary.
To illustrate the advantages of different six-sigma analysis methods, the results based on MCS and
AIS are compared in Table 7. The main difference between these two different robust optimization
methods is the computing time. The computing time of robust optimization with the MCS method
is 1307 min, while the computing time with the AIS method is only 391 min; thus, the computational
efficiency has improved greatly. The results show that robust optimization with the AIS method has
better efficiency, while achieving the same accuracy as the MCS method.
The results of the design variables obtained with the deterministic optimization and robust opti-
mization were applied in an FE model. The simulation results are listed in Table 8, where the error
between the optimal solution and the FE simulation result meets the requirement of precision.
The results of the FE simulations with the optimized variables were selected to indicate the
improvements in the robust design. Figure 11 shows a comparison of the acceleration curves. The line
through the circles represents the initial acceleration curve and the peak acceleration is 52.8 g. The
line through the triangles represents the deterministic optimization acceleration curve and the peak
14 Z. LI ET AL.

Table 6. Six-sigma (6σ ) robust optimization results.

Deterministic Robust optimization Robust optimization


optimization based on MCS based on AIS
Results σ Level Results σ Level Results σ Level
Design variable x1 Al-6060_T6 – Al-6060_T6 – Al-6060_T6 –
x2 Al-5182 – Al-5182 – Al-5182 –
x3 Al-6060_T6 – Al-5182 – Al-5182 –
x4 Al-5182 – Al-6060_T6 – Al-6060_T6 –
x5 Al-5052 – Al-5182 – Al-5182 –
x6 Al-5182 – Al-5182 – Al-5182 –
x7 Al-6060_T6 – Al-5182 – Al-5182 –
x8 2.5 – 2.6 – 2.7 –
x9 1.8 – 2.0 – 2.0 –
x10 1.6 – 1.8 – 1.7 –
x11 2.4 – 2.4 – 2.4 –
x12 1.6 – 2.0 – 2.0 –
x13 1.5 – 1.6 – 1.6 –
x14 1.8 – 1.8 – 1.8 –
Constraint function T 44.3 8 42.2 8 41.8 8
D1 346.8 8 340.3 8 332.7 8
D2 129.6 8 125.7 8 125.3 8
Objective function αB 39.6 8 42.4 8 42.5 8
m 7.01 5.5 7.43 8 7.46 8
Note: MCS = Monte Carlo sampling; AIS = adaptive importance sampling.

Table 7. Results with different robust optimization methods.

Deterministic Robust Robust


optimization design (MCS) design (AIS)
Computing time (min) 0.3 1307 391
Standard deviation of αB 0.438 0.070386 0.13849
Standard deviation of m 0.036 0.028831 0.029606
Note: MCS = Monte Carlo sampling; AIS = adaptive importance sampling.

Table 8. Comparison of optimal solution with finite element (FE) simulation results.
Deterministic optimization Robust design (AIS)
Description Optimal FE Error % Optimal FE Error %
αB (g) 39.6 39.9 0.08 42.5 42.9 0.09
m (kg) 7.01 6.86 2.14 7.46 7.28 2.41
Note: AIS = adaptive importance sampling.

acceleration is 39.9 g. The line through the squares represents the robust optimization acceleration
curve and the peak acceleration is 42.9 g. The peak acceleration of the B-pillar decreases from 52.8 g
to 42.9 g after robust optimization, which will help to decrease occupant injury. The peak accelera-
tion of robust optimization is larger than that of deterministic optimization, but the curve has better
stability. The simulation results from the robust design show that appropriate redistribution of the
thickness of key components can enhance the crashworthiness.
Figure 12 compares the deformation of the front rail before and after optimization. In the deter-
ministic optimization model, the deformation of the front rail is not ideal as it has undergone severe
bending deformation during the impact. In the robust optimization model, the deformation of the
front rail structures has a more appropriate axial deformation mode under frontal impact. The results
show that the stability of the deformation mode has been greatly improved.
Figure 13 presents the deformation of the battery compartment before and after optimization. In
the initial model, the deformation of the front beam of the battery compartment is serious and the
ENGINEERING OPTIMIZATION 15

Figure 11. Comparison of crash pulses before and after optimization.

Figure 12. Comparison of front rail deformation: (a) front rail deformation; (b) initial model; (c) deterministic optimization model;
(d) robust optimization model.

Figure 13. Comparison of battery compartment deformation: (b) initial model; (c) deterministic optimization model; (d) robust
optimization model.

maximum X-direction deformation is 91.8 mm. The power battery wire harness is severely extruded;
it is likely to hit the battery pack and further damage the battery pack during the collision. In the
deterministic optimization model, the deformation of the front beam of the battery compartment
has been decreased greatly, and the maximum X-direction deformation is 45.7 mm. In the robust
optimization model, the maximum X-direction deformation of the battery compartment is 41.7 mm.
Thus, both battery safety and passenger safety will be improved. The results show that battery packs
should be placed in well-protected areas and away from the crumple zone.
16 Z. LI ET AL.

The smaller the deformation of the battery compartment, the higher the safety of the battery
pack. It can be concluded that using the optimal material and thickness for key energy-absorbing
components can not only greatly reduce the weight of the EV but also enhance its crashworthiness.

5. Conclusions and further work


In this work, the lightweight and crashworthiness design of an EV using six-sigma robust optimiza-
tion methods is studied. The following conclusions can be summarized:

1) Compared with deterministic optimization results, the robustness of robust optimization results
has been improved significantly. It can be concluded that it is necessary to consider robust
optimization during the process of optimization design.
2) Two kinds of sampling method were used in the six-sigma robust optimization. The main differ-
ence between these robust optimization methods is the computing time. The computing time of
robust optimization with the MCS method is 1307 min, while the computing time with the AIS
method is only 391 min. The results indicate that the robust optimization method based on AIS
is more suitable for the optimization of large nonlinear collision problems.
3) The optimal design variables obtained using the robust optimization based on AIS were applied
in an FE model. The results show that the crashworthiness of the EV was greatly improved and the
lightweight effect of the BIW was acceptable. The optimized peak acceleration of the B-pillar was
reduced by 18.9% and the X-direction deformation of the battery compartment was decreased.
These factors will vastly improve the safety of the EV.

This study was conducted under only one load-case, whereas crash scenarios of EVs usually involve
offset impacts, side impacts, etc. These crash load-cases are often considered simultaneously in the
development of vehicle products. Therefore, further research will consider other load-cases in the
lightweight and safety design of EVs.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This work was supported by the International Science & Technology Cooperation Program of China [contract number
2016YFE0102200]. The authors also acknowledge the project National Natural Science Foundation of China [grant
number 51805221], Research Project of State Key Laboratory of Mechanical Systems and Vibration [grant number
MSV201711] and the International Cooperation Technology Center Platform Project [number 20170204].

References
Ang, A. H.-S. 2006. Probability Concepts in Engineering Planning and Design. New York: John Wiley & Sons.
Arora, S., W. Shen, and A. Kapoor. 2016. “Review of Mechanical Design and Strategic Placement Technique of a Robust
Battery Pack for Electric Vehicles.” Renewable and Sustainable Energy Reviews 60: 1319–1331.
Boria, Simonetta, and Silvia Pettinari. 2014. “Mathematical Design of Electric Vehicle Impact Attenuators: Metallic vs
Composite Material.” Composite Structures 115: 51–59.
Chen, R., K.-S. Choi, A. Daniello, and H. Gabler. 2015. “An Analysis of Hybrid and Electric Vehicle Crashes in the
U.S.” Paper presented at the 24th International Technical Conference on the Enhanced Safety of Vehicles (ESV),
Gothenburg, Sweden, June 8–11.
Cui, J., W. Zhang, L. Xie, and W. Chang. 2013. “A Robustness Analysis on Vehicle Crashworthiness Based on Kriging
Metamodel.” Automotive Engineering 35 (1): 51–55.
Delogu, M., L. Zanchi, C. A. Dattilo, and M. Pierini. 2017. “Innovative Composites and Hybrid Materials for Electric
Vehicles Lightweight Design in a Sustainability Perspective.” Materials Today Communications 13: 192–209.
Du, X., and W. Chen. 2004. “Sequential Optimization and Reliability Assessment Method for Efficient Probabilistic
Design.” Journal of Mechanical Design 126: 225–233.
ENGINEERING OPTIMIZATION 17

Duan, L. B., G. Li, A. Cheng, G. Sun, and K. Song. 2017a. “Multiobjective System Reliability-Based Optimization
Method for Design of a Fully Parametric Concept Car Body.” Engineering Optimization 49 (7): 1247–1263.
Duan, L. B., G. Sun, J. Cui, T. Chen, A. Cheng, and G. Li. 2016. “Crashworthiness Design of Vehicle Structure with
Tailor Rolled Blank.” Structural and Multidisciplinary Optimization 53: 321–338.
Duan, L. B., N.-c. Xiao, Z. Hu, G. Li, and A. Cheng. 2017b. “An Efficient Lightweight Design Strategy for Body-
in-White Based on Implicit Parameterization Technique.” Structural and Multidisciplinary Optimization 55 (5):
1927–1943.
Duan, L. B., N.-c. Xiao, G. Li, F. Xu, T. Chen, and A. Chang. 2017c. “Bending Analysis and Design Optimization of
Tailor-Rolled Blank Thin-Walled Structures with Top-Hat Sections.” International Journal of Crashworthiness 22
(3): 227–242.
Fang, J. G., Y. Gao, G. Sun, C. Xu, and Q. Li. 2015. “Multiobjective Robust Design Optimization of Fatigue Life for a
Truck Cab.” Reliability Engineering & System Safety 135: 1–8.
Fang, J. G., G. Sun, N. Qiu, N. H. Kim, and Q. Li. 2017. “On Design Optimization for Structural Crashworthiness and
Its State of the Art.” Structural and Multidisciplinary Optimization 55 (3): 1099–1119.
GB11551-2014. 2014. The Protection of the Occupants in the Event of a Frontal Collision for Motor Vehicle. Beijing: China
National Standardization Management Committee.
Ge, D. D., Q. Wang, W. Liu, and F. Zhao. 2013. “Frontal Crashworthiness Structure Analysis and Optimization of Electric
Vehicle.” Journal of Mechanical & Electrical Engineering 30 (3): 325–328.
Gedikli, Hasan. 2013. “Crashworthiness Optimization of Foam-Filled Tailor-Welded Tube Using Coupled Finite
Element and Smooth Particle Hydrodynamics Method.” Thin-Walled Structures 67: 34–48.
Gu, X. G., J. Dai, X. Huang, and G. Li. 2017. “Reliable Optimisation Design of Vehicle Structure Crashworthiness Under
Multiple Impact Cases.” International Journal of Crashworthiness 22 (1): 26–37.
Gu, X. G., G. Sun, G. Li, L. Mao, and Q. Li. 2013. “A Comparative Study on Multiobjective Reliable and Robust Opti-
mization for Crashworthiness Design of Vehicle Structure.” Structural and Multidisciplinary Optimization 48 (3):
669–684.
Guo, S. L., H. Jin, X. Liu, and L. Liu. 2009. “Shape Parameter Optimization for Wing and Wingtip Based on RBF.”
Journal of Nanjing University of Aeronautics & Astronautics 41 (3): 308–313.
Halquist, J. 2007. LS-DYNA Keyword User’s Manual Version 971. Livermore, CA: Livermore Software Technology
Corporation.
Hao, F., X. Lu, Y. Qiao, and X. Chen. 2017. “Crashworthiness Analysis of Electric Vehicle With Energy-Absorbing
Battery Modules.” Journal of Engineering Materials and Technology 2: 1–4.
He, Z. C., X. Xiao, and E. Li. 2017. “Design for Structural Vibration Suppression in Laminate Acoustic Metamaterials.”
Composites Part B: Engineering 111: 237–252.
Hennessey, Barbara C., and N. T. Nguyen. 1990. “Status of NHTSA’s Hydrogen and Fuel Cell Vehicle Safety Research
Program.” National Highway Traffic Safety Administration USA. 07-0046.
Holtz, J., I. Eguia, M. Pedersen, R. Molinero, M. Tress, R. Torres, and J. Finez. 2017. “Crashworthiness Enhancement
of a Composite Intensive, Multimaterial Fully- Electric Urban Car.” Paper presented at the 6th Hybrid and Electric
Vehicles Conference (HEVC 2016), November 2–3.
Hou, S., X. Han, G. Sun, S. Long, W. Li, X. Yang, and Q. Li. 2011. “Multiobjective Optimization for Tapered Circular
Tubes.” Thin-Walled Structures 49 (7): 855–863.
Hsu, J., A. Kunihiro, A. Yasin, H. Nagumo, T. Tajima, and K. Sugimoto. 2007. “Development of a Robust Design
Approach for Occupant Protection Performance in Frontal Impact using Design for Six Sigma.” SAE Technical Paper
2007-01-0992.
Jiang, X., and L. Q. Fang. 2012. Isight Parameter Optimization Theory and Example Explanation. Beijing: Beijing
University of Aeronautics and Astronautics Press.
Jin, R., W. Chen, and T. W. Simpson. 2001. “Comparative Studies of Metamodelling Techniques Under Multiple
Modelling Criteria.” Structural and Multidisciplinary Optimization 23 (1): 1–13.
Kitayama, S., M. Arakawa, and K. Yamazaki. 2011. “Sequential Approximate Optimization Using Radial Basis Function
Network for Engineering Optimization.” Optimization and Engineering 12: 535–557.
Koch, P. N., R.-J. Yang, and L. Gu. 2004. “Design for Six Sigma through Robust Optimization.” Structural and
Multidisciplinary Optimization 26 (3-4): 235–248.
Kurtaran, H., A. Eskandarian, D. Marzougui, and N. E. Bedewi. 2002. “Crashworthiness Design Optimization Using
Successive Response Surface Approximations.” Computational Mechanics 29: 409–421.
Lei, Z. B., T.-x. Li, and R. Wang. 2015. “Multi-objective Topology Optimization Design of Pure Electric Vehicle Body.”
Journal of Dalian University of Technology 55 (5): 484–491.
Li, W. L. 2013. Front Impact Safety Analysis and Optimization of Electric Vehicle. Wu Han, Hu Bei: Wu Han University
of Technology.
Liao, X. T., Q. Li, X. Yang, W. Li, and W. Zhang. 2008a. “A Two-Stage Multi-objective Optimisation of Vehicle
Crashworthiness Under Frontal Impact.” International Journal of Crashworthiness 13 (3): 279–288.
Liao, X. T., Q. Li, X. Yang, W. Zhang, and W. Li. 2008b. “Multiobjective Optimization for Crash Safety Design of Vehicles
Using Stepwise Regression Model.” Structural and Multidisciplinary Optimization 35 (6): 561–569.
18 Z. LI ET AL.

Liu, Q., Y. Lin, Z. Zong, G. Sun, and Q. Li. 2013c. “Lightweight Design of Carbon Twill Weave Fabric Composite Body
Structure for Electric Vehicle.” Composite Structures 97: 231–238.
Liu, Q., J. Wang, X. Huang, T. Wu, Z. Zong, and Q. Li. 2018. “In-Plane and Out-of-Plane Bending Responses of Alu-
minum Mortise-Tenon Joints in Lightweight Electric Vehicle Inspired by Timber Structures.” Thin-Walled Structures
127: 169–179.
Murugan, P., S. Kannan, and S. Baskar. 2009. “NSGA-II Algorithm for Multi-objective Generation Expansion Planning
Problem.” Electric Power Systems Research 79: 622–628.
Ni, S. Y., J. Q. Wang, S. Wang, and Y. Z. Wang. 2016. “The Application of Lightweight Materials in Lightweight Design
of Electric Vehicles.” Automotive Technology 7: 55–57.
Pesaran, Ahmad A., G-H. Kim, and M. Keyser. 2009. “Integration Issues of Cells Into Battery Packs for Plug-In
and Hybrid Electric Vehicles.” Presented at EVS-24 International Battery, Hybrid and Fuel Cell Electric Vehicle
Symposium, Stavanger, Norway, May 13–16.
Rao, Z., S. Wang, M. Wu, Z. Lin, and F. Li. 2013. “Experimental Investigation on Thermal Management of Electric
Vehicle Battery with Heat Pipe.” Energy Conversion and Management 65: 92–97.
Sahraei, E., J. Campbell, and T. Wierzbicki. 2012. “Modeling and Short Circuit Detection of 18650 Li-Ion Cells Under
Mechanical Abuse Conditions.” Journal of Power Sources 220: 360–372.
Shayanfar, M. A., M. A. Barkhordari, M. Barkhori, and M. Barkhori. 2018. “An Adaptive Directional Importance
Sampling Method for Structural Reliability Analysis.” Structural Safety 70: 14–20.
Shi, L., R.-J. Yang, and P. Zhu. 2013. “An Adaptive Response Surface Method for Crashworthiness Optimization.”
Engineering Optimization 45 (11): 1365–1377.
Smerd, R., S. Winkler, C. Salisbury, M. Worswick, D. Lloyd, and M. Finn. 2005. “High Strain Rate Tensile of Automotive
Aluminum Alloy Sheet.” International Journal of Impact Engineering 32: 541–560.
Smith, B. 2012. “Chevy Volt Battery Incident Summary Report.” Technical report, NHTSA. Report Number: DOT HS
811573.
Song, X. G., G. Sun, G. Li, W. Gao, and Q. Li. 2013. “Crashworthiness Optimization of Foam-Filled Tapered
Thin-Walled Structure Using Multiple Surrogate Models.” Structural and Multidisciplinary Optimization 47 (2):
221–231.
Sun, G. Y., G. Y. Li, and T. Chen. 2008. “Sheet Metal Forming Based Six Sigma Robust Optimization Design.” Chinese
Journal of Mechanical Engineering 44 (11): 248–254.
Sun, G. Y., G. Li, S. Zhou, H. Li, S. Hou, and Q. Li. 2011. “Crashworthiness Design of Vehicle by Using Multiobjective
Robust Optimization.” Structural and Multidisciplinary Optimization 44: 99–110.
Sun, G. Y., H. Zhang, J. Fang, G. Li, and Q. Li. 2017. “Multi-objective and Multi-case Reliability-Based Design
Optimization for Tailor Rolled Blank (TRB) Structures.” Structural and Multidisciplinary Optimization 55 (5):
1899–1916.
Tang, Y. C., and J. Chen. 2009. “Robust Design of Sheet Metal Forming Process Based on Adaptive Importance
Sampling.” Structural and Multidisciplinary Optimization 39: 531–544.
Wang, D. Z., and C. X. Yu. 2010. “Frontal Crashworthiness Structure Design of Mini Electric Passenger Cars.” Journal
of Automotive Safety and Energy 1 (1): 49–52.
Wu, Y.-T. 1992. “An Adaptive Importance Sampling Method for Structural System Reliability Analysis, Reliability Tech-
nology.” In Proceedings of the Symposium on Reliability Technology, Vol. 28, 217–231. New York: American Society
of Mechanical Engineers.
Wu, Y. H., Q. Liu, J. Fu, Q. Li, and D. Hui. 2017. “Dynamic Crash Responses of Bio-inspired Aluminum Honeycomb
Sandwich Structures with CFRP Panels.” Composites Part B: Engineering 121: 122–133.
Wu, F., X. Xiao, J. Yang, and X. Gao. 2018. “Quasi-static Axial Crushing Behavior and Energy Absorption of Novel
Metal Rope Crochet-Sintered Mesh Tubes.” Thin-Walled Structures 127: 120–134.
Xia, Y., Tomasz Wierzbicki, Elham Sahraei, and Xiaowei Zhang. 2014. “Damage of Cells and Battery Packs Due to
Ground Impact.” Journal of Power Sources 267: 78–97.
Xiao, X., Z. C. He, Eric Li, and A. G. Cheng. 2019. “Design Multi-stopband Laminate Acoustic Metamaterials for
Structural-Acoustic Coupled System.” Mechanical Systems and Signal Processing 115: 418–433.
Xiao, N. C., Yan-Feng Li, Yuanjian Yang, Le Yu, and Hong-Zhong Huang. 2014. “A Novel Reliability Method for
Structural Systems with Truncated Random Variables.” Structural Safety 50: 57–65.
Xiao, N. C., Ming J. Zuo, and Chengning Zhou. 2018. “A New Adaptive Sequential Sampling Method to Construct
Surrogate Models for Efficient Reliability Analysis.” Reliability Engineering and System Safety 169: 330–338.
Xie, Q. X., W. G. Zhang, and Z. H. Zhong. 2005. “The Hidden Troubles and Their Counter Measures of Electric Vehicle
Safety.” Bus Technology and Research 27 (2): 8–11.
Xie, L. J., W. G. Zhang, W. B. Chang, and J. Cui. 2011. “Frontal Crashworthiness Structure Research of Electric Car.”
The 9th International forum of automotive traffic safety (INFATS), Changsha, China.
Yang, R., and L. Gu. 2004. “Experience with Approximate Reliability-Based Optimization Methods.” Structural and
Multidisciplinary Optimization 26: 152–159.
Zhao, G. F., Ming Min Liu, and Yan Li. 2013. “Safety Research and Lightweight Design of Electric Vehicle Battery
Protection Device.” Advanced Materials Research 724-725: 1366–1373.
ENGINEERING OPTIMIZATION 19

Zhu, G. H., A. Cheng, Z. Wang, and F. Lei 2016. “Analysis of Lightweight Composite Body Structure for Electrical
Vehicle Using the Multiscale Approach.” Journal of Mechanical Engineering 52 (6): 145–152.
Zhu, H., L. Zhu, and J.-H. Chen. 2007. “Influence of Stress Triaxiality and Strain Rate on the Mechanics Behavior of
6063 Aluminum Alloy and Material Characterization.” Journal of Materials Science and Engineering 25 (3): 358–362.
Zhu, P., Lei Shi, Ren-Jye Yang, and Shih-Po Lin. 2015. “A New Sampling-Based RBDO Method via Score Function with
Reweighting Scheme and Application to Vehicle Designs.” Applied Mathematical Modelling 39 (15): 4243–4256.
Zhu, P., Y. Zhang, and G.-L. Chen. 2009. “Metamodel-Based Lightweight Design of an Automotive Front-Body
Structure Using Robust Optimization.” Proceedings of the Institution of Mechanical Engineers, Part D: Journal of
Automobile Engineering 223: 1133–1147.

You might also like