Struct Multidisc Optim (2017) 55:1091–1119

DOI 10.1007/s00158-016-1579-y

REVIEW ARTICLE

On design optimization for structural crashworthiness

and its state of the art
Jianguang Fang 1 & Guangyong Sun 1 & Na Qiu 2 & Nam H. Kim 2 & Qing Li 1

Received: 30 July 2015 / Revised: 21 August 2016 / Accepted: 23 August 2016 / Published online: 16 September 2016
# Springer-Verlag Berlin Heidelberg 2016

Abstract Optimization for structural crashworthiness and en- 1 Introduction

ergy absorption has become an important topic of research
attributable to its proven benefits to public safety and social 1.1 Motivation of crashworthiness optimization
economy. This paper provides a comprehensive review of the
important studies on design optimization for structural crash- Motorization brings two significant challenges to the modern
worthiness and energy absorption. First, the design criteria society. Firstly, road and vehicle safety becomes increasingly
used in crashworthiness and energy absorption are reviewed important, which has notably heightened legislative require-
and the surrogate modeling to evaluate these criteria is ments by introducing more effective protective systems to the
discussed. Second, multiobjective optimization, optimization vehicle. Secondly, there is an ever-growing concern in envi-
under uncertainties and topology optimization are reviewed ronment and sustainability, which largely push up the
from concepts, algorithms to applications in relation to crash- lightweighting standards to reduce fuel consumption. For
worthiness. Third, the crashworthy structures are summarized, these reasons, the automotive industry has devoted a substan-
from generically novel structural configurations to industrial tial effort to deliver more crashworthy vehicles for addressing
applications. Finally, some conclusions and recommendations these two competing issues concurrently.
are provided to enable academia and industry to become more Vehicle crash brings increasing concerns from socioeco-
aware of the available capabilities and recent developments in nomic aspects. Each year vehicle crash leads to about 1.2
design optimization for structural crashworthiness and energy million deaths and many more injuries worldwide, becoming
absorption. one of the biggest public health issues facing modern society.
In USA, the most motorized country, for example, the crash-
induced fatality has remained at a considerably high level,
Keywords Literature review . Crashworthiness optimization . though certain reduction, over the past decade (Fig. 1), leading
Surrogate model . Uncertainty . Multiobjective optimization . to the direct annual cost of US$277 billion, equivalent to near-
Topology optimization . Energy absorption ly US$900 per head or 1.9 % of real Gross Domestic Product
(GDP) of the country (http://www.nhtsa.gov/). Fig. 1 also pro-
* Guangyong Sun vides the fatality data in China, a rapidly developing motori-
[email protected] zation country. A total of 58,539 people were killed in acci-
* Qing Li dents in 2013 (http://www.stats.gov.cn/.) Such high
[email protected] socioeconomic burden places increasing attention in road
Jianguang Fang
and vehicle safety. As an important means, crashworthiness
[email protected] design of vehicle structures has proven highly effective,
through which at least 43 % of potential fatalities could have
1
School of Aerospace, Mechanical and Mechatronic Engineering, The been prevented and much more injuries avoided as per the
University of Sydney, Sydney, NSW 2006, Australia study by O’Neill (2009), making crashworthiness research
2
Department of Mechanical & Aerospace Engineering, University of and development draw growing interest over recent years.
Florida, Gainesville, FL 32611, USA For example, front rails of automobiles and trains (Fig. 2) have
1092 J. Fang et al.

possible crashworthiness and lightest possible structure for

various vehicles.

1.2 Theme of this review

In general, a problem of crashworthiness optimization can be

formulated mathematically as:
8
< min f ðxÞ
s:t: gðxÞ ≤ 0 ð1Þ
:
xL ≤ x ≤ x U

where f(x) and g(x) are the objective vector and constraint
vector, respectively. x denotes the vector of design variables.
Fig. 1 Fatalities caused by motor vehicle crashes in USA and China This review of literature will be conducted based on (1) as
follows. (1) classification of crashworthiness criteria adopted
been extensively studied to protect passengers from fatal or in various optimization objectives and constraints; (2) formu-
severe injuries during the collision. To rate crashworthiness of lations of crashworthiness criteria (objectives and constraints)
vehicle, the new car assessment program (NCAP) was for optimization; (3) optimization strategies for size, shape
established to provide consumers with comparative ratings. and topology under single-/multiobjective with/without un-
In many ways, NCAP has largely lifted the standard of con- certainties; and (4) applications of crashworthiness optimiza-
sumer’s expectation and vehicle crashworthiness, virtually tion, ranging from novel structural configurations, such as
making all the new vehicles become five-star (highest) in or- sectional profile of energy absorbers, to entire vehicular and
der to enter the market (O’Neill (2009)). other engineering structures.
On the other hand, lightweighting of the vehicle has been
driven by emergent concerns in fuel consumption (over 1
billion Liters per day for light vehicles such as passenger cars) 2 Design criteria for crashworthiness and energy
and environment due to rapid motorization (Davis et al. absorption
(2013)). The statistics show that the fuel consumption is di-
rectly proportional to vehicle weight; specifically, 10 % reduc- 2.1 Classification of crashworthiness criteria
tion of weight could lead to 6–8 % saving in fuel consumption
(Zhang et al. (2007b)). Furthermore, lightweighting increases 2.1.1 Injury-based metrics
the range of the vehicle, which is a determinant for the use of
the electric vehicle (EV). Lured by such a heightening require- From a biomechanics point of view, occupants’ responses to a
ment, a series of recent articles in Nature and Science clearly crash can be measured by such indices as head injury criteria
indicated a replenished research interest and significant en- (HIC), chest acceleration, chest deflection and femur loads
deavor in pursuing lightweight materials and structures (Kim under impact (Du Bois et al. (2004)). These indices are affect-
et al. (2015)). Yet the question remaining to be addressed is ed by the vehicle crash pulse (CP), the magnitude of intrusion
whether their crashworthiness meets the requirements or there (Intr) into the occupant compartment, restraint system, and
is a solution in balancing crashworthiness and lightweighting vehicle interiors. Among them, CP and Intr are the direct
most effectively. Over the past two decades, design optimiza- consequences of structural crashworthiness. As a variant of
tion has been developed as a powerful tool to seek a highest Intr, intrusion velocity (IntrV) is also utilized as a design

Fig. 2 Energy absorbers in

automobile/train structures
(Marsolek and Reimerdes (2004))

Energy absorbers
On design optimization for structural crashworthiness 1093

criterion in the literature. In general, a high acceleration im- element to be as uniform as possible over the whole design
plies a large impact force exerted on occupants and can result domain. Table 2 in Appendix summarizes the energy-based
in a high risk of injury to occupants. For this reason, peak metrics in literature.
acceleration (amax) and peak crash force (Fmax) during impact
are extensively employed as design criteria for optimization. 2.2 Remarks on design criteria
Table 1 in Appendix summarizes commonly used injury-
based metrics in literature. Note that the selection of design criteria prior to optimization
has been considered critically important in order to obtain a
most ‘beneficial’ optimal design. However, the researchers
2.1.2 Energy-based metrics have not reached a consensus yet to date. For example,
Marzbanrad and Ebrahimi (2011) and Shakeri et al. (2007)
The crashworthy structures are expected to absorb as much preferred a long effective distance to make full use of the
energy as possible so as to reduce the kinetic energy transmit- energy-absorbing capacity, whereas the others (e.g. Fang
ting to the occupants. Hence, the amount of energy absorption et al. (2014a), Shi et al. (2013a), Shi et al. (2013b), Wang
(EA) has been drawn exhaustive attention by researchers. and Shi (2014)) preferred a short effective distance with the
Z d expectation of a lowest possible intrusion. Taking the front
EAðd Þ ¼ F ðsÞds ð2Þ side rails in frontal impact as an example, if they deform too
0 much, the passenger compartment might be intruded severely
by the engine booth. On the other hand, if the rails crash a
where F(s) is the instantaneous impact force at the crash dis-
short distance, only a little kinetic energy can be dissipated
tance s and d the total crash displacement concerned for mea-
through progressive deformation and thus high deceleration
suring the energy absorption.
could be yielded. Also some researchers (e.g., Zhang et al.
To take into account the mass efficiency, the specific ener-
(2008), Bi et al. (2010)) constrained Favg above a certain level
gy absorption (SEA) defined as the EA per unit mass has also
to maintain high energy absorption, while others (e.g., Zarei
been widely used.
and Kroger (2006), Yang and Qi (2013), Zarei and Kroger
EAðd Þ (2008a), Zarei and Kroger (2008b), Toksoy and Güden
SEAðd Þ ¼ ð3Þ
M (2011)) suggested to keep Favg at a low level to reduce the
risk of occupant injury. Horstemeyer et al. (2009) compared
where M is the mass of the structure. the two different criteria under side impact, i.e. the energy
Other criteria in relation to energy absorption capacity in- absorption of collapsed components and an injury-based met-
clude the crash load efficiency (CFE), defined as the ratio of ric (in terms of accelerations) of the dummy. They revealed
the mean crash force (Favg) to peak force Fmax; and the load that the injury-based design, which differs a lot from the
uniformity (LU), which is a reciprocal of CFE. energy-based design, could achieve a much safer structure.
F avg ðd Þ The uniform internal energy density (IED) criterion in topol-
C FE ðd Þ ¼  100% ð4Þ ogy optimization was also challenged by Witowski et al.
F max ðd Þ
(2012) because uniform IED only represents the whole struc-
F max ðd Þ ture absorbs energy uniformly, rather than the maximum
LU ðd Þ ¼  100% ð5Þ
F avg ðd Þ amount of EA in the whole structure.
EAðd Þ For industrial applications, the selection of design criteria is
F avg ðd Þ ¼ ð6Þ closely related to the loading scenario considered (please see
d
Section 5.2 for further details). As the ultimate goal of crash-
The usage ratio (UR) of the energy absorber can also be worthiness optimization is to ensure the passenger’s safety, it
employed to assess the crashworthiness, can be more judicious to consider the damage of the dummy
during crashes although its modeling is fairly complicated
d
U Rðd Þ ¼  100% ð7Þ (Horstemeyer et al. (2009)). When it comes to the safety com-
l
ponents used in real world, more realistic loading conditions
where l is the total length of the structure. The effective crash and design criteria should be considered altogether to realize
distance deff is calculated as the deformation at which their required functionality in a context of the entire vehicle
CFE(d) × UR(d) has the maximum value (Hanssen et al. system. For example, design criteria under more realistic
(2000)). Then, the other criteria can be calculated during the loads, such as lateral bending (Kim et al. (2002), Xiang et al.
displacement from 0 to deff. (2006), Zarei and Kroger (2008b), Zhang et al. (2009), Fang
In order to maximize the material usage, topology optimi- et al. (2014b) ) and oblique loads (Reid and Reddy (1986),
zation often seeks internal energy density (IED) of each Zarei and Kroger (2008a), Qi et al. (2012), Tarlochan et al.
1094 J. Fang et al.

(2013), Zhang et al. (2014b)) should also be taken into ac- To overcome this difficulty, gradient-free or namely zeroth-
count for crashworthiness design of crashing components. In order methods, which do not need gradient information, seem
addition, from the practical point of view, it is recommended to be more suitable. Of them, population-based algorithms
that the roles of each safety component/system should be an- such as Monte Carlo simulation (MCS) methods, simulated
alyzed based on the whole vehicle safety (Zhu et al. (2016)). annealing (SA) and genetic algorithms (GA) could be a useful
As such, the design criteria can be integrated to achieve the choice. The advantage of gradient-free methods is that they
best performance of the vehicular structure as a whole. may converge to the global optimum. However, computation-
al cost could be prohibitive as they often require a large num-
ber of function evaluations before convergence, and there is
3 Formulations of crashworthiness criteria no universally good criterion to determine the convergence. A
in optimization practical compromise is to limit the number of nonlinear FEA
runs by the predefined population size and number of
3.1 Analytical functions generations, which may however make it difficult to yield a
global optimum.
It is indispensable to formulate the crashworthiness perfor- Nevertheless, in an industrial context, the mathematical
mance quantitatively for optimization. The first approach is optimum may be of less practical interest while an efficient
perhaps to develop proper analytical models of these perfor- improvement is intended. In this regard, Rzesnitzek et al.
mance criteria with respect to design variables (Chen (2001), (2002) proposed a two-stage optimization method for crash-
Kim (2002), Hanssen et al. (2001), Kim et al. (2002)). For worthiness problems, in which the first stage conducts a sto-
example, Chen (2001) derived the closed-form theoretical ex- chastic optimization using MCS for a large number of design
pressions of EA during a bending collapse and combined variables to acquire an optimal solution away from the initial
compression/bending deformation for a thin-walled beam design and identify a small number of significant design
with a ultralight filler. Kim (2002) derived the analytical func- variables. Redhe et al. (2004) pointed out that the stochastic
tion of SEA and then incorporated it into the optimization of a optimization should not be used for problems with less than
multi-cell tube. Hanssen et al. (2001) used several formulas in 10-15 design variables; and the more design variables the
terms of foam density, wall thickness, column width, wall problem has, the more efficient the stochastic optimization
material strength and total component length to optimize the is. In the multidisciplinary optimization of car bodies,
square column with foam filler. Note that analytical functions Duddeck (2008) recommended to couple population-based
can only be applied to such tubes with simple geometries (e.g., algorithms with FEA for the design of the frontal impact prob-
square and circular tubes) subject to strong mechanical as- lem, which has a highly non-regular crash responses due to
sumptions, and was more extensively used at earlier days high nonlinearity and bifurcations. More recently, Xu et al.
when the computational resources were not affordable to ac- (2014) used the benchmark problems to conduct direct cou-
complish the crashing simulation for design optimization. pling based optimization. From their study, when sufficient
computational resources are available, direct coupling method
3.2 Direct coupling with finite element analysis could be promising in terms of the performance and feasibility
of optimization. Xu et al. (2015) developed a data mining-
With the increasing computational capacity, numerical based strategy to improve the efficiency of population-based
methods, represented by nonlinear finite element analysis algorithms. The historic information was utilized to identify
(FEA), have proven effective to predict the crashworthiness. and eliminate low-quality and repetitive designs based on
The question is if it is possible to directly couple FEA with clustering analysis. The algorithm considers both the explor-
optimization algorithms for crashworthiness problems. Or, ative search at the earlier stage and exploitative search at the
whether is it realistic to iteratively call FEA evaluations in later stage. The authors also recommended using larger pop-
the optimization process. Mathematical programming-based ulation and fewer generations to take advantages of parallel
structural optimization often requires gradient information of computing.
the objectives and constraints to determine a searching Besides, the equivalent static load (ESL) method (Choi and
direction towards an optimal solution. In a very early work, Park (1999), (Kang et al. (2001), Shin et al. (2007), Park
Yang et al. (1994) showed that it is feasible but fairly costly to (2011)) offers a simplified and approximate approach to the
conduct the optimization for crash simulations, where the de- evaluation of crashworthiness performance. The applications
sign sensitivities were calculated by using the forward finite can be found in Jeong et al. (2008), Jeong et al. (2010), Yi
difference method with 1 % step size for each design variable. et al. (2012) and Lee et al. (2015). The major advantage of
For highly nonlinear problems, the simulated responses often ESL is that a nonlinear crashworthiness problem is converted
contain numerical noise, making it difficult to calculate gradi- to a linear static problem, which is much cheaper and more
ents accurately (Zabaras et al. (2003)). stable, computationally. However, its limitations are also
On design optimization for structural crashworthiness 1095

obvious due to the intrinsic nature of simplifications and as- Considering the fact that the number of unknown coeffi-
sumptions (Witowski et al. (2012)). cients is ns = (n + 1) × (n + 2)/2, it is recommended to generate
more than twice more samples than ns to prevent overfitting of
coefficients, which implies the number of design variables in
3.3 Surrogate modeling
PRS model can be critical for determining the computational
cost. To address this issue, stepwise regression (Draper and
In crashworthiness optimization, direct coupling method
Smith (1981), Yang et al. (2000), Gu et al. (2001)) can be
may be inefficient (if not impossible) since iterative non-
implemented to screen the terms in PRS that have relatively
linear FEA during optimization usually require enormous
little contribution to the design criteria.
computational efforts and take the high risk of premature
simulation failure prior to a proper convergence. As a
result, surrogate models (or metamodels) are more often 3.3.2 Radial basis function (RBF) model
used as an alternative for formulating the design criteria in
terms of an explicit function of design variables in ad- Radial basis function model was developed for scattered mul-
vance of optimization, which has proven an effective tivariate data interpolation by using a series of basis functions
and sometimes a unique approach (Wang and Shan that are symmetric and centered at each sampling point. Radial
(2007), Forrester et al. (2008)). The idea of surrogate basis functions are typically formulated as (Hardy (1971)):
modeling is to construct an approximate function based
X
m
X
ns
upon a series of sampling evaluations, in which design ^yðxÞ ¼ c j p j ð xÞ þ λi φðrðx; xi ÞÞ ð9Þ
space is typically sampled using the design of experiment j¼1 i¼1
(DoE) methods (Fig. 3). Then, the FEA is performed at
these sample points to establish surrogate models with a where m is the number of the polynomial terms, cj is the
certain confidence of approximation for crashworthiness coefficient for polynomial basis function pj(x), and ns is the
optimization. number of sample points. λi is the weighted coefficient for the
term for the i-th variable, r(x, xi) is the Euclidean distance
expressed in terms of ‖x ‐ xi‖. φ(r) is the radial basis function.
3.3.1 Polynomial response surface (PRS) model

In many cases, the polynomial basis functions are found sim- 3.3.3 Kriging (KRG) model
ple yet effective to establish a surrogate model. For example, a
quadratic PRS model can be expressed as (Montgomery The Kriging model (Sacks et al. (1989)) was originally devel-
(1996)), oped for mining and geostatistical applications involving spa-
tially and temporally correlated data. The KRG model is com-
X
n
X
n
X
n−1
X
n posed of a global model f(x) and a local departure Z(x):
^yðxÞ ¼ b0 þ b i xi þ bii x2i þ bi j x i x j ð8Þ
i¼1 i¼1 i¼1 j>i y ð xÞ ¼ f ð xÞ þ Z ð xÞ ð10Þ

where b0, bi, bii and bij are the unknown coefficients, xi is the i- where y(x) is the unknown function of interest, f(x)
th design variable, and n is the total number of design vari- models the global trend of the function of interest, and
ables. ŷ(x) is the approximation to the actual value y(x) from Z(x) models the correlation between the points by a sto-
FEA. chastic process whose mean is zero and variance is σ2.

Fig. 3 Surrogate modeling: (a) x2

Design of experiment, (b)
Construction of surrogate models
P2 ŷ
P1 P2
P1 yˆ =f ( x1 , x2 )

x2
x1

x1
(a) (b)
1096 J. Fang et al.

Z(x) provides local deviations, and the covariance be- was able to produce a satisfactory approximation to EA, while
tween different points is modeled as: the RBF models performed better to approximate amax. Also,
      the RBF models can yield more accurate optimization results.
Cov Z ðxi Þ; Z x j ¼ σ2 R R xi ; x j ð11Þ Forsberg and Nilsson (2006) compared the linear PRS and
KRG with the same updating scheme in the region of
where R is a correlation matrix defined by correlation
interest. KRG was found to enable to improve the sequential
function Rðxi ; x j Þ as follows:
behavior of the optimization algorithm at earlier iterations of
" n #
  X  2

the optimization process. However, KRG could be
R xi ; x j ¼ exp − θk xki −xkj  ð12Þ problematic if a constraint was violated after several
k¼1 iterations and linear PRS seemed more easily to find a
where θk is the unknown correlation parameter used to fit feasible solution. Shi et al. (2012) proposed to select the best
the model, and xki and xkj are the k-th components of sam- surrogate model using a Bayesian metric under data uncertain-
ple points xi and xj, respectively. ty, thereby determining a proper sample size for large scale
real-life problems. It can be concluded that the selection of a
surrogate model is largely case dependent. In other words, no
3.3.4 Artificial neural network (ANN) model unique surrogate model is able to produce the most accurate
result for all cases (Yang et al. (2005)). Furthermore, the most
Artificial neural network is a powerful method to formulate accurate surrogate model may not necessarily provide the
the relationship between a set of input variables and output most promising optimum (Song et al. (2013)). Therefore, the
results in complex systems (Zurada (1992)). It is composed of concurrent use of multiple surrogate models are recommended
some parallel numerical computing units, namely neural ele- by Song et al. (2013) to seek for a better optimum since the
ments. The neural elements are linked according to specific time of constructing surrogate models is negligible compared
topological network functions. Thus, a neural model will be to that of acquiring DoE data. Another practice of using mul-
defined by using connection weights and biases parameters. tiple surrogates is to construct ensembles of surrogate models
These parameters are trained from the training sample data set as follows.
by using specific optimization algorithm, which adjusts the
values of the weights between elements. The training data 3.3.6 Ensemble of surrogates
consists of pairs of design variables and output responses.
Other types of surrogate models used in crashworthiness Typically, obtaining data required for developing surrogate is
optimization include support vector regression (SVR) (Smola computationally expensive, and the use of an ensemble was
and Schölkopf (2004)) and multivariate adaptive regression first introduced by Bishop (1995) to take full advantage of all
splines (MARS) (Friedman (1991)) etc. the individual surrogates to extract as much information as
possible with a relatively low computational cost. Using the
3.3.5 Comparison of different surrogate models weighted-sum formulation, the ensemble of surrogates can be
expressed as:
Regarding the selection of surrogate models, researchers have
provided some general guidelines, which may also be useful X
N
^yEns ðxÞ ¼ ωi ðxÞ^yi ðxÞ ð13Þ
in crashworthiness optimization. Simpson et al. (2001) point-
i¼1
ed out that PRS works well in the problems with < 10 variables
and the problems with random errors. They also claimed that where ŷEns(x) denotes the predicted response by the ensemble
ANN should be good for very large design problems (~10,000 of surrogates ŷi(x), N is the number of the individual surrogate
variables) while KRG is able to handle the problems with < 50 in the ensemble, ŷi(x) and ωi(x) are the surrogate response and
variables. Jin et al. (2001) compared surrogate models under the corresponding weight factor of the i-th surrogate models,
multiple modeling criteria. RBF was found most insensitive to respectively.
DoE sample size in most situations in terms of accuracy and To determine the weight factors for surrogate models, dif-
robustness, while KRG is very sensitive to the noise because it ferent strategies have been developed. Zerpa et al. (2005) set
interpolates the sample data. the values of the weight factors for each surrogate model to be
There have been some comparative studies of surrogate inversely proportional to the estimate of the prediction vari-
models in relation to crashworthiness problem in literature ance. Goel et al. (2007) proposed a heuristic weight scheme
(e.g., Fang et al. (2005), Forsberg and Nilsson (2006), Zhu based on the generalized mean square cross-validation error
et al. (2009) , Shi et al. (2012)). For instance, Fang et al. (GMSE). Viana et al. (2009b) proposed to select the weight
(2005) compared quadratic PRS and RBF for fitting factors following an approach based on minimizing the mean
nonlinear responses in a frontal collision and found that PRS square error. Acar and Rais-Rohani (2009) proposed an
On design optimization for structural crashworthiness 1097

optimal weighted surrogate approach to determining the On the other hand, exploring design space is a strategy to
weight factors via optimization. The difference between the increase the global accuracy of a surrogate model. It is
last two is that Acar and Rais-Rohani’s approach obtains the straightforward to add sequential samples to the sparse regions
weights through an optimization process, while Viana’s ap- of design space. If error estimates are available for the surro-
proach obtains the weights through an analysis expression gate model, those points with large errors can be a candidate
(Zhou et al. (2011)). Note that while the determination of for increasing the accuracy of the surrogate. For example,
proper weight factors associated with individual surrogates Chen et al. (2014) and Sun et al. (2014a) used the maximum
can be based upon global and/or local measures (Acar mean squared error of KRG model to determine new sampling
(2010b)), the above-mentioned strategies used the global points in the framework of sequential optimization.
method (i.e., ω i (x) = ω i ). The more sophisticated local Considering both exploitation and exploration, Efficient
methods, in which ωi(x) varies over the design space, were Global Optimization (EGO) (Schonlau (1998)) has been pro-
also proposed by Acar (2010b). posed to add new sampling points iteratively which contribute
When coming to crashworthiness optimization, ensembles toward global optimization. The EGO algorithm uses KRG
of surrogates have also shown their advantages over individ- models because they provide not only the surrogate prediction
ual surrogates (Acar (2010b); Acar and Solanki (2009a); but also error estimates. The expected improvement (EI) is
Hamza and Saitou (2012); Pan and Zhu (2011a); Yin et al. maximized to find the sequential sampling points at each iter-
(2014a)). For example, Acar and Solanki (2009a) examined ation as (Schonlau (1998)),
ensembles of surrogates in two cases of offset frontal and side 8 ! !
>
>   f −^yðxÞ f min −^yðxÞ
impacts and found that for all crash responses of interest the < f þ ^sðxÞϕ if ^sðxÞ > 0
min −^
yðxÞ Φ min
EI ðxÞ ¼ ^sðxÞ ^sðxÞ
ensemble of surrogates outperformed all individual >
>
:
surrogates. Pan and Zhu (2011a) demonstrated that forming 0 if ^sðxÞ ¼ 0

an ensemble of surrogates could help avoid a misleading op- ð14Þ

timum in a design optimization of vehicle roof structures.
The ideal scenario for using ensembles of surrogates would where ϕ(⋅) and Φ(⋅) denote the probability density and the
be that individual surrogates have different prediction values cumulative distribution functions of the standard normal dis-
but similar overall prediction accuracies on the entire design tribution. ŝ(x) is the KRG prediction error, which is also called
domain, so that prediction errors cancel out when aggregation as the mean squared error. If ŝ(x) > 0, the first term in . (14) is
of the prediction is performed. That is to say, one should the difference between the current minimum and the predicted
ensure the accuracy and diversity of individual surrogates to value multiplied by the probability that y(x) is smaller than the
better make advantage of surrogate ensembles. Unfortunately, current best fmin. Therefore, the first term becomes large when
surrogates with a comparable accuracy were found often high- ŷ(x) is likely smaller than fmin. The second term is the standard
ly correlated (Viana et al. (2009a)). One should also keep in deviation of y(x) multiplied by the probability density at y-
mind that when applying ensembles of surrogates to crash- (x) = fmin. This term becomes great when there is a high un-
worthiness optimization, the global accuracy is less interesting certainty of the prediction (which probably appears far away
than the ability to lead to the global optimum (Viana et al. from the existing samples as the KRG model goes exactly
(2014)). through samples). To take into account the constraints, . (14)
can be multiplied by the probabilities P(g(x) ≤ 0) to obtain the
constrained version of EI (EIc) so that each constraint is met.
3.3.7 Efficient global optimization EI c ðxÞ ¼ EI ðxÞPðgðxÞ ≤ 0Þ ð15Þ

When performing a surrogate-based optimization, a basic as- Schonlau (1998) proposed to generate sequential sam-
sumption is that the surrogate model is sufficiently accurate pling scheme by maximizing EI to yield the sequential
and all we need to do is to find the optimum design using the points. This criterion balances the exploration of design
established surrogate model (Forrester and Keane (2009)). space and exploitation of the local region around an opti-
However, the surrogate model constructed using initial sam- mum, which has been applied to the crashworthiness de-
ples will probably not be accurate in the local region of the sign of a cylindrical tube by Lee et al. (2002). The EI
final optimum. It is common to exploit this local region by criterion assumes that KRG model parameters were esti-
sequentially positioning additional samples inside. These infill mated accurately based on the existing sample data.
points are then used to update the surrogate model until the Otherwise, the iteration process could converge very
optimum converges to the final location properly, which slowly or even not at all (Forrester and Keane (2009)).
seems to be attractive to more accurately locate a local opti- Besides, since a KRG model often underestimates the un-
mum rather than the true global optimum (Forrester and certainty, extra care should be taken to avoid the prema-
Keane (2009)). ture termination by setting an overly stringent threshold.
1098 J. Fang et al.

Another extension of EGO is the multiobjective versions of

EI, which were established recently (e.g., Keane (2006),
Shimoyama et al. (2013), Couckuyt et al. (2014)). Unlike
the definition of single objective EI in (14) and (15),
multiobjective EI could become cumbersome as the dimen-
sion of the objective function increases.
Traditionally, EGO adds one single point to the sample set
at a time, which may not be thought efficient when computa-
tional resources are available for parallel computing and the
main concern is the wall-clock time (rather than the number of
simulations). To take advantage of parallel computing capa-
bilities, Schonlau (1998) sought the global maximum of the EI
criterion and then temporarily added the KRG predicted value
at this point to sample data (assuming that the model is correct
at this location). The KRG model is then constructed with the Fig. 4 Updating process of RoI
sample data and the maximization of EI criterion is continued
to seek other new samples until a predefined number of new
samples has been obtained. Sobester et al. (2004) proposed to iteration. The fraction parameter λi for the i-th design variable
locate a number of maxima of EI (which is usually extremely is calculated based upon the distance between the optimum
multimodal) using either a gradient-based optimization algo- and the center of the current RoI:
rithm with multiple restarts or a genetic algorithm with    
 ðk Þ* ðk Þ ðk Þ 
clustering and sharing. Then, those locations of the maxima  xi − xil þ xiu =2
ðkþ1Þ
λi ¼ η þ ðγ−ηÞ   ð16Þ
can be evaluated in parallel and the process is repeated until ðk Þ ðk Þ
xiu −xil =2
convergence. Viana et al. (2013) proposed a multiple surro-
gate efficient global optimization algorithm to add multiple
T h e m a x i m u m v a l u e o f λ (i k + 1 ) ( λ ( k +
samples per optimization cycle, in which uncertainty esti- 1)
= max λ(k
i
+ 1)
(i = 1,.., n)) is applied to all design variables
mates of other models was imported from the KRG model.
during the iterations. Then, the lower and upper bounds of
However, these parallel versions of EGO have not been sub-
the i-th design variable of (k + 1)-th subregion can be deter-
stantially validated for their effectiveness in crashworthiness
mined by:
optimization. Hamza and Shalaby (2014) used three infill
8 n   o
criteria to generate new multiple samples, and their < xðkþ1Þ ¼ max xði k Þ* −λðkþ1Þ xðiuk Þ −xðk Þ =2; xð0Þ
il il
algorithm was successfully applied to a crashworthiness n   il o ð17Þ
problem after being tested on four benchmark mathematical : xðkþ1Þ ¼ min xðk Þ* þ λðkþ1Þ xðk Þ −xðk Þ =2; xð0Þ
iu i iu il iu
functions. Most recently, Haftka et al. (2016) conducted a
comprehensive survey on parallel surrogate-assisted global where x(0) (0)
il and xiu are the lower and upper bounds of the entire
optimization and the interested readers are strongly recom- design space.
mended to acquire more insightful information from this While SSM has been demonstrated to be able to iden-
article. tify the optimum region for various crashworthiness prob-
lems (Kurtaran et al. (2002), Craig et al. (2005), Liang
3.3.8 Successive surrogate modeling (SSM) and Le (2009), Liu et al. (2014)), iterative resampling in
SSM might be prohibited in practice as crashworthiness
The general idea of the EGO is to add sequential sampling simulations are rather expensive computationally.
points iteratively at the regions of interest so that the accuracy Implementation of inherited Latin hypercube design
of surrogate models is improved locally and globally. The (Wang (2003)), which is a technique to inherit previous
other option for sequential sampling is to use successive re- sample points, might help reduce the required number of
sponse surface method (SRSM or successive surrogate model- sample points in subsequent iterations. The other limita-
ing, SSM) (Kurtaran et al. (2002)), in which the region of tion might be that the continuity between subsequent ap-
interest (RoI) is gradually shrunk to a smaller area around proximations is not well guaranteed and the information
the optimum by panning and zooming within the design space obtained at the previous iterations is difficult to be taken
(original RoI) during the iterations (Fig. 4). into account (Naceur et al. (2006)). With recent develop-
In the successive surrogate modeling method, the center of ments in high-performance computing (HPC), parallel
RoI at the (k + 1)-th iteration is the optimum x(k)* of the k-th computing has become a trend in optimization. Thanks
iteration, and its size of RoI is a fraction of the size of the k-th to its parallel nature within each iteration, the speed and
On design optimization for structural crashworthiness 1099

efficiencies of SSM can be realized by using multiple Secondly, one can conduct the multiobjective optimi-
resources (Sheldon et al. (2011)). Recently, Stander zation using population-based algorithms directly without
(Stander (2012), Stander (2013)) extended SSM to solve formulating a combined cost function, of which
multiobjective optimization problems by introducing a multiobjective particle swarm optimization (MOPSO)
Pareto Domain Reduction (PDR) technique. Irregular (Raquel and Naval (2005)) and non-dominated sorting
sub-regions of the Pareto optimal front were used as sam- genetic algorithm II (NSGA-II) (Deb et al. (2002)) are
pling domains, which shrunk iteratively to explore the two popular algorithms frequently used in crashworthi-
neighborhood of the Pareto optimal front. This technique ness problems. Table 4 in Appendix summarizes the pre-
was demonstrated to have a comparable accuracy to the vious works on crashworthiness optimization with multi-
direct genetic algorithm while using a much smaller num- ple objectives and Fig. 5 displays the percentage of the
ber of simulations. different optimization methods used in the literature,
For readers’ reference, Table 3 in Appendix summarizes where MOGA stands for general multi-objective genetic
the studies on surrogate modeling techniques used in the lit- algorithm and NSGA-II is a special version of MOGA. It
erature for crashworthiness optimization. is noted that more than 60 % publications have adopted
the direct evolutionary algorithms to seek non-dominated
solutions.
4 Optimization strategies Although it is of considerable limitation to generation
of preferable Pareto solutions, combined cost function
4.1 Multiobjective optimization (MOO) methods still contributed to around 40 % of
multiobjective crashworthiness optimization in the litera-
Like most real-life engineering applications, crashworthi- ture for its simplicity. Note that a linear weighted sum
ness optimization can be often characterized by a number method is impossible to obtain a proper solution in the
of design criteria. In the literature, many crashworthiness non-convex portions on the Pareto frontier. Theoretical
optimization articles have dealt with multiple design ob- reasons for this deficiency have been given by Das and
jectives. Due to conflicts between these objectives, often a Dennis (1997) and Messac et al. (2000b). If nonlinear
rational approach to such a problem is to generate a set of weighted sum method is used, this limitation might be
solutions (namely Pareto solutions) that provide accept- avoided but the form of the function to be used is difficult
able overall performance in terms of all these objectives to decide consistently. In addition, in the linear weighted
rather than a single one. These solutions are compared sum method, varying the weighting factor from 0 to 1
using the non-dominated approach which does not intro- homogeneously cannot guarantee an even distribution of
duce preference on any of objective functions in prior. In Pareto points. Das and Dennis (1997) illustrated the nec-
this approach, solution x(1) dominates solution x(2) if: (1) essary conditions for a series of weighted sum interactions
x(1) is feasible and x(2) is not, or both of them are infea- to create an even spread of points on the Pareto curve in
sible but x(1) is closer to the feasible boundary; or (2) the objective space.
feasible solution x(1) is not worse than feasible solution In crashworthiness optimization, lightweighting of
x(2) in all the objectives and x(1) is strictly better than x(2) structure and its crashing performance are frequently con-
in at least one objective (Coello et al. (2004)). Otherwise, flictive, and design criteria (e.g. SEA and Fmax,) could
none of the solutions dominates the other and they are also strongly compete with each other during optimization
both non-dominated. (e.g., Khakhali et al. (2010), Fang et al. (2014b)). Under
Currently, there appear to be two popular ways of deal- this circumstance, generating a complete representation of
ing with multiobjective optimization in crashworthiness non-dominated Pareto solutions in objective space is
problems. Firstly, one can formulate a combined cost meaningful and could provide insightful information for
function F(x) to indirectly represent the contributions of decision-making.
multiple objectives through a single function; and then
performs a single objective optimization (Forrester and
Keane (2009)). In this approach, the linearly weighted MOPSO
Combined
sum technique is the most commonly-used formulation 28%
cost
function
in crashworthiness optimization, as: 39%

Xk Xk MOGA
F ðxÞ ¼ i¼1
wi f i ðxÞ; 0 ≤ wi ≤ 1; i¼1
wi ¼ 1 ð18Þ 33%

where wi is the weighting factor (to emphasize the relative Fig. 5 Percentage of MOO methods used for crashworthiness
importance) of the i-th objective function fi(x). optimization in literature
1100 J. Fang et al.

4.2 Optimization under uncertainties hand, the reliable optimum xre moves away to create a gap
from the boundary of the constraint so that it can still be within
4.2.1 Definition of optimization under uncertainties the feasible region when uncertainties are present.
Moreover, conventional design likely leads to a large scat-
Most (if not all) real-life engineering problems involve some ter of optimal performance due to uncertainties, which may
degree of uncertainties in loading conditions, material proper- not only cause significant fluctuations from the desired per-
ties, geometries, manufacturing tolerances and actual usage, formance, but also increase life-cycle costs, including inspec-
etc. It must be pointed out that usually a deterministic optimi- tion, repair and other maintenance expenses (Fang et al.
zation tends to push a design toward one or more constraints (2015a)). Thus, the concept of robust design optimization
until the constraints become active, thereby leaving no room (RDO) is to reduce the scatter of the structural performance
for accommodating various uncertainties. Therefore, without eliminating the source of uncertain variability. This
reliability-based optimization (RBO), which aims to seek a approach has drawn increasing attention for solving real-
reliable optimum by converting the deterministic constraints world problems recently (e.g. Park et al. (2006), Beyer and
into probabilistic counterparts representing that probability of Sendhoff (2007), Yao et al. (2011)). As shown in Fig. 6b, let
design infeasibility is restricted to a pre-specified level, has the x-axis represent the uncertain parameter, e.g., random de-
been widely applied to engineering problems. As shown in sign variable or noise factor, and the vertical axis represents
Fig. 6a, let xd represent the deterministic optimum and xre the value of an objective function f(x) to be minimized. Of
represent the reliable optimum in the design space (x1-x2 these two optimal solutions xd and xro as pointed, x2 is con-
space), which is divided into infeasible and feasible regions sidered more robust as a variation of ± Δx in the design vari-
by the constraints. Since the deterministic optimum xd is lo- able and/or noise factor does not alter the objective function
cated on the boundary of the constraint, it may fall to the too much (Δfro < <Δfd). On the contrary, xd appears highly
infeasible region when uncertainties are present. On the other sensitive to the parametric perturbation and often cannot be
recommended as a design in practice, even though it has a
better nominal value than xro. It is noted that a robust-based
optimization places more emphasis on the stability of the ob-
jective, while a reliability-based optimization pays more atten-
tion to the feasibility of the constraint.
To accommodate uncertainties, reliability-based optimiza-
tion (RBO) has been adopted in crashworthiness problems. A
general RBO problem can be expressed mathematically as:
8
< min f ðxÞ
s:t: PðgðxÞ ≤ 0Þ ≥Rt ð19Þ
:
xL ≤ x ≤ x U

(a)
where Rt denotes the reliability level and P(⋅) stands for the
probability function of satisfying the constraints (g(x) ≤ 0).
It is commonly acknowledged that a robust design was
firstly proposed by Japanese engineer Genichi Taguchi,
named as the Taguchi method to improve the quality of
manufactured goods and makes the product performance less
sensitive to variations of variables beyond the control of de-
signers. A general robust design optimization (RDO) problem
can be formulated mathematically as:

8  
< min F μ f ðxÞ; σ f ðxÞ
s:t: gðxÞ ≤0 ð20Þ
:
x L ≤ x ≤ xU
(b)
Fig. 6 Illustrations of design optimization with uncertainties (a)
Reliability-based optimization (RBO) and (b) Robust design where μf(x) and σf(x) are the vectors of mean and standard
optimization (RDO) deviation of the objectives, respectively.
On design optimization for structural crashworthiness 1101

To enhance a design in terms of both reliability and robust- of the probability distribution of input random variables.
ness, RBO and RDO can be combined and referred to as Thus, MCS is also referred to as sampling-based method
reliability-based robust design optimization (RBRDO), which (Helton et al. (2006)). MCS is also a conventional method
can be formulated as: of quantifying robustness (Yao et al. (2011)), allowing
8   determining the means and standard deviations of objec-
< min F μ f ðxÞ; σ f ðxÞ tives in . (19)- (21).
s:t: PðgðxÞ≤ 0 Þ ≥ Rt ð21Þ 8
: XQ
x ≤x≤x
L U
>
>
>
> f ð xi Þ
>
< μ f ð xÞ ≅ i¼1
Q
In the literature, design for six sigma (DFSS) can be ð25Þ
>
> X Q  2
regarded as a special case of RBRDO. Six sigma is a >
> 1
: σ f ðxÞ≅ Q−1 f ðxi Þ−μ f ðxÞ
2
>
quality philosophy involving the statistical tools within i¼1
a structured methodology for gaining the knowledge
needed to achieve better, faster and less expensive prod- If xi is independent, the laws of large numbers allow us to
ucts than competitors (Breyfogle III (2003)). The term achieve any degree of accuracy by increasing Q. The accuracy
sigma refers to as standard deviation (σ), measuring the of MCS estimation can be quantified with the standard error
dispersion of a set of data around the mean value (μ) of defined as:
the data.
While the six sigma approach aims to reduce the num- σ f ð xÞ
err ¼ pffiffiffiffi ð26Þ
ber of defects, DFSS offers a powerful tool to optimize Q
the products in a cost-effective and simple fashion to meet
the customer’s requirements (Antony (2002)). The two The error is, therefore, unrelated to the problem dimension
goals in DFSS are: (1) striving to maintain performance (i.e., the number of design variables), which is very appealing
for large-scale problems. And the error is proportional to
within acceptable limits consistently (reliability); and (2) pffiffiffiffi
striving to reduce performance variation and thus increase 1= Q, implying that the improvement of accuracy by one
robustness. With this concept, (21) can be revised as order of magnitude will require 100 times more samples.
8 Such computational cost can be prohibitive in the application
< min μ f ðxÞ þ 6σ f ðxÞ for complex and highly nonlinear problems such as crashwor-
s:t: μg ðxÞ þ 6σg ðxÞ ≤ 0 ð22Þ thiness analysis.
:
xL ≤ x ≤ x U On the other hand, the minimum sampling size required for
the desired reliability level P(g(x) ≤ 0 ), as suggested by Tu
et al. (1999), is:
4.2.2 Methods of uncertainty analysis in optimization
10
Q¼ ð27Þ
Monte Carlo simulation The problems defined in 1‐P½g ðxÞ≤ 0
(19)–(22) involve a procedure to obtain the values of
which indicates that for a 10 % estimated probability of fail-
probabilistic objectives and constraints. One of the effec-
ure; about 100 function evaluations (e.g., nonlinear FEA runs
tive yet simple approaches could be Monte Carlo simula-
in crashworthiness analysis) are required with some confi-
tion (MCS). Using a large number of samples, Monte
dence on the first digit of failure prediction. To verify an event
Carlo simulation allows the estimation to the probability
having a 1 % failure probability; about a 1000 structural anal-
of feasibility as follows,
yses are required, which would be usually considered too
expensive and some alternatives may be needed.
1X
Q

P ð g ð xÞ ≤ 0 Þ ¼ I ð xÞ ð23Þ To apply MCS to crashworthiness optimization, the use of

Q i¼1 surrogate models has been advocated by many researchers
(e.g., Acar and Solanki (2009b), Fang et al. (2014a), Gu
et al. (2001), Gu et al. (2013), Khakhali et al. (2010), Koch
where Q is the total number of samples and I(x) is an indicator et al. (2004), Lönn et al. (2011), Shi et al. (2013b)). After
function defined as validation, the surrogate models can be used to evaluate the
function values with a very large number (e.g., up to millions)
1 if gðxÞ ≤ 0
I ð xÞ ¼ ð24Þ of times around each design point at a relatively low compu-
0 if g ðxÞ > 0
tational cost.
Note that in (23), Q independent sets of design vari- In conventional design optimization, the accuracy of a sur-
ables are obtained from sampling techniques on the basis rogate model is a major concern because the purpose of a
1102 J. Fang et al.

surrogate model is to find the optimum function value. In where σ2xi represents the variances of the i-th variable xi and x
reliability analysis, however, the region where the limit state denotes the mean of the variable vector. To analytically eval-
changes its sign, as in (23), is more important than the value of uate the statistical information using (28) and (29), function
the function itself. Therefore, it is more important to identify f(x) should be known. However, such a function f(x) may not
the feasible region that satisfies the constraint conditions than probably be easy to derive for various crashworthiness
accuracy in predicting function values. Ramu et al. (2007) criteria. Again, surrogate models can serve as an alternative
explored the situation when the function is insensitive to input to approximating f(x) (e.g. Sun et al. (2014a), Sinha (2007),
variables, the error in reliability tends to be amplified due to a Chen et al. (1996)). This could lead to significant improve-
relatively large error in identifying the failure (infeasible) ment in computational efficiency. However, it might be prob-
region. lematic as the combination of these two kinds of approxima-
When surrogate models are used for reliability analysis, tion methods (i.e., Taylor’s expansions and surrogate models)
random variables are selected as input. For design optimiza- could cause inaccurate results. Alternatively, derivatives of
tion, on the other hand, design variables are selected as input responses with respect to random variables can be solved by
to the surrogate models. When the mean of a random variable numerical methods (i.e. finite difference methods). Note that
is used as a design variable, it is possible to use the surrogate (n + 1) and (n + 1)(n + 2)/2 analyses are needed for the first and
model for both reliability analysis and design optimization. In second order Taylor’s expansions, respectively. Therefore,
many cases, however, the design variables are different from Taylor’s series approximations will become more expensive
random variables. In such a case, it is necessary to extend the with increasing n, but can still be more efficient than MCS
dimension of surrogate models to include both design and (Koch et al. (2004)). Other drawbacks of AMA can be found
random variables (Qu and Haftka (2004)). in Youn and Choi (2004b).
The main issue in surrogate models is the accuracy. Once
an accurate surrogate model is available, any method can be
Dual surrogate model (DSM) Following the work by
applied to calculate reliability or to perform RBDO because
Vining and Myers (1990), dual surrogate models (DSM)
the function evaluation using the surrogate model is extremely
have been used in crashworthiness (e.g., Sun et al. (2011),
cheap. Therefore, the conventional MCS can be used to cal-
Lönn et al. (2010)) in which two surrogate models are
culate the reliability. However, due to sampling uncertainty,
created, one for the mean and the other for the variance
using the conventional MCS can cause some difficulties in
or standard deviation of a response. Two types of vari-
calculating sensitivity information during RBDO. Lee et al.
ables are considered in such a system: namely design
(2011) proposed a method of calculating sensitivity using
(controllable) variables and noise (uncontrollable) vari-
score functions especially when input variables are correlated.
ables (Jin et al. (2003)). For constructing DSMs, a cross
product array needs to be generated, where design vari-
ables are arranged in the inner array while noise variables
Approximate moment approach (AMA) When a function is
in the outer array. In each set of design variables, the
linear and input variables are normally distributed, the function
simulation is repeated several times to capture the mean
is also normally distributed. In such a case, it is much easier to
and standard deviation of responses. Then, they are
calculate the reliability. When a function is mildly nonlinear, it is
approximated as the functions of the design variables
possible to approximate the function as a linear or a quadratic
using surrogate modeling for the optimization. One issue
function. Taylor series methods can be implemented to approx-
of this approach is how to reasonably generate the outer
imate statistical moments of system output. The statistical ap-
array to capture accurate statistical information in an
proximations of f(x) using the first-order and second-order
efficient manner. Besides, only noise variables can be
Taylor’s expansions are expressed in (28) and (29), respectively:
uncertain and design variables are assumed to be
8 deterministic, which could limit spectrum of applications
>
> μ ðxÞ ≅ f ðxÞ
< f x whose uncertainties in design variables may not be
n 
X ∂f ðxÞ ð28Þ neglected. For this purpose, Aspenberg et al. (2012) pro-
>
: σ f ðxÞ≅ σ2
2
>
i¼1
∂xi x xi posed a method to use global and local surrogate models
for constructing DSMs, in which uncertainties in both
design variables and noise variables can be considered.
8  
1X
n
>
> ∂2 f
>
> μ f ðxÞ ≅ f ðxÞ þ σ2xi
>
< x 2 i¼1 ∂x2i
x  ð29Þ Most probable point based reliability analysis In addition
> n   n  2
>
> X ∂f ðxÞ 1X X
n
∂ f
2
to the above-mentioned quantification methods for uncer-
>
> σ 2
ð Þ≅ σ 2
þ σ2 σ 2
: f x
∂xi x xi
2 i¼1 j¼1 ∂xi ∂x j x xi x j tainty, the most probable point (MPP) based reliability
i¼1
(29) analysis has been employed to the crashworthiness
On design optimization for structural crashworthiness 1103

optimization (Youn et al. (2004), Rais-Rohani et al. (thickness, shape), material properties (Young’s modulus,
(2010), Sinha (2007)). First Order Reliability Method Poisson’s ratio, density, yield stress, tangent modulus, etc)
(FORM) (Hohenbichler and Rackwitz (1982)) is the most of crashing structures.
commonly used MPP based method in practical applica- 2) Operational uncertainties. Uncertainties present in differ-
tions, which fits a tangential hyperplane to the limit state ent operational conditions upon crashing, such as occu-
hypersurface at MPP. In RBO, the probabilistic constraint pant mass, impact speed, impact position, impact angle,
can be assessed to check whether or not the relibility and barrier, etc.
index has been achieved (greater than the target value), 3) Modeling uncertainties. These are related to mathematical
which is called reliability index approach (RIA) (Sinha and numerical modeling techniques for extracting crash-
(2007)). The RIA is often found to be associated with ing performances. For example, numerical errors in FEA
high computational cost and/or with lack of robustness and uncertainties in surrogate modeling (Zhang et al.
(sometimes simply fails to converge). That is, when the (2013a), Zhang et al. (2013b), Zhu et al. (2013) ) should
system is very safe, the reliability index approaches infin- be considered in crashworthiness design.
ity. To overcome this difficulty, an alternative perfor-
mance measure approach (PMA) was proposed (Tu et al. Note that although various approaches of uncertainty opti-
(1999); Youn et al. (2004)). In PMA, instead of finding mization have been developed and adopted in various crash-
MPP point that satisfies the limit state constraint, the val- worthiness designs, the optimum results have seldom been
ue of limit state is minimized on the points that satisfies validated experimentally in a statistical fashion. The valida-
the reliability constraint. The advantage of PMA is that it tion of uncertainty optimization in crashworthiness should be
can always find a solution, while its disadvantage is that an important topic in the future, which could lay a solid foun-
the value of the reliability is not available. dation to the widespread use in industries.
It should be noted that even with the abovementioned
methods for evaluating the probabilistic constraint, the 4.3 Topology optimization in crashworthiness
implementation of RBO could be still fairly challenging
for large-scale problems (Youn and Choi (2004a), Youn The topology optimization for structural crashworthiness
and Choi (2004b)). Moreover, for such application areas began with Mayer et al’s work (Mayer et al. (1996)).
as crashworthiness design, lack of sensitivity information They used the homogenization technique and optimality
leads to considerable difficulties for performing RBO criteria algorithm to distribute elemental material in a
(Youn and Choi (2004a), Youn and Choi (2004b)). To progressive fashion. In their work, internal energy was
tackle this problem, surrogate modeling has been integrat- accounted as the objective function subject to a mass
ed into the reliability analysis for complex problems constraint. Their method was applied to the design of a
(Youn and Choi (2004a), Youn and Choi (2004b), Youn three-dimensional automotive rear rail. Pedersen (2003a)
et al. (2004)). proposed a topology optimization method for two-
Different from the complexity of reliability analysis, some dimensional frame ground structure. The objective aimed
alternative simplification methods have also been proposed to to obtain a desired energy absorption history for a
translate the constraint with uncertainty into a quasi- crushed structure, where the plastic beam elements could
deterministic constraint so as to balance the computational undergo large rotations and translation. Analytical
cost and accuracy. One of these methods, namely worst case sensitivity was derived to avoid expensive calculation
analysis (Parkinson et al. (1993)), has been used in crashwor- of numerical gradients. However, the contact between
thiness design by Zhang et al. (2007a), Zhu et al. (2009) and elements was ignored because of the number of
Baril et al. (2011). discontinuities and numerical instabilities associated
with the highly nonlinear phenomena. The further work
4.2.3 Uncertainty based optimization for crashworthiness done by the same author can also be found in Pedersen
(2003b) and Pedersen (2004).
Table 5 in Appendix summarizes the research works on crash- Soto (2004) presented a heuristic non-gradient method-
worthiness optimization with uncertainties in the literature, ology to vary the density within the design domain for a
from which we can classify them according to the sources of prescribed distribution of plastic strains and stresses with
uncertainties as follows: a mass constraint. This methodology utilizes a density
approach with two base materials (i.e., stiff and extremely
1) Manufacturing uncertainties. The uncertainties induced soft ones), to represent a foam-like structure. Forsberg and
by manufacturing processes account for the discrepancy Nilsson (2007) devised another non-gradient technique by
between the nominal design and corresponding real prod- using thickness as the design variable with one base ma-
uct. They may include parameters such as geometry terial. However, through varying the thickness of each
1104 J. Fang et al.

element, this methodology can only handle plate or shell effective plastic strain to the limiting value so as to ensure
structures. Huang et al. (2007) used the bi-directional evo- the integrity of the structure. Bandi et al. (2013) presented
lutionary optimization (BESO) technique to design energy a new method in the HCA framework to optimize the
absorbing structures, where a discrete sensitivity of ele- crashworthy structures with controlled energy absorption.
ment was derived to address two principal design criteria, Again, the design domain was divided into two
i.e. absorbed energy per unit volume and absorbed energy subdomains for different requirements. That is to say that
ratio. the flexible subdomain close to the incident end was de-
Ortmann and Schumacher (2013) proposed a graph and vised to provide cushioning effect (lower peak force),
heuristic based topology optimization technique for the while the stiff subdomain close to the support (distal)
design of profile cross-sections of crashing structures. end was to maintain the integrity of the entire structure.
They divided the optimization problem into two different Topological optimization is perhaps one of the most
loops. In the outer loop, the topology and shape of the difficult problems being addressed in crashworthiness de-
structure are optimized based on expert knowledge; while sign to date. That is because of considerable complexity
in the inner loop the size and shape optimization takes of obtaining topological sensitivity or optimality criteria
place. This method is only capable of addressing the to- for effectively addressing the crash dynamic process in-
pological design of cross-sections (i.e., it can only solve volving material and geometric nonlinearities, contact,
the problems with a 2D design space though the structure strain rate etc. For this reason, alternative methods have
investigated can be 3D). been adopted to simplify such dynamic nonlinear problem
Based on the hybrid cellular automaton (HCA) method through equivalent static and/or linear counterpart.
(Tovar et al. (2007)), Patel et al. (2009) more recently Christensen et al. (2012) used the inertia relief method
proposed a heuristic (non-gradient) approach to address- as a practical tool for crashworthiness topology optimiza-
ing continuum-based topology optimization for structural tion of a body-in-white. Chuang and Yang (2012) pointed
crashworthiness. Similarly to a fully-stressed design, all out that the inertia relief method fails to fully support
elements in the structure were expected to contribute to crashworthiness topology optimization and special atten-
the energy absorption through plastic deformation; and tion is required for the definition of the loads in applica-
thus the optimum was achieved to obtain a uniform inter- tions. Despite its limitations as mentioned in Section 3.2,
nal energy density in the whole structure. Based on their the ESL method has also been employed for simplifying
work, commercial software LS-TaSC was developed crashworthiness topology optimization problems recently
(Roux (2011)), which allows generating optimal crash- (Kaushik and Ramani (2014)).
worthy configurations. However, this method still needs
to overcome the following limitations (Witowski et al.
(2012)): (1) While the uniform IED likely helps produce 5 Optimization of crashworthy structures
a better topology with more even distribution of energy
absorption in the material, it may unnecessarily ensure an 5.1 Configurations for energy-absorbing structures
overall maximum of energy absorption. (2) The inclusion
of constraint (e.g. the maximum displacement) is realized To achieve better crashworthiness performance, various novel
indirectly by using the mass constraint. (3) Since HCA configurations of structures have been proposed and further
accumulates material distribution in the areas with high optimized as an energy absorber during crashes, as summa-
stresses and strains, the risk of rupture in these areas rized in Table 6 of Appendix. The following sub-sections will
should be taken into account. (4) Although this method briefly outline these categories of studies.
is in heuristic nature without sensitivity information, it
takes a large number of iterations prior to convergence. 5.1.1 Thin-walled tubes
To address the third problem, Guo et al. (2011) proposed
a strain-based, dynamic multi-domain topology optimiza- Thin-walled tubes have been exhaustively investigated in
tion technique for crashworthy structures, in which the crashworthiness design by using analytical, numerical and
optimization was reformulated for dynamically dividing experimental methods. Alexander (1960) was amongst
two different subdomains in terms of the plastic strain one of the pioneers who derived a closed-form formula
limit. During the optimization, the material in low-strain for calculating average crushing force. Wierzbicki and
subdomain was distributed by driving the IED of each Abramowicz (1983), Abramowicz and Jones (1984) and
material element to the prescribed target. The material in Abramowicz and Jones (1986) also carried out experi-
high-strain subdomain was distributed to reduce the mental and theoretical studies on the axial crushing of
On design optimization for structural crashworthiness 1105

tubes subjected to static and dynamic loads in the early 5.1.4 Tailor-welded blank (TWB) and tailor-rolled blank
stage. More recently, thin-walled tubes with various geo- (TRB) structures
metric sections have been studied for crashworthiness,
such as circular, square/polygonal, conical/ tapered and To maximize the functionality of material in crashworthiness
hat etc, to seek for optimal designs (please refer to and energy absorption, substantial efforts have been devoted
Table 6). to the applications of proper tailor-welded blanks (TWB)
structures (Ahmetoglu et al. (1995), Abdullah et al. (2001),
5.1.2 Multi-cell tubes Kinsey et al. (2000)). The TWB technology consists of laser-
welded sheet metals with different thicknesses and different
In general, the number of angular elements (corners) in a materials for a single workpiece. Crashworthiness
tubal cross-section largely determines the energy absorp- optimization of TWB structures often aims to seek the best
tion and crashing behaviors (Wierzbicki and Abramowicz partition of different materials and thicknesses of each blank
(1983) and Abramowicz and Wierzbicki (1989)). It is for both lightweighting and crash behaviors. For example, Pan
therefore expected to design thin-walled tubes with mul- et al. (2010) optimized a TWB based B-pillar structure to
tiple cells and internal webs for achieving better energy- minimize the weight subject to the crashworthiness constraints
absorbing characteristics. Crashworthiness optimization of vehicular roof crush and side impact. Xu et al. (2013) dem-
has been also introduced to the design of various multi- onstrated that the multi-component TWB structure can be op-
cell tubes (Table 6). For example, Hou et al. (2008b) timized to further enhance the crashworthiness and reduce the
adopted surrogate modeling to optimize the single, dou- weight.
ble, triple and quadruple cell sectional columns, aiming to Different from TWB, the TRB technology varies the blank
maximize the SEA and minimize the peak force Fmax. thickness by a rolling process, which leads to a continuous
Zhang et al. (2008) found that for bitubal columns with thickness variation in the sheet. Chuang et al. (2008) demon-
internal ribs, an appropriate combination of the side strated the feasibility of design optimization for TRB technol-
length of the inner profile, inner and outer walls and ogy to achieve a better functional performance and reduce the
strong ribs are preferred for best energy absorption. Liu mass of a vehicle structure.
et al. (2014) pointed out that the multi-cell section with
double vertical internal stiffeners can absorb more energy 5.1.5 Composite structures
and they further optimized this novel structure for the
application to automotive front rails. One option to achieve lightweight design is to replace
heavy metallic materials with light composites. Although
5.1.3 Foam-filled structures most composite materials display little plastic characteris-
tics, properly designed composite materials could absorb
Substantial research efforts on foam-filled structures have more energy per unit mass than the conventional metals
been devoted through various experimental (e.g., Seitzberger (Ramakrishna (1997)). Lanzi et al. (2004b) optimized the
et al. (1997), Gupta and Velmurugan (1999), Santosa et al. shape of a composite cylindrical energy absorber and
(2000)), analytical (e.g., Gupta and Velmurugan (1999)) and found that the moderate eccentricity and conicity led the
numerical methods (e.g., Seitzberger et al. (1997), Santosa structures to have higher energy absorption efficiency and
et al. (2000)). These studies demonstrated that foam-filled less mass. Zarei et al. (2008) found that the optimized
structures can undergo large deformation at nearly constant composite crash box could absorb around 17 % more
load. The presence of the foam-filler materials in thin-walled energy with 26 % lower weight than the optimized alumi-
structures helps improve crashing stability and collapse num counterpart. Belingardi et al. (2013) optimized the
modes, thereby enhancing the overall crashworthiness cross-sectional shape, wall thickness and transverse cur-
(Borvik et al. (2003), Seitzberger et al. (2000), Santosa et al. vature of the E-Glass pultruded bumper and they achieved
(2000)). However, the crashworthiness performance is highly comparable energy absorption with steel and E-Glass fab-
dependent on the foam density and geometrical configurations ric bumpers but better progressive failure mode with re-
(Seitzberger et al. (1997), Reyes et al. (2004)). To address this duced peak load. In the work by Paz et al. (2014), the
issue, optimization techniques were used to select best possi- optimal GFRP honeycomb-filled tube improved the spe-
ble combination of tube geometry and foam density in both cific energy absorption by 40 % with a similar peak load
simple tubes (e.g., Yang and Qi (2013), Hou et al. (2009), Bi or a lowered the peak load (by 37 %) with similar mass
et al. (2010)) and complex structures (e.g., Kim (2001), and energy absorption capacity. Duan et al. (2014) studied
Hanssen et al. (2006) and Villa et al. (2011). the crashworthiness optimization of a tapered sinusoidal
1106 J. Fang et al.

specimen made of fiber reinforced polymers materials. deformation with combined bending and membrane defor-
Their optimal results showed that a smaller ratio of the mation often takes place near the corners of tubes (Kim
thickness to the radius of the specimen was often benefi- (2002)). Zhang et al. (2014a) extended the concept of
cial to the enhancement of specific energy absorption and transverse FGT to the square tube by placing more mate-
reduction of the peak force. rial to the corners and discovered 30–35 % increase in
The energy absorption performance of composite struc- energy absorption efficiency. To take the advantages of
tures can be tailored by controlling various material struc- FGT and multi-cell structures, Fang et al. (2015b) pro-
tures and parameters, such as fiber type, matrix configu- posed the transverse FGT multi-cell tubes; and they found
ration, fiber architecture, specimen geometry, process con- that the graded structures can generate more competent
dition and fiber volume fraction (Jacob et al. (2002)). Pareto solutions in terms of F max and SEA (Fig. 7a).
However, the existing optimization literature considered Under almost the same value of Fmax, the crashing force
only geometrical parameters as design variables and it of the FGT tube maintains at a higher level overall than
remains unknown whether or not crashworthiness optimi- that of the uniform tube seen in Fig. 7b, where the shaded
zation can be applied to the problems involving other area between these two curves is the additional energy
material and process parameters; and from this perspec- absorbed by the FGT structure. More recently, there has
tive, composite structures still have considerable room to been also research work to investigate the optimization for
be further optimized for better crashworthiness. double gradient structures (i.e., an FGF-filled tube with
FGT) (Fang et al. (2016)). The study of these functionally
graded structures should be extended to engineering
5.1.6 Functionally graded structures applications.

To tailor the crashworthiness performance, functionally-

graded materials and structures are drawing increasing 5.2 Industrial applications
attention more recently. For example, Sun et al. (2010a) in-
vestigated the crashing characteristics of functionally graded Over the past two decades, structural optimization has been
foam (FGF)-filled columns, in which the foam has a gradient widely used for crashworthiness design of a full-scale vehicle
density along the axial direction, and they sought the best in the automotive industry. For example, Liao et al. (2008b)
possible exponential gradient via a multiobjective optimiza- investigated a multiobjective optimization problem of frontal
tion. Fang et al. (2014b) compared the bending behavior of crash safety of a full-scale vehicle, where both full frontal and
different FGF-filled structures and found the FGF structures is 40 % offset frontal crashes were considered based on polyno-
able to generate more competent designs than the uniform mial surrogate modeling method with stepwise regression. A
counterpart. set of Pareto solutions was generated using NSGA-II, which
In addition to FGF, the concept of functionally graded provided the decision maker with insightful information. Acar
thickness (FGT) structures has also been introduced to and Solanki (2009b) performed a system reliability based op-
crashworthiness optimization. Sun et al. (2014b) proposed timization of an automotive structure for crashworthiness and
a novel tube with a longitudinal variation of thickness and analyzed the reliability allocation in different failure modes.
identified the best possible thickness gradient for achiev- They evaluated the effect of various uncertainty reduction
ing the highest SEA and lowest Fmax using multiobjective measures and plotted the tradeoff of uncertainty reduction
optimization. Note that during crashing process severe measures, system reliability, and structural weight. Liang

Fig. 7 Optimization results of

FGT and uniform multi-cell tubes
(Fang et al. (2015b)): a) Pareto
front; b) crushing force
On design optimization for structural crashworthiness 1107

and Le (2009) investigated an SSM based crashworthiness models have been commonly recognized as an effective
optimization for bus structures to maintain survival space alternative in industry applications. However, the “curse of
and reduced occupant injury when a rollover occurs. The re- dimensionality” problem arises when the number of design
sults showed the side wall deformations were reduced by 49.2 variables increases. In other words, the computational cost
and 39.4 % for upper and lower frames respectively, while the for obtaining a large number of training points can make
bus weight was only increased by 1.6 %. Goel et al. (2010) the surrogate modeling less attractive and even infeasible
proposed an efficient resource allocation of NSGA-II with (Koch et al. (1999)). There seems to be two kinds of ap-
1,000 simulations and a real-life vehicle model comprised of proaches to address this issue in literature. First, direct cou-
58,000 elements was used for design testing. Use of a moder- pling population-based optimization algorithms with FEA
ate population size would provide a reasonable trade-off could carry out the optimization more efficiently when the
among convergence to Pareto front, diversity of non- number of design variables becomes large (Duddeck
dominated solutions and computational cost in a parallel (2008), Xu et al. (2014), Xu et al. (2015)). It was also
framework. The results also showed that for the problems with recommended to use hybrid approaches to combine both
a small feasible region, the number of feasible solutions can be surrogate modeling and direct coupling methods (Redhe
significantly increased in the first few generations involving et al. (2004)), in which the optimal result from the direct
about 200 simulations. The optimal design could lead to more coupling method was used as a starting point for subsequent
than 50 % reduction in the peak acceleration and almost 6 % further surrogate-based optimization. Second, variable
increase in the time to reach zero velocity while remaining the screening techniques could be used to reduce the dimen-
mass and maximum intrusive displacement unchanged. Kiani sionality of the problem before performing a surrogate mod-
et al. (2013) investigated an RBF-based optimization of auto- el based optimization (Simpson et al. (2004)). In this regard,
motive structures considering multiple crashes and vibration Craig et al. (2005) identified some important variables
scenarios. While the intrusion, acceleration, internal energy in through an analysis of variance (ANOVA) based on linear
full frontal, offset frontal and side impacts and three frequen- surrogate models. Liang and Le (2009) investigated the ca-
cies in vibration characteristics satisfied the constraints, their pability of energy absorption of the components to reduce
study achieved a weight saving of 3.6 % by optimizing 20 the dimension of the optimization problem. Su et al. (2011)
components of the vehicle. simply considered the components with high strain energy
Real-world loading conditions of crashworthiness optimi- during the bus rollover as the influential structures. Hou
zation mainly include the front crash, side crash, and rollover et al. (2012) proposed a method of unreplicated saturated
in the literature (as summarized in Table 7). In front impact, factorial design to screen out the less important variables for
deformable yet stiff front structures with crumple zones are a vehicle crash. In industrial applications, the difficulty with
required to absorb the kinetic energy so as to reduce the crash variable screening may be due to the presence of multiple
energy transmitting to occupants. Besides, intrusion into the responses (Simpson et al. (2001)). In other words, complex
occupant compartment should also be prevented, especially in industry problems should inevitably consider multiple per-
the case of offset crashes. As a result, crash pulse (CP) (e.g., formance responses and each response may unfortunately
Craig et al. (2005), Goel et al. (2010), Liao et al. (2008a)) or its require different important design variables, which results
variants peak acceleration amax (e.g., Hamza and Saitou in the breakdown of the variable screening process.
(2012), Parrish et al. (2012), Zhu et al. (2012), Gu et al. Automotive safety components/subsystems have been
(2013)) and peak force Fmax (e.g., Zhu et al. (2009), Wang separated from the vehicle for crashworthiness optimiza-
et al. (2011b)) and intrusion Intr (e.g., Redhe and Nilsson tion in literature. Longitudinal rails, bumpers, side pillar
(2004), Yang et al. (2005), Craig et al. (2005)) are often used structures are some typical examples for optimizing the
as design criteria in optimization for a front impact. In the side crashworthiness of components as listed in Table 8.
impact, side structures are expected to minimize intrusion and Real-world road systems were considered in crashworthiness
to prevent doors from opening. Due to little space for defor- optimization for vehicular structures. The guardrail is a common
mation, Intr is the emphasis in structural optimization under traffic facility on roads, and its design affects the functionality of
side impact (e.g., Bojanowski and Kulak (2011), Wang et al. absorbing the kinetic energy and redirecting the errant vehicle
(2011a), Aspenberg et al. (2012), Xu et al. (2013)). To ensure (Hou et al. (2014b), Kurtaran et al. (2002), Yi et al. (2012)).
the rollover safety, a large resistance force and/or a small in- As discussed above, there are three different levels of systems
trusion in the roofing structure, are commonly adopted to be in crashworthiness optimization of automotive structures: com-
the design criteria (e.g., Liang and Le (2009), Pan and Zhu ponent level, vehicle level, and vehicle-road facility level. A
(2011a), Bojanowski and Kulak (2011), Su et al. (2011)). futuristic problem arising is how to integrate the designs of dif-
Because of the complexity of crashworthiness in the con- ferent system levels to ultimately improve the occupant safety.
text of full-scale vehicles, the computational cost of FEA in As a broad topic, crashworthiness optimization also applies to
optimization iterations can be rather high. Thus, surrogate other industries in real-life scenarios. In the aerospace
1108 J. Fang et al.

engineering, crashworthiness optimization has been applied to & Population-based optimization algorithms are recom-
subfloor structures of helicopters by several researchers mended for dealing with multiobjective problems for their
(Bisagni et al. (2002), Hajela and Lee (1997), Lanzi et al. capability of producing well-distributed Pareto solutions
(2004a), Astori and Impari (2013)). In the maritime industry, for decision making.
crashworthiness optimization can enhance the energy-absorbing & Due to their limitations, equivalent methods (i.e., inertia relief
capacity of the fender structure (Jiang and Gu (2010)). and equivalent static loads) might not be a general strategy
for crashworthiness topology optimization. Rigorous formu-
lation still needs further studies for addressing nonlinearity
and stability issues in topological sensitivity analysis.
6 Conclusions
& When uncertainties are presented, crashworthiness opti-
mization could become challenging. Surrogate-based
With the growing socioeconomic concerns on transport safety
Monte Carlo simulation (MCS) is recommended to quan-
and fuel consumption, significance of designing crashworthy
tify the data of uncertainties during optimization itera-
structures has been more and more recognized. Computational
tions. Besides, the experimental validation of robust and/
optimization provides a powerful tool to achieve best possible
or reliable optimization results is recommended, whenever
crashworthiness with lightest structures. This article provides a
possible, as a focus of future studies prior to applying such
state-of-the-art review of a range of key issues from structural
uncertainty optimizations to real world more extensively.
configurations, crashworthiness criteria, modeling strategies, to
& Various novel structures and materials have been proposed
multiobjective, uncertainties, and industrial applications. From
and optimized to enhance the crashworthiness. While rath-
this comprehensive review, we can draw the following conclu-
er promising, composite structures may still need to be
sions and recommendations for futuristic crashworthiness
further studied for tailoring better crashing performance.
optimization.
The concept of functionally graded structures has proven
considerable effectiveness in the crashworthiness and their
& Due to the complexity of crashworthiness problem, surrogate
further applications in the industry could be a promising
modeling has been the most popular and feasible approach to
topic of study in the future.
formulating the design criteria for optimization, especially in
& For designs of each component, their individual roles and
small scale problems (e.g. < 20 design variables). Use of
design optimization should be addressed in an integrated
multiple surrogate models is recommended for addressing
form subject to proper design criteria so that they can
modeling accuracy and optimization effectiveness. To make
perform the best as a whole.
full use of high-performance computing (HPC), efficient
& Some natural hierarchical materials have demonstrated re-
global optimization (EGO) and successive surrogate model-
markable energy absorption and impact resistance
ing (SSM) methods can be conducted with parallel comput-
(McKittrick et al. (2010)). Bio-inspired design and micro-
ing to save the wall-clock time.
structural optimization of composites could be a promising
& For real-word problems with a large number of design
area of study, which could open up a new avenue pushing the
variables, direct coupling population-based optimization
lightweighting for crashworthiness to a new level.
algorithms with finite element analysis (FEA) may be
more suitable; and data mining techniques are drawing
increasingly attention for enhancing the optimization effi- Acknowledgments This work was supported by Australian Research
ciency of population-based optimization algorithms. Council (ARC) DECRA and future fellowship scheme.
On design optimization for structural crashworthiness 1109

Appendix

Table 1 Injury-based metrics

Metrics Publications

HIC Hong and Park (2003), Youn et al. (2004), Zhu et al. (2008), Oman and Nilsson (2010), Yi et al. (2012)
CP Wu et al. (2002), Craig et al. (2005), Goel et al. (2010), Liao et al. (2008a), Liao et al. (2008b), Stander (2012),
Stander (2013)
Intr Rakheja et al. (1999), Kurtaran et al. (2002), Redhe et al. (2004), Redhe and Nilsson (2004), Craig et al. (2005),
Forsberg and Nilsson (2006), Redhe and Nilsson (2006), Zhang et al. (2007a), Duddeck (2008), Liao et al. (2008a),
Liao et al. (2008b), Shin et al. (2008), Zhu et al. (2008), Acar and Solanki (2009b), Acar and Solanki (2009a),
Horstemeyer et al. (2009), Liang and Le (2009), Acar (2010b), Acar (2010a), Goel et al. (2010), Bojanowski
and Kulak (2011), Pan and Zhu (2011b), Su et al. (2011), Wang et al. (2011a), Zhu et al. (2011), Hamza
and Saitou (2012), Hou et al. (2012), Parrish et al. (2012), Yi et al. (2012), Yildiz and Solanki (2012),
Zhu et al. (2012), Gu et al. (2013), Kiani et al. (2013), Shi et al. (2013a), Shi et al. (2013b), Xu et al. (2013),
Zhang et al. (2013b), Hamza and Shalaby (2014), Hou et al. (2014a), Kiani et al. (2014), Wang and Shi (2014),
Lönn et al. (2009), Aspenberg et al. (2012), Stander (2012), Stander (2013)
Intrusion Velocity Yang et al. (2000), Blumhardt (2001), Gu et al. (2001), Fu and Sahin (2004), Koch et al. (2004), Youn et al. (2004),
(IntrV) Sinha (2007), Sinha et al. (2007), Zhu et al. (2008), Baril et al. (2011), Wang et al. (2011a), Rangavajhala
and Mahadevan (2013), Xu et al. (2013), Zhang et al. (2013a), Zhu et al. (2013), Hou et al. (2014a),
Marklund and Nilsson (2001)
amax Rakheja et al. (1999), Blumhardt (2001), Kurtaran et al. (2002), Pedersen (2003b), Redhe et al. (2004),
Fang et al. (2005), Forsberg and Nilsson (2006), Redhe and Nilsson (2006), Cristello and Kim (2007),
Shin et al. (2008), Horstemeyer et al. (2009), Goel et al. (2010), Jeong et al. (2010), Rais-Rohani et al. (2010),
Sun et al. (2010b), Wang et al. (2010), Pan and Zhu (2011b), Sun et al. (2011), Zhu et al. (2011), Gu et al.),
Hamza and Saitou (2012), Parrish et al. (2012), Zhu et al. (2012), Gu et al. (2013), Ingrassia et al. (2013),
Kiani et al. (2013), Zhang et al. (2013b), Zhu et al. (2013), Abbasi et al. (2014), Hamza and Shalaby (2014),
Hou et al. (2014a), Hou et al. (2014b), Kiani et al. (2014), Mohammadiha and Beheshti (2014),
Aspenberg et al. (2012)
Fmax Hanssen et al. (2001), Kurtaran et al. (2002), Lee et al. (2002), Pedersen (2003b), Redhe et al. (2004), Forsberg
and Nilsson (2006), Hou et al. (2007), Huang et al. (2007), Zhang et al. (2007a), Hou et al. (2008a),
Liu (2008a), Liu (2008c), Liu (2008b), Hou et al. (2009), Zhang et al. (2009), Zhu et al. (2009), Jiang
and Gu (2010), Kaya and Oeztuerk (2010), Khakhali et al. (2010), Liu (2010b), Liu (2010a), Shariati et al. (2010),
Sun et al. (2010a), Wang et al. (2010), Allahbakhsh et al. (2011), Hou et al. (2011), Pan and Zhu (2011b),
Wang et al. (2011b), Wang et al. (2011a), Yin et al. (2011b), Zhu et al. (2011), Najafi and Rais-Rohani (2012),
Qi et al. (2012), Zhang et al. (2012), Zhu et al. (2012), Belingardi et al. (2013), Esfahlani et al. (2013),
Gedikli (2013), Song et al. (2013), Tang et al. (2013), Yang and Qi (2013), Yin et al. (2013), Duan et al. (2014),
Fang et al. (2014b), Fang et al. (2014a), Hou et al. (2014c), Mohammadiha and Beheshti (2014),
Najafi et al. (2014), Nguyen et al. (2014), Paz et al. (2014), Qi and Yang (2014), Sun et al. (2014a),
Sun et al. (2014b), Tran et al. (2014), Yin et al. (2014a), Yin et al. (2014b), Zheng et al. (2014), Redhe et al. (2002),
Jansson et al. (2003), Hunkeler et al. (2013), Bisagni et al. (2002), Lanzi et al. (2004a)
1110 J. Fang et al.

Table 2 Energy-based metrics

Metrics Publications

EA Mayer et al. (1996), Yamazaki and Han (1998), Rakheja et al. (1999), Shi and Hagiwara (2000), Hanssen et al. (2001),
Kim (2001), Kim et al. (2002), Kurtaran et al. (2002), Lee et al. (2002), Lanzi et al. (2004b), Fang et al. (2005),
Hanssen et al. (2006), Zarei and Kroger (2006), Yang and Qi (2013), Zhang et al. (2007a), Zhang et al. (2007c),
Liu (2008a), Liu (2008c), Liu (2008b), Zarei and Kroger (2008a), Zarei and Kroger (2008b),
Zarei and Kroger (2008c), Acar and Solanki (2009b), Acar and Solanki (2009a), Horstemeyer et al. (2009),
Hou et al. (2009), Zhang et al. (2009), Zhu et al. (2009), Kaya and Oeztuerk (2010), Lönn et al. (2010), Wang et al. (2010),
Marzbanrad and Ebrahimi (2011), Pan and Zhu (2011b), Sun et al. (2011), Zhu et al. (2011), Ghamarian and Zarei (2012),
Gu et al.), Parrish et al. (2012), Zhu et al. (2012), Esfahlani et al. (2013), Gu et al. (2013), Tang et al. (2013),
Zhang et al. (2013b), Paz et al. (2014), Redhe et al. (2002), Milho et al. (2004)
SEA Zarei and Kroger (2006), Hou et al. (2007), Zarei and Kroger (2007), Hou et al. (2008a), Zarei and Kroger (2008a),
Zarei and Kroger (2008c), Zhang et al. (2009), Bi et al. (2010), Jiang and Gu (2010), Khakhali et al. (2010),
Liu (2010b), Liu (2010a), Shariati et al. (2010), Sun et al. (2010a), Sun et al. (2010b), Acar et al. (2011),
Allahbakhsh et al. (2011), Hou et al. (2011), Guo et al. (2011), Marzbanrad and Ebrahimi (2011), Toksoy and
Güden (2011), Yin et al. (2011a), Yin et al. (2011b), Ghamarian and Zarei (2012), Qi et al. (2012), Zhang et al. (2012),
Belingardi et al. (2013), Gedikli (2013), Song et al. (2013), Yang and Qi (2013), Yin et al. (2013), Chen et al. (2014),
Duan et al. (2014), Fang et al. (2014b), Fang et al. (2014a), Hou et al. (2014c), Nguyen et al. (2014), Qi and Yang (2014),
Sun et al. (2014a), Sun et al. (2014b), Tran et al. (2014), Yin et al. (2014a), Yin et al. (2014b), Zhang et al. (2014a),
Zhang et al. (2014b), Zheng et al. (2014), Hunkeler et al. (2013), Costas et al. (2014), Lanzi et al. (2004a), Zhu et al. (2016)
CFE/ LU Avalle et al. (2002), Chiandussi and Avalle (2002), Avalle and Chiandussi (2007), Shakeri et al. (2007), Acar et al. (2011),
Marzbanrad and Ebrahimi (2011), Gedikli (2013), Zhang et al. (2014b), Zhou et al. (2014), Costas et al. (2014)
Favg Xiang et al. (2006), Zarei and Kroger (2006), Yang and Qi (2013), Zarei and Kroger (2008a), Zarei and Kroger (2008c),
Zhang et al. (2008), Bi et al. (2010), Toksoy and Güden (2011), Ghamarian and Zarei (2012), Najafi and
Rais-Rohani (2012), Song et al. (2013), Fang et al. (2014a), Najafi et al. (2014), Sun et al. (2014a),
Bisagni et al. (2002), Lanzi et al. (2004a)
deff / UR Shakeri et al. (2007), Marzbanrad and Ebrahimi (2011), Zhang et al. (2012), Shi et al. (2013a), Shi et al. (2013b),
Wang and Shi (2014)
IED Forsberg and Nilsson (2007), Patel et al. (2009)

Table 3 Surrogate models in crashworthiness optimization

Surrogate models Publications

Individual surrogate models PRS Blumhardt (2001), Kim (2001), Han and Yamazaki (2003), Hong and Park (2003), Xiang et al. (2006),
Zarei and Kroger (2006), Zhang et al. (2007a), Zhang et al. (2007c), Liu (2008a), Liu (2008c),
Liu (2008b), Shin et al. (2008), Zarei and Kroger (2008a), Zarei and Kroger (2008b), Zarei and
Kroger (2008c), Hou et al. (2009), Bi et al. (2010), Kaya and Oeztuerk (2010), Liu (2010b),
Liu (2010a),Shariati et al. (2010), Sun et al. (2010a), Zhang et al. (2010), Allahbakhsh and
Saetni (2011), Baril et al. (2011), Hou et al. (2011), Guo et al. (2011), Sun et al. (2011),Toksoy
and Güden (2011), Yin et al. (2011b), Bae and Huh (2012), Ghamarian and Zarei (2012),
Hou et al. (2012), Qi et al. (2012), Yin et al. (2013), Zhu et al. (2013), Duan et al. (2014),
Hou et al. (2014a), Hou et al. (2014c), Kiani et al. (2014), Mohammadiha and Beheshti (2014),
Nguyen et al. (2014), Sun et al. (2014b), Tran et al. (2014), Zhang et al. (2014a), Lönn et al.
(2009), Redhe et al. (2002), Avalle et al. (2002), Chiandussi and Avalle (2002), Avalle and
Chiandussi (2007), Marklund and Nilsson (2001), Jansson et al. (2003), Esfahlani et al. (2013),
Yang et al. (2000), Gu et al. (2001), Fu and Sahin (2004), Koch et al. (2004), Youn et al. (2004),
Sinha (2007),Sinha et al. (2007), Liao et al. (2008b), Liao et al. (2008a), Aspenberg et al. (2012),
Lönn et al. (2011), Sun et al. (2010b), Redhe and Nilsson (2004)
KRG Redhe and Nilsson (2006), Zhang et al. (2012),Yang and Qi (2013), Zhang et al. (2013a), Zhang et al.
(2013b), Fang et al. (2014b), Qi and Yang (2014), Zhang et al. (2014b), Fang et al. (2014a)
RBF Hamza and Shalaby (2014), Lanzi et al. (2004b), Horstemeyer et al. (2009), Rais-Rohani et al. (2010),
Bojanowski and Kulak (2011), Su et al. (2011), Farkas et al. (2012), Najafi and Rais-Rohani (2012),
Yildiz and Solanki (2012), Gu et al. (2013), Kiani et al. (2013), Hou et al. (2014b), Acar (2010a),
Aspenberg et al. (2012)
ANN Sun et al. (2010b), Hajela and Lee (1997),Shi and Hagiwara (2000), Khakhali et al. (2010), Jiang
and Gu (2010), Zhu et al. (2008), Marzbanrad and Ebrahimi (2011), Bisagni et al. (2002),
Lanzi et al. (2004a)
SVR Aspenberg et al. (2012), Pan et al. (2010), Wang et al. (2010), Paz et al. (2014)
Comparison of surrogate models Wang et al. (2011a), Fang et al. (2005), Forsberg and Nilsson (2006), Costas et al. (2014),
Acar and Solanki (2009b), Gedikli (2013), Xu et al. (2013),Yin et al. (2014b), Shi et al. (2013a),
Shi et al. (2013b) , Yin et al. (2011a), Zhu et al. (2012)
Multiple surrogate models Zhu et al. (2009), Pan and Zhu (2011b), Song et al. (2013), Zheng et al. (2014), Yin et al. (2014a),
(including ensemble of surrogates) Pan and Zhu (2011a), Zhu et al. (2011), Hamza and Saitou (2012), Parrish et al. (2012)
Sequential sampling Najafi et al. (2014), Chen et al. (2014), Lee et al. (2002), Kurtaran et al. (2002), Craig et al. (2005),
Hou et al. (2007), Liang and Le (2009), Sun et al. (2014a), Fang et al. (2014a), Sheldon et al. (2011),
Stander (2012), Stander (2013)
On design optimization for structural crashworthiness 1111

Table 4 Algorithms used in crashworthiness Optimization

Methods Publications

Combined cost function Hamza and Shalaby (2014), Cristello and Kim (2007), Zhang et al. (2009), Parrish et al. (2012),
Rais-Rohani et al. (2010), Mohammadiha and Beheshti (2014), Wang et al. (2010), Ghamarian and
Zarei (2012), Kaya and Oeztuerk (2010), Zarei and Kroger (2006), Acar et al. (2011), Yildiz and
Solanki (2012), Hou et al. (2012), Farkas et al. (2012), Hou et al. (2014b), Tran et al. (2014), Fang et al.
(2005), Ingrassia et al. (2013), Marzbanrad and Ebrahimi (2011), Shakeri et al. (2007), Zhu et al. (2008),
Hou et al. (2008a), Costas et al. (2014)
MOGA Hou et al. (2011), Paz et al. (2014), Kiani et al. (2014), Lanzi et al. (2004b), Najafi et al. (2014), Xu et al. (2013),
Hamza and Saitou (2012), Gu et al. (2013), Liao et al. (2008a), Sun et al. (2014b), Guo et al. (2011),
Zheng et al. (2014), Hou et al. (2014a), Bojanowski and Kulak (2011), Liao et al. (2008b), Sinha et al. (2007),
Zhang et al. (2012), Sinha (2007), Goel et al. (2010), Jiang and Gu (2010), Xu et al. (2015), Stander (2012),
Stander (2013)
MOPSO Hou et al. (2009), Qi and Yang (2014), Sun et al. (2010a), Yin et al. (2011a), Fang et al. (2014a), Sun et al. (2011),
Yin et al. (2014b), Yin et al. (2011b), Qi et al. (2012), Duan et al. (2014), Yin et al. (2014a), Zhang et al. (2014b),
Nguyen et al. (2014), Fang et al. (2014b), Yin et al. (2013), Yang and Qi (2013)

Table 5 Research works on crashworthiness optimization with uncertainties

Publications Sources of uncertainties Uncertainty

Optimization Uncertainty analysis

classification

Robust Reliability MCS AMA DSM MPP Others

Yang et al. (2000) Barrier height, barrier hitting position √ √

Gu et al. (2001) Barrier height, barrier hitting position √ √
Youn et al. (2004) Sheet thickness, material property, barrier height, √ √
barrier hitting position
Fu and Sahin (2004) Sheet thickness, material property, barrier height, √ √ √
barrier hitting position
Koch et al. (2004) Sheet thickness, material property, barrier height, √ √ √
barrier hitting position
Sinha (2007) Sheet thickness, material yield stress √ √ √
Sinha et al. (2007) Sheet thickness, material yield stress √ √ √
Zhang et al. (2007a) Sheet thickness, material yield stress √ √ √
Acar and Solanki (2009b) Material parameter, occupant mass, impact speed, √ √
error in FEA, error in our ignorance
Zhu et al. (2009) Sheet thickness, yield limit √ √ √ √
Rais-Rohani et al. (2010) Stress-strain parameter, impact speed, offset distance, √ √
occupant mass
Lönn et al. (2010) Geometrical parameters √ √
Khakhali et al. (2010) Material density, material yield stress, plastic modulus, sheet √ √ √
thickness
Zhu et al. (2011) Sheet thickness, material yield stress √ √
Baril et al. (2011) Sheet thickness, material yield stress √ √
Sun et al. (2011) Material density, material yield stress, Young’s modulus √ √ √
Farkas et al. (2012) Geometrical parameters, sheet thickness √ √
Aspenberg et al. (2012) Sheet thickness, material scaling √ √
Shi et al. (2013b) Sheet thickness √ √
Zhu et al. (2013) Surrogate modeling uncertainty √ √
Gu et al. (2013) Sheet thickness √ √ √
Rangavajhala and Mahadevan Sheet thickness, yield stress √ √
(2013)
Zhang et al. (2013a) Sheet thickness, surrogate modeling uncertainty √ √ √
Zhang et al. (2013b) Sheet thickness, surrogate modeling uncertainty √ √ √
Chen et al. (2014) Sheet thickness, yield stress √ √
Najafi et al. (2014) Young’s modulus, yield stress, tangent modulus √ √ √
Fang et al. (2014a) Sheet thickness, foam density √ √ √
Sun et al. (2014a) Sheet thickness, foam density √ √ √
1112 J. Fang et al.

Table 6 Crashworthiness optimization of energy absorbers with different configurations

Structures Works

Thin-walled tubes Cylindrical Sun et al. (2014a), Zarei and Kroger (2006), Kurtaran et al. (2002), Lee et al. (2002), Wang et al. (2010),
Marzbanrad and Ebrahimi (2011), Lanzi et al. (2004b), Ghamarian and Zarei (2012)
Square/ polygonal Yamazaki and Han (1998), Lönn et al. (2010), Allahbakhsh et al. (2011), Liu (2008c), Redhe et al. (2002),
Jansson et al. (2003), Kaya and Oeztuerk (2010), Liu (2008a), Hou et al. (2007), Liu (2010b)
Conical/ Tapered Liu (2010a), Chiandussi and Avalle (2002), Avalle and Chiandussi (2007), Hou et al. (2011),
Avalle et al. (2002), Liu (2008b), Qi et al. (2012), Acar et al. (2011), Zhang et al. (2014b)
Hat-sectional Qi and Yang (2014), Najafi and Rais-Rohani (2012), Najafi et al. (2014), Xiang et al. (2006)
Multi-cell/honeycomb Tang et al. (2013), Hou et al. (2014c), Zhang et al. (2008), Tran et al. (2014), Hou et al. (2008a),
Sun et al. (2010b), Yin et al. (2011a), Esfahlani et al. (2013), Chen et al. (2014),Zarei and
Kroger (2008a), Yin et al. (2011b)
Foam-filled Yang and Qi (2013), Zarei and Kroger (2008b), Zarei and Kroger (2008c), Toksoy and Güden (2011),
Hou et al. (2009), Zhang et al. (2012), Yin et al. (2014b), Bi et al. (2010), Hanssen et al. (2006),
Kim (2001), Kim et al. (2002), Fang et al. (2014a), Zarei and Kroger (2008a), Sun et al. (2014a),
Shariati et al. (2010), Hanssen et al. (2001), Yang and Qi (2013), Zarei and Kroger (2007),
Zarei and Kroger (2008c), Zheng et al. (2014), Zarei and Kroger (2008b), Song et al. (2013),
Zhang et al. (2014b)
TWB/ TRB Qi and Yang (2014), Tang et al. (2013), Gedikli (2013), Zhu et al. (2008), Xu et al. (2013),
Pan et al. (2010), Shi et al. (2007)
Composite Chuang et al. (2008), Zarei et al. (2008), Belingardi et al. (2013), Duan et al. (2014), Paz et al. (2014),
FGT/FGF structure Lanzi et al. (2004b), Fang et al. (2014b), Fang et al. (2016), Mohammadiha and Beheshti (2014),
Yin et al. (2013), Yin et al. (2014a), Sun et al. (2010a), Sun et al. (2014b)

Table 7 Crashworthiness optimization for automotive structures

Loading scenario Works

frontal impact Zhang et al. (2014a), Wang et al. (2011b), Wang and Shi (2014), Kiani et al. (2014), Shi et al. (2013a),
Shi et al. (2013b), Craig et al. (2005), Fang et al. (2005), Duddeck (2008), Liao et al. (2008a),
Liao et al. (2008b), Acar and Solanki (2009b), , Acar and Solanki (2009a), Hou et al. (2012),
Zhang et al. (2013b), Zhu et al. (2013), Abbasi et al. (2014), Hamza and Shalaby (2014), Yildiz
and Solanki (2012), Kiani et al. (2013), Parrish et al. (2012), Gu et al. (2013), Redhe and Nilsson (2004),
Zhu et al. (2012), Sun et al. (2011), Acar (2010a), Goel et al. (2010), Rais-Rohani et al. (2010),
Wang et al. (2010), Acar (2010b), Redhe and Nilsson (2006), Pan and Zhu (2011b), Hamza and Saitou (2012),
Zhu et al. (2009), Jansson et al. (2003), Zhu et al. (2011), Forsberg and Nilsson (2006), Sheldon et al. (2011),
Stander (2012), Zhu et al. (2016)
Side impact Yang et al. (2005), Bojanowski and Kulak (2011), Yildiz and Solanki (2012), Kiani et al. (2013),
Parrish et al. (2012), Zhu et al. (2012), Rangavajhala and Mahadevan (2013), Zhang et al. (2013a),
Baril et al. (2011), Yang et al. (2000), Fu and Sahin (2004), Koch et al. (2004), Youn et al. (2004),
Sinha (2007), Sinha et al. (2007), Horstemeyer et al. (2009), Hou et al. (2014a), Wang et al. (2011a),
Zhang et al. (2010), Xu et al. (2013) Zhu et al. (2008), Sheldon et al. (2011), Xu et al. (2015)
Roof strength and Rollover safety Gu et al. (2001), Pan and Zhu (2011a), Zhang et al. (2013a), Zhu et al. (2012), Christensen et al. (2012),
Christensen et al. (2013), Bojanowski and Kulak (2011), Su et al. (2011)
On design optimization for structural crashworthiness 1113

Table 8 Crashworthiness optimization for automotive components

Component Works

Longitudinal rail Liang and Le (2009), Yang et al. (1994), Soto (2004), Hanssen et al. (2006), Forsberg and Nilsson (2006),
Redhe et al. (2002), Shi and Hagiwara (2000), Redhe et al. (2004), Cho et al. (2006), Zhang et al. (2007a),
Zhang et al. (2007c), Cho et al. (2008), Aspenberg et al. (2012), Kim (2001), Wang et al. (2011b),
Wang et al. (2011a), Redhe and Nilsson (2006), Mayer et al. (1996), Khakhali et al. (2010), Nguyen et al. (2014)
Pillars Shi et al. (2007), Hanssen et al. (2006), Marklund and Nilsson (2001), Bae and Huh (2012)
Crash box Pan et al. (2010), Lee et al. (2013)
Bumper Costas et al. (2014), Farkas et al. (2012), Shin et al. (2008), Jeong et al. (2010), Gu et al. (2012)

References Alexander J (1960) An approximate analysis of the collapse of thin cylin-

drical shells under axial loading. Quarter J Mech Appl Math 13:10–15
Allahbakhsh HR, Saetni J (2011) Design optimization of square and
Abbasi M, Ghafari-Nazari A, Reddy S, Fard M (2014) A new approach
for optimizing automotive crashworthiness: concurrent usage of circular aluminium extrusion damage columns with crashworthiness
criteria. Indian J Eng Mater Sci 18:341–350
ANFIS and Taguchi method. Struct Multidisc Optim 49:485–499
Abdullah K, Wild P, Jeswiet J, Ghasempoor A (2001) Tensile testing for Allahbakhsh HR, Saemi J, Hourali M (2011) Design optimization of
weld deformation properties in similar gage tailor welded blanks square aluminium damage columns with crashworthiness criteria.
using the rule of mixtures. J Mater Process Technol 112:91–97 Mechanika:187–192
Abramowicz W, Jones N (1984) Dynamic axial crushing of square tubes. Antony J (2002) Design for six sigma: a breakthrough business improve-
Int J Impact Eng 2:179–208 ment strategy for achieving competitive advantage. Work Stud 51:6–8
Abramowicz W, Jones N (1986) Dynamic progressive buckling of circu- Aspenberg D, Jergeus J, Nilsson L (2012) Robust optimization of front
lar and square tubes. Int J Impact Eng 4:243–270 members in a full frontal car impact. Eng Optim 45:245–264
Abramowicz W, Wierzbicki T (1989) Axial crushing of multicorner sheet Astori P, Impari F (2013) Crash response optimisation of helicopter seat
metal columns. J Appl Mech 56:113–120 and subfloor. Int J Crashworth 18:570–578
Acar E (2010a) Optimizing the shape parameters of radial basis functions: Avalle M, Chiandussi G (2007) Optimisation of a vehicle energy absorb-
an application to automobile crashworthiness. Proc Institution Mech ing steel component with experimental validation. Int J Impact Eng
Eng Part D-J Automobile Eng 224:1541–1553 34:843–858
Acar E (2010b) Various approaches for constructing an ensemble of Avalle M, Chiandussi G, Belingardi G (2002) Design optimization by
metamodels using local measures. Struct Multidisc Optim 42:879–896 response surface methodology: application to crashworthiness de-
Acar E, Rais-Rohani M (2009) Ensemble of metamodels with optimized sign of vehicle structures. Struct Multidisc Optim 24:325–332
weight factors. Struct Multidisc Optim 37:279–294 Bae GH, Huh H (2012) Comparison of the optimum designs of center pillar
Acar E, Solanki K (2009a) Improving the accuracy of vehicle crashwor- assembly of an auto-body between conventional steel and ahss with a
thiness response predictions using an ensemble of metamodels. Int J simplified side impact analysis. Int J Automot Technol 13:205–213
Crashworth 14:49–61 Bandi P, Schmiedeler JP, Tovar A (2013) Design of crashworthy struc-
Acar E, Solanki K (2009b) System reliability based vehicle design for tures with controlled energy absorption in the hybrid cellular autom-
crashworthiness and effects of various uncertainty reduction mea- aton framework. J Mech Des 135:091002
sures. Struct Multidisc Optim 39:311–325 Baril C, Yacout S, Clement B (2011) Design for six sigma through col-
Acar E, Guler MA, Gerceker B, Cerit ME, Bayram B (2011) Multi- laborative multiobjective optimization. Comput Ind Eng 60:43–55
objective crashworthiness optimization of tapered thin-walled tubes Belingardi G, Beyene AT, Koricho EG (2013) Geometrical optimization
with axisymmetric indentations. Thin-Walled Struct 49:94–105 of bumper beam profile made of pultruded composite by numerical
Ahmetoglu MA, Brouwers D, Shulkin L, Taupin L, Kinzel GL, Altan T simulation. Compos Struct 102:217–225
(1995) Deep drawing of round cups from tailor-welded blanks. J Beyer H-G, Sendhoff B (2007) Robust optimization–a comprehensive
Mater Process Technol 53:684–694 survey. Comput Methods Appl Mech Eng 196:3190–3218
1114 J. Fang et al.

Bi J, Fang H, Wang Q, Ren X (2010) Modeling and optimization of foam- hydroformability. Proc Instit Mech Eng Part D-J Automobile Eng
filled thin-walled columns for crashworthiness designs. Finite Elem 221:511–526
Anal Des 46:698–709 Das I, Dennis JE (1997) A closer look at drawbacks of minimizing
Bisagni C, Lanzi L, Ricci S (2002) Optimization of helicopter subfloor weighted sums of objectives for Pareto set generation in multicriteria
components under crashworthiness requirements using neural net- optimization problems. Struct Optimization 14:63–69
works. J Aircr 39:296–304 Davis SC, Diegel SW, Boundy RG (2013) Transportation energy data
Bishop CM (1995) Neural networks for pattern recognition. Oxford uni- book. 32 edn., Oak Ridge National Lab
versity press Deb K, Pratap A, Agarwal S, Meyarivan T (2002) A fast and elitist
Blumhardt R (2001) FEM - crash simulation and optimization. Int J Veh multiobjective genetic algorithm: NSGA-II. Evol Comput, IEEE
Des 26:331–347 Trans 6:182–197
Bojanowski C, Kulak RF (2011) Multi-objective optimisation and sensi- Draper NR, Smith H (1981) Applied regression analysis 2nd ed
tivity analysis of a paratransit bus structure for rollover and side Du Bois P et al. (2004) Vehicle crashworthiness and occupant protection
impact tests. Int J Crashworth 16:665–676 Duan S, Tao Y, Han X, Yang X, Hou S, Hu Z (2014) Investigation on
Borvik T, Hopperstad OS, Reyes A, Langseth M, Solomos G, Dyngeland structure optimization of crashworthiness of fiber reinforced poly-
T (2003) Empty and foam-filled circular aluminium tubes subjected mers materials. Composit Part B-Eng 60:471–478
to axial and oblique quasi-static loading. Int J Crashworth 8:481– Duddeck F (2008) Multidisciplinary optimization of car bodies. Struct
494 Multidisc Optim 35:375–389
Breyfogle III FW (2003) Implementing six sigma: smarter solutions using Esfahlani SS, Shirvani H, Shirvani A, Nwaubani S, Mebrahtu H, Chirwa
statistical methods. John Wiley & Sons C (2013) Hexagonal honeycomb cell optimisation by way of meta-
Chen WG (2001) Optimisation for minimum weight of foam-filled tubes model techniques. Int J Crashworth 18:264–275
under large twisting rotation. Int J Crashworth 6:223–241 Fang H, Rais-Rohani M, Liu Z, Horstemeyer MF (2005) A comparative
Chen W, Allen JK, Tsui K-L, Mistree F (1996) A procedure for robust study of metamodeling methods for multiobjective crashworthiness
design: minimizing variations caused by noise factors and control optimization. Comput Struct 83:2121–2136
factors. J Mech Des 118:478–485 Fang J, Gao Y, Sun G, Zhang Y, Li Q (2014a) Crashworthiness design of
Chen Z, Qiu H, Gao L, Li X, Li P (2014) A local adaptive sampling foam-filled bitubal Structures with uncertainty. Int J Non-Linear
method for reliability-based design optimization using Kriging mod- Mech
el. Struct Multidisc Optim 49:401–416 Fang J, Gao Y, Sun G, Zhang Y, Li Q (2014a) Parametric analysis and
multiobjective optimization for functionally graded foam-filled thin-
Chiandussi G, Avalle M (2002) Maximisation of the crushing perfor-
wall tube under lateral impact. Comput Mater Sci 90:265–275
mance of a tubular device by shape optimisation. Comput Struct
Fang J, Gao Y, Sun G, Xu C, Li Q (2015a) Multiobjective robust design
80:2425–2432
optimization of fatigue life for a truck cab. Reliab Eng Syst Saf 135:
Cho Y-B, Bae C-H, Suh M-W, Sin H-C (2006) A vehicle front frame
1–8
crash design optimization using hole-type and dent-type crush initi-
Fang J, Gao Y, Sun G, Zheng G, Li Q (2015b) Dynamic crashing behav-
ator. Thin-Walled Struct 44:415–428
ior of new extrudable multi-cell tubes with a functionally graded
Cho Y-B, Bae C-H, Suh M-W, Sin H-C (2008) Maximisation of crash
thickness. Int J Mech Sci 103:63–73
energy absorption by crash trigger for vehicle front frame using the
Fang J, Gao Y, An X, Sun G, Chen J, Li Q (2016) Design of transversely-
homogenisation method. Int J Veh Des 46:23–50
graded foam and wall thickness structures for crashworthiness
Choi W, Park G (1999) Transformation of dynamic loads into equivalent criteria. Compos Part B 92:338–349
static loads based on modal analysis. Int J Numer Methods Eng 46: Farkas L, Moens D, Donders S, Vandepitte D (2012) Optimisation study
29–43 of a vehicle bumper subsystem with fuzzy parameters. Mech Syst
Christensen J, Bastien C, Blundell MV (2012) Effects of roof crush load- Signal Process 32:59–68
ing scenario upon body in white using topology optimisation. Int J Forrester A, Keane AJ (2009) Recent advances in surrogate-based opti-
Crashworth 17:29–38 mization. Prog Aerosp Sci 45:50–79
Christensen J, Bastien C, Blundell MV, Batt PA (2013) Buckling consid- Forrester A, Sobester A, Keane A (2008) Engineering design via surro-
erations and cross-sectional geometry development for topology gate modelling: a practical guide. John Wiley & Sons
optimised body in white. Int J Crashworth 18:319–330 Forsberg J, Nilsson L (2006) Evaluation of response surface methodolo-
Chuang C, Yang R (2012) Benchmark of topology optimization methods gies used in crashworthiness optimization. Int J Impact Eng 32:759–
for crashworthiness design. In: 12th International LS-DYNA Users 777
Conference, Dearborn, Michigan, USA Forsberg J, Nilsson L (2007) Topology optimization in crashworthiness
Chuang C, Yang R, Li G, Mallela K, Pothuraju P (2008) design. Struct Multidisc Optim 33:1–12
Multidisciplinary design optimization on vehicle tailor rolled blank Friedman JH (1991) Multivariate adaptive regression splines. Ann Stat:
design. Struct Multidisc Optim 35:551–560 1–67
Coello CAC, Pulido GT, Lechuga MS (2004) Handling multiple objec- Fu Y, Sahin KH (2004) Better Optimization of Nonlinear Uncertain
tives with particle swarm optimization. Evol Comput, IEEE Trans 8: Systems (BONUS) for vehicle structural design. Ann Oper Res
256–279 132:69–84
Costas M, Díaz J, Romera L, Hernández S (2014) A multi-objective Gedikli H (2013) Crashworthiness optimization of foam-filled tailor-
surrogate-based optimization of the crashworthiness of a hybrid im- welded tube using coupled finite element and smooth particle hy-
pact absorber. Int J Mech Sci 88:46–54 drodynamics method. Thin-Walled Struct 67:34–48
Couckuyt I, Deschrijver D, Dhaene T (2014) Fast calculation of Ghamarian A, Zarei H (2012) Crashworthiness investigation of conical
multiobjective probability of improvement and expected improve- and cylindrical end-capped tubes under quasi-static crash loading.
ment criteria for Pareto optimization. J Glob Optim 60:575–594 Int J Crashworth 17:19–28
Craig KJ, Stander N, Dooge DA, Varadappa S (2005) Automotive crash- Goel T, Haftka R, Shyy W, Queipo NV (2007) Ensemble of surrogates.
worthiness design using response surface-based variable screening Struct Multidisc Optim 33:199–216
and optimization. Eng Comput 22:38–61 Goel T, Stander N, Lin Y-Y (2010) Efficient resource allocation for ge-
Cristello N, Kim IY (2007) Multidisciplinary design optimization of a netic algorithm based multi-objective optimization with 1,000 sim-
zero-emission vehicle chassis considering crashworthiness and ulations. Struct Multidisc Optim 41:421–432
On design optimization for structural crashworthiness 1115

Gu L, Yang R, Tho CH, Makowski M, Faruque O, Li Y (2001) Hou S, Dong D, Ren L, Han X (2012) Multivariable crashworthiness
Optimization and robustness for crashworthiness of side impact. optimization of vehicle body by unreplicated saturated factorial de-
Int J Veh Des 26:348–360 sign. Struct Multidisc Optim 46:891–905
Gu J, Li GY, Dong Z (2012) Hybrid and adaptive meta-model-based Hou S, Liu T, Dong D, Han X (2014a) Factor screening and multivariable
global optimization. Eng Optim 44:87–104 crashworthiness optimization for vehicle side impact by factorial
Gu X, Sun G, Li G, Mao L, Li Q (2013) A Comparative study on design. Struct Multidisc Optim 49:147–167
multiobjective reliable and robust optimization for crashworthiness Hou S, Tan W, Zheng Y, Han X, Li Q (2014b) Optimization design of
design of vehicle structure. Struct Multidisc Optim 48:669–684 corrugated beam guardrail based on RBF-MQ surrogate model and
Guo L, Tovar A, Penninger CL, Renaud JE (2011) Strain-based topology collision safety consideration. Adv Eng Softw 78:28–40
optimisation for crashworthiness using hybrid cellular automata. Int Hou S, Zhang Z, Yang X, Yin H, Li Q (2014c) Crashworthiness optimi-
J Crashworth 16:239–252 zation of new thin-walled cellular configurations. Eng Comput 31:
Gupta NK, Velmurugan R (1999) Axial compression of empty and foam 4–4
filled composite conical shells. J Compos Mater 33:567–591 Huang X, Xie YM, Lu G (2007) Topology optimization of energy-
Haftka R, Villanueva D, Chaudhuri A (2016) Parallel surrogate-assisted absorbing structures. Int J Crashworth 12:663–675
global optimization with expensive functions–a survey. Struct Hunkeler S, Duddeck F, Rayamajhi M, Zimmer H (2013) Shape optimi-
Multidisc Optim 54:3–13 sation for crashworthiness followed by a robustness analysis with
Hajela P, Lee E (1997) Topological optimization of rotorcraft subfloor respect to shape variables. Struct Multidisc Optim 48:367–378
structures for crashworthiness considerations. Comput Struct 64:65– Ingrassia T, Nigrelli V, Buttitta R (2013) A comparison of simplex and
76 simulated annealing for optimization of a new rear underrun protec-
Hamza K, Saitou K (2012) A co-evolutionary approach for design opti- tive device. Eng Comput 29:345–358
mization via ensembles of surrogates with application to vehicle Jacob GC, Fellers JF, Simunovic S, Starbuck JM (2002) Energy absorp-
crashworthiness. J Mech Design 134 tion in polymer composites for automotive crashworthiness. J
Hamza K, Shalaby M (2014) A framework for parallelized efficient glob- Compos Mater 36:813–850
al optimization with application to vehicle crashworthiness optimi- Jansson T, Nilsson L, Redhe M (2003) Using surrogate models and re-
zation. Eng Optim 46:1200–1221 sponse surfaces in structural optimization – with application to
Han J, Yamazaki K (2003) Crashworthiness optimization of S-shape crashworthiness design and sheet metal forming. Struct Multidisc
square tubes. Int J Veh Des 31:72–85 Optim 25:129–140
Hanssen AG, Langseth M, Hopperstad OS (2000) Static and dynamic Jeong SB, Yi SI, Kan CD, Nagabhushana V, Park G (2008) Structural
crushing of circular aluminium extrusions with aluminium foam optimization of an automobile roof structure using equivalent static
filler. Int J Impact Eng 24:475–507 loads. Proc Institution Mech Eng Part D-J Automobile Eng 222:
Hanssen AG, Langseth M, Hopperstad OS (2001) Optimum design for 1985–1995
energy absorption of square aluminium columns with aluminium Jeong SB, Yoon S, Xu S, Park G (2010) Non-linear dynamic response
foam filler. Int J Mech Sci 43:153–176 structural optimization of an automobile frontal structure using
Hanssen AG, Stobener K, Rausch G, Langseth M, Keller H (2006) equivalent static loads. Proc Institution Mech Eng Part D-J
Optimisation of energy absorption of an A-pillar by metal foam Automobile Eng 224:489–501
insert. Int J Crashworth 11:231–241 Jiang Z, Gu M (2010) Optimization of a fender structure for the crash-
Hardy RL (1971) Multiquadric equations of topography and other irreg- worthiness design. Mater Des 31:1085–1095
ular surfaces. J Geophys Res 76:1905–1915 Jin R, Chen W, Simpson TW (2001) Comparative studies of
Helton JC, Johnson JD, Sallaberry CJ, Storlie CB (2006) Survey of metamodelling techniques under multiple modelling criteria. Struct
sampling-based methods for uncertainty and sensitivity analysis. Multidisc Optim 23:1–13
Reliability Eng Syst Saf 91:1175–1209 Jin R, Du X, Chen W (2003) The use of metamodeling techniques for
Hohenbichler M, Rackwitz R (1982) First-order concepts in system reli- optimization under uncertainty. Struct Multidisc Optim 25:99–116
ability. Struct Saf 1:177–188 Kang B, Choi W, Park G (2001) Structural optimization under equivalent
Hong UP, Park G (2003) Determination of the crash pulse and optimiza- static loads transformed from dynamic loads based on displacement.
tion of the crash components using the response surface approxi- Comput Struct 79:145–154
mate optimization. Proc Institution Mech Eng Part D-J Automobile Kaushik A, Ramani A (2014) Topology optimization for nonlinear dy-
Eng 217:203–213 namic problems: considerations for automotive crashworthiness.
Horstemeyer MF, Ren XC, Fang H, Acar E, Wang PT (2009) A compar- Eng Optim 46:487–502
ative study of design optimisation methodologies for side-impact Kaya N, Oeztuerk F (2010) Multi-objective crashworthiness design opti-
crashworthiness, using injury-based versus energy-based criterion. misation of thin-walled tubes. Int J Veh Des 52:54–63
Int J Crashworth 14:125–138 Keane AJ (2006) Statistical improvement criteria for use in multiobjective
Hou S, Li Q, Long S, Yang X, Li W (2007) Design optimization of design optimization. AIAA J 44:879–891
regular hexagonal thin-walled columns with crashworthiness Khakhali A, Nariman-zadeh N, Darvizeh A, Masoumi A, Notghi B
criteria. Finite Elem Anal Des 43:555–565 (2010) Reliability-based robust multi-objective crashworthiness op-
Hou S, Li Q, Long S, Yanga X, Li W (2008a) Multiobjective optimization timisation of S-shaped box beams with parametric uncertainties. Int
of multi-cell sections for the crashworthiness design. Int J Impact J Crashworth 15:443–456
Eng 35:1355–1367 Kiani M, Gandikota I, Parrish A, Motoyama K, Rais-Rohani M (2013)
Hou S, Li Q, Long SY, Yang XJ, Li W (2008b) Multiobjective optimiza- Surrogate-based optimisation of automotive structures under multi-
tion of multi-cell sections for the crashworthiness design. Int J ple crash and vibration design criteria. Int J Crashworth 18:473–482
Impact Eng 35:1355–1367 Kiani M, Motoyama K, Rais-Rohani M, Shiozaki H (2014) Joint stiffness
Hou S, Li Q, Long S, Yang X, Li W (2009) Crashworthiness design for analysis and optimization as a mechanism for improving the struc-
foam filled thin-wall structures. Mater Des 30:2024–2032 tural design and performance of a vehicle. Proc Instit Mech Eng Part
Hou S, Han X, Sun G, Long S, Li W, Yang X, Li Q (2011) Multiobjective D-J Automobile Eng 228:689–700
optimization for tapered circular tubes. Thin-Walled Struct 49:855– Kim HS (2001) Analysis of crash response of aluminium foam-filled
863 front side rail of a passenger car. Int J Crashworth 6:189–207
1116 J. Fang et al.

Kim HS (2002) New extruded multi-cell aluminum profile for maximum Lönn D, Bergman G, Nilsson L, Simonsson K (2011) Experimental and
crash energy absorption and weight efficiency. Thin-Walled Struct finite element robustness studies of a bumper system subjected to an
40:311–327 offset impact loading. Int J Crashworth 16:155–168
Kim HS, Chen W, Wierzbicki T (2002) Weight and crash optimization of Marklund PO, Nilsson L (2001) Optimization of a car body component
foam-filled three-dimensional “S” frame. Comput Mech 28:417– subjected to side impact. Struct Multidisc Optim 21:383–392
424 Marsolek J, Reimerdes HG (2004) Energy absorption of metallic cylin-
Kim S-H, Kim H, Kim NJ (2015) Brittle intermetallic compound makes drical shells with induced non-axisymmetric folding patterns. Int J
ultrastrong low-density steel with large ductility. Nature 518:77–79 Impact Eng 30:1209–1223
Kinsey B, Liu Z, Cao J (2000) A novel forming technology for tailor- Marzbanrad J, Ebrahimi MR (2011) Multi-objective optimization of alu-
welded blanks. J Mater Process Technol 99:145–153 minum hollow tubes for vehicle crash energy absorption using a
Koch PN, Simpson TW, Allen JK, Mistree F (1999) Statistical approxi- genetic algorithm and neural networks. Thin-Walled Struct 49:
mations for multidisciplinary design optimization: the problem of 1605–1615
size. J Aircr 36:275–286 Mayer RR, Kikuchi N, Scott RA (1996) Application of topological opti-
Koch PN, Yang R, Gu L (2004) Design for six sigma through robust mization techniques to structural crashworthiness. Int J Numer
optimization. Struct Multidisc Optim 26:235–248 Methods Eng 39:1383–1403
Kurtaran H, Eskandarian A, Marzougui D, Bedewi NE (2002) McKittrick J et al (2010) Energy absorbent natural materials and
Crashworthiness design optimization using successive response sur- bioinspired design strategies: A review. Mater Sci Eng C 30:331–
face approximations. Comput Mech 29:409–421 342
Lanzi L, Bisagni C, Ricci S (2004a) Crashworthiness optimization of Messac A, Puemi-Sukam C, Melachrinoudis E (2000a) Aggregate objec-
helicopter subfloor based on decomposition and global approxima- tive functions and Pareto frontiers: required relationships and prac-
tion. Struct Multidisc Optim 27:401–410 tical implications. Optim Eng 1:171–188
Lanzi L, Castelletti LML, Anghileri M (2004b) Multi-objective optimi- Messac A, Sundararaj GJ, Tappeta RV, Renaud JE (2000b) Ability of
sation of composite absorber shape under crashworthiness require- objective functions to generate points on nonconvex Pareto fron-
ments. Compos Struct 65:433–441 tiers. AIAA J 38:1084–1091
Lee Y, Ahn J-S, Park G Crash optimization of automobile frontal and side Milho JF, Ambrósio JAC, Pereira MFOS (2004) Design of train crash
structures using equivalent static loads.In: 11th World Congress on experimental tests by optimization procedures. Int J Crashworth 9:
Structural and Multidisciplinary Optimization, Sydney, NSW, 483–493
Australia, 2015 Mohammadiha O, Beheshti H (2014) Optimization of functionally grad-
Lee SH, Kim HY, Oh SI (2002) Cylindrical tube optimization using ed foam-filled conical tubes under axial impact loading. J Mech Sci
response surface method based on stochastic process. J Mater Technol 28:1741–1752
Process Technol 130:490–496 Montgomery DC (1996) Desing and analysis of experiments. Wiley, New
Lee I, Choi KK, Noh Y, Zhao L, Gorsich D (2011) Sampling-based York, EUA
stochastic sensitivity analysis using score functions for RBDO prob- Naceur H, Guo YQ, Ben-Elechi S (2006) Response surface methodology
lems with correlated random variables. J Mech Des 133:021003 for design of sheet forming parameters to control springback effects.
Lee S-J, Lee H-A, Yi S-I, Kim D-S, Yang HW, Park G (2013) Design Comput Struct 84:1651–1663
flow for the crash box in a vehicle to maximize energy absorption. Najafi A, Rais-Rohani M (2012) Sequential coupled process-
Proc Inst Mech Eng Part D-J Automobile Eng 227:179–200 performance simulation and multi-objective optimization of thin-
Liang C-C, Le G-N (2009) Bus rollover crashworthiness under European walled tubes. Mater Des 41:89–98
standard: an optimal analysis of superstructure strength using suc- Najafi A, Acar E, Rais-Rohani M (2014) Multi-objective robust design of
cessive response surface method. Int J Crashworth 14:623–639 energy-absorbing components using coupled process–performance
Liao X, Li Q, Yang X, Li W, Zhang W (2008a) A two-stage multi-objec- simulations. Eng Optim 46:146–164
tive optimisation of vehicle crashworthiness under frontal impact. Nguyen VS, Wen G, Yin H, Van TP (2014) Optimisation design of rein-
Int J Crashworth 13:279–288 forced S-shaped frame structure under axial dynamic loading. Int J
Liao X, Li Q, Yang X, Zhang W, Li W (2008b) Multiobjective optimiza- Crashworth 19:385–393
tion for crash safety design of vehicles using stepwise regression Oman M, Nilsson L (2010) Structural optimization of product families
model. Struct Multidisc Optim 35:561–569 subjected to multiple crash load cases. Struct Multidisc Optim 41:
Liu Y (2008a) Crashworthiness design of multi-corner thin-walled col- 797–815
umns. Thin-Walled Struct 46:1329–1337 O’Neill B (2009) Preventing passenger vehicle occupant injuries by ve-
Liu Y (2008b) Design optimisation of tapered thin-walled square tubes. hicle design—a historical perspective from IIHS. Traffic Injury
Int J Crashworth 13:543–550 Prevent 10:113–126
Liu Y (2008c) Optimum design of straight thin-walled box section beams Ortmann C, Schumacher A (2013) Graph and heuristic based topology
for crashworthiness analysis. Finite Elem Anal Des 44:139–147 optimization of crash loaded structures. Struct Multidisc Optim 47:
Liu Y (2010a) Crashworthiness design of thin-walled curved beams with 839–854
box and channel cross sections. Int J Crashworth 15:413–423 Pan F, Zhu P (2011a) Design optimisation of vehicle roof structures:
Liu Y (2010b) Thin-walled curved hexagonal beams in crashes - FEA benefits of using multiple surrogates. Int J Crashworth 16:85–95
and design. Int J Crashworth 15:151–159 Pan F, Zhu P (2011b) Lightweight design of vehicle front-end structure:
Liu S-T, Tong Z-Q, Tang Z-L, Zhang Z-H (2014) Design optimization of contributions of multiple surrogates. Int J Veh Des 57:124–147
the S-frame to improve crashworthiness. Acta Mech Sin 30:589– Pan F, Zhu P, Zhang Y (2010) Metamodel-based lightweight design of B-
599 pillar with TWB structure via support vector regression. Comput
Lönn D, Öman M, Nilsson L, Simonsson K (2009) Finite element based Struct 88:36–44
robustness study of a truck cab subjected to impact loading. Int J Park G (2011) Technical overview of the equivalent static loads method
Crashworth 14:111–124 for non-linear static response structural optimization. Struct
Lönn D, Fyllingen Ø, Nilssona L (2010) An approach to robust optimi- Multidisc Optim 43:319–337
zation of impact problems using random samples and meta-model- Park G, Lee T-H, Lee KH, Hwang K-H (2006) Robust design: an over-
ling. Int J Impact Eng 37:723–734 view. AIAA J 44:181–191
On design optimization for structural crashworthiness 1117

Parkinson A, Sorensen C, Pourhassan N (1993) A general approach for Rzesnitzek T, Müllerschön H, Günther FC, Wozniak M (2002) Two-
robust optimal design. J Mech Des 115:74–80 Stage stochastic and deterministic optimization. In: 1st
Parrish A, Rais-Rohani M, Najafi A (2012) Crashworthiness optimisation DYNAMORE Users’ Forum, Bad Mergentheim, German
of vehicle structures with magnesium alloy parts. Int J Crashworth Sacks J, Welch WJ, Mitchell TJ, Wynn HP (1989) Design and analysis of
17:259–281 computer experiments. Stat Sci 4:409–423
Patel NM, Kang B-S, Renaud JE, Tovar A (2009) Crashworthiness de- Santosa SP, Wierzbicki T, Hanssen AG, Langseth M (2000) Experimental
sign using topology optimization. J Mech Design 131 and numerical studies of foam-filled sections. Int J Impact Eng 24:
Paz J, Díaz J, Romera L, Costas M (2014) Crushing analysis and multi- 509–534
objective crashworthiness optimization of GFRP honeycomb-filled Schonlau M (1998) Computer experiments and global optimization.
energy absorption devices. Finite Elem Anal Des 91:30–39 University of Waterloo
Pedersen CBW (2003a) Topology optimization design of crushed 2D- Seitzberger M, Rammerstorfer FG, Degischer HP, Gradinger R (1997)
frames for desired energy absorption history. Struct Multidisc Crushing of axially compressed steel tubes filled with aluminium
Optim 25:368–382 foam. Acta Mech 125:93–105
Pedersen CBW (2003b) Topology optimization for crashworthiness of Seitzberger M, Rammerstorfer FG, Gradinger R, Degischer HP,
frame structures. Int J Crashworth 8:29–39 Blaimschein M, Walch C (2000) Experimental studies on the
Pedersen CBW (2004) Crashworthiness design of transient frame struc- quasi-static axial crushing of steel columns filled with aluminium
tures using topology optimization. Comput Methods Appl Mech foam. Int J Solids Struct 37:4125–4147
Eng 193:653–678 Shakeri M, Mirzaeifar R, Salehghaffari S (2007) New insights into the
Qi C, Yang S (2014) Crashworthiness and lightweight optimisation of collapsing of cylindrical thin-walled tubes under axial impact load.
thin-walled conical tubes subjected to an oblique impact. Int J Proc Instit Mech Eng Part C-J Mech Eng Sci 221:869–885
Crashworth 19:334–351 Shariati M, Allahbakhsh HR, Saemi J, Sedighi M (2010) Optimization of
Qi C, Yang S, Dong F (2012) Crushing analysis and multiobjective crash- foam filled spot-welded column for the crashworthiness design.
worthiness optimization of tapered square tubes under oblique im- Mechanika:10-16
pact loading. Thin-Walled Struct 59:103–119 Sheldon A, Helwig E, Cho Y-B (2011) Investigation and application of
Qu X, Haftka R (2004) Reliability-based design optimization using prob- multi-disciplinary optimization for automotive body-in-white devel-
abilistic sufficiency factor. Struct Multidisc Optim 27:314–325 opment. In: 8th European LS-DYNA Users conference, Strasbourg,
23–24
Rais-Rohani M, Solanki KN, Acar E, Eamon CD (2010) Shape and sizing
Shi Q, Hagiwara I (2000) Optimal design method to automobile problems
optimisation of automotive structures with deterministic and proba-
using holographic neural network’s approximation. Jpn J Ind Appl
bilistic design constraints. Int J Veh Des 54:309–338
Math 17:321–339
Rakheja S, Balike M, Hoa SV (1999) Study of an energy dissipative
Shi Y, Zhu P, Shen L, Lin Z (2007) Lightweight design of automotive
under-ride guard for enhancement of crashworthiness in a car-
front side rails with TWB concept. Thin-Walled Struct 45:8–14
truck collision. Int J Veh Des 22:29–53
Shi L, Yang R, Zhu P (2012) A method for selecting surrogate models in
Ramakrishna S (1997) Microstructural design of composite materials for
crashworthiness optimization. Struct Multidisc Optim 46:159–170
crashworthy structural applications. Mater Des 18:167–173
Shi L, Yang R, Zhu P (2013a) An adaptive response surface method for
Ramu P, Kim NH, Haftka R (2007) Error amplification in failure proba- crashworthiness optimization. Eng Optim 45:1365–1377
bility estimates of small errors in surrogates. In: SAE 2007 World Shi L, Zhu P, Yang R-J, Lin S-P (2013b) Adaptive sampling-based
Congress. Detroit, Michigan, USA RBDO method for vehicle crashworthiness design using Bayesian
Rangavajhala S, Mahadevan S (2013) Design optimization for robustness metric and stochastic sensitivity analysis with independent random
in multiple performance functions. Struct Multidisc Optim 47:523– variables. Int J Crashworth 18:331–342
538 Shimoyama K, Shinkyu J, Obayashi S (2013) Kriging-surrogate-based
Raquel C, Naval P (2005) An effective use of crowding distance in optimization considering expected hypervolume improvement in
multiobjective particle swarm optimization. In: Proceedings of the non-constrained many-objective test problems. In: Evolutionary
2005 conference on Genetic and evolutionary computation, Computation (CEC), 2013 I.E. Congress on, 658–665
Washington DC, USA. New York, NY, USA, pp 257-264 Shin MK, Park KJ, Park G (2007) Optimization of structures with non-
Redhe M, Nilsson L (2004) Optimization of the new Saab 9-3 exposed to linear behavior using equivalent loads. Comput Methods Appl
impact load using a space mapping technique. Struct Multidisc Mech Eng 196:1154–1167
Optim 27:411–420 Shin MK, Yi SI, Kwon OT, Park G (2008) Structural optimization of the
Redhe M, Nilsson L (2006) A multipoint version of space mapping op- automobile frontal structure for pedestrian protection and the low-
timization applied to vehicle crashworthiness design. Struct speed impact test. Proc Instit Mech Eng Part D-J Automobile Eng
Multidisc Optim 31:134–146 222:2373–2387
Redhe M, Forsberg J, Jansson T, Marklund PO, Nilsson L (2002) Using Simpson TW, Poplinski J, Koch PN, Allen JK (2001) Metamodels for
the response surface methodology and the D-optimality criterion in computer-based engineering design: survey and recommendations.
crashworthiness related problems. Struct Multidisc Optim 24:185– Eng Comput 17:129–150
194 Simpson TW, Booker AJ, Ghosh D, Giunta AA, Koch PN, Yang R
Redhe M, Giger M, Nilsson L (2004) An investigation of structural op- (2004) Approximation methods in multidisciplinary analysis and
timization in crashworthiness design using a stochastic approach - a optimization: a panel discussion. Struct Multidisc Optim 27:302–
comparison of stochastic optimization and the response surface 313
methodology. Struct Multidisc Optim 27:446–459 Sinha K (2007) Reliability-based multiobjective optimization for automo-
Reid SR, Reddy TY (1986) Static and dynamic crushing of tapered sheet tive crashworthiness and occupant safety. Struct Multidisc Optim
metal tubes of rectangular cross-section. Int J Mech Sci 28:623–637 33:255–268
Reyes A, Hopperstad O, Hanssen A, Langseth M (2004) Modeling of Sinha K, Krishnan R, Raghavendra D (2007) Multi-objective robust op-
material failure in foam-based components. Int J Impact Eng 30: timisation for crashworthiness during side impact. Int J Veh Des 43:
805–834 116–135
Roux W (2011) Topology Design using LS-TaSC™ Version 2 and LS- Smola A, Schölkopf B (2004) A tutorial on support vector regression.
DYNA® Stat Comput 14:199–222
1118 J. Fang et al.

Sobester A, Leary SJ, Keane AJ (2004) A parallel updating scheme for 14th International Conference on Material Forming Esaform, 2011
approximating and optimizing high fidelity computer simulations. Proceedings 1353:1656-1661
Struct Multidisc Optim 27:371–383 Vining GG, Myers RH (1990) Combining Taguchi and response surface
Song X, Sun G, Li G, Gao W, Li Q (2013) Crashworthiness optimization philosophies: a dual response approach. J Qual Technol 22
of foam-filled tapered thin-walled structure using multiple surrogate Wang GG (2003) Adaptive response surface method using inherited latin
models. Struct Multidisc Optim 47:221–231 hypercube design points. J Mech Des 125:210–220
Soto CA (2004) Structural topology optimization for crashworthiness. Int Wang GG, Shan S (2007) Review of metamodeling techniques in support
J Crashworth 9:277–283 of engineering design optimization. J Mech Des 129:370–380
Stander N (2012) An efficient new sequential strategy for multi-objective Wang X, Shi L (2014) A new metamodel method using Gaussian process
optimization using LS-OPT®. Paper presented at the 12th based bias function for vehicle crashworthiness design. Int J
International LS-DYNA Users Conference, Dearborn, Michigan, Crashworth 19:311–321
USA Wang H, Li GY, Li E (2010) Time-based metamodeling technique for
Stander N An Adaptive Surrogate-Assisted Strategy for Multi-Objective vehicle crashworthiness optimization. Comput Methods Appl Mech
Optimization. In: 10th World Congress on Structural and Eng 199:2497–2509
Multidisciplinary Optimization, Orlando, Florida, USA, 2013. Wang H, Li E, Li GY (2011a) Probability-based least square support
Su R, Gui L, Fan Z (2011) Multi-objective optimization for bus body with vector regression metamodeling technique for crashworthiness op-
strength and rollover safety constraints based on surrogate models. timization problems. Comput Mech 47:251–263
Struct Multidisc Optim 44:431–441 Wang H, Shan S, Wang GG, Li G (2011b) Integrating least square support
Sun G, Li G, Hou S, Zhou S, Li W, Li Q (2010a) Crashworthiness design vector regression and mode pursuing sampling optimization for
for functionally graded foam-filled thin-walled structures. Mater Sci crashworthiness design. J Mech Des 133
Eng A-Struct Mater Properties Microstruct Process 527:1911–1919 Wierzbicki T, Abramowicz W (1983) On the crushing mechanics of thin-
Sun G, Li G, Stone M, Li Q (2010b) A two-stage multi-fidelity optimi- walled structures. J Appl Mech-T Asme 50:727–734
zation procedure for honeycomb-type cellular materials. Comput Witowski K, Erhart A, Schumacher P, Müllerschön H (2012) Topology
Mater Sci 49:500–511 optimization for crash
Sun G, Li G, Zhou S, Li H, Hou S, Li Q (2011) Crashworthiness design of Wu JP, Nusholtz GS, Bilkhu S (2002) Optimization of vehicle crash
vehicle by using multiobjective robust optimization. Struct pulses in relative displacement domain. Int J Crashworth 7:397–413
Multidisc Optim 44:99–110 Xiang YJ, Wang Q, Fan ZJ, Fang HB (2006) Optimal crashworthiness
design of a spot-welded thin-walled hat section. Finite Elem Anal
Sun G, Song X, Baek S, Li Q (2014a) Robust optimization of foam-filled
thin-walled structure based on sequential Kriging metamodel. Struct Des 42:846–855
Multidisc Optim 49:897–913 Xu F, Sun G, Li G, Li Q (2013) Crashworthiness design of multi-
component tailor-welded blank (TWB) structures. Struct Multidisc
Sun G, Xu F, Li G, Li Q (2014b) Crashing analysis and multiobjective
Optim 48:653–667
optimization for thin-walled structures with functionally graded
Xu H, Majcher MT, Chuang C, Fu Y, Yang R (2014) Comparative bench-
thickness. Int J Impact Eng 64:62–74
mark studies of response surface model-based optimization and di-
Tang L, Wang H, Li G, Xu F (2013) Adaptive heuristic search algorithm
rect multidisciplinary design optimization. SAE Technical Paper
for discrete variables based multi-objective optimization. Struct
Xu H, Chuang C, Yang R (2015) A data mining-based strategy for direct
Multidisc Optim 48:821–836
multidisciplinary optimization. SAE Int J Mater Manufact 8:357–
Tarlochan F, Samer F, Hamouda AMS, Ramesh S, Khalid K (2013)
363
Design of thin wall structures for energy absorption applications:
Yamazaki K, Han J (1998) Maximization of the crushing energy absorp-
Enhancement of crashworthiness due to axial and oblique impact
tion of tubes. Struct Optim 16:37–46
forces. Thin-Walled Struct 71:7–17
Yang S, Qi C (2013) Multiobjective optimization for empty and foam-
Toksoy AK, Güden M (2011) The optimisation of the energy absorption filled square columns under oblique impact loading. Int J Impact
of partially Al foam-filled commercial 1050H14 and 6061T4 Al Eng 54:177–191
crash boxes. Int J Crashworth 16:97–109 Yang R, Tseng L, Nagy L, Cheng J (1994) Feasibility study of crash
Tovar A, Patel NM, Kaushik AK, Renaud JE (2007) Optimality condi- optimization. ASME 2:549–556
tions of the hybrid cellular automata for structural optimization. Yang R, Akkerman A, Anderson DF, Faruque O, Gu L (2000)
AIAA J 45:673–683 Robustness optimization for vehicular crash simulations. Comput
Tran T, Hou S, Han X, Tan W, Nguyen N (2014) Theoretical prediction Sci Eng 2:8–13
and crashworthiness optimization of multi-cell triangular tubes. Yang R, Wang N, Tho C, Bobineau J, Wang B (2005) Metamodeling
Thin-Walled Struct 82:183–195 development for vehicle frontal impact simulation. J Mech Des
Tu J, Choi KK, Park YH (1999) A new study on reliability-based design 127:1014–1020
optimization. J Mech Des 121:557–564 Yao W, Chen X, Luo W, van Tooren M, Guo J (2011) Review of
Viana FA, Haftka R, Steffen V Jr (2009a) Multiple surrogates: how cross- uncertainty-based multidisciplinary design optimization methods
validation errors can help us to obtain the best predictor. Struct for aerospace vehicles. Prog Aerosp Sci 47:450–479
Multidisc Optim 39:439–457 Yi SI, Lee JY, Park G (2012) Crashworthiness design optimization using
Viana FAC, Haftka R, Steffen V (2009b) Multiple surrogates: how cross- equivalent static loads. Proc Instit Mech Eng Part D-J Automobile
validation errors can help us to obtain the best predictor. Struct Eng 226:23–38
Multidisc Optim 39:439–457 Yildiz AR, Solanki KN (2012) Multi-objective optimization of vehicle
Viana FC, Haftka R, Watson L (2013) Efficient global optimization al- crashworthiness using a new particle swarm based approach. Int J
gorithm assisted by multiple surrogate techniques. J Glob Optim 56: Adv Manuf Technol 59:367–376
669–689 Yin H, Wen G, Gan N (2011a) Crashworthiness design for honeycomb
Viana FA, Simpson TW, Balabanov V, Toropov V (2014) Metamodeling structures under axial dynamic loading. Int J Comput Methods 8:
in multidisciplinary design optimization: how far have we really 863–877
come? AIAA J 52:670–690 Yin H, Wen G, Hou S, Chen K (2011b) Crushing analysis and
Villa A, Strano M, Mussi V (2011) Optimization of Design and multiobjective crashworthiness optimization of honeycomb-filled
Manufacturing Process of Metal Foam Filled Anti-Intrusion Bars. single and bitubular polygonal tubes. Mater Des 32:4449–4460
On design optimization for structural crashworthiness 1119

Yin H, Wen G, Hou S, Qing Q (2013) Multiobjective crashworthiness Zhang Z, Liu S, Tang Z (2009) Design optimization of cross-sectional
optimization of functionally lateral graded foam-filled tubes. Mater configuration of rib-reinforced thin-walled beam. Thin-Walled
Des 44:414–428 Struct 47:868–878
Yin H, Wen G, Fang H, Qing Q, Kong X, Xiao J, Liu Z (2014a) Zhang B, Yang J, Zhong Z (2010) Optimisation of vehicle side interior
Multiobjective crashworthiness optimization design of functionally panels for occupant safety in side impact. Int J Crashworth 15:617–
graded foam-filled tapered tube based on dynamic ensemble 623
metamodel. Mater Des 55:747–757 Zhang Y, Sun G, Li G, Luo Z, Li Q (2012) Optimization of foam-filled
Yin H, Wen G, Liu Z, Qing Q (2014b) Crashworthiness optimization bitubal structures for crashworthiness criteria. Mater Des 38:99–109
design for foam-filled multi-cell thin-walled structures. Thin- Zhang S, Zhu P, Chen W (2013a) Crashworthiness-based lightweight
Walled Struct 75:8–17 design problem via new robust design method considering two
Youn BD, Choi KK (2004a) A new response surface methodology for sources of uncertainties. Proc Instit Mech Eng Part C-J Mech Eng
reliability-based design optimization. Comput Struct 82:241–256 Sci 227:1381–1391
Youn BD, Choi KK (2004b) Selecting probabilistic approaches for Zhang S, Zhu P, Chen W, Arendt P (2013b) Concurrent treatment of
reliability-based design optimization. AIAA J 42:124–131 parametric uncertainty and metamodeling uncertainty in robust de-
Youn BD, Choi KK, Yang R, Gu L (2004) Reliability-based design op- sign. Struct Multidisc Optim 47:63–76
timization for crashworthiness of vehicle side impact. Struct Zhang X, Wen Z, Zhang H (2014a) Axial crushing and optimal design of
Multidisc Optim 26:272–283 square tubes with graded thickness. Thin-Walled Struct 84:263–274
Zabaras N, Ganapathysubramanian S, Li Q (2003) A continuum sensi- Zhang Y, Sun G, Xu X, Li G, Li Q (2014b) Multiobjective crashworthi-
tivity method for the design of multi-stage metal forming processes. ness optimization of hollow and conical tubes for multiple load
Int J Mech Sci 45:325–358 cases. Thin-Walled Struct 82:331–342
Zarei HR, Kroger M (2006) Multiobjective crashworthiness optimization
Zheng G, Wu S, Sun G, Li G, Li Q (2014) Crushing analysis of foam-
of circular aluminum tubes. Thin-Walled Struct 44:301–308
filled single and bitubal polygonal thin-walled tubes. Int J Mech Sci
Zarei HR, Kroger M (2007) Crashworthiness optimization of empty and
87:226–240
filled aluminum crash boxes. Int J Crashworth 12:255–264
Zhou XJ, Ma YZ, Li XF (2011) Ensemble of surrogates with recursive
Zarei H, Kroger M (2008a) Optimum honeycomb filled crash absorber
arithmetic average. Struct Multidisc Optim 44:651–671
design. Mater Des 29:193–204
Zhou Q, Wu X, Xia Y, Cai W (2014) Spot weld layout optimization of
Zarei HR, Kroger M (2008b) Bending behavior of empty and foam-filled
tube crash performance with manufacturing constraints. J Manufact
beams: structural optimization. Int J Impact Eng 35:521–529
Sci Eng-Trans Asme 136
Zarei HR, Kroger M (2008c) Optimization of the foam-filled aluminum
tubes for crush box application. Thin-Walled Struct 46:214–221 Zhu P, Shi YL, Zhang KZ, Lin ZQ (2008) Optimum design of an auto-
Zarei H, Kröger M, Albertsen H (2008) An experimental and numerical motive inner door panel with a tailor-welded blank structure. Proc
crashworthiness investigation of thermoplastic composite crash box- Instit Mech Eng Part D-J Automobile Eng 222:1337–1348
es. Compos Struct 85:245–257 Zhu P, Zhang Y, Chen G (2009) Metamodel-based lightweight design of
Zerpa LE, Queipo NV, Pintos S, Salager J-L (2005) An optimization an automotive front-body structure using robust optimization. Proc
methodology of alkaline–surfactant–polymer flooding processes Instit Mech Eng Part D-J Automobile Eng 223:1133–1147
using field scale numerical simulation and multiple surrogates. J Zhu P, Zhang Y, Chen G (2011) Metamodeling development for
Pet Sci Eng 47:197–208 reliability-based design optimization of automotive body structure.
Zhang Y, Zhu P, Chen G (2007a) Lightweight design of automotive front Comput Ind 62:729–741
side rail based on robust optimisation. Thin-Walled Struct 45:670– Zhu P, Pan F, Chen W, Zhang S (2012) Use of support vector regression in
676 structural optimization: application to vehicle crashworthiness de-
Zhang Y, Zhu P, Chen G, Lin Z (2007b) Study on structural lightweight sign. Math Comput Simul 86:21–31
design of automotive front side rail based on response surface meth- Zhu P, Pan F, Chen W, Viana FAC (2013) Lightweight design of vehicle
od. J Mech Des 129:553–557 parameters under crashworthiness using conservative surrogates.
Zhang Y, Zhu P, Chen G, Lin ZQ (2007c) Study on structural lightweight Comput Ind 64:280–289
design of automotive front side rail based on response surface meth- Zhu F, Jiang B, Chou CC (2016) On the development of a new design
od. J Mech Des 129:553–557 methodology for vehicle crashworthiness based on data mining the-
Zhang X, Cheng G, Wang B, Zhang H (2008) Optimum design for energy ory. SAE Technical Paper
absorption of bitubal hexagonal columns with honeycomb core. Int J Zurada JM (1992) Introduction to artificial neural systems vol 8. West
Crashworth 13:99–107 publishing company, St. Paul