Reliability Analysis and Design Optimization of Mechanical Systems under Various Uncertainties - Research Article

Advances in Mechanical Engineering

2018, Vol. 10(12) 1–13
Ó The Author(s) 2018
Optimization design for foam-filled DOI: 10.1177/1687814018811239
journals.sagepub.com/home/ade
double cylindrical tubes under multiple
lateral impacts

Ping Jiang1, Qidong Wang1,2, Andong Yin1, Jinfang Hu1

and Xianguang Gu1,2

Abstract
Nowadays, thin-walled foam-filled structures have already been excessively used in automobile industry due to the
superior energy absorption capacity and relatively light weight. The components in vehicle probably subject to lateral
impact at any position in practice; however, most of the previous literature focused only on the bending behavior of
structures under lateral impact at the mid-span. In this study, a hybrid structure of the structural epoxy foam TerocoreÒ
and two cylindrical tubes is comprehensively investigated under various lateral impact positions. The finite element
model of the hybrid structure is established and then validated by the experimental results. From a numerical study, sev-
eral design parameters, including the thicknesses of outer and inner tubes, the diameters of inner tubes, and the foam
densities, are explored to exhibit great effects on the bending resistance of the hybrid structure. To find the optimal
designs of the hybrid structure under different load cases, a system methodology, which is constructed by optimal Latin
hypercube sampling, radial basis function model and multi-objective particle swarm optimization algorithm, is implemen-
ted. Compared with the original design, the optimization designs of different load cases perform better bending resis-
tance, namely, higher specific energy absorption and lower peak crushing force. Therefore, the optimal hybrid structure
can be considered as a practical candidate for energy absorbing under lateral impact.

Keywords
Hybrid structures, multiple lateral impacts, energy absorption, optimization design, optimization algorithm

Date received: 23 April 2018; accepted: 15 October 2018

Handling Editor: MA Hariri-Ardebili

Introduction rectangle and square section tube and Wierzbicki et al.4

The increasing requirements in vehicle safety and emis-
sion have led to a series of investigations into energy-
absorbing structures with light weight. Thin-walled
tubes have been widely used as the key structural com-
ponents of the majority of transportation vehicles
because of the outstanding energy absorption (EA) effi-
ciency and light weight.1 In the real world, about 90%
of the involved structural members failed in bending
collapse mode.2 For this reason, it is practically signifi-
cant to study the bending behavior of thin-walled struc-
tures under lateral impacts. Kecman3 pioneered the
investigation of the bending collapse behavior of
extended the concept of a superfolding element, which
was developed originally for axially loaded columns, to
the case of bending and combined bending and
1
School of Automobile and Traffic Engineering, Hefei University of
Technology, Hefei, P.R. China
2
Institute of Intelligent Manufacturing Technology, Hefei University of
Technology, Hefei, P.R. China
Corresponding author:
Xianguang Gu, School of Automobile and Traffic Engineering, Hefei
University of Technology, 230009 Hefei, Anhui, P.R. China.
Email: gxghfut@163.com

Creative Commons CC BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(http://www.creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 Advances in Mechanical Engineering
compression loading. Since then, a series of experimen-
tal and theoretical studies5–14 have been performed in
this attractive field to investigate square and circular
tubes subjected to static and dynamic loads.
To further improve the bending resistance and
achieve higher EA to weight ratio, a series of research
works were conducted by filling the hollow structures
with cellular materials. The initial work of Santosa and
Wierzbicki15 showed the effect of low-density metal fil-
ler on the bending collapse resistance of thin-walled
prismatic columns and found that the low-density
metal core could increase the bending resistance. In
addition, the presence of adhesive could improve the
specific energy absorption (SEA) significantly. Sun
et al.16 carried out an experimental program including
24 tests to study the bending behavior of square alumi-
num extrusions with aluminum foam filler under quasi-
static loading conditions. The results showed that the
foam filler could alter the local deformation pattern of
the beams. The bending behavior of empty and foam-
filled square aluminum beams has been compared by
Zarei and Kröger17 using experiments and finite ele-
ment (FE) simulations. The optimization techniques
have been implemented to find the optimum foam-
filled beam that absorbs the same energy as optimum
empty tubes with lower weight.
Apart from the conventional prismatic tubes, more
foam-filled thin-walled structures with new configura-
tions were discussed in the following studies. Chen18
conducted an in-depth research of the bending collapse
of thin-walled empty and foam-filled hat profiles
experimentally and numerically and found that foam-
filled members provided a 30%–40% increase in the
SEA compared to the empty ones. Sun et al.19 com-
pared the EA ability of functionally graded thickness
(FGT) structure and uniform thickness (UT) under lat-
eral loading and carried out a multi-objective optimiza-
tion method on the FGT structure. Yin et al.20
investigated nine kinds of foam-filled multi-cell thin-
walled structures (FMTS) under lateral crushing load
conditions using the FE method and obtained the
excellent three structures by metamodel-based multi-
objective optimization approach. In order to improve
the crashworthiness of foam-filled structures while
keeping high bending resistance, a new composite
structure, namely, double cylindrical tubes filled with
closed-cell aluminum foams, was investigated under
quasi-static and dynamic three-point bending by Guo
et al.21,22 and Li et al.23 The comprehensive studies
declared that the new structures provided better SEA
efficiency compared to the traditional foam-filled single
tubes. The foam-filled double cylindrical tube can be
considered as an excellent candidate for energy absorb-
ing under the bending load.
Most of the published studies mainly focused on the
hybrid structures with aluminum foam filler, while
there was a limited concern about the epoxy foam-filled
structures. Herbst et al.24 utilized the expanding epoxies
and rigid polyurethane foams to fill the sheet metal sec-
tions. The results demonstrated that the foam core
could delay wall buckling through localized confine-
ment and the epoxy filler could enhance peak force and
EA significantly. In recent years, an epoxy-based poly-
mer foam TerocoreÒ was utilized to fill the double
cylindrical tubes due to its advantages of a relatively
simple manufacturing process by Rathnaweera et al.25
and Bilston et al.26 The quasi-static and dynamic bend-
ing performances and deformation modes of the hybrid
structures were investigated using experiments and FE
analyses. It was found that the hybrid structure with
outer tube of 6060 T5 of 2-mm wall thickness and car-
rier tube of 25-mm diameter had the largest increase in
EA and SEA of 85% and 14%, respectively. The results
also suggested that an appropriate design of the hybrid
structure could be a more mass-efficient solution. Since
the design parameters greatly affect the bending perfor-
mance of the structures, the optimal hybrid structure
remains to be explored by the optimization approach.
Yet, the vast majority of published research works
only aimed at the optimization solution of structure
under the mid-span lateral impact without considering
the variation of impact positions. However, in the real
vehicle collision accidents, the structures may rarely
subject to the lateral impact just in the mid-span. To
take into account the lateral impact with three punch
positions, a bio-inspired tube was proposed by Yin
et al.27 and was optimized using a multi-objective deter-
ministic optimization (MDO) method. The study
revealed that the bending resistance behavior might dif-
fer dramatically when the punch positions changed.
Hence, it is of critical importance to study the bending
behavior under multiple load cases in order to find a
more practical EA structure.
With the aim of addressing the above issues, an
intensive study is provided to investigate the bending
resistance of the Terocore foam-filled double cylindri-
cal tubes subjected to multiple lateral impacts. The
main advantages of the Terocore foam-filled double
cylindrical tubes over the traditionally used foam filler
(aluminum foam) are curing convenience, strong adhe-
sion, and excellent mechanical properties. After validat-
ing the FE model of the hybrid structure, a parametric
study is carried out to compare the bending perfor-
mance of hybrid structures with different configura-
tions and materials. Moreover, a multi-objective
optimization method composed of the optimal Latin
hypercube sampling (OLHS), the radial basis function
(RBF) model, and the multi-objective particle swarm
optimization (MOPSO) algorithm is adopted to seek
for the optimal design parameters for single load case
(SLC) and multiple load cases (MLC), respectively.
Jiang et al.
FE modeling and experimental validation
Crashworthiness indicators
To address the issue of evaluating the performance of
the structures under bending collapse, it is necessary to
define the crashworthiness indicators. Three indicators
are used in the following study, including EA, SEA,
and peak crushing force (PCF).
EA can be expressed as follows
ðd
EA = F(x)dx ð1Þ
0
where F(x) is the instantaneous bending force, and d
denotes the displacement of the punch.
SEA is adopted to evaluate the EA ability of the
structure by taking mass into account, which can be
written as follows
EA
SEA = ð2Þ
M accuracy of the model and minimize the computational
where M is the total mass of structure. As an EA struc-
ture, the value of SEA should be maintained at a high
level.
Note that the indicators EA and SEA are closely
related to the bending distance, which is defined as the
distance when the catastrophic failure occurs in the
outer tubes in this study.
Another important indicator PCF is the maximum
bending force during the whole loading process. In
vehicle applications, a large value of PCF may lead to
a high acceleration and serious injuries of occupants in
the process of vehicle crash. For this reason, further
effort is required to reduce the value of PCF.
Geometry and material description
The double cylindrical tubes filled with an epoxy-based
polymer foam Terocore under lateral impact are inves-
tigated as shown in Figure 1. The total length of the
structure is 340 mm. The structure is supported by two
cylindrical supports with a span of 260 mm. A cylindri-
cal punch impacts onto the structure at an initial speed
of 7.8 m/s in the mid-span. The diameters of the sup-
ports and the punch are 19.05 mm. The cross section of
the structure is plotted in Figure 2. The diameter and
thickness of the outer tube are 38 and 3.4 mm, respec-
tively, and the diameter and thickness of inner tube are
16 and 0.9 mm, respectively.
The materials of the outer and inner tubes are alumi-
num alloy 6060 T5 and 6061 T6, respectively. The foam
core is made of the structural epoxy foam Terocore.
The mechanical properties of the above-mentioned
materials are summarized in Table 1.
3
Figure 1. Schematic representation of the hybrid structures
for the dynamic three-point bending test.
FE modeling and experimental validation
The FE model of the structure is established using
explicit non-linear FE code LS-DYNA. Constant stress
solid elements are employed to model the hybrid struc-
ture, involving the two cylindrical tubes and the foam
core. In order to reduce the effect of mesh size on the
time, a mesh convergence test was conducted by Bilston
et al.26 The test found that the most suitable element
sizes for the outer tube, foam core, and inner tube were
1.5 3 1.5 3 1.5 mm3, 1.25 3 1.25 3 1.25 mm3, and
0.45 3 0.45 3 0.45 mm3, respectively. The power law
plasticity material model (*MAT_018) is adopted to
represent the aluminum alloys 6060 T5 and 6061 T6. In
order to simulate the material behavior of the Terocore
foam, the piecewise linear plasticity material model
(*MAT_024) is used. The punch and two supports are
constructed by thin shell elements and set as rigid bod-
ies (*MAT_020). The velocity of the punch is set as a
constant value 7.8 m/s.
The interfaces between the foam and two tubes are
modeled using the ‘‘tied surface to surface offset’’ con-
tacts with the friction coefficients 1.0, whereas the
‘‘automatic surface to surface smooth’’ contact is cho-
sen to simulate the contacts between the outer tube and
supports/punch. Note that for this kind of contacts, the
static and dynamic coefficients of friction are both set
to 0.5. The control card of *MAT_ADD_EROSION is
applied to simulate the material failure by deleting the
elements when they reach a defined criterion. The volu-
metric strain (ev ) is activated to model the failure of the
outer tube. Since the failure of the outer tube is the
main factor contributing to the loss of load carrying
capacity of the structure and the failure of foam will
not occur until the most of load carrying capacity of
the structure disappears, the foam failure criterion is
neglected in this study.
Bilston et al.26 have conducted a three-point bending
experiment on the Terocore foam-filled cylindrical
tubes. The structure is divided into three layers: the
inner layer and the outer layer, which are made up of
4 Advances in Mechanical Engineering

Figure 2. Schematic representation of the hybrid structure.

Table 1. Dimensions and material properties of the hybrid structure.26

Outer tube Inner tube Foam core

r Density g/cm3 2.7 2.7 0.5

sy Yield stress MPa 186 308 20a
E Young’s modulus GPa 70 70 1
v Poisson’s ratio % 0.3 0.3 0.23
k Strength coefficient MPa 275.18 369.02 –
n Hardening exponent 0.08 0.05 –
eV Volumetric strain 0.0023 N/Ab –
a
The value herein is the plateau stress of Terocore foam.
b
Failure is not observed in the experiment.

Figure 3. Deformation pattern of the hybrid structure26: (a) experiment and (b) FE analysis.

aluminum alloy, and the middle sandwich, which is a
foam material. Therefore, the result of the experiment
is capable of validating the accuracy of the FE model
established in the study. The deformation patterns of
the experiment and the simulation are compared in
Figure 3. It can be seen that the deformation modes of
simulation agree well with those of the experiment. The
slight difference of the foam between simulation and
experiment is because the material failure of foam is
ignored. As shown in Figure 4, the force–displacement
curve of the simulation has fairly good agreement with
that of the experiment. It can be found that the
Jiang et al. 5

Figure 5. Schematic representation of the hybrid structures

under different loading positions.

Figure 4. Comparisons of force–displacement curves between

experiments and FE analysis.

Table 2. Crashworthiness indicators of experiments and FE

analysis.

Experiment FE analysis Error (%)

Maximum displacement 51.6 51.5 –0.19

(mm)
PCF (kN) 20.9 20.8 –0.48
EA (J) 734.7 772.3 5.12
SEA (J/kg) 1677.3 1632.3 –2.62

FE: finite element; PCF: peak crushing force; EA: energy absorption; SEA:
specific energy absorption.

difference of PCF, EA, and the maximum displacement

Figure 6. Force–displacement curves under different impact
with respect to the experiment and simulation is less
positions.
than 6% (Table 2). To measure the accuracy of simula-
tion to the experiment results, the relative error (RE)
can be evaluated as follows Table 3. Crashworthiness indicators of the hybrid structures
under positions 1, 2, and 3.
^f (x)  f (x)
RE = ð3Þ Position 1 Position 2 Position 3
f (x)
PCF (kN) 20.9 23.7 30.4
where ^f (x) represents the simulation results and f (x) EA (J) 772.3 488.2 190.8
denotes the experiment value. Based on the above vali- SEA (J/kg) 1632.3 1031.8 403.2
dation, the FE model is accurate enough to predict the
bending behavior of the hybrid structure. FE: finite element; PCF: peak crushing force; EA: energy absorption; SEA:
specific energy absorption.
Different lateral impacts with three loading positions
are considered in this study. A series of different impact
impact positions are compared in Figure 6, and the
locations, that is, the impact at the center point (posi-
crashworthiness indicators (i.e. PCF, EA, and SEA) of
tion 1) and offsetting from the center with 50 mm (posi-
different lateral impacts are summarized in Table 3.
tion 2) and 90 mm (position 3) are tested as presented
in Figure 5. It is obvious that the impact at position 1
is equal to the initial load condition, so the validated
Parametric study
FE model is taken to simulate the structure subject to
the impact at position 1. The FE models of remaining In this section, a parametric study is carried out to
two load positions are established based on the vali- understand the relationship between the design para-
dated model. The force–displacement curves of three meters and the bending characteristics of the hybrid
6 Advances in Mechanical Engineering

Table 4. Detailed settings of the variables.

Variables Baseline Detailed settings

T1 (mm) 3.4 2.5 3.0 3.4 4.0 4.5 –

T2 (mm) 0.9 0.5 0.7 0.9 1.1 1.3 1.5
D2 (mm) 16 10 12 14 16 18 20
rf (g/cm3) 0.5 0.2 0.3 0.4 0.5 – –

Figure 7. Effects of (a) T1, (b) T2, (c) D2, and (d) rf on PCF under different impact positions.

structures under various loading positions (i.e. position Comparisons of PCF with different design
1, position 2, and position 3). Four important struc- parameters
tural and material parameters are chosen as the inde-
The relationships between the PCF and four variables
pendent variables for analysis, including the thicknesses
under different loading positions are plotted in
of outer tubes T1, the thicknesses of inner tubes T2, the
Figure 7. Overall, it is notable that the PCF increases
diameters of inner tubes D2, and the foam densities rf .
as the impact location offsets from position 1 to posi-
The details of the variables are given in Table 4. The
tion 3. As shown in Figure 7(a), PCF grows dramati-
baseline of the variables is set as T1 = 3.4 mm,
cally with the increase in T1 under all the loading
T2 = 0.9 mm, D2 = 16 mm, and rf = 0:5 g=cm3 . It
positions. As shown in Figure 7(b), the value of PCF
should be mentioned that the diameters of outer tubes
generally increases as the inner tube thickens, especially
D1 are fixed as 38 mm in the following analyses and
under position 3. The change trends of PCF with T2
optimizations.
Jiang et al.
Figure 8. Effects of (a) T1, (b) T2, (c) D2, and (d) rf on SEA under different impact positions.
and rf are presented in Figure 7(c) and (d). Compared
to the other parameters, PCF is not very sensitive to D2
and rf under all the loading positions.
Comparisons of SEA with different design
parameters
As shown in Figure 8, all the variables of the structure
play vital roles in absorbing the energy. The main fea-
tures of Figure 8 can be listed as follows:
1. As shown in Figure 8(a), when the impact
occurs in position 1, the SEA of the structure
falls off sharply with the increase in T1 when
2.5 mm \ T1 \ 3.0 mm, but goes up slowly
when 3.0 mm \ T1 \ 4.5 mm. However, the
SEA of position 2 changes irregularly when T1
increases. In the case of position 3, the SEA
declines intensively as T1 varies from 2.5 to
3.4 mm and tends to be flattened when
3.4 mm \ T1 \ 4.5 mm. It is observed that the
structure with T1 = 2.5 has an outstanding
capacity in absorbing energy under position 3.
This advantage may probably be caused by the
7
change in the collapse mode of the structure. To
visually illustrate the phenomenon, the collapse
modes of the structures with T1 = 2.5 mm and
T1 = 3.0 mm are compared in Figure 9. As for
the structure with T1 = 2.5 mm under lateral
impact at position 3, a compression fracture
occurs at the top surface of the outer tube ini-
tially before a catastrophic failure occurs.
Nevertheless, only the catastrophic failure of the
outer tube can be observed when T1 . 2.5 mm
during the whole load process.
2. It is demonstrated in Figure 8(b) that under
positions 1 and 2, the patterns of SEA variation
are fairly irregular, while for position 3, the spe-
cific EA ability of the structure is enhanced by
increasing T2.
3. As shown in Figure 8(c), the SEA of position 1
shows the tendency to increase as D2 increases
from 10 to 14 mm but declines dramatically
when 14 mm \D2 \ 20 mm. However, the SEA
of position 2 is not sensitive to the variation of
D2.
4. Figure 8(d) displays that the SEA of position 2
clearly shows a descending tendency as rf
8 Advances in Mechanical Engineering
Figure 9. Collapse mode of the hybrid structure with (a) T1 = 2.5 mm and (b) T1 = 3.0 mm under position 3.
increases. For position 3, the effects of rf on the
SEA are negligible.
Notably, the EA indicator SEA of the structure is
heavily affected by the loading position as shown in
Figure 8. On one hand, the structure with higher SEA
at the particular loading position does not necessarily
perform well when the impact position changes. On the
other hand, the relationship between design parameters
and SEA values also changes with the change in impact
position.
To conveniently compare the SEA between different
impact positions, an integrated indicator of EA (i.e.
SEAm) is introduced in this study. Given the specific
meaning of the EA indicator, the value of SEAm is
defined as the average of SEA under three lateral
impacts. The influence of four variables on SEAm is pre-
sented in Figure 10. As shown in Figure 10(a), the SEAm
drops significantly when 2.5 mm \ T1 \ 3.4 mm, but
reaches a plateau in the region 3.4 mm \ T1 \ 4.5 mm.
The structure with T1 = 2.5 mm shows a huge advantage
in absorbing the impact energy. As shown in Figure
10(b), the SEAm monotonously increases when T2 varies
from 0.5 to 1.1 mm, and the structure with T2 = 1.5 mm
presents the best SEAm performance. It can be observed
from Figure 10(c) that there is a small growth of SEAm
as D2 increases when 10 mm \ D2 \ 14 mm. While D2
varies from 14 to 20 mm, the value of SEAm declines
considerably. Figure 10(d) depicts that the lower value
of rf presents an inverse effect on the specific EA capac-
ity of the structure.
Multi-objective optimization design
Optimization problem formulation
As an EA component, the hybrid structure is supposed
to absorb as much energy as possible per unit mass.
Besides, the excessive high-impact force may lead to
severe structural deformation and occupant injury in
the traffic collision. Therefore, SEA and PCF are
defined as the objective functions to be maximized and
minimized, respectively. The parameters discussed in
section ‘‘Parametric study’’ are chosen as variables
herein, including the thickness of outer tube T1, the
thickness of inner tube T1, the diameter of the inner
tube D2, and the foam density rf , while the diameter of
the outer tube remains the same. The optimization
design is conducted under both SLC and MLC.
Ignoring the variability of impact location, the
multi-objective optimization design under SLC only
focuses on the typical three-point bending (i.e. position
1) since it is the most commonly used loading condi-
tion. The optimization problem under SLC can be
expressed mathematically as follows
8
>
> min ( SEA1 , PCF1 )
>
> s:t: 2:5 mm ł T1 ł 4:5 mm
<
0:5 mm ł T2 ł 1:5 mm ð4Þ
>
> 10 mm ł D2 ł 20 mm
>
>
:
0:2 g=cm3 ł rf ł 0:5 g=cm3
Taking into account the multiple lateral impacts
with variable loading positions, the original objectives
involving SEA and PCF should be adjusted corre-
spondingly. As mentioned in section ‘‘Parametric
study,’’ the EA indictor SEA under MLC has been set
to SEAm. Another indicator PCF is modified to PCFm
under MLC with reference to its specific meaning.
SEAm and PCFm can be calculated as follows

SEAm = (SEA1 + SEA2 + SEA3 )=3
ð5Þ
PCFm = max (PCF1 , PCF2 , PCF3 )
where SEA1, SEA2, and SEA3 are the SEAs of the
hybrid structure under three loading positions,
Jiang et al. 9

Figure 10. Effects of (a) T1, (b) T2, (c) D2, and (d) rf on SEAm under different impact positions.

respectively. Similarly, PCF1, PCF2, and PCF3 denote Surrogate models

the PCF under each position. Thus, the corresponding Surrogate model techniques have been widely employed
optimization problem under MLC can be expressed as to solve the engineering optimization problems owing
follows to the advantages of replacing the complex implicit sys-
8 tems with mathematical models.31
>
> min ( SEAm , PCFm )
> s:t: 2:5 mm ł T ł 4:5 mm
> As one of the most extensively used surrogate model
< 1
methods, RBF has shown overall advantages on
0:5 mm ł T2 ł 1:5 mm ð6Þ
>
> 10 mm ł D2 ł 20 mm
accuracies, robustness, sample size, problem types, effi-
>
>
: ciency, and simplicity. RBF, which was originally
0:2 g=cm3 ł rf ł 0:5 g=cm3
developed for scattered multivariate data interpolation,
employs a series of basis functions that are symmetric
and centered at each sample points.32 A general form
Design of experiment of RBF can be written as follows
In order to minimize the effect of sampling on the
accuracy of surrogate model, the design of experi- X
S X
K

ments (DOE) method is often adopted to generate
design samples.28 The most commonly used DOE
methods include full factorial design, orthogonal
design, and central composite design, Latin hypercube
sampling (LHS) and OLHS.29 Taking advantages of
providing space-filling and uniformly distributed sam-
pling points in the design region efficiently,30 the
OLHS is employed to construct the surrogate model
in the following optimizations.
y(x) = ls f(jjx  xs k) + Ck pk (x) ð7Þ
s=1 k =1
where S is the number of sample points, x is the vector
of design variables, xs is the vector of design variables at
the Sth sample points, ls is the unknown weighting fac-
tor at the Sth sample points, ||x – xs|| is the Euclidean
distance, K is the total number of polynomial terms,
pk(x) is the polynomial terms, Ck is the corresponding
10
coefficient for pk (x), and f is the basis function. In the
study, the RBF can be calculated as follows
c
f(jjx  xs jj) = jjx  xs jj
where c is a shape function variable between 0.2 and 3.
As shown in equation (6), it can be found that an RBF
model is a linear combination of S RBFs with an added
polynomial function.
To assess the accuracies of the established RBF
models, three numerical estimators, namely, R-square
(R2), root mean square error (RMSE), and maximum
absolute relative error (MARE), are introduced as
follows
P
m
(yi  ^yi )2
2 i=1
R =1  P
m ð9Þ
2 of objectives, the RBF models were constructed. Since
(yi  y)
i=1
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
uP
u m
u (yi  ^yi )2
ti = 1
RMSE =
m
 
yi  ^yi 
MARE = max   3 100%, i = 1, 2, . . . , m
yi 
where m is the number of validation points, yi is the true
response, ^yi is the corresponding predicted value, and y
is the mean of all yi. Generally, the higher the level of
R2, and the lower the values of RMSE and MARE, the
more accurate is the RBF model.
Optimization algorithm
In the multi-objective optimization problem, a multi-
objective optimization algorithm is required to make
trade-offs between two or more conflicting objectives
and generate a set of the non-dominated optimal solu-
tions called the Pareto frontier.33 Compared to other
multi-objective optimization algorithms, such as non-
dominated sorting genetic algorithms (NSGA) and
Pareto archived evolution strategy (PEAS), the
MOPSO34,35 has superiorities of its fast convergence
and well-distributed Pareto frontier.36 In this study,
MOPSO is implemented to optimize the hybrid struc-
tures under different positions.
Optimization procedure
The flow chart of multi-objective optimization problem
under different load cases is presented in Figure 11.
First, 50 initial samples are generated with a homoge-
neous distribution in the interior of the design space by
OLHS. Moreover, 10 validation points are produced
Advances in Mechanical Engineering
Table 5. Accuracies of the RBF surrogate models.
SLC MLC
ð8Þ SEA1 PCF1 SEAm PCFm
R2 0.951 0.989 0.973 0.960
RMSE 0.062 0.033 0.057 0.059
MARE (%) + 5.859 23.502 27.899 23.965
RBF: radial basis function; SLC: single load case; MLC: multiple load
cases; RMSE: root mean square error; MARE: maximum absolute
relative error.
for the purpose of verifying the accuracy of the surro-
gate models. Second, the FE models of the initial sam-
ples and validation points are established and analyzed
under three impact positions. Based on the responses
the prediction accuracy of the surrogate models herein
is not enough, additional 14 sampling points are chosen
to improve the accuracy of the RBF models. The error
estimators of the rebuilt RBF models are summarized
ð10Þ in Table 5. It can be found that the models are capable
of the following optimizations with sufficient accura-
cies. Finally, the adequate Pareto solutions of the opti-
mizations under different load cases are obtained by
ð11Þ MOPSO after 100 iterations.
Results and discussion
Two Pareto fronts of the PCF against SEA for the
hybrid structures under SLC and MLC are plotted in
Figure 12. First of all, the Pareto fronts directly depict
that the two objectives PCF and SEA are conflicting
with each other in all load cases. In other words,
increase in SEA may lead to increase in PCF, which is
undesirable in the real-world application.
Obviously, the Pareto fronts of different load cases
have no overlap region. The Pareto fronts of SLC and
MLC optimization locate in the left and right parts of
the graph, respectively. It is because the loading posi-
tion has a huge influence on the bending resistance of
the structures as mentioned in section ‘‘Parametric
study.’’ The great gap between the two curves reveals
that the optimal hybrid under a specific impact position
does not necessarily satisfy the design demands of the
other load cases and also verifies the importance of
applying the MLC in the optimization.
Although the Pareto fronts provide the designers
with numerous available schemes to choose, the best
compromised design point must be selected from the
Pareto solutions in terms of the specific design
requirements. For example, emphasis on the occupant
safety may lead to the sacrifice of the EA capacity. In
contrast, if the designers pay more attention to the
EA capacity of the structure, the design schemes with
Jiang et al. 11

Figure 11. Flow chart of multi-objective optimization under SLC and MLC.

Table 6. Optimal design parameters for the hybrid structures

under SLC and MLC.

Baseline SLC MLC

T1 (mm) 3.4 2.500 2.500

T2 (mm) 0.9 0.922 0.500
D2 (mm) 16 11.377 16.426
rf (g/cm3) 0.5 0.256 0.388

SLC: single load case; MLC: multiple load cases.

point) from the Pareto sets. TMDSM provides a per-

Figure 12. Pareto fronts of the hybrid structures under SLC
and MLC.
relatively high SEA should be considered first. Sun
et al.37 proposed a method called the minimum dis-
tance selection method (TMDSM) in order to deter-
mine a most satisfactory solution (namely, a knee
fect way to make a balance between the conflicting
objectives. Therefore, the TMDSM is taken to choose
the optimal designs under SLC and MLC and the
selected points are marked in Figure 12. The detailed
parameters of the optimal designs under two different
load cases are listed in Table 6. In the meantime, the
FE models based on the optimal designs are estab-
lished and tested. The corresponding responses pre-
dicted by RBF surrogate models and obtained by FE
models are given in Table 7. It can be seen that the
12
Table 7. Results of optimal designs under SLC and MLC.
Baseline
SLC SEA1 (J/kg) 1632.3
PCF1 (kN) 20.9
MLC SEAm (J/kg) 1022.5
PCFm (kN) 30.4
SLC: single load case; MLC: multiple load cases; RBF: radial basis function; FEA: finite element analysis.
errors between the FE results and the corresponding
predictions are less than 5%, demonstrating that the
accuracies of RBF models for SLC and MLC are
acceptable. As shown in Table 7, the optimization
results also show that SEA1 and SEAm are reinforced
by 57.174% and 44.457%, respectively. Compared to
the original design, PCF1 and PCFm of optimal
designs are reduced by 25.755% and 16.235%, respec-
tively. Hence, the optimized hybrid structures can be
regarded as the best configurations for energy absorp-
tion under SLC and MLC, respectively.
Conclusion
In this article, the bending resistance of the Terocore
foam-filled double cylindrical hybrid structures is
investigated by simulation analysis and optimization
design. Unlike most of the existing research works
that sorely concentrated on the lateral impact at mid-
span, different impact positions are taken into consid-
eration in the study. A parametric analysis is con-
ducted to discover the influences on SEA and PCF
with respect to the thicknesses of outer and inner
tubes, the diameters of inner tubes, as well as the
foam densities. Furthermore, the hybrid structure is
optimized under SLC and MLC using a combination
method of OLHS, RBF model, and MOPSO. The
main conclusions from the investigation can be listed
as follows:
1. The loading position has a significant effect on
the crashworthiness indicators SEA and PCF,
resulting in the optimal design which highly
depends upon the particular load case.
Therefore, the optimization result considering
different impact positions is more applicable
and practical than that under a specific loading
position.
2. The bending behavior of the hybrid structure is
strongly influenced by several design parameters
under all loading positions, including thick-
nesses of inner and outer tubes, the diameter of
inner tube, and the foam density. It calls for the
optimizations to improve the bending resistance
of the structures under SLC and MLC.
Advances in Mechanical Engineering
RBF FEA Error (%) Improvement (%)
2455.393 2565.578 24.487 + 57.174
15.854 15.407 + 2.822 225.755
1414.549 1477.003 24.415 + 44.457
24.891 25.445 22.228 216.235
3. The optimal structure under SLC improves the
SEA value by 57.174% and reduces the PCF by
25.755% than the original one. The optimiza-
tion design for MLC shows a 44.457% increase
in SEAm and a 16.235% reduction in PCFm.
For that reason, an appropriate design of the
hybrid structure can be highly recommended as
an efficient EA application to protect vehicle
occupants under lateral impacts.
This article is restricted to the double cylindrical
tubes filled with Terocore foam subjected to three-point
bending. More EA structures with various parameters,
materials, cross sections, as well as loading conditions
still need to be fully explored in the future.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
article.
Funding
The author(s) disclosed receipt of the following financial sup-
port for the research, authorship, and/or publication of this
article: This project is supported by National Natural Science
Foundation of China (51505115).
ORCID iD
Xianguang Gu https://orcid.org/0000-0002-1988-3953
References
1. Abramowicz W. Thin-walled structures as impact energy
absorbers. Thin Wall Struct 2003; 41: 91–107.
2. Kallina I, Zeidler F, Baumann K, et al. The offset crash
against a deformation barrier, a more realistic frontal
impact. In: Proceedings of the 14th international technical
conference on enhanced safety of vehicles, Washington,
DC, 23–26 May 1995, pp.1300–1304.
3. Kecman D. Bending collapse of rectangular and square
section tubes. Int J Mech Sci 1983; 25: 623–636.
4. Wierzbicki T, Recke L, Abramowicz W, et al. Stress
profiles in thin-walled prismatic columns subjected to
crash loading-II. Bending. Compos Struct 1994; 51:
625–641.
Jiang et al.
5. Sun G, Liu T, Fang J, et al. Configurational optimiza-
tion of multi-cell topologies for multiple oblique loads.
Struct Multidiscip O 2017; 57: 469–488.
6. Sun G, Liu T, Huang X, et al. Topological configuration
analysis and design for foam filled multi-cell tubes. Eng
Struct 2018; 155: 235–250.
7. Sun G, Zhang H, Fang J, et al. A new multi-objective
discrete robust optimization algorithm for engineering
design. Appl Math Model 2017; 53: 602–621.
8. Fang J, Sun G, Qiu N, et al. On design optimization for
structural crashworthiness and its state of the art. Struct
Multidiscip O 2017; 55: 1091–1119.
9. Fang J, Sun G, Qiu N, et al. Topology optimization of
multicell tubes under out-of-plane crushing using a modi-
fied artificial bee colony algorithm. J Mech Design 2017;
139: 071403.
10. Fang J, Sun G, Qiu N, et al. On hierarchical honeycombs
under out-of-plane crushing. Int J Solids Struct 2018; 135:
1–13.
11. Fang J, Gao Y, Sun G, et al. On design of multi-cell tubes
under axial and oblique impact loads. Thin Wall Struct
2015; 95: 115–126.
12. Qiu Na, Gao Y, Fang J, Feng Z, et al. Crashworthiness
analysis and design of multi-cell hexagonal columns
under multiple loading cases. Finite Elem Anal Des 2015;
104: 89–101.
13. Qiu N, Gao Y, Fang J, et al. Topological design of multi-
cell hexagonal tubes under axial and lateral loading cases
using a modified particle swarm algorithm. Appl Math
Model 2018; 53: 567–583.
14. Fang J, Gao Y, Sun G, Zhang Y, et al. Parametric analy-
sis and multiobjective optimization for functionally
graded foam-filled thin-wall tube under lateral impact.
Comput Mater Sci 2014; 90: 265–275.
15. Santosa S and Wierzbicki T. Effect of an ultralight metal
filler on the bending collapse behavior of thin-walled pris-
matic columns. Int J Mech Sci 1999; 41: 995–1019.
16. Sun G, Pang T, Zheng G, et al. On energy absorption of
functionally graded tubes under transverse loading. Int J
Mech Sci 2016; 115–116: 465–480.
17. Zarei HR and Kröger M. Bending behavior of empty and
foam-filled beams: structural optimization. Int J Impact
Eng 2008; 35: 521–529.
18. Chen W. Experimental and numerical study on bending
collapse of aluminum foam-filled hat profiles. Int J Solids
Struct 2001; 38: 7919–7944.
19. Sun G, Tian X, Fang J, et al. Dynamical bending analy-
sis and optimization design for functionally graded
thickness (FGT) tube. Int J Impact Eng 2015; 78:
128–137.
20. Yin H, Xiao Y, Wen G, et al. Multiobjective optimization
for foam-filled multi-cell thin-walled structures under lat-
eral impact. Thin Wall Struct 2015; 94: 1–12.
13
21. Guo L and Yu J. Bending behavior of aluminum foam-filled
double cylindrical tubes. Acta Mech 2011; 222: 233–244.
22. Guo L and Yu J. Dynamic bending response of double
cylindrical tubes filled with aluminum foam. Int J Impact
Eng 2011; 38: 85–94.
23. Li Z, Zheng Z, Yu J, et al. Crashworthiness of foam-filled
thin-walled circular tubes under dynamic bending. Mater
Design 2013; 52: 1058–1064.
24. Herbst B, Hock D, Meyer SE, et al. Epoxy reinforcing
for rollover safety. In: Proceedings of the AMSE interna-
tional mechanical engineering congress and exposition,
Anaheim, CA, 13–19 November 2004, pp.279–280. New
York: ASME.
25. Rathnaweera G, Ruan D, Hajj M, et al. Performance of
aluminum/TerocoreÒ hybrid structures in quasi-static
three-point bending: experimental and finite element
analysis study. Mater Design 2014; 54: 880–892.
26. Bilston D, Rathnaweera G, Ruan D, et al. Parametric
study of the bending properties of lightweight tubular
metal/polymer foam hybrid structures. Compos Part B:
Eng 2016; 105: 101–110.
27. Yin H, Xiao Y, Wen G, et al. Multi-objective robust opti-
mization of foam-filled bionic thin-walled structures. Thin
Wall Struct 2016; 109: 332–343.
28. Yu Z, Sun Z, Wang J, et al. A new Kriging-based DoE
strategy and its application to structural reliability analy-
sis. Adv Mech Eng 2018; 10: 767682.
29. Park JS. Optimal Latin-hypercube designs for computer
experiments. J Stat Plan Infer 1994; 39: 95–111.
30. Qiu N, Gao Y, Fang J, et al. Theoretical prediction and
optimization of multi-cell hexagonal tubes under axial
crushing. Thin Wall Struct 2016; 102: 111–121.
31. Ran R, Li W and Li Y. A two-level global optimization
method based on hybrid metamodel for expensive prob-
lems. Adv Mech Eng 2018; 10: 769542.
32. Hardy RL. Multiquadric equations of topography and
other irregular surfaces. J Geophys Res 1971; 76:
1905–1915.
33. Liao X, Li Q, Yang X, et al. Multi-objective optimization
for crash safety design of vehicles using stepwise regres-
sion model. Struct Multidiscip O 2008; 35: 561–569.
34. Coello CAC, Pulido GT and Lechuga MS. Handling
multiple objectives with particle swarm optimization.
IEEE T Evolut Comput 2004; 8: 256–279.
35. Zhang C, Li Q, Chen P, et al. An improved multi-
objective particle swarm optimization and its application
in raw ore dispatching. Adv Mech Eng 2018; 10: 17.
36. Yin H, Fang H, Xiao Y, et al. Multi-objective robust
optimization of foam-filled tapered multi-cell thin-walled
structures. Struct Multidiscip O 2015; 52: 1051–1067.
37. Sun G, Li G, Zhou S, et al. Crashworthiness design of
vehicle by using multi-objective robust optimization.
Struct Multidiscip O 2010; 44: 99–110.