Thin-Walled Structures 117 (2017) 165–177

Contents lists available at ScienceDirect

Thin-Walled Structures
journal homepage: www.elsevier.com/locate/tws

Full length article

Multiobjective crashworthiness optimization of thin-walled structures with MARK

functionally graded strength under oblique impact loading
⁎
Liang Yinga, , Minghua Daib, Sizhu Zhanga, Haolin Maa, Ping Hua
a
School of Automotive Engineering, Dalian University of Technology, Dalian 116024, China
b
School of Mechanical Engineering, Dalian University of Technology, Dalian 116024, China

A R T I C L E I N F O A B S T R A C T

Keywords: In this paper, the crashworthiness of a new introduced thin-walled structure made of hot stamping high strength
Functionally graded strength (FGS) columns steel with functionally graded strength (FGS), i.e. wall strength varying along the axial direction with a specific
Hot stamping gradient, is investigated. The FGS columns are comprehensively studied under both axial crushing and oblique
Oblique impact impact loading in the nonlinear FE mode LS_DYNA. The numerical simulation result shows that parameters of
Crashworthiness
gradient exponent m and top strength S of FGS columns have a remarkable effect on the crashing behavior
Multi-objective optimization
indicators such as critical load angle, energy absorption (EA) and peak crash force (PCF). To optimize the
crashworthiness of the FGS columns, multi-objective optimization based on surrogate model of Radial Basis
Function (RBF) and algorithm of Non-dominated Sorting Genetic Algorithm II (NSGA-II) are performed. To
effectively consider the load angle uncertainty effect and obtain a more robust design, four schemes are
employed to evaluate the comprehensive crashworthiness with different weight coefficient distributions. The
result shows that all the Pareto fronts of FGS columns indicate considerably better crashworthiness compared to
that of the counterpart uniform strength (US) columns. The consistent optimization result under different
evaluation schemes not only provide guidance for the FGS column design, but also declare a good robustness for
Pareto designs obtained by multi-objective optimization design (MOD) optimization. Finally, the obtained
Pareto fronts of FGS columns are obviously found to consist of two parts. The first part contains the columns that
possess gradient exponent ranging from 0 to 3 with top strength keeping a constant value near 480 MPa. The
second part consists of the columns that possess gradient exponent keeping constants close to 0 with the top
strength ranging from 700 to 950 MPa. This optimum results is different from that only obtained from pure axial
crushing analyze in the previous researches and shows a better reference for engineering practice.

1. Introduction
In recent years, road and vehicle safety becomes increasingly
important and has notably heightened legislative requirement by
introducing more effective systems to the vehicle, while a growing
concern in environment and sustainability has also largely pushed up
the lightweight standard to reduce fuel consumption. Thus, various
novel configurations of structures have been proposed and further
optimized as an energy absorber during crashes, such as the thin-walled
structures, multi-cell tubes [1–3], foam-filled structures [4–6], compo-
site structures [7–9], functionally graded structures [10–13] and
structures made of tailor-welded blank [14,15] and tailor-rolled blank,
which have been widely and deeply investigated in crashworthiness
design by using analytical, numerical and experimental methods.
As an energy absorber in crashworthiness applications like trains,
cars, ships, airplanes and other high-volume industrial products, the
thin-walled structures have been widely used to ensure crash safety due
to their lightweight, low cost and high energy absorption (EA), but also
exhaustively investigated and designed in order to achieve better
crashworthiness performance. Among those, some researchers focused
on structures with various cross-sections including circle-[16,17],
square-[18], polygonal- hat-shaped [19,20] and multi-cell cross-sec-
tions that affect crashing performance. Some other researchers tried to
improve the EA of structures by adjusting the thickness or mechanical
properties on side walls through different technologies, in order to
make full use of the material on structures. For example, Hou et al. [21]
optimized the single, double, triple and quadruple cell sectional
columns under longitudinal impact loading so as to maximize the
specific energy absorption (SEA) and minimize the peak crashing force
(PCF). Sun et al. introduced a functionally graded thickness [10,22,23]
to the longitudinal direction and cross section of single and multi-cell
tubes to obtain better crashworthiness. And results showed that all the

⁎
Correspondence to: School of Automotive Engineering, Dalian University of Technology, Linggong Road 2#, Ganjingzi District, Dalian 116024, Liaoning Province, China.
E-mail address: yingliang@dlut.edu.cn (L. Ying).

http://dx.doi.org/10.1016/j.tws.2017.04.007
Received 23 October 2016; Received in revised form 6 March 2017; Accepted 5 April 2017
Available online 26 April 2017
0263-8231/ © 2017 Elsevier Ltd. All rights reserved.
L. Ying et al.
functionally graded thickness tubes produced more competent Pareto
solutions than the conventional uniform thickness counterparts. Even
more interesting, a crash-box modeled as a deformable body in full
detail was designed and optimized by defining the shapes of the cross-
sectional and longitudinal profiles as well as the thickness as the
optimization variables, aiming at maximizing its crashworthiness [24].
In addition, along with the development of technologies such as
tailor-welded blank, tailor-rolled blank and hot stamping technology
for boron steel, researches have also paid great attention to thin-walled
structures with graded mechanical properties. Tailor-welded blank
technology consists of laser welded sheet metals with different thick-
nesses and different materials for a single part. Crashworthiness
optimization of tailor-welded blank structures [14,15,25,26] often aims
to seek the best partition of different materials and thicknesses of each
blank for both crash behaviors and lightweight. While the tailor-rolled
blank technology varies the blank thickness by a rolling process, which
has been demonstrated feasible to be applied in vehicle structures to
achieve better functional performance and reduce the mass [27]
simultaneously. Employ the hot stamping technology, Ying et al. [28]
introduced a thin-walled structure with functionally graded strength
(FGS) and demonstrated that the FGS columns could enhance the SEA
and lower the PCF concurrently. It is more competent than the
traditional uniform strength (US) columns when configured with
suitable gradient parameters.
As the hot stamping technology developed, studies have been
carried out on how to manufacture the hot stamping structure with
graded strengths. Technologies such as the stage cooling by forced air
[29], the bypass resistance heating of blank and the tailored tempering
process using heated and cooled tools [30–32] have been brought up.
Among them, the tailored tempering process is the most popular
technology that has been studied by many researchers. Based on this
process, parts with graded strengths have been successfully manufac-
tured by R. George [30,33]. And low conductivity tools were proved to
be better than that with high conductivity in consideration of the
robustness [29]. What's more, a numerical model of the tailored
tempering process was developed by B.T. Tang et al. [31]. It can use
the commercial FE code Forge™ to accurately predict the part's Vickers
hardness distribution and microstructural evolution before it is manu-
factured, so the temperature distribution on tools can be reverse
predicted according to the strength distribution on designed part.
Besides, K. Mori et al. [34,35] have developed a hat-shaped tailored
die and successfully formed the Advanced High Strength Steel parts
with gradient strength distribution using bypass resistance heating
technology, too. The above research basis has provided us confidence to
design structures with functionally graded strengths (FGS), though,
however, very limited studies on crashworthiness design of thin-walled
FGS structures have been conducted so far. Along with the development
of the tailored hot stamping technology, the optimal FGS columns can
be manufactured someday in future.
However, all the above-mentioned studies have focused on the
crushing response and energy absorption characteristics of thin-walled
structures under pure axial loads. In practical crash event, especially in
the context of automobile crashes, energy absorbers such as the side
rails rarely experience pure axial, instead, they often deform under a
combination of axial and off-axis or oblique loads. Compared to pure
axial loading, oblique loading is much more complicated because both
axial progressive and global bending deformation would happen, which
often leads to unstable reduction in energy absorption. Thus, lots of
researches have concentrated on the design and optimization of thin-
walled structures regarding oblique loading. According to requirements
in the automotive industry, the bumper system should endure a load
applied with a load angle of 30° to the longitudinal axis [36].
Investigations on thin-walled columns [37–39] subjected to oblique
loading showed that the response of columns could be divided into
three stages: axial progressive collapse, global bending collapse and a
transition zone, as load angle changes. And a critical load angle that lies
166
Thin-Walled Structures 117 (2017) 165–177
within the transition zone can be found for each column. Studies
[40,41] on straight columns also found that the crashworthiness
indicators, such as EA, peak crashing force (PCF) and mean crashing
force (MCF), drop dramatically with an increasing loading angle,
especially when the global bending collapse happened.
As a consequence, thin-walled structure, as energy-absorber, needs
to not only meet the requirements of structural collapse and decelera-
tion under axial crushing, but also needs to maintain proper crash-
worthiness under oblique impact loading. In other words, the structures
should be intended to minimize PCF, absorb maximum energy, and
generate stable folding lobes under oblique impact loading [5,39] as
well. So far, the novel developed FGS thin-walled structure has only
been investigated based on the axial crushing [28], which has been
proved to have superior performance compared to the traditional US
columns. However, it is still unknown whether it is able to maintain the
superiority when subjected to oblique impact loading under different
load angles. Therefore, the purpose of this study is to investigate the
crashworthiness of the FGS thin-walled structure under oblique impact
and seek out the optimal gradient parameters for FGS columns, aiming
to minimize the PCF and maximize the energy absorption capability.
To achieve the above purpose, numerical simulations of FGS
columns configured with different gradient exponents and top strength
subjected to oblique impact under different angles in range of 0°~40°
were carried out in LS_DYNA. To sample the design points, a full
factorial design of experiments (DoE) method was employed. When
analyze and optimize the crashworthiness performance of FGS columns
considering the load angle uncertainty, four evaluate schemes were
adopted to calculate the indicator comprehensive EA. In order to search
for the best FGS columns with optimal parameters of gradient exponent
m and top strength S, the surrogate model of Radial Basis Function
(RBF) and algorithm of algorithm of Non-dominated Sorting Genetic
Algorithm II (NSGA-II) were implemented in solving multi-objective
optimization design (MOD) problems. Based on the simulation and
optimization result, factors influencing critical load angle, PCF and EA
were analyzed, together with the comparison study between FGS
columns and US columns. Results under four different evaluation
schemes all demonstrated that the FGS columns were more preferable
to US columns.
2. Material and method
2.1. Material and geometry
It is assumed that the thin-walled columns investigated in this paper
were made of quenched boron steel 22MnB5 [42], which can obtain
different material properties when quenched at different cooling rates
[42–45]. According to the constitutive model established by A.
Bardelcik [42], a series of material's flow stress curves of USIBOR@
1500P under different Vickers hardness and strain rates can be
obtained. They had been adopted by L. Ying [28] and validated to be
feasible for numerical simulation of FGS thin-walled structure. The
detail process to obtain the flow stress curves are elaborately intro-
duced in literature [28] and the typical true stress-effective plastic
strain curves are shown in Fig. 1.
The thin-walled structure is deemed to be made of quenched
22MnB5 steel sheet with functionally graded strength along the long-
itudinal direction of the wall. The strength gradient can be defined as
[11,23]:
⎡ y ⎤m
σf (y ) = σf 1 + (σf 2 − σf 1) ⎢ ⎥
⎣L⎦ (1)
where σf 1 and σf 2 are the strengths at the top and bottom ends,
respectively, L is the total length of the square column, y is the distance
from the top end of the column, and m is the exponent to determine the
change pattern of steel strength. When m = +∞, FGS columns turn out
L. Ying et al. Thin-Walled Structures 117 (2017) 165–177

Fig. 1. Flow stress curves of quasi-static for a variety of as-quenched yield strength at
different cooling rates.

Fig. 3. Schematic of experimental set-up for dynamic tests and geometry description of
square column.

where F (δ ) is an instantaneous impact force as a function of the

displacement δ . The SEA is defined as the ratio of the absorbed energy
to the mass M of the structure, defined as SEA=EA/M. In this study, the
indicator EA instead of SEA is adopted because all the columns in this
study have the same geometric dimension and mass. In order to
evaluate the comprehensive energy absorbing characteristics and take
load angle uncertainty effect into structural EA calculation under
oblique loading, a comprehensive crashworthiness index, EAα is pro-
posed and defined as follows:
n
EAα = ∑i =1 EAαi W αi (3)

Fig. 2. Variation in strength versus normalized distance (y) with the schematic of
ascending case.
to be US ones, which can be seen as a special FGS columns. In this
study, the ascending case with σf 1 < σf 2 implying that the steel strength
increases along the length from top to bottom, as shown in Fig. 2, was
adopted for the FGS columns. Four top strengths of 480 MPa, 593 MPa,
706 MPa and 819 MPa were employed, while the bottom strength of all
the columns was fixed at 950 MPa.
The thin-walled structure used here is a straight square column,
whose cross-section has rounded corners and average dimensions of
60 mm×60 mm, wall thickness of 1.5 mm, as shown in Fig. 3 [19]. The
total length is 410 mm. The bottom part with L2=98 mm is fixed, the
top part with L1=312 mm is impacted by a rigid wall with a constant
velocity of 10 m/s. The normal of the rigid wall is in the X–Z plane and
has an oblique angle α with the axis of the square column. The added
impacting mass is 600 kg in order to imitate the real process in vehicle
crash event.
where the symbol α indicates that the EAα is a comprehensive index
involving the structural EA under multiple oblique loadings; EAαi
denotes the structural EA under the ith oblique load with angle αi ,
and W αi is the corresponding weight coefficient to this EAαi . In this
paper, five different load angles αi = 0°, 10°, 20°, 30°, 40° and four
evaluation schemes with different weight coefficient distribution, as
shown in Table 1, are employed to consider the load angle uncertainty.
Apparently, the higher the EA and EAα is, the better the energy
absorption capacity of a structure would be. Similarly, to include the
load angle uncertainty, the comprehensive PCF can be defined as:
PCFmax = Max (PCFαi, i = 1, ... ,n ) (4)
where PCFαi is the PCF when the load angle is ai , normally, the PCFmax is
usually happened in the axial impact process. As we all know, the PCF
of an absorber should be reduced and constrained to the survivable
levels. However, as an energy absorber in engineering, the structure
with higher MCF is preferred. The MCF of a column for a given
deformation can be described as:

2.2. Crashworthiness indicators under oblique loading EA

MCF =
δ (5)
Generally, there are four key crashworthiness indicators for evalu-
ating the crash behaviors of a structure, i.e. energy absorption (EA), Table 1
specific energy absorption (SEA), peak crashing force (PCF), and mean Four evaluation schemes with different weight coefficient distribution.
crashing force (MCF). Usually, an absorber is expected to absorb as
Evaluation schemes W0
°
W 10
°
W 20
°
W 30
°
W 40
°
much strain energy as possible in order to reduce the impact force. The
EA of a structure measures the capacity of absorbing impact energy,
EAα1 0.20 0.20 0.20 0.20 0.20
which can be determined mathematically as, EAα2 0.10 0.15 0.20 0.25 0.30
δ EAα3 0.30 0.25 0.20 0.15 0.10
EA = ∫0 F (δ ) dδ
(2)
EAα4 0.15 0.15 0.40 0.15 0.15

167
L. Ying et al. Thin-Walled Structures 117 (2017) 165–177

2.3. Numerical modeling and its validation

2.3.1. FE modeling
The FGS thin-walled structure subjected to oblique impact loading
here is implemented in the finite element (FE) program LS_DYNA. The
structure is modeled using Belytschko-Tsay four-node shell elements
with five integration points through thickness and six degrees of
freedom at each node. We model the FGS column in a layered structure
and each layer assumed to have the same depth. After the mesh
convergence test and comparison study, a mesh size of 3.0 mm and a
graded layered number of 26 for L1=312 mm is demonstrated to be
sufficient for numerical simulation. The oblique load exerted to the
tubes is modeled by the RIGID_WALL_GEOMETRIC_FLAT_MOTION card
in LS_DYNA. When dealing with contact condition between column and
mass block surface, contact automatic surface to surface model is
employed with the static and dynamic friction coefficient setting to 0.2
and 0.15 respectively. But frictional effect is neglected when dealing
with the self-contact of the column. The material model used here is the
same as that in literature [28]. That is, the elasto-plastic constitutive
model (MAT_24 in LS_DYNA) with material constants values C=8000
and P=0.8 is adopted. The yield stresses together with their corre-
sponding flow stress curves are shown in Fig. 1. Other mechanical
properties are: Young's modulus, E=210 GPa; Poisson's ratio, v=0.3
and density, ρ=7890 kg/m3.

2.3.2. Validation of the FE models

Since the FGS columns have not been properly available in literature
yet though it is promising to be manufactured in the future, along with
the development of the tailored hot stamping technology [29,31]. Thus,
to validate the FE models, finite element analysis (FEA) results of US
columns with uniform strength under axial dynamic loading were
compared with the theoretical solutions. The theoretical expression
form is [38,39]: Fig. 5. Comparison of numerical and experimental results [41] of square columns under
oblique crash.
MCF = 13.06σ0 b1/3t 5/3 (6)

where σ0 is the energy equivalent flow stress of column material

calculated by Eq. (7). b and t are the width and wall thickness of a
column, respectively.
σy σu
σ0 =
1+n (7)

where σy and σu are the yield strength and ultimate tensile strength,
respectively. n is the hardening coefficient of the material. For US
columns with yield strength σy = 480MPa (US480) and
σy = 950MPa (US950), the calculated theoretical MCF values under axial
dynamic crash are 60.1 kN and 108.9 kN, respectively. The numerical
and theoretical results of crash force and MCF are compared in Fig. 4.
Results in this figure show a reasonable agreement between the

Fig. 6. Flowchart of multi-objective optimization based on RBF for FGS column.

theoretical solutions and numerical simulation for both US480 and

Fig. 4. Numerical and theoretical results for US columns under axial dynamic crash.
US950 columns.
In addition to the theoretical validation described above, the
numerical FE model should be verified by comparing with the experi-
mental results, too. As described in Section 2.1, the material model of
quenched boron steel BR1500HS had been validated to be feasible after
comparing with the experiment implement by K. Omer [46], a member
from the exact research team that developed the material model [42].
So what we have to do here is to validate the availability of the FE

168
L. Ying et al. Thin-Walled Structures 117 (2017) 165–177

Table 2
Sampling design points and the summary of PCF and EA for all columns under different impact load angles.

No. Top strength Exponent m 0° 10° 20° 30° 40°`

S (MPa) PCF/KN EA/KJ PCF/KN EA/KJ PCF/KN EA/KJ PCF/KN EA/KJ PCF/KN EA/KJ

1 480 0.125 181.50 19.51 175.49 20.40 140.59 17.73 115.17 6.38 77.27 3.57
2 480 0.25 173.73 18.24 147.65 18.28 141.21 16.72 106.77 8.27 71.62 3.76
3 480 0.5 153.31 16.87 149.14 16.07 143.22 15.21 134.39 11.03 64.90 3.81
4 480 1 144.66 15.13 137.83 14.65 122.08 13.43 123.11 9.95 60.57 4.09
5 480 2 141.72 13.13 117.49 12.77 102.58 11.67 102.29 8.89 58.47 3.82
6 480 4 141.77 11.98 102.84 11.79 108.49 10.82 98.63 8.42 58.45 5.03
7 593 0.125 194.62 19.12 163.55 19.56 145.26 17.20 111.68 6.10 79.48 3.43
8 593 0.25 182.00 18.25 149.76 18.57 147.27 16.63 107.90 6.56 75.86 3.52
9 593 0.5 168.31 17.21 132.87 17.08 140.52 15.66 94.92 8.18 71.24 3.67
10 593 1 159.29 16.00 153.81 15.57 143.78 14.38 122.91 10.54 66.95 3.71
11 593 2 157.31 14.44 127.86 14.08 127.50 12.77 110.69 9.67 65.21 3.59
12 593 4 157.35 13.43 120.06 13.21 119.98 11.91 83.78 7.11 65.18 3.44
13 706 0.125 217.20 20.44 170.49 20.68 158.28 18.44 118.61 5.94 85.65 3.34
14 706 0.25 204.60 19.80 162.63 19.99 150.58 17.36 112.17 6.11 82.70 3.55
15 706 0.5 192.36 18.80 172.50 18.55 153.74 16.71 104.54 6.59 79.06 3.55
16 706 1 183.51 17.82 170.40 17.33 130.66 15.89 103.40 8.19 76.84 3.68
17 706 2 180.98 16.42 136.80 16.09 141.47 14.70 96.09 9.16 74.79 3.66
18 706 4 181.03 15.58 136.21 15.19 127.40 13.97 96.08 6.95 74.73 3.39
19 819 0.125 230.06 20.95 174.21 21.00 165.17 18.66 119.84 5.95 90.64 3.29
20 819 0.25 224.01 20.56 185.86 20.41 159.52 18.70 116.72 5.93 89.27 3.30
21 819 0.5 216.82 19.99 167.25 19.70 152.49 18.03 113.85 5.90 87.57 3.31
22 819 1 210.45 19.13 157.45 18.96 146.52 17.14 110.43 6.04 86.04 3.41
23 819 2 209.95 18.49 156.32 18.30 147.71 16.39 109.17 5.98 85.18 3.38
24 819 4 209.98 17.94 156.31 17.68 149.95 16.16 109.15 5.79 85.15 3.23
25 480 +∞ 142.15 11.47 104.91 11.31 97.49 10.50 80.84 5.32 59.60 2.60
26 593 +∞ 157.40 12.75 116.48 12.47 110.06 11.53 83.71 5.69 64.96 2.69
27 706 +∞ 181.08 14.66 135.93 14.55 127.41 13.41 96.00 5.28 74.59 2.85
28 819 +∞ 210.02 17.60 156.05 17.32 149.18 16.45 109.06 5.42 85.08 2.95
29 950 +∞ 249.21 22.15 180.97 21.28 168.61 19.36 124.28 5.63 94.84 3.66

Columns of 1–24 are FGS ones, while the others are US ones.

Table 3
The summary of PCFmax and EAα of all columns considering load angle uncertainty.

No. Top strength Exponent m PCFmax /KN EAα1/KJ EAα2 /KJ EAα3 /KJ EAα4 /KJ
S (MPa)

1 480 0.125 181.50 13.52 11.22 15.81 14.57

2 480 0.25 173.73 13.05 11.11 15.00 13.97
3 480 0.5 153.31 12.60 11.04 14.16 13.25
4 480 1 144.66 11.45 10.11 12.79 11.94
5 480 2 141.72 10.06 8.93 11.18 10.46
6 480 4 141.77 9.61 8.74 10.47 9.91
7 593 0.125 194.62 13.08 10.84 15.32 14.11
8 593 0.25 182.00 12.70 10.63 14.78 13.69
9 593 0.5 168.31 12.36 10.56 14.16 13.18
10 593 1 159.29 12.04 10.56 13.52 12.62
11 593 2 157.31 10.91 9.60 12.21 11.37
12 593 4 157.35 9.82 8.52 11.12 10.34
13 706 0.125 217.20 13.77 11.32 16.22 14.94
14 706 0.25 204.60 13.36 11.04 15.68 14.36
15 706 0.5 192.36 12.84 10.72 14.96 13.81
16 706 1 183.51 12.58 10.71 14.45 13.41
17 706 2 180.98 12.01 10.38 13.63 12.68
18 706 4 181.03 11.02 9.39 12.65 11.76
19 819 0.125 230.06 13.97 11.45 16.49 15.14
20 819 0.25 224.01 13.78 11.33 16.23 15.01
21 819 0.5 216.82 13.53 11.17 15.88 14.83
22 819 1 210.45 12.94 10.72 15.15 13.99
23 819 2 209.95 12.51 10.38 14.64 13.48
24 819 4 209.98 12.16 10.10 14.23 13.16
25 480 +∞ 142.15 8.24 7.05 9.43 8.81
26 593 +∞ 157.40 9.03 7.68 10.37 9.65
27 706 +∞ 181.08 10.15 8.50 11.79 10.96
28 819 +∞ 210.02 11.95 9.89 14.01 13.07
29 950 +∞ 249.21 14.42 11.79 17.05 15.65

Columns of 1–24 are FGS ones, while the others are US ones.

169
L. Ying et al. Thin-Walled Structures 117 (2017) 165–177

Table 4 ⎧ Min − EAαj (m, S ), PCFmax (m, S )

Accuracy of RBF surrogate model for FGS and US columns under four different evaluation ⎨
⎩ s.t. 480MPa ≤ S ≤ 950MPa, j = 1, 2, 3, 4 (9)
schemes.

Columns Evaluation R2 RAE RME

To search for the best FGS structure with optimal strength gradient
exponent and top strength, i.e. seek the Pareto designs for the
Scheme PCFmax EAα PCFmax EAα PCFmax EAα formulated MOD problems, the surrogate model of RBF [47,48] and
algorithm of NSGA-II [49,50] are applied. The flow chart of the NSGA-II
FGS 1 0.9954 0.9695 0.0212 0.0469 0.033 0.1208
2 0.9563 0.9578 0.0508 0.0509 0.1466 0.1438
algorithm is shown in Fig. 6. In multi-optimization, 24 points, the
3 0.9559 0.9516 0.0514 0.0468 0.1273 0.1422 former 24 sampling points as shown in Table 2, representing for FGS
4 0.9854 0.9861 0.0313 0.0284 0.0896 0.0588 columns were selected by using a full factorial DoE method with
US 1 0.9955 0.9980 0.0240 0.0160 0.0198 0.0143 different gradient exponent m and top strength S.
2 0.9955 0.9975 0.0198 0.0175 0.0240 0.0181 To increase the accuracy of the optimization, it is necessary to
3 0.9955 0.9978 0.0198 0.0137 0.0240 0.0240 increase the number of sampling points, but an increasing of sample
4 0.9955 0.9848 0.0198 0.0368 0.0240 0.0446 points would sacrifice the calculation efficiency. Thus, before the
optimization, criteria to verify the accuracy of the RBF models should
be calculated. For instance, the R square values R2 , the relative average
modeling method for thin-walled structures under oblique impact
error (RAE) and relative maximum error (RME). Their normalized
loading. However, as dynamic precision tests of oblique loading are
equations can be described as
quite difficult to accomplish, quasi-static tests are usually carried out to
validate a numerical model. Thus, the typical square column under ∑ (yi − yˆi )2
oblique loading experiment with loading angles of 5°, 15° and 30 °con- R2 = 1 −
∑ (yi − yi )2 (10)
ducted by A. Reyes et al. [41] was numerically modeled here in
software HyperMesh and LS_DYNA. The square column is made of ∑ yi − yˆi
aluminum AA6060-T4. Its length and width are constant, 199 mm and RAE =
∑ yi − yi (11)
80 mm, respectively. Wall thickness of 1.9 mm and 2.5 mm is used. The
load is applied at the upper end of the specimen, through a rigid body
max yi − yˆi
modeled with shell elements. The length of rigid body is 70 mm. All the RME = n
∑i =1 yi − yi / n (12)
degrees of freedom are fixed at the lower end, while the upper end is
fixed to the rigid body. The geometric and material parameters of the where ŷi and yi are the corresponding predicted and mean values
square columns as well as the loading and boundary conditions are all respectively for the exact function value yi at each checking point i, n
from Ref. [41]. The crash force is the sum of the component force represents the number of checking points. In general, for a good fitting
extracted from all the nodes in column's bottom end, to keep consistent model, the R2 value should be close to 1.0 while the RAE and RME
with the force recorded in the experiment. Fig. 5(a) compares the crash should be as small as possible.
force versus deformation with initial imperfections, together with the
final deformed profile of the present FE models and experiment result.
Fig. 5(b) shows the correlation between PCF and MCF of 1.9 mm and 3. Results and discussion
2.5 mm square columns from the experiments and simulation analysis
with initial imperfections. As one can see, there is an acceptable 3.1. Sampling design points and the accuracy of surrogate model
agreement between the numerical and experimental results for the
crash loads. This acceptable validation result demonstrated that the FE To investigate the effect of parameters on FGS columns under
modeling method adopted here was feasible for further investigation on oblique impact loading, a full factorial DoE method was selected to
crashworthiness design of square columns. sample design points. As shown in Table 2, the EA is calculated up to a
deformation of 200 mm, an effective length that taking up 62.4% of the
total length. Within this distance, a sufficient deformation has occurred
2.4. Multi-objective optimization models while the maximum crash force was extracted as PCF for each column.
To consider the effect of load angle uncertainty, a simplified
FGS tubes should be designed for good crashworthiness perfor- approach [38] is proposed here, which is easy to be implemented in
mance under all possible load angles when subjected to oblique engineering applications. In this approach, a limited number of
impacts. Thus, the two crashworthiness indicators, EAα and PCFmax , representative load angles are first selected and a specified weighting
as introduced above will be included in the multi-objective optimiza- factor is assigned to each representative load angle to account for the
tion design (MOD) process and taken as the optimization objectives. probability of the load angle around it. Distribution of the load angle in
The strength gradient exponent m and top strength S of the FGS the defined range can be then realized by adjusting the weighting
columns are chosen as the two design variables, while the column's factors. The exact calculation method is defined by Eq. (3) and four
bottom strength and geometrical dimension remain the same in the evaluation schemes with different weighting factors adopted in this
design process. Therefore, the MOD problem of the FGS columns can be study are listed in Table 1. Thus, to provide a more comprehensive
formulated as follows: analysis and more robust design in the MOD problem, the load angle
uncertainty is also considered in the design process.
⎧ Min − EAαj (m, S ), PCFmax (m, S ) Based on the calculation method in Eq. (3) and the data list in
⎪
⎪ s.t.
⎪ 0 ≤ m ≤ 10, 480MPa ≤ S ≤ 950MPa Table 1 and Table 2, crashworthiness indices under four different
⎨ − EAαj (m, S ) ≤ −EAαj 480MPa evaluation schemes of all columns are calculated and summarized in
⎪ Table 3. It establishes the database for multi-objective optimization.
⎪ PCFmax (m, S ) ≤ PCFmax 950MPa
⎪
⎩ j = 1, 2, 3, 4 (8) According to Eqs. (10)–(12), the accuracy of the established RBF
surrogate model can be determined. The value of R2 , RAE and RME of
where EAαj480MPa and PCFmax 950MPa are constants obtained from columns PCFmax and EAα are calculated and summarized in Table 4. As can be
of US480 and US950 under j evaluation schemes, respectively. For the seen in this table, all the obtained R2 values are above 0.95, RAE values
US columns, only the material strength is taken as variable, so the MOD are below 0.1 and RME values are below 0.2, which indicate adequate
problem can be formulated as: accuracy.

170
L. Ying et al.
Fig. 7. Effect of load angle on the crushing force-deformation response of different FGS columns.
3.2. Parametric study
Based on the DoE results, FE models of columns representing the
sampling design points are created and used for the crashworthiness
analyses with an impact velocity of 10 m/s. From the FEA results, the
response functions of each FGS columns in terms of EAα and PCFmax are
determined. In the following study, the parameters of gradient ex-
ponent m and top strength S are assumed to be independent parameters,
which are decoupled in affecting the columns’ crash responses.
3.2.1. Effect of gradient exponent on critical load angle under oblique
impact loading
Fig. 7 shows the effect of load angle on the crushing force-
deformation response for FGS columns with different gradient exponent
171
Thin-Walled Structures 117 (2017) 165–177
and constant top strength of 480 MPa. In these figures, the deformation
equals the vertical displacement of the rigid wall with various load
angle a , as depicted in Fig. 3. It can be found from Fig. 7 that the load
angle α in oblique impact has a remarkable effect on the FGS columns.
Firstly, the PCF reduces in magnitude as load angle increases. This is
because only the PCF of columns loaded axially, i.e.a = 0°, occurs at the
beginning of the crushing, while others loaded under a certain angle
such as 10°, 20°, 30° and 40° have no predominate initial PCF. So it
leads to a significant PCF drop as response. Secondly, as shown in
Fig. 7(a), when the load angle a ≤ 20° , the impact force of the columns
is always stay in a high level and fluctuated with increasing deforma-
tion. But when a ≥ 30°, the impact force with big magnitude and
relative small fluctuations only happens at the initiation range of the
deformation and then gradually decreases to zero. Combined with the
L. Ying et al.
Fig. 8. Different collapse mode of FGS columns under oblique impact: (a) global
symmetric collapse mode, (b) global bending collapse mode.
analysis of the final stage deformation mode as illustrated in Fig. 8,
columns with a load angle a ≤ 20° were found to deform by progressive
buckling at the impacted end, which was called global symmetric
collapse mode [51]. However, when the load angle a ≥ 30°, columns
showed a tendency to deform in a global bending collapse mode [52],
i.e. the tension flange buckles at the impacted end initially, after which
a plastic hinge forms on the compression flange close to the fixed
bottom end. Obviously, the former collapse mode is preferred due to its
higher energy absorption while the latter should be avoided in crash as
much as possible. Thus, there exists a critical load angle that makes a
transition of the deformation mode from progressive to global bending
collapse. For the column depicted in Fig. 7(a), it should be in the
transition range of 20°~30°.
A similar mode has also been observed on FGS columns with
different gradient exponent as shown in Fig. 7(b), (c), (d). It can be
observed from Fig. 7 that the critical load angle range show an
increasing tendency as gradient exponent increases from 0.125 to 4.
However, the critical load angle is still an unknown variable and cannot
be expressed as one value because the continuous load angles cannot be
Fig. 9. Final deformed profiles of different FGS columns under oblique impact in their critical load angles.
172
Thin-Walled Structures 117 (2017) 165–177
used for numerical simulations. Therefore, to further examine the
critical load angle for these six columns, the range of the critical angle,
which contains the critical load angle, was investigated with a 1°
increment of the load angle. And the upper limit of the obtained
smallest range was assumed as the critical load angle in this study.
Result found that the critical load angle for FGS columns with
m=0.125, 0.25, 0.5, 1, 4 and m = +∞ were 27°,28°,32°,37°,33°and
27°, respectively. Their corresponding final deformation modes are
shown in Fig. 9. Obviously, with the increasing of exponent m, the
critical load angle will rise to the highest value until m reaches 1, then
decreases with the further increasing of m. The reason is, for the FGS
columns, the critical load angle is affected by both the strength of the
last plastic hinge layer on the upper part and the bending layer on the
bottom part. After mechanical analysis [37], it was found that a
relatively lower strength ratio in these two layers will result in bigger
critical load angle and bigger distance between these two layers, which
are in consistent with the result as shown in Fig. 9. Whatever, all the
FGS columns have critical load angles no smaller than that of the
counterpart US columns, which indicates a better crashworthiness
under oblique impact loading for FGS columns.
3.2.2. Effect of gradient exponent and top strength of FGS column on energy
absorption
Firstly, the effect of strength gradient exponent and load angle on
EA of FGS columns under oblique impact is shown in Fig. 10. For each
column, five load angles were applied and the load angle was found to
have a remarkable effect on EA of the FGS columns. As load angle
increases, the EA value showed a tendency to decrease for all the FGS
columns, especial during the load angle change from 20° to 40°. But for
columns with higher gradient exponent m, the effect of load angle on
EA was not so obvious, which indicate a better energy absorption
stability for columns when facing impact loading from different
directions.
Compare FGS columns with different top strength to each other, it's
easy to find that the gradient exponent affect the columns’ energy
absorption a lot, too. When the load angle α ≤ 20°, the EA value
decreased gradually with an increasing gradient exponent. But when
the load angle α ≥ 30° , the regulation was disrupted. The reason is that,
when the load angle α ≤ 20° , all the FGS columns deformed in a global
symmetric collapse mode, which is a good energy absorption mode. So
L. Ying et al.
Fig. 10. Effect of strength gradient exponent and load angle on EA of FGS columns under oblique impact: Top strength is (a) 480 MPa; (b) 593 MPa; (c) 706 MPa; (d) 819 MPa.
Fig. 11. Effect of top strength on comprehensive EA of FGS columns under oblique
impact.
FGS column with smaller gradient exponent m means greater equiva-
lent average strength on column and results in higher EA during the
progressive buckling process. However, when the load angle α ≥ 30°,
some of the FGS columns deformed in a global bending collapse mode,
which was a bad deformation mode that would result in a sharp energy
absorption drop in column. Thus, it turns up the different phenomenon
that FGS columns with smaller gradient exponent m have lower EA
value in constant, for instance, the FGS480 columns with m=0.125 and
m=0.25, as shown in Fig. 10(a). The collapse mode of these two
columns was just in consistent with the result analyzed in Section 3.2.1.
In addition, observed from the four figures as shown in Fig. 10, the
effect of gradient exponent becomes smaller and smaller as top strength
increases.
173
Thin-Walled Structures 117 (2017) 165–177
Fig. 12. Response surface based on RBF approximation method for -EAα1 of FGS columns
under oblique impact.
Secondly, the top strength of the FGS columns is found to have a
remarkable effect on EA, too. Given the load angle uncertainty, the
indicator EAα1 is adopted here. As shown in Fig. 11, for FGS columns
with a certain gradient exponent, there is a rising tendency for EAα1
when the top strength increases. This is easy to understand, since a
higher top strength leads to a higher equivalent average strength on
FGS column, so it can absorb more energy during the impact process in
general. However, there are several special cases that are not matched
well with the rising tendency as well due to the difference in collapse
mode under different load angles.
In addition, it is clearly seen from Fig. 11 that the bigger the
gradient exponent is, the more obvious the rising trend shown. When m
L. Ying et al.
Fig. 13. Effect of gradient exponent m on PCF of FGS columns under oblique impact.
Fig. 14. Effect of gradient exponent and top strength on the PCFmax of FGS columns under
oblique impact.
Fig. 15. Response surface based on RBF approximation method for PCFmax of FGS
columns under oblique impact.
is no more than 1.0, the curves’ strength-dependent variation is no
longer obvious. By analyzing the strength gradient curve shown in
Fig. 2, the strength gradient curves with gradient exponent no more
than 1.0 are all convex, while the others are concave. Comparing to the
concave curves, the convex strength gradient curves with the same
gradient exponent are close to each other as the top strength changes. It
implies a relatively small difference in equivalent strength for such a set
of FGS columns. Finally, compared to the US columns, the superiority of
the FGS tubes gradually decrease as the top strength increases.
174
Thin-Walled Structures 117 (2017) 165–177
Fig. 16. Comparison of -EAα1 and PCFmax between FGS and US columns.
Fig. 17. Comparison of the Pareto fronts between FGS and US columns under four
different evaluation schemes.
To view the effect of both gradient exponent and top strength
synthetically, the constructed response surface based on RBF approx-
imation method for EAα1 and PCFmax prediction is depicted in Fig. 12. It
is found that the EAα1 of FGS columns increases along with the
decreasing gradient exponent and increasing top strength. FGS columns
with the highest top strength and the smallest strength gradient
exponent have the best energy absorption behavior when subjected to
oblique impact loading.
3.2.3. Effect of gradient exponent and top strength of FGS columns on PCF
Fig. 13 shows that the PCF value of FGS columns reduces in general
when the load angle increases due to the decreasing component force
along the longitudinal direction of columns. However, special cases
exist in the columns with gradient exponent m bigger than 0.5, either.
For instance, FGS column with m=1 obtained higher PCF value
(123.11 kN) when load angle is 30° than that value (122.08 kN) when
load angle is 20°. They have even bigger PCF when load angle is bigger.
This phenomenon also relates to the deformation mode of the exact
columns as analyzed for the special cases in Section 3.2.2, which will be
omitted here for the sake of brevity. Moreover, the downtrend of PCF
becomes less significant as the gradient exponent m increases, which
imply a better robustness for FGS columns with higher gradient
exponent.
Fig. 14 shows both the effect of gradient exponent and top strength
on the maximum PCF of the FGS columns. Both the gradient exponent
and top strength have a significant effect on PCFmax , i.e. the PCF when
load angle a = 0° as illustrated in Fig. 13. For columns under a certain
gradient exponent, the PCFmax increases along with the increasing top
L. Ying et al. Thin-Walled Structures 117 (2017) 165–177

Fig. 18. 3D Solution space and Pareto front of MOD problem for FGS columns in terms of (a)-EAα1 and (b) PCFmax.

the optimal FGS columns were found to be conflicting in terms of the

two indicators of EA and PCF. Thus, a multi-objective optimization
design should be conducted to obtain simultaneous optimum columns,
which will be further discussed in a later section.

3.3. Comparative study

3.3.1. Comparison study of the crashworthiness between FGS and US

Fig. 19. Pareto fronts of the FGS columns under the four evaluation schemes with load
angle uncertainty.
strength obviously. And it decreases as the exponent m increases for a
certain top strength. This is because the increasing top strength and
decreasing exponent m is equivalent to reducing the column's stiffness,
especial the stiffness at the top impact part. However, when m becomes
bigger than 1.0, the PCFmax ceases to decrease, since the exponent m can
hardly affect the strength of a column within a certain distance from the
impact end. Thus, when FGS columns subjected to oblique impact
under different load angles, the best peak crash force behavior could be
obtained with a lowest top strength and the gradient exponent above a
certain extent, for example 2.0, as depicted in Fig. 15.
Since a desired thin-walled structure is supposed to have lower PCF
and higher EA simultaneously, based on the above parametric study,
columns
As analyzed in Section 3.2, the FGS columns were found to be better
than the US ones in terms of critical load angle, while the US columns
were found to have better robustness when subjected to oblique impact
under different load angles as illustrated in Fig. 10 and Fig. 13.
To compare the crashworthiness of FGS and US columns visually,
both the EAα1 and PCFmax data for different columns are presented in
Fig. 16. Apparently, the point representing columns located in the left
bottom are supposed to have better crashworthiness. US columns with
five different uniform strength, as labeled along the curve next to the
black hollow square, and FGS columns with four different top strength
and six different gradient exponent (m=0.125, 0.25, 0.5, 1, 2, 4) are
presented. It is observed that the EAα1 is nearly in line with the PCFmax
for US columns as the strength changes. For FGS columns, the relation-
ship is obviously different, but all the points representing FGS columns
have better crash performance than the US ones due to the double-
direction change of top strength and gradient exponent. This is the
exact reason why FGS columns can obtain much better crashworthiness.
Moreover, FGS columns with lower top strength are seem to have even
better crashworthiness.
3.3.2. Comparison study of the Pareto fronts between FGS and US columns
Given the practical condition, it is necessary to take the load angle
uncertainty into optimization process of such energy absorbers to

Table 5
Ideal optimums of the two single objective functions for FGS columns with different evaluation schemes.

Evaluation Scheme Single objective Gradient exponent/m Top strength/S (MPa) PCFmax（KN） − EAα (KJ)

1 Ideal min. PCF 2.899E+00 480.000 141.286 −9.715

Ideal max. SEA 5.901E-04 938.665 249.027 −14.388

2 Ideal min. PCF 2.936E+00 480.002 141.280 −8.723

Ideal max. SEA 4.042E-06 947.904 249.158 −11.591

3 Ideal min. PCF 2.976E+00 480.007 141.288 −10.685

Ideal max. SEA 9.657E-05 939.762 249.182 −17.181

4 Ideal min. PCF 2.901E+00 480.002 141.286 −10.060

Ideal max. SEA 4.659E-05 938.803 249.053 −15.725

175
L. Ying et al.
obtain a more robust design under oblique impact loading. Thus, all the
four evaluation schemes with different weight coefficient distribution
are employed here in the optimization process.
Depending on the procedure illustrated in Fig. 6, the generated
models are calculated through NSGA-Ⅱmethod to solve the MOD
problems of FGS and US columns. Based on a population size of 40
and iterated for 50 generations, Pareto fronts of the FGS and US
columns under four different evaluation schemes are obtained and
compared in Fig. 17. It is important to note that, all the Pareto fronts of
FGS columns indicate considerably better crashworthiness compared to
that of the US ones under oblique impact loading no matter what the
evaluation scheme is.
3.4. Optimization results of the FGS columns
For a deeper analysis, the 3D solution space and Pareto front of
MOD problem for FGS columns in terms of EAα1 and PCFmax are shown
in Fig. 18, in which the Pareto designs are projected onto three planes
in different colors simultaneously. Obviously, the Pareto designs consist
of two parts. The first part with -EAα1 ranging from −14 KJ to −10 KJ
are the columns that possess gradient exponent ranging from 0 to 3 and
the top strength keeping constant in 480 MPa. The other part that have
even higher energy absorption capabilities are the columns that possess
gradient exponent keeping constant close to 0 with the top strength
ranging from about 700 MPa to 950 MPa. The above optimal designs
are obviously different from the result obtained by pure axial crushing
as deliberated in literature [28]. So it is necessary to optimal design the
FGS columns under both axial crushing and oblique impact. Moreover,
it also can be found from the green curves projected onto the back
planes that both EAα1 and PCF decrease with the increasing gradient
exponent, indicating that the EAα1 and PCF conflict with each other,
since a desired structure is supposed to have lower PCF and higher EA.
To investigate the robustness of Pareto designs, four Pareto front
curves under four evaluation schemes are compared and shown in
Fig. 19. The ideal optimum designs for single objective optimizations,
PCFmax and the comprehensive EA, are marked in hollow magenta star
in Fig. 19, while the value is given in Table 5. It is clearly shown that
concave Pareto front curves exist for all the four evaluation schemes,
which means the trade-off relationship between EAα and PCFmax .
Results in Table 5 showed that the columns to obtain minimum PCF
under different evaluation schemes nearly had the consistent gradient
parameters m and top strength S (m≈2.9, S≈480 MPa), while the
columns to obtain maximum EA should have the similar rather small
gradient exponent with rather big top strength around 940 MPa. Thus,
the consistent optimization result under different evaluation schemes
declares a good robustness for the Pareto designs obtained by the MOD
optimization.
4. Conclusions
In this study, the crashworthiness of FGS and US columns with
different configurations was investigated under oblique impact loading.
Numerical simulation models were established in nonlinear FE code
LS_DYNA and then validated by both theoretical and experimental
methods. Based on the parametric study, the parameters of gradient
exponent m and top strength S of FGS columns were found to have a
remarkable effect on the crashing performance. Firstly, with the
increase of exponent m, the critical load angle rises to the highest
value until m reaches 1, then decreases with the further increase of m.
And all the FGS columns have critical load angles no smaller than that
of the counterpart US columns, indicating a better crashworthiness
under oblique impact loading for FGS columns. Secondly, the FGS
columns with the smallest strength gradient exponent and the highest
top strength were found to have the best energy absorption behavior.
When FGS columns subjected to oblique impact under different load
angles, the best peak crash force behavior could be obtained with a
176
Thin-Walled Structures 117 (2017) 165–177
lowest top strength and the gradient exponent above a certain extent,
such as 2.0. Since a desired thin-walled structure is supposed to have
lower PCF and higher EA simultaneously, the optimal FGS columns
were found to be conflicting in terms of the two indicators of EA and
PCF, Thus, a multi-objective optimization design should be conducted
to obtain simultaneous optimum columns.
Based on the comparative study, it is observed that all the FGS
columns have better crash performance than the US ones due to the
double-direction change of top strength and gradient exponent on FGS
columns. The FGS columns with lower top strength are seem to have
even better crashworthiness. To consider the effect of load angle
uncertainty into optimization process and obtain a more robust design,
four evaluation schemes with different weight coefficient distributions
were employed in the MOD process. The result shows that all the Pareto
fronts of FGS columns indicate considerably better crashworthiness
compared to the US columns under oblique impact loading. The
consistent optimization result under four different evaluation schemes
declares a good robustness for the Pareto designs obtained by the MOD
optimization. Moreover, the obtained Pareto fronts of FGS columns
obviously consist of two parts. The first part contains the columns that
possess gradient exponent ranging from 0 to 3 with top strength
keeping a constant value near 480 MPa. The second part contains the
columns that possess gradient exponent keeping constants close to 0
with the top strength ranging from 700 MPa to 950 MPa. The study in
this work under oblique impact presents different optimum results
compared to that only obtained from pure axial crushing analyze in the
previous researches, which demonstrates the necessary to optimal
design the FGS thin-walled columns under oblique impact.
Acknowledgement
This work is funded by the Key Project of the National Natural
Science Foundation of China (No. 11272075), China's Post-doctoral
Science Fund (2014M561223) and Basic Scientific Research Fund of the
Central University (DUT16RC(4)28).
References
[1] X. Zhang, G. Cheng, A comparative study of energy absorption characteristics of
foam-filled and multi-cell square columns, Int. J. Impact Eng. 34 (11) (2007)
1739–1752.
[2] G. Sun, T. Pang, J. Fang, et al., Parameterization of criss-cross configurations for
multiobjective crashworthiness optimization, Int. J. Mech. Sci. 124–125 (5) (2017)
145–157.
[3] S. Tabacu, Analysis of circular tubes with rectangular multi-cell insert under
oblique impact loads, Thin-Walled Struct. 106 (2016) 129–147.
[4] X. An, Y. Gao, J. Fang, et al., Crashworthiness design for foam-filled thin-walled
structures with functionally lateral graded thickness sheets, Thin-Walled Struct. 91
(2015) 63–71.
[5] Q. Gao, L. Wang, Y. Wang, et al., Crushing analysis and multiobjective crash-
worthiness optimization of foam-filled ellipse tubes under oblique impact loading,
Thin-Walled Struct. 100 (2016) 105–112.
[6] H. Yin, G. Wen, S. Hou, et al., Multiobjective crashworthiness optimization of
functionally lateral graded foam-filled tubes, Mater. Des. 44 (0) (2013) 414–428.
[7] Y. Zhang, M. Lu, G. Sun, et al., On functionally graded composite structures for
crashworthiness, Compos. Struct. 132 (2015) 393–405.
[8] M.R. Bambach, H.H. Jama, M. Elchalakani, Static and dynamic axial crushing of
spot-welded thin-walled composite steel–CFRP square tubes, Int. J. Impact Eng. 36
(9) (2009) 1083–1094.
[9] M. Vesenjak, Z. Ren, T. Fiedler, et al., Impact behavior of composite hollow sphere
structures, J. Compos. Mater. 43 (22) (2009) 2491–2505.
[10] G. Li, F. Xu, G. Sun, et al., A comparative study on thin-walled structures with
functionally graded thickness (FGT) and tapered tubes withstanding oblique impact
loading, Int. J. Impact Eng. 77 (2015) 68–83.
[11] G. Sun, G. Li, S. Hou, et al., Crashworthiness design for functionally graded foam-
filled thin-walled structures, Mater. Sci. Eng.: A. 527 (7–8) (2010) 1911–1919.
[12] L. Cui, S. Kiernan, M.D. Gilchrist, Designing the energy absorption capacity of
functionally graded foam materials, Mater. Sci. Eng.: A 507 (1–2) (2009) 215–225.
[13] M.S. Attia, S.A. Meguid, H. Nouraei, Nonlinear finite element analysis of the crush
behaviour of functionally graded foam-filled columns, Finite Elem. Anal. Des. 61
(2012) 50–59.
[14] B. Tang, Z. Yuan, G. Cheng, et al., Experimental verification of tailor welded joining
partners for hot stamping and analytical modeling of TWBs rheological constitutive
in austenitic state, Mater. Sci. Eng.: A. (2013).
L. Ying et al.
[15] F. Xu, G. Sun, G. Li, et al., Experimental study on crashworthiness of tailor-welded
blank (TWB) thin-walled high-strength steel (HSS) tubular structures, Thin-Walled
Struct. 74 (0) (2014) 12–27.
[16] W. Abramowicz, N. Jones, Dynamic progressive buckling of circular and square
tubes, Int. J. Impact Eng. 4 (4) (1986) 243–270.
[17] C.P. Gameiro, J. Cirne, Dynamic axial crushing of short to long circular aluminium
tubes with agglomerate cork filler, Int. J. Mech. Sci. 49 (9) (2007) 1029–1037.
[18] D. Karagiozova, G.N. Nurick, S. Chung Kim Yuen, Energy absorption of aluminium
alloy circular and square tubes under an axial explosive load, Thin-Walled Struct.
43 (6) (2005) 956–982.
[19] V. Tarigopula, M. Langseth, O.S. Hopperstad, et al., Axial crushing of thin-walled
high-strength steel sections, Int. J. Impact Eng. 32 (5) (2006) 847–882.
[20] M. Yamashita, H. Kenmotsu, T. Hattori, Dynamic axial compression of aluminum
hollow tubes with hat cross-section and buckling initiator using inertia force during
impact, Thin-Walled Struct. 50 (1) (2012) 37–44.
[21] S. Hou, Q. Li, S. Long, et al., Multiobjective optimization of multi-cell sections for
the crashworthiness design, Int. J. Impact Eng. 35 (11) (2008) 1355–1367.
[22] J. Fang, Y. Gao, G. Sun, et al., Dynamic crashing behavior of new extrudable multi-
cell tubes with a functionally graded thickness, Int. J. Mech. Sci. 103 (2015) 63–73.
[23] G. Sun, F. Xu, G. Li, et al., Crashing analysis and multiobjective optimization for
thin-walled structures with functionally graded thickness, Int. J. Impact Eng. 64 (0)
(2014) 62–74.
[24] N. Tanlak, F.O. Sonmez, Optimal shape design of thin-walled tubes under high-
velocity axial impact loads, Thin-Walled Struct. 84 (2014) 302–312.
[25] D. Meric, H. Gedikli, Energy absorption behavior of tailor-welded tapered tubes
under axial impact loading using coupled FEM/SPH method, Thin-Walled Struct.
104 (2016) 17–33.
[26] M. Merklein, M. Johannes, M. Lechner, et al., A review on tailored blanks—pro-
duction, applications and evaluation, J. Mater. Process. Technol. 214 (2) (2014)
151–164.
[27] C. Chuang, R. Yang, G. Li, et al., Multidisciplinary design optimization on vehicle
tailor rolled blank design, Struct. Multidiscip. Optim. (2008).
[28] L. Ying, X. Zhao, M. Dai, et al., Crashworthiness design of quenched boron steel
thin-walled structures with functionally graded strength, Int. J. Impact Eng. 95
(2016) 72–88.
[29] A. Abdollahpoor, X. Chen, M.P. Pereira, et al., Sensitivity of the final properties of
tailored hot stamping components to the process and material parameters, J. Mater.
Process. Technol. 228 (2016) 125–136.
[30] R. George, A. Bardelcik, M.J. Worswick, Hot forming of boron steels using heated
and cooled tooling for tailored properties[J], J. Mater. Process. Technol. 212 (11)
(2012) 2386–2399.
[31] B.T. Tang, S. Bruschi, A. Ghiotti, et al., Numerical modelling of the tailored
tempering process applied to 22MnB5 sheets, Finite Elem. Anal. Des. 81 (0) (2014)
69–81.
[32] S.Y. Deng, G.Z. Shen, P. Hu, et al.Ivestigations on numerical simulation of tailored
tempering process based on related experiments, in: Proceedings of the 11th
International Conference on Numerical Methods in Industrial Forming Processes,
Shenyang, China, 2013.
[33] R. G, Hot Forming of Boron Steels With Tailored Mechanical Properties:
177
Thin-Walled Structures 117 (2017) 165–177
Experiments and Numerical Simulations, University of Waterloo, Waterloo, 2011.
[34] K. Mori, T. Maeno, K. Mongkolkaji, Tailored die quenching of steel parts having
strength distribution using bypass resistance heating in hot stamping, J. Mater.
Process. Technol. 213 (3) (2013) 508–514.
[35] K. Mori, Y. Okuda, Tailor die quenching in hot stamping for producing ultra-high
strength steel formed parts having strength distribution, Ann. - Manuf. Technol. 59
(1) (2010) 291–294.
[36] A. Reyes, M. Langseth, O.S. Hopperstad, Square aluminum tubes subjected to
oblique loading, Int. J. Impact Eng. 28 (10) (2003) 1077–1106.
[37] D.C. Han, S.H. Park, Collapse behavior of square thin-walled columns subjected to
oblique loads, Int. J. Impact Eng. 28 (2003) 1077–1106.
[38] C. Qi, S. Yang, F. Dong, Crushing analysis and multiobjective crashworthiness
optimization of tapered square tubes under oblique impact loading, Thin-Walled
Struct. 59 (0) (2012) 103–119.
[39] S. Yang, C. Qi, Multiobjective optimization for empty and foam-filled square
columns under oblique impact loading, Int. J. Impact Eng. 54 (0) (2013) 177–191.
[40] Z. Ahmad, D.P. Thambiratnam, A.C.C. Tan, Dynamic energy absorption character-
istics of foam-filled conical tubes under oblique impact loading, Int. J. Impact Eng.
37 (5) (2010) 475–488.
[41] A. Reyes, M. Langseth, O.S. Hopperstad, Crashworthiness of aluminum extrusions
subjected to oblique loading: experiments and numerical analyses, Int. J. Mech. Sci.
44 (9) (2002) 1965–1984.
[42] A. Bardelcik, M.J. Worswick, S. Winkler, et al., A strain rate sensitive constitutive
model for quenched boron steel with tailored properties, Int. J. Impact Eng. 50 (0)
(2012) 49–62.
[43] T. Nishibata, N. Kojima, Effect of quenching rate on hardness and microstructure of
hot-stamped steel, J. Alloy. Compd. 577 (Suppl.1) (2013) S549–S554.
[44] D.J. Mun, E.J. Shin, Y.W. Choi, et al., Effects of cooling rate, austenitizing
temperature and austenite deformation on the transformation behavior of high-
strength boron steel, Mater. Sci. Eng.: A. 545 (0) (2012) 214–224.
[45] L. Ying, M. Dai, P. Hu, et al., Strength and hardness prediction based on cooling rate
for hot forming high strength steel, J. Jilin Univ. (Eng. Technol. Ed.). (06) (2014)
1716–1722.
[46] K. Omer, Development and Testing of a Hot Stamped Axial Crush Member With
Tailored Properties, University of Waterloo, Waterloo, 2014.
[47] F. Xu, G. Sun, G. Li, et al., Crashworthiness design of multi-component tailor-
welded blank (TWB) structures, Struct. Multidiscip. Optim. 48 (3) (2013) 653–667.
[48] J.P.C. Kleijnen, An overview of the design and analysis of simulation experiments
for sensitivity analysis], Eur. J. Oper. Res. 164 (2) (2005) 287–300.
[49] K. Deb, A. Pratap, S. Agarwal, et al., A fast and elitist multiobjective genetic
algorithm: NSGA-II], IEEE Trans. Evolut. Comput. 2 (6) (2002) 182–197.
[50] C.E. Mohn, W. Kob, A genetic algorithm for the atomistic design and global
optimisation of substitutionally disordered materials, Comput. Mater. Sci. 45 (1)
(2009) 111–117.
[51] W. Abramowicz, N. Jones, Dynamic axial crushing of square tubes, Int. J. Impact
Eng. 2 (2) (1984) 179–208.
[52] J. Fang, Y. Gao, G. Sun, et al., On design of multi-cell tubes under axial and oblique
impact loads, Thin-Walled Struct. 95 (2015) 115–126.