Thin-Walled Structures 59 (2012) 103119

Contents lists available at SciVerse ScienceDirect

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

Crushing analysis and multiobjective crashworthiness optimization

of tapered square tubes under oblique impact loading
Chang Qi, Shu Yang n, Fangliang Dong
State Key Laboratory of Structural Analysis for Industrial Equipment School of Automotive Engineering, Dalian University of Technology,
B1211 Chuangxinyuan High-rise Building, No. 2 Linggong Road, Ganjingzi District, Dalian 116024, China

a r t i c l e i n f o

abstract

Article history:
Received 9 March 2012
Received in revised form
10 May 2012
Accepted 14 May 2012
Available online 14 June 2012

In this paper, a class of axisymmetric thin-walled square (ATS) tubes with two types of geometries
(straight and tapered) and two kinds of cross-sections (single-cell and multi-cell) are considered as
energy absorbing components under oblique impact loading. The crash behavior of the four types of
ATS tubes, namely single-cell straight (SCS), single-cell tapered (SCT), multi-cell straight (MCS) and
multi-cell tapered (MCT), are rst investigated by nonlinear nite element analysis through LS-DYNA. It
is found that the MCT tube has the best crashworthiness performance under oblique impact regarding
both specic energy absorption (SEA) and peak crushing force (PCF). Sampling designs of the MCT tube
are created based on a four-level full factorial design of experiments (DoE) method. Parametric studies
are performed using the DoE results to investigate the inuences of the geometric parameters on the
crash performance of such MCT tubes under oblique impact loading. In addition, multiobjective
optimization design (MOD) of the MCT tube is performed by adopting multiobjective particle swarm
optimization (MOPSO) algorithm to achieve maximum SEA capacity and minimum PCF with and
without considering load angle uncertainty effect. During the MOD process, accurate surrogate models,
more specically, response surface (RS) models of SEA and PCF of the MCT tubes are established to
reduce the computational cost of crash simulations by nite element method. It is found that the
optimal designs of the MCT tubes are different under different load angles. It is also found that the
weighting factors for different load angles are critical in the MOD of the MCT tubes with load angle
uncertainty.
& 2012 Elsevier Ltd. All rights reserved.

Keywords:
Tapered tube
Crashworthiness
Energy absorption
Multiobjective optimization
Oblique impact
Load uncertainty

1. Introduction
Thin-walled metal tubes have been widely used as energy
absorbing devices for decades in trains, passenger cars, ships and
other high-volume industrial products since they are relatively
cheap and weight efcient. For instance, the crash box of an
automotive body in white (BIW) is often made of thin-walled
tubes which can absorb the kinetic energy of the vehicle through
plastic deformation during an impact event.
Extensive efforts have been made to investigate the crushing
behavior of thin-walled tubes through analytical, numerical and
experimental methods. Among those, some focused on tubes with
various cross-sections including circular [13], polygonal (e.g.
square, rectangular, etc.) [46] and top-hat like [7], while other
researchers tried to improve the energy absorption of thin-walled
tubes by lling them with different materials including metallic
[811] and polymer foams [12,13]. Besides that, thin-walled

Corresponding author. Tel./fax: 86 411 84706475.

E-mail address: yangshu@dlut.edu.cn (S. Yang).

0263-8231/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.tws.2012.05.008

tubes with multiple cells have been shown to have desirable

energy absorption with weight efciency [14,15]. Zhang and
Cheng [16] showed through numerical simulations that the
energy absorption efciency of multi-cell (MC) columns was
about 50%100% higher than that of foam-lled columns. More
recently, Tang et al. [17] further showed that by introducing more
corners into the structure, the energy absorption of thin-walled
tubes can be further increased.
All these above-mentioned studies have been focused on the
crushing response and energy absorption characteristics of thinwalled structures under pure axial loads. However, in real-world
impact event, especially in the context of automobile crashes, the
energy absorbers such as the shotguns and side rails rarely
experience pure axial or pure bending loads, rather they deform
under a combination of axial and off-axis or oblique loads. Such
loading causes the thin-walled tube to deform via a combination
of both axial progressive and global bending modes. Compared to
progressive axial collapse, the global bending deformation of a
thin-walled structure is generally unstable with an associated
reduction in impact energy absorption. The experimental and
numerical analyses on the quasi-static oblique loading behavior

104

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

of both empty [18,19] and foam-lled [20] square aluminum

columns were carried by Reyes et al. [20]. The studies showed
that the energy absorption drops drastically when a global
bending mode is initiated instead of progressive buckling, and it
decreases further with increasing load angle. Another numerical
investigation on the oblique crush behavior of square thin-walled
column made of mild steel showed that there exists a critical load
angle at which a transition takes place from the axial progressive
collapse mode to the global bending collapse mode while the
latter yields a signicant reduction in the mean load [21]. In
aware of this, increased focus has been given to tapered tubes,
namely tapered rectangular and conical tubes, which have been
found preferable to straight tubes since they are capable of
withstanding oblique impact loads as effectively as axial loads,
and appear to have fewer chances to fail via global bending [22].
Despite this, compared to straight tubes, information in the open
literature on the oblique loading of tapered thin-walled energy
absorbers is limited. Nagel and Thambiratnam [2325] compared
the energy absorption response of straight and tapered thinwalled rectangular tubes subjected to both axial and oblique
quasi-static and dynamic impact loading. They showed the
tapered tubes have more advantages in applications when oblique impact is inevitable. Ahmad et al. also [26] showed that foamlled conical tubes appear to be advantageous in impact applications where oblique impact load is expected.
To seek the optimal congurations of empty and foam or
honeycomb-lled thin-walled tubes, structural optimization techniques have been applied on the crashworthiness design of such
components under either pure axial [2734] or lateral loads
[35,36]. The design objectives in theses crashworthiness optimizations usually include the specic energy absorption (SEA,
absorbed energy per unit structural mass) and the peak crushing
force (PCF). The SEA should always be maximized to improve
energy absorption efciency, while the PCF should be decreased
for the safety of the passengers or goods protected by such
structures. To deal with these multiple objectives, more recently,
multiobjective optimization design (MOD) method has been
employed in the crashworthiness design of tapered circular tubes
under axial loads. Acar et al. [37] carried out MOD of tapered
circular thin-walled tubes with axisymmetric indentations. The
optimum values of the number of axisymmetric indentations, the
taper angle and the tube thickness were sought for maximum
crush force efciency (CFE, the ratio of the average crushing load
to the PCF), and maximum SEA. Hou et al. [38] optimized the
energy absorption characteristic of tapered circular tubes with
three different congurations. Moreover, surrogate modeling
techniques, such as the response surface method (RSM), are often
used in these studies in lieu of nonlinear nite element analysis
(FEA) for fast iteration. However, in all these aforementioned
optimization studies, only pure axial or lateral loading conditions
were considered despite the fact that oblique loadings are much
more common in real crash events. In other words, the load angle
uncertainty effect has not been systematically considered in the
crashworthiness design optimization of such thin-walled structures. To date, crashworthiness optimization of tapered tubes
under oblique loading are not available in the literature to the
best of authors knowledge.
In the present study, a new class of tapered square tubes with
MC cross-sections is proposed. These tubes are expected to have
high energy absorptions efciency as the MC straight tubes as
well as better capability of withstanding oblique impact loading
as shown by the tapered tubes. The crushing responses of this
type of thin-walled tubes under both axial and oblique impact
loading are analyzed using the nonlinear explicit FEA code LSDYNA [39]. Towards this end, nite element (FE) models validated
against theoretical and experimental results in the literature are

established. Design information for such tubes as energy absorbers in oblique impact applications are developed through parametric study based on the sampling designs. A four-level full
factorial design of experiments (DoE) method is used to determine those sampling design points. Also based on the DoE results,
quartic polynomial functions are used to build response surface
(RS) surrogate models that relate SEA and PCF to the geometric
design variables associated with the MC tapered square tubes
under oblique impact loading. Multiobjective particle swarm
optimization (MOPSO) algorithm is used as the optimizer for
solving the MOD problems both with and without load angle
uncertainty.

2. Problem description
2.1. Structures of the crushing analysis under oblique impact loading
The structures considered in this study are tapered axisymmetric
thin-walled square (ATS) tubes with single-cell (SC) and multi-cell
(MC) cross-section congurations. For simplicity, they are named as
SCT (abbreviate of single-cell tapered) tubes and MCT (abbreviate of
multi-cell tapered) tubes afterwards, respectively. For comparison
purposes, their counterparts of straight tubes are named as SCS and
MCS tubes, which stand for single-cell straight and multi-cell straight
tubes, respectively. All tubes have the same baseline geometry with
length L 250 mm, largest cross-sectional side length A80 mm and
wall thickness t 2.0 mm besides that the tapered tubes have an
angle y 51. This geometry is determined from typical dimensions of
the front side rail of a passenger car [15] as well as similar thinwalled structures investigated in the literatures [25,26,37]. The
schematic diagram of the computational model is shown in Fig. 1.

Fig. 1. Tube dimensions and schematic of the computational model.

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

105

computed to be 11 kJ/kg using Eq. (4). The weighting factor

reects the relative importance of individual SEAai in the structural SEA calculation involving load angle uncertainty effect. Here,
the normalized weighting factors can be applied as
Xn
W ai 1, W ai Z 0
5
i1
Varying the weights allows us to obtain a different SEAa value
with emphasizing on different load angles. In the above example, if
3
3
more weights are given to large load angles with W 0 0:1, W 10
203
0:3, and W 0:6, then we have SEAa 8.7 kJ/kg based on Eq. (4).

3. Numerical simulations
Fig. 2. Cross-section of the ATS tube: (a) SC and (b) MC.

3.1. Finite element (FE) modeling

The large end of the tube is clamped, while the other end is impacted
by a rigid wall with a constant velocity of V10 m/s, which is
typically encountered in passenger vehicle crash event. The normal of
the rigid wall is in the XZ plane and has an oblique angle a with the
axis of the tube. The SC and MC cross-sections of the ATS tubes are
shown in Fig. 2(a) and (b), respectively.
2.2. Structural crashworthiness indices under oblique loading
Generally, there are four key crashworthiness indices for
evaluating the performance of the energy absorbing structure,
i.e., mean crushing force (MCF), peak crushing force (PCF), energy
absorption (EA) and specic energy absorption (SEA). The MCF for
a given structural deformation is dened as
Z
1 d
MCF
Fxdx
1

3.1.1. Material properties

The material of the ATS tubes is aluminum alloy AA6060 T4
with mass density r 2700 kg/m3, Youngs modulus E68.2 GPa,
Poissons ratio n 0.3, yield strength sy 80 MPa and ultimate
strength su 173 MPa. To take strain hardening effects into
account, the energy equivalent ow stress can be calculated by
using [41]
r
sy su
6
s0
1n
where n is the strain hardening exponent of the material, n 0.23
and the ow stress for AA6060T4 adopted here is 106 MPa. The
tensile stressstrain curve is shown in Fig. 3. The strain rate effect
is neglected in the FE modeling as the aluminum alloy is strain
rate insensitive [42]. The ATS tubes are modeled with MAT_123
(Modied Piecewise Linear Plasticity) provided by LS-DYNA [39].

where F(x) is the instantaneous crushing load and d is the

structural deformation at a specic time when MCF is calculated.
PCF that can closely relate to the deceleration of the structure is
another critical criterion in crashworthiness design. For instance,
PCF should be reduced or constrained under certain level in
automobile crash safety design since a large PCF usually results
in high acceleration of the occupant [40].
The EA is typically indicated by the total strain energy
absorbed during the plastic deformation (note: the elastic deformation energy is neglected), which can be calculated from the
crushing forcedeection curve, as
Z d
EAd
Fxdx
2

3.1.2. FE models
The FE models of the four types of ATS tubes, namely SCS, SCT,
MCS, MCT tubes, are shown in Fig. 4. The tubes are modeled with
BelytschkoTsay four-node shell element with three integration
points through the thickness and one integration point in the
element plane.
To nd the optimum mesh size for the numerical simulation, a
mesh convergence analysis is performed. The peak crushing force
(PCF) of the four types of ATS tubes under axial impact loading
predicted by different element size and the relative errors with
respect to the results of element size 2.5 mm are summarized in
Table 1. Fig. 5 further depicts the convergence of the PCF with

where d is the effective stroke length [14], which is taken as 0.6l

in the current study, and l is the total length of the energy
absorbing structure.
The SEA is dened as the ratio of the absorbed energy to the
mass of the structure, m, as follows:
SEA EA=m

In this study, in order to bring the load angle uncertainty effect

into structural SEA calculation under oblique loading, a new
crashworthiness index, SEAa, is proposed and dened as follows:
Xn
SEAai W ai
4
SEAa
i1
where the symbol a implies that SEAa is a composite index
involving the structural SEA under multiple oblique loadings;
SEAai denotes the structural SEA under the ith oblique load with
angle ai, and W ai is the corresponding weighting factor to this
SEAai . For example, if ai are selected as 01, 101 and 201, respec3
3
tively, and we assume that SEA0 15 kJ/kg, SEA10 12 kJ/kg and
203
03
103
203
SEA 6 kJ/kg with W W W 1=3, then SEAa can be

Fig. 3. Tensile stressstrain curve of AA6060 T4.

106

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

Fig. 4. FE models of the ATS tubes (Overall view and top view): (a) SCS, (b) SCT, (c) MCS and (d) MCT.
Table 1
PCF of the four types of ATS tubes under axial impact loading predicted by different element size.
ATS tubes

Element size (mm)

4.5

SCS
SCT
MCS
MCT

4.0

3.5

3.0

2.5

PCF (kN)

Relative error (%)

PCF (kN)

Relative error (%)

PCF (kN)

Relative error (%)

PCF (kN)

Relative error (%)

PCF (kN)

Relative error (%)

60.677
33.792
114.693
91.184

3.82
5.53
3.91
6.43

59.612
33.157
113.273
87.558

1.99
3.55
2.62
2.19

58.767
32.979
111.472
86.911

0.55
2.99
0.99
1.44

58.539
32.255
110.903
85.902

0.16
0.73
0.48
0.26

58.446
32.02
110.376
85.678

0
0
0
0

increased mesh density. It is clear from both Table 1 and Fig. 5

that convergent solutions of PCF are achieved for all tubes when
an element size of smaller than 3.0  3.0 mm has been adopted.
Similar results are obtained for other crashworthiness indices and
load cases and are omitted here for the sake of brevity. Hence,
mesh size adopted for all models of 2.5  2.5 mm is sufcient.
The nonlinear explicit FE code LS-DYNA 971 [39] is used to
simulate the crash behavior of the ATS tubes subjected to oblique
impact loading. The oblique load exerted to the tubes is modeled
by the RIGID_WALL_GEOMETRIC_FLAT_MOTION card in LS-DYNA.
An automatic single-surface contact is adopted to account for the
contact of the tube itself during deformation, while an automatic
node-to-surface contact is used between tube and rigid wall. For
both static and dynamic friction, the friction coefcient of 0.2 is
adopted for all contact condition.
CAD model of the tube is created using Pro/Engineer through
parametric modeling technique. The CAD model is imported into
the preprocessor of FE software package ANSYS [43] to automatically generate the LS-DYNA model using APDL (ANSYS Parametric Design Language). The post-processor LS-PREPOST is used
for visualization and data acquisition.

theoretical prediction of MCF under quasi-static loading as

3.2. Numerical results and discussion

3.2.1. Validation of the numerical models
In order to validate the developed FE models, FEA results of
the SC and MC straight square tubes under axial dynamic loading
are rst compared with the theoretical solutions available in the
literatures. For SC tubes, Abramowicz and Jones [2] developed
the expression of mean crushing force (MCF) theoretically with
the form
1=3 5=3

MCF 13:06s0 b

Fig. 5. Convergence of peak crushing force versus mesh density for ATS tubes
under axial loading.

where s0 is the energy equivalent ow stress of tube material

given by Eq. (6), b and t are width and wall thickness of tube,
respectively. For MC tubes, Zhang et al. [44] derived the

MCF s0 t

p
Nc 4No 2NT pLc t

where Nc, No and NT denote the number of corner, crisscross and

T-shape in the cross-section of the MC tube, and Lc is the total
length of the wall of the cross-section. An enhancing coefcient
(EC) was proposed by Langseth and Hopperstad [42] to account
for the inertia and strain rate effects of tubes under dynamic
loading. According to their test results, the EC is in the range of
1.31.6 for AA6060 T4 tubes under the loading rate of 820 m/s.
Thus, the MCF for MC tubes under axial dynamic loading can be
written as:
MCF ECUs0 t

p
N c 4No 2NT pLc t

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

107

Fig. 6. Forcedeection curves for straight square tubes under axial dynamic loading: (a) SC section and (b) MC section.

Table 2
Comparison of the FEA results for SCT tube under quasi-static oblique loads with
the experimental results [25].
Load
angle
(deg)

10
30

Peak crushing force (kN)

Mean crushing force (kN)

Nagel
Present FE Relative
et al. [25] model
error (%)

Nagel
Present
et al. [25] FE model

Relative
error (%)

64.8
39.4

46.7
21.4

0.86
0.93

66.4
39.8

2.47
1.02

47.1
21.6

crushing force versus deection and the nal deformed prole of

each tube predicted by the present FE models and the test results.
More detailed comparison of the PCF and MCF results from the
present FE simulations with those from the literature [25] for the
SCT tube under oblique loads is tabulated in Table 2, and indicates
good agreement between the two sets of results.
These satisfactory correlations show that it is feasible to
extend the SC and MC straight tube models to tapered tube
models under both axial and oblique loading conditions.

Fig. 7. Numerical and experimental results (Nagel and Thambiratnam [25])

comparison of SCT tubes under quasi-static oblique loadings: (a) a 101 and (b)
a 301.

For simplicity, the EC is set to 1.4 for a crushing velocity of

10 m/s in this study.
The FEA results of SC and MC straight square tubes under axial
dynamic loading can be veried by the theoretical solutions of
Eqs. (7) and (9), respectively. The comparison of the MCF versus
deection of the SC (t 2 mm, b80 mm) and MC (t2 mm,
Nc 4, No 1, NT 4, Lc 480 mm) straight square tubes under
axial dynamic loading rate of 10 m/s are shown in Fig. 6(a) and
(b), respectively. It is seen that present FEA results are in good
agreement with the theoretical solutions for both SC and MC
straight square tubes.
Second, the FEA results of a single-cell tapered (SCT) tube
under quasi-static crushing with load angles of 101 and 301 are
compared with the test results in Ref. [25]. The geometric and
material parameters of the tube as well as the loading and
boundary conditions are all from Ref. [25]. Fig. 7 compares the

3.2.2. Numerical results comparison of four types of ATS tubes under

oblique impact loading
In this study, it is of special interest to investigate the
inuences of load angle and important geometric parameters on
the response of the ATS tubes under oblique impact loading. Such
knowledge provides a guide as to the design of such structures in
applications involving oblique loads. Using the validated FE
models, numerical analyses are rst carried out to compare the
crush behavior of the four types of ATS tubes, i.e., SCS, SCT, MCS
and MCT tubes, under dynamic oblique impact with variations in
loading angles. In performing the numerical simulations, the
tubes are actually impacted by a rigid wall with an increasing
load angle a from 01 (axial crushing) to 451 with a typical interval
of 51. Particular attention is given to the inuence of load angle on
the SEA and PCF of the ATS tubes under investigation.
The dynamic crushing forcedeection curves for the four
types of ATS tubes under different impact angles are shown in
Fig. 8. In the gure, the deection equals the vertical displacement d of the rigid wall with various load angle a, as depicted in
Fig. 1. It is clearly seen that the load angle has a signicant effect
on the response of the ATS tubes. For the SCS tube, there exists a
critical load angle between 201 and 301 at which a transition takes
place from the axial progressive collapse mode to the global
bending collapse mode, as shown in Fig. 8(a). When the load angle
is less than 201, the load oscillates between local minimum and
maximum values with mean load level about 20 kN. When the load
angle is increased to 301, the load does not oscillate, rather it increases

108

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

Fig. 8. Effect of load angle on the crushing forcedeection response of ATS tubes: (a) SCS, (b) SCT, (c) MCS and (d) MCT.

Fig. 9. Final deformed proles of ATS tubes under oblique impact with different load angles.

up to a certain deection, before decreasing continuously for the

remainder of the crushing process. Similar observation can be made
for the force history of the SCT tube as shown in Fig. 8(b).

Nevertheless, the critical load angle for the SCT tube is in the range
of 301401, implying that tapering the square tube helps to stabilize
the tube under oblique loading. Same observations have also been

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

made for both single and double tapered rectangular tubes [25].
Moreover, for the straight tube, the PCF occurs at the beginning of the
crushing when the tube is loaded axially, i.e., a 01, corresponding to
the initiation of buckling at the impacted end of the tube. As the load
angle increases, the PCF reduces in magnitude. In contrast, there is no
predominate initial PCF for the SCT tube due to the tapering feature of
the tube which will be further discussed in a later section. Furthermore, it is seen from Fig. 8(c) and (d) that both straight and tapered
MC tubes show higher crushing force level than their SC counterparts
for all the simulated load angles. It is also seen that both MCS and
MCT tubes exhibit progressive collapse when the load angle is less
than 101. The force histories of the MC tubes show more uctuations
due to the constraining effect of the crisscross part in the tubes
resulting in more lobes during the impact. Similar to the SCT tube, the
MCT tube shows no obvious initial PCF.
The above observed load-deection characteristics of the ATS
tubes can be further understood by looking at the nal stage
deformation modes of the tubes at typical load angles as shown in
Fig. 9. All four types of tubes for a load angle of 101 deform by
progressive buckling initiating at the impacted end. Folds in all
tubes form via the global symmetric (Type I) collapse mode, as
dened in Ref. [6]. Different from the SC tubes with unchanged
central axes, the MC tubes show a tendency to slide sideways
slightly in a direction towards the oblique loading direction due
to their relatively higher stiffness. This is discussed in detail later
on. In fact, for a load angle smaller than 101 each tube fails via
progressive collapse. When the load angle reaches 201, the SC
tubes still fail via progressive collapse but the MC tubes fail via
global bending collapse. Initially the tension ange buckles at the
impacted end, after which a plastic hinge or lobe forms on the
compression ange either close to the clamped end (MCS tube) or
a little bit higher (MCT tube). When the load angle reaches 301,
only the SCT tube fails via progressive collapse, and when the load
angle nally reaches 401, all tubes fail via global bending collapse.
A single bending mode is exhibited for all tubes, in which the
compression ange buckles inward while the two side-walls
buckle outward to form the lobe. A similar mode has been
observed on the oblique loading of square [18] and rectangular
[25] tubes. Thus, there exists a critical load angle for each tube
conguration which makes a transition of the deformation mode
from progressive to global bending collapse. This is further
examined with additional data at 11 increments of the load angle
in the transition range, to determine the critical load angle of each
tube conguration.
Fig. 10 shows the effect of load angle on the SEA calculated up
to a deection of 150 mm for each tube under oblique impact
loading. As seen from the gure, for each tube conguration, three
regions arise from the relationship between the SEA and the load
angle: a region (1) where progressive axial collapse dominates
(progressive region), a region (2) where there is a transition from
progressive to global bending collapse (transition region), and
nally a region (3) where global bending collapse dominates the
response (global region). This relationship has also been found for
the oblique loading of mild steel straight square columns [21],
tapered rectangular tubes [25] and empty and foam-lled conical
tubes [26]. From the detailed observation, the critical load angle
lies within the transition range of 253 r a r 273 for the SCS tube,
393 r a r 403 for the SCT tube, 153 r a r173 for the MCS tube and
143 r a r 163 for the MCT tube. It is evident that the critical load
angles of the SC tubes are generally higher than that of the MC
tubes. The reason is that, with the same wall thickness, the SC
tubes relative lower stiffness causes its loaded end more likely to
deform under oblique loading instead of being pushed laterally
away from the central axis of the tube as in the case of the MC
tube. Thus, the deformed portion of the loaded end serves as a
crush trigger to initiate the progressive collapse of the

109

Fig. 10. Effect of load angle on the SEA of different types of ATS tubes.

undeformed portion of the tube, as shown in Fig. 9. For each

ATS tube conguration, when the load angle is smaller than the
lower bound of the transition range, the tube fails progressively to
give a high SEA. But when the load angle is larger than the upper
bound of the transition range, global bending is initiated due to
the formation of plastic hinge in the tube (see Fig. 9), causing an
abrupt drop of the absorbed energy (and hence SEA) over the
transition region. As seen from Fig. 10, the SEA drops over the
transition region by 37%, 17%, 45%, and 53% for the SCS, SCT, MCS
and MCT tube, respectively. It is also seen that this drop is more
signicant for the MC tubes. However, despite the increased
structural mass and the relatively small critical load angles, it is
evident in Fig. 10 that SEA of the MC tubes are much higher than
that of the SC tubes across the full range of load angles (01401)
under investigation, especially in the progressive region when the
load angle is less than 151. This declares a great advantage of
using MC tubes in applications involving oblique loading and
when weight is an important consideration for comparing the
energy absorption capacity of structures, such as in the automotive industry. Similar observations have been found for MC
square columns under pure axial crushing [14,16,44].
To take the load angle uncertainty effect into account, without
loss of generality, we select load angles ai 01,51,101,y,401
(i 1,y,9) in Eq. (4) to calculate SEAa of the four types of ATS
tubes with W ai 1=9 for each ai. The normalized SEAa of different
ATS tubes is compared in Fig. 11. SEAa of the MCT tube is 3.30068
times that of the SCS tube, indicating that the MCT tube is
superior to other ATS tube congurations for energy absorption
and light weight applications involving oblique impact loading
conditions.
Fig. 12 depicts the effect of load angle on the peak crushing
force (PCF) for different ATS tubes. It is evident that the PCF
decreases with the increasing of load angle for each tube, which
means the maximum PCF occurs when the tube is loaded pure
axially (a 01). Compared with the corresponding straight tubes,
tapered tubes have smaller PCF for both SC and MC section
congurations, meaning that tapered tubes outperform straight
tubes regarding PCF reduction.

110

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

Table 3
Range and step size of design variables and load angle in the oblique impact
analyses of MCT tubes.

Fig. 11. Normalized SEAa of different types of ATS tubes under oblique impact
loading.

Parameter

Lower bound

Upper bound

Step size

Wall thickness, t (mm)

Taper angle, y (deg)
Load angle, a (deg)

1.0
1
0

2.5
7
30

0.5
2
5

For the purpose of doing this, FEA are performed for the selected
sampling design points determined through the design of experiments (DoE) method since the selection of sampling design points
can be critical for the parametric study as well as the construction
of the responses surface (RS) models for MOD. Many different DoE
methods are available, such as the factorial, Koshal, composite,
Latin Hypercube, D-optimal, etc. [45]. In the following study, a set
of sampling design points are created using a four-level full
factorial DoE method. Consequently, 4  4 sampling design points
are selected, which evenly distribute in the design domain. The
responses of SEA and PCF of the MCT tubes are evaluated at these
16 sampling points with the load angle varying from 01 to 301 at a
step size of 51 such that a total number of 112 FEA are carried out.
The range and step size of the two design variables of the MCT
tubes and the load angle for the DoE are listed in Table 3. The
dimensions are chosen to cover the typical range of tube sizes
commonly used in crashworthiness applications, such as in
automobiles [21,25].
4.2. Parametric study

Fig. 12. Effect of load angle on the peak crushing force of different ATS tubes.

From the results shown above, it appears that the MCT tube
conguration shows a better performance in both SEA and PCF
under oblique impact loading, compared with the other three
structural congurations of the ATS tube. The MCT tube is thus
selected as the best structural component in the present study. In
the following, the inuences of important geometric parameters
on the crushing response of MCT tube are rst examined through
parametric study, followed by multiobjective optimization design
(MOD) to further improve its crashworthiness under oblique
impact loading.

4. Parametric study and response surface models of MCT

tubes
4.1. Sampling design points
Before performing MOD of the MCT tube, the inuence of
geometric parameters (design variables) on the tube response
under oblique impact loading are rst studied based on numerical
simulations, including the wall thickness, t and the taper angle, y.

Based on the DoE results, detailed FE models are created for

the tubes representing the sampling design points and used for
the computer crashworthiness analyses with an impact velocity
V10 m/s. From the FEA results, the response functions of SEA
and PCF are determined as well as the critical load angle, if any, at
which there is a transition from progressive to global bending
collapse of the tube. The baseline tube geometry is used for the
parametric study with L250 mm, A 80 mm, t 2.0 mm and
y 51. It is assumed in the following study that the load angle, the
wall thickness and taper angle of the MCT tubes are independent
parameters which are decoupled in affecting the tubes crash
responses.
4.2.1. Effect of wall thickness on tube response under oblique impact
Initially the effect of the wall thickness on the tube response,
across the range of load angle as listed in Table 3, is evaluated.
The SEA is calculated up to a deection of 150 mm (60% of the
initial length) for each tube as sufcient axial compression or
global bending has occurred at this deection, this deection is
chosen for adequate comparison of results.
Fig. 13 shows the effects of wall thickness and load angle on
the SEA and PCF of the MCT tube under oblique impact loading. It
is seen that both SEA and PCF continuously decrease as the load
angle increase for each wall thickness. The critical load angle of
the tube decreases with increased wall thickness as seen from
Fig. 13(a). For instance, the critical load angle for wall thickness
1.0, 1.5 and 2.5 mm lies within the transition range of 201251,
151201 and 101151, respectively. The reason is that increased
stiffness of the tube with a larger wall thickness makes it more
prone to be pushed laterally away from the central axis of the
tube by forming a plastic hinge at the xed end. In contrast, the
tube with a thinner wall thickness deforms more easily at the
impacted end to form the lobes which work as a trigger to initiate
the progressive collapse of the tube. The reduction in stiffness

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

111

Fig. 13. Effect of wall thickness and load angle on tube response: (a) SEA and (b) PCF.

Fig. 14. Effect of taper angle and load angle on tube response: (a) SEA and (b) PCF.

with decreased wall thickness also helps to reduce the peak

crushing force as shown in Fig. 13(b). However, a smaller wall
thickness reduces the SEA of the MCT tube at small load angles.
For instance, at load angle a 51, when the wall thickness reduces
from 2.5 to 1.0 mm, the corresponding SEA is decreased by 53.78%
from 25.603 to 11.834 kJ/kg. Moreover, the SEA and PCF are more
sensitive to wall thickness when progressive collapse dominates
the response as opposed to global bending. Furthermore, it is
interesting to note that the reduction in SEA and PCF as the load
angle increases is more signicant when the wall thickness is
larger.

4.2.2. Effect of taper angle on tube response under oblique impact

Fig. 14 shows the SEA and PCF of each tube versus load angle
for varying taper angle, y. Again, both SEA and PCF continuously
decrease as the load angle increases for each taper angle. From
Fig. 14(a), the taper angle has no signicant inuence on the
critical load angle of the tube when y r 53 as shown in a similar
study of tapered rectangular tubes [25]. However, the MCT tube
with taper angle y 71 gives a relatively smooth SEA curve in the
complete range of load angle, which means there is no obvious
transition region from progressive to global bending collapse. This
property makes its SEA higher than the MCT tube with a smaller
taper angle, especially in the upper range of the load angle. For
instance, at load angle a 251, the SEA of the 51 MCT tube is
4.476 kJ/kg, while the SEA of the 71 MCT tube is 6.695 kJ/kg,
which is 1.50 times of the former. Also seen from Fig. 14(b),
increasing the taper angle of the tube reduces the peak crushing force.
Therefore, it seems that larger taper angle is preferred for both SEA
maximization and PCF minimization in the current study.

Fig. 15. Relative effect of wall thickness, taper angle and load angle on the
crushing response of MCT tube under oblique impact.

Fig. 15 further compares the relative effect of the parameters

on SEA and PCF. It can be obviously observed that the load angle
has more inuence than the wall thickness and taper angle on
SEA. However, the load angle is an uncertainty parameter and
cannot be controlled in the real applications. As such, for practical
applications, only the wall thickness and taper angle can be
controlled. From Fig. 15, the PCF is more sensitive to wall
thickness than to taper angle. In addition to this, the wall
thickness has positive effect on both SEA and PCF, while the taper
angle has positive inuence on SEA and negative inuence on PCF.
From the results shown thus far it appears that both the wall
thickness and taper angle have signicant effects on SEA and PCF
of the MCT tube. More specically, an increase in SEA due to a

112

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

parameter change often leads to an increase in PCF as well. In

other words, SEA and PCF could conict with each other and one
has to impose the optimum in a Pareto sense. As a practical
outcome, multiobjective optimization design (MOD) is carried out
in the following to search for the optimal MCT tube congurations
with respect to SEA and PCF under oblique impact loading by
controlling both the wall thickness and taper angle across the
range of load angles considered in this study.
4.3. Response surface models
4.3.1. Response surface method (RSM) [45]
In order to achieve optimized responses from the MCT tubes
under oblique impact loading, the surrogate model approach, e.g.
RSM, is used which has been proven particularly effective under
such circumstance [34,38]. In this study, RSM is employed to
predict the crashworthiness indices of SEA and PCF of the MCT
tubes under oblique impact loading. These indices are then to be
maximized or minimized when oblique impact occurs on
such tubes.
In our problem, the responses of the MCT tube are SEA(t,y) and
PCF(t,y), which are assumed a priori in terms of relatively simple
basis functions as
XN
~
yx
SEAt, yorPCFt, y
a j t, y
10
i1 i i
where N represents the number of basis functions ji(t,y). In
general, the selected basis functions should ensure sufcient
accuracy and fast convergence. In Eq. (10), the unknown parameters a (a1,a2,y,aN), known as the regression coefcients, are
determined using the least-square technique. Suppose a number
of sampling design points x(i) (i1,2,y,M) (M4N) are selected
and the FEA results y(i) (i1,2,y,M) at these sampling design
points (t, y)i are obtained, the least square function can be
expressed as
i2
XM
XM h
XN
L
e2
yi 
a j t, y
11
i1 i
i1
j1 j j
where the sampling design points (t, y)i are selected from the
specied design domain, ei is the error between the response yi
observed at these points and the RS approximation at that point.
Afterwards, the unknown parameter vector a (a1,a2,y,aN) can
be determined by @L/@a 0, which is
a FT F1 FT y

12

where F denotes the matrix consisting of basis functions evaluated at the M sampling design points as
2
3
j1 t, y1    jN t, y1
6
7
^
&
^
F4
13
5
j1 t, yM    jN t, yM
By substituting Eq. (12) into Eq. (10), the RS models are created
and the response functions SEA(t,y) and PCF(t,y) can be fully
dened.
The accuracies of these RS models can be measured using
several criteria. The maximum absolute error (MAX), the relative
error (RE) and the R square values are written as
i

MAX max9yi y^ 9,

14



yi y^ i 


RE  i ,
 y 

15

R2 1 PM

PM

i1

i1

yi yi 2

i 1,2,. . .,M

16

i
where yi is the FEA result, y^ is the approximation based on the
i
regression function, and y is the mean value of yi. The smaller the
values of MAX and RE, or the larger the value of R2, the more
accurate the RS model.

4.3.2. RS models of MCT tube under different load angles

Based on the methodology of Section 4.3.1, in this section,
surrogate RS models for SEA and PCF prediction of the MCT tubes
under oblique impact loading with respect to design variables t
and y are constructed for four representative load angles based on
the DoE results summarized in Table 4.
In generating these RS models, the polynomials are selected as
the basis functions. By using Eqs. (14)(16), the accuracies of
different polynomial functions with orders ranging from 1 to 4 for
SEA and PCF prediction under load angle a 01 are summarized in
Table 5. From the data listed in Table 5, it is found that the quartic
polynomial functions provide the best approximation on the
tubes responses under axial loading. It is also found that the
quartic polynomial functions are the most accurate models for
SEA and PCF prediction under the other three load angles. The
accuracy data for these functions are omitted here for the sake of
brevity. Thus, these quartic polynomial functions are used as RS
models in the following optimization design process. The quartic

Table 4
Sampling design points and crashworthiness indices of MCT tubes under different impact angles.
No.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

Thickness t
(mm)

1.0
1.0
1.0
1.0
1.5
1.5
1.5
1.5
2.0
2.0
2.0
2.0
2.5
2.5
2.5
2.5

Taper angle y
(deg)

1.0
3.0
5.0
7.0
1.0
3.0
5.0
7.0
1.0
3.0
5.0
7.0
1.0
3.0
5.0
7.0

Load angle a 03

Load angle a 103

Load angle a 203

Load angle a 303

SEA
(kJ/kg)

PCF
(kN)

SEA
(kJ/kg)

PCF
(kN)

SEA
(kJ/kg)

PCF
(kN)

SEA
(kJ/kg)

PCF
(kN)

11.020
11.820
12.608
14.087
14.511
15.472
16.967
19.884
17.736
19.223
21.666
27.512
21.065
22.678
27.091
34.329

41.180
34.066
27.011
25.472
70.421
58.021
49.293
49.413
101.30
85.012
85.678
83.106
136.65
123.26
129.56
107.77

9.013
10.402
10.632
12.824
11.544
12.974
14.958
15.266
14.532
16.053
18.670
18.658
17.676
19.235
21.902
16.034

27.446
24.854
20.492
20.312
48.224
43.877
39.264
39.482
77.078
68.147
66.498
63.518
108.19
97.826
99.601
88.439

3.926
7.412
8.104
5.695
4.705
4.929
4.904
8.808
5.677
5.802
5.879
8.782
6.826
6.554
6.080
9.215

23.732
22.031
17.502
15.420
44.728
37.146
32.698
23.516
58.064
54.232
48.244
30.481
76.910
73.698
61.988
38.087

2.672
2.504
2.766
3.221
3.218
3.360
3.403
6.567
4.022
4.182
3.988
5.553
4.652
4.607
4.338
5.916

15.610
15.309
14.573
10.042
28.704
27.081
23.109
15.473
39.991
36.132
33.055
20.198
50.647
51.309
42.102
25.331

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

polynomial functions for SEA and PCF prediction for the four
representative load angles are provided in the Appendix.
The constructed quartic response surfaces for SEA and PCF
prediction at different impact angles are depicted in Figs. 16 and
17, respectively. It is observed from Fig. 16 that the SEA favors
different taper angle and wall thickness for different load angles.
The PCF, however, favors large taper angle and small wall
thickness for all load angles as shown in Fig. 17. In general, it is
hardly possible to have these two objectives be optimized
simultaneously. Such a typical feature calls for different multiobjective optimization schemes. In this study, the MOPSO algorithm is used to determine the optimum tube congurations
under various design circumstances, which is discussed in the
following section.

Table 5
Accuracy of different polynomial RS models for the MCT tube under load angle
a 01.
Response surface model
SEA
Linear polynomial
Quadratic polynomial
Cubic polynomial
Quartic polynomial
PCF
Linear polynomial
Quadratic polynomial
Cubic polynomial
Quartic polynomial

Maximum RE (%)

R2

4.2336
0.8677
0.3426
0.3031

24.111
7.8736
1.3891
1.3989

0.9206
0.9924
0.9994
0.9996

10.991
10.240
6.8526
2.1966

24.1280
7.9036
12.817
4.1530

0.9808
0.9885
0.9917
0.9993

MAX

113

5. Multiobjective optimization design (MOD) of MCT tubes

5.1. MOD approach and optimization algorithm
5.1.1. MOD approach with load angle uncertainty
The MCT tubes should be designed for maximum crashworthiness
performance under all possible load angles including both axial and
oblique impacts. The crashworthiness performance is evaluated by
the two indices (SEA and PCF) at different load angles, which should
be included in the MOD process. The wall thickness, t, and taper
angle, y of the MCT tubes are chosen as the two design variables. The
tube length (L 250 mm) and large end side length (A80 mm)
remain as constants in the design process.
Therefore, the MOD problem of the MCT tube under a specic
load angle ai without considering load angle uncertainty with two
crashworthiness design objectives can be formulated as follows:
(

min

SEAai t, y, PCF ai t, y

s:t: 1:0 mm rt r 2:5 mm, 11 r y r 71

17

To include load angle uncertainty in the MOD problem

mathematically, one needs to employ the stochastic based methods such as the reliability-based design optimization (RBDO)
method [46]. However, these methods require a pre-known
distribution function of the random variable (in the current case,
the load angle a) as well as a large number of crash simulations to
provide the mean and/or standard deviation of the response (i.e.,
SEA) to be optimized or constrained. To overcome these difculties, a simplied approach is proposed herein which is more
easily to be implemented in engineering applications. In this

Fig. 16. Quartic RSs of SEA for the MCT tube under different load angles. The bold black dots in the plots show the training points. (a) a 01, (b) a 101, (c) a 201 and
(d) a 301.

114

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

Fig. 17. Quartic RSs of PCF for the MCT tube under different load angles. The bold black dots in the plots show the training points. (a) a 01, (b) a 101, (c) a 201 and
(d) a 301.

approach, instead of dening the distribution function over the

complete range of load angle variation and performing large
amount of FE simulations, a limited number of representative
load angles are rst selected. A specied weighting factor is
assigned to each representative load angle to account for the
probability of the load angle around it. Distribution of the load angle
in the dened range can be then realized by adjusting the weighting
factors. For example, if the load angle varies between 01 and 301 and
we select four representative load angles ai 01, 101, 201 and 301.
3
3
3
3
Then, a set of weighting factors W 0 W 10 W 20 W 30 0:25
can approximate a uniform distribution, while another set of weigh3
3
3
3
ing factors W 0 0:1, W 10 0:4, W 20 0:4, W 30 0:1 may
approximate a normal distribution of the load angle in the dened
range. Accordingly, a composite performance function SEAa dened
by Eq. (4) can be obtained based on the weighting factors and used as
one of the objectives in the MOD problem. Thus, the load angle
uncertainty is considered in the design process to give a more robust
design.
Based on the proposed approach, the MOD problem with load
angle uncertainty can be written as
(
min SEAa t, y,PCF a t, y
18
s:t:
1:0 mm rt r 2:5 mm, 13 r y r 73
where SEAa is dened by Eq. (4) and PCFa is the maximum of PCF ai .
5.1.2. Optimization algorithm
For a MOD problem, the design task is to search for a cloud of
best trade-off solutions, called Pareto front, in the design space

[47]. Further decision can be made based on the Pareto front

according to the designers subjective preferences later on. In our
study, the multiobjective particle swarm optimization (MOPSO)
algorithm is used to search for the Pareto front of the two
conicting objectives of SEA and PCF. The MOPSO algorithm that
incorporates the mechanism of crowding distance computation
has been successfully used in a wide variety of optimization tasks
including crashworthiness designs [34,38] due to its relatively
low computational cost and fast convergence [48]. Additionally,
compared with other multiobjective optimization algorithms
such as NSGA-II [49], PAES [50], MOPSO was shown to be able
to generate the best set of Pareto solutions close to the true Pareto
front [48]. The details of MOPSO can be consulted from Refs.
[48,51].
The owchart of the implementation of MOD for the MCT
tubes with load angle uncertainty based on the proposed
approach is explained in Fig. 18.
5.2. Results and discussion
5.2.1. MOD of MCT tube without load angle uncertainty
The MOD problems of the MCT tube under four different
impact angles are rst solved based on the corresponding RS
models. The MOPSO algorithm is used to search for the optimal
solutions to the MOD problems with a population size of 50. After
iterating for 20 generations, the Pareto fronts for the crashworthiness optimization problems of the MCT tube under different load
angles dened as Eq. (17) are obtained and shown in Fig. 19. It is
clearly seen that  SEA and PCF are negatively correlated for all

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

115

Table 6
Ideal optimums of the two single objective functions for the MCT tube under
different load angles.
Load angle
(deg)
0
10
20
30

Fig. 18. Flowchart of the implementation of the MOD for the MCT tubes with load
angle uncertainty.

Fig. 19. Pareto fronts of PCF and  SEA for the MCT tube under different load
angles.

four load angles, which means increase of SEA (decrease of SEA)

always leads to increase of PCF and decrease of PCF will result in
reduced SEA regardless of load angle.
The optimum tube congurations (see columns 3 and 4) and
the corresponding ideal optimal values of the two single objective
functions SEA (see column 5) and PCF (see column 6) predicted by
the RS models are provided in Table 6. Here, the ideal means
only a single objective (SEA or PCF) is pursued, i.e., a singleobjective optimization design (SOD) problem is solved. These
eight single ideal optimums listed in Table 6 correspond to the
eight special points in the Pareto space, which lie at each end of
the Pareto curves and marked as solid star in Fig. 19. We can see
that for these two single objective functions, there exist two
different optimal solutions for a specic load angle. In three cases
(i.e., ideal max SEA for a 01, ideal min PCF for a 201 and 301),
the RS models give the same optimum congurations which are
exactly at the sampling design points (design Nos. 16 and 4 in
Table 4, respectively). Thus, no extra FEA is required for a
conrmation run. For the other ve cases, the RS models give

Single objective

t (mm)

y (deg)

SEA (kJ/kg)

PCF (kN)

Ideal
Ideal
Ideal
Ideal
Ideal
Ideal
Ideal
Ideal

2.5000
1.0000
2.5000
1.0000
2.2228
1.0000
1.5742
1.0000

7.0000
6.2925
4.7611
5.9742
7.0000
7.0000
7.0000
7.0000

34.403
13.351
21.701
11.247
9.384
5.910
6.293
3.352

108.270
23.496
99.138
18.453
34.213
15.381
16.318
9.937

Max. SEA
Min. PCF
Max. SEA
Min. PCF
Max. SEA
Min. PCF
Max. SEA
Min. PCF

different optimum congurations. For these cases, extra FEA are

necessary for design conrmation. More importantly, as seen
from Table 6, the optimum congurations of the MCT tube for
the same objective function are generally different at different
load angles. For instance, the optimum wall thickness of the tube
to maximize SEA at a 01 is 2.5 mm, while the optimum wall
thickness for SEA maximization at a 301 is 1.5742 mm. This
implies an optimal solution for a specic load angle does not
necessarily yield the best solution when the load angle changes
such as in an automobile crash event. Thus, it is very important
and of practical meaning to include the load angle uncertainty
effect in the optimization process of such energy absorbers, so as
to obtain a more robust design for as much as possible load
angles. This is discussed in the next section.
In practical applications of MOD, one can impose any feasible
constraint on the Pareto front to nail down only a single optimum
design. Here, a constraint of PCF is set as 70 kN as suggested in
Ref. [40]. The Pareto points corresponding to the optimal designs
for load angle 01 and 101 are marked as solid square and solid
circle in Fig. 19, and their detailed design parameters are listed in
Table 7. Meanwhile, FE models for the optimal designs of the MCT
tube are established. The comparison between the FEA results and
the approximation results are shown in Table 7. It can be found
that the errors between FEA results and the RS models are less
than 0.4% for load angle a 01. This indicates that the quartic
polynomial functions for MCT tube responses under pure axial
loading are accurate. We can also nd that the errors of SEA and
PCF of the optimal design for load angle a 101 are bigger than
that of pure axial loading (a 01), respectively. More importantly,
the optimum tube conguration given by the RS models with
maximum SEA (19.671 kJ/kg) for load angle a 101, however,
yields a PCF of 72.899 kN by FEA. This load exceeds the constrained load level (70 kN) and makes the design actually infeasible. Thus, it is concluded that design validation through FEA is
required to approve any optimal design obtained by using the
surrogate models.

5.2.2. MOD of MCT tube with load angle uncertainty

To include the load angle uncertainty effect in the MOD of the
MCT tubes under oblique impact, two design cases are conducted
herein for the MOD problem dened as Eq. (18). In this study, it is
assumed that the load angle a varies between 01 and 301. Four
representative load angles ai 01, 101, 201 and 301 are considered
in each design case for SEAa calculation using Eq. (4). In design
case 1, the weighting factors W ai for each load angle are taken
3
3
3
3
equally, i.e. W 0 W 10 W 20 W 30 0:25. In design case 2,
3
more weights are given to larger load angles with W 0 0:1,
103
203
303
W 0:2, W 0:3 and W 0:4. The values of SEAa in each
design case are obtained using Eq. (4) based on the DoE data in
Table 4 and the weighting factors given above. The quartic
polynomial functions are used again to create the RS models of

116

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

SEAa for each design case as shown in Fig. 20. It should be noted
again that the maximum PCF occurs when the tube is loaded pure
3
axially (a 01), i.e. PCF a PCF 0 . Therefore, the RS models for PCFa
prediction are the same as that shown in Fig. 17(a) for both
design cases.
By using the RS models and the MOPSO algorithm, Pareto
fronts of the two design cases are obtained and shown in Fig. 21.
It is clearly shown that a concave Pareto front curve exists for
each design case, which means the trade-off relationship between
SEAa and PCFa. On the other hand, the range of SEAa is different in
the two design cases, the SEAa value in design case 2 is relatively
small since more weights were given to larger load angles under
which the tube has less SEA.
The optimum tube congurations (see columns 3 and 4) and
the corresponding ideal optimal values of the two single objective
functions SEAa (see column 5) and PCFa (see column 6) predicted
by the RS models are provided in Table 8 and marked as solid star
in Fig. 21. Notice that, for both design cases, the maximum SEAa of
the tube occurs at the value of taper angle locating at its upper
boundary (y 7.01). This indicates that larger taper angle is
benecial for increasing SEA of MCT tubes under oblique impact
loading as demonstrated previously by parametric study. This
observation is the same as the optimal results of the tapered
circular tubes of Refs. [37] and [38]. It is interesting to note that
even though equal weighting factors were assigned to the four
load angles to calculate SEAa in design case 1, the optimum tube
conguration (y 7.01 and t 2.5 mm) to maximize SEAa in design
case 1 is the same as that to maximize SEA when the load angle
a 01 (refer to Table 6). This can be explained as follows: since the
absolute value of tube SEA at a small load angle is much higher
than at a large load angle (e.g., the MCT tube with y 7.01 and
t 2.5 mm gives SEA of 34.329 and 5.916 kJ/kg at load angle a 01
and 301, respectively), the SEA value obtained at a small load
angle plays a leading role in SEAa calculation and dominates the
Table 7
Optimal designs of the MCT tube with PCF constrained under 70 kN at different
load angles.
Load angle
(deg)

t
(mm)

SEA (kJ/kg)

0
10

FEA

Fig. 21. Pareto fronts of the MCT tube for the two design cases with load angle
uncertainty.

Table 8
Ideal optimums of the two single objective functions for MCT tube with load angle
uncertainty.

PCF (kN)

(deg)
RS
model

optimization process. This can also be observed from the response

surfaces of SEA at a 01 (Fig. 16(a)) and SEAa in design case 1
(Fig. 20(a)). The two RSs show similar shape and both reach their
vertexes at y 7.01 and t2.5 mm. When more weights are given
to large load angles as in design case 2, however, the optimal wall
thickness to maximize SEAa is 2.3717 mm, which can be considered as a compromise between the separate optimal values for
the four load angles as listed in Table 6. We can then conclude
that the selection of weighting factors for different load angles
can affect the location of the optimal solution to SEAa maximization of the MCT tube, and thus plays an important role in the
crashworthiness optimization design of such tubes with load

Error
(%)

1.8063 6.9760 24.377 24.458 0.33

2.0605 5.7733 19.889 19.671 1.11

RS
model

FEA

Error
(%)

69.768 69.563 0.29

69.929 72.899 4.07

Design case

Single objective

t (mm)

y(deg)

SEAa (kJ/kg)

PCFa (kN)

Ideal
Ideal
Ideal
Ideal

2.5000
1.0000
2.3717
1.0000

7.0000
6.1725
7.0000
6.2118

16.422
8.544
11.790
6.7759

108.270
23.397
102.860
23.433

Max. SEAa
Min. PCFa
Max. SEAa
Min. PCFa

Fig. 20. Quartic RSs of SEAa for the MCT tube with load angle uncertainty. The bold black dots in the plots show the training points. (a) design case 1 and (b) design case 2.

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

Table 9
Optimal designs of the MCT tube with PCF constrained under 70 kN with load
angle uncertainty.
Design
case

t (mm)

y(deg) SEAa (kJ/kg)

RS
model

1
2

1.8060
1.8067

7.0000 14.283
6.9943 11.077

FEA

PCFa (kN)
Error
(%)

14.518 1.62
11.246 1.50

RS
model

FEA

Error
(%)

69.676
69.741

69.029 0.94
69.442 0.43

angle uncertainty. In designing energy absorbers in real applications such as the automobile front side rails, the weighting factors
for different load angles should be decided based on the crash
probabilities of each angle in a statistical sense. However, this is
out of the scope of the current paper and could be studied
separately.
Similarly, a larger taper angle is preferred for minimizing PCFa
in both design cases as listed in Table 8. In contrast, the minimum
PCFa occurs at the value of wall thickness locating at its lower
boundary (t 1.0 mm). This conrms the ndings in the parametric study that SEA and PCF show obvious trade-off characteristics with respect to the wall thickness of the MCT tubes.
The Pareto optimal designs of the MCT tube with load angle
uncertainty and PCF constrained under the level of 70 kN are
marked as solid square and solid circle in Fig. 21 for design case
1 and 2, respectively. Meanwhile, the detailed design parameters
for these optimal designs and their corresponding FEA results are
summarized in Table 9. We can nd that the errors between FEA
results and the results obtained by using the RS models are less
than 2%. The error of SEAa is bigger than that of PCFa for the
optimal design of the MCT tubes in both design cases. This implies
3
that the RS model for PCFa, i.e., PCF 0 (Fig. 17a) is more accurate
than those for SEAa (Fig. 20) in the current design cases. Moreover, the optimal design parameters in the two design cases are
very close to the optimal values in the case when load angle a 01
3
(refer to Table 7). This again conrms that SEA0 is the major
contributor in SEAa calculation and the PCF at load angle a 01 is
the maximum among PCF ai for all four load angles considered in
3
the current study, i.e., PCF a PCF 0 .

117

also been presented for the MCT tube with and without load angle
uncertainty effect. It is found that SEA and PCF could be conicting with each other and generally not obtain a simultaneous
optimum as demonstrated in other studies, e.g. Refs. [29,38]. It is
also found that the optimum tube congurations are in general
different for different load angles, either with or without constraint on the PCF level. Therefore, it is necessary to include the
load angle uncertainty effect in the optimization process of such
energy absorbers to obtain a more robust design under oblique
impact loading. Last but not the least, the MOD results of two
design cases with load angle uncertainty show that the selection
of weighting factors for different load angles can affect the
location of the optimal solution to maximize SEAa of the MCT
tube, and thus plays an important role in the crashworthiness
optimization design of such energy absorbers with load angle
uncertainty.

Acknowledgments
This work is supported by the National Natural Science
Foundation of China (no. 50905024, 51105053), Liaoning Provincial Natural Science Foundation of China (no. 20102026) and the
Research Fund for the Doctoral Program of Higher Education of
China (no. 20090041120032, 20110041120022). The nancial
supports are gratefully acknowledged.

Appendix
The quartic polynomial functions of SEA and PCF for the MCT
tube under different dynamic load angles in this study are
provided below
3

SEA0 t, y 0:0020368 0:15702y 0:50283t0:11462y

3

0:45353yt0:1008t 2 0:028059y 0:54254y t

4

0:8581yt 2 0:068249t 3 0:11887y 0:02737y t

2 2

0:0021246y t 0:47783yt 0:098441t

PCF 0 t, y 0:135720:024578y 0:88987t0:35094y

3

A:1

2
2

2:75yt0:15259t 0:060687y 4:4467y t

2

2:7206yt 0:085946t 0:30209y 2:2473y t

6. Conclusions

1:323y t 2 0:96754yt 3 0:21236t 4

The crashworthiness characteristics of axisymmetric thinwalled square (ATS) tubes under oblique impact loading with
four different congurations (SCS, SCT, MCS and MCT) have been
investigated by using the nonlinear FE code LS-DYNA. The
numerical results show that the SEA of multi-cell (MC) tubes
are higher than that of the single-cell (SC) tubes for the complete
range of load angle (01401) under investigation. It is also found
that the PCF is lower for the tapered tubes than the straight tubes,
especially in MC cases. Thus, we focus on the ATS tubes with MCT
conguration for the crashworthiness optimization under oblique
impact loading.
Then, based on the DoE results, parametric study has been
presented for the MCT tubes. It is found that the load angle, as
well as wall thickness and taper angle have signicant effects on
SEA and PCF of the MCT tubes. A larger taper angle is preferred for
both SEA maximization and PCF minimization. The wall thickness,
however, shows conicting effects on SEA and PCF. Moreover, the
quartic polynomial functions have been found as the suitable
response surface (RS) models for the MCT tubes SEA and PCF
prediction under different load angles.
Multiobjective optimization design (MOD) using the multiobjective particle swarm optimization (MOPSO) algorithm has

A:2

SEA10 t, y 0:00573 0:58025y 0:71238t0:68918y

3

2:7837yt0:1726t 0:1896y 6:4973y t

4

3:6358yt 2 0:075538t 3 0:59207y 4:168y t

2 2

1:5599y t 2:0382yt 3 0:21818t 4

A:3
2

PCF 10 t, y 0:077622 0:068823y 0:61687t0:3698y

3

1:3159yt 0:35242t 2 0:080819y 2:9103y t

2

0:37186yt 0:08195t 0:30563y 1:6406y t

2 2

0:51262y t 0:042206yt 0:12975t

A:4

SEA20 t, y 0:016202 2:8036y0:39783t2:2671y

3

8:1525yt 1:7804t 2 1:1905y 7:7776y t

4

3:8253yt 2 0:46035t 3 1:0129y 2:8204y t

2 2

8:5627y t 2:3938yt 1:3437t

A:5

PCF 20 t, y 0:14260:14929y 1:129t0:011866y

2

2
2

0:85647yt0:48862t 0:012199y 0:59569y t

118

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

1:6574yt 2 0:21024t 3 0:0057241y 0:62923y t

2

0:45603y t 2 0:80693yt 3 0:42761t 4

A:6

SEA30 t, y 0:045075 0:053495y0:15042t0:40109y

3

2:7856yt 1:3448t 0:0030794y 0:054892y t

2

8:0949yt 0:31367t 0:50827y 1:9286y t

2

2:7183y t 2 6:2059yt 3 1:0497t 4

A:7
2

PCF 30 t, y 0:134890:081497y 1:1036t 0:18318y

3

0:72237yt0:36532t 0:00096215y 0:26089y t

4

1:4303yt 2 0:19145t 3 0:24008y 0:25339y t

2

0:81597y t 2 0:37107yt 3 0:30884t 4

A:8

The quartic polynomial functions of SEAa for the MCT tube

with load angle uncertainty in the two design cases of this study
are provided below.
Design case 1:
2

SEAa t, y 0:00042445 0:67448y 0:45923t0:64892y

3

1:9984yt 0:27791t 0:2443y 3:534y t

4

1:3373yt 2 0:033143t 3 0:45441y 0:77975y t

2 2

1:985y t 0:058025yt 0:17133t

A:9

Design case 2:
2

SEAa t, y 0:0018769 0:97277y 0:32273t0:96457y

3

2:5358yt 0:65597t 0:37212y 4:6447y t

4

0:64983yt 2 0:13391t 3 0:65085y 0:50234y t

2 2

3:3447y t 1:1453yt 0:46221t

A:10

References
[1] Alexander JM. An approximate analysis of the collapse of thin cylindrical
shells under axial loading. Quarterly Journal of Mechanics and Applied
Mathematics 1960;13(1):105.
[2] Abramowicz W, Jones N. Dynamic progressive buckling of circular and square
tubes. International Journal of Impact Engineering 1986;4(4):24370.
[3] Guillow SR, Lu GX, Grzebieta RH. Quasi-static axial compression of thinwalled circular aluminum tubes. International Journal of Mechanical Sciences
2001;43:210323.
[4] Wierzbicki T, Abramowicz W. On the crushing mechanics of thin-walled
structures. Journal of Applied Mechanics 1983;50:72734.
[5] Mamalis AG, Manolakos DE, Baldoukas AK, Viegelahn GL. Energy dissipation
and associated failure modes when axially loading polygonal thin-walled
cylinders. Thin-Walled Structures 1991;12:1734.
[6] Abramowicz W, Jones N. Dynamic axial crushing of square tubes. International Journal of Impact Engineering 1984;2(2):179208.
[7] White MD, Jones N, Abramowicz W. A theoretical analysis for the quasi-static
axial crushing of top-hat and double-hat thin-walled sections. International
Journal of Mechanical Sciences 1999;41:20933.
[8] Hanssen AG, Langseth M, Hopperstad OS. Static and dynamic crushing of
circular aluminium extrusions with aluminium foam ller. International
Journal of Impact Engineering 2000;24:475507.
[9] Hanssen AG, Langseth M, Hopperstad OS. Static and dynamic crushing of
square aluminium extrusions with aluminium foam ller. International
Journal of Impact Engineering 2000;24:34783.
[10] Seitzberger M, Rammerstorfer FG, Degischer HP, Gradinger R. Crushing of
axially compressed steel tubes lled with aluminium foam. Acta Mechanica
1997;125(14):93105.
[11] Seitzberger M, Rammerstorfer FG, Gradinger R, Degischer HP, Blaimschein M,
Walch C. Experimental studies on the quasi-static axial crushing of steel
columns lled with aluminum foam. International Journal of Solids and
Structures 2000;37(30):412547.
[12] Aktay L, Toksoy A, Guden M. Quasi-static axial crushing of extruded
polystyrene foam-lled thin-walled aluminum tubes: Experimental and
numerical analysis. Materials and Design 2006;27(7):55665.
[13] Ghamarian A, Zarei HR, Abadi MT. Experimental and numerical crashworthiness investigation of empty and foam-lled end-capped conical tubes. ThinWalled Structures 2011;49(10):13129.

[14] Chen WG, Wierzbicki T. Relative merits of single-cell,multi-cell and foamlled thin-walled structures in energy absorption. Thin-Walled Structures
2001;39(4):287306.
[15] Kim H-S. New extruded multi-cell aluminum prole for maximum crash
energy absorption and weight efciency. Thin-Walled Structures 2002;40:
31127.
[16] Zhang X, Cheng G. A comparative study of energy absorption characteristics
of foam-lled and multi-cell square columns. International Journal of Impact
Engineering 2007;34(11):173952.
[17] Tang Z, Liu S, Zhang Z. Energy absorption properties of non-convex multicorner thin-walled columns. Thin-Walled Structures 2012;51:11220.
[18] Reyes A, Langseth M, Hopperstad OS. Crashworthiness of aluminum extrusions subjected to oblique loading experiments and numerical analyses.
International Journal of Mechanical Sciences 2002;44:196584.
[19] Reyes A. Square aluminum tubes subjected to oblique loading. International
Journal of Impact Engineering 2003;28(10):1077106.
[20] Reyes A, Hopperstad OS, Langseth M. Aluminum foam-lled extrusions
subjected to oblique loading: experimental and numerical study. International Journal of Solids and Structures 2004;41(56):164575.
[21] Han DC, Park SH. Collapse behavior of square thin-walled columns subjected
to oblique loads. Thin-Walled Structures 1999;35(3):16784.
[22] Reid SR, Reddy TY. Static and dynamic crushing of tapered sheet metal tubes
of rectangular cross-section. International Journal of Mechanical Sciences
1986;28(9):62337.
[23] Nagel G. A numerical study on the impact response and energy absorption of
tapered thin-walled tubes. International Journal of Mechanical Sciences
2004;46(2):20116.
[24] Nagel G, Thambiratnam D. Computer simulation and energy absorption of tapered
thin-walled rectangular tubes. Thin-Walled Structures 2005;43(8):122542.
[25] Nagel G, Thambiratnam D. Dynamic simulation and energy absorption of
tapered thin-walled tubes under oblique impact loading. International
Journal of Impact Engineering 2006;32(10):1595620.
[26] Ahmad Z, Thambiratnam DP, Tan ACC. Dynamic energy absorption characteristics of foam-lled conical tubes under oblique impact loading. International Journal of Impact Engineering 2010;37(5):47588.
[27] Hou S, Li Q, Long S, Yang X, Li W. Design optimization of regular hexagonal
thin-walled columns with crashworthiness criteria. Finite Elements in
Analysis and Design 2007;43(6-7):55565.
[28] Hou S, Li Q, Long S, Yang X, Li W. Multiobjective optimization of multi-cell
sections for the crashworthiness design. International Journal of Impact
Engineering 2008;35(11):135567.
[29] Hou S, Li Q, Long S, Yang X, Li W. Crashworthiness design for foam lled thinwall structures. Materials and Design 2009;30(6):202432.
[30] Liu Y. Optimum design of straight thin-walled box section beams for crashworthiness analysis. Finite Elements in Analysis and Design 2008;44(3):13947.
[31] Liu Y. Crashworthiness design of multi-corner thin-walled columns. ThinWalled Structures 2008;46(12):132937.
[32] Zarei H, Kroger M. Optimization of the foam-lled aluminum tubes for crush
box application. Thin-Walled Structures 2008;46(2):21421.
[33] Zarei H, Kroger M. Optimum honeycomb lled crash absorber design.
Materials and Design 2008;29(1):193204.
[34] Yin H, Wen G, Hou S, Chen K. Crushing analysis and multiobjective
crashworthiness optimization of honeycomb-lled single and bitubular
polygonal tubes. Materials and Design 2011;32(8-9):444960.
[35] Zarei H, Kroger M. Bending behavior of empty and foam-lled beams: Structural
optimization. International Journal of Impact Engineering 2008;35(6):
5219.
[36] Zhang Z, Liu S, Tang Z. Design optimization of cross-sectional conguration of
rib-reinforced thin-walled beam. Thin-Walled Structures 2009;47(89):86878.
[37] Acar E, Guler MA, Gerc- eker B, Cerit ME, Bayram B. Multi-objective crashworthiness optimization of tapered thin-walled tubes with axisymmetric
indentations. Thin-Walled Structures 2011;49(1):94105.
[38] Hou S, Han X, Sun G, Long S, Li W, Yang X, et al. Multiobjective optimization
for tapered circular tubes. Thin-Walled Structures 2011;49(7):85563.
[39] Hallquist J. LS-DYNA keyword users manual, version: 970. Livermore Software Technology Corporation; 2003.
[40] Kurtaran H, Eskandarian A, Marzougui D, Bedewi NE. Crashworthiness design
optimization using successive response surface approximations. Computational Mechanics 2002;29(45):40921.
[41] Santosa SP, Wierzbicki T, Hanssen AG, Langseth M. Experimental and
numerical studies of foam-lled sections. International Journal of Impact
Engineering 2000;24(5):50934.
[42] Langseth M, Hopperstad OS. Static and dynamic axial crushing of square
thin-walled aluminium extrusions. International Journal of Impact Engineering 1996;18(78):94968.
[43] ANSYS I: Ansys documentation. ANSYS LS-DYNA users guide, version 8.1;
2004.
[44] Zhang X, Cheng G, Zhang H. Theoretical prediction and numerical simulation
of multi-cell square thin-walled structures. Thin-Walled Structures
2006;44(11):118591.
[45] Myers RH, Montgomery DC. Response surface methodology:process and
product optimization using designed experiments. New York: Wiley; 1995.
[46] Yang RJ, Gu L. Experience with approximate reliability-based optimization
methods. Structural and Multidisciplinary Optimization 2004;26(12):
1529.

C. Qi et al. / Thin-Walled Structures 59 (2012) 103119

[47] Papalambros PY, Wilde DJ. Principles of optimal design: modeling and
computation. New York: Cambridge; 2000.
[48] Coello Coello CA, Pulido GT, Lechuga MS. Handling multiple objectives with
particle swarm optimization. IEEE Transactions on Evolutionary Computation
2004;8(3):25679.
[49] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjective
genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation
2002;6(2):18297.

119

[50] Knowles JD, Corne DW. Approximating the nondominated front using the
Pareto archived evolution strategy. Evolutionary Computation 2000;8:
14972.
[51] Raquel CR, Naval Jr. PC. An effective use of crowding distance in multiobjective particle swarm optimization. Association for Computing Machinery;
2005:25764.