International Journal of Mechanical Sciences 106 (2016) 107–116

Contents lists available at ScienceDirect

International Journal of Mechanical Sciences

journal homepage: www.elsevier.com/locate/ijmecsci

Bending collapse of square tubes with variable thickness

Xiong Zhang a,c,n, Hui Zhang b, Zong Wang a
a
Department of Mechanics, Huazhong University of Science and Technology, Wuhan 430074, Hubei, PR China
b
School of Mechanical Engineering and Automation, Wuhan Textile University, Wuhan 430073, Hubei, PR China
c
Hubei Key Laboratory of Engineering Structural Analysis and Safety Assessment, Luoyu Road 1037, Wuhan 430074, PR China

art ic l e i nf o a b s t r a c t

Article history: This paper aims to study the bending collapse behavior of thin-walled square tubes with variable
Received 8 October 2015 thickness in the cross-section. Three-point bending test is carried out for thin-walled square beams with
Received in revised form different thickness for flanges and webs. The relative strength of the flanges versus the web plates is
25 November 2015
found to have significant influence on the force response and deformation pattern. Numerical simulation
Accepted 10 December 2015
of the experimental test is also performed and the numerical results compared very well with the
Available online 17 December 2015
experimental results. Moreover, the difference between the quasi-static and dynamic responses of the
Keywords: square tubes under transverse loading is analyzed. A response surface method is finally employed to
Bending optimize two thin-walled beams under impact loading. The optimization results show that putting less
Square tube
material in flanges and more in the web plates is an efficient way to improve the bending resistance of
Energy absorption
the beams under transverse loading. Adopting graded thickness in the web plates is an effective and
Variable thickness
Optimization design promising approach to further increase the energy absorption efficiency of the beams.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction
Thin-walled metallic tubes are widely applied as structural
components in various engineering fields, especially in moving
vehicles such as automobiles, ships, aircrafts and etc. The perfor-
mance of them under accidental impact events is therefore of
interest to the researchers for the occupant safety considerations.
The energy absorption characteristics of metallic tubes under
various load conditions [1] have received extensive investigations
including axial crushing [2–5], lateral crushing [6–9], transverse
bending [10–16] and etc. Various approaches were proposed by
researchers to improve the performance of them under impact
loading. For example, introducing multi-cells [4,5], foam filler [10–
13], grooves [17,18], reinforced ribs [19] and et al. The purpose of
these methods is generally to increase the energy absorption
efficiency of them during deformation and to attenuate the peak
force which might cause damage or injury.
Recently, introducing thickness variables in thin-walled struc-
tures to improve their crashworthiness performance has attracted
increasing interests in the researcher community [20–27]. The
reason is twofold: first, the manufacture of such structures
becomes more and more cheap and convenient. New emerging
material processing technology, such as tailor rolled blank (TRB)
n
Corresponding author at: Department of Mechanics, Huazhong University of
Science and Technology, Wuhan 430074, Hubei, PR China. Tel.: þ86 27 87543538;
fax: þ86 27 87543501.
E-mail address: zhangxiong@hust.edu.cn (X. Zhang).
[20] or tube tapering machine [21], can easily produce plates or
tubes with graded thickness. Actually, early studies were per-
formed by researchers [28,29] to investigate the inversion of cir-
cular tubes with variable thickness. However, the uncontrollable
thickness distribution hampered further investigations and appli-
cations. Second, the performance of structures with variable
thickness is definitely not worse than that of those with uniform
thickness due to larger design domain. This has been validated by
recent experimental or numerical studies that investigated thin-
walled tubes with graded thickness in the cross-section [22] or
along the longitudinal direction [23–27].
Under transverse loading, the bending collapse mechanisms of
square tubes are similar as those subjected to axial crush. The
energy is primarily dissipated by bending along static hinge lines,
rolling at moving hinge lines and stretching in toroidal regions
[2,14–16]. The theoretical model proposed by Kecman [14] is
shown in Fig. 1(a). It can be noticed that the deformation is not
uniformly distributed in different regions of the tube. For axial
crush, the contribution of each plate in energy dissipation of a
square tube is approximately equal, while the webs of the tube
apparently dissipate more energy than the top and bottom flanges
during bending since the latter just bends along static hinge lines.
Therefore, a natural idea to increase the energy absorption effi-
ciency of the tube is to put more material in the webs and
accordingly reduce that in top and bottom flanges.
In the present work, the bending collapse of square tubes with
variable thickness in the cross-section is investigated. Three-point
bending test for square tubes with uniform but different wall

http://dx.doi.org/10.1016/j.ijmecsci.2015.12.006
0020-7403/& 2015 Elsevier Ltd. All rights reserved.
108 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116

Fig. 1. (a) Collapse mechanism proposed by Kecman, (b) Web plates of square tubes.

Fig. 2. Specimens of square tubes with different wall thickness for side plates.

thickness in the four side plates are firstly conducted. The non-
linear explicit finite element code LS-DYNA is then employed to
simulate the test and the numerical model is validated by the
experimental results. The dynamic analyses of square tubes under
transverse loading are then performed and the influence of load
speed is analyzed. Two relevant optimization problems are for-
mulated for square tubes under impact loading and the optimal
designs of thickness variables (distribution of the material) in the
cross-section are finally found out by a surrogate optimization
method. The advantage of tubes with variable thickness over those
with uniform thickness is revealed and some suggestions are
offered for design of structures against transverse loading.
2. Specimen preparation and experimental setup
Square tubes with uniform but different wall thickness in the
four side plates are experimentally investigated in this section. To
keep the symmetry of structure during deformation, same thick-
ness is assigned to the opposite side plates of the tube. The thin-
walled square beam specimens are fabricated by using commercial
square tubes with inside width b¼26 mm and wall thickness
t¼2 mm. The tubes are firstly cut to L ¼250 mm and the thickness
of two opposite side plates of the tubes is then reduced by a Wire
cut Electrical Discharge Machining (WEDM) technique. The pre-
cision of the cutting is below 20 mm. The original square tube and
the specimen after cutting are shown in Fig. 2. Three different wall
thicknesses ta: 0.8, 1.4 and 2 mm are investigated here and three
specimens are prepared for each case. The sectional dimensions of
the specimens are also presented in Fig. 2.
The structural material of the square tubes is aluminum alloy
AA1100-O. The tensile test is performed on a 10 kN capacity Zwick
Z010 universal testing machine to obtain the stress–strain relation
of the material. The specimens are taken parallel to the axial
direction of square tubes and standard dimensions as specified in
the ASTM E8M-04 for tensile test are adopted [30]. The tests are
conducted in room temperature and the engineering stress–strain
characteristics of the material are shown in Fig. 3. The mechanical
properties of it are given as follows: Young's modulus E ¼68.0 GPa,
initial yield stress σy ¼30.5 MPa, the ultimate stress σu ¼90.5 MPa,
Poisson's ratio ν ¼0.33.
Quasi-static three point bending tests are also performed in
ZWICK Z010 machine with computer control and data acquisition
systems. The experimental setup and geometry for three-point
bending test are shown in Fig. 4. The diameter of the cylindrical
punch and supports is 10 mm and the span between the supports
 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116
Fig. 3. Engineering stress–strain curve of AA1100-O.
Fig. 4. Experimental setup and geometry for three point bending test.
is D¼200 mm. The testing is displacement controlled with the
punch being moved vertically downward to load the specimens
and the loading velocity of the punch is 0.5 mm/s. As shown in
Fig. 2, if ta atb, the bending resistance will be different when
loaded around x- or y-axis. Therefore, both cases are tested. The
test plan is given in Table 1.
3. Finite element modeling
The simulation of the experimental test is performed by com-
mercial explicit finite element code LS-DYNA. As shown in Fig. 5,
the thin-walled square beam is loaded by a rigid cylindrical punch
with a prescribed velocity in the transverse direction and sup-
ported by two stationary rigid cylinders. Due to variable thickness
in the cross-section, solid elements instead of shell elements are
employed to discretize the structure. The beams are discretized by
58464 eight-node brick elements with one point integration and a
total of 23040 Belytschko-Tsay 4-node shell elements are
employed for the punch and supports. Three elements are
assigned along the thickness direction and the characteristic size
of the mesh along the length direction is 2.0 mm in the two end
regions. The center segment where the major bending deforma-
tion is expected is modeled with finer mesh size of 1.0 mm. By
109
Table 1
Test plan for three-point bending.
Test # ta (mm) tb (mm) Bending axis
1 0.8 2.0 x
2 0.8 2.0 x
3 0.8 2.0 y
4 1.4 2.0 x
5 1.4 2.0 x
6 1.4 2.0 y
7 2.0 2.0 x or y
8 2.0 2.0 x or y
9 2.0 2.0 x or y
Fig. 5. Representative finite element model for three-point bending test.
mesh convergence analysis, these mesh configurations are found
to achieve enough accuracy.
The structural material AA1100 O is modeled by piecewise
linear elasto-plastic material #24 in LS-DYNA [31] and strain rate
effect is not considered since it is strain rate insensitive. Automatic
single-surface and automatic surface-to-surface contact are
employed to account for the self-contact of the beam and the
contact between the beam and the rigid punch or supports,
respectively. The friction coefficient in all the contacts is assumed
to be 0.3. The quasi-static loading is impossible to be simulated in
the real time scale and the loading rate has to be increased to
accelerate the analysis. During the simulation, the principles pro-
posed by Santosa et al. [32] are followed, just as done in Ref. [22].
4. Experimental and numerical results
The characteristics of the deformation patterns and punch
force–displacement responses of the specimens are analyzed here.
Fig. 6 presents the deformation pattern of the square tubes. A
general view is given in Fig. 6(a) and more details are shown in
Fig. 6(b). The deformation pattern is found to be highly correlated
with the wall thickness of the plates in the square section. A
parameter FW is defined as the ratio of the thickness of flanges to
that of webs. The parameter FW of the specimens is therefore
equal to ta/tb or tb/ta considering bending around different axes. A
total of five FW values: 0.4, 0.7, 1.0, 1.43 and 2.5 are
considered here.
When the webs are relatively stronger or weaker than the
flanges, the deformation patterns are quite different. As shown in
Fig. 6(b), for test #1 with FW ¼0.4, the inward fold deformation is
small and multiple folds are formed and propagate along the axial
direction. It is interesting to find that this type of deformation
mode is similar to the tubes with foam-fillers [10,13]. According to
the investigations by Santosa [13], the lateral support of the foam
resulted in a higher bending strength and retarded the formation
of inward fold, which prevented the drop of load carrying capacity
of the section during deformation. This special deformation mode
formed more plastic hinge lines and therefore dissipated more
bending energy. Apparently, the thicker web plates also provide
stronger lateral support and hamper the inward collapse of the top
flange. An illustration for the formation of multiple folds along the
axial direction in square tubes with foam filler or thicker side
supports is presented in Fig. 7. During bending of the section, the
110 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116

Fig. 6. Deformation patterns of thin-walled square beams: experimental and numerical.

Fig. 7. An illustration of deformation for square tube with foam filler and thicker side support.
 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116 111

top flange is subjected to compression deformation. The localized

inward collapse in the middle region is the typical buckling mode
of square tubes, while if the inward collapse is restricted by the
lateral support, the deformation switches to another buckling
mode: "multiple-fold mode". The multiple folds formed in the top
flange plate are major differences between these two modes of
thin-walled square beams.
The multiple-fold mode is developed only for FW ¼0.4 and the
square tubes switch to inward "single fold mode" for FW Z0.7.
With the increase of FW value, the extent of inward intrusion of
the top flange to the webs is increased gradually. As shown in the
left of Fig. 6(b), the intrusion displacement increases up to almost
the half height of the square beam for FW¼ 2.5. Besides the top
flange, the deformations of the bottom flange and web plates also
show some interesting variations with the change of FW value. As
shown in the right of Fig. 6(b), the bended specimens are reversely
placed with the bottom flange in the top and viewed from the
axial direction. It can be observed that the bottom flange is warped
and similar as the top flange, it deforms inwardly toward the
section. The curvature is very high for small FW value (for
instance, FW¼ 0.4) and it decreases gradually with the increase of
FW. When FW ¼2.5, the bottom flange almost undergoes no warp
and bends only. With the inward moving of top and bottom
flanges, the web plates protrudes outward in the same time. The
extent of outward protrusion is also found to increase with the
increase of FW value.
The deformation modes of the specimens obtained by numer-
ical simulation are also presented in Fig. 6(b). We can find that the
simulated patterns are almost completely the same as the
experimental results. For FW¼0.4, both the special multiple folds
in top flange and the dramatic warp of bottom flange are captured
by the numerical model. The variation tendencies of deformation
features with the increase of FW value, including extent of inward
intrusion of top flange, warp curvature of bottom flange and the
outward protrusion of the webs, are all well reflected in the
numerical results.
The punch force–displacement responses of the square tube
specimens are shown in Fig. 8. It is noted that the numerical
results compare fairly well with the measured data. The features of
the force response curves also vary with the variation of FW ratio.
The curves are more flat for small FW values while the forces drop
quickly after an initial peak for large FW values. That is, the beams
with small FW values are more resistive to bending loads. The
energy absorption indices of the specimens, including total energy
dissipation E, peak force Pmax, mean crushing force Pm, SEA (spe-
cific energy absorption) and crushing force efficiency CE (the ratio
of Pm to Pmax), are listed in Table 2. The mean crushing forces are
calculated at punch distance of δ ¼38.0 mm.
It can be noticed that the SEA values of specimens with
FWr 1.0 are comparable, while those of specimens with FW4 1.0
are remarkably lower than others. The CE index gets the highest
value 0.9 for specimens with FW¼0.4 and it decreases gradually Fig. 8. Punch force–displacement responses for thin-walled square beams.
with the increase of FW ratio. The data in Table 2 indicate that if
the thickness of top and bottom flanges is reduced, the energy Table 2
absorption efficiency of a thin-walled square beam will not be Energy absorption indices for square tube specimens.
significantly varied while the crushing force efficiency will be
Test # FW E (J) Pm (kN) Pmax (kN) Mass (g) SEA (J/kg) CE
increased. On the other hand, if the thickness of web plates is
reduced, both indices will decrease. 1 0.4 70.6 1.86 2.07 102.6 688.3 0.90
In Table 2, the comparison between the specimens with dif- 2 0.4 71.7 1.89 2.09 102.6 698.9 0.90
ferent structural mass is to some extent not fair. This is due to the 3 2.5 22.5 0.59 1.26 102.6 218.8 0.47
4 0.7 84.6 2.23 2.56 126.9 667.0 0.87
fact that the mass is linearly increased with the increase of wall
5 0.7 85.8 2.26 2.58 126.9 676.4 0.88
thickness while the variation of force, moment or SEA is generally 6 1.43 60.6 1.59 2.25 126.9 477.5 0.71
nonlinear with the increase of wall thickness. For example, the 7 1.0 103.3 2.72 3.35 151.2 683.3 0.81
fully plastic bending moment is proportional to the square of wall 8 1.0 106.7 2.81 3.33 151.2 705.9 0.84
thickness. Therefore, if the mass of a beam is given during design 9 1.0 104.8 2.76 3.29 151.2 692.9 0.84

process, the results may be different. To find out the most efficient
112 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116

thickness distribution scheme of square tubes with constant mass,
structural optimization designs are necessitated for the beams
under transverse loading and relevant problems will be solved in
later section.
5. Comparison of static and dynamic loading
In accidental impact events, the transverse loading is always
dynamic and therefore, it is necessary to understand the difference
of deformation and force responses between quasi-static and
dynamic loading. The dynamic loading of specimens with different
FW values in the test is analyzed in this section and three different
load speeds: V¼5, 10 and 15 m/s are employed.
In dynamic loading, the deformation mode of the beams is
almost the same with quasi-static loading and it is therefore not
provided here. The difference in force responses is large and the
force curves versus punch displacement for the specimens are
given in Fig. 9. The results for quasi-static loading are also pre-
sented in the figure for comparison. As the increase of loading rate,
the major difference is the appearance of an initial peak force. The
magnitude of the peak force increases gradually with the increase
of load speed. After the initial stage, the tendency of the force
response curve is almost the same for various load speeds. The
force level in this stage is just increased slightly with the increase
of loading rate. Apparently, this is due to the inertia effect of the
beams under dynamic loading. However, an exception occurs at
FW¼2.5.

Fig. 9. Force responses and deformation of square tubes under different load velocities. (a) Specimen for test #1, (b) Specimen for test #3, (c) Specimen for test #4,
(d) Specimen for test #6, (e) Specimen for test #7, and (f) Deformation of test #7 at V ¼ 10 and 15 m/s.
 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116
As shown in Fig. 9(e), the force level in the latter stage at
V¼15 m/s is lower than that for V¼10 m/s. This is caused by the
switch of deformation mode at higher load speed. The deformed
shapes at punch displacement of 50 mm are presented in Fig. 9(f) for
V¼10 and 15 m/s. When V r10 m/s, the web plates protrude out-
ward during bending while they deform inward to the section at
V¼15 m/s. Actually at V415 m/s, for example V¼ 20 or 25 m/s, the
web plates are also found to deform inward to the section. A slight
decrease in impact force is resulted by this switch of fold direction of
the web plates. The reason for this switch is not quite clear but
apparently deformation mode with smaller resistance is preferred
and it is triggered by thin web plates (large FW value) and high
load speed.
6. Structural optimization with thickness variables
To find out the most efficient scheme for the thickness dis-
tribution in the section of square tubes with constant mass,
structural optimization design problems are solved in this section.
To better simulate the real design of accidental impact events,
dynamic loading with a constant speed of 10 m/s is applied. Two
optimization problems are analyzed here. Firstly, square tubes
with uniform but different wall thickness in the four plates are
optimized. Secondly, graded thickness variables are introduced to
the webs of the square beams. The geometric and loading condi-
tions are the same as experiment in Section 2 and the finite ele-
ment model described in Section 3 is adopted since it has been
validated by experiments to give very good predictions. The lower
and upper bounds of the design variables are set to 0.8 and
2.0 mm since no fracture of material was found in the experiment
of square tubes with the thickness in this range.
Structural crashworthiness optimization problems are gen-
erally solved by surrogate methods due to its highly nonlinear
characteristics. The response surface method (RSM) [33] is one of
the most effective and efficient surrogate methods in solving
relevant problems. Accompanying with design of experiment
(DOE) technology, RSM was widely employed in various engi-
neering fields including crashworthiness [22,34–35], tube hydro-
forming [36], laser cutting [37], welding processes [38–39] and etc.
The response surface method is employed here to solve the pre-
sent optimization problem and the basic process of RSM optimi-
zation is as follows: the form of the basis functions is selected first
and design of experiment techniques are then employed to
determine sampling points in the design domain. Numerical ana-
lyses are carried out in the selected points and the basis functions
are then fitted to the responses at these points by the least-squares
Fig. 10. Illustrations for optimization problems with thickness variables: (a) Problem 1, and (b) Problem 2.
113
method. Finally, the approximate surrogate optimization problem
is constructed and solved.
6.1. Formulation of optimization problem 1
An illustration for optimization problem 1 is shown in Fig. 10(a).
The beam section is eudipleural along the central axis. The thick-
ness of each plate is uniform and a total of three thickness variables
t1, t2 and t3 are employed. Maximizing the SEA value during the
transverse impact loading is the objective while as a constraint, the
mass m of the structure is kept constant during design. The SEA
value and mean crushing force are determined at a stroke distance
of δ ¼50 mm. The constant mass is set to the value when each plate
gets the average thickness of 1.4 mm. Therefore, the mass constraint
can be expressed as t1 þ t2 þ2t3 ¼ 5.6 mm.
The optimization problem is formulated as:
Pm δ
Maximize SEA ¼
m
Subjected to : 0:8 mm r t 1 ; t 2 ; t 3 r 2:0 mm
t 1 þ t 2 þ 2t 3 ¼ 5:6 mm ð1Þ
To further reduce the design variables, an alternative problem
with two design variables can be solved.
Pm δ
Maximize SEA ¼
m
Subjected to : 0:8 mm r t 1 ; t 2 r 2:0 mm
8 mm r t 3 ¼ ð5:6  t 1  t 2 Þ=2 r 2:0 mm ð2Þ
A full set of quadratic polynomial functions is chosen as basis
functions here. The basis functions are therefore 1, t1, t2, t12, t1t2,
t22. Central composite experimental design is utilized for RSM
sampling since it is powerful when combined with full quadratic
polynomial model. A total of 9 experimental points are required
for central composite design with two design variables. The sam-
pling points are listed in Table 3 and numerical analyses are per-
formed on these points. The relevant energy absorption indices
and force responses are also presented in Table 3. By least-squares
data fitting, the approximate response surface function of SEA is
expressed as
SEA ¼ f ðt 1 ; t 2 Þ ¼ 1768:544  764:354t 1  195:432t 2 þ 162:986t 1 2
þ 15:560t 1 t 2  36:913t 2 2 ð3Þ
Substitute Eq. (3) into Eq. (2) and solve the optimization pro-
blem, the optimum point is (0.8, 0.8). That is, t1, t2, t3 are 0.8,
0.8 and 2 mm, respectively. Substitute t1 and t2 into the approx-
imate function Eq. (3), the predicted optimal SEA is 1091.4 J/kg. A
114 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116

Table 3 SEA value, the CE index of it is not maximized. In experiment, the

Design points and relevant numerical results in cycle 1 of problem 1. CE index of the specimen with FW ¼0.4 gets the maximum value.
Apparently, this should be attributed to the high initial peak force
Point t1 (mm) t2 (mm) Mass (g) Pm (kN) Pmax (kN) SEA (J/kg) CE
in dynamic loading. Nevertheless, the CE index 0.62 of the optimal
1 1.400 1.400 103.6 1.46 2.74 702.4 0.53 design is still very good. It is only slightly lower than the max-
2 1.824 1.824 103.1 1.02 2.14 492.5 0.47 imum value 0.64. It should be mentioned that the objective of the
3 0.976 1.824 103.6 1.50 2.72 726.4 0.55
present work is to maximize the SEA value, other optimization
4 1.824 0.976 103.6 1.49 2.80 720.5 0.53
5 0.976 0.976 103.1 1.99 3.19 965.6 0.62
problems, such as multi-objective optimization or optimization
6 2.000 1.400 103.3 1.21 2.45 586.0 0.49 with other constraints, can also be done. However, no effort is
7 1.400 2.000 103.3 1.08 2.23 520.9 0.48 made here to address these problems.
8 0.800 1.400 103.3 1.93 3.16 934.1 0.61
9 1.400 0.800 103.3 1.77 3.04 855.3 0.58
Optimal 0.800 0.800 102.6 2.06 3.30 1003.0 0.62
6.2. Formulation of optimization problem 2

As expected, the optimization results of problem 1 show that

Table 4 putting more material in the webs and less in top and bottom
Design points and relevant numerical results in cycle 2 of problem 1. flanges will increase the energy absorption efficiency of thin-
Point t1 (mm) t2 (mm) Mass (g) Pm (kN) Pmax (kN) SEA (J/kg) CE
walled square beams. This is due to the fact that the web plates
dissipate more energy than flanges. However, the deformation in
1 1.100 1.100 103.3 1.84 3.09 890.2 0.59 the web plates is also not uniform. According to the theoretical
2 1.312 1.312 103.6 1.56 2.86 755.3 0.55 analyses by researchers [2,14,16], the upper region of the webs
3 0.888 1.312 103.3 1.91 3.15 924.9 0.61
undergoes more severe deformation than the lower region. A
4 1.312 0.888 103.3 1.78 3.05 863.4 0.58
5 0.888 0.888 102.9 2.06 3.21 999.9 0.64 direct observation on deformation patterns in Fig. 6 can also
6 1.400 1.100 103.5 1.62 2.91 784.3 0.56 indicate this feature. Therefore, an optimization problem with
7 1.100 1.400 103.5 1.68 2.96 813.6 0.57 graded thickness in the web plates is analyzed in this subsection.
8 0.800 1.100 103.0 2.06 3.23 999.5 0.64
An illustration of the optimization problem is shown in Fig. 10(b).
9 1.100 0.800 103.0 1.95 3.17 944.5 0.61
Optimal 0.800 0.800 102.6 2.06 3.30 1003.0 0.62 The beam section is also eudipleural along the central axis. Three
thickness variables t1, t2 and t3 are applied to the web plates, while
the top and bottom flange thickness is set to the lower bound
verification analysis at the optimum point is performed and the 0.8 mm considering the optimal design in last problem. The objective
calculated SEA value of the optimal design is 1003.0 J/kg. The SEA is still to maximize SEA value and the relevant results are also cal-
predicted by response surface function is about 9% higher than the culated at the stroke distance of δ ¼50 mm. The mass constraint of
verification value for the optimal design. In order to reduce this the structure is set to the value when the web plates get the uniform
error, sequential RSM method [40] is employed here. The purpose thickness of 1.4 mm. Consequently, the mass constraint is expressed
of the sequential RSM is to allow convergence of the solution to a as t1 þ t2 þt3 ¼4.2 mm. Different from problem 1, the thickness t1, t2
prescribed tolerance by adjusting the region of interest in sub- and t3 are independent during sampling and the constant mass is
sequent approximations. By setting the upper bound of t1, t2 to applied as constraint during the optimization solution of RSM
1.4 mm, the design region is reduced and the optimization pro- surrogate model.
blem is The optimization problem is then formulated as:
Pm δ Pm δ
Maximize SEA ¼ Maximize SEA ¼
m m

Subjected to : 0:8 mm r t 1 ; t 2 r 1:4 mm Subjected to : 0:8 mm r t 1 ; t 2 ; t 3 r 2:0 mm

t 3 ¼ ð5:6 t 1  t 2 Þ=2
The design points and relevant results are listed in Table 4. By
fitting the data, the approximation function of SEA is now
SEA ¼ f ðt 1 ; t 2 Þ ¼ 1180:215  204:105t 1 þ 253:342t 2 þ 21:275t 21
 183:923t 1 t 2  121:817t 22
Substitute Eq. (5) into Eq. (4) and solve the problem, the
optimum point is still (0.8, 0.8). The predicted value of SEA by
Eq. (5) at the optimum point is 1037.6 J/kg. The difference
between predicted value and the verification value is now
smaller than 3.5%.
The optimization results show that the maximum SEA is
achieved when the thickness of top and bottom flanges is set to
the lower bound and at the same time the thickness of the webs is
equal to the upper bound. A comparison of the responses between
the uniform design (1.4, 1.4) and the optimal design (0.8, 0.8) is
plotted in Fig. 11. The SEA value of the optimal design is more than
40% higher than that of the uniform design. The deformation
patterns are presented in Fig. 11(b). Similar as quasi-static cases,
the optimal design (FW ¼0.4) develops multiple-fold mode while
single inward fold is formed for uniform design (FW ¼1.0).
The CE index of the design points is also listed in Tables 3 and 4.
It is noted that although the optimal design shows the maximum
ð4Þ t 1 þ t 2 þ t 3 ¼ 4:2 mm ð6Þ
The basis functions are set to 1, t1, t2, t3, t1 , 2
t1t2, t1t3, t22, t2t3, t32
and D-optimal design is employed for RSM sampling. A total of 16
experimental points are required for D-optimal design with three
design variables. The design points and the results are presented in
ð5Þ
Table 5. The approximation function of SEA is fitted as
SEA ¼ f ðt 1 ; t 2 ; t 3 Þ ¼  388:144 þ159:210t 1  104:145t 21 þ856:089t 2
þ 105:238t 1 t 2  342:975t 22 þ331:927t 3  12:887t 1 t 3
þ 140:106t 2 t 3  109:851t 23 ð7Þ
Substitute Eq. (7) into Eq. (6) and solve the problem, the
optimum point is at (0.8, 1.553, 1.847) and the predicted max-
imum SEA value at the point is 926.8 J/kg. After verification
analysis, the calculated SEA value at the point is 918.4 J/kg. The
difference between predicted value and verification value is
smaller than 1.0%. It should be mentioned that the design points
listed in Table 5 generally have different structural masses and
therefore a direct comparison between different points is
meaningless.
The optimization indicates that thin-walled square beam with
graded thickness in the webs can achieve higher energy
absorption efficiency. A comparison of the responses between
 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116 115

Fig. 11. Comparison between uniform and optimal design (a) force responses, (b) Deformation patterns.

Table 5
Design points and relevant numerical results for problem 2.

Point t1 (mm) t2 (mm) t3 (mm) Mass (g) Pm (kN) Pmax (kN) SEA (J/kg) CE

1 1.40 1.40 1.40 80.2 1.34 2.38 833.9 0.56

2 2.00 0.80 0.80 69.1 0.52 1.31 373.7 0.39
3 1.16 0.80 0.80 61.2 0.66 1.28 536.2 0.51
4 2.00 2.00 0.80 91.4 1.23 2.05 673.0 0.60
5 1.16 2.00 0.80 83.6 1.13 1.97 676.3 0.57
6 0.80 0.80 1.52 64.6 0.73 2.02 564.4 0.36
7 0.80 2.00 1.52 87.0 1.51 2.65 867.4 0.57
8 2.00 2.00 1.88 101.5 2.01 3.19 989.3 0.63
9 0.80 1.40 0.80 69.1 0.80 1.61 580.5 0.50
10 2.00 0.80 2.00 80.2 0.88 2.57 547.4 0.34
11 1.52 2.00 2.00 98.1 1.93 3.24 985.4 0.60
12 0.80 2.00 2.00 91.4 1.71 3.15 935.5 0.54
13 0.80 0.80 0.80 57.9 0.61 1.26 527.2 0.48
14 0.80 1.28 2.00 78.0 1.47 2.76 941.1 0.53
15 0.80 0.80 2.00 69.1 0.85 2.37 612.1 0.36
16 2.00 1.28 1.40 83.6 1.39 2.44 829.5 0.57
Optimal 0.80 1.553 1.847 81.7 1.50 2.77 918.4 0.54

Fig. 12. Comparison between uniform and optimal design (a) force responses, and (b) deformation patterns.

the uniform design (1.4, 1.4, 1.4) and the optimal design (0.8,
1.553, 1.847) is plotted in Fig. 12. The SEA value of the optimal
design is about 10% higher than that of the uniform design. The
deformation patterns are presented in Fig. 12(b). Both designs
develop multiple-fold mode, whereas there are some differences
in the extent of deformation. More material in the upper region
of optimal design results in smaller inward collapse of the top
flange.
7. Conclusion
Three-point bending collapse of square tubes with variable
thickness in the cross-section is investigated in the present work.
The primary works and conclusions are summarized as follows:
1) Three-point bending tests are performed quasi-statically for
square tubes with uniform but different wall thickness in the
116 X. Zhang et al. / International Journal of Mechanical Sciences 106 (2016) 107–116
four side plates. The experimental results show that relative
strength of the flanges versus the web plates has significant
influence on force responses and deformation pattern. When
web plates are relatively stronger, the force response curves are
more flat (high crush force efficiency) and a multiple-fold mode
that can dissipate energy more efficiently is developed.
2) Numerical simulation of the three-point bending test is con-
ducted and validated by experiments. The numerical model
discretized by eight-node brick elements can well reflect the
behavior of the thin-walled specimens. Both the force responses
and deformation modes compare fairly well with the experi-
mental results.
3) Two structural optimization design problems are formulated
and solved by the response surface method. The optimization
results validate that the beams with thin flanges and thick webs
can achieve the highest SEA value. The SEA of a thin-walled
square beam is increased by more than 40% by reducing the
thickness of flanges and accordingly increasing that of webs.
Strong web plates and relatively weak flanges are suggested in
the design of beam structures against transverse loading.
4) The optimization also indicates that adopting graded thickness
in the web plates could further improve the SEA of thin-walled
square beams. The introduction of thickness variables in the
cross-section of beams is validated to be an effective way to
improve the crashworthiness performance of structures under
transverse loading.
5) The force responses of beams under dynamic loading are quite
different from those under quasi-static loading. The dynamic
force responses showed a very high initial peak and a subse-
quent relatively flat plateau region. When FW ratio is small, the
crush force efficiency of the beam is very high under quasi-
static loading, while this advantage is impaired in dynamic
loading due to the high initial peak force.
Acknowledgments
The present work was supported by National Natural Science
Foundation of China (Nos. 11002060, 11372115, 11502117), Natural
Science Foundation of Hubei Province (2013CFA130) and the
Fundamental Research Funds for the Central Universities, HUST
(No. 2014TS132).
References
[1] Lu G, Yu T. Energy absorption of structures and materials, Boca Raton.
Cambridge: CRC Press; Woodhead; 2003.
[2] Wierzbicki T, Abramowicz W. On the crushing mechanics of thin-walled
structures. J Appl Mech-T ASME 1983;50(4a):727–34.
[3] Abramowicz W, Jones N. Dynamic progressive buckling of circular and square
tubes. Int J Impact Eng 1986;4(4):243–70.
[4] Zhang X, Zhang H. Axial crushing of circular multi-cell columns. Int J Impact
Eng 2014;65(0):110–25.
[5] Zhang X, Zhang H. Energy absorption of multi-cell stub columns under axial
compression. Thin Wall Struct 2013;68(0):156–63.
[6] Morris E, Olabi A, Hashmi M. Lateral crushing of circular and non-circular tube
systems under quasi-static conditions. J Mater Process Technol 2007;191
(1):132–5.
[7] Olabi A, Morris E, Hashmi M, Gilchrist M. Optimised design of nested circular
tube energy absorbers under lateral impact loading. Int J Mech Sci 2008;50
(1):104–16.
[8] Baroutaji A, Morris E, Olabi AG. Quasi-static response and multi-objective
crashworthiness optimization of oblong tube under lateral loading. Thin Wall
Struct 2014;82:262–77.
[9] Baroutaji A, Gilchrist MD, Smyth D, Olabi AG. Crush analysis and multi-
objective optimization design for circular tube under quasi-static lateral
loading. Thin Wall Struct 2015;86:121–31.
[10] Chen W, Wierzbicki T, Santosa S. Bending collapse of thin-walled beams with
ultralight filler: numerical simulation and weight optimization. Acta Mech
2002;153(3–4):183–206.
[11] Yu JL, Wang X, Wei ZG, Wang EH. Deformation and failure mechanism of
dynamically loaded sandwich beams with aluminum-foam core. Int J Impact
Eng 2003;28(3):331–47.
[12] Yu J, Wang E, Li J, Zheng Z. Static and low-velocity impact behavior of sand-
wich beams with closed-cell aluminum-foam core in three-point bending. Int
J Impact Eng 2008;35:885–94.
[13] Santosa S, Banhart J, Wierzbicki T. Experimental and numerical analyses of
bending of foam-filled sections. Acta Mech 2001;148(1–4):199–213.
[14] Kecman D. Bending collapse of rectangular and square section tubes. Int
J Mech Sci 1983;25(9–10):623–36.
[15] Wierzbicki T, Recke L, Abramowicz W, Gholami T, Huang J. Stress profiles in
thin-walled prismatic columns subjected to crush loading-II. Bending. Comput
Struct 1994;51(6):625–41.
[16] Kim TH, Reid SR. Bending collapse of thin-walled rectangular section columns.
Comput Struct 2001;79(20–21):1897–911.
[17] Hosseinipour SJ, Daneshi GH. Energy absorbtion and mean crushing load of
thin-walled grooved tubes under axial compression. Thin Wall Struct 2003;41
(1):31–46.
[18] Zhang X, Huh H. Energy absorption of longitudinally grooved square tubes
under axial compression. Thin Wall Struct 2009;47(12):1469–77.
[19] Zhang Z, Liu S, Tang Z. Design optimization of cross-sectional configuration of
rib-reinforced thin-walled beam. Thin Wall Struct 2009;47(8–9):868–78.
[20] Meyer A, Wietbrock B, Hirt G. Increasing of the drawing depth using tailor
rolled blanks—numerical and experimental analysis. Int J Mach Tool Manuf
2008;48(5):522–31.
[21] Peng X. Steel-pipe conical-tube machine, China, No. CN20032119113U
20031212, 2004.
[22] Zhang X, Wen Z, Zhang H. Axial crushing and optimal design of square tubes
with graded thickness. Thin Wall Struct 2014;84:263–74.
[23] Sun G, Xu F, Li G, Li Q. Crashing analysis and multiobjective optimization for
thin-walled structures with functionally graded thickness. Int J Impact Eng
2014;64(0):62–74.
[24] Zhang X, Zhang H, Wen Z. Axial crushing of tapered circular tubes with graded
thickness. Int J Mech Sci 2015;92(0):12–23.
[25] Xu F, Tian X, Li G. Experimental Study on Crashworthiness of Functionally
Graded Thickness Thin-Walled Tubular Structures. Exp Mech 2015;55
(7):1339–52.
[26] H. Yin, H. Fang, G. Wen, Q. Wang and Y. Xiao, An adaptive RBF-based multi-
objective optimization method for crashworthiness design of functionally
graded multi-cell tube, Struct Multidiscp Optim. 2015, http://dx.doi.org/
10.1007/s00158-015-1313-1.
[27] Zhang X, Zhang H. Relative merits of conical tubes with graded thickness
subjected to oblique impact loads. Int J Mech Sci 2015;98:111–25.
[28] Chirwa EC. Theoretical analysis of tapered thin-walled metal inverbucktube.
Int J Mech Sci 1993;35(3–4):325–51.
[29] Gupta PK. A study on mode of collapse of varying wall thickness metallic
frusta subjected to axial compression. Thin Wall Struct 2008;46(5):561–71.
[30] ASTM. Standard E8M-04: Standard Test Methods for Tension Testing of
Metallic Materials, Testing of Metallic Materials, ASTM; 2004.
[31] Hallquist J. LS-DYNA user's manual. LS-DYNA 970 ed.. Livermore, California:
Livermore Software Technology Corporation; 2003.
[32] Santosa SP, Wierzbicki T, Hanssen AG, Langseth M. Experimental and
numerical studies of foam-filled sections. Int J Impact Eng 2000;24(5):509–34.
[33] Myers RH, Montgomery DC. Response surface methodology: process and
product optimization using designed experiments. 1st ed.. New York: John
Wiley & Sons; 1995.
[34] Zhang X, Cheng G, Wang B, Zhang H. Optimum design for energy absorption of
bitubal hexagonal columns with honeycomb core. Int J Crashworthines
2008;13(1):99–107.
[35] Zhang X, Zhang H. Optimal design of functionally graded foam material under
impact loading. Int J Mech Sci 2013;68:199–211.
[36] Alaswad A, Benyounis K, Olabi A. Employment of finite element analysis and
response surface methodology to investigate the geometrical factors in T-type
bi-layered tube hydroforming. Adv Eng Softw 2011;42(11):917–26.
[37] Eltawahni H, Olabi A-G, Benyounis K. Effect of process parameters and opti-
mization of CO2 laser cutting of ultra high-performance polyethylene. Mater
Des 2010;31(8):4029–38.
[38] Anawa E, Olabi A-G. Optimization of tensile strength of ferritic/austenitic
laser-welded components. Opt Laser Eng 2008;46(8):571–7.
[39] Benyounis K, Olabi A. Optimization of different welding processes using sta-
tistical and numerical approaches–a reference guide. Adv Eng Softw 2008;39
(6):483–96.
[40] Kurtaran H, Eskandarian A, Marzougui D, Bedewi NE. Crashworthiness design
optimization using successive response surface approximations. Comput
Mech 2002;29(4–5):409–21.