Safety Science 117 (2019) 263–275

Contents lists available at ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/safety

Design and injury analysis of the seated occupant protection posture in train T
collision
⁎
Weilin Yanga,b, Suchao Xiea, , Haihong Lia, Zengtao Chenb
a
Key Laboratory of Traffic Safety on Track, Ministry of Education, Central South University, Changsha, Hunan 410075, PR China
b
Department of Mechanical Engineering, University of Albera, Edmonton, AB T6G 2G9, Canada

A R T I C LE I N FO A B S T R A C T

Keywords: In order to minimize injuries and protect safety of seated occupants in train collision, this study proposed a self-
Frontal collision protective posture with hands laced behind head and body curled up for occupants. Through the local sled
Dynamic response simulation test, the effectiveness of hands laced behind head posture was verified. In order to obtain the optimal
Injury protective posture with body curled up, First, the waist angle A, leg angle B and backward rotation angle C
Occupant’s protective posture
around the hip-point H of a dummy were selected as factors to design orthogonal test with different factor levels.
Then, by using the direct analysis method, the influence laws of each factor on dynamic response and injury of
occupants were analyzed. Furthermore, synthesized each kind of critical injury value of human body, the radial
basis function surrogate model was constructed. Combining with NSGA-II multi-objective genetic algorithm, the
parameters of curled-up body posture of occupants were optimized. The balance optimal protective posture of
occupants was at 40-0-08 (A-B-C). Finally, the optimal scheme was verified in the simulation system for standard
collision of train at the speed of 48 km/h. The results show that after optimization, head injury of occupants
reduces by 88% and axial force and bending moment of neck separately decrease by nearly 50% and 80%.
Moreover, stresses on thigh bone and tibia bone decline by 53% and 56%, respectively.

1. Introduction plastic platform forces of main energy-absorbing structures in the ends

of a vehicle and combined with genetic algorithm to realize maximum
It is an important issue to protect the safety of seated occupants in ordered deformation length of energy-absorbing components. Xie et al.
the train collision. At present, passive protection is one of the important (2017a, 2017b, 2017c, 2018), Liu et al. (2018), Peng et al. (2015, 2017)
ways to minimize impact injuries that occupants may suffer from and Gao et al. (2017) proposed different energy-absorbing structures to
(Zhang et al., 2018; Chen et al., 2018; Luo and Fan, 2018; Praveen improve the crashworthiness performances.
et al., 2018). By considering that collision force hurls occupants against interior
The structural crashworthiness of railway vehicles is one of research parts of the compartment, scholars conducted a lot of studies on interior
focuses (Lu, 2002; Xue et al., 2005; Hosseini and Bayat, 2011; Hosseini structure of the trains (Kirkpatrick et al., 2001; Simons and Kirkpatrick
and Nankali, 2010; Gao and Tian, 2007; Xie et al., 2017a, 2017b). In 1999; Tyrell, 2002; Parent et al., 2004; Severson et al., 2006, 2004).
the early 20th century, European Union successively carried out large- Rail Safety and Standards Board (RSSB) carried out a series of re-
scale collision projects of real vehicles, such as TrainCol (Wei et al., searches on secondary collision of structure of passenger compartment
2012), SafeTrain (European Union Contract, 2001) and SafeTram and occupants in trains and issued a series of standards in view of in-
(Hecht, 2004) and proposed the basic requirements of energy adsorp- terior structure, such as train materials, safety belt and seat structure
tion and crushing distance for energy-absorbing components in the (AV/ST9001, 2002; GM/RT2100, 2010). Tyrell et al. (1995) and Anton
front end and middle of vehicles (EN15227:2008, 2009). Jacobsen et al. et al. (2017) focused on the influences of seat stiffness and occupant
(2004) designed the crushable components by using crash energy restraint pulse on fatalities resulting from secondary collision. Seat
management (CEM) method. The method was usefully to crush the end stiffness variation studies had showed positive effects of plastic de-
of the car in a controlled manner and limit the structural damage from formation in the seat structure which can decrease concentrated forces
intruding into the passenger compartment. Based on characteristic on the occupant’s abdomen. Besides, lap and shoulder belts can help the
curve of energy-absorbing structures, Zhang et al. (2016) changed the seated occupants survive in all the collision scenarios. Wei and Zhang

⁎
Corresponding author.
E-mail address: [email protected] (S. Xie).

https://doi.org/10.1016/j.ssci.2019.04.028
Received 31 October 2018; Received in revised form 18 March 2019; Accepted 19 April 2019
Available online 24 April 2019
0925-7535/ © 2019 Elsevier Ltd. All rights reserved.
W. Yang, et al. Safety Science 117 (2019) 263–275

(2017) added the buffer springs to seats, which is an effective com-

partmentalization strategy to limit the occupant’s range of motion. Cao
et al. (2014) investigated the impact of backrest angle change of seats
on passenger’s safe. The results indicate that with the increase of
backrest angle, the injury becomes more severe. Xie and Tian (2013)
studied the influences of table height, distance between seats and table
to seat distance on occupant impact injury. By placing the seats and
table reasonably close together, the occupants will have fewer rates of
fatality, and making table height judicious lower can effectively reduce
injuries. Furthermore, Kazuma et al. (2012) installed handrails and
partitions along bench seats to separate passengers can mitigate the
injuries. And, analyses clearly showed that change of the shape of
bench-end partitions and reduction of the length of overhead baggage
racks in relation to the bench seats can improve passengers’ safety. Koji
et al. (2008) found chest injury with panel dividers was lower than that
with tubular dividers. Fig. 1. Posture with human body curled up.
Apparently, countermeasures now being taken are broadly in opti-
mization of crashworthiness of main energy-absorbing components in
body show the bending action, as the curled-up posture demonstrated
the front end of trains and researches on interior structure in the train.
in Fig. 1.
Prochowski and Zuchowski (2014) proved that relatively small changes
At this time, contact area of occupants exposed to the collision en-
of an occupant’s position have a serious influence on a risk of injuries
vironment greatly decreases. All important organs, like thorax and in-
during a road accident. Railway vehicles are different from single car
ternal organs of human body curl up inside, while parts bearing strong
structure, because the whole train comprises multiple carriages con-
impact force, such as arms, legs and feet are mainly exposed. In this
nected through coupler and draft gears and geometric dimension of a
way, the possibility of injuring human body caused by secondary col-
single carriage is much larger than that of a car. Therefore, unlike car
lision reduces.
crashes that happen in so short a time that passengers therein have no
Therefore, the enlightenment is to design a curled-up protective
time to response, occupants in running trains can timely find the im-
posture with hands laced behind head to guide occupants to effectively
pending dangers and make response. Therefore, how to make self-
protect themselves when accidents occur, thus reducing collision injury.
protective postures and improve survival probability when occupants in
When occupants in a high-speed bullet train protect themselves, there
running trains are warned of the accidents has important research
are three angles between a sitting dummy and a seat, as displayed in
significance. Wei et al. (2015) found that the angle between the di-
Fig. 2. The angles include bending angle A of waist, bending angle B of
rection faced by the standing passengers and running direction of trains
legs and back leaning angle C around the hip-point H of the dummy.
significantly affects the degree of casualty of occupants. When occu-
The three angles, i.e. A, B and C are used as design variables and are in
pants have an angle larger than 90° with the running direction of a train
the design ranges of 0° ⩽ A ⩽ 40°, 0° ⩽ B ⩽ 40° and 0° ⩽ C ⩽ 20°.
and are against the running direction, they are more likely to be in-
jured. Through accident reconstruction, Peng et al. (2012) found that
frequency and degree of injuries of people in the posture of bending and 2.2. Finite element model of seat
riding are obviously lower than those of standing occupants under the
same collision occasion. Koji et al. (2002) suggested that standing By taking a first-class seat in a certain high-speed bullet train as the
passengers could drastically minimize the risk of head and thorax in- prototype, a finite element model was established. The seat mainly
juries by squatting on the floor at the time of collision. Based on these included a headrest, a backrest, a cushion, a chassis, two armrests, a
researches, this study proposed a protective posture with hands laced backrest adjustment unit and connectors. The seat frame was divided
behind head and body curled up for occupants. Through sled simulation into quadrilateral shell elements, and backrest, headrest and cushion
test, the influence laws of different protective postures on head, neck, were mainly composed of foam materials. In order to better simulate
thorax and legs of occupants were analyzed. Furthermore, the multi- the deformation of foam in collision, the hexahedral elements were
objective optimization design was conducted on parameters for curled- utilized for building the model. Based on the specifications on metal
up body posture of occupants. Finally, a coupling collision simulation materials for producing seats of Japanese industrial standard, cushion,
system of train body, seat and occupants was built to verify the effec- side supports of backrest and junction plates were made mainly
tiveness of the optimized scheme. adopting the SPHC hot rolled steel (HRS), and the SAPH HRS was used
for fabricating side supports of backrest, supports of cushion and
2. Numerical simulation scenario junction plates for adjusting. The SPCC cold rolled steel (CRS) was
utilized for producing cushion plate and backrest plate. Adjusting tubes
2.1. Design of protective posture in seats were manufactured with the STKM carbon steel tubes. Each
parameter of materials is displayed in Table 1.
Biological researches show that animals and plants in the ecology The model was built by using *MAT_24 PIECEWISE_LINEAR_PLAS-
can automatically shrink and curl up when they are stimulated by the TICITY. The elastic modulus was 210 GPa and Poisson’s ratio was 0.3.
external environment or meet immediate dangers. The higher the class Foam materials for fabricating the seat were low-density polyurethane
of animals is, the more obvious the self-protective behaviors. and *MAT_57 LOW_DENSITY_FOAM was used for modeling. Fig. 3
As the highest class in the ecological environment, human body can shows the characteristic curve of foam materials for the seat (Hou et al.,
automatically bend limbs and shrink joints when being stimulated by 2011).
external collision, which is known as buckling response in biology. The In order to prevent distortion of foam materials, damping coefficient
curled-up body posture of human depends on the change of position of of the materials was set to be DAMP = 5. In the meanwhile, a layer of
human head in stimulation. When head leans back, limbs and extensor *SET_SEGMENT materials was used to wrap the cushion and backrest
of back suddenly contract and human body shows a posture in which structures and defined as *MAT_9 NULL. The internal nodes of foam
the limbs and back are stretched straight. When head moves forward, materials were connected to the *SET_SEGMENT through
limbs and extensor of back suddenly relax and back and limbs of human *CONTACT_TIED_NODE_TO_SURFACE (Lin et al., 2005).

264
W. Yang, et al. Safety Science 117 (2019) 263–275

Fig. 2. Variables of curling angles.

Table 1
Performance parameters of metal materials of a seat.
Material SPHC SAPH SPCC STKM

Yield limit (MPa) 205 305 215 305

Ultimate strength (MPa) 270 440 270 440
Elongation (%) 29 29 39 13

0.06

0.05

0.04
Stress/MPa

0.03
Fig. 4. Finite element model for seats.
0.02

0.01

0.00
0.0 0.2 0.4 0.6 0.8 1.0
Strain/mm
Fig. 3. Stress-strain curve of the foam materials.

*CONTACT_INTERIOR was selected to define the contact of foam ele-

ments inside which were defined to be in type-2 sticky hourglass con-
trol mode.
At present, the first-class seats in trains of China show the spacing of
1080–1160 mm, so the seats are arranged in the spacing of 1160 mm
here. Therefore, the finite element model for the first-class seats was
built, as shown in Fig. 4.

2.3. Anthropomorphic test set-up

Fig. 5. Anthropomorphic test set-up with hybrid III 50th.

The validated (Sarba et al., 2011) Hybrid III 50th semi-rigid dummy
which was developed by Livermore Software Technology Corporation of the front seat and ground in surface-to-surface collision mode. In
(LSTC) was used. National Aeronautics and Space Administration (Al addition, the dummy itself was defined as one side self-contact and
Tabiei et al., 2009) have validated it has good prediction to the results. friction coefficient was set to be 0.2.
Import semi-rigid dummy into LS-PrePost processing software, through Triangular acceleration pulse shown in Fig. 6 was applied in the test
Dummy Position interface adjusted the posture of the dummy, so that it so as to measure injury degrees of each part of the dummy. The pulse
sat naturally on the seat model (Fig. 5). Finally, import it into the Altair was applied to dummy model installed on local seat structure by Volpe
HyperWorks graphical Pre-processor, coupling with seats finite element Center in the United States (Deb et al., 2002). Triangular acceleration
model. The contact relationships of the dummy with the seat and pulse reached the peak, being 8 g lasting for 1000 ms in total after
ground were established. Surface-to-surface contact was applied be- acting for 125 ms. In the meanwhile, acceleration of gravity was per-
tween the dummy and external Null elements of cushion and backrest. formed on the dummy and the seat to simulate the scenario of occu-
It was defined that the dummy contacted with the armrests and chassis pants moving forward in collision, the computations were carried out

265
W. Yang, et al. Safety Science 117 (2019) 263–275

Fz My
Nij = +
8 Fzc Myc (2)

where Fzc and Myc are constants relating to axial force and bending
moment of neck, respectively. FMVSS 208 and GM/RT2100 Issue Five
acceleration/g

6
present that when axial force is tensile force, Fzc = 6806 N , while
Fzc = 6160 N when axial force is compressive force. Myc corresponding
to bending and stretching of neck values 310 Nm and 135 Nm sepa-
4 rately. It is stipulated that the tolerance limit of biomechanical neck
injury predictor is 1 in collision accidents.
At present, contiguous_3ms injury criterion and the maximum
2 compression of thorax are the key indexes for evaluating injury degree
of thorax of unrestrained occupants in frontal collision. The tolerance
limit of thorax of occupants is that the maximum resultant linear ac-
celeration of thorax cannot be higher than 60 g at continuous 3 ms, that
0 is, C3ms ⩽ 60 g .
0 50 100 150 200 250
Leg injury is evaluated through tibia index (TI) and its calculation
Time/ms expression is shown as
Fig. 6. Finite element model for seat-dummy sled test. M (t ) F (t )
TI = +
Mc Fc (3)
using LS-DYNA.
Critical bending moment and critical compressive force are set to be
Mc = 240 Nm and Fc = 12 kN . Where, M (t ) represents the instantaneous
2.4. Railway injury criteria resultant bending moment of tibia and M (t ) = (Mx )2 + (My )2 , while
F (t ) indicates the instantaneous axial compressive force of tibia. GM/
In view of the survey report on the high-speed bullet train accident RT2100 standard regulates that the limit of TI is 1.3 in the whole col-
occurring in Wenzhou, Zhejiang province, China on July 23, 2011 (Wen lision.
et al., 2011), the easily injured body parts mainly include head, neck, In collision, each body part shows different injury degrees. In order
thorax and limbs in frontal collision accident. to comprehensively evaluate injury degree of human body, weighted
Head injury criterion (HIC) proposed by the United States is an injury criterion (WIC) is introduced to evaluate the comprehensive in-
acceptable and most commonly used parameter for evaluating head juries of head, thorax and legs of occupants. The calculation formula for
injury at present and its theoretical expression is WIC is
HIC15 ⎞ C D⎞ F + FR
2 t 2.5 WIC = 0.6 ⎛ + 0.35 ⎛ 3ms + /2 + 0.05 ⎛ L ⎞
⎡ ⎛ ∫t A (t ) dt ⎞ ⎤ ⎝ 700 ⎠ ⎝ 60 63 ⎠ ⎝ 20 ⎠ (4)
HIC = max ⎢ (t2 − t1 ) ⎜ 1 ⎥
TS⩽ t 1⩽ t2⩽ TE ⎢ (t2 − t1 ) ⎟ ⎥ where FL and FR denote the maximum axial forces of the left and right
⎣ ⎝ ⎠ ⎦ (1)
thigh bones, respectively and the unit is kN.
where TS and TE represent the starting moment and ending moment of
simulation, respectively. t1 and t2 indicate the starting and ending time 3. Results and discussion
of arbitrary integral in the whole calculation separately. A (t ) stands for
the resultant acceleration value (unit: g) of occupant’s head in X, Y and 3.1. Effectiveness of curled-up posture with hands laced behind head
Z directions in integral time in the whole collision. The maximum value
calculated in integral time is the injury value of head in the whole The whole dynamic response process of occupants sitting naturally
collision. According to the crashworthiness standard GM/RT2100 Issue with hands normally placing on the both sides of legs and in the pro-
Five of railway vehicles, it is regulated that HIC15 ⩽ 500 . tective posture with hands laced behind head in the sled test is shown in
Biomechanical neck injury predictor (Nij ) is an effective index for Fig. 7. As shown in Fig. 7(a), after collision, knees of the dummy firstly
evaluating injuries of occupants. contact with backrest and then the dummy continuously moves forward
due to inertial effect. The neck and head of the dummy directly collide

(a) Natural sitting posture

(b) Siitting posture with hands laced behind head

Fig. 7. Dynamic response of natural sitting posture and posture with hands laced behind head.

266
W. Yang, et al. Safety Science 117 (2019) 263–275

Table 2 body injuries were studied. The analysis results are shown in Table 3. In
Injury values of each part of the dummy in sled test. the table, Ki (i = 1, 2, 3, 4 and 5) is the sum of test results under each
HIC Nij CSI (g) D (mm) TI (left) TI (right) WIC factor and level. K̄i (i = 1, 2, 3, 4 and 5) indicates the average value of
test results under each factor and level. R represents the range of test
Natural sitting 624 1.08 103 5.06 1.29 0.64 0.88 results under the same factor and different levels.
posture
By comparing range R, angle change of legs of curled-up occupants
Hand laced behind 202 0.66 118 7.36 1.32 0.63 0.58
head
most significantly affects head, neck and thorax of occupants, while the
angle change of waist has the profound influences on legs. In the
meanwhile, the head and lower limbs of occupants are least sensitive to
with the backrest of the front seat until response ends. It can be ob- back leaning angle at H point and angle change of legs separately, and
viously seen from Fig. 7(b) that after knees of the dummy in the posture injuries to neck and thorax of occupants are least significantly influ-
of hands laced behind head contact with backrest, the arms of the enced by angle change of waist.
dummy are pressed against the backrest before the head collides, which In order to more intuitively obtain the influence laws of each factor
avoids direct violent collision between head and backrest. After that, and different levels on injuries of the dummy, the range analysis results
the elbows of the dummy constantly press the seat, thus protecting the are drawn through the diagram of orthogonal points to obtain the in-
head. fluences of each factor on injuries of head, neck, thorax and legs of
Based on dynamic response, in the sled test, occupants with hands dummies.
laced behind head protect their heads with arms. In order to better As demonstrated in Fig. 9, with the increase of bending angle A of
verify protection effects, the injuries on head, neck, thorax and legs of waist, injuries of head and legs of occupants decrease, while neck in-
the dummies in sitting posture with hands laced behind head and with juries constantly fluctuate. Moreover, injury degree of the thorax re-
original sitting posture were compared. The injury values of each duces after constant increase as the changing angle is less than 20°.
dummy are presented in Table 2. As shown in Fig. 10, with the constant increase of bending angle of
As demonstrated in the table, the head, neck, thorax and tibia of the tibia, injury values of head and neck constantly rise. However, change
dummy sitting naturally respond dramatically to the collision. HIC and trends of injuries to thorax and legs are contrary. Injury value of thorax
left TI are approximate to the injury limits of 700 and 1.3, while neck of the dummy firstly increases and then decreases, while injury to legs
injury predictor of the dummy and CSI of thorax are larger than the firstly constantly reduces and then rises after the changing angle is
thresholds of 1 and 60 g. Injuries of head and neck of the dummy in the larger than 30°.
posture with hands laced behind head are much slight in comparison As presented in Fig. 11, as the dummy leans back further, injuries to
with the above. Injury value is much lower than tolerance limit, while thorax and legs of occupants improve, while leg injury of occupants
injuries of thorax and legs are slightly larger than those of the dummy increases slightly after angle of H point increases to 15°. In addition,
in original posture, showing a small difference. The weighted injury of injury values of head and neck of occupants rise dramatically and then
the surviving dummy in the posture with hands laced behind head is far change gently after leaning back to 5°. The difference is that injury
smaller than that of the surviving dummy in original sitting posture. value of head of occupants suddenly reduces after rising the angle to
Therefore, posture with hands laced behind head is an effective self- 15°.
protective posture and can obviously reduce injuries of head and neck Based on the response laws of each injury value to variables of
of dummies. The original setting of the curled-up posture of dummies is curled-up posture of occupants, it can be seen that when a certain ob-
correct and effective. jective reaches the optimal, the other injury objective probably reaches
the worst state. This results in that occupants cannot reach the optimal
injury value on the whole. Therefore, by using Isight optimization
3.2. Influences of shape parameters on human-body injury software, the multi-objectives of occupants are optimized at the same
time, so as to obtain the compromised optimal posture.
The orthogonal experimental design (OED) with three factors st five
levels was applied which involved 25 sample points, and the spatial
4. Multi-objective optimization of the curled-up posture with
distribution of sample points obtained by using OED is shown in Fig. 8.
hands laced behind head
Through the direct analysis method the injuries of occupants in
different curled-up protective postures with hands laced behind head
The weighted injury values of head, thorax and legs of occupants
were analyzed and the influences of each design variable on human-
after comprehensive consideration were selected and neck injury pre-
dictor was used as design objective. Moreover, change angles of waist,
legs and H point of parameters of the curled-up protective posture oc-
cupants were adopted as design variables. Based on this, by using or-
thogonal test design method, the samples points were selected to con-
struct the model.

4.1. Approximate model of radial basis function and error analysis

For radial basis function model (RBF), Euclidean distance between

the points to be measured and sample points is used as independent
variable of the function. In the design space, the response function of
design variables can be approximately defined as
N
⎛ ⎞
Y (x ) = ∼
y (x ) + ε = ⎜∑ ai gi ( ∥x − x i ∥ci + aN + 1) ⎟ + ε
⎝ i=1 ⎠ (5)

where ∼ y (x ) represents the approximate function of the objective. ε , N,

ai and x − x i indicate the relative error (RE) between true value and
Fig. 8. Distribution of samples generated by Orthogonal experimental. approximate value, the number of terms of basis function gi , the

267
W. Yang, et al. Safety Science 117 (2019) 263–275

Table 3
Simulation result of injury values of each part of the dummy.
Index HIC Nij CSI (g) TI

A B C A B C A B C A B C

K1 2398 1091 2059 3.86 2.95 2.93 617 608 803 3.18 2.53 2.72
K2 3060 1695 2285 5.17 2.83 4.64 727 554 785 1.82 1.88 1.87
K3 2050 2329 2270 3.88 3.72 4.44 779 655 774 1.49 1.70 1.75
K4 2306 2980 2544 4.90 5.34 4.66 758 910 641 1.55 1.53 1.49
K5 1396 3196 2052 3.54 6.51 4.68 646 800 524 1.39 1.79 1.60
K̄ 1 480 218 412 0.77 0.59 0.59 123 122 161 0.64 0.51 0.54
K̄ 2 612 339 457 1.03 0.57 0.93 145 111 157 0.36 0.38 0.37
K̄ 3 410 466 454 0.78 0.74 0.89 156 131 155 0.30 0.34 0.35
K̄ 4 461 596 509 0.98 1.07 0.93 152 182 128 0.31 0.31 0.30
K̄ 5 279 639 410 0.71 1.30 0.94 129 160 105 0.28 0.36 0.32
R 333 421 99 0.32 0.73 0.35 33 71 56 0.36 0.20 0.24

650 1.05
600
1.00
550
0.95
500
Nij 0.90
HIC

450
0.85
400
0.80
350
0.75
300
0.70
250
0 10 20 30 40 0 10 20 30 40
A/angle A/angle
(a) Head injury (b) Neck injury
160
0.65
155
0.60
150
0.55
145 0.50
CSI/g

Ti

140 0.45

135 0.40

130 0.35

125 0.30

120 0.25
0 10 20 30 40 0 10 20 30 40
A/angle A/angle
(c) Thorax injury (d) Leg injury
Fig. 9. Influence laws of change angle A of waist on injuries to key parts of the dummy.

undetermined coefficient of basis function and Euclidean distance, re- value range of c generally changes from 0.2 to 3.0.
spectively. ci denotes the shape parameter. Different shape parameters Through approximation modulus of Isight optimization software,
result in different RBF model accuracies. When Isight optimization the response model of radial basis function of WIC and neck injury
software is used to build RBF approximate model, shape parameters are predictor Nij of the dummy with change angles of waist, legs and H
optimized inside. The values of shape parameter c constantly change, so point in the curled-up posture was built.
that the error of the established approximate model is minimum and the For the constructed agent function model, the response value and

268
W. Yang, et al. Safety Science 117 (2019) 263–275

1.50
715

650 1.35

585 1.20
520
1.05
HIC

Nij
455
0.90
390

325 0.75

260 0.60
195
0.45
0 10 20 30 40 0 10 20 30 40
B/angle B/angle
(a) Head injury (b) Neck injury
187
0.51
176
0.48
165
0.45
154
CSI/g

0.42
Ti

143
0.39
132
0.36
121
0.33
110
0.30
0 10 20 30 40 0 10 20 30 40
B/angle B/angle
(c) Thorax injury (d) Leg injury
Fig. 10. Influence laws of changing angle B of legs on injuries to key parts of the dummy.

the true value have random error. In order to evaluate whether the the higher the accuracy of the model. For the model, the allowable
fitted response surface is reasonable, the four indexes, i.e. Relative maximum error of RE, MR and RMSE are 0.2, 0.3 and 0.2, respectively
Error (RE), Maximum Error (ME), Root Mean Square Error (RMSE) and and the smaller the value is, the better. Errors of the selected 25 sample
Coefficient of Determination (R2 ) are used to evaluate response surface points were analyzed. Table 4 shows that the constructed response
fitted by the radial basis function model. surface model of radial basis function has a high accuracy, which can be
used to further analyze multi-objective optimization.
|∼
yi − yi |
RE = 100% ×
yi (6)
4.2. Multi-objective optimization model
ME = 100% × (∼
yi − yi ) (7)
The established mathematical model for multi-objective optimiza-
∑i = 1 (∼
M
yi − yi )2 tion is
RMSE =
M (8)
⎧ min[WIC (A, B, C ), Nij (A, B, C )]
⎪ 0° ⩽ A ⩽ 40°
(∼
M
SSE ∑i = 1 yi − yi )2 ⎨ 0° ⩽ B ⩽ 40°
R2 = 1 − =1− M ⎪
SST ∑i = 1 (yi − y¯)2 (9) ⎩ 0° ⩽ C ⩽ 20° (10)

where M represents the number of the samples for testing accuracy of By using the multi-objective genetic algorithm NSGA-II, the multi-
the model. ∼
yi and yi indicate the predicted value of the model and si- objective Pareto frontier of the optimal curled-up protective posture of
mulation analysis value of the ith response, respectively. ȳ stands for occupants was obtained, as shown in Fig. 12. The figure demonstrates
the average value of samples in simulation analysis. The value range of the optimal solutions of WIC and Nij obtained through response surface
the determined coefficient R2 is [0, 1] and the closer the value to 1 is, model of radial basis function and WIC and Nij always present inverse

269
W. Yang, et al. Safety Science 117 (2019) 263–275

525
0.95
510
0.90
495
0.85
480
0.80
HIC

Nij
465
0.75
450
0.70
435
0.65
420
0.60
405
0.55
0 5 10 15 20 0 5 10 15 20
C/angle C/angle
(a) Head injury (b) Neck injury
170
0.56
160 0.52
150 0.48
140 0.44
CSI/g

Ti

130 0.40

120 0.36

110 0.32

100 0.28

90 0.24
0 5 10 15 20 0 5 10 15 20
C/angle C/angle
(c) Thorax injury (d) Leg injury
Fig. 11. Influence laws of changing angle C around point H on injuries to key parts of the dummy.

Table 4
0.38 D
Error estimation of response surface model.
Response RE ME RMSE R2
0.36

WIC 5.39E−13 1.15E−12 6.25E−13 0.9968 0.34

Nij 3.83E−13 1.07E−12 4.64E−13 0.9977
0.32

0.30
WIC

proportions. The reduction of neck injury predictor Nij can constantly

increase WIC. Similarly, with the constant decrease of WIC, neck injury 0.28
predictor Nij continuously rises.
Two end points of Pareto frontiers are the optimal single-objective 0.26 C
values corresponding to different response surface models, that is, the
minimum Nij value (D point) and the minimum WIC value (F point). At 0.24 F
the optimization point D, the minimum value of neck injury obtained
0.22
through response surface model of radial basis function is Nij = 0.271
(A = 40°, B = 0° and C = 14°). At this time, by establishing simulation 0.27 0.28 0.29 0.30 0.31 0.32
model for sled test at the response of 40-0-14, simulation result is 0.287
and error values −5.57%. At optimization point F, weighted injury
Nij
response value of the dummy is the minimum, namely WIC = 0.231 Fig. 12. Pareto frontier of the optimal curled-up protective posture of occu-
(A = 40°, B = 0° and C = 6°). In addition, through simulation, the pants.

270
W. Yang, et al. Safety Science 117 (2019) 263–275

weighted injury value of the dummy in the curled-up posture at 40-0-06 40-0-08 in collision of two carriages at the speed of 48 km/h was si-
is calculated to be 0.239 and error is −3.35%. mulated.
Besides the optimal single-objective values at the two end points of The whole train was composed of eight carriages (four were pow-
Pareto frontier in collision, it generally needs to comprehensively ered while the other four were not). Except for the first three carriages,
consider two response values, so that each part of the dummy is injured the rest ones were simplified into mass points for simulation and the
slightly on the whole. In order to comprehensively consider the optimal coupler and draft gear between ends of two carriages was simulated by
neck injury and weighted injury of the dummy, the scheme of mini- using the generally used nonlinear spring elements. The buffer perfor-
mization of distance was introduced to find the equilibrium point be- mances of the first and middle carriages were simulated. The con-
tween Nij and WIC in optimization. According to the principle of structed finite element model for carriages is displayed in Fig. 14.
minimization of distance, the comprehensive optimal point is obtained Seats and dummies were horizontally placed at the three re-
when it meets the condition that the sum of distances from the two presentative positions, i.e. the front, middle and rear in longitudinal
response values of the Pareto optimal point to respective minimum direction of the first carriage. For convenience of data statistics, all
values in solution set is the smallest (Guan et al., 2018). The equation is dummies in running and static trains were numbered, as displayed in
n
Fig. 15.
⎛ ⎞ In the simulation of collision scenario of a train running collides
min Z = ⎜∑ (fik − min(fi ))2⎟
⎝ i=1 ⎠ (11) with the same type of static train on a straight track at the speed of
48 km/h, the total calculation time was set to be 1.5 s and the accel-
where n represents the number of optimized objectives and values 2
eration of gravity in Z direction was performed on the whole model.
here, which means the optimization objectives of Nij and WIC. fik in-
Through the analysis on injuries of occupants sitting naturally in the
dicates the response value of the ith optimization objective at kth re-
running train in Table 6, the most dangerous position in the running
sponse point in Pareto. According to the calculated optimal solution set
train is at dummy 6. At the same position, the collision was simulated to
of multiple objectives, the midpoint C in Fig. 12 calculated through
explore injury of the dummy in the posture with hands laced behind
Formula (11) is the equilibrium solution of the conflict between the two
head and body curled up at 40-0-08 and to compare injuries.
response values. At equilibrium point C, neck injury obtained through
response model is Nij = 0.306 and weighted injury response value is
5.2. Results
WIC = 0.249 (A = 40°, B = 0° and C = 8°). Then, the simulation model
of sled test at the response of 40-0-8 was built. The simulation results
Fig. 16 compares dynamic response processes of the dummy in
are 0.326 and 0.262 and errors are −6.13% and −4.96%. The optimal
normal sitting posture and dummy 6 applying the multi-objective op-
posture with body curled up is shown in Fig. 13.
timization scheme at 40-0-08.
The optimization schemes for obtaining optimal solutions to single
The dummy in the optimized curled-up posture contacts the
objectives and the equilibrium solution of the curled-up posture in
backrest of the front seat at the moment of secondary collision and the
Pareto frontier is comprehensively arranged in Table 5.
upper limbs and legs support the body and share the stress on head of
dummy sitting naturally. Acceleration curves of head and thorax under
5. Verification test
the two states are shown in Fig. 17. Injury value of head of the dummy
in the protective posture with body curled up greatly reduces in com-
5.1. Simulation system for standard frontal collision of trains at the speed of
parison with that of the dummy sitting normally. Compared with
48 km/h
HIC = 834 in the original posture, injury of head of dummy in the
optimized posture decreases by 88%, which shows good protection
It is also necessary to verify the effectiveness of the multi-objective
effects. In collision, the upper limbs of the dummy press and transfer
optimal posture with body curled up at 40-0-08. For this purpose, the
forces to the thorax, so that thorax and head are curled up as a whole to
finite element model for a whole train with eight carriages was built by
suffer the collision. Therefore, stress on the thorax increases, while
selecting a structure of high-speed bullet train as prototype according to
injures to the thorax because stress exceeds the tolerance limit do not
crashworthiness standard of Federal Railway Administration (FRA).
happen.
Through the model, the injury process of occupants sitting naturally
Fig. 18 demonstrates comparisons of axial stress curves and bending
and occupants in the curled-up posture with hands laced behind head at
moment curves of neck of occupant dummy 6 in frontal collision of
trains at the speed of 48 km/h. Injuries of neck of dummy 6 in the two
postures are different. Neck of the occupant sitting naturally shows
compression-stretch injury, while the dummy in the posture with body
curled up at 40-0-08 presents compression-bending injury to neck. It
can be obviously seen from the figure that axial stress and bending
moment of the neck of the dummy in the posture with body curled up
decrease by nearly 50% and 80% separately, implying that stress on the
neck greatly reduces.
Fig. 19 displays stress-time curves of thigh bone and tibia. When the
dummy moves in the 40-0-08 posture, stress on thigh bone obviously
decreases. The response posture of the dummy in Fig. 19(a) shows that
in the 40-0-08 posture, the uplift of feet helps the lower limbs resist
impacts, which avoids knees directly colliding with seats, thus resulting
in that thigh bone bears all impacts. Stress on thigh bone reduces after
optimization. In similar, due to the uplift of lower limbs, the resistance
of feet to seats in collision reduces direct impacts on tibias. It can be
intuitively seen from the stress curves of tibias of the dummy in colli-
sion in Fig. 19(b) that the tibias of the dummy in the optimized curled-
up posture are protected in collision.
Fig. 13. The optimal curled-up posture at 40-0-08 in multi-objective optimi- After optimization, acceleration and stress of head, neck and legs of
zation. the dummy are improved obviously. The acceleration of the thorax

271
W. Yang, et al. Safety Science 117 (2019) 263–275

Table 5
The optimization schemes under different constraints.
Posture Nij Error WIC Error

Response value Simulation value Response value Simulation value

40-0-14 0.271 0.287 −5.57% 0.377 0.382 −1.31%

40-0-08 0.306 0.326 −6.13% 0.249 0.262 −4.96%
40-0-06 0.321 0.325 −1.23% 0.231 0.239 −3.35%

Fig. 14. Finite element model for the whole

train.

Fig. 15. Numbering of the dummies.

Table 6 injury criteria of each part, the values in natural sitting posture and
Injury values of each part of the dummies in collision at the speed of 48 km/h. optimized posture at 40-0-08 were obtained and the results are dis-
HIC Nij CSI (g) D (mm) TI (left) TI (right) WIC played in Table 7.
It can be observed that the optimized posture obviously protects
Dummy1 700 2.24 36.21 4.27 0.84 0.55 0.78 head, neck, thorax and legs as well as the whole weighted injury and all
Dummy2 704 1.94 37.03 4.24 0.71 0.53 0.79 parts of the dummy are not injured. By comprehensively considering
Dummy3 715 1.47 32.54 3.68 0.94 0.91 0.78
Dummy4 654 2.20 36.09 4.30 0.49 0.78 0.74
dynamic response of the dummy in the optimized posture and injury
Dummy5 665 2.05 37.18 4.00 0.60 0.64 0.75 indexes of each key part and weighted injury, it is verified that the
Dummy6 834 1.36 30.97 3.75 0.50 0.64 0.88 optimized curled-up posture is correct and effective.

slightly increases but does not exceed tolerance limit of the thorax. 6. Conclusions
Therefore, the optimization scheme of curled-up posture of the dummy
has obvious protection to each main part of the dummy. By combining The curled-up posture with hands laced behind head can effectively

(a) Response process of the dummy in natural sitting posture

(b) Response process of dummy 6 in curled-up posture at 40-0-08

Fig. 16. Dynamic response posture of dummy 6 in collision at the speed of 48 km/h.

272
W. Yang, et al. Safety Science 117 (2019) 263–275

natural
natural 35
200 40-0-08
40-0-08
30

150 25
R su ltantacce l/g

Rsultantaccel/g
20
100
15

10
50
5

0 0
0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30 0.32 0.34 0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30 0.32 0.34

Time/s Time/s
(a) Acceleration curves of the head (b) Acceleration curve of the thorax
Fig. 17. Acceleration curves of head and thorax of occupant dummy 6 in the natural and 40-0-08 posture.

reduce injury degrees of head and neck of occupants in collision. In after leaning back for 5°. The difference is that injury to head of
frontal collision at the speed of 48 km/h, occupants in the posture at 40- occupants suddenly reduces after increasing the back leaning angle
0-08 is obviously protected from injuries of head, neck, thorax and legs to 15°.
as well as the whole weighted injury. Therefore, the seats can be de- (3) In the optimized curled-up posture with hands laced behind head at
signed to turn to the optimal curling angle automatically in collision, so 40-0-08, injury to head of occupants reduces by 88% and axial
as to assist occupants reaching the optimal protective state. stress and bending moment of neck decrease by nearly 50% and
80% separately. CSI of thorax slightly increases but does not exceed
(1) The changing angle of legs significantly affects head, neck and tolerance limit to injure thorax.
thorax of occupants, and changing angle of waist has the most
profound influences on legs. In addition, head and lower limbs of Acknowledgements
occupants are least sensitive to back leaning angle at H point and
changing angle of legs separately, while injuries to neck and thorax This research was undertaken at the Key Laboratory for Traffic
of occupants are affected least by changing angle of waist. Safety on Track of the Ministry of Education, Central South University,
(2) With the increase of changing angle of waist of occupants in the China and Department of Mechanical Engineering University of
posture with body curled up, injuries to head and legs decrease, Alberta, Edmonton, Canada. The author W.L. Yang gratefully ac-
while neck injury constantly fluctuates. Moreover, injury degree of knowledge the support from the China Scholarship Council (Grant no.
thorax reduces after the initial increase. As changing angle of legs 201806370212) and the National Natural Science Foundation of China
rises, injury degrees of head and neck of occupants constantly in- (Grant no. 51775558). This paper also supported by Innovation-Driven
crease, while injury value of the thorax firstly decreases and then Program of the Central South University (No. 2018CX023), the Nature
dramatically rises and finally reduces again. Injury to legs increases Science Foundation for Excellent Youth Scholars of Hunan Province
after constant decrease. The increase of angle around H point al- (Grant No. 2019JJ30034) and the Shenghua Yu-ying Talents Program
leviates injuries of thorax and legs of occupants. Injury values of of the Central South University (Principle Investigator: Pro. Suchao
head and neck of occupants significantly rise and become gentle Xie).

2000 natural 80000 natural

40-0-08 40-0-08
1000
60000
0

40000
-1000
My/Nmm

My/Nmm

-2000 20000

-3000
0
-4000
-20000
-5000

-6000 -40000
0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30 0.32 0.34 0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30 0.32 0.34

Time/s Time/s
(a) Axial stress curve of the neck (b) Bending moment curve of the neck
Fig. 18. Axial stress and bending moment curves of the neck of dummy 6 in the natural and 40-0-08 posture.

273
W. Yang, et al. Safety Science 117 (2019) 263–275

4000 4000
natural natural
2000 40-0-08 40-0-08
2000
0
0
-2000

-4000 -2000

Fz/N
Fz/N

-6000
-4000
-8000

-10000 -6000

-12000
-8000
-14000
0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30 0.32 0.34 0.16 0.18 0.20 0.22 0.24 0.26 0.28 0.30 0.32 0.34

Time/s Time/s
(a) Stress on thigh bone (b) Stress on tibias
Fig. 19. Stress curves of legs of dummy 6 in the natural and 40-0-08 posture.

Table 7 of passengers on bench seats in a train collision. Quart. Rep. RTRI 49 (1), 47–52.
Natural and optimized posture injury results comparison. Liu, G.W., Xie, J., Xie, S.C., 2018. Experimental and numerical investigations of a new U-
shaped thin plate energy absorber subjected to bending and friction. Thin-Walled
Scheme HIC Nij CSI (g) D (mm) TI (left) TI (right) WIC Struct. 131, 258–273.
Lin, Y., Yao, W.M., Sun, D.D., 2005. Study on automobile seat safety performance during
Natural 834 1.36 30.97 3.75 0.50 0.64 0.88 impact. Trans. Beijing Instit. Technol. 25 (1), 18–26.
Optimized 96 0.38 52.29 5.70 0.28 0.28 0.26 Lu, G., 2002. Energy absorption requirement for crashworthy vehicles. Proc. Instit. Mech.
Eng. Part F: J. Rail Rapid Trans. 216 (1), 31–39.
Luo, Y.H., Fan, H.L., 2018. Energy absorbing ability of rectangular self-similar multi-cell
sandwich-walled tubular structures. Thin-Walled Struct. 124, 88–97.
References Parent, D., Tyrell, D.C., Perlman, A.B., 2004. Crashworthiness analysis of the Placentia,
CA rail collision. Int. J. Crashworth. 9 (5), 527–534.
Peng, Y., Chen, Y., Yang, G.K., Dietmar, O., Remy, W., 2012. A study of pedestrian and
Al Tabiei, Lawrence, Charles, Fasanella, Edwin L., 2009. Validation of Finite Element
bicyclist exposure to head injury in passenger car collisions based on accident data
Crash Test Dummy Models for Predicting Orion Crew Member Injuries During a
and simulations. Saf. Sci. 50, 1749–1759.
Simulated Vehicle Landing. NASA/TM-2009-215476.
Peng, Y., Deng, W.Y., Yao, S.G., 2015. Study on the collision performance of a composite
Anton, K., Igor, T., Christine, Q.W., 2017. Numerical parametric study on factors affecting
energy-absorbing structure for subway vehicles. Thin-Walled Struct. 94, 663–672.
passenger safety in motorcoach frontal collision. Int. J. Crashworth. 22, 214–226.
Peng, Y., Wang, S.M., Yao, S.G., Xu, P., 2017. Crashworthiness analysis and optimization
AV/ST9001, 2002. Vehicle Interior Crashworthiness. Association of Train Operating
of a cutting-style energy absorbing structure for subway vehicles. Thin-Walled Struct.
Companies, London.
120, 225–235.
Cao, L.B., Dai, H.W., Zhang, R.F., Li, Z.G., Zhang, G.J., 2014. An analysis on the influence
Praveen, Kumar A., Nalla, Mohamed M., Jusuf, A., Dirgantara, T., Gunawan, L., 2018.
of seat backrest angle on occupant crash safety. Automot. Eng. 36 (12), 1461–1465.
Axial crash performance of press-formed open and end-capped cylindrical tubes-a
Chen, S.J., Yu, H.Y., Fang, J.G., 2018. A novel multi-cell tubal structure with circular
comparative analysis. Thin-Walled Struct. 124, 468–488.
corners for crashworthiness. Thin-Walled Struct. 122, 329–343.
Prochowski, A., Zuchowski, L., 2014. Analysis of the influence of passenger position in a
Deb, K., Pratap, A., Agarwal, S., Meyarivan, T., 2002. A fast and elitist multiobjective
car on a risk of injuries during a car accident. Mainten. Reliabil. 16 (3), 360–366.
genetic algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6 (2), 182–197.
Sarba, G., Dilip, B., Jacob, K., 2011. LSTC Hybrid III 50th Fast Dummy Positioning&Post-
EN15227:2008, 2009. Railway Applications-Crashworthiness Requirements for Railway
Processing. LSTC, USA.
Vehicle Bodies. BS, British.
Severson, K.J., Parent, D.P., Tyrell, D.C., 2004. Two-car impact test of crash-energy
European Union Contract, 2001. No. BRPR-CT97_0 457, Project No. BE-3092. SafeTrain
management passenger rail cars: analysis of occupant protection measurements. In:
Train Crashworthiness for Europe, Final report.
ASME 2004 International Mechanical Engineering Congress and Exposition, pp.
Gao, G.J., Tian, H.Q., 2007. Train's crashworthiness design and collision analysis. Int. J.
87–96.
Crashworth. 12 (1), 21–28.
Severson, K.J., Parent, D.P., 2006. Train-to-train impact test of crash energy management
Gao, G.J., Guan, W.Y., Li, J., Dong, H.P., Zou, X., Chen, W., 2017. Experimental in-
passenger rail equipment: occupant experiments. 2006 ASME International
vestigation of an active-passive integration energy absorber for railway vehicles.
Mechanical Engineering Congress and Exposition. American Society of Mechanical
Thin-Walled Struct. 117, 89–97.
Engineers.
GM/RT2100, 2010. Requirements for Rail Vehicle Structures. Rail Safety and Standards
Simons, J.W., Kirkpatrick, S.W., 1999. High-speed passenger train crashworthiness and
Board, London.
occupant survivability. Int. J. Crashworth. 4 (2), 121–132.
Guan, W.Y., Gao, G.J., Li, J., Yu, Y., 2018. Crushing analysis and multi-objective opti-
Tyrell, D.C., 2002. US rail equipment crashworthiness standards. Proc. Instit. Mech. Eng.
mization of a cutting aluminum tube absorber for railway vehicles under quasi-static
Part F: J. Rail Rapid Trans. 216 (2), 123–130.
loading. Thin-Walled Struct. 123, 395–408.
Tyrell, David C., Severson, Kristine J., Marquis, Brain P., 1995. Crashworthiness and
Hecht, Markus, 2004. Safetram addresses crashworthiness of trams and light rail vehicles.
Occupant Protection in Transportation System. vol. 210 ASME, AMD.
Railw. Gazette Int. 160 (5), 288–290.
Wei, L., Zhang, L.L., Cui, J., Song, Y., 2015. Response simulation and injury prediction of
Hosseini, T.P., Nankali, A., 2010. Study on characteristics of a crashworthy high-speed
standing. J. China Railw. Soc. 37 (1), 16–23.
train nose. Int. J. Crashworth. 15 (2), 161–173.
Wei, L., Zhang, L.L., 2017. Evaluation and improvement of crashworthiness for high-
Hosseini, T.P., Bayat, V., 2011. Study on crashworthiness of wagon's frame under frontal
speed train seats. Int. J. Crashworth. 1–8.
impact. Int. J. Crashworth. 16 (1), 25–39.
Wen, H., Lin, L.T., Chen, D.Q., 2011. Feature of survived casualties and treatment after
Hou, J.J., Zhang, X.R., Ren, L.H., 2011. Simulation analysis of vehicle seat seating re-
“July 23” EMU railway accident at. Wenzhou 20 (12), 1248–1250.
ference point and pressure distribution. Mod. Transp. Technol. 8 (4), 65–67.
Wei, Q., Wang, Y.M., Wen, B., 2012. Safety research on collision of railway locomotives
Jacobsen, K., Tyrell, D., Perlman, B., 2004. Impact test of a crash-energy management
and rolling stock aboard. For. Roll. Stock 49 (5), 39–46.
passenger rail car. In: Proceedings of the ASME/IEEE Joint Railroad Conference, pp.
Xie, S.C., Tian, H.Q., 2013. Influencing factors and sensitivity analysis of occupant impact
1–10.
injury in passenger compartment. Traffic Inj. Prev. 14, 816–822.
Kazuma, N., Koji, O., Hiroaki, S., Daisuke, Suzuki, 2012. Simulation of passenger behavior
Xie, S.C., Yang, W.L., Li, H.H., Wang, L., 2017a. Impact characteristics and crash-
in board a commuter train in the event of a level crossing accident. Quart. Rep. RTRI
worthiness of multi-cell, square, thin-walled, structures under axial loads. Int. J.
53 (4), 235–240.
Crashworth. 22 (5), 503–517.
Kirkpatrick, S.W., Schroeder, M., Simons, J.W., 2001. Evaluation of passenger rail vehicle
Xie, S.C., Yang, W.L., Wang, L., Li, H.H., 2017b. Crashworthiness analysis of multi-cell
crashworthiness. Int. J. Crashworth. 6 (1), 95–106.
square tubes under axial load. Int. J. Mech. Sci. 121, 106–118.
Koji, O., Hiroaki, S., Ayano, S., 2002. Estimation of passenger movements against the
Xie, S.C., Li, H.H., Yang, W.L., Wang, L., 2017c. Crashworthiness optimisation of a
impact in train collision. Quart. Rep. RTRI 43 (2), 11–82.
composite energy-absorbing structure for railway vehicles. Struct. Multidiscip.
Koji, Omino, Hiroaki, Shiroto, Ayano, Saitoh, Hiroharu, Endoh, 2008. Behavior analysis

274
W. Yang, et al. Safety Science 117 (2019) 263–275

Optim. 8, 1–15. 125–136.

Xie, S.C., Wang, L., Yang, W.L., Li, H.H., 2018. Energy absorption performance of thin- Zhang, K., Xu, P., Yao, S.G., 2016. Optimization on electric passenger locomotive
walled metal plate due to upheaval deformation based on experiments and numerical crashworthiness design based on genetic algorithm. J. Railw. Sci. Eng. 13 (6),
simulation. Thin-Walled Struct. 131, 258–273. 1186–1192.
Xue, X., Smith, R.A., Schmid, F., 2005. Analysis of crush behaviors of a rail cab car and Zhang, Y., Ge, P., Lu, M., Lai, X., 2018. Crashworthiness study for multi-cell composite
structural modifications for improved crashworthiness. Int. J. Crashworth. 10 (2), filling structures. Int. J. Crashworth. 23 (1), 1–15.

275