Thin-Walled Structures 131 (2018) 286–296

Contents lists available at ScienceDirect

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

Full length article

Multi-response optimization design of tailor-welded blank (TWB) thin- T

walled structures using Taguchi-based gray relational analysis
⁎
Fengxiang Xua,b, , Suo Zhanga,b, Kunying Wua,b, Zhinan Donga,b
a
Hubei Key Laboratory of Advanced Technology of Automotive Components, Wuhan University of Technology, Wuhan 430070, China
b
Hubei Collaborative Innovation Center for Automotive Components Technology, Wuhan University of Technology, Wuhan 430070, China

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

Keywords: In order to further improve crashworthiness and reduce weight, tailor-welded blanks (TWBs) have been widely
Crashworthiness applied in auto-body design. In this paper, the discrete optimization design of TWBs structures with top-hat thin-
Thin-walled structures walled section subjected to front dynamic impact is performed by using Taguchi-based gray relational analysis.
Tailor-welded blank (TWB) Material grades and thicknesses with three levels are taken as discrete design variables. The total energy ab-
Taguchi method
sorption (EA), the total weight (Mass) and the peak crashing force (Fmax) are chosen as optimization indicators.
Gray relational analysis
Considering the uncertain weight ratio of responses, four different cases would be analyzed. In order to de-
termine the optimal parameter combination more accurately and eliminate errors from range analysis, the
analysis of variance (ANOVA) would be performed. The optimized results demonstrate that it is feasible to
increase the crashworthiness of TWBs by increasing the gray correlation of the structure. Compared to initial
structure, case 1 (w(Fmax):w(EA):w(Mass)= 1/3:1/3:1/3) has the largest improvement among the four cases, i.e.,
the Fmax and the Mass are reduced by 29.3% and 2.7%, respectively, while the EA is increased by 3.5%. The
discrete optimization method with only 27 iterations is a low computing cost or cost-effective and provides some
guidance for some similar structural design. More comprehensive studies are essential to optimize performance
of multi-components with more discrete variables.

1. Introduction time, high productivity, and no secondary treatment, etc [24].

Lightweight design of the vehicle is a significant important aspect
when the indispensable requirements such as stiffness, strength,
crashworthiness and NVH, etc have been satisfied. Generally, there are
three main ways to achieve lightweight. The first is from the light
material such as widely-applied high strength steel (HSS) [1], magne-
sium [2,3], and composite materials [4,5], etc. The second is from the
manufacturing technologies such as tailor welded blank (TWB) [6–8],
and hot thermoforming [9], etc. And the third is from the structural
optimization design such as functional gradient structure [10–13],
novel section structures [14–17], topological optimization structures
[18,19], and some advanced optimization methods for its [20,21], etc.
Those lightweight ways have been widely conducted to reduce weight
and improve corresponding performance [22].
Among those ways, TWBs are defined as effective energy-absorbing
structures with different materials or thicknesses welded together to
form an integral component through the laser welding process [23].
This processing can be achieved due to the advantages of laser welding
technology such as the narrow welds width, short welding processing
Based on abovementioned advantages of TWB processing tech-
nology, many researchers have studied the formability and solder joint
distribution of TWB. For example, Chan et al. [25] performed a form-
ability analysis concerning the mechanical characteristics of the weld
zones of TWBs made of cold rolled steel sheets with different thick-
nesses. They analyzed the effect of the thickness ratios on the forming
limit diagram (FLD) for TWBs. Xu [26] numerically compared the
modeling strategies of the weld line under quasi-static and dynamic
events, and proposed a novel method with crossover scheme. Gaied
et al. [27] proposed a new method to overcome the formability of TWB
and was able to accurately predict the unique characteristics for TWB
formation early in the design process.
Apart from the forming of TWBs, their crashworthiness performance
is also one of the most important cases, which reduces injury risk for
passengers. Therefore, it is essential to study the crashworthiness of
TWB structures for improving the safety. Song et al. [28] investigated
the crashworthiness of three different types of TWB with hat-shaped
section and concluded that the optimized TWB tubes could improve
energy absorption and enhance the reliability. Xu et al. [29] carried out

⁎
Corresponding author at: Hubei Key Laboratory of Advanced Technology of Automotive Components, Wuhan University of Technology, Wuhan 430070, China.
E-mail address: xufx@whut.edu.cn (F. Xu).

https://doi.org/10.1016/j.tws.2018.07.007
Received 15 April 2018; Received in revised form 22 June 2018; Accepted 3 July 2018
0263-8231/ © 2018 Elsevier Ltd. All rights reserved.
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

multi-objective optimization design of a multi-component TWB struc- 2.1. Taguchi method

ture that involved both the B-pillar and inner door system subjected to a
side impact. Zhu et al. [30] proposed an integrated approach using fi- Taguchi method takes signal to noise ratios (S/N ratios) as the
nite element analysis, an artificial neural network (ANN), and a genetic characteristic values of the responses to measure the current values
algorithm (GA) for the optimization design of an inner door panel with deviating from the desired values [31,32]. The quality characteristic in
a TWB structure to reduce weight reduction and improve crash- the analysis of the S/N ratios is divided into the-larger-the-better type
worthiness. characteristic (Eq. (1)), the-smaller-the-better type characteristic (Eq.
There are also a great number of researchers to investigate opti- (2)) and nominal-the-better type characteristic.
mization algorithms in discrete space. Those advanced optimization
algorithm combined with finite element analysis can reduce the number 1
x i (k ) = −10 lg ⎜⎛ 2 ⎟⎞ For the−larger−the−better response
of experiment tests and save a lot of time compared with traditional ⎝ i (k ) ⎠
y (1)
optimization algorithm to a considerable extent. Important crucial ex-
periments combinations which could comprehensively reflect situation x i (k ) = −10 lg(yi2 (k )) For the−smaller−the−better response (2)
of various factors were conduct by using Taguchi orthogonal table where “the-larger-the-better” means that the greater the value of the
[31,32]. However, a single Taguchi analysis only applies to mono-ob- response, the better the collision system, such as total absorbed energy
jective optimization, which greatly limits its use. At the same time, and specific energy absorption, etc. yi (k ) is the original experimental
there are increasingly attentions to gray correlation due to the out- data or responses for kth response at ith trial, x i (k ) is the value of S/N
standing application of its in multi-objectives. Eskandari et al. [33] si- ratios for kth response at ith trial.
multaneously optimized surface roughness, tool wear and volume of
material by using gray relational analysis, and finally obtained the best 2.2. Gray relational method
cutting parameters. Cai et al. [34] studied multi-responses optimization
of S-rail extracted from the frontal body to improve its crashworthiness Gray relational analysis (GRA) is a measurement method in gray
by using gray relational analysis, and the proposed method reduced the system theory that analyzes the degree of relation for different datasets
peak collision force to 26.81% as well as increased the corresponding in a discrete space [33]. It is based on the gray relational coefficient
specific energy absorber by 176.06%. Most researchers used gray ana- between sample dataset and ideal dataset, and then uses the gray cor-
lysis alone for multi-objective optimization without considering the relation to rank order of relation among factors or responses. The ori-
robustness of the system, let alone combine it with Taguchi analysis. ginal responses about S/N ratios should be linearly normalized in the
Regarding the design variables of TWB, the main and obvious fea- range of 0–1 (called the normalization of S/N ratios). Thus, according
ture is the discrete such as the material grades and even material types, to the different characteristics of the dataset, GRA has different nor-
etc. To the author's best knowledge, the crashworthiness optimization malization formulas illustrated in Eqs. (3)–(5) [35,36].
design of TWB structures considering discrete variables was seldom
reported, to say nothing of Taguchi-based gray relational analysis for it. x i (k ) − min x (k )
x i′ (k ) = For the−larger−the−better response
In this paper, Taguchi-based gray relational analysis would be fur- max x (k ) − min x (k )
ther performed to multi-response optimization design of TWB thin- (3)
walled structures under crashing event. Firstly, a finite element model
x i (k ) − min x (k )
(FEM) about TWB crashing verified by physical experiment is con- x i′ (k ) = 1 − For the−smaller−the
max x (k ) − min x (k )
structed to simulate actual collision behavior. Then, orthogonal ex-
periments are carried out by changing design variables with different −better response (4)
thicknesses and material combinations based on Taguchi orthogonal
x i (k ) − x 0 (k )
array. Furthermore, analysis of Taguchi method, gray relational ana- x i′ (k ) = 1 − For the−nominal−the−best response
max x (k ) − x 0 (k )
lysis and analysis of variance are applied into optimization design of
HSS TWBs thin-walled structures. Finally, the optimal results obtained (5)
from which significant effect factors and optimal level combinations are where min x(k) is the minimum original experimental data or simula-
presented. The optimum results demonstrate that the optimized struc- tion data for kth response at all trials, max x(k) is the maximum original
ture can reduce the weight and improve the crashworthiness to a cer- experimental data or simulation data for kth response at all trials, x0(k)
tain extent, and provide some guidance for some similar structural is the nominal for kth response, xi(k) is normalized value for kth re-
design. sponse at ith trial, i is the number of trials. In this paper, k is 1, 2 or 3
indicating Fmax, EA and Mass, respectively.
2. Design procedure of TWB structures Thus, the gray relational coefficient could be further calculated as
follows:
With regard to TWB structures, the inherent properties of the
structure have a relatively large effect on energy absorption and min ⎛⎜min x i (0) − x i′ (k ) ⎞⎟ + ρ max ⎛⎜max x i (0) − x i′ (k ) ⎞⎟
lightweight, especially thickness and material grade. Material grades k
⎝ i ⎠ k
⎝ i ⎠
ξi (k ) =
and thicknesses of base materials fabricated into TWB structures are
based on a series of constant discrete values. Therefore, the discrete x i (0) − x i′ (k ) + ρ max ⎛⎜max x i (0) − x i′ (k ) ⎞⎟
k
optimization design problem can be defined as determining the optimal ⎝ i ⎠ (6)
values of the some discrete design parameters in order to obtain max- where xi(0) is ideal value at ith trial, ρ is resolution coefficient
imum or minimum of objective functions or responses. To our best ( ρ ∈ [0, 1]), the smaller the ρ ,the greater the resolution is. In this ex-
knowledge, Taguchi method is one of the most popular discrete opti- periment, the ρ value is set as 0.5 and the reference data for the ideal
mization methods applied in engineering problems, but it's limited by a value is taken as 1.
problem with a single quality characteristic. To analyze and solve a For a practical engineering system, different performance responses
problem with multiple responses, the Taguchi-based gray relational have different effects on the system. The different effects on perfor-
analysis is considered an effective method [35,36]. Hence, it is crucial mance can be transformed by using Eq. (7) with diverse weight ratios.
that discrete optimization algorithm based on Taguchi-based gray re- n n
lational analysis is investigated. In this study, the procedure of discrete γi (m) = ∑ ωk ξi (k ); ∑ ωk = 1
optimization design for TWB structures is showed in Fig. 1. k=1 k=1 (7)

287
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Fig. 1. Flowchart of the discrete design for TWB structures.

where ωk is the weight ratio for kth response, γi (m) is gray correlation SSa is the sum of squares of deviation for certain parameter, SSe is the
for case m at ith trial. sum of squares of deviation for certain error, ravg(m) is mean gray
correlation, MS is the mean squares, F is significant value, C is con-
2.3. Analysis of variance method tribution rate and D is freedom.

The method of GRA is based on range analysis, which is simple and

easy to understand. But range analysis cannot distinguish between data 3. Numerical modeling and variation
fluctuations caused by changes in test conditions (changes in the level
of the factors) and those by experimental errors. Therefore, the analysis 3.1. Geometry and materials description
of variable (ANOVA) would be performed to solve this problem by in-
vestigating design parameters significantly affecting the quality char- The structure considered in this paper is the TWB thin-walled
acteristic and giving a precise quantitative estimate for its contribution. column with a top-hat section fabricated by four parts as shown in
The key indicators of ANOVA are the contribution rate and F ratio Fig. 2. The structural geometry is similar to that of a typical design for
shown in Eqs. (8)–(12). the front side rail of a passenger car and the detail dimensions are given
n
in Table 1. The parameters which affect the crashworthiness of the TWB
SST = ∑ (γi (m) − γavg (m)) structures include the base materials and thickness ratios.
i=1 (8) Dual-phase steel as a combining high-strength and superplastic HSS
has been widely applied in the automobile field. Therefore, dual-phase
t
steels with various material grades are utilized to investigate the
SSa = ∑ ⎛⎜γa (m) − γavg
t
(m) ⎞⎟
crashing performance of TWBs in this study. It requires special attention
i=1 ⎝ ⎠ (9)
to the fact that high rates of deformation always occur for energy ab-
SS sorbing structures during the dynamic crashing. Therefore, the strain
MS =
D (10) rate effect of HSS materials must be considered to accurately simulate
ssa its stress-strain relationships. The stress-strain curves available for finite
F= element (FE) software are obtained by transforming engineering stress-
sse (11)
strain relations into true stress strain curves based on standard tensile
ssa tests. Here two materials (DP590 and DP780) are used as examples to
C=
ssT (12) illustrate the effect of strain rate on material strength. Their relation-
ship curves are given in Fig. 3.
where n is the total number of experiments, t is the number of experi-
ments at that level, SST is the sum of squares of deviation for all factors,

288
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Fig. 2. Geometry dimensions of the top-hat specimen (welded via two individual parts) for impact test: (a) top view; (b) cross section shape (side view).

Table 1 car and floor is defined as 0.3, the friction coefficient between the TWB
Geometry parameters of top-hat structures for dynamic impact (Unit: mm). and rigid wall is defined 0.6 and dynamic and static friction coefficient
L D d x a b f R1 = R2 t
for TWB itself both are set to 0.2.

400 5 20 0–400 80 80 30 4 0.8–1.8

3.3. Validation of FE model

To perform parameter analysis for the next step, it is necessary to

3.2. Finite element modeling
verify the accuracy of the model. The TWB structure with top-hat sec-
tion is divided into identical two pats, i.e., the weld line is situated in
Based on the FE crashing simulation system developed in LS-DYNA
the middle of the whole structure. In this experiment test, one typical
971, the structure of TWB with top-hat shaped section on the front of
case is considered here: the materials with two parts keep changed
the sled car is used to simulate crashing process under an initial velocity
(DP590 and DP780) and the thickness values of two parts are set as
of 30 km/h. Before simulating crashing process, FE model shown in
1.0 mm (front or impacting end) and 1.4 mm (rear or fixed end), re-
Fig. 4 should be built up to approximate the physical experiment tests.
spectively. To verify the established FE model, the collapse mode and
In this task, TWBs is meshed using single-point integration
the corresponding acceleration-time relationships are compared with
Belytschko-Lin-Tsay shell elements with 2 mm, which makes the cal-
tests as shown in Fig. 5. It can be observed that a little different in
culation speed very fast as well as stable, and is suitable for analyze
deformation modes and accelerations occur due to manufacturing error.
large deformation problems. In addition, in order to maintain energy
However, overall acceleration level and deformation mode of FE ana-
conservation, hourglass energy control and interface energy control
lysis is in agreement with experiments. Therefore, the results well de-
caused by reducing integration shell and initial penetration must be
monstrate that the established FE model can successfully predict the
considered. Moreover, the rigidity of the obstacle wall is much greater
crashworthiness behaviors of the TWB columns with an acceptable
than the deformed structures. Therefore, the deformation of obstacle
accuracy and will be used for the next parametric analysis.
wall defined as a rigid wall could be negligible at the stage of the
crashing.
Meantime, as for connection and contact of TWBs, the upper and 3.4. Structural crashing indicators
lower parts are connected by a flexible welding spots, and the front and
rear parts are rigidly connected by weld seam. In addition, in order to There have been some effective indicators available to evaluate the
prevent the penetration of the components, the components themselves crashworthiness performance for typical thin-walled top-hat structures,
are set to automatic single-surface contact algorithm, an automatic the energy absorption (EA), the total weight (Mass), and the peak
nodes-to-surface contact algorithm is defined between components. crashing force (Fmax) as the main indicators are usually used to evaluate
At the same time, taking into account the actual weight of the car, a comprehensive capacity of reducing weight and improving crash-
mass element was added to the center of the sled car to ensure that it worthiness.
reached 536 kg. As for friction, the friction coefficient between the sled Through the simulation analysis, the axial force, displacement and
time relation curves can be obtained. With the force-displacement

900 1000
Effective Stress (MPa)

900
Effective Stress (MPa)

800

700 800

700
600 16.150/s
15.596/s
600 78.730/s
500 69.316/s
389.591/s 437.100/s
500
400 709.390/s 603.320/s
400
300
0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 0.00 0.02 0.04 0.06 0.08 0.10 0.12
Effective Plastic Strain Effective Plastic Strain
(a) DP 590 (b ) DP78 0
Fig. 3. Effective stress-effective stain curves at different strain rates.

289
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Fig. 4. Finite element model of the crashing system.

relationship, the energy absorption (EA) can be calculated as follows:

δe
EA = ∫0 F (δ ) dδ (13)

where δe is the effective crashing displacement, F(δ ) is instantaneous

crashing force at each displacement.
At the same time of pursuing to improve crashworthiness, light-
weight should be not ignored. The specific energy absorbed is a key
indicator to distinguish energy absorption capabilities of different ma-
terials and weights. The energy absorption of unit mass can be calcu-
lated as follows:

EA
SEA =
m (14)

4. Results and discussion

4.1. Design variables
The optimum object is made up by four different parts with re-
spective material property and thickness called as Up1, Low1, Up2, and
Low2 and is illustrated in Fig. 6. The material types and the thicknesses
of each part are taken as independent discrete variables with three le-
vels for selection, presented in Table 2.
After determining the design variables, the orthogonal array
L27(38) is selected to assign 8 design variables with three levels.
Fig. 6. Schematic of discrete optimization objectives.
4.2. Analysis of the S/N ratios
The large-the-better type characteristic is chosen for the total energy
absorption as well as the small-the-better type characteristic is chosen
for the peak crashing force and mass according to quality character-
istics. The results of multiple responses and corresponding S/N ratios
calculated by Eqs. (1) and (2) are shown in Table 3.
Based on the S/N ratios for each response, the average S/N ratios of

35
Simulation result
30 Experiment-1
Experiment-2
25
Acceleration (g)

20

15

10

0
0.00 0.01 0.02 0.03 0.04 0.05
Time (s)

Fig. 5. Comparison between experiments and FE analysis.

290
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Table 2
Design variables and corresponding candidate choices of individual part.
Factor Description (Unit) Level 1 Level 2 Level 3
(low) (medium) (high)

A Material grade of Up1 DP590 TR590 DP780

B Material grade of DP590 TR590 DP780
Low1
C Material grade of Up2 DP780 TR780 DP980
D Material grade of DP780 TR780 DP980
Low2
E Thickness of Up1 0.8 1.0 1.2
(mm)
F Thickness of 0.8 1.0 1.2
Low1(mm)
G Thickness of Up2 1.2 1.4 1.6
(mm)
H Thickness of 1.2 1.4 1.6
Low2(mm)

Fig. 7. Single objective optimization for each response.

Table 3
Simulation results and corresponding S/N ratios.
each design variable corresponding to levels could be calculated to
Scenario Fmax (kN) S/N EA (kJ) S/N Mass (kg) S/N
compare the effect of parameters on different responses. The compar-
1 111.523 − 40.947 22.62 27.090 2.687 − 8.585 ison results are quantitatively presented in Table 4. It is noted that the
2 138.417 − 42.824 23.609 27.462 3.243 − 10.219 larger the range (Δ) of S/N ratios is, the more significant the effect of
3 197.859 − 45.927 24.137 27.654 3.780 − 11.550 this factor on responses is. By analysis the ranking of S/N ratios for
4 159.226 − 44.040 20.827 26.373 2.793 − 8.921
different factors, it can be found that the thickness of Up2 has the
5 245.332 − 47.795 22.735 27.134 3.330 − 10.449
6 179.576 − 45.085 21.553 26.670 3.606 − 11.141 greatest influence on the Fmax (Δ = 8.461), which is followed by the
7 170.218 − 44.620 21.599 26.689 2.880 − 9.188 thickness of Low1 (Δ = 5.304). On another hand, the thickness of Low2
8 237.724 − 47.521 22.638 27.097 3.155 − 9.980 and material grade of Up1 has little effect on Fmax. Among all the factors
9 216.244 − 46.699 24.167 27.664 3.693 − 11.348 that influence the EA, thicknesses of Low1 and Up2 have a significant
10 235.933 − 47.456 21.941 26.825 3.155 − 9.980
effect with a range of 3.125 and 2.768; it is followed by the thickness of
11 164.267 − 44.311 22.634 27.095 3.149 − 9.963
12 194.102 − 45.761 23.827 27.541 3.424 − 10.691 Up1 and material grade of Low2 with a range of 0.549 and 0.465.
13 297.416 − 49.467 24.132 27.652 3.243 − 10.219 However, the effects of material grade of Low1 (0.063) and thickness of
14 195.640 − 45.829 23.946 27.585 3.236 − 10.200 Low2 (0.161) are relatively small on EA. In terms of lightweight, it is
15 163.433 − 44.267 24.937 27.937 3.249 − 10.235
also worthy noting that thickness of TWB has a more significant effect
16 209.228 − 46.412 23.420 27.392 3.330 − 10.449
17 140.864 − 42.976 24.348 27.729 3.061 − 9.717 than the material grade.
18 123.784 − 41.853 21.416 26.615 3.337 − 10.467 As for the selection of the optimal level, the level corresponding to a
19 259.704 − 48.290 21.988 26.844 3.061 − 9.717 large S/N ratio is the optimal parameter level. Fig. 7 shows the optimal
20 278.517 − 48.897 21.838 26.784 3.337 − 10.467 parameters level combination for single response. It can be seen that the
21 105.349 − 40.453 23.920 27.575 3.330 − 10.449
optimal parameters combination setting for the mass was
22 207.818 − 46.354 23.083 27.266 3.149 − 9.963
23 204.974 − 46.234 22.448 27.024 3.424 − 10.691 A1B1C1D1E1F1H1G1. Thus, the best combination values for Mass are
24 158.138 − 43.981 24.547 27.800 3.155 − 9.980 material grade of DP590 and minimizing thickness, which is also in line
25 230.499 − 47.253 21.220 26.535 3.236 − 10.200 with the actual situation. The analysis of the results also showed that
26 279.325 − 48.922 23.008 27.238 3.249 − 10.235
the optimum combination of design parameters for Fmax is
27 106.790 − 40.571 25.547 28.147 3.243 − 10.219
A1B1C1D2E3F1G1H3, namely, DP 590–1.2 mm, DP590–0.8 mm,
DP780–1.2 mm and TR780–1.6 mm for up1, low1, up2 and low2.
Meanwhile, A2B2C1D1E3F2G3H2 is the optimum combination of para-
meters for the EA. From the above analysis, it is found that various

Table 4
Average S/N ratios for each level of parameters.
Factors A B C D E F G H

Fmax Level1 − 45.051 − 44.985 − 44.168 − 45.112 − 46.093 − 40.554 − 38.773 − 45.494
Level2 − 45.370 − 45.895 − 45.688 − 45.089 − 46.146 − 45.164 − 41.192 − 45.419
Level3 − 45.662 − 45.203 − 46.227 − 45.882 − 43.844 − 45.858 − 47.233 − 45.170
Δ 0.611 0.910 2.059 0.793 2.302 5.304 8.461 0.324
Rank 7 5 4 6 3 2 1 8
EA Level1 27.092 27.208 27.337 27.478 26.963 24.329 24.608 27.248
Level2 27.375 27.271 27.062 27.013 27.239 27.454 24.443 27.313
Level3 27.246 27.234 27.314 27.222 27.511 27.227 27.211 27.152
Δ 0.282 0.063 0.275 0.465 0.549 3.125 2.768 0.161
Rank 5 8 6 4 3 1 2 7
Mass Level1 − 10.153 − 10.180 − 10.180 − 10.185 − 9.691 − 8.948 − 8.722 − 9.952
Level2 − 10.213 − 10.200 − 10.200 − 10.197 − 10.213 − 10.216 − 9.192 − 10.368
Level3 − 10.213 − 10.200 − 10.200 − 10.198 − 10.675 − 10.422 − 10.675 − 10.260
Δ 0.060 0.020 0.020 0.013 0.984 1.474 1.953 0.416
Rank 5 6 7 8 3 2 1 4

291
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Table 5 G > E > D > F > C > B > A > H, F > E > G > D > C > A
Normalization of S/N ratios (NOR) and Gray relational coefficient (GRC) for > H > B and G > E > D > H > F > C > A > B for case 1, 2, 3
each performance characteristic. and 4, respectively. It is obvious to see that the thickness of Up2 and the
Scenario Fmax EA Mass thickness of Up1 have a large effect on comprehensive crash char-
acteristics, but the material grade of low1 have only a slight effect on
NOR GRC NOR GRC NOR GRC comprehensive crash characteristics in either case. Compared case 2
with other cases, it is worth noting that the thickness of Low1 has a
1 0.945 0.901 0.404 0.456 1.000 1.000
2 0.737 0.655 0.614 0.564 0.449 0.476 larger effect on the EA. With the dynamic change of weight ratios, the
3 0.393 0.452 0.722 0.643 0.000 0.333 thickness of Low2 has an increased effect on the performance of mass.
4 0.602 0.557 0.000 0.333 0.887 0.815 In the aspect of the EA, the material grade of Up1 and the thickness of
5 0.185 0.380 0.429 0.467 0.371 0.443
Low1 have an increased influence, thickness of Up2 and material grade
6 0.486 0.493 0.168 0.375 0.138 0.367
7 0.538 0.520 0.178 0.378 0.797 0.711
of Low2 have a little effect. Meanwhile, it can be known that the mean
8 0.216 0.389 0.408 0.458 0.530 0.515 gray correlations for cases 1, 2, 3 and 4 are 0.521, 0.523, 0.525 and
9 0.307 0.419 0.728 0.648 0.068 0.349 0.515, respectively. Consequently, the initial optimal scenario can be
10 0.223 0.392 0.255 0.402 0.530 0.515 set as A3B1C1D1E3F2G1H1, A3B1C1D1E3F2G1H2, A2B3C1D1E3F2G1H2 and
11 0.572 0.539 0.407 0.458 0.535 0.518
A1B1C1D1E1F2G1H1 for case 1, 2, 3 and 4, respectively.
12 0.411 0.459 0.659 0.594 0.290 0.413
13 0.000 0.333 0.721 0.642 0.449 0.476
14 0.404 0.456 0.683 0.612 0.455 0.479 4.4. Analysis of variance
15 0.577 0.542 0.882 0.809 0.444 0.473
16 0.339 0.431 0.574 0.540 0.371 0.443
In order to determine the optimal parameter combination more
17 0.720 0.641 0.765 0.680 0.618 0.567
18 0.845 0.763 0.137 0.367 0.365 0.441 accurately and eliminate errors from range analysis mentioned above,
19 0.131 0.365 0.266 0.405 0.618 0.567 the ANOVA would be performed by using Eqs. (8)–(12) shown in the
20 0.063 0.348 0.232 0.394 0.365 0.441 Table 6.
21 1.000 1.000 0.678 0.608 0.371 0.443 If the ratio of F exceeds the critical value of a certain significance
22 0.345 0.433 0.503 0.502 0.535 0.518
level, this factor is considered significant at this significance level,
23 0.359 0.438 0.367 0.441 0.290 0.413
24 0.609 0.561 0.805 0.719 0.530 0.515 otherwise this factor is considered to have no significant effect at this
25 0.246 0.399 0.092 0.355 0.455 0.479 level. For example, the ratio of F for Thickness of Up2 at case 4 can be
26 0.060 0.347 0.488 0.494 0.444 0.473 calculated:
27 0.987 0.974 1.000 1.000 0.449 0.476
F0.01 = 99.01 > FG = 42.05 > F0.025 (2, 2) = 39.00 (15)
Therefore, the factor can be considered as having a confidence of
responses mean different optimal parameter combinations. Therefore, it
97.5%, other cases and factors can be analyzed in the similar manner.
is necessary to introduce the gray relational analysis to deal with the
One of the important indicators is contribution rate in the ANOVA.
multi-objective optimization problem.
Taking the case 1 as an example, the contribution rate of factors are
0.04%, 1.36%, 4.31%, 9.66%, 9.57%, 12.01%, 40.75% and 0.94% for
4.3. Analysis of gray correlation A, B, C, D, E, F, G and H, respectively. In addition, since the con-
tribution rate of A is too small (C = 0.04), the effect of material grade of
The S/N ratios would be further normalized through Eq. (3) ac- Up1 on comprehensive performance (Fmax, Mass and EA) in case 1 can
cording to its large-the-better characteristics (see Table 5). Then, the be ignored. Combined with the above analysis, the final optimal level
gray relational coefficients are computed by Eq. (6) and those results combination with G1F2D1E3C1B1H1A3 can be obtained, i.e., thicknesses
are also summed in Table 5. In the cases of uncertain weight ratio of of Up1, Low1, Up2 and Low2 are 1.2 mm, 1.0 mm, 1.2 mm and 1.2 mm,
responses, different situations can be discussed with four cases herein: material grades of Up1, Low1, Up2 and Low2 are DP590, DP590,
Case 1- w(Fmax):w(EA):w(Mass)= 1/3:1/3:1/3; Case 2- w(Fmax):w(EA):w DP780 and DP780. Similarly, the final optimal level combination for
(Mass)= 0.6:0.2:0.2; Case 3- w(Fmax):w(EA):w(Mass)= 0.2:0.6:0.2; Case case2, 3 and 4 are G1E3F2D1C1B1A3H2, G1F2E3D1C1A2H2B3 and
4- w(Fmax):w(EA):w(Mass)= 0.2:0.2:0.6. G1E1F2D1C1A1B1H1.
According to the Eq. (7), the gray correlations of all cases are In addition, Fig. 10 gives a more intuitive view of the overall con-
computed and displayed in Fig. 8. It is obvious that when the gray tribution rates of each factor to multi-response in four cases in order to
correlation increases, the specific energy absorption also increases fully understand the sensitivity of each parameter. It is observed that
while the peak force decreases for every case from the overall trend. the thickness of Up2 is the most significant factor whose contribution is
Therefore, it is feasible to increase the crashworthiness of the tailor- 40.75%, 36.04%, 26.67% and 44.53% respectively for four cases. The
weld structures by increasing the gray correlation. However, it is no- secondary influencing factors are material grade of Low2, thickness of
ticed that the fluctuations of the relationships are various with different Up1 and thickness of Low1. Furthermore, with the dynamic change of
response weight ratios. In cases 1 and 2, the relationships between the weight ratios for the three responses, it can be found that the thickness
specific energy absorption and the gray correlation are less fluctuating of Up1 is the main effects variable on Fmax, which mainly attribute to
and has stronger consistency compared to other cases. Meanwhile, the reason that Up1 is the direct contact area for crashing. The impact
there has a good fit for the peak force in case 4. In addition, the re- of these three variables mentioned above on EA increase, but the in-
lationships between responses and the gray correlation are the most fluence of thickness of Up1 on it decreases. Hence, the properties of the
volatile and unstable in case 3. Therefore, the accuracy of prediction front end of TWB have a great effect on the EA. In addition, the
will be greatly affected by the allocation of different response weight thickness of Up2 with a large proportion has a more significant effect on
ratios. mass. It is precisely because the lightweight is considered, so no matter
In order to obtain the global optimization parameters, the average in any case, it is the main influencing factor.
gray correlations of the different levels for each influencing factor are
calculated (see Fig. 9). The larger the range (Δ) of gray correlation is, 4.5. Analysis of optimized results
the more obvious effect of this factor is. Therefore, it can be seen from
Fig. 9 that the significances of each factor by order on the compre- The final optimization results and corresponding parameter level
hensive crash performance are: G > E > D > F > C > B > H > A, are shown in Table 7. There are different optimal parameter

292
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Fig. 8. Relationships between Gray correlation and responses: (a) case 1; (b) case 2; (c) case 3; (d) case 4.

0.58 0.60
Case2
Mean Gray Correlation
Mean Gray Correlation

0.56 Case1 0.58

0.56
0.54
0.54
0.52 0.521 0.52 0.523
0.50 0.50
0.48
0.48
0.46
0.46
0.44
0.44 0.42
A1A2A3B1B2B3C1C2C3D1D2D3E1E2E3F1F2F3G1G2G3H1H2H3-- A1A2A3B1B2B3C1C2C3D1D2D3E1E2E3F1F2F3G1G2G3H1H2H3-- --

Parameter Level Parameter Level

(a) Cas e 1 ( b) Case 2

0.60 0.56
Mean Gray Correlation
Mean Gray Correlation

Case3 Case4
0.58
0.54
0.56
0.52
0.54 0.515
0.52 0.525 0.50
0.50 0.48
0.48
0.46
0.46
0.44
0.44
A1A2A3B1B2B3C1C2C3D1D2D3E1E2E3F1F2F3G1G2G3H1H2H3-- -- 0.42
A1A2A3B1B2B3C1C2C3D1D2D3E1E2E3F1F2F3G1G2G3H1H2H3-- --
Parameter Level Parameter Level

(c) Cas e 3 ( d) Case 4

Fig. 9. Mean gray correlation for each level of parameters.

293
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Table 6
The ANOVA table for the material/thickness design parameters.
Fac. A B C D E F G H Error Total

Case 1: w(Fmax):w(EA):w(Mass)= 1/3:1/3:1/3

D 2 2 2 2 2 2 2 2 10 26
SS 0.00012 0.00415 0.01316 0.02952 0.02922 0.03669 0.12448 0.00286 0.00012 0.30543
MS – 0.00207 0.00658 0.01476 0.01461 0.01834 0.06224 0.00143 0.00001 –
F – 34.41 109.27 245.05 242.60 304.55 1033.37 23.76 – –
C/% 0.04 1.36 4.31 9.66 9.57 12.01 40.75 0.94 – –
Case 2: w(Fmax):w(EA):w(Mass)= 0.6:0.2:0.2
D 2 2 2 2 2 2 2 2 10 26
SS 0.00132 0.01650 0.02460 0.03387 0.06409 0.04489 0.17306 0.00085 0.00085 0.48016
MS 0.00066 0.00825 0.01230 0.01694 0.03205 0.02245 0.08653 – 0.00009 –
F 1.56 19.50 29.08 40.04 75.77 53.07 204.59 – – –
C/% 0.27 3.44 5.12 7.05 13.35 9.35 36.04 0.18 – –
Case 3: w(Fmax):w(EA):w(Mass)= 0.2:0.6:0.2
D 2 2 2 2 2 2 2 2 10 26
SS 0.00532 0.00147 0.01121 0.05343 0.06577 0.06574 0.08797 0.00507 0.00147 0.33485
MS 0.00266 – 0.00560 0.02671 0.03289 0.03287 0.04399 0.00253 0.00001 –
F 3.62 – 7.63 36.35 44.75 44.73 59.86 3.45 – –
C/% 1.59 0.44 3.35 15.96 19.64 19.63 26.27 1.51 – –
Case 4: w(Fmax):w(EA):w(Mass)= 0.:0.2:0.6
D 2 2 2 2 2 2 2 2 10 26
SS 0.00603 0.00300 0.00866 0.01488 0.02309 0.01817 0.12613 0.01080 0.00300 0.28325
MS 0.00302 – 0.00433 0.00744 0.01155 0.00909 0.06307 0.00540 0.0003 –
F 2.01 – 2.89 4.96 7.70 6.06 42.05 3.60 – –
C/% 2.13 1.06 3.06 5.25 8.15 6.42 44.53 3.81 – –

combinations for different optimization cases. Table 8 and Fig. 11 show
detailed comparisons of the initial and optimized results. Case 1 has the
largest improvement in the four cases, that is, the peak force and mass
were reduced by 29.3% and 2.7%, respectively, while the total ab-
sorbed energy increased by 3.5%. Followed by the case 2, the peak
force is reduced by 25.4% and the specific energy absorption is in-
creased by 1.2%. Although the total absorbed energy was reduced by
9.3% in case 4, its specific energy absorption increased by 5.3%, which
was mainly due to its greatly reduced mass. Fig. 11 shows that the peak
force of the initial structure occurs at the later stage of the collision,
which mainly lies in the unreasonable design of the rear parts for their
material grades or thicknesses. The optimized design can greatly reduce
the peak force while increasing the specific energy absorption.
With regarding to gray correlation, case 2 has the highest gray
correlation with the improvement of 0.267 in the four cases as illu-
strated in Table 7 and has increased by 49.2% shown in Table 8. It is
worth noting that the case where the maximum gray correlation (case
2) appears is inconsistent with the simulation optimal solution (case 1).
The main reason for this is that there are some fluctuations in predicting
optimal responses using the gray correlation as shown in Fig. 8, which

Fig. 10. Contribution rates of design parameters on multi-response.

294
F. Xu et al. Thin-Walled Structures 131 (2018) 286–296

Table 7
Final optimization results and corresponding parameter level.
Initial results Case 1 Case 2 Case 3 Case 4

Design variables A DP590 DP780 DP780 TR590 DP590

B DP590 DP590 DP590 DP780 DP590
C DP780 DP780 DP780 DP780 DP780
D DP780 DP780 DP780 DP780 DP780
E 1.0 1.2 1.2 1.2 0.8
F 1.0 1.0 1.0 1.0 1.0
G 1.4 1.2 1.2 1.2 1.2
H 1.4 1.2 1.4 1.4 1.2
Responses Fmax(kN) 138.417 97.876 100.225 113.937 112.730
EA(kJ) 23.609 24.430 23.739 23.779 21.402
Mass(kg) 3.243 3.155 3.224 3.243 2.793
Gray Correlation Simulations 0.543 0.737 0.810 0.741 0.739
Improvement of gray correlation 0.194 0.267 0.198 0.196

Table 8 10 ms 20 ms 30 ms 40 ms
Percentage increase of final optimization results.
Fmax(kN) EA(kJ) Mass(kg) SEA(kJ/kg) Gray correlation Initiaal
Case 1 29.3% 3.5% 2.7% 6.4% 35.7%
Case 2 25.4% 0.6% 0.6% 1.2% 49.2%
Case 3 17.7% 0.7% 0 7.1% 36.5% Case 1
Case 4 18.6% − 9.3% 15.5% 5.3% 36.1%

is inevitable. However, the crashworthiness of the structure can be Case 2

improved overall. There just has a slight fluctuation in the degree of
improvement.
In addition, the optimal structures deformation mode is compared
with different cases (see Fig. 12). In these four cases, the deformation of
the TWB in the case 4 is the largest, which mainly attribute to its small
average thickness and low average material strength. However, the
deformation in the case 3 is the smallest due to its negative deformation
mode. It can be seen that the TWB begins to deform from the middle
and then continues to the front end. This is mainly due to the sudden
thickness change in the middle, resulting in a great deal of stress con-
centrated. This deformation mode does not meet the requirements of
progressive deformation characteristics.
5. Conclusion remarks
In this paper, the discrete optimization design of TWBs structures
with top-hat thin-walled section based on Taguchi-based gray relational
analysis was presented. Some key findings conclusions were summar-
ized:
Case 3
Case 4
Fig. 12. Deformation modes of optimal structures with four cases.
(1) Multi-objective discrete optimization design was successfully de-
veloped through the gray correlation analysis of the S/N ratios. The
method with only 27 iterations was a low computing cost or cost-
effective discrete optimization deign. Therefore, it was more sui-
table to be applied into complex optimization problems.
(2) It was feasible to optimize by the gray correlation. The greater the
gray correlation means the better the system response character-
istic. The gray correlation of case 1 with the best crashworthiness
was 0.194, which was obviously greater than the original value.
Therefore, the comprehensive performance of the structure could

140 Initiaal result 4.0

Case1
120 3.5 Case2
SEA (kJ/Kg)

3.0 Case3
100
F (kN)

2.5 Case4
80 2.0
60 1.5
Casee1 1.0
40
Casee2
20 0.5
Casee3 Iniitial resultt
Casee4 0.0
0
0 10
0 20 30 40 50 0 50
0 100 150 200 25 50
Time (m
ms) D
Displacemen
nt (mm)
(a) (b)
Fig. 11. Comparisons of responses for initial and optimal results: (a) crashing force; (b) SEA.

295
F. Xu et al.
be significantly increased by the Taguchi-based gray relational
analysis method.
(3) According to influencing factors analysis, it was concluded that the
thickness of Up2 had a significant effect on the crashworthiness
characteristics, which was mainly attributed to its effects on mass.
The thickness of Up1 was the main effects variable for the peak
crashing force. The material grade of Low2, thickness of Up1 and
thickness of Low1 all had a huge influence on the total absorbed
energy.
(4) The optimized results showed that case 1 had the largest im-
provement in the four cases; the peak crashing force and mass were
reduced by 29.3% and 2.7%, respectively, while the total absorbed
energy increased up to 24.430 kJ and by 3.5%. It significantly im-
proved the crashworthiness of TWBs structures.
The obtained remarks can give some useful guidance for light-
weighting and crashworthiness analysis of TWBs which are two im-
portant objectives of car body design. It would be meaningful that
further more comprehensive studies are needed to optimize perfor-
mance of more components with more discrete variables.
Acknowledgments
The support of this work by the National Natural Science
Foundation of China (51605353, U1564202) is greatly appreciated. The
work was also supported by the China Postdoctoral Science Foundation
(2018T110812), 111 Project (B17034) and the Fundamental Research
Funds for the Central Universities (2018-JL-024).
References
[1] N. Peixinho, N. Jones, A. Pinho, Experimental and numerical study in axial crushing
of thin walled sections made of high-strength steels, J. Phys. IV 110 (2003)
717–722.
[2] Z. Gronostajski, P. Kaczynski, S. Polak, B. Bartczak, Energy absorption of thin-
walled profiles made of AZ31 magnesium alloy, Thin-Walled Struct. 122 (2018)
491–500.
[3] Kiani Morteza, Gandikota Imtiaz, Rais-Rohani Masoud, Motoyama Keiichi, Design
of lightweight magnesium car body structure under crash and vibration constraints,
J. Magnes. Alloy. 2 (2) (2014) 99–108.
[4] Priem Cyril, Othman Ramzi, Rozycki Patrick, Guillon Damien, Experimental in-
vestigation of the crash energy absorption of 2.5D-braided thermoplastic composite
tubes, Compos. Struct. 116 (2014) 814–826.
[5] M. Kathiresan, K. Manisekar, V. Manikandan, Crashworthiness analysis of glass
fibre/epoxy laminated thin walled composite conical frusta under axial, Compos.
Struct. 108 (2014) 584–599.
[6] F.X. Xu, C. Wang, Dynamic axial crashing of tailor-welded blanks (TWBs) thin-
walled structures with top-hat shaped section, Adv. Eng. Softw. 96 (2016) 70–82.
[7] Hasan Gedikli, Numerical investigation of axial crushing behavior of a tailor welded
tube, Mater. Des. 44 (2013) 587–595.
[8] Fengxiang Xu, Guangyong Sun, Guangyao Li, Qing Li, Experimental study on
crashworthiness of tailor-welded blank (TWB) thin-walled high-strength steel (HSS)
tubular structures, Thin-walled Struct. 74 (2014) 12–27.
[9] Ying Liang, Zhao Xi, Dai Minghua, Zhang Sizhu, Hu Ping, Crashworthiness design of
quenched boron steel thin-walled structures with functionally graded strength, Int.
J. Impact Eng. 95 (2016) 72–88.
[10] Zhang Yong, Xu Xiang, Liu Shuran, Chen Tengteng, Hu Zhongwei, Crashworthiness
design for bi-graded composite circular structures, Constr. Build. Mater. 168 (2018)
633–649.
[11] F. Xu, Enhancing material efficiency of energy absorbers through graded thickness
296
Thin-Walled Structures 131 (2018) 286–296
structures, Thin-Walled Struct. 97 (2015) 250–265.
[12] Erdin Muhammed Emin, Baykasoglu Cengiz, Cetin Merve Tunay, Quasi-static axial
crushing behavior of thin-walled circular aluminum tubes with functionally graded
thickness, Procedia Eng. 149 (2016) 559–565.
[13] F.X. Xu, X.J. Wan, Y.S. Chen, Design optimization of thin-walled circular tubular
structures with graded thickness under later impact loading, Int. J. Automot.
Technol. 18 (3) (2017) 439–449.
[14] Xu Xiang, Zhang Yong, Wang Jin, Wang Chun H, Jiang Feng, Crashworthiness
design of novel hierarchical hexagonal columns, Compos. Struct. 194 (2018) 36–48.
[15] B. Azimi Mohammadbagher, Asgari Masoud, A new bi-tubular conical–circular
structure for improving crushing behavior under axial and oblique impacts, Int. J.
Mech. Sci. 105 (2016) 253–265.
[16] Zhang Yong, Xu Xiang, Wang Jin, Wang Chun H, Chen Tengteng, Crushing analysis
for novel bio-inspired hierarchical circular structures subjected to axial load, Int. J.
Mech. Sci. 140 (2018) 407–431.
[17] Jin Wang, Yong Zhang, Ning He, Chun H. Wang, Crashworthiness behavior of Koch
fractal structures, Mater. Des. 144 (2018) 229–244.
[18] Aulig Nikola, Nutwell Emily, Menzel Stefan, Detwiler Duane, Preference-based to-
pology optimization for vehicle concept design with concurrent static and crash
load cases, Struct. Multidiscip. Optim. 57 (1) (2018) 251–266.
[19] P. Hajela, E. Lee, Topological optimization of rotorcraft subfloor structures for
crashworthiness considerations, Comput. Struct. 64 (1) (2017) 65–76.
[20] X.G. Gu, G.Y. Sun, G.Y. Li, L.C. Mao, Q. Li, A comparative study on multi-objective
reliable and robust optimization for crashworthiness design of vehicle structures,
Struct. Multidiscip. Optim. 48 (3) (2013) 669–684.
[21] Zhang Yong, Xu Xiang, Sun Guangyong, Lai Xiongming, Li Qing, Nondeterministic
optimization of tapered sandwich column for crashworthiness, Thin-Walled Struct.
122 (2018) 193–207.
[22] F.X. Xu, X. Zhang, H. Zhang, A review on functionally graded structures and ma-
terials for energy absorption, Eng. Struct. 171 (2018) 309–325.
[23] M. Merklein, M. Johannes, M. Lechner, A. Kuppert, A review on tailored blank-
s—production, applications and evaluation, J. Mater. Process. Technol. 214 (2)
(2014) 151–164.
[24] H.Y. Wang, Z.J. Li, Investigation of laser beam welding process of AZ61 of mag-
nesium-based alloy, Acta Metall. Sin. 19 (4) (2006) 287–294.
[25] L.C. Chan, C.H. Cheng, S.M. Chan, T.C. Lee, C.L. Chow, Formability analysis of
tailor-welded blanks of different thickness ratios, J. Manuf. Sci. Eng.-Trans. ASME
127 (4) (2005) 743–751.
[26] F.X. Xu, On modeling strategy of the weld line for tailor-welded structures under
quasi-static and dynamic scenarios, Int. J. Veh. Des. 72 (3) (2016) 230–247.
[27] S. Gaied, J.M. Roelandt, F. Pinard, F. Schmit, M. Balabane, Experimental and nu-
merical assessment of tailor-welded blanks formability, J. Mater. Process. Technol.
209 (1) (2009) 387–395.
[28] X.G. Song, G.Y. Sun, Q. Li, Sensitivity analysis and reliability based design opti-
mization for high-strength steel tailor welded thin-walled structures under crash-
worthiness, Thin-Walled Struct. 109 (2016) 132–142.
[29] F.X. Xu, G.Y. Sun, G.Y. Li, Q. Li, Crashworthiness design of multi-component tailor-
welded blank (TWB) structures, Struct. Multidiscip. Optim. 48 (3) (2013) 653–667.
[30] P. Zhu, Y.L. Shi, K.Z. Zhang, Z.Q. Lin, Optimum design of an automotive inner door
panel with a tailor-welded blank structure, Proc. Inst. Mech. Eng. Part D-J.
Automob. Eng. 54 (D8) (2008) 1337–1348.
[31] Shrestha Sanjay, Manogharan Guha, Optimization of binder jetting using Taguchi
method, JOM 69 (3) (2017) 491–497.
[32] Asiltürk İlhan, Akkuş. Harun, Determining the effect of cutting parameters on
surface roughness in hard turning using the Taguchi method, Measurement 44 (9)
(2011) 1697–1704.
[33] Eskandari Behzad, Davoodi Behnam, Ghorbani Hamid, Multi-objective optimization
of parameters in turning of N-155 iron-nickel-base super alloy using gray relational
analysis, J. Braz. Soc. Mech. Sci. Eng. 40 (4) (2018) 1–12.
[34] Cai Kefang, Wang Dengfeng, Optimizing the design of automotive S-rail using grey
relational analysis coupled with grey entropy measurement to improve crash-
worthiness, Struct. Multidiscip. Optim. 56 (6) (2017) (1539-155).
[35] Mondal Subrata, C.P. Paul, L.M. Kukreja, Bandyopadhyay Asish, Pal Pradip Kumar,
Application of Taguchi-based gray relational analysis for evaluating the optimal
laser cladding parameters for AISI1040 steel plane surface, Int. J. Adv. Manuf.
Technol. 66 (1) (2013) 91–96.
[36] N. Ahmad, S. Kamal, Z.A. Raza, T. Hussain, F. Anwar, Multi-response optimization
in the development of oleo-hydrophobic cotton fabric using Taguchi based grey
relational analysis, Appl. Surf. Sci. 367 (2017) 370–381.