Research Article

Advances in Mechanical Engineering

2017, Vol. 9(9) 1–11
Ó The Author(s) 2017
Optimization of fiber-orientation DOI: 10.1177/1687814017719221
journals.sagepub.com/home/ade
distribution in fiber-reinforced
composite injection molding by
Taguchi, back propagation neural
network, and genetic
algorithm–particle swarm
optimization

Kun Li1,2, Shi-Lin Yan1, Wen-Feng Pan1 and Gang Zhao2

Abstract
Fiber orientation induced by injection molding of short-fiber-reinforced composites causes anisotropy in material proper-
ties and produces warping. Fiber-orientation distribution is very important to research for mold design and quality to
produce sound molded parts. In this study, three kinds of methods are used to solve the optimization problem. Fiber-
orientation distribution is described by fiber-orientation tensor variation. The objective function is a minimum problem
of the fiber-orientation tensor variation. Parameters such as fiber content, fiber aspect ratio, melting temperature, injec-
tion pressure, holding pressure, and filling time are considered as design variables. Based on orthogonal experiment
design, Moldflow software is used in the fiber-reinforced acrylonitrile butadiene styrene composite injection molding.
The effects of process parameters for the plastic part are studied using the signal-to-noise ratio. The most important
design parameter influencing fiber-orientation tensor variation is determined by finite element analysis results based on
the analysis of variance. The optimization model is established on the basis of the back propagation neural network. The
Taguchi, the particle swarm optimization, and genetic algorithm–particle swarm optimization hybrid algorithm are used
to find the minimum fiber-orientation tensor variation value. Results show that the quality index of the fiber-orientation
tensor variation in the part is improved.

Keywords
Optimization, injection molding, Taguchi method, back propagation neural network, genetic algorithm–particle swarm
optimization hybrid algorithm

Date received: 17 February 2017; accepted: 13 June 2017

Academic Editor: David R Salgado

1
Hubei Key Laboratory of Theory and Application of Advanced Materials
Introduction Mechanics, Wuhan University of Technology, Wuhan, China
2
School of Mathematics and Information Sciences, Nanchang Hangkong
In the injection molding product production, in order University, Nanchang, China
to enhance the performance of the products, the most
common method is the addition of short fibers such as Corresponding author:
Shi-Lin Yan, Hubei Key Laboratory of Theory and Application of
glass and carbon fibers in the polymer-based. The exis-
Advanced Materials Mechanics, Wuhan University of Technology, Wuhan
tence of the fiber will ultimately affect the products’ 430070, China.
elastic modulus, tensile strength, thermal conductivity, Email: yanshlin@sohu.com

Creative Commons CC-BY: This article is distributed under the terms of the Creative Commons Attribution 4.0 License
(http://www.creativecommons.org/licenses/by/4.0/) which permits any use, reproduction and distribution of the work without
further permission provided the original work is attributed as specified on the SAGE and Open Access pages (https://us.sagepub.com/en-us/nam/
open-access-at-sage).
2 Advances in Mechanical Engineering
electrical conductivity, optical properties, and so on. In
the injection molding filling, the fiber is suspended in
the melt because of the deformation and effect of flow
field, and the orientation will orient along a certain
direction and change with different positions and dif-
ferent time. Ultimately, the fiber orientation of the
products will present in a complex condition, which
leads to the anisotropy of mechanical properties.1
Therefore, the research of fiber orientation has impor-
tant significance.
In recent years, extensive research on the fiber orien-
tation has been conducted using the numerical simula-
tion. Hung and Shen2 predicted the fiber orientation
based on Jeffery’s model during injection molding using
the finite element method. Aurich and Mennig3 pre-
sented a thorough experimental study of flax fiber
orientation. Comparison of the measured and calcu-
lated modulus in tension shows good agreement. Chang
et al.4 presented an effective methodology for the opti-
mization injection molding process parameters of short-
glass-fiber-reinforced polycarbonate composites. A
computer-aided engineering (CAE) flow simulation
software was used to simulate the injection molding
process and to predict the fiber orientation. On the
basis of the CAE analysis, the optimum process para-
meters are obtained for the thickest shear layer.
As previously mentioned, an extensive research on
the fiber orientation has been conducted using the
numerical simulation method. Advanced methods are
highly demanded to model and optimize the injection
molding process with the purpose of manufacturing
high-quality plastic parts. Many researchers have been
devoted to the optimization research by their experi-
ence to apply a trial-and-error method or the Taguchi
approach. Although the methods are time-consuming
and costly, it is important to evaluate design para-
meters. Kuram et al.5 researched that the glass-fiber-
reinforced nylon 6 was reprocessed in the five
processing cycles using Taguchi method. The result
showed that the method can improve effectively on
mechanical properties. Huang et al.6 presented a hybrid
swarm intelligence with Taguchi-genetic algorithm
(GA) to optimize the inverse kinematics. Wang et al.7
investigated the injection molding process parameters
for manufacturing a brake booster valve body. The
optimal parameters are determined by the Taguchi
method integrating with the CAE to improve the com-
pressive property of the valve body. KC et al.8
researched the optimization of process parameters of
sisal-glass fiber hybrid biocomposite in injection mold-
ing by Taguchi method. The results show the optimiza-
tion parameters meet the dimensional requirements of
automotive parts. Chen et al.9 proposed an effective
methodology integrating the finite element and
Taguchi methods to determine the optimal design
conditions in injection molding for short-fiber-
reinforced polycarbonate composites. The effects of
four main parameters—the filling time, melt tempera-
ture, mold temperature, and injection speed—on the
fiber orientation were investigated and discussed. It is
found that the dominant parameter is the filling time.
In addition, many researchers are committed to
building the relationship model between the objective
function and the influence factors by artificial neural
network, Kriging model, and response surface metho-
dology. Kurtaran et al.10 established a predictive model
in injection molding using the artificial neural network.
It is validated that neural network model is valid for
predictive capability. Zhang et al.11 employed the
response surface methodology to obtain the relation-
ship model between the objective function and indepen-
dent variables. Compared with the reference model, the
performance of the optimized model was significantly
improved. Ozcelik and Erzurumlu12 optimized process
conditions to minimize the warpage of the part in injec-
tion molding using analysis of variance (ANOVA), arti-
ficial neural network, and GA. Chow et al.13 researched
the global optimization of absorption chiller system.
The neural network was used to model the system char-
acteristics. The results confirmed that this methodology
could be effective.
In recent research, the global optimization approach
has become a new trend in optimizing the process para-
meters. GAs and particle swarm optimization (PSO)
have been widely used to solve optimization problems.
To minimize the warpage on the thin shell plastic
part,14 an effective GA was used to obtain the optimum
process parameters. Chen et al.15 proposed the hybrid
GA-PSO to solve the optimization problem in plastic
injection. The final optimal combination of process
parameters was obtained by the hybrid GA-PSO. The
confirmation results show that the proposed method
not only enhances the stability but also reduces the
costs of and time spent in the injection molding.
As Taguchi method is a robust parameter design tech-
nology, the neural network has become a powerful and
practical method for modeling highly complex nonlinear
systems. GA and PSO have the broaden application in
various optimization research fields. Therefore, these
methods have been successfully applied in system model-
ing and process optimization. This study uses the statisti-
cal method for the first time. In order to determine the
best levels for design parameters, the Taguchi design of
experiment (DOE) technique is used to obtain the opti-
mal injection molding conditions. The ANOVA is used
to find out the main control factors of the injection mold-
ing process. The neural network method is used to estab-
lish the relationship model. Finally, PSO and GA-PSO
are, respectively, used in the parameter optimization of
fiber-reinforced composite injection molding.
Li et al.
Relative theory
The back propagation neural network
Back propagation neural network (BPNN) has the
powerful ability of nonlinear interpolation to obtain
the mathematical mapping reflecting the internal law of
the experimental data. Thus, neural network has been
widely used in engineering applications for prediction
and optimization. In this study, BPNN is used to estab-
lish the relationship model between the design para-
meters and the fiber-orientation tensor variation. The
algorithm process of the BPNN is divided into BPNN
training and BPNN for prediction.
Before making prediction with the BPNN, it should
be trained first. By training, the network will own the
ability of associative memory and prediction. The
BPNN is a multilayered architecture made up of one or
more hidden layers placed between the input and out-
put layers.
Based on the input variable x, connection weights
wsj between input layer and hidden layer, hidden layer
threshold a, and the outputs H of hidden layer are cal-
culated as follows
!
X
n between 0 and 1.
Hj = f vsj xs  aj j = 1, 2, . . . , l
i=1
where l is the node number of hidden layer of neural
network and f is the activation function. The activation
function is expressed as follows
1 the later iterations. However, the GA uses selection,
f (x) =
1 + ex
According to the hidden layer outputs H, connection
weight wjt, and threshold value b, the predictive outputs
O of the BPNN are calculated as follows
X
l
Ot = Hj wjt  bt t = 1, 2, . . . , m
j=1
The PSO algorithm
PSO algorithm inspires from population behavior char-
acteristics and is used to solve optimization problems.
Each particle in the algorithm represents a potential
solution of the problem and corresponds with a fitness
which is determined by the fitness function. The parti-
cle’s velocity determines the direction and distance, and
the speed is dynamically adjusted by themselves and
movement experience of other particles, so as to realize
the optimization of individual in the solution space.
We assume that a searching space of D dimension
has the species composed of n particles X = (X1, X2, .,
Xn), where the ith particle is expressed as a D dimen-
sion vector Xi = [xi1, xi2, ., xiD]T, representative of
3
position of the ith particle in D dimension of search
space, which also represents a potential solution of a
problem. According to the objective function, the posi-
tion of each particle Xi corresponding fitness values can
be calculated. The speed of the ith particle can be
expressed as Vi = [Vi1, Vi2, ., ViD]T, and its individual
extremum is Pi = [Pi1, Pi2, ., PiD]T and populations of
the global extremum is Pg = [Pg1, Pg2, ., PgD]T.
In the process of each iteration, the particle updates
the velocity and position by individual extremum and
global extremum updates. Updating formula is as
follows
 
 
Vidk + 1 = vVidk + c1 r1 Pkid  Xidk + c2 r2 Pkgd  Xidk ð4Þ
Xidk + 1 = Xidk + Vidk + 1 ð5Þ
where Vid and Xid represent the current velocity and the
position of the ith particle, w is the inertia weight, and
k and k + 1 refer to the recent and the next iterations.
d = 1, 2, ., D; i = 1, 2, ., n. c1 and c2 represent
‘‘trust’’ parameters indicating how much confidence the
current particle has in itself and how much confidence
it has in the swarm; r1 and r2 are two random numbers
ð1Þ
The GA-PSO algorithm
The PSO algorithm has fast convergence and strong
versatility; however, there are some deficiencies, such as
easy to premature, low precision, and low efficiency of
ð2Þ crossover, and mutation operators in the process of
function optimization to enhance the global optimiza-
tion ability, accelerate the speed of evolution algorithm,
and improve the precision of convergence. But this
mechanism leads to increase in the convergence time of
the algorithm. In order to make the two kinds of algo-
ð3Þ rithms effectively combine, the search mechanism of
the GA is introduced to form the initial particle swarm.
Thus, the GA-PSO optimization algorithm is proposed.
When the particle swarm algorithm stagnates or
embeds local optimal solution, the algorithm is tested
and genetic operation is carried on to the particles,
namely, genetic PSO algorithm. Genetic operation com-
monly uses selection, crossover, and mutation opera-
tions, and in this article, the crossover and mutation
operations are chosen to high particle searchability. In
the process of crossover operation, the position and
speed of the particle that possesses the highest fitness
value keep unchanged. Crossover operation of the velo-
city and position is expressed in the following equation
Vidk + 1 = a1 Vidk + ð1  a1 ÞVik+ 1, d
ð6Þ
Vik++1,1 d = ð1  a1 ÞVidk + a1 Vik+ 1, d
4 Advances in Mechanical Engineering
Xidk + 1 = a1 Xidk + ð1  a1 ÞXik+ 1, d
ð7Þ
Xik++1,1 d = ð1  a1 ÞXidk + a1 Xik+ 1, d
where a1 is a random number between 0 and 1.
After crossover operation is completed, the mutation
is carried on. The articles that have no searchability are
randomly initialized at a certain probability so as to
break the equilibrium state which the particle hovers
near the local optimal solution and improve the popu-
lation diversity.
The GA-PSO algorithm steps can be outlined as
follows:
Step 1. Initialize each particle in the population ran-
domly and set the initial position and speed.
Step 2. Calculate the fitness value of each particle
and initialize the individual optimal value and the
global optimum value of the particle swarm.
Step 3. Update the local and global optimal values
of the individual according to the particle’s fitness
value.
Step 4. Update the particle velocity and position.
Step 5. Determine whether the particles fall into
local optimum. If yes, return to step 3. Otherwise,
enter step 6 to continue the operation.
Step 6. Check whether the end condition is satisfied.
If the current iteration number reaches the maxi-
mum number of times, then stop the iteration and
output the optimal value.
Experimental design
Problem description
Figure 1 shows the simulation model of a plastic part
with mesh geometry. Its length, width, and height are
146, 190, and 10 mm, respectively. The geometry of the
plastic part is discretized using the ‘‘fusion’’ grid. The
global grid side length is 6.45 mm, and the string is
0.1 mm. The number of triangle units is 3520. The
material is acrylonitrile butadiene styrene (ABS) (man-
ufacturers for the UMG ABS Ltd, brand UMG ABS
GF20). The thermoplastic material for conducting the
finite element analysis is selected from the Moldflow
software. Its properties are given in Table 1.
Fiber-orientation tensor variation measurement
Fiber-orientation tensor results show the probability of
fiber orientation along the first main direction (i.e. flow
direction). Specifying the high probability of fiber
orientation along the first main direction is signified by
the range value that is close to 1, whereas the low prob-
ability is close to zero value. The results of the fiber-
orientation tensor along the first principal direction are
useful for research.
Table 1. Properties of the part material.
Solid density (g/cm3) 1.1839
Melt density (g/cm3) 1.0819
Recommended mold temperature (°C) 50
Recommended melt temperature (°C) 240
Material characteristics ABS
Elastic modulus (MPa) 7000
Shear modulus (MPa) 2573
Poisson’s ratio 0.36
ABS: acrylonitrile butadiene styrene.
Figure 1. Base and mesh geometry together.
In the fiber-reinforced composite injection molding
process, fiber orientation significantly influences the
mechanical property of molding products and produces
the products’ current anisotropic properties. It is very
important to research the fiber-orientation distribution.
In this study, the fiber-orientation distribution is
denoted by the fiber-orientation tensor variation (the
difference between the maximum and the minimum
fiber-orientation tensor). When the fiber-orientation
tensor variation is smaller, fiber-orientation distribu-
tion is more homogeneous.
Therefore, the fiber-orientation tensor variation in
the plastic part is selected as the objective function, and
the objective is the minimum problem.
Orthogonal design
In this study, Taguchi’s experiment design is used to
identify the optimal settings of process conditions and
the effect of the process parameters on the fiber-
orientation tensor variation. It is composed of the para-
meters of the fiber-reinforced composite injection
molding process. Each experimental factor is researched
at three levels between the lowest and highest values.
Fiber content ranges from 10% to 30%, fiber aspect
ratio ranges from 10 to 50, melting temperature ranges
from 220°C to 260°C, injection pressure ranges from 60
Li et al. 5

Table 2. Factor and level set.

Level Factor
Fiber Fiber Melting Injection Holding Filling time (s)
content (%) aspect ratio temperature (°C) pressure (MPa) pressure (MPa)
A B C D E F

1 10 10 220 60 40 1
2 20 30 240 70 50 2
3 30 50 260 80 60 3

Table 3. Results of the L27 experiment.

No. A B C D E F Fiber-orientation S/N value (dB)

tensor variation

1 10 10 220 60 40 1 0.5108 5.835

2 10 10 220 60 50 2 0.6114 4.273
3 10 10 220 60 60 3 0.6045 4.372
4 10 30 240 70 40 1 0.5368 5.404
5 10 30 240 70 50 2 0.6077 4.326
6 10 30 240 70 60 3 0.6131 4.249
7 10 50 260 80 40 1 0.5057 5.922
8 10 50 260 80 50 2 0.5143 5.776
9 10 50 260 80 60 3 0.5072 5.896
10 20 10 240 80 40 2 0.5083 5.878
11 20 10 240 80 50 3 0.6102 4.291
12 20 10 240 80 60 1 0.5181 5.712
13 20 30 260 60 40 2 0.6202 4.149
14 20 30 260 60 50 3 0.6294 4.021
15 20 30 260 60 60 1 0.5054 5.927
16 20 50 220 70 40 2 0.6117 4.269
17 20 50 220 70 50 3 0.6083 4.318
18 20 50 220 70 60 1 0.5112 5.828
19 30 10 260 70 40 3 0.6245 4.080
20 30 10 260 70 50 1 0.5061 5.915
21 30 10 260 70 60 2 0.5074 5.893
22 30 30 220 80 40 3 0.507 5.900
23 30 30 220 80 50 1 0.5059 5.919
24 30 30 220 80 60 2 0.6107 4.283
25 30 50 240 60 40 3 0.6152 4.220
26 30 50 240 60 50 1 0.5064 5.910
27 30 50 240 60 60 2 0.6196 4.158

S/N: signal-to-noise ratio.

to 80 MPa, holding pressure ranges from 40 to 60 MPa, Results and discussion

and filling time ranges from 1 to 3 s. Six variable factors
with three levels are studied, as shown in Table 2.
According to the orthogonal experiment design, the
number of experimental trials is 27. The signal-to-noise
ratio (S/N) for each experiment is computed by
! method is used to determine the effect of the process
S 1X n
=  10 3 log y2
N n i=1 i
where n is the number of fiber-orientation tensor varia-
tion data sets, which is equal to 27, and yi is the fiber-
orientation tensor variation value for the ith data set.
Taguchi method
The Taguchi method is used during the design of
parameters for product control and procedure opti-
mization to find the best results. Thus, the Taguchi
parameters on the fiber-orientation tensor variation.
ð8Þ For each of the 27 experimental trials, the finite ele-
ment results for the fiber-orientation tensor variation
and the S/N results are shown in Table 3. The S/N is
an important quality indicator that designers use to
evaluate the effect of process parameters on the
6 Advances in Mechanical Engineering

Table 4. Response table of the S/N ratio.

Level Factors
A B C D E F

1 5.1170 5.1398 4.9997 4.7628 5.074 5.8191

2 4.9326 4.9087 4.9053 4.9212 4.9721 4.7783
3 5.143 5.1441 5.2876 5.5086 5.1464 4.5951
Effect 0.2104 0.2354 0.3823 0.7458 0.1743 1.2240
Rank 5 4 3 2 6 1

S/N: signal-to-noise ratio.

Figure 2. Response diagram of the S/N ratio.

products’ quality index. As the objective of the
research is to minimize the fiber-orientation tensor
variation, the S/N ratio related to the smaller-the-
better quality characteristic is selected when calculat-
ing the S/N ratio. The response table of the S/N ratio
and diagram are given in Table 4 and Figure 2,
respectively. The table is used to determine the opti-
mal control factor settings in minimizing the fiber-
orientation tensor variation. Table 4 presents the S/N
response values at three levels and the effect of each
factor on the S/N ratio. Figure 2 illustrates the effects
of control factors by showing S/N values at three
level settings for each factor.
In Table 4, the best set of combination parameters
can be determined by selecting the level with the highest
value for each factor. The analysis of the results shows
that the optimal process parameter combinations for
the fiber-orientation tensor variation are A3, B3, C3,
D3, E3, and F1. These variations correspond to a fiber
content of 30%, a fiber aspect ratio of 50, a melting
temperature of 260°C, an injection pressure of 80 MPa,
a holding pressure of 60 MPa, and a filling time of 1 s
for the fiber-orientation tensor variation.
ANOVA
The purpose of ANOVA is to determine the statistically
significant parameters influencing the fiber-orientation
tensor variation in the designed experimental study.
The ANOVA results for fiber-orientation tensor varia-
tion are given in Table 5. As shown in Table 5, the
F-values of factors D and F are all greater than
F0.05, 2, 23 = 3.42. Therefore, the injection pressure and
filling time are important parameters. In particular, the
filling time is the most influential factor for fiber-
orientation tensor variation in injection molding, fol-
lowed by the injection pressure and melting tempera-
ture. Chen et al.15 studied optimization of the injection
Li et al. 7

Table 5. ANOVA for fiber-orientation tensor variation.

Factors Degree of Sum of squares Mean Square F-value p value

freedom

A 2 0.00099 0.000497 0.294 0.74965

B 2 0.00137 0.000683 0.404 0.67542
C 2 0.00279 0.001393 0.824 0.45894
D 2 0.01163 0.005816 3.439 0.06095
E 2 0.00059 0.000294 0.174 0.84204
F 2 0.03281 0.016407 9.701 0.00227
Residual 14 0.02368 0.001691
Total 26 0.07386

ANOVA: analysis of variance.

molding process for short-fiber-reinforced composites
and found that the filling time is the most influential
factor. This result is consistent with the results of this
study. Moreover, the fiber content and holding pressure
are not effective factors of minimum fiber-orientation
tensor variation.
BPNN model for fiber-orientation distribution
variation
Based on the above-mentioned results, the fiber content
and the holding pressure do not significantly affect the
fiber-orientation tensor variation. Therefore, factors,
such as fiber aspect ratio, melting temperature, injec-
tion pressure, and filling time, are selected as the design
parameters in establishment of BPNN model for fiber
orientation tensor variation because they potentially
affect the fiber-orientation tensor variation. The objec-
tive is to optimize the process parameters such that the
fiber-orientation tensor variation is minimized. The
neuron number of the input layer of BPNN is deter-
mined by the number of design variables selected, and
the neuron number of the output layer is determined by
the number of the objective indexes. Thus, the number
of input parameters is 4 and the output parameters is 1,
that is, the node number of the input layer is 4, and the
node number of the output layer is 1. The node number
of the hidden layer is determined with train trials, and
the final value obtained is 10. Therefore, the artificial
neural network configuration is 4-10-1. In this study, a
four-layer BPNN model with two hidden layers is used.
The configuration of BPNN is shown in Figure 3.
To obtain the training sample library containing
enough samples, the way of circulation exchange is
used to extend the 27 data sets of orthogonal design in
Table 3 to 81 training sample data sets.
Among the 81 samples in the previous section, 61
samples are used to train the BPNN. Then, the remain-
ing 20 samples are used to test the BPNN performance.
In Figure 4, the difference of the BPNN forecast and
the Moldflow simulation value is shown. The predictive
value of the established BPNN is in good agreement
with the simulation results. The developed BPNN
model gives a satisfactory approximation and can be
used as an efficient predictive tool for fiber-orientation
tensor variation.
Optimization through PSO and GA-PSO
Based on the established relationship model using
BPNN, the PSO and GA-PSO are, respectively, used to
figure out the minimum fiber-orientation tensor varia-
tion. The optimum value of fiber aspect ratio (B), melt-
ing temperature (C), injection pressure (D), and filling
time (F) toward minimum fiber-orientation tensor var-
iation can be efficiently obtained by formulating it as
an optimization problem, which can be stated as
follows
Find : B, C, D, and F ð9aÞ
Minimize: fiber-orientation tensor variation ðB, C, D, and FÞ
ð9bÞ
Subjected to constraints:
ð9cÞ
fiber-orientation tensor variation \0:52
In order to figure out the optimal solution, PSO and
GA-PSO are used to solve the minimum problem. In
this study, a population size of 20, the dimension of the
particle is 2, and generation number of 100 are
employed. The evolution of the generations is shown in
Figure 5.
Based on PSO algorithm, the figure shows that after
50 iterations, the optimal fitness value is 0.4725.
However, based on GA-PSO algorithm, the figure
shows that after 10 iterations, the optimal fitness value
is 0.4307. The corresponding optimized parameter
combinations are, respectively, given in Table 6.
Verification of the optimization result
The confirmation experiment is very important in para-
meter design. The numerical experiments have been
performed based on the optimized process parameters
8 Advances in Mechanical Engineering

Figure 3. Configuration of the BPNN model.

Figure 4. Comparison of fiber-orientation tensor variation with (a) FE analyses and (b) neural network for training data set.

by GA and GA-PSO method. The calculated fiber- and the minimum value is 0.5289 and 0.5051. The mini-
orientation tensor distribution is shown in Figure 6(c) mum value of the fiber-orientation tensor variation is
and (d). The difference between the maximum value 0.5289 and 0.5051. Therefore, the GA-PSO enhanced
Li et al. 9

Figure 5. Fitness curve with (a) PSO algorithm and (b) GA-PSO algorithm.

Table 6. Result verification.

Parameter Fiber aspect Melting Injection Filling time (s) Fiber-orientation

ratio temperature (°C) pressure (MPa) tensor variation

Recommended parameter 25 240 65 1 0.5462

Taguchi method 50 260 80 1 0.5368
PSO optimization 30 250 60 1 0.5289
GA-PSO optimization 18 224 65 1 0.5051

GA: genetic algorithm; PSO: particle swarm optimization.

the global optimization ability of PSO algorithm to Conclusion

speed up the evolution of the algorithm, and the con-
In this study, an efficient optimization methodology
vergence precision is improved.
using BPNN and GA-PSO is developed to optimize the
When the minimum fiber-orientation tensor varia-
fiber-orientation distribution in short-fiber-reinforced
tion of the part is considered, the fiber-orientation ten-
ABS composite injection molding. A plastic part under
sor variation (0.5462 before the optimization) is
the design parameters is used as the example. On the
reduced to 0.5051. This value is larger than the BPNN-
basis of the Taguchi experiment design and Moldflow
predicted result under the optimization parameters but
numerical simulation, the S/N ratio is used to deter-
smaller than all the samples. The fiber-orientation ten-
mine the optimal set of process parameters. ANOVA is
sor distribution before optimization is shown in
used to determine the significance of each factor in the
Figure 6(a). The corresponding parameters and result
experimental study. This method determines the signifi-
are shown in Table 6. At the same time, the confirma-
cance of each factor to minimize the fiber-orientation
tion experiment is conducted based on the Taguchi
tensor variation. The dominant process parameter of
method. The result demonstrates that the fiber-
injection molding is the filling time, followed by the
orientation tensor variation is 0.5368 based on the opti-
injection pressure and melting temperature. An efficient
mum process parameters obtained by Taguchi method,
predictive model for fiber-orientation tensor variation
which is smaller than the value before optimization;
is established based on the most important process
however, the variation is larger than the value obtained
parameters (fiber aspect ratio, melting temperature,
by GA and GA-PSO method. The fiber-orientation
injection pressure, and filling time) using BPNN to
tensor distribution is shown in Figure 6(b). The result
reduce the computational cost of the optimization pro-
shows that the fiber-orientation tensor distribution of
cess. GA-PSO reduces the fiber-orientation tensor var-
the fiber-reinforced composite injection molded part
iation, and satisfactory results are achieved. The results
has been improved under the optimized process condi-
showed that fiber aspect ratio of 18, melting tempera-
tions using GA-PSO.
ture of 224°C, injection pressure of 65 MPa, and filling
10
Figure 6. Fiber-orientation tensor distribution: (a) recommended parameter, (b) Taguchi method, (c) PSO optimization, and (d)
GA-PSO optimization.
time of 1 s provide the minimum fiber-orientation ten-
sor variation. Furthermore, this study indicates that
the optimization methodology can also improve the
other parts manufactured in fiber-reinforced composite
injection molding.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with
respect to the research, authorship, and/or publication of this
article.
Funding
The author(s) disclosed receipt of the following financial sup-
port for the research, authorship, and/or publication of this
article: This work was supported by the Fundamental
Research Funds for the Central Universities (WUT:
2017III020) and Science and Technology Research Project of
Jiangxi Provincial Department of Education (No.
GJJ160711).
References
1. Yang YK. Optimization of injection molding process for
mechanical properties of short glass fiber and polytetra-
fluoroethylene reinforced polycarbonate composites: a
case study. J Reinf Plast Comp 2006; 25: 1279–1290.
Advances in Mechanical Engineering
2. Hung CF and Shen YK. Numerical simulation of fiber
orientation in injection mold filling. Int Commun Heat
Mass 1995; 22: 791–802.
3. Aurich T and Mennig G. Flow-induced fiber orientation
in injection molded fit fiber reinforced polypropylene.
Polym Composite 2001; 22: 680–689.
4. Chang SH, Hwang JR and Doong JL. Manufacturing
process optimization of short glass fiber reinforced poly-
carbonate composites in injection molding. J Reinf Plast
Comp 2000; 19: 301–321.
5. Kuram E, Tasci E, Altan AI, et al. Investigating the
effects of recycling number and injection parameters on
the mechanical properties of glass-fibre reinforced nylon 6
using Taguchi method. Mater Design 2013; 49: 139–150.
6. Huang HC, Xu SD and Wu CH. A hybrid swarm intelli-
gence of artificial immune system tuned with Taguchi-
genetic algorithm and its field-programmable gate array
realization to optimal inverse kinematics for an articu-
lated industrial robotic manipulator. Adv Mech Eng
2016; 8: 1–10.
7. Wang YQ, Kim JG and Song JI. Optimization of plastic
injection molding process parameters for manufacturing
a brake booster valve body. Mater Design 2014; 56:
313–317.
8. KC B, Faruk O, Agnelli JAM, et al. Sisal-glass fiber
hybrid biocomposite: optimization of injection molding
parameters using Taguchi method for reducing shrink-
age. Compos Part A 2016; 83: 152–159.
Li et al.
9. Chen CS, Chen TJ, Chen SC, et al. Optimization of the
injection molding process for short-fiber-reinforced com-
posites. Mech Compos Mater 2011; 47: 359–368.
10. Kurtaran H, Ozcelik B and Erzurumlu T. Warpage opti-
mization of a bus ceiling lamp base using neural network
model and genetic algorithm. J Mater Process Tech 2005;
169: 314–319.
11. Zhang X, Yu S, Gong Y, et al. Optimization design for
turbodrill blades based on response surface method. Adv
Mech Eng 2016; 8: 1–12.
12. Ozcelik B and Erzurumlu T. Comparison of the warpage
optimization in the plastic injection molding using
11
ANOVA, neural network model and genetic algorithm. J
Mater Process Tech 2006; 171: 437–445.
13. Chow TT, Zhang GQ, Lin Z, et al. Global optimization
of absorption chiller system by genetic algorithm and
neural network. Energ Buildings 2002; 34: 103–109.
14. Kurtaran H and Erzurumlu T. Efficient warpage optimi-
zation of thin shell plastic parts using response surface
methodology and genetic algorithm. Int J Adv Manuf
Tech 2006; 27: 468–472.
15. Chen WC, Nguyen MH, Chiu WH, et al. Optimization
of the plastic injection molding process using the Taguchi
method, RSM, and hybrid GA-PSO. Int J Adv Manuf
Tech 2016; 83: 1–14.