Int J Adv Manuf Technol

DOI 10.1007/s00170-015-7683-0

ORIGINAL ARTICLE

Optimization of the plastic injection molding process using

the Taguchi method, RSM, and hybrid GA-PSO
Wen-Chin Chen 1 & Manh-Hung Nguyen 2 & Wen-Hsin Chiu 3 & Te-Ning Chen 1 &
Pei-Hao Tai 1

Received: 18 March 2015 / Accepted: 3 August 2015

# Springer-Verlag London 2015

Abstract This paper proposes a systematic optimization 1 Introduction

model of process parameters in plastic injection molding
(PIM). Firstly, the Taguchi method is employed for experi- Over time, plastic injection molding has become one of the
mentation and data analysis, in which the quality characteris- main methods for producing plastic parts. Many engineers and
tics for the plastic injection product are length and warpage. others with experience in the field consider plastic injection
The control factors for the process are melt temperature, in- molding to be a simple manufacturing process which requires
jection velocity, packing pressure, packing time, and cooling no adjustments. However, since plastic injection molding is
time. Moreover, the signal-to-noise (S/N) ratio and analysis of actually a complex manufacturing process, and product qual-
variance (ANOVA) are used to obtain a combination of pa- ity depends on the choice of materials, mold design, and pro-
rameter settings. Experimental data are set for the response cess parameters, finding the optimal set of parameters is, in
surface methodology (RSM) in order to analyze and create fact, of utmost importance. As such, there has been a great
two quality predictors and two S/N ratio predictors. The two need to seek quick and effective solutions, especially for glob-
quality predictors are associated with genetic algorithms (GA) al optimal process parameter settings, and the determination of
to search for an optimal combination of process parameters the optimal parameter setting is now recognized as a crucial
that meets multiple-objective quality characteristics. Finally, step if the quality of plastic injection molding products is to be
four predictors are combined with the hybrid GA-PSO to find improved.
the final optimal combination of process parameters. The con- Nonetheless, in order to reach better process parameter
firmation results show that the proposed model not only en- settings for individual products, engineers must often rely on
hances the stability in the injection molding process, including their general know-how to apply a trial-and-error method, or
the quality in length and warpage, but also reduces the costs of they use the Taguchi approach, both of which are time-
and time spent in the PIM process. consuming as numerous experiments are needed in order to
achieve the suitable parameter combinations. Additionally, the
Taguchi method in the extant literature requires the utilization
Keywords PIM . Taguchi method . ANOVA . RSM . GA .
of the signal-to-noise (S/N) ratio to seek the initial process
Hybrid GA-PSO
parameter settings, and considerable effort is needed to seek
them in this way. Specifically, Wu and Chen [1] used the
* Pei-Hao Tai Taguchi L9(34) orthogonal array table in injection molding
[email protected] and injection extrusion molding to investigate the optimal
process parameters, where the main factors were melt temper-
1
Department of Industrial Management, Chung Hua University, 707 ature, injection velocity, mold temperature, and packing pres-
Wu Fu Rd., Sec. 2, Hsinchu 30012, Taiwan sure. Teng and Xu [2], in an attempt to improve the whole-cell
2
Ph.D. Program of Technology Management, Chung Hua University, lipase production in submerged fermentation by Rhizopus
707 Wu Fu Rd., Sec. 2, Hsinchu 30012, Taiwan chinensis, combined the Taguchi method and response surface
3
Department of Physical Education, National Hsinchu University of methodology. In a noteworthy study, Aggarwal et al. [3] con-
Education, 521 Nanda Rd., East Dist., Hsinchu 30014, Taiwan ducted experiments on the setting of process parameters using
 Int J Adv Manuf Technol

Identify experimental factors

and quality characteristics

Conduct PIM experiments through

Taguchi orthogonal array table

Implement ANOVA analysis and

determine the significant factors
Fig. 3 Finished experimental product
stage I
Develop S/N and quality
characteristics predictors via RSM [5] conducted experiments using the Taguchi method and
Multi-objective regression models ANOVA in an attempt to reduce injection molding shrinkage.
two-stage stage II Zhai and Xie [6], using the Moldflow software and the
optimization Taguchi method in injection molding, carried out optimization
Search for the optimal multi-
quality parameter settings using experiments to seek the optimal convoluted positions. In line
S/N and quality predictors with with this, Ng et al. [7] employed the Taguchi method and
GA and hybrid GA-PSO
ANOVA in their attempt to optimize the injection molding
process parameters. Moreover, Öktem [8] used the Moldflow
Meet optimal multi-quality No
software with Taguchi experiments and ANOVA to identify
parameter settings? the more important process parameters. Further elaborating,
Wang et al. [9] used the Taguchi method to investigate the
Yes influential relationships between process parameters and
Proceed to run confirmation warpage, where the main parameters were melt temperature,
tests for two optimal solutions injection time, packing pressure, packing time, and cooling
time. However, the results of such Taguchi experiments re-
vealed only discrete combinations rather than the optimal pro-
Achieve the quality No
requirements cess parameters. In practice, it has been noted that inappropri-
ate process parameter combinations can lead to product de-
Yes
fects, the process instability phenomenon, and so on.
Obtain the optimal process In order to solve these problems, many scholars have used
parameter settings
computer-aided engineering (CAE) simulation software (e.g.,
Fig. 1 Flowchart of the proposed method Moldex3D, Moldflow, C-MOLD) to carry out injection mold-
ing simulation experiments, or they have used a combination
the Taguchi method, experimental design, and response sur- of simulation software, experimental design, and algorithms
face methodology and found that response surface methodol- to conduct optimizations. For instance, Shi et al. [10] applied
ogy had obvious advantages over the Taguchi method. Simi- the Moldflow software to perform simulations in which ex-
larly, Kurt et al. [4] employed Taguchi orthogonal array ex- perimental data were processed using the backpropagation
periments, S/N ratio values, analyses of variance (ANOVA), neural network (BPNN) combined with genetic algorithms
and regression models to seek the optimal parameters. Altan (GA). In their study, mold temperature, melt temperature, in-
jection time, and injection pressure were taken as control fac-
tors, while maximum shear stress was a key quality

Fig. 2 Experimental mold Fig. 4 Length specification measurements

Int J Adv Manuf Technol

which the significant parameters were melt temperature, injec-

tion velocity, injection pressure, packing pressure, and pack-
ing time. However, the CAE simulations neglected the inter-
nal and external noise interference of the control factors and
the actual injection error still existed.
Therefore, numerous studies utilized actual injection mold-
ing experiments to collect data and establish multiobjective
optimization models for the plastic injection molding process.
Chen et al. [15] constructed a parameter optimization system
integrating Moldflow analysis, the Taguchi method, ANOVA,
BPNN, GA, and the Davidon–Fletcher–Powell (DFP) method
to determine the optimal process parameter settings for the
multiple-input and single-output (MISO) plastic injection
molding process, in terms of the weight of the push-button
housing and runner. Kurtaran and Erzurumlu [16] further in-
Fig. 5 Warpage specification measurements tegrated DOE, RSM, and GA to obtain the optimal thin plastic
warpage parameters and utilized ANOVA to find the more
characteristic. Yin et al. [11] also presented a hybrid optimi- important factors, as well as combining the RSM and GA to
zation method combining BPNN and GA to optimize the pro- find the optimal process parameters. Park and Dang [17] com-
cess parameters of a plastic cover on the basis of Moldflow bined DOE and RSM together with ANOVA and numerical
analysis. Ozcelik and Erzurumlu [12] applied the Moldflow simulations to conduct optimization experiments. Similarly,
software to perform simulations. Warpage was employed as a Sun et al. [18] utilized simulation software combined with
crucial quality characteristic, and the seven control factors RSM and GA to find the optimal injection molding parame-
chosen for the Taguchi orthogonal array experiments were ters. Of note, Zhao et al. [19] proposed a fast strip analysis
mold temperature, barrel temperature, packing pressure, pack- model combined with particle swarm optimization (PSO)
ing time, cooling time, runner type, and nozzle location. After which effectively reduced the cooling time in the same pack-
this, ANOVA was used to investigate the impacts of control ing pressure. Xu et al. [20] conducted multiobjective optimi-
factors on quality characteristics, and a neural network model zation research, integrating Taguchi’s parameter design meth-
with genetic algorithms was employed to search for the opti- od and neural networks with a multiobjective PSO algorithm
mal parameter combinations. Subsequently, Akbarzadeh and to identify the optimal process conditions.
Sadeghi [13] indicated that the change in size due to shrinkage In addition, Mostafa et al. [21] used RSM methodology
during the plastic injection molding process was an important and an effective simulated annealing (SA) algorithm to iden-
issue. Their study used regression and ANOVA to explore the tify the optimum process parameter values so as to efficiently
relationships between input and output, with the significant minimize the shrinkage and warpage on the fuel filter in the
experimental parameters being melt temperature, injection injection molding process. Tzeng and Chen [22] developed a
pressure, packing pressure, and packing time. Chen et al. hybrid method combining BPNN, GA, and RSM to determine
[14], in their optimization of the process parameters, used the optimal parameter settings of the electrical discharge ma-
design of experiment (DOE), response surface methodology chining (EDM) process using Taguchi orthogonal array exper-
(RSM), and GA. Significant plastic injection molding (PIM) iments of specimens to obtain the BPNN training data;
process parameters were demonstrated by the ANOVA and ANOVA was used to identify significant factors. Subsequent-
DOE screening experiments via the CAE simulations, in ly, Chen et al. [23] developed an approach in a soft computing
paradigm for the process parameter optimization of the
Table 1 Control factors and the standard settings of levels
multiple-input and multiple-output (MIMO) plastic injection
molding process, which integrated Taguchi’s parameter de-
Melt Injection Packing Packing Cooling sign method, BPNN, GA, and engineering optimization con-
temperature velocity pressure time (s) time (s) cepts. Furthermore, Tzeng et al. [24] constructed a hybrid
(°C) (mm/s) (MPa)
method in conjunction with Taguchi orthogonal array experi-
Level 1 249 30 27 0.9 11 ments, ANOVA, RSM, BPNN, and GA to predict the quality
Level 2 252 34 31 1.2 14 characteristics of SGF- and PTFE-reinforced PC composites,
Level 3 255 38 35 1.5 17 such as ultimate strength, flexural strength, and impact resis-
Level 4 258 42 39 1.8 20 tance, and finally generate an optimal parameter setting of the
Level 5 261 46 43 2.1 23 injection molding process under a MIMO consideration. Re-
cently, Huang et al. [25] proposed a hybrid optimization
 Int J Adv Manuf Technol

Table 2 Experimental results for length

No. Y1 Y2 Y3 Y4 Y5 Average Standard deviation S/N ratio Sensitivity

1 170.16 170.12 170.11 170.12 170.10 170.12 0.0228 8.4344 44.6152

2 170.16 170.23 170.22 170.24 170.20 170.21 0.03162 10.7007 44.6197
3 170.36 170.33 170.34 170.33 170.32 170.34 0.01517 15.6661 44.6261
4 170.52 170.51 170.50 170.51 170.52 170.51 0.00837 36.6959 44.6351
5 170.76 170.74 170.73 170.71 170.72 170.73 0.01924 12.6605 44.6463
6 170.32 170.31 170.30 170.31 170.31 170.31 0.00707 14.4189 44.6248
7 170.41 170.42 170.42 170.44 170.43 170.42 0.0114 22.2871 44.6306
8 170.31 170.30 170.30 170.29 170.29 170.30 0.00837 13.8855 44.6242
9 170.16 170.12 170.13 170.14 170.18 170.15 0.02408 8.9999 44.6164
10 170.23 170.19 170.19 170.18 170.20 170.20 0.01924 10.3823 44.6191
11 170.34 170.33 170.32 170.32 170.34 170.33 0.01 15.376 44.6258
12 170.14 170.14 170.13 170.12 170.14 170.13 0.00894 8.7278 44.6158
13 170.37 170.36 170.35 170.37 170.36 170.36 0.00837 17.1865 44.6275
14 170.40 170.36 170.35 170.35 170.36 170.36 0.02074 17.2294 44.6276
15 170.24 170.23 170.20 170.19 170.18 170.21 0.02588 10.6583 44.6196
16 170.31 170.33 170.34 170.33 170.34 170.33 0.01225 15.3685 44.6258
17 170.16 170.12 170.10 170.11 170.12 170.12 0.0228 8.4344 44.6152
18 170.20 170.18 170.20 170.18 170.18 170.19 0.01095 10.1116 44.6186
19 170.41 170.40 170.41 170.38 170.39 170.40 0.01304 19.7576 44.6293
20 170.27 170.26 170.24 170.23 170.22 170.24 0.02074 11.8068 44.6214
21 170.53 170.52 170.52 170.50 170.50 170.51 0.01342 34.2481 44.6352
22 170.39 170.38 170.37 170.39 170.34 170.37 0.02074 17.8765 44.6281
23 170.25 170.24 170.23 170.23 170.21 170.23 0.01483 11.424 44.6208
24 170.05 170.04 170.03 170.00 170.02 170.03 0.01924 6.514 44.6104
25 170.24 170.24 170.23 170.23 170.23 170.23 0.00548 11.5005 44.6209
26 170.44 170.42 170.42 170.41 170.41 170.42 0.01225 21.8376 44.6304
27 170.45 170.43 170.43 170.41 170.41 170.43 0.01673 22.3988 44.6307
28 170.36 170.34 170.32 170.33 170.31 170.33 0.01924 15.4373 44.6259
29 170.26 170.22 170.25 170.22 170.22 170.23 0.01949 11.4791 44.6209
30 170.36 170.37 170.33 170.33 170.34 170.35 0.01817 16.1896 44.6266

approach integrating BPNN with embedded SA into the GA differential evolution (ODE), harmony search (HS), GA, fire-
to improve its local searching ability algorithm and optimize fly swarm optimization (FSO), artificial bee colony (ABC),
the thickness of the blow-molded polypropylene bellows used gravitational search algorithm (GSA), brainstorm optimiza-
in cars. The above approach has shown great potential in the tion algorithm (BSOA), chaotic inertia weight PSO
search for the optimal process parameter settings for the die (CIWPSO), fixed inertia weight PSO (FIWPSO), time-
gap profile needed to achieve the desired thickness distribu- varying acceleration coefficient PSO (TVACPSO), random
tion in the final bellows with a minimum of experiments and inertia weight PSO (RIWPSO), and constricted PSO
so avoid getting trapped at a local optimum in complicated (COPSO). The experimental results confirmed that ELPSO
manufacturing processes. performed well in all the above terms. Ranjania and
In view of current PSO developments, Jordehi [26] devel- Murugesan [27] proposed a PSO-based fuzzy controller pa-
oped a new PSO variant, named enhanced leader PSO rameter optimization to overcome the drawbacks of the con-
(ELPSO), based on a five-stage successive mutation strategy ventional controller suffering from uncertain parameters and
to solve the problem of how to mitigate premature conver- the nonlinear qualities of the quasi-Z source converter, as well
gence and minimize getting trapped in the local optima prob- as computational inefficiency in optimizing the fuzzy control-
lems of conventional PSO (CPSO). The terms of accuracy, ler parameters. The PSO algorithm was exploited to identi-
scalability, and convergence rate were also compared via com- fy the optimal fuzzy parameters for minimizing the ob-
petitive differential evolution (CDE), opposition-based jective (cost) functions and enhancing its feasibility. The
Int J Adv Manuf Technol

Table 3 Experimental results for warpage

No. Y1 Y2 Y3 Y4 Y5 Average Standard deviation S/N ratio Sensitivity

1 0.421 0.417 0.513 0.557 0.489 0.479 0.06031 6.3179 −6.3998

2 0.376 0.269 0.321 0.317 0.346 0.326 0.03954 9.6775 −9.7538
3 0.137 0.152 0.098 0.115 0.112 0.123 0.02149 18.0851 −18.2427
4 0.173 0.143 0.135 0.193 0.132 0.155 0.02665 16.056 −16.2079
5 0.143 0.171 0.152 0.135 0.104 0.141 0.02465 16.8849 −17.0422
6 0.183 0.235 0.149 0.16 0.181 0.182 0.0331 14.6757 −14.8466
7 0.102 0.144 0.147 0.177 0.15 0.144 0.02692 16.6836 −16.8632
8 0.188 0.185 0.179 0.213 0.179 0.189 0.01408 14.4559 −14.4848
9 0.243 0.253 0.267 0.281 0.27 0.263 0.01491 11.5935 −11.6103
10 0.272 0.303 0.27 0.295 0.289 0.286 0.01441 10.8677 −10.881
11 0.193 0.147 0.161 0.162 0.158 0.164 0.01717 15.6453 −15.702
12 0.164 0.234 0.144 0.181 0.228 0.19 0.03954 14.2321 −14.4535
13 0.164 0.208 0.244 0.188 0.205 0.202 0.02936 13.8106 −13.92
14 0.202 0.128 0.228 0.213 0.254 0.205 0.04725 13.5401 −13.8113
15 0.289 0.318 0.288 0.244 0.281 0.284 0.02649 10.896 −10.9412
16 0.186 0.109 0.164 0.137 0.143 0.148 0.02901 16.4423 −16.6401
17 0.356 0.47 0.367 0.332 0.317 0.368 0.06009 8.5596 −8.6968
18 0.376 0.376 0.375 0.32 0.358 0.361 0.02417 8.8304 −8.8538
19 0.155 0.118 0.14 0.136 0.107 0.131 0.01889 17.5522 −17.6594
20 0.31 0.292 0.342 0.249 0.262 0.291 0.03731 10.6513 −10.7364
21 0.173 0.195 0.155 0.153 0.145 0.164 0.02003 15.6284 −15.7055
22 0.078 0.103 0.088 0.105 0.034 0.082 0.02883 21.2553 −21.876
23 0.225 0.334 0.169 0.237 0.364 0.266 0.08087 11.1244 −11.5901
24 0.342 0.335 0.378 0.327 0.335 0.343 0.02006 9.2692 −9.287
25 0.258 0.291 0.242 0.25 0.361 0.28 0.04876 10.9151 −11.0708
26 0.195 0.25 0.146 0.198 0.19 0.196 0.03694 14.0119 −14.1948
27 0.173 0.22 0.243 0.19 0.245 0.214 0.032 13.2878 −13.403
28 0.192 0.156 0.105 0.172 0.161 0.157 0.03229 15.8915 −16.1078
29 0.268 0.402 0.354 0.355 0.418 0.359 0.05842 8.7752 −8.9114
30 0.227 0.157 0.155 0.185 0.242 0.193 0.03988 14.0987 −14.317

experimental results confirmed that the proposed fuzzy best particle does not share the same niche with the global
PSO-based controller was more effective than the con- optimum.
ventional fuzzy and GA-based fuzzy controllers.
While the conventional PSO has often been the cause of
premature convergence, it has the advantage of fast searching 2 Optimization model, experimental work, and data
and converging rates. However, the more the particles of the analysis
swarm move forward to the direction of yielding the best
fitness value, the more the diversity of the whole swarm drops. The aim of this study was to propose a systematic
Thus, particles can easily get trapped in a local optimum if the multiobjective quality optimization model for the injection

Table 4 Parameter combination of the highest S/N ratio for length and warpage

No. Melt temperature Injection velocity Packing pressure Packing time (s) Cooling time (s) Average Standard Highest S/N
(°C) (mm/s) (MPa) deviation ratio

4 249 42 39 1.8 20 170.51 0.00837 36.6959

22 261 34 27 2.1 20 0.082 0.02883 21.2553
 Int J Adv Manuf Technol

Table 5 ANOVA results for quality characteristic of length Table 7 Taguchi experimental optimal parameters

Source DF Sep SS Adj MS F P value Melt Injection Packing Packing Cooling

temperature velocity pressure time (s) time (s)
Melt temperature (°C) 4 0.050494 0.012624 5.24 0.069 (°C) (mm/s) (MPa)
Injection velocity (mm/s) 4 0.017130 0.004282 1.78 0.295
Length 249 42 39 1.8 20
Packing pressure (MPa) 4 0.048133 0.012033 5.00 0.074
Warpage 261 34 27 2.1 20
Packing time (s) 4 0.211573 0.052893 21.98 0.006
Taguchi optimal 255 38 33 1.95 20
Cooling time (s) 4 0.221355 0.055339 22.99 0.005
parameters
Error 4 0.009627 0.002407
Total 24 0.558312
R2 =98.28 % R2(adj)=89.65 %
range setting, short injection shots or overflow shot situations
can easily occur. To prevent this phenomenon, practical refer-
molding process, the intention being to determine the values ences from field engineers and experimental results were
consistent with multiobjective product qualities, which would employed to determine the experimental control factors’ stan-
result in less quality variation and the ability to maintain prod- dard range.
uct quality within an acceptable range. As such, product qual- In the next step, the Taguchi method was utilized for
ity would be stabilized. Accordingly, this study integrated the the experiments and data analysis, followed by the S/N
Taguchi method, ANOVA, response surface methodology, ge- ratio method and ANOVA, which were used to identify
netic algorithms, and the hybrid (GA-PSO) algorithm with the most significant process parameters for the initial
related technologies to carry out the optimization, whereby optimal combinations. The experimental data formed
the hybrid GA-PSO algorithm employed the alteration of the the basis for the RSM analysis via the multiquality re-
particles’ position via crossover and mutation of GA to retain gression models (S/N ratio and quality characteristic
the best particles and abandon the worst particles, which predictors) and combined with GA and hybrid GA-
would prevent the particles of the swarm from getting trapped PSO to determine the optimal process parameter combi-
in a local optimum so a global optimum solution could be nations in compliance with multiobjective product qual-
obtained. ity. In this way, the achieved process parameter combi-
First, the experimental factors, as well as the quality char- nations were expected not only to enhance the stability
acteristics, were determined and the Taguchi orthogonal array of the injection molding process and ensure that the
table was arranged. Since the injection-molded pieces used in product length met the specifications but also to effec-
this study were pieces of plastic, any slight warpage could tively reduce product warpage. The flowchart of the
result in unusable or defective products. Therefore, the proposed method is shown in Fig. 1.
multiobjective quality optimization model in this study was
in accordance with manufacturing specifications by selecting
2.1 Experimental equipment
length and warpage, respectively, in terms of product appear-
ance and specifications, as well as vendor requirements, as the
In this study, a Victor Taichung VS-80 injection mold-
two major quality characteristics. Experimental control factors
ing machine was used for the injection molding
comprised melt temperature, injection velocity, packing pres-
sure, packing time, and cooling time. If the process parameter
values in the control factors’ standard settings exceed the Table 8 ANOVA for sensitivity of length

Table 6 ANOVA results for quality characteristic of warpage Source DF Sep SS Adj MS F P value

Source DF Sep SS Adj MS F P value Melt temperature 4 0.0001311 0.0000328 5.22 0.069
(°C)
Melt temperature (°C) 4 0.62466 0.15616 5.53 0.063 Injection velocity 4 0.0000445 0.0000111 1.75 0.297
(mm/s)
Injection velocity (mm/s) 4 0.21908 0.05477 1.94 0.569
Packing pressure 4 0.0001248 0.0000312 4.97 0.075
Packing pressure (MPa) 4 0.57593 0.14398 5.1 0.072 (MPa)
Packing time (s) 4 1.68223 0.42056 14.88 0.011 Packing time (s) 4 0.0005503 0.0001376 21.89 0.006
Cooling time (s) 4 1.00653 0.25163 8.91 0.029 Cooling time (s) 4 0.0005757 0.0001439 22.91 0.005
Error 4 0.11303 0.02826 Error 4 0.0000251 0.0000063
Total 24 4.22145 Total 24 0.0014516
R2 =97.32 % R2(adj)=83.94 % R2 =98.27 % R2(adj)=89.61 %
Int J Adv Manuf Technol

Table 9 ANOVA for sensitivity of warpage Table 11 An optimal parameter combination of RSM-GA

Source DF Sep SS Adj MS F P value Melt Injection Packing Packing Cooling

temperature velocity pressure time time
Melt temperature (°C) 4 0.0006288 0.01572 4.83 0.078
Injection velocity (mm/s) 4 0.0013665 0.03416 1.35 0.642 Optimal 250.7049 32.0459 35.9207 2.1 23
parameter
Packing pressure (MPa) 4 0.0015928 0.03982 3.82 0.111
combination
Packing time (s) 4 0.0116749 0.29187 9.82 0.024
Cooling time (s) 4 0.0137127 0.34282 8.43 0.035
Error 4 0.0035940 0.08958 as shown in Table 1. The length and warpage experi-
Total 24 0.0325697 mental results are shown in Tables 2 and 3, respectively.
R2 =88.97 % R2(adj)=82.74 % As apparent from the achieved L25(56) Taguchi orthog-
onal array experiment via Tables 2 and 3, the fourth
parameter combination had the highest S/N ratio for
experiments. PBT-2100 plastic material was chosen due the length quality characteristic. Similarly, the 22nd pa-
to its characteristics of high hardness, low shrinkage, rameter combination had the highest S/N ratio for the
and greater resistance to high temperature. The product warpage quality characteristic. The parameter combina-
nominal–the best length was 170.5 mm, and the small- tions are shown in Table 4. In the next step, experimen-
er–the best warpage value was 0 as expected. The ex- tal data for length and warpage were used in the
perimental mold and the finished product are shown in ANOVA; the results are presented in Tables 5 and 6,
Figs. 2 and 3, respectively. respectively. Accordingly, the significant influential fac-
Product length was measured using a Mitutoyo elec- tors toward length and warpage were identified. The P
tric caliper with a range of 300 mm and precision of value was less than 0.05. Then, the significant values
0.01 mm. Warpage was measured using the NEW were utilized to determine the Taguchi optimal parame-
VISION-II automatic optical image measuring instru- ter combinations, as in Table 7. In addition, the sensi-
ment with a precision of 0.001 mm. Product length tivity of the length and warpage was assessed by
and warpage were regarded as the product quality char- ANOVA; the results are shown in Tables 8 and 9, re-
acteristics in this study and, as such, determinants of the spectively. The significant process factors were packing
optimal process parameter settings. Length specifications time (x4) and cooling time (x5). The P value was less
and measurement positions are presented in Fig. 4. In than 0.05. According to the ANOVA results for the
addition, it can be observed from Fig. 5 that the warp- sensitivity of the quality characteristics and the product
age measurement positions were targeted to the scope of quality, packing time and cooling time were chosen as
1 cm inward from the cap. The measurement of the the control factors in the second stage of numerical
amount of change and the length of the same side analysis.
(i.e., the broken line) has also been shown.

2.2 Taguchi experiments and ANOVA 2.3 Response surface methodology

This study employed Taguchi parameter standard setting In accordance with the response surface methodology,
values to arrange an L25(56) orthogonal array experi- the second-order model analysis was conducted to con-
ment, in which no. 1 to no. 25 were Taguchi experi- struct the quality regression models L(X) and W(X),
mental data and no. 26 to no. 30 randomly generated where x1 is the melt temperature, x2 the injection veloc-
testing data within the level range. Accordingly, the ity, x3 the packing pressure, x4 the packing time, x5 the
control factors’ range settings were given five levels, cooling time, L(X) the length quality predictor, and

Table 10 Parameter setting ranges between the upper and lower limits
using GA Table 12 Two-stage optimal process parameters
Setting range Melt Injection Packing Packing Cooling Melt Injection Packing Packing Cooling
temperature velocity pressure time (s) time (s) temperature velocity pressure time time
(°C) (mm/s) (MPa)
Two-stage 250.7049 32.0459 35.9207 1.94 21.19
Upper level 261 46 43 2.1 23 optimal
Lower level 249 30 28 1.95 12 parameters
 Int J Adv Manuf Technol

Fig. 6 Surface plot for length

over the packing pressure and 170.7
packing time domain
171
170.6

170.8

170.6 170.5

Length (mm)
170.4
170.4

170.2

170 170.3
25

30
2. 170.2
35 2
1.8
Packing pressure (MPa) 1.6
40 1.4
1.2 170.1
1 Packing time (Sec)
45 0.8

W(X) the warpage quality predictor. The quality regres- the S/N ratio quality predictor for length, and W(X)SN the S/N
sion equations are shown as Eqs. (1) and (2). ratio quality predictor for warpage. The S/N ratio quality re-
In this study, product length and warpage were the quality gression equations are shown as Eqs. (3) and (4):
characteristics measured, in which the target length values
were expected to be the nominal–the best, and the warpage
values were expected to be the smaller–the better throughout LðX Þ ¼ − 1:14146 þ 1:30857x1 þ 0:46305x2 −0:11675x3 þ 1:31403x4 ð1Þ
the experiment range. In accordance with the Taguchi method, − 0:42655x5 −0:00171x1 x2 þ 0:00027x1 x3 −0:0092x1 x4
the second-order model analysis was conducted to construct þ 0:00222x1 x5 þ 0:00027x2 x3 −0:00112x2 x4 −0:00097x2 x5
the S/N ratio regression models, L(X)SN and W(X)SN, where x1
þ 0:02673x3 x4 þ 0:00002x3 x5 þ 0:0277x4 x5 −0:0025x21
is the melt temperature, x2 the injection velocity, x3 the pack-
ing pressure, x4 the packing time, x5 the cooling time, L(X)SN − 0:00015x22 þ 0:00005x23 −0:04241x24 −0:00372x25

Fig. 7 Surface plot for warpage

over the packing time and 0.22
packing pressure domain
0.2

0.18
0.2
0.16
Warpage (mm)

0.15 0.14

0.12
0.1

0.1

0.05
0.08

1 0.06
30
1.5 35
2 40 0.04
Packing time (Sec) Packing pressure (MPa)
Int J Adv Manuf Technol

Fig. 8 Surface plot for length

over the cooling time and packing
pressure domain 170.9

171

170.9 170.8

170.8

Length (mm)
170.7 170.7

170.6

170.6
170.5

170.4
170.5
170.3
25

45
20 170.4
40

15 35
Cooling time (Sec) 30
Packing pressure (MPa)
10 25

W ðX Þ ¼ −48:3015 þ 0:28623x1 þ 0:07625x2 þ 0:10492x3 ð2Þ The mutation operation was conducted using mutation
−0:58078x4 þ 0:9682x5 −0:00032x1 x2 −0:00063x1 x3 points with a mutation rate of 0.7 and a convergence
threshold of 1.000e-006 or after 50,000 iterations. For this
þ 0:00545x1 x4 −0:00387x1 x5 þ 0:00035x2 x3 −0:00307x2 x4
phase, the Optimization Toolbox of the MATLAB soft-
−0:0005x2 x5 þ 0:00423x3 x4 þ 0:00021x3 x5 −0:03524x4 x5 −0:00036x21
ware was utilized. The parameter setting ranges are shown
−0:00011x22 þ 0:00042x23 −0:11096x24 þ 0:00116x25 in Table 10 and the fitness function is presented as fol-
lows:
LðX ÞSN ¼ −14; 908 þ 94:1077x1 þ 45:9647x2 þ 49:6343x3 þ 1246:01x4 ð3Þ
þ 10:0615x5 − 0:12918x1 x2 − 0:15088x1 x3 − 4:21883x1 x4 Min F ðX Þ ¼ ðLðX Þ−170:5Þ2
þ 0:06723x1 x5 − 0:13372x2 x3 − 1:48379x2 x4 − 0:16649x2 x5 Min W ðX Þ
þ 0:36733x3 x4 − 0:12204x3 x5 þ 0:71716x4 x5 − 0:152475x21

− 0:036292x22 − 0:05419x23 − 0:4388x24 − 0:479452x25

s.t.

249≤ x1 ≤ 261
W ðX ÞSN ¼ 3479:4 − 22:7489x1 − 12:4748x2 þ 6:8635x3 þ 89:7046x4 ð4Þ 30≤ x2 ≤ 46
− 55:6622x5 þ 0:04771x1 x2 − 0:01574x1 x3 − 0:480376x1 x4 28≤ x3 ≤ 43
þ 0:21404x1 x5 − 0:04031x2 x3 þ 0:15468x2 x4 − 0:00559x2 x5 1:95≤ x4 ≤ 2:1
12≤ x5 ≤ 23
− 0:35563x3 x4 − 0:00293x3 x5 þ 2:09276x4 x5 þ 0:03571x21

þ 0:01892x22 − 0:00806x23 þ 2:53153x24 − 0:04929x25 where X=(x1,x2,x3,x4,x5), L(X) is the predicted value of the
length response, W(X) the predicted value of the warpage re-
2.4 Multiobjective optimization sponse, 170.5 the length quality characteristic target value, x1
the melt temperature, x2 the injection velocity, x3 the packing
In this stage, the quality predictors were combined with pressure, x4 the packing time, and x5 the cooling time. The
GA in order to carry out the parameter optimization of the obtained optimal parameter combination of the RSM-GA
injection molding process. Search range settings for the model is shown in Table 11.
parameters were determined using the length and warpage In the second stage, the optimal parameter combination of
process parameter settings, with each of the max and min RSM-GA was utilized for the initial parameter values. Ac-
values added to or subtracted by ½ standard values in cording to the ANOVA results for the sensitivity of the quality
case these max and min values reached the standard characteristics and the product quality, packing time and
values. The normalized range was 0.1∼0.9, the crossover cooling time had significant effects and were chosen as control
point was 100, and the single point as well as the cross- factors. The PIM process optimization was conducted using
over rate of 0.8 was employed for the crossover operation. the hybrid GA-PSO with S/N ratio, quality predictors, and
 Int J Adv Manuf Technol

Fig. 9 Surface plot for warpage

0.4
over the cooling time and packing
pressure domain
0.45

0.4 0.35

0.35

Warpage (mm)
0.3 0.3

0.25

0.2
0.25

0.15

0.1
0.2
12
14
16
18 30
20 35
Cooling time (Sec) 22 40 0.15
Packing pressure (MPa)

initial values of the process parameters as in the first stage. where X = (x1,x2, x3, x4, x5), L(X) is the quality predicted
The objective function can be expressed as in Eq. (5): value of the length, W(X) the quality predicted value of

2 the warpage, 170.5 the target of the quality character-
Min GðX Þ ¼ ðLðX Þ−170:5Þ2 þ LðX ÞSN −36:6959 istic (length), 0.082 the target of quality characteristic

2 (warpage), L(X) SN the predicted S/N ratio for the
þ W ðX ÞSN −21:2553 ð5Þ
length using the S/N ratio predictor, W(X)SN the pre-
dicted S/N ratio for the warpage using the S/N ratio
s.t. predictor, 36.6959 the target of the S/N ratio (length),
21.2553 the target of the S/N ratio (warpage), x4 the
W ðX Þ ≤ 0:082 packing time, and x5 the cooling time. The numerical
1:7 ≤ x4 ≤ 2:05 solutions of the two-stage optimal process parameters
19≤ x5 ≤ 22 are shown in Table 12.

Fig. 10 Surface plot for length

over the cooling time and packing
time domain 170.7 170.6

170.6
170.5
170.5
L e n g th (m m )

170.4 170.4

170.3
170.3
170.2

170.1 170.2

170
25
170.1
20 2 2.2
1.6 1.8
15 1.4
1.2
10 0.8 1
Cooling time (Sec) Packing time
Int J Adv Manuf Technol

Fig. 11 Surface plot for warpage

over the cooling time and packing
time domain 0.4

0.5

0.4
0.3

0.3

Warpage (mm)
0.2
0.2
0.1

0
0.1
-0.1

-0.2
25
0
2.2
20 2
1.8
1.6
15 1.4
1.2 -0.1
1
Cooling time (Sec) 10 0.8
Packing time (Sec)

2.5 Relation between process parameters and quality and length than the packing pressure; the cooling time in-
characteristics creased and the quality of the warpage and length improved.
As shown in Figs. 10 and 11, the warpage and length exhibited
There exists a very complicated nonlinear relationship be- better quality trends as the packing time or cooling time in-
tween process parameters and quality characteristics. Through creased. The results of the correlation analysis between the
the multiobjective two-stage optimization analysis, the con- process parameters and the quality characteristics indicated
trolling factors chosen were the three adjustment factors of that the packing time should be adjusted earlier than the
the process parameters: packing pressure, packing time, and cooling time, or that the above two factors be raised at the
cooling time; the two fixed factors of the process parameters same time to generate a better and lower warpage.
were melt temperature (254 °C) and injection velocity
(33 mm/s). According to the RSM regression models with
MATLAB tools, the surface plots of the regression functions 3 Confirmation experiment
over the different domains are illustrated in Figs. 6, 7, 8, 9, 10,
and 11. As shown in Figs. 6 and 7, at a fixed cooling time, the In this study, after conducting the injection molding process
warpage decreased (better quality), while the packing time parameters optimization, the Taguchi parameter design meth-
increased. The packing time was raised about 1.7 to 1.9 s, od, response surface optimization methodology, and the opti-
and the length reached the target state; meanwhile, the packing mal parameter combinations were utilized in carrying out the
pressure had a more significant influence on the warpage. confirmation experiments. The obtained process parameters
As shown in Figs. 8 and 9, at a fixed packing time, the were rounded off to two decimals. However, due to potential
cooling time had a more significant influence on the warpage precision mistakes in the machine settings, the optimal

Table 13 Machine parameter setting values for the confirmation experiment

Setting range Melt temperature (°C) Injection velocity (mm/s) Packing pressure (MPa) Packing time (s) Cooling time (s)

Taguchi optimal parameters 255 38 33 1.95 20

Machine settings 255 38 33 2 20
RSM-GA optimal parameters 250.7049 32.0459 35.9207 2.1 23
Machine settings 251 32 36 2.1 23
Two-stage optimal parameter settings 250.7049 32.0459 35.9207 1.94 21.19
Machine settings 251 32 36 1.9 21.2
 Int J Adv Manuf Technol

Table 14 Measured values of the

confirmation experiment Taguchi method RSM-GA Two-stage optimization

Run Length Warpage Run Length Warpage Run Length Warpage

1 170.63 0.113 1 170.45 0.030 1 170.51 0.029

2 170.62 0.073 2 170.47 0.032 2 170.48 0.016
3 170.61 0.095 3 170.48 0.034 3 170.51 0.025
4 170.63 0.101 4 170.44 0.037 4 170.50 0.030
5 170.59 0.067 5 170.48 0.034 5 170.47 0.030
6 170.61 0.109 6 170.47 0.029 6 170.49 0.023
7 170.61 0.095 7 170.44 0.045 7 170.49 0.036
8 170.58 0.065 8 170.47 0.039 8 170.47 0.019
9 170.58 0.110 9 170.44 0.034 9 170.48 0.032
10 170.57 0.105 10 170.46 0.026 10 170.47 0.024
11 170.63 0.064 11 170.45 0.034 11 170.48 0.035
12 170.57 0.066 12 170.45 0.056 12 170.49 0.033
13 170.58 0.122 13 170.48 0.017 13 170.47 0.014
14 170.59 0.068 14 170.44 0.057 14 170.50 0.033
15 170.59 0.067 15 170.45 0.027 15 170.47 0.021
16 170.58 0.118 16 170.44 0.030 16 170.47 0.028
17 170.58 0.086 17 170.43 0.035 17 170.47 0.028
18 170.63 0.116 18 170.44 0.025 18 170.48 0.021
19 170.57 0.104 19 170.48 0.029 19 170.49 0.014
20 170.61 0.101 20 170.43 0.023 20 170.48 0.016
21 170.62 0.104 21 170.49 0.029 21 170.48 0.011
22 170.57 0.065 22 170.47 0.027 22 170.50 0.018
23 170.62 0.108 23 170.48 0.052 23 170.47 0.017
24 170.62 0.066 24 170.43 0.045 24 170.47 0.019
25 170.58 0.109 25 170.43 0.057 25 170.48 0.012
Average 170.599 0.092 Average 170.456 0.033 Average 170.483 0.025
SD 0.022 0.021 SD 0.020 0.013 SD 0.013 0.007

parameter values were in compliance with the machine setting the length in conjunction with Cpk when conducting the anal-
restrictions concerning rounding modes, as shown in Table 13. ysis; additionally, the product specifications were in accor-
Additionally, the machine setting parameters were employed dance with industry product length standards, set at 170.5±
for the injection experiments, 25 treatments, during which the 0.2 mm.
length and warpage were measured. The measured values of After conducting measurements and calculations through
the optimal parameters as proposed are shown in Table 14. 25 actual experimental treatments, in terms of length quality
The process capability index (Cpk) is an important indicator characteristics, the optimized length average value of 170.483
when assessing process stability. In manufacturing industries, was acknowledged to be closest to the target value (170.5).
the threshold for a practical minimum Cpk is 1.33. Values less Additionally, the Cpk value increased from 1.53 to 4.69, and
than 1.33 are unable to help achieve effective output and in- the standard deviation decreased from 0.022 to 0.013.
ferior products may be the result. Therefore, this study took Concerning warpage quality characteristics, the warpage

Table 15 Optimization comparison of quality length Table 16 Optimization comparison of quality warpage

Average SD Cpk Average SD

Taguchi method 170.599 0.022 1.53 Taguchi method 0.092 0.021

RSM-GA 170.456 0.02 2.60 RSM-GA 0.033 0.013
Two-stage optimization 170.483 0.013 4.69 Two-stage optimization 0.025 0.007
Int J Adv Manuf Technol

Fig. 12 Comparison of quality

characteristic (length) through
systematic optimization

average value decreased from 0.092 to 0.025, and the standard to be costly in terms of both time and money, and the obtained
deviation reduced from 0.021 to 0.007 as shown in Tables 15 process parameters are not usually optimal. The ongoing use
and 16, respectively, indicating that after optimization, the of the Taguchi method of discrete optimization has resulted in
length value approached the target value, the warpage values the achieved parameters remaining nonoptimal. Therefore,
were lower, and the process became more stable. The compar- this study has proposed a two-stage optimization system for
isons of the quality characteristics (warpage and length) identifying the best process parameter combination, in which
through systematic optimization are illustrated in Figs. 12 length and warpage serve as the two major quality character-
and 13. istics, the aims being to stabilize the compliance length in the
manufacturing process and to reduce the amount of product
warpage. Using the two-stage optimization parameter combi-
4 Conclusion nation, 25 injection molds were processed. The results showed
that the parameter combination for Cpk length values for the
Injection molding parameter settings are affected by the cost Taguchi approach and the two-stage optimization was 1.53
of production. According to previous studies, in and 4.69, respectively. For the Taguchi method, the warpage
premanufacturing, the parameter setting combinations are average value was 0.092 with a standard deviation of 0.021.
generally determined based on the engineers’ practical field- whereas the warpage average value was 0.025 with a standard
related experiences or through empirical methods, trial-and- deviation of 0.007 for the two-stage optimization. Hence, it
error methods, and experimental designs. These methods tend could be assumed that the achieved results of this study will

Fig. 13 Comparison of quality

characteristic (warpage) through
systematic optimization
 Int J Adv Manuf Technol

help enhance the overall stability of the injection molding 12. Ozcelik B, Erzurumlu T (2006) Comparison of the warpage opti-
mization in the plastic injection molding using ANOVA, neural
process by generating a much greater Cpk value than those
network model and genetic algorithm. J Mater Process Technol
of the Taguchi method and industry standard; as a result, over- 171(3):437–445
all warpage will be reduced, which will effectively reduce 13. Akbarzadeh A, Sadeghi M (2011) Parameter study in plastic injec-
injection costs and the time required for mold designs for tion molding process using statistical methods and IWO algorithm.
manufacturing processes. Int J Model Optim 1(2):141–145
14. Chen WC, Kurniawan D, Fu GL (2012) Optimization of process
parameters using DOE, RSM and GA in plastic injection molding.
Acknowledgments The research was conducted as part of a project Adv Mater Res 472–475:1220–1223
sponsored by Polyprecision Industrial Co. Ltd., Hsinchu, Taiwan.
15. Chen WC, Wang MW, Chen CT, Fu GL (2009) An integrated
parameter optimization system for MISO plastic injection molding.
Int J Adv Manuf Technol 44:501–511
References 16. Kurtaran H, Erzurumlu T (2006) Efficient warpage optimization of
thin shell plastic parts using response surface methodology and
genetic algorithm. Int J Adv Manuf Technol 27(5–6):468–472
1. Wu CH, Chen WS (2006) Injection molding and injection compres- 17. Park HS, Dang XP (2010) Optimization of conformal cooling chan-
sion molding of three-beam grating of DVD pickup lens. Sensors nels with array of baffles for plastic injection mold. Int J Precis Eng
Actuators A Phys 125(2):367–375 Manuf 11(6):879–890
2. Teng Y, Xu Y (2008) Culture condition improvement for whole-cell 18. Sun B, Wu Z, Gu B, Huang X (2010) Optimization of injection
lipase production in submerged fermentation by Rhizopus molding process parameters based on response surface methodolo-
chinensis using statistical method. Bioresour Technol 99(9):3900– gy and genetic algorithms. In: Proc of the int’l conf. on computer
3907 engineering and technology, pp 397–400
3. Aggarwal A, Singh H, Kumar P, Singh M (2008) Optimizing power
19. Zhao P, Zhou H, Li Y, Li D (2010) Process parameter optimization
consumption for CNC turned parts using response surface method-
of injection molding using a fast strip analysis as a surrogate model.
ology and Taguchi’s technique—a comparative analysis. J Mater
Int J Adv Manuf Technol 49:949–959
Process Technol 200(1–3):373–384
20. Xu G, Yang Z, Long G (2012) Multi-objective optimization of
4. Kurt M, Bagci E, Kaynak A (2009) Application of Taguchi
MIMO plastic injection molding process conditions based on par-
methods in the optimization of cutting parameters for surface finish
ticle swarm optimization. Int J Adv Manuf Technol 58:521–531
and hole diameter accuracy in dry drilling processes. Int J Adv
Manuf Technol 40(5–6):458–469 21. Mostafa JJ, Mohammad MA, Ehsan M (2011) A hybrid response
5. Altan M (2010) Reducing shrinkage in injection moldings via the surface methodology and simulated annealing algorithm: a case
Taguchi ANOVA and neural network method. Mater Des 31:599– study on the optimization of shrinkage and warpage of a fuel filter.
640 World Appl Sci J 13(10):2156–2163
6. Zhai M, Xie Y (2010) A study of gate location optimization of 22. Tzeng CJ, Chen RY (2013) Optimization of electric discharge ma-
plastic injection molding using sequential linear programming. Int chining process using the response surface methodology and genet-
J Adv Manuf Technol 49(14):97–103 ic algorithm approach. Int J Precis Eng Manuf 14(5):709–717
7. Ng CF, Kamaruddin S, Siddiquee AN, Khan ZA (2011) 23. Chen WC, Fu GL, Tai PH, Deng WJ (2009) Process parameter
Experimental investigation on the recycled HDPE and optimization optimization for MIMO plastic injection molding via soft comput-
of injection moulding process parameters via Taguchi method. Int J ing. Expert Syst Appl 36:1114–1122
Mech Mater Eng 6(1):81–91 24. Tzeng CJ, Yang YK, Lin YH, Tsai CH (2012) A study of optimi-
8. Öktem H (2012) Optimum process conditions on shrinkage of an zation of injection molding process parameters for SGF and PTFE
injected-molded part of DVD-ROM cover using Taguchi robust reinforced PC composites using neural network and response sur-
method. Int J Adv Manuf Technol 61:518–528 face methodology. Int J Adv Manuf Technol 63:691–704
9. Wang X, Zhao G, Wang G (2013) Research on the reduction of sink 25. Huang HX, Li JC, Xiao CL (2015) A proposed iteration optimiza-
mark and warpage of the molded part in rapid heat cycle molding tion approach integrating back-propagation neural network with
process. Mater Des 47:779–792 genetic algorithm. Expert Syst Appl 42:146–155
10. Shi F, Lou ZL, Lu JG, Zhang YQ (2003) Optimization of plastic 26. Jordehi R (2015) Enhanced leader PSO (ELPSO): a new PSO var-
injection molding process with soft computing. Int J Adv Manuf iant for solving global optimisation problems. Appl Soft Comput
Technol 21:656–661 26:401–417
11. Yin F, Mao H, Hua L (2011) A hybrid of back propagation neural 27. Ranjania M, Murugesan P (2015) Optimal fuzzy controller param-
network and genetic algorithm for optimization of injection mold- eters using PSO for speed control of Quasi-Z Source DC/DC con-
ing process parameters. Mater Des 32:3457–3464 verter fed drive. Appl Soft Comput 27:332–356