Journal of Manufacturing Processes 32 (2018) 782–790

Contents lists available at ScienceDirect

Journal of Manufacturing Processes

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

Numerical optimization of process parameters in plastic injection molding T

for minimizing weldlines and clamping force using conformal cooling
channel
⁎
Satoshi Kitayamaa, , Kanako Tamadab, Masahiro Takanoc, Shuji Aibad
a
Kanazawa University, Kakuma-machi, Kanazawa, 920-1192, Japan
b
Graduate School of Natural Science & Technology, Kanazawa University, Kakuma-machi, Kanazawa, 920-1192, Japan
c
Industrial Research Institute of Ishikawa, 2-1, Kuratsuki, Kanazawa, 920-8203, Japan
d
Sodick Co., Ltd., Ka-1-1, Miya-machi, Kaga-shi, Ishikawa, 922-0595, Japan

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

Keywords: Weldlines are one of the major defects in plastic injection molding (PIM). Since the weldlines have an influence
Plastic injection molding on not only the appearance of product but also the strength, it is important to reduce the weldlines as much as
Weldlines possible. The melt plastic will be quickly solidified with the low weldline temperature, which results in long
Clamping force weldlines. The weldline temperature is one of the important factors for the weldlines reduction. Clamping force
Radial basis function
also affect the product quality, but the relationship between the weldlines and the clamping force is rarely
Sequential approximate optimization
discussed in the literature. In this paper, the minimum weldline temperature is maximized for the weldlines
reduction, whereas the clamping force is minimized for the high product quality. Therefore, a multi-objective
design optimization is performed and the pareto-frontier between them is identified. Numerical simulation in
PIM is generally so intensive that a sequential approximate optimization (SAO) using a radial basis function
(RBF) network is used to identify the pareto-frontier. Through the numerical simulation, the trade-off between
the minimum weldline temperature and the clamping force is clarified.

1. Introduction particular, a sequential approximate optimization (SAO) that the re-

Plastic injection molding (PIM) is one of the most widely used in-
dustrial technologies for producing plastic products with high pro-
ductivity. In the PIM, there are several process parameters such as melt
temperature, mold temperature, injection time, packing pressure, and
cooling time. In filling phase, melt plastic is filled into the die cavity
with the injection pressure. Then, the melt plastic is packed with an
appropriate packing pressure in packing phase. Finally, the melt plastic
is cooled down for the solidification in cooling phase, and the solid
plastic is then ejected. Conventionally, the trial and error method is
widely used to determine the process parameters. The inappropriate
process parameters easily cause major defects such as warpage, wel-
dlines, and short shot, so it is important to determine the optimal
process parameters to minimize these defects. Recently, computer aided
engineering (CAE) coupled with design optimization is recognized as
one of the powerful tools available [1–3], and the product quality such
as the warpage and the weldlines can be numerically evaluated. The
numerical simulation in PIM is so intensive that response surface ap-
proach is widely used to determine the optimal process parameters. In
sponse surface is repeatedly constructed and optimized is a popular
approach [4]. Let us briefly review several representative papers for
determining the optimal process parameters in PIM.
Shi et al. optimized several process parameters for minimizing the
maximum shear stress of a product [5]. Kurtaran et al. optimized the
process parameters for minimizing the warpage of a bus ceiling lamp
base using neural network (NN) and genetic algorithm (GA) [6]. They
also used the quadratic polynomial as the approximation technique [7].
Ozcelik and Erzurumlu adopted the similar approach for minimizing
the warpage of thin shell plastic product [8]. Chiang and Chang per-
formed a multi-objective optimization of process parameters for mini-
mizing warpage and shrinkage of a cell phone shell cover [9]. Note that
above researches adopt one-step optimization without iteration for
determining the optimal process parameters. In the one-step optimi-
zation, the optimum process parameters completely depend on the
accuracy of the response surface. In order to obtain a highly accurate
response surface, a big-size design of experiment (DOE) is generally
required. Gao and Wang adopted the SAO using the Kriging [10,11] in
which the warpage of a cellular phone cover was minimized. Zhang

⁎
Corresponding author.
E-mail addresses: kitayama-s@se.kanazawa-u.ac.jp (S. Kitayama), tamada@stu.kanazawa-u.ac.jp (K. Tamada), takano@irii.jp (M. Takano), aibas@sodick.co.jp (S. Aiba).

https://doi.org/10.1016/j.jmapro.2018.04.007
Received 26 July 2017; Received in revised form 10 March 2018; Accepted 5 April 2018
1526-6125/ © 2018 The Society of Manufacturing Engineers. Published by Elsevier Ltd. All rights reserved.
S. Kitayama et al. Journal of Manufacturing Processes 32 (2018) 782–790

Table 1
Summary of representative papers of process parameter optimization using approximation technique.
Refs. Process parameters in PIM Approximation technique Objective function

Shi et al. [5]
Kurtaran et al. [6]
Kurtaran and Erzurumlu [7]
Chiang and Chang [9]
Gao and Wang [10]
Gao and Wang [11]
Zhang et al. [12]
Deng et al. [13]
Zhao et al. [14]
Li et al. [15]
Shi et al. [16]
Xia et al. [17]
Cheng et al. [18]
Shi et al. [19]
Kitayama et al. [20]
Zhao et al. [21]
Mold temperature Injection time Back propagation neural network Maximum shear stress
Melt temperature Injection pressure
Mold temperature Packing time Back propagation neural network Warpage
Melt temperature Cooling time
Packing pressure
Mold temperature Packing pressure Quadratic polynomial Warpage
Melt temperature
Mold temperature Packing time Quadratic polynomial Warpage
Packing pressure Cooling time Shrinkage
Melt temperature Injection time Kriging Warpage
Mold temperature Packing pressure
Melt temperature Packing profile Kriging Warpage
Mold temperature
Mold temperature Injection time Quadratic polynomial Warpage
Melt temperature
Mold temperature Injection time Kriging Warpage
Melt temperature Packing pressure
Mold temperature Fast Strip Analysis Cativy pressure
Melt temperature Melt temperature
Injection time Cooling time
Packing profile Radial basis function Shrinkage
Mold temperature Packing pressure Back propagation neural network Warpage
Melt temperature Packing time
Injection time Cooling time
Packing pressure Melt temperature Gaussian process Warpage
Packing time Injection time
Injection pressure Cooling time
Diameter of runner Cooling time Back propagation neural network Volume shrinkage
Packing pressure Total volume of runner system
Packing time Cycle time
Mold temperature Packing time Back propagation neural network Warpage
Melt temperature Packing pressure
Injection time Cooling time
Pressure profile in injection and packing phase Radial basis function Warpage
Mold temperature Melt temperature
Injection time Packing time Kriging Volume shrinkage
Melt temperature Cooling temparature Sink marks
Packing pressure Cooling time

et al. and Deng et al. adopted the mode-pursuing sampling method for
the SAO and the warpage of a scanner frame was minimized [12,13].
Zhao et al. developed the fast strip analysis (FSA) for the surrogate
model and the pareto-optimal solutions among three objectives were
determined [14]. Other representative papers using the SAO on the
process parameters in PIM can be found in Refs. [15–21]. The summary
is listed in Table 1.
Weldlines which are formed when two or more melt fronts meet are
also one of the important product qualities in PIM. Since the weldlines
influence not only the appearance of products but also the strength, it is
preferable to reduce the weldlines as much as possible. Li et al. used the
Taguchi method to examine the effects of process parameters on the
weldlines [22], and they clarified that the melt temperature, the in-
jection velocity and the injection pressure had an influence on the
weldlines. Chen et al. controlled the weldline positions by changing the
gate location [23]. Wu et al. adopted distributed multi-population GA
to optimize the process parameters for minimizing the warpage of LCD
panel under the weldlines constraint [24]. Deng et al. optimized the
process parameters for minimizing four objective functions (the tem-
perature distribution, the shear stress distribution, the weldlines, and
the total volume of air traps), and the pareto-optimal solutions were
determined [25]. Kim et al. optimized both the gate locations and the
process parameters for minimizing the weldlines of an automotive front
bumper [26]. It is found from above brief review that the process
parameters in PIM affect the weldlines.
Clamping force also plays an important role in PIM. As suggested by
Yin et al. [27], energy consumption is also an important issue in the
PIM. Small clamping force can save the energy consumption, and will
lead to high productivity. Zhai and Xie optimized the process
parameters for minimizing the filling time of a toy table [28] in which
the clamping force was handled as the design constraint. Kitayama and
Natsume performed a multi-objective design optimization for mini-
mizing the volume shrinkage and the clamping force of a cup [29].
Zhang et al. optimized several process parameters for minimizing the
warpage and the clamping force of an oil cooler cove of diesel engine
[30]. It is found from above brief review that the clamping force is also
useful for improving the product quality.
Finally, the cooling channel in PIM needs to be considered.
Conventionally, straight-type cooling channels are widely used. Due to
the recent advancement in 3D printing technology, it is possible to
produce conformal cooling channel [31–33]. Dimla et al. reported that
conformal cooling channel could drastically reduce cycle time [34]. Au
and Yu designed various scaffold cooling channels and evaluated the
cooling performance [35] and found that the conformal cooling channel
could offer a more uniform thermal distribution. Kitayama et al. have
examined the cooling performance of conformal cooling channel nu-
merically and experimentally [36] where the process parameter opti-
mization for short cycle time and warpage reduction was performed.
The objective of this work is summarized as the following:
1 Process parameter optimization in PIM is still a crucial issue. Several
process parameters are optimized for weldlines reduction and
clamping force minimization.
2 Weldlines that cannot be completely eliminated in PIM are one of
the major defects. The melt plastic will be quickly solidified with the
low weldline temperature, which results in long weldlines. Then, we
consider that the minimum weldline temperature should be max-
imized for the weldlines reduction.

783
S. Kitayama et al. Journal of Manufacturing Processes 32 (2018) 782–790

3 The clamping force should also be minimized. As far as we survey,

the relationship between the minimum weldline temperature and
the clamping force is not discussed. If the trade-off between them
can be observed, this indicates that the clamping force will be useful
for wedlines reduction.
4 Conformal cooling channel is used to evaluate the weldlines and the
clamping force. To compare the effectiveness of the conformal
cooling channel, the conventional cooling channel is also used and
the effect to the weldlines and the clamping force is discussed.

The minimum weldline temperature and the clamping force are the
implicit function of process parameters, and Moldex3D (R14) is used to
numerically evaluate them. The numerical simulation is so intensive
Fig. 2. Overview of conformal cooling channel.
that the SAO using a radial basis function (RBF) network is used to
identify the trade-off with a small number of simulation. We have al-
ready developed the SAO using the RBF network [37], which is used as
the design optimization tool.
The rest of this paper is organized as follows. In Section 2, the nu-
merical simulation model is described. The multi-objective design op-
timization is formulated in Section 3. The flow of SAO is briefly de-
scribed in Section 4. The numerical result is shown in Section 5, in
which the comparison between the conformal and the conventional
cooling channel is made. Based on the numerical result, the experiment
using PIM machine (GL30-LP, Sodick) is carried out in order to examine
the validity of the proposed approach.
Fig. 3. Enlarged view of conformal cooling channel.

2. Overview of numerical simulation model Table 2

Material property of polycarbonate.
In this paper, a plastic product for a small LED panel shown in Fig. 1
Melt density [g/cm3] 1.02
is considered. Fig. 2 shows the overview of conformal cooling channel Solid density [g/cm3] 1.2
in the numerical simulation. Topology optimization is one of the useful Eject temperature [ °C] 91
approaches to develop the conformal cooling channel [38,39], but the Maximam shear stress [MPa] 0.3
simple application of topology optimization will generate the conformal Thermal conductivity [W/(m °C)] 0.228
Elastic module [GPa] 2.37
cooling channel without outlet (or inlet) [40]. Then, in this paper the
Poisson ratio 0.4
conformal cooling channel is developed by the designer’s experience. Specific module [J/(kg °C)] 2355
The enlarged view enclosed by the black circle in Fig. 2 is shown in Material characteristics Non-crystalline
Fig. 3. Polycarbonate (PC) is used as the material, and the property is Recommended mold temperature [°C] 60–90
listed in Table 2. Recommended melt temperature [°C] 270–300

Fig. 1. Plastic product and dimension.

784
S. Kitayama et al.
3. Multi-objective optimization of weldline temperature and
clamping force
3.1. Multi-objective design optimization
A multi-objective design optimization is generally formulated as
follow:
(f1 (x), f2 (x), ⋯, fK (x)) → min ⎫
⎪
x iL ≤ x i ≤ x iU i = 1, 2, ⋯, n
⎬
gk (x) ≤ 0 k = 1, 2, ⋯, ncon ⎪ (1)
⎭
where fi(x) is the i-th objective function to be minimized, K represents
the number of objective functions, xi denotes the i-th design variable, xiL
and xiU are the lower and upper bounds of the i-th design variable, n
represents the number of design variables. gk(x) denotes the k-th design
constraint, and ncon represents the number of design constraints. When
an objective function fj(x) is to be maximized, it is equivalent to
minimize the function –fj(x).
3.2. Objective functions
An illustrative example of weldlines is shown in Fig. 4, where the
black curved lines in Fig. 4(a) represent the weldlines. In addition,
Fig. 4(b) shows the distribution of weldline temperature.
The melt plastic will be quickly solidified with the low weldline
temperature, which results in the long weldlines. We consider that the
weldline temperature is one of the important factors for the weldlines
reduction. In particular, the minimum weldline temperature is taken as
the first objective function f1(x), and is maximized.
Next, the clamping force is taken as the second objective function
f2(x), and is minimized. The clamping force is evaluated as the product
between the maximum pressure of melt plastic and the projected area of
cavity, which can be numerically evaluated through the numerical si-
mulation of Moldex3D(R14).
3.3. Design variables
Process parameters in PIM strongly affect the weldline temperature
as well as the clamping force. In this paper, the injection time (tinj), the
melt temperature (Tmelt), the mold temperature (Tmold), the packing
time (tp), the packing pressure (Pp), the cooling time (tc), and the
cooling temperature of the coolant (Tc) are taken as the design variables
x. The lower and upper bounds of the design variables are set as fol-
lows:
0.1 ≤ tinj ≤ 0.5[s] 270 ≤ Tmelt60 ≤ Tmold ≤ 90[oC] ⎫
o ⎪
≤ 300[ C] ⎪ (STEP2) Numerical simulation is carried out at the sampling points,
5 ≤ tp ≤ 25[MPa] 50 ≤ Pp ≤ 200[MPa] 10 ≤ tc ≤ 40[s] ⎬
⎪ (STEP3) Objective functions and design constraints are approxi-
60 ≤ Tc ≤ 90[oC] ⎪ (2)
⎭ mated by the RBF network. Here, the approximated objective functions
785
Journal of Manufacturing Processes 32 (2018) 782–790
Fig. 5. Illustrative example of short shot.
3.4. Design constraint
Short shot that the melt plastic is not filled into the cavity is a fatal
defect in PIM. Inappropriate process parameters will easily lead to the
short shot, but the short shot is rarely discussed in the literature. Then,
the short shot should be handled as the design constraint. The short shot
Vs is numerically evaluated as follow:
V
Vs = 1 −
V0 (3)
where V and V0 denote the melt plastic volume and the cavity volume,
respectively. An illustrative example of the short shot in the numerical
simulation and the experiment is shown in Fig. 5 respectively, where
the gray part in Fig. 5(a) represents the melt plastic volume V0. It is
found from Eq. (3) that the positive value implies the short shot. In
other words, Eq. (3) represents the ratio of the unfilled volume to the
cavity volume. The design constraint g1(x) is evaluated as follow:
g1 (x) = { Vs Vs > 0
0 otherwise (4)
4. Sequential approximate optimization for multi-objective
optimization
Sequential approximate optimization (SAO) that response surface is
repeatedly constructed and optimized is one of the effective approaches
for determining the optimal process parameters in PIM. In this paper, a
multi-objective design optimization is formulated, and the main ob-
jective is to identify the pareto-frontier between the minimum weldline
temperature and the clamping force. The RBF network is used
throughout the SAO system. See Appendix about the RBF network and
the density function. In this section, we briefly describe the procedure
to identify the pareto-frontier.
(STEP1) Initial sampling points are determined by using the latin
hypercube design (LHD).
and objective functions are numerically evaluated.
Fig. 4. Illustrative example of weldlines and
distribution of weldline. Temperature.
S. Kitayama et al. Journal of Manufacturing Processes 32 (2018) 782–790

Fig. 7. Trade-off between minimum weldline temperature and clamping force.

Fig. 6. Flow of sequential approximate optimization for multi-objective opti-

mization.

are denoted as fi͠ (x) (i = 1, 2, ⋯, K ), and the approximated design

constraints are denoted as g͠ k (x)(k = 1, 2, ⋯, ncon) .
(STEP4) We find a pareto-optimal solution of response surface using
the weighted lp norm method formulated as follows:
K 1 p
[∑i = 1 (αi fi͠ (x)) p] → min ⎫
g͠ k (x) ≤ 0 k = 1, 2, ⋯, ncon ⎬ (5)
⎭
where αi (i = 1, 2, ⋯, K ) represents the weight of the i-th objective
function. p is the parameter, and is set to 4 in this paper. In order to Fig. 8. Weldlines using conformal cooling channel.
obtain a set of pareto-optimal solutions, various weights are assigned.
The optimal solution of Eq. (5) is taken as the new sampling point for
updating the response surface. As the result, the accuracy of the pareto-
selected for the comparison of the weldlines. The weldlines at three
frontier will be improved.
points are shown in Fig. 8, from which the weldlines at point A is
(STEP5) To find the unexplored region, the density function de-
shorter than others.
scribed in Appendix is used. The density function is constructed and
First, let us consider the area enclosed by the dashed line in Fig. 8.
minimized. The optimal solution of the density function is added as a
The weldline at points A and B is separated from the holes, whereas the
new sampling point. This step is repeated till a terminal criterion is
weldline at point C is connected. As the result, the weldline at point C is
satisfied. This step is introduced for uniform distribution of the sam-
long. It seems that the weldline at point A is long, compared with the
pling points. As the result, global approximation can be achieved.
one at point B. Then, let us consider the other area enclosed by the sold
(STEP6) If terminal criterion is satisfied, the SAO algorithm will be
line in Fig. 8. The weldline at points B and C is qualitatively similar, but
terminated. Otherwise, it will return to STEP 2. The average error be-
the weldine at point A is completely different. Therefore, the weldline
tween the response surface and the numerical simulation at the pareto-
at point A is more straight. As the result, the weldlines at point A are
optimal solutions obtained in STEP 4 is taken as the terminal criterion.
shorter than others. It is found from Figs. 7 and 8 that the weldlines are
The SAO algorithm will be terminated when the average error is within
so closely related the weldline temperature and the maximization of the
5%.
minimum weldline temperature is valid to the weldlines reduction. In
Fig. 6 shows the flow of the SAO for the multi-objective optimiza-
addition, high clamping force leads to high weldline temperature,
tion problem.
which results in short weldlines. Therefore, clamping force plays an
important role for weldlines reduction.
5. Numerical and experimental result Finally, the weldline temperature is listed in Table 3. In particular,
let us compare the weldline temperature at points A and C. At point A,
5.1. Numerical result

Fifteen initial sampling points are generated by the LHD, and the
pareto-frontier between the minimum weldline temperature and the Table 3
clamping force is identified using the SAO with the RBF network. The Comparison of weldline temperature.
pareto-frontier is shown in Fig. 7 where the black circles denote the Temperature [°C] Point A Point B Point C
pareto-optimal solutions. In order to identify the pareto-frontier, 48
sampling points (simulations) are required. It is found from Fig. 7 that Minimum 261 259 234
Maximum 291 290 287
the trade-off between them can be observed. Note that the minimum
Average 278 275 268
weldline temperature is maximized, whereas the clamping force is Standard deviation 7.56 8.10 12.92
minimized. Among the pareto-optimal solutions, points A, B, and C are

786
S. Kitayama et al. Journal of Manufacturing Processes 32 (2018) 782–790

Table 4
Optimal process parameters at points A, B and C in Fig. 7.
Point Melt Mold Injection time Packing time Packing pressure Cooling time Cooling Weldline Clamping force
temperature [°C] temperature [°C] [s] [s] [MPa] [s] temperature [°C] temperature [°C] [t]

A 287.0 64.8 0.17 11.3 170.3 27.0 77.2 275.3 9.7

B 291.2 74.2 0.26 19.3 161.3 15.0 80.2 258.4 7.7
C 288.9 69.0 0.28 13.5 166.6 16.0 78.6 227.3 6.6

the standard deviation of weldline temperature is small and the average

is high, which results in short weldlines. At point C, compared with
other two points, the standard deviation of weldline temperature is
large and the average is low, which results in long weldlines. It is found
from this result that the uniform weldline temperature distribution also
plays an important role for the weldlines reduction. The optimal pro-
cess parameters at points A, B and C are listed in Table 4.

5.2. Comparison of weldlines using conventional cooling channel

The conventional cooling channel shown in Fig. 9 is used for the

comparison. The pareto-frontiers between the conformal and the con-
Fig. 9. Overview of conformal cooling channel.
ventional cooling channel are shown in Fig. 10, where the white tri-
angles denote the pareto-optimal solutions of conventional cooling
channel. It is found from Fig. 10 that the pareto-frontier of conventional
cooling channel is so close to the one of conformal cooling channel.
However, the pareto-optimal solutions of conventional cooling channel
are not diverse, compared with the ones of conformal cooling channel.
It is important to find a set of solutions as diverse as possible in multi-
objective design optimization [41] and this result indicates that the
conventional cooling channel has clearly the limitation for improving
the weldline temperature and the clamping force. Furthermore, it is
assumed that the weldlines will not be effectively reduced by the con-
ventional cooling channel. Points A’, B’, and C’ in Fig. 10 are selected,
and the weldlines are shown in Fig. 11. It is clear from Fig. 11 that the
weldlines are much similar. In particular, the weldline enclosed by the
dashed line connects two holes, and this consequently results in the
long weldlines. It is clear from above discussions that the conformal
cooling channel is valid for the weldlines reduction, compared with the
conventional cooling channel.
Fig. 10. Comparison of pareto-frontiers between conformal and conventional
cooling channel.
5.3. Experimental result

The molding die of conformal cooling channel shown in Figs. 2 and

3 is developed by metal 3D printer (OPM250L, Sodick), and the over-
view is shown in Fig. 12. The cavity and cavity core are also developed

Fig. 11. Weldlines using conventional cooling channel.

Fig. 12. Overview of molding die in experiment.

787
S. Kitayama et al. Journal of Manufacturing Processes 32 (2018) 782–790

numerical result, the experiment using the PIM machine (GL30-LP,

Sodick) is carried out. It is very difficult to measure the length of the
weldlines, the weldline temperature, and the clamping force exactly in
the experiment. Then, the weldlines at points A, B, and C in Fig. 7 are
shown in Fig. 14, where, the red lines represent the weldlines. It is
found from Fig. 14 that the experimental results are qualitatively si-
milar to Fig. 8.

6. Conclusion

This paper investigated the relationship between clamping force and

Fig. 13. Cavity and cavity core developed by metal 3D printer.
Fig. 14. Experimental result of weldlines.
weldlines. The melt plastic would be quickly solidified with the low
weldline temperature, which resulted in long wedlines. To reduce the
weldlines, the minimum weldline temperature was maximized for the
weldlines reduction. In addition to the weldline temperature, the
clamping force was considered and was minimized. The injection time
(tinj), the melt temperature (Tmelt), the mold temperature (Tmold), the
packing time (tp), the packing pressure (Pp), the cooling time (tc), and
the cooling temperature of the coolant (Tc) were optimized. The nu-
merical simulation in PIM was so intensive that the SAO using the RBF
network was used to identify the trade-off. The numerical result showed
that the trade-off between the minimum weldline temperature and the
clamping force was observed. The weldlines were so closely related the
weldline temperature, and the maximization of the minimum weldline
temperature was valid to the weldlines reduction. In addition, high
clamping force led to high weldline temperature, which resulted in the
short wedlines. Therefore, clamping force played an important role for
weldlines reduction. We also identified the pareto-frontier using the
conventional cooling channel. It was found from the numerical result
that the conventional cooling channel had clearly the limitation for
weldlines reduction, and the validity of conformal cooling channel for
weldlines reduction wad clarified. Based on the numerical result, the
experiment was carried out. Through the numerical and experimental
result, the validity of the proposed approach has been confirmed.

by the metal 3D printer, which are shown in Fig. 13. Based on the

Appendix A. Radial basis function network

As described in Section 4, the RBF network is used throughout the SAO procedure. The RBF network and the density function are briefly
described.

(A1) Radial basis function network

The RBF network is a three-layer feed-forward network. The response surface using the RBF network is given by
m
yˆ (x) = ∑ j=1 wj K (x, xj) (A1)
where m denotes the number of sampling points, K(x, xj) is the j-th basis function, and wj denotes the weight of the j-th basis function. The following
Gaussian kernel is generally used as the basis function:
(x − xj)T (x − xj)
K (x, xj) = exp(− )
r j2 (A2)
In Eq. (A2), xj represents the j-th sampling point, and rj is the width of the j-th basis function. The response yj is calculated at the sampling point
xj. The learning of RBF network is usually accomplished by solving
m m
E= ∑ j=1 (yj − yˆ (xj))2 + ∑ j=1 λj wj2 → min (A3)
where the second term is introduced for the purpose of the regularization. It is recommended that λj in Eq. (A3) is sufficient small value (e.g.
λj = 1.0 × 10−2). The necessary condition of Eq. (A3) result in the following equation.
w = (HT H + Λ)−1HT y (A4)
where H, Λ and y are given as follows:

788
S. Kitayama et al. Journal of Manufacturing Processes 32 (2018) 782–790

⎡ K (x1, x1) K (x1, x2) ⋯ K (x1, x m) ⎤

⎢ K (x2, x1) K (x2, x2) ⋯ K (x2, x m) ⎥
H=⎢ ⎥
⋮ ⋮ ⋱ ⋮
⎢ ⎥
⎣ K (x m , x1) K (x m, x2) ⋯ K (x m, x m) ⎦ (A5)

⎡ λ1 0 ⋯ 0⎤
⎢0 λ2 ⋯ 0⎥
Λ=⎢
⋮ ⋮ ⋱ ⋮⎥
⎢ ⎥
⎣0 0 0 λm ⎦ (A6)

y = (y1 , y2 , ⋯, ym )T (A7)
It is clear from Eq. (A4) that the weigh vector w can be obtained by the matrix inversion. The following simple estimate is adopted to determine
the width in Eq. (A2) [31]:
dj,max
rj = nn m−1
j = 1, 2, ⋯, m (A8)
where n denotes the number of design variables, m the number of sampling points, dj,max the maximum distance between the j-th sampling point and
another one in the sampling points.

(A9) Density function to find an unexplored region

In the SAO, it is important to find out the unexplored region for global approximation. In order to find out the unexplored region with the RBF
network, we have developed a function called the density function [31]. The procedure to construct the density function is summarized as follows:
(D-STEP1) The following vector yD is prepared at the sampling points.
y D = (1, 1, ⋯, 1)Tm × 1 (A9)
(D-STEP2) The weight vector w of the density function D(x) is calculated as follows:
D

wD = (HT H + Λ)−1HT y D (A10)

(D-STEP3) The density function D(x) is minimized.
m
D (x) = ∑ j=1 wjD K (x, xj) → min (A11)
(D-STEP4) The point minimizing D(x) is taken as the new sampling point.
Fig. A1 shows an illustrative example in one dimension. The black dots denote the sampling points. It is found from Fig. A1 that local minima are
generated around the unexplored region. The RBF network is basically the interpolation between sampling points: therefore, points A and B in Fig.
A1 are the lower and upper bounds of the design variables of the density function.

Fig. A1. Illustrative example of density function in one dimension.

References
[1] Chang RY, Yang WH. Numerical simulation of mold filling in injection molding
using a three-dimensional finite volume approach. Int J Numer Methods Fluids
2001;37:125–48.
[2] Bikas A, Pantelelis N, Kanarachos A. Computational tools for the optimal design of
the injection moulding process. J Mater Process Technol 2002;122:112–26.
[3] Lam YC, Britton GA, Deng YM. A computer-aided system for an optimal moulding
conditions design using a simulation-based approach. Int J. Adv Manuf Technol
2003;22:574–86.
[4] Dang XP. General frameworks for optimization of plastic injection molding process
parameters. Simul Modell Pract Theory 2014;41:15–27.
[5] Shi F, Lou ZL, Lu JG, Zhang YQ. Optimisation of plastic injection moulding process
with soft computing. Int J Adv Manuf Technol 2003;21:656–61.
[6] Kurtaran H, Ozcelik B, Erzurumlu T. Warpage optimization of a bus ceiling lamp
base using neural network and genetic algorithm. J Mater Process Technol
2005;169:314–9.
[7] Kurtaran H, Erzurumlu T. Efficient warpage optimization of thin shell plastic parts
using response surface methodology and genetic algorithm. Int J Adv Manuf
Technol 2006;27:468–72.
[8] Ozcelik B, Erzurumlu T. Comparison of the warpage optimization in the plastic
injection molding using ANOVA, neural network model and genetic algorithm. J
Mater Process Technol 2006;171:437–45.
[9] Chiang KT, Chang FP. Analysis of shrinkage and warpage in an injection-molded

789
S. Kitayama et al.
part with a thin shell feature using the response surface methodology. Int J Adv
Manuf Technol 2007;35:468–79.
[10] Gao Y, Wang X. An effective warpage optimization method in injection molding
based on the Kriging model. Int J Adv Manuf Technol 2008;37:953–60.
[11] Gao Y, Wang X. Surrogate-based process optimization for reducing warpage in in-
jection molding. J Mater Process Technol 2009;209:1302–9.
[12] Zhang Y, Deng YM, Sun BS. Injection molding warpage optimization based on a
mode-pursuing sampling method. Polym-Plast Technol Eng 2009;48:767–74.
[13] Deng YM, Zhang Y, Lam YC. A hybrid of mode-pursuing sampling method and
genetic algorithm for minimization of injection molding warpage. Mater Des
2011;31:2118–23.
[14] Zhao P, Zhou H, Li Y, Li D. Process parameters optimization of injection molding
using a fast strip analysis as a surrogate model. Int J Adv Manuf Technol
2010;49:949–59.
[15] Li C, Wang FL, Chang YQ, Liu Y. A modified global optimization method based on
surrogate model and its application in packing profile optimization of injection
molding process. Int J Adv Manuf Technol 2010;48:505–11.
[16] Shi H, Gao Y, Wang X. Optimization of injection molding process parameters using
integrated artificial neural network model and expected improvement function
method. Int J Adv Manuf Technol 2010;48:955–62.
[17] Xia W, Luo B, Liao XP. An enhanced optimization approach based on Gaussian
process surrogate model for process control in injection molding. Int J Adv Manuf
Technol 2011;56:929–42.
[18] Cheng J, Liu Z, Tan J. Multiobjective optimization of injection molding parameters
based on soft computing and variable complexity method. Int J Adv Manuf Technol
2013;66:907–16.
[19] Shi H, Xie S, Wang X. A warpage optimization method for injection molding using
artificial neural network with parametric sampling evaluation strategy. Int J Adv
Manuf Technol 2013;65:343–53.
[20] Kitayama S, Onuki R, Yamazaki K. Warpage reduction with variable pressure profile
in plastic injection molding via sequential approximate optimization. Int J Adv
Manuf Technol 2014;72:827–38.
[21] Zhao J, Cheng G, Ruan S, Li Z. Multi-objective optimization design of injection
molding process parameters based on the improved efficient global optimization
algorithm and non-dominated sorting-based genetic algorithm. Int J Adv Manuf
Technol 2015;78:1813–26.
[22] Li H, Guo Z, Li D. Reducing the effects of weldlines on appearance of plastic pro-
ducts by Taguchi experimental method. Int J Adv Manuf Technol 2007;32:927–31.
[23] Chen MY, Tzeng HW, Chen YC, Chen SC. The application of fuzzy theory for the
control of weld line positions in injection-molded part. ISA Trans 2008;47:119–26.
[24] Wu CY, Ku CC, Pai HY. Injection molding optimization with weld line design
constraint using distributed multi-population genetic algorithm. Int J Adv Manuf
Technol 2011;52:131–41.
[25] Deng YM, Zheng D, Lu XJ. Injection moulding optimization of multi-class design
790
Journal of Manufacturing Processes 32 (2018) 782–790
variables using a PSO algorithm. Intl J Adv Manuf Technol 2008;39:690–8.
[26] Kim KH, Park JC, Suh YS, Koo BH. Interactive robust optimal design of plastic
injection products with minimum weldlines. Int J Adv Manuf Technol
2017;88:1333–44.
[27] Yin F, Mao H, Hua L. A hybrid of back propagation neural network and genetic
algorithm for optimization of injection molding process parameters. Mater Des
2011;32:3457–64.
[28] Zhai M, Xie Y. A study of gate location optimization of plastic injection molding
using sequential liner programming. Int J Adv Manuf Technol 2010;49:97–103.
[29] Kitayama S, Natsume S. Multi-objective optimization of volume shrinkage and
clamping force for plastic injection molding via sequential approximate optimiza-
tion. Simul Modell Pract Theory 2014;48:35–44.
[30] Zhang J, Wang J, Lin J, Guo Q, Chen K, Ma L. Multiobjective optimization of in-
jection molding process parameters based on opt LHD, EBFNN, and MOPSP. Int J
Adv Manuf Technol 2016;85:2857–72.
[31] Sachs E, Wylonis E, Allen S, Cima M, Guo H. Production of injection molding
tooling with conformal cooling channels using the three dimensional printing
process. Polym Eng Sci 2000;40(5):1232–47.
[32] Xu X, Sachs E, Allen S. The design of conformal cooling channels in injection
molding tooling. Polym Eng Sci 2001;41(7):1265–79.
[33] Ferreira JC, Mateus A. Studies of rapid soft tooling with conformal cooling channels
for plastic injection moulding. J Mater Process Technol 2003;142:508–16.
[34] Dimla DE, Camilotto M, Miani F. Design and optimisation of conformal cooling
channels in injection moulding tools. J Mater Process Technol 2005;164-
165:1294–300.
[35] Au KM, Yu KM. A scaffolding architecture for conformal cooling design in rapid
plastic injection moulding. Int J Adv Manuf Technol 2007;34:496–515.
[36] Kitayama S, Miyakawa H, Takano M, Aiba S. Multi-objective optimization of in-
jection molding process parameters for short cycle time and warpage reduction
using conformal cooling channel. Int J Adv Manuf Technol 2017;88:1735–44.
[37] Kitayama S, Arakawa M, Yamazaki K. Sequential approximate optimization using
radial basis function network for engineering optimization. Optim Eng
2011;12:535–57.
[38] Zhou M, Alexandersen J, Sigmund O, Pedersen CB. Industrial application of to-
pology optimization for combined conductive and convective heat transfer pro-
blems. Struct Multidiscip Optim 2016;54:1045–60.
[39] Coffin P, Maute K. Level set topology optimization of cooling and heating devices
using a simplified convection model. Struct Multidiscip Optim 2016;53:985–1003.
[40] Lohan DJ, Dede EM, Allison JT. Topology optimization for heat conduction using
generative design algorithms. Struct Multidiscip Optim 2017. http://dx.doi.org/10.
1007/s0158-016-1563-6.
[41] Miettinen KM. Nonlinear multiobjective optimization. Kluwer Academic Publishers;
1998.