Upload
others
View
7
Download
0
Embed Size (px)
Citation preview
For Review Only
The Novel Simulation of Bottle Blow Molding to Determine
Suitable Parison Thicknesses for Die Shaping
Journal: Songklanakarin Journal of Science and Technology
Manuscript ID SJST-2017-0497.R1
Manuscript Type: Original Article
Date Submitted by the Author: 09-Apr-2018
Complete List of Authors: Suvanjumrat, Chakrit; Mahidol University Faculty of Engineering, Mechanical Engineering
Keyword: Simulation, Bottle, Blow Molding, Parison, Die Shaping
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
For Review Only
Original Article
The Novel Simulation of Bottle Blow Molding to Determine Suitable Parison
Thicknesses for Die Shaping
Chakrit Suvanjumrat1,2*, Nathaporn Ploysook
2, Ravivat Rugsaj
2 and
Watcharapong Chookaew1
1Department of Mechanical Engineering, Faculty of Engineering, Mahidol
University, Nakhon Pathom, 73170, Thailand
2Laboratory of Computer Mechanics for Design (LCMD), Department of
Mechanical Engineering, Faculty of Engineering, Mahidol University,
Nakhon Pathom, 73170, Thailand
* Corresponding author, Email address: [email protected]
Abstract
This research proposed artificial intelligence (AI) method to determine suitable
parison thicknesses at the horizontal cross-section of bottle for the die shaping. The
cross-section of the bottle was emphasized for the blow molding simulation by the finite
element method (FEM) in order to achieve the accurate results more than the full bottle
shape simulation. The absolute average error of cross-section simulation was 17.02%
when it was compared with the experimental data. Subsequently, the artificial neural
network (ANN) was trained using the FEM data of cross-section blow molding. The
ANN was good agreement with the experimental data for the more complex shape of
two implementable bottles and obtained an absolute average error of 15.66%.
Eventually, the genetic algorithm (GA) method was developed to determine the uniform
perimeter thickness of bottle which passed the top load testing. The predictable parison
thickness by AI gave an absolute average error of 4.18% from the desirable thickness.
Page 2 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Keywords: Simulation, bottle, blow molding, parison, die shaping
1. Introduction
Lubricant oil bottles are attractive with the color and shape. There are many
complex shapes of lubricant oil bottle which were launched in the automotive market.
These bottles are produced using the extrusion blow molding process. There are two
extrusion methods which comprise of the die-gap programming and die shaping have
been used to prepare the plastic tube or parison for blowing lubricant oil bottles. The
suitable parison thickness is important to use for forming the lubricant oil bottles by
inflation to completely contact on the surface of bottle mold cavity. Parison thickness is
regulated by the extrusion die methods to obtain the bottle thickness regarding to the
requirement of bottle design. However, the blow molding theory is employed to
calculate the initial thickness of parison (Lee, 1998), the trial and error method is often
used to adjust the die gap for the complex shape of bottles. The thickness of parison is
difficult to control properly to form the lubricant oil bottles which principally have a
reflective symmetry shape. It is often depended on the experience of die makers to
shape the extrusion die. Consequently, the significant amount of wasteful plastic, cost
and time are lost at the parison setting.
In the present days, the computer aided engineering (CAE) which involves the
finite element method (FEM) to simulate the extrusion blow molding process is referred
for the engineering design to reduce the trial and error method for setting up the parison
thickness (Marcmann, Verron, & Peseux, 2001). Nevertheless, the FEM is started by
preparing the extrusion blow molding model which is difficult to perform with the
Page 3 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
complex shape bottles. Particularly, it has limitation to suddenly optimize the parison
thickness for the desirable thickness of bottle design; therefore, the simulating time is
also entirely consumed for the trial and error method to prepare the blow molding
model and to determine the suitable thickness of parison.
The optimal techniques had been applied to the finite element analysis (FEA) to
search an appropriated parison thickness of hollow products (Biglione, Bereaux,
Charmeau, Balcaen, & Chhay, 2016). The gradient-based optimization scheme had been
proposed to determine the optimal bottle thickness that satisfied to resist two types of
loading, top load and internal pressure (Gauvin, Thibault, & Laroche, 2003). The
parison thickness had been controlled using parison programming or die gap opening.
The variable thickness of parison enforced researchers to prepare many results of FEA
for the optimization of each shape of bottles. The optimal die gap programming of
extrusion blow molding process could use the neural network to determine for the
uniform part thickness after the parison inflation. Taguchi’s method was applied to
reduce the number of die gap data of FEA to prepare for training by the artificial neural
network (ANN). The parison thickness which was assumed to be equal in the cross
section area was predicted by the neural network. The genetic algorithm (GA) was then
applied to search for the optimal parison thickness (Jyh-Cheng, Xiang-Xian, Tsung-
Ren, & Thibault, 2004). The fuzzy neural-Taguchi network with GA (FUNTGA) was
established to help the parison programming of the bottle which should be an axis
symmetric shape. Huang and Huang (2007) used POLYFLOW 3.10 software to prepare
the blow molding data of bottles for the ANN which was combined to find the optimal
parison thickness using the GA. Yu and Juang (2010) proposed fuzzy reasoning of the
prediction reliability to evolve the ANN training to guide the GA searching. The small
Page 4 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
number of training samples was performed using Taguchi’s orthogonal array. This
methodology balance reliability and optimality for the FEA of extrusion blow molding
process. Even though the AI method (ANN, fuzzy logic and GA) was used to optimize
the parison thickness for extrusion blow molding process, it was just effective to the
axis symmetry bottle shape. Otherwise, the previous AI method did not perform to
prepare a suitable parison from the die shaping (Lee, 1990) for the blow molding
process. Unfortunately, the AI method could not be accomplished when the FEA could
not perform to prepare the training samples for the complex bottle shape such as
rectangular shape bottles, one handle bottles, and two handle bottles.
This research proposed the novel simulation technique for the die shaping of the
complex shape bottles in the blow molding process. The horizontal cross-section of
bottle shape was reasonable to simulate using FEM instead of the full shape bottle for
the blow molding process with the shaping die. The ANN and GA were applied for the
optimal parison thickness to obtain the uniform cross-section thickness of bottle which
could resist the top load specification (Suvanjumrat & Chaichanasiri, 2014). The flow
chart of the novel approach which is used to determine the suitable parison thickness for
the die shaping is shown in Fig. 1.
2. Experiment
The lubricant oil bottle is shown in Fig. 2a. This lubricant oil bottle has a
reflective symmetry shape. It was produced with the high density polyethylene (HDPE)
grade 6140B of SCG Plastic Company Ltd. The extrusion blow molding process was
employed to produce the lubricant oil bottles and set up the producing conditions which
were comprised the parison outer diameter of 52 mm, initial temperature of 180 °C, die
Page 5 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
gap opening of 150 mm, die closing speed of 2.2 mm/sec, blow-up pressure of 1 bar,
and blow-up time of 0.3 sec. The blow molding machine was set three shaping dies
without the parison programming. The five parisons which were extruded through the
shaping dies had been selected to measure the thickness (Fig. 2b). The twelve columns
(A to F) around parison perimeter were assigned to be the reference line for measuring
parison thickness. There were started by A at 15 degrees from the front parting line of
bottle in the counterclockwise (CCW) rotation and started by A at -15 degrees from the
front parting line of bottle in the clockwise (CW) rotation. The thickness of parison
every level of 20 mm along its height was assigned points to measure the thickness. The
thickness gauge model MiniTest 7200 FH of ElektroPhysik which had the resolution of
±3 µm had been used to measure the parison thickness at the reference points. Figure 3a
shows the average thickness graph of parisons which are blew to be the lubricant oil
bottles. The parison had been the circular hollow cylinder with the outerdiameter of
52.00 mm and did not have the uniform thickness around its perimeter. The thickness of
parison increased from top to bottom which was affected from the gravity force. The
correlation of parison height and thickness can be written as Eq. 1.
�� = �ℎ + ��(1) where �� is the parison thickness, � is the parison height, � is 2.492×10
-3, and �� is
constant regarding to the initial thickness of parison flow through the die gap.
The label area on bottle surface at the bottle height of 115.00 mm was happened
the maximum blow ratio of parison along the bottle height and affected by the
unsuitable die shaping which did not blow the uniform thickness around the bottle
perimeter. The fifteen bottle samples were selected and measured the wall thickness.
The bottle also assigned column positions regarding to the parison columns. Figure 3b
Page 6 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
shows the average thickness graphs of the lubricant oil bottles along the bottle height.
The die shaping had been performed to control the parison thickness to form the
complete shape of bottle; therefore the parison thickness which was regarded to be the
bottle corner was more thicker than the other column in parison tube. The column of
parison which was formed to be the bottle wall as far from the parison axis had
thickness more than other parison column. The parison thickness which was obtained by
the die shaping method depended on the experience of die maker. The number of
shaping to obtain the final shape of die gap related to the time and cost consuming for
setting up the extrusion blow molding process of lubricant oil bottles.
3. The Bottle Blow Molding Models
The 3D model of 1.0 liter lubricant oil bottle as shown in Fig. 4 is carried out to
create finite element model of the blow molding process. The full shape bottle would be
created by starting at the mold clamping on parison until the parison was blown to from
the bottle shape. The cross-section shape bottle would be investigated at the horizontal
section which was high from the bottle bottom of 115.00 mm. The bottle mold did not
clamp on parison to blow the bottle wall at the height of 115.00 mm; therefore, the
clamping process did not consider for creating the finite element model. The bottle blow
molding simulation was performed using the personal computer with Core-i5 CPU 3.3
MHz and 4 GB DDR III SDRAM memory.
3.1 Full Shape Bottle
The blow molding model was prepared using three parts of finite element model,
two mold parts and one parison part. Figure 4c shows the finite element model of full
Page 7 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
shape bottle blow molding process. The parison part was generated by rectangular
elements which were total amount of 2,880 elements. The plate element had the width
and length of 3.75 and 4.53 mm, respectively. The mold parts were placed on two sides
of parison for clamping and blowing, respectively. The surface model was assigned to
be the bottle mold because the parison was clamped and blown to contact just only the
inner surface of mold. The material model of parison was assigned regarding to
Thusneyapan and Rugsaj (2014). The thickness of parison used the average thickness of
the measurement results around parison perimeter to assign the parison thickness model.
The fixed boundary condition was defined on nodes at the top edge of parison
model. The surface molds were assigned to clamp parison and to hold until the parison
was blown to form the final shape of bottle. The blowing pressure of 607.95 kPa was
defined on the inner surface of parison. The temperature of parison was 180 °C while
the environmental temperature was 35 °C.
The simulation result has been shown by the sequent images of blow molding
process in Fig. 5a. The color contour was represented by the thickness of the bottle from
start to finish blow molding process. The final thickness of bottle of the simulation
result was compared with the experimental data. Figure 5b shows graphs of bottle
thickness for experiment and simulation. The simulation graphs have trended as like as
the experimental graphs. The finite element analysis (FEA) obtained an absolute
average error of 32.35%.
3.2 Cross-Section Shape Bottle
The finite element model of cross-section shape bottle blow molding process
was more advanced than the full shape bottle blow molding simulation. The interesting
Page 8 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
horizontal section which was high from the bottom of 115.00 mm was cut to simulate
using FEM. This bottle section did not concern the clamping step in the extrusion blow
molding process. The initial section of parison was in the clamped mold at the initial
step of simulation; therefore, the parison model was prepared in 2D finite element
model. The hollow circular shape of parison model was divided by the rectangular plate
elements. The parison thicknesses were defined as same as the experimental data at the
bottle height of 115.00 mm. Total elements to create the parison model were 120
elements. The material property of parison was defined as same as the parison model in
the full shape bottle blow molding process. Otherwise conditions comprised internal
pressure and environment temperature also defined as same as the full shape bottle.
Figure 6a shows the simulation results of the cross-section shape bottle blow
molding process. The cross-section of parison was inflated to form the cross-section
shape of bottle at its height of 115.00 mm. The deformation of parison could be
represented by the color contour. The thinner area would be yellow and the thickener
area was blue. The bottle thickness could be measured by the distance between inner
and outer perimeter of finite element model at the final step of simulation. Figure 6b
shows the comparison graph of bottle thickness between experiment and cross-section
shape bottle simulation. The FEA results of the cross-section shape bottle were more
agreement with the experiment than the full shape bottle blow molding simulation. The
absolute average error of the cross-section shape bottle blow molding simulation which
compared with the experiment was 17.02%.
The blow molding simulation using the cross-section shape bottle can improve
for more accuracy by setting other material models such as K-BKZ, following works by
(Rasmussen & Yu, 2008). However, the cross-section shape bottle blow molding
Page 9 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
method has the final thickness of bottle well and can use the cross-section parison
thickness at the interested level for determining the initial parison thickness around the
die gap using Equation 1. Particularly, this FEM can use to guide mold makers for die
shaping of the extrusion blow molding process in the future work.
4. Artificial Neural Network
The cross-section bottle blow molding simulation was trained using ANN. The
back propagation learning method which improved using Levenberg-Marquardt
algorithm was defined for the training technique. The ANN architecture with three
hidden layers was designed to train the cross-section blow molding process. The
multiple input neuron and three hidden layers network was applied to learn the blow
molding simulation of the cross-section bottle. The ANN gave the final thickness of
cross-section bottle at the desirable height when the parison thicknesses were entered
for the input of network. The thicknesses of parison and bottle at 15°, 45°, 75°, 105°,
135° and 165° were defined to be the input and output data, respectively. The Log-
Sigmoid transfer function (Rajasekaran & Vijayalakshmi Pai, 2008) which is written in
Equation 2 is used to be the first and second hidden layer. The linear transfer function
was the third hidden layer of network. The first, second and third hidden layer has 25,
15 and 10 neurons, respectively.
�(�) = �1 + ��(���) (2)
where �� is the x-value of the sigmoid’s midpoint, � is the curve’s maximum value, and
� is the steepness of the curve.
This work was created the novel orthogonal array to reduce the input data
preparation from the FEA. The orthogonal array had a simple concept to arrange
Page 10 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
thickness of parison for the input data. The three level thicknesses around parison
perimeter were defined which comprised of the high, middle and low thickness. The
parison thickness at two consecutive columns was random for three level thicknesses
meanwhile the residual columns were the middle thickness. The level of thickness was
arranged for each set of the input data; therefore, the input data of 54 set for training the
ANN were defined to prepare for prediction of the cross-section thickness of bottle wall
at the interesting bottle height.
The ANN which was trained using orthogonal array could be used to predict
thickness of the cross-section shape bottle blow molding simulation. The thicknesses
around cross-section perimeter of ANN prediction are compared with the FEA which is
shown by graphs in Fig. 7. The absolute average error of ANN when compared with the
FEA using five unseen initial parison thicknesses was less than 8.88%. The ANN
showed the good agreement of result with the FEA and could use for predicting of the
bottle thickness rapidly.
The parison tubes of two different shapes of one-liter bottle from the extrusion
blow molding process were collected for the ANN testing. The A-shape and B-shape
bottle is shown positions to measure parison thickness in Fig. 8 (right). These parisons
are blown to be the A-shape and B-shape bottle which is shown in Fig. 8 (left). The
interesting cross-section of A-shape and B-shape bottle was at the bottle height of 112.5
and 118.0 mm, respectively. The ANN was tested by input the average thickness of
parisons, five parison tubes for each bottle shape, for the input data. The output data of
ANN which was the bottle thickness at the interesting horizontal cross-section was
compared with the experimental data. Figure 9a and 9b shows the comparison graph
between the ANN and experiment of A-shape and B shape bottle, respectively. The
Page 11 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
ANN could predict the bottle thickness in good agreement with experiment. The
absolute average error of ANN was 12.28% and 19.05% when compared cross-section
thickness with the A-shape and B-shape bottle, respectively.
5. Genetic Algorithm
The top load simulation was performed on the one-liter bottle model to
determine the target thickness. The one-liter bottle thickness of 1.164 mm was required
to support the 25 kg top load. The bottle thickness around perimeter of the interesting
cross-section which was equal to the desirable thickness was the optimal solution for the
GA. The objective function as the fitness function in GA was designed for determining
the initial parison column thicknesses which gave the final bottle thickness as equal to
the desirable thickness. This study used the trained ANN prediction model for the
fitness function. The goal output of ANN prediction can be written by the following
fitness function.
�(�) =��(�� − ��)�∑ �� �!"#
�!"(3)
where �� is the desirable bottle thickness, �� is the bottle column thickness, and % is the
number of bottle column.
The GA was applied to search the parison thickness which was blown the uniform
cross-section bottle thickness. The parison thicknesses at A to F column of the one-liter
bottle are described in Table 1. The suitable parison thicknesses were specified for the
finite element model of the cross-section shape bottle blow molding. The results of
blow molding simulation are shown in Fig. 10. The final thicknesses of parison which
were blown to be the one-liter bottle were the uniform thickness. The bottle thickness
could be predicted using the ANN. The final thicknesses of bottle at the bottle height of
Page 12 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
115.00 mm by the FEA and ANN by specifying the suitable parison thicknesses using
GA are described in Table 2. The thickness of bottle of ANN had an absolute average
difference from the FEA of 0.199 mm and was deviated from the target (1.164 mm) by
the absolute value of 4.18%.
The suitable parison thicknesses which were blown to be the uniform bottle
thickness at the desirable bottle height can use to calculate the initial parison thicknesses
at the die gap using Equation 1. The parison thicknesses at the die gap will be used for
die shaping of the blow molding process in the further work. It is the good guidance for
the die maker of the lubricant oil bottle manufacturing.
6. Conclusions
This research had been developed a novel approach to optimize the parison
thickness using hybrid method which was combined the FEM, ANN, and GA. The
horizontal cross-section of bottle was developed to simulate the blow molding process
using the FEM. It was demonstrated that this FEM was more advantage than the finite
element model of the full shape bottle. Otherwise, the cross-section of bottle blow
molding simulation was validated with the physical experiment. The thicknesses of
cross-section bottle at the interesting height were defined the error and were obtained
the simulated results in the good agreement with the experimental data which had an
absolute average error less than 17.02%. The horizontal cross-section bottle blow
molding simulation was used to prepare the initial parison and final bottle thickness data
to train ANN instead the physical experiment. The novel orthogonal array was
developed based on logical connectivity of parison at each zone to optimize amount of
data which was required to train ANN. The ANN was demonstrated by comparing with
Page 13 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
the FEA of the cross-section bottle blow molding process using five unseen sets of
initial parison thickness. It was found that the absolute average error of ANN was
8.88%. Particularly, it was validated with two complex shape bottles and had an
absolute average error less than 15.66% when compared with the physical experiment.
Subsequently, the trained ANN was used as the fitness function for the parison
thickness optimization using GA. The objective of the GA is the searching of suitable
initial parison thickness which could give the uniform thickness around the horizontal
cross-section perimeter at the interesting height of the reflective shape bottle. The
optimized parison thickness which was obtained by GA was validated using FEM. The
cross-section bottle blow molding was simulated using the initial parison thickness
which was optimized and searched by GA. The simulated result indicated that the
optimized parison thicknesses were successful well to blow the uniform bottle thickness
regarding to a thickness target. The absolute deviation of thickness was 4.18% from the
target. The optimization of parison thickness by the novel method which was developed
in this research could be a guide for shaping the extrusion die by using with the
simplified relationship equation between the parison height and thickness. This novel
approach to determine initial thickness of parison which was extruded through the die
will be useful to guide the die makers for shaping of the die gap in the further works.
Acknowledgments
The authors wish to thank Panjawattana Plastic Public Company Limited for
supporting the experimental devices.
References
Page 14 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Biglione, J., Bereaux, Y., Charmeau, J. Y., Balcaen, J., & Chhay, S. (2016). Numerical
simulation and optimization of the injection blow molding of polypropylene
bottles – a single stage process. International Journal of Material Forming,
9(4), 471-487.
Gauvin, C, Thibault, F., & Laroche, D. (2003). Optimization of blow molded part
performance through process simulation. Polymer Engineering & Science,
43(7), 1407-1414.
Huang, G. Q., & Huang, H. X. (2007). Optimizing parison thickness for extrusion blow
molding by hybrid method. Journal of Materials Processing Technology,
182(1-3), 512-518.
Jyh-Cheng, Y., Xiang-Xian, C., Tsung-Ren, H., & Thibault, F. (2004). Optimization of
extrusion blow molding processes using soft computing and Taguchi’s method.
Journal of Intelligent Manufacturing, 15(5), 625-634.
Lee, N. C. (1990). Plastic Blow Molding Handbook. New York, USA: Van Nostrand
Reinhold.
Lee, N. C. (1998). Blow Molding Design Guide. Munich, Germany: Hanser.
Marcmann, G., Verron, E., & Peseux, B. (2001). Finite element analysis of blow
molding and thermoforming using a dynamic explicit procedure. Polymer
Engineering & Science, 41(3), 426-439.
Rajasekaran, S., & Vijayalakshmi Pai, G. A. (2008). Neural Networks, Fuzzy Logic,
and Genetic Algorithms. New Delhi, India: Prentice-Hall.
Rasmussen, H. K., & Yu, K. (2008). On the burst of branched polymer melts during
inflation. Rheologica Acta, 47(2), 149-157.
Page 15 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Suvanjumrat, C., & Chaichanasiri, E. (2014). Finite element method for creep testing of
high density polyethylene lubricant oil bottles. Kasetsart Journal (Natural
Science), 48(3), 487-497.
Thusneyapan, S., & Rugsaj, R. (2016). Non-linear finite element analysis of time-
varying thickness and temperature during the extrusion blow molding process.
Journal of Research and Applications in Mechanical Engineering, 4(1), 12-24.
Yu, J. C., & Juang, J. Y. (2010). Design optimization of extrusion-blow-molded parts
using prediction-reliability-guided search of evolving network modelling.
Journal of Applied Polymer Science, 117, 222-234.
Page 16 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Figure 1 The parison thickness optimization flow chart.
Figure 2 (a) The 1.0 liter lubricant oil bottle, (b) The samples of parison for
blowing the 1.0 liter lubricant oil bottle.
Figure 3 The height vs. the average wall thickness of: (a) parison and (b) lubricant
oil bottles.
Figure 4 The 3D model of 1.0 liter lubricant oil bottles in: (a) an isometric view, (b)
the detailed dimension, and (c) the finite element model of extrusion blow molding.
Figure 5 (a) The sequent image of simulation result of extrusion blow molding
process, (b) the height vs. the average wall thickness of lubricant oil bottles
between experiment and simulation.
Figure 6 (a) The sequent image of simulation result of cross-section shape bottle
extrusion blow molding, (b) the column line position vs. the average wall thickness
of lubricant oil bottles between experiment and simulation.
Figure 7 The column line position vs. the wall thickness of lubricant oil bottles
between ANN and simulation at bottle height of 115.00 mm.
Figure 8 The samples of lubricant oil bottle with its cross-section (left) and parison
for blowing (right) of: (a) the A-shape and (b) the B-shape lubricant oil bottle.
Page 17 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Figure 9 The column line position vs. the wall thickness of: (a) A-shape lubricant
and (b) B-shape lubricant oil bottle between ANN and experiment at bottle height
of 112.5 mm and 118.0 mm, respectively.
Figure 10 The sequent image of simulation result of cross-section shape bottle
extrusion blow molding process using the suitable parison thickness of GA.
Page 18 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Table 1 The suitable parison thicknesses for blowing the uniform thickness of
cross-section bottle.
Table 2 The final thicknesses of bottle at the bottle height of 115.00 mm of the FEA
and ANN by specifying the suitable parison thickness using GA.
Page 19 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Figure 1 Theparison thickness optimization flow chart.
Fitness function Evaluation
Selection
Crossover
Mutation
Satisfied?
Initial parison thickness Final bottle thickness
Parison thickness vs height relationship function
�� = �ℎ + ��
Optimized die shape
FEA
ANN
GA
Optimized parison thickness
Input
Output
Targeted uniform bottle thickness
Page 20 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
(a) (b)
Figure 2 (a) The one-liter lubricant oil bottle, (b) The samples of parison for
blowing the one-liter lubricant oil bottle.
(a) (b)
Figure 3 The height vs. the average wall thickness of: (a) parison and (b) lubricant
oil bottles.
Page 21 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Figure 4 The3D model of one-liter lubricant oil bottles in: (a) an isometric view, (b)
the detailed dimension, and (c) the finite element model of extrusion blow molding.
Parison
Mold B
Mold A
235.5
mm
58 mm
110 mm
(a) (b) (c)
Mold A
Parison
Page 22 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Thickness (mm)
t = 0 sec t = 0.2 sec t = 0.4 sec t = 2 sec t = 4 sec t = 20 sec
(a)
(b)
Figure 5 (a) The sequent image of simulation result of extrusion blow molding
process, (b) the height vs. the average wall thickness of lubricant oil bottles
between experiment and simulation.
Page 23 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Strain
t = 0 sec t = 0.2 sec t = 0.4 sec
t = 0.6 sec t = 1 sec t = 2 sec
(a)
(b)
Figure 6 (a) The sequent image of simulation result of cross-section shape bottle
extrusion blow molding, (b) the column line position vs. the average wall thickness
of lubricant oil bottles between experiment and simulation.
15°
45° 75° 105° 135°
165°
15°
45° 75° 105° 135°
165°
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
Page 24 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Figure 7 The column line position vs. the wall thickness of lubricant oil bottles
between ANN and simulation at bottle height of 115.00 mm.
15°
45° 75° 105° 135°
165°
15°
45° 75° 105° 135°
165°
Page 25 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
(a)
(b)
Figure 8 The samples of lubricant oil bottle with its cross-section (left) and parison
for blowing (right) of: (a) the A-shape and (b) the B-shape lubricant oil bottle.
Page 26 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
(a) (b)
Figure 9 The column line position vs. the wall thickness of: (a) A-shape lubricant
and (b) B-shape lubricant oil bottle between ANN and experiment at bottle height
of 112.5 mm and 118.0 mm, respectively.
Strain
t = 0 sec t = 0.2 sec t = 0.4 sec
t = 0.6 sec t = 1 sec t = 2 sec
Figure 10 The sequent image of simulation result of cross-section shape bottle
extrusion blow molding process using the suitable parison thickness of GA.
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
15°
45° 75° 105° 135°
165°
15°
45° 75° 105° 135°
165°
15°
45° 75° 105° 135°
165°
15°
45° 75° 105° 135°
165°
Page 27 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
For Review Only
Column A
(15 deg)
B
(45 deg)
C
(75 deg)
D
(105 deg)
E
(135 deg)
F
(165 deg)
Thickness
(mm) 2.86 2.69 2.26 1.98 2.89 2.83
Table 1 The suitable parison thicknesses for blowing the uniform thickness of
cross-section bottle.
Method
Bottle Column Thickness (mm) Average
Thickness
(mm) A
(15 deg)
B
(45 deg)
C
(75 deg)
D
(105 deg)
E
(135 deg)
F
(165 deg)
FEM 1.365 1.362 1.298 1.315 1.319 1.475 1.356
ANN 1.200 1.085 1.181 1.197 1.075 1.202 1.157
Different
Thickness
(mm)
0.165 0.277 0.117 0.118 0.244 0.273 0.199
Table 2 The final thicknesses of bottle at the bottle height of 115.00 mm of the FEA
and ANN by specifying the suitable parison thickness using GA.
Page 28 of 28
For Proof Read only
Songklanakarin Journal of Science and Technology SJST-2017-0497.R1 Suvanjumrat
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960