University of South Australia

PROCESS OPTIMISATION OF INJECTION

MOULDING USING COMPUTER SIMULATION

Research Proposal

Submitted by: Kyaw Lin

---

Abstract

Determining optimal values of process control factors is critical in injection moulding process

because of their influences on product quality, productivity and cost of production. In the past,

many researchers exploited traditional and artificial methods but the limitations of different

methods prevented to achieve optimal level of process design variables for multiple-input

multiple-output (MIMO) injection moulding process. To bridge the gap, this study aims to

develop a computer integrated optimisation system (CIOS). In this research, virtual reality

(VR) technology is employed, in the first phase, for simulation purpose and combination of

design of experiments (DOE) and SAPSO-based ANN method is used for optimising the

simulation results in order to achieve global optimal solution for the control parameters.

Apparently, the proposed approach is a new integration system that can help the users

determine optimal parameter settings to accomplish the MIMO injection moulding process

with competitive benefits of cost and production efficiency.

ii

---

Table of contents

Page

Title

Page

i

Abstract

ii

Figures

v

Tables

v

Symbols

v

Acronyms

vi

1. Introduction

1

1.1.

Project

background

1

1.2. Project aims and problem statement

1

1.3.

Scope

of

the

project

2

1.4.

Hypotheses

2

2.

Literature Review and Discussions

3

2.1. Review on work about process control variables 3

2.2. Review on previous work about process optimisation 4

2.3.

Review on optimisation techniques used for injection moulding 6

2.3.1.

Computer simulation 6

2.3.2.

Integration of computer simulation and optimisation algorithms 8

2.4.

Conclusion and discussions 9

3. Methodology

10

3.1. Sampling control factors 10

3.2. Data collection and data analysis 11

3.3. Framework of proposed approach 12

3.4. Virtual Simulation Model (VSM) 13

iii

---

3.4.1. Virtual environment 14

3.4.2. Pre-processing 14

3.4.3. Post-processing 15

3.4.4. Application of VSM 15

3.5. Coupled Optimisation Model (COM) 16

3.5.1. DOE 16

3.5.2. SAPSO-based ANN 18

4. Requirements of the project

20

4.1. Timeframe 20

4.2. Supervision 20

4.3. Facilities 21

4.4. Estimated cost 21

5. Significance of the project

21

6.

Conclusion

and

future

work

21

7. References

23

Appendix A

Table of quality response comparison of different approach

27

Appendix B

Alias Relationship for Fractional 25 Designs 28

Appendix C

25-1 Design

29

Appendix D

Timeframe of the project 30

iv

---

Figures

Figure.1.

Flow diagram of proposed CIOS 13

Figure.2.

Structure of VSM 14

Figure.3.

Flow diagram of COM 16

Figure.4.

SAPSO logical sequence 20

Tables

Table.1.

Process control factors and levels 11

Symbols

Xa1, Xa2 = variables of injection velocity level 1 and 2

Xb1, Xb2 = variables of melt temperature level 1 and 2

Xc1, Xc2 = variables of mould temperature level 1 and 2

Xd1, Xd2 = variables of packing pressure level 1 and 2

Xe1, Xe2 = variables of gate location level 1 and 2

E1, E2 = error term of level 1 and 2

k = the number of factors

nl = the experiment level

m = the number of replicates

n

g = the number of groups (levels),

nj =

the number of individuals in the

j

th group,

xij =

the

i

th individual in the

j

th group,

v

---

x

=

the mean of the

j

th group,

x

′ = the grand mean value.

yi

= expected output.

oij

= the predicted output of the network.

no

= number of output nodes.

np

= number of training set samples.

w = inertia weight.

v = velocity of the particle.

c1, c2 = acceleration constants.

r1, r2 = random numbers in the range [0, 1].

xi(t) = the current position of a particular particle.

pid = the best one of the solution that the current particle has reached.

pgd = the best one of the solutions that all particles have reached.

prob

= pre ­ determined probability .

temp

= temperature.

Itemp

= initial temperature (constant).

Acronyms

MIMO - multiple-input multiple-output

CIOS - computer integrated optimisation system

VR - virtual reality

DOE - design of experiments

vi

---

ANNs - Artificial Neural Networks

BPNN - Back Propagation Neural Network

GRNN - General Regression Neural Network

GA - Genetic Algorithm

DFP - Davidon-Fletcher-Powell

SA - Simulated Annealing

DEA - Data Envelopment Analysis

MRSN Ratio - Multi-Response

Signal to Noise Ratio

ANP - Analytic Network Process

PSO - particle swarm optimisation

FDM - Finite Difference Method

FEM - Finite Element Method

FVM - Finite Volume Method

BEM - Boundary Element Method

CPU - Central Processing Unit

CVFEM - Control Volume Finite Element Method

EMM - Equivalent Medium Method

CV - Control Volume

FE - Finite Element

VIMS - Virtual Injection Moulding System

VR - Virtual Reality

FEA - Finite Element Analysis

MS - Motion Simulation

SV - Scientific Visualisation

FV - Finite Volume

vii

---

RSM - Response Surface Methodology

CFD - Computational Fluid Dynamics

MPC - Model Predictive Control

AMD - absolute mean deviation

VSM - Virtual Simulation Model

CAD - Computer Aided Design

ML - Model Loader

MS - Moulding Simulator

MD - Motion Definer

COM - Coupled Optimisation Model

ANOVA - analysis of variance

UniSA - University of South Australia

LTU - Language and Teaching Unit

viii

---

PROCESS OPTIMISATION OF INJECTION MOULDING USING

COMPUTER SIMULATION

1. Introduction

1.1. Project background

Injection moulding is one of the most important processes for manufacturing commercial

items due to its advantages such as short product cycles and easily moulded complicated

shapes. It can be basically divided into four stages: plasticisation, filling, packing and cooling,

and ejection (Rosato & DV Rosato, 1995). Of four stages, filling phase includes the process

parameters that influence on the product quality as well as the productivity. The high demand

of three dimensional parts for new application areas in different industries has promoted the

development of its efficiency. Previously, trial-and-error approach based on the engineer′s

experience and intuition was employed to identify the significant parameter setting. However,

it is costly, time-consuming and inappropriate for complex manufacturing processes (Lam,

Deng & Au 2004). Manzione (ed. 1987) suggested that a key to improve the capacity of

handling important aspects of injection moulding process is the utilisation of computer

simulation. However, there remains a gap between simulation results and optimisation

objectives owing to the complexity and dynamic interaction of the injection moulding process.

Moreover, the issues regarding materials, product design and quality, production time, and

cost effectiveness make it more complicated. Accordingly, the requirement of more flexible

optimisation method draws attention of many researchers. The ideal method is to check and

measure the effects of process parameters in the real production line, but it has many

disadvantages such as time-consuming, high-cost and inaccuracy (Bikas, Pantelelis &

Kanarachos 2002, p.112). Therefore, in most cases, optimisation algorithms are integrated

with computer simulation to bridge the gap between simulation and reality.

1.2. Project aims and problem statement

The purpose of this research is to develop a new sophisticated computer integration system to

optimise the injection moulding process using computer simulation and optimisation

algorithms. In dealing with this project, the study will focus on the three sub-problems:

1

---

i.

The first sub-problem is to determine the process design variables that are dominant

in the injection moulding process.

ii.

The second sub-problem is to decide on the optimisation algorithms that are able to

quickly search the best results for the moulding process performance.

iii.

The third sub-problem is to discover the most efficient computer simulation package

for injection moulding.

1.3. Scope of the project

To eliminate irrelevancies to the identified objective and the research problem, the scope of

the project is clearly stated as follows:

i.

The study will be limited to plastic injection moulding process for most of

manufacturing enterprises use plastic as a major material for both household and

industrial products.

ii.

The study will not attempt to optimise all stages of the injection moulding process but

filling stage as it is critical.

iii.

The study will be limited to the dominant control process design variables of the

filling phase.

iv.

The study will not evaluate the economic factors of the injection moulding process.

v.

The study will not appraise simulation software used in other industries.

1.4. Hypotheses

i.

The appropriate combination of process control factors influence on the production

efficiency and product quality.

ii.

The preferred simulation package can facilitate the optimisation purpose, giving more

flexibility in understanding and decision making of the injection moulding process.

2

---

iii.

The optimisation algorithms designated are able to effectively search the global

optimum solution for the entire process.

2. Literature Review and Discussions

In the literature, a large number of papers regarding optimisation of injection moulding

process can be found. The word "optimisation" in different papers means making an

improvement of process factors affecting the product quality and productivity within the

feasible limits that satisfy all specified constraints. These factors include design variables

relating with process parameter settings and process design. Besides, the use of optimisation

methods and computer simulations plays an important role for optimising the plastic injection

moulding process.

2.1. Review on work about process control variables

A number of researches have been carried out with different aspects of moulding quality to

optimise the process parameters that comprise not only physical and chemical parameters

such as melt temperature, mould temperature, injection velocity, injection pressure, and

packing pressure, but also time parameters such as injection time, packing time and cooling

time (Chen et al. 2007; Kurtaran & Erzurumlu 2006). In the past studies, different

combination of process parameters had been used. Wu and Liang (2005) used six control

process parameters: melt temperature, mould temperature, packing pressure, injection

velocity, injection acceleration and packing time to examine the effects of parameters on the

width of weld line of an injection moulded plastic part. Chen, Wang, Fu, & Chen (2008)

employed four process control parameters: Injection Time, Velocity Pressure Switch Position,

Packing Pressure, and Injection Velocity to identify the optimum parameter setting for the

moulded plastic push-button housing piece under consideration of single quality response:

product weight. Chiang and Chang (2006) applied four process factors: melt temperature,

mould temperature, injection pressure, and injection time to determine the optimal process

parameter for a thin-shell plastic part with multiple quality characters. Zhou, & Turng (2007)

exploited seven process parameters: melt temperature, mould temperature, packing pressure,

injection time, packing time, cooling time, and velocity/pressure switch-over (V/P) by

volume to optimise the volume shrinkage for Plexiglass optical lens with different thickness

at the centre and outer rim.

3

---

In terms of process design optimization, a series of studies have been found, which focus on

different perceptive of the moulding process design: gate location, runner system, cooling

system, and part geometry. Subramanian, Tingyu, & Seng, (2005) attempted to optimise the

gate location in order to minimize the warpage of the injection moulded part, analysing the

distortion of the legs and reference pads in a plastic optical housing for a CD optical pickup.

Zhai, Lam, & Au (2006) tried to specify the optimum solutions of gate locations and runner

size of a multi gated moulding process, considering two critical affects: weld line and

warpage. Lee, & Lin (2006) targeted to determine the optimal runner and gating system

parameters for a multi-cavity injection mould in order to minimise the warp formation. Deng,

Zheng, & Lu (2008) aimed to optimise multi-class variables including process parameter,

gating system, runner system, cooling system and part geometry under consideration of

multi-response moulding qualities: part warpage, weld lines, air traps and so on. Reviewing

the above papers, one can conclude that appropriate combination of process parameters and

process design variables is a very important issue to achieve the objective quality

requirements.

2.2. Review on previous work about process optimisation

Concerning the optimisation methods used for the injection moulding process, there are a

significant number of papers in the literature, some of which focus on single criterion while

another on multiple criteria. The techniques used for optimisation issue include traditional

methods and artificial intelligence methods. Previously, trial-and-error method was used to

determine the process parameter settings, depending on the experience and intuition of the

engineers. However, it has many drawbacks and is unsuitable for complex processes to get

the actual optimum results (Lam et al. 2004). Next, Taguchi′s parameter design method has

been used for experimental design and process improvement as a central one in many papers.

Liao et al. (2004) exploited L27 orthogonal array experimental tests based on Taguchi′s

method to optimise the process conditions of a thin-wall injection moulding of a cellular

phone cover made of amorphous PC/ABS resin plastics, taking into account the interaction

effects between process parameters, and quality targets: shrinkage and warpage. Oktem,

Erzurumlu & Uzman (2007) utilised Taguchi optimization tool to determine plastic injection

moulding process parameters for thin shell parts. In this case, material property, part design,

and injection-moulding conditions were considered as the factors or variables needed to be

changed to minimize shrinkage. Nonetheless, Taguchi′s method is only able to find the best

4

---

combination of parameter level that includes discrete values so it falls short when the

parameter values are continuous (Chen et al. 2008). To deal with the continuous parameter

variables, Artificial Neural Networks (ANNs) has been introduced as an alternative means

(Chiu et al. 1997). Chen et al. (2009) claims that ANNs can map the relationship between

input factors and output responses and it has different sub-categories such as Back

Propagation Neural Network (BPNN), General Regression Neural Network (GRNN).

However, it finds difficult to search the final optimal variables. Later on, the robust

optimisation method Genetic Algorithm (GA) has been extensively used for random

searching of the global optimum value in large dimensional space without being trapped in

the local optimum (Tseng 2006; Shen, Wang & Li 2007). Lam, Den & Au (2006) argues that

GA is opportunistic but not deterministic alone. Therefore, many researchers combine

different nature of optimisation methods to approach the optimal design and process

conditions. Chen et al. (2008) integrated Taguchi′s parameter design method, BPNN, and

Davidon-Fletcher-Powell (DFP) method to optimize the process parameter settings of plastic

injection moulding, considering single product quality (weight). Furthermore, Chen et al.

(2009) combined Taguchi′s parameter design method, BPNN, GA and engineering

optimization concepts to optimize the process parameters, for an experiment on a standard

plastic piece, under multiple-input multiple-out (MIMO) consideration. They claim that their

approach can effectively help engineers determine optimal process parameter settings and

achieve competitive advantages of product quality and the costs.

On the other hand, some researchers used different optimisation methods for their particular

objectives. Su & Chang (2000) proposed the combination of Neural Network (NN) and

Simulated Annealing (SA) to optimise the parameter design of the injection moulding process.

Loera et al. (2008) used Data Envelopment Analysis (DEA) optimisation strategy for a

thermoplastic injection moulding, considering multiple criteria: design and process variables

to meet several performances. Upon the application of their approach, they discussed the

optimum parameter settings for the moulding of rear automotive lamps to control the part

dimension and surface properties (aesthetic) using seven control process inputs. Moreover,

Park & Ahn (2004) used DOE to reduce the cooling time and the injection pressure. This

research proved that DOE is useful to discover the cause and effect relationship between the

inputs and outputs of a process, and to determine the optimal process parameters with fewer

testing trials. Furthermore, Deng (2008) coupled Multi-Response

Signal to Noise Ratio

(MRSN Ratio) with Analytic Network Process (ANP) to optimize the process parameters of

5

---

multiple-response process in order to achieve higher production efficiency. MRSN is one of

reformed Taguchi′s methods for multiple-quality response process parameter optimization

and ANP is a systemic process that applies ratio scales to evaluate internal relationship of

dimensions, criterions, and alternatives. They used five control parameters: mould

temperature, pipe temperature, injection velocity, injection pressure and packing time to

achieve the targeted four quality responses: weight, length (dimensional), warpage &

shrinkage (surface property) for case study of a bottom cover of polypropylene Modem. In

addition, Da & Xiurun (2004) presented SAPSO-based ANN method using particle swarm

optimisation (PSO) with simulated annealing (SA) in order to achieve global optimum. Their

approach modelled the relationship between confining pressure, peak stress and

corresponding strain, and showed its flexibility in escaping local optimum. For that reason, it

can be concluded that proper selection and/or combination of different techniques to meet

quality criterion and production efficiency becomes one of the most important factors in

optimisation of injection moulding process.

2.3. Review on optimisation techniques used for injection moulding

Towards the optimisation of injection moulding process, development of computer

simulation makes it easier to analyse the process in the early design stage before real

implementation. As a result, it becomes essential tool in the injection moulding industry.

Quite a large number of different simulation software has been introduced in the last three

decades, and a number of applications have been published. While some researchers used

only computer simulation for optimisation purposes, others integrated simulation tools with

optimisation algorithms.

2.3.1. Computer simulation

Basically, computer simulation is built up with sequential development of mathematical

equation groups to analyse physical features of material inside the mould cavity. Kim and

Turng (2004) outlined the development process of computer­aided simulation technique from

traditional 2.5-D Hele-Shaw approach to 3-D simulation applications. Disadvantages and

limitations of the 2.5-Dimensional solution evidently pointed the necessity of building a new

method for moulding simulation. As a result, they introduced a typical procedure

implemented with four numerical methods: Finite Difference Method (FDM), Finite Element

6

---

Method (FEM), Finite Volume Method (FVM) and Boundary Element Method (BEM),

which are considered as key supplementary tools for fulfilling simulation objectives. FDM is

relatively effective and quite simple for solving partial and differential set of algebraic

equations. It is, however, difficult to apply this method to a highly complicated boundary.

Hence, the application of it is restricted to regular and simple domains. Conversely, FEM has

excellent flexibility in solving complex boundaries and irregular geometries. Nevertheless,

the global matrix system employed by this method requires large memory space and CPU

(Central Processing Unit) time to process the data. Meanwhile, BEM successfully tackles the

excessive effort of computation, an inherent weak point of most simulation algorithm, but it

is impaired when non-linear problems for the matrix system of algebraic equations are dense

and non-symmetric. Although FVM is a particular case of FDM, it can overcome the troubles

of FDM by subdividing physical domain into small control volumes. Therefore, the

combination of this method and FEM can create an application of simulation algorithm to

solve complicated domain and complex geometries but less CPU time and computational

effort.

Alternatively, the two other methods, Control Volume Finite Element Method (CVFEM) and

Equivalent Medium Method (EMM) were introduced by Dong (2005) to solve the simulation

problem in a vacuum assisted moulding process. CVFEM bases on the matrix of permeability

tensors and the pressure gradients of the three directions combined with Gauss′s theorem and

Darcy′s law to generate simulation data. Similarly, EMM solution is generated by

constructing a matrix of elements based on 3-D mesh generation of the model. This method

requires fewer elements than that of traditional CVFEM but much complex mesh generation

algorithm. To support and fulfil for the EMM method, a mesh generation algorithm is also

organisationally produced and presented in the same article. EMM enhances the reduction of

time consumption for computational activities with 85% time savings compared to that of

other traditional method (Dong, C 2006, p.1212).

Consequently, based on the steps of a basic simulation procedure, Shojaei et al. (2003)

presented two methods for tracking simulation algorithm, exploiting Quasi-Steady State and

Nodal Partial Saturation approach. Quasi-Steady State approach is constructed to generate the

algorithm for isothermal models while Nodal Partial Saturation approximation is for non-

isothermal solutions. More than algorithms, Quasi-Steady State and Nodal Partial Saturation

approach are intensively developed into higher applications of computer code with given

7

---

names RTMS and RAPFIL, respectively. Both of those numerical schemes are, then,

combined together and implemented in FORTRAN 77. Moreover, Quasi-Steady State and

Nodal Partial Saturation algorithm are suitable for personal computer but can not reach high

accuracy. There is an error up to 1.52% in filling time compared to practical experiment

(Shojaei et al. 2003).

Additionally, Shojaei (2006) studied numerical simulation of filling process of moulding in

full three-dimensional domain using CVFEM and developed numerical algorithm to progress

the flow front based on a quasi-steady state approach. The numerical algorithm he presented

was coded and the resulting computer code was used to predict the necessary parameters of

the filling stage such as flow progression, pressure distribution and mould clamping force for

the mould containing both the single and multi-layer performs. Shen, Wang, & Li (2007)

integrated a hybrid FE/ FD method with a Control Volume (CV) to simulate the filling stage

of injection molding. Also, they developed a Visual C++ program for compressible flow

analysis of the filling stage, generalising the Hele­Shaw flow of a compressible viscous fluid

under non-isothermal conditions with consideration of the effects of compressibility and

phase change. Llado´ & Sa´nchez (2008) employed Finite Element (FE) Moldflow numerical

simulation software to determine the cause of blush that appears around the gate in the

injection of PVC fittings and to predict the relationship between blush and injection rate and

melt temperature. Zhou, Shi, & Ma (2009) proposed a Virtual Injection Moulding System

(VIMS) integrated with Virtual Reality (VR), Finite Element Analysis (FEA), Motion

Simulation (MS) and Scientific Visualisation (SV). VR technology can provide the

imperative view of the model and process, offering interactive (what-if) studies. Obviously,

their research shows that VIMS can highlight the problematic area and influences of the

design variables for the target product requirements. Conversely, application of computer

simulation alone cannot solve the very complicated manufacturing cases because of its nature:

only to run the program not to solve.

2.3.2. Integration of computer simulation and optimisation algorithms

Subsequently, in most cases, several researchers integrate the computer simulation with

optimisation algorithms. Chang & Yang (2001) proposed a collocated implicit Finite Volume

(FV) approach with the SIMPLE segregated algorithm to solve the three-dimensional

injection mould filling problems. The numerical model they described dealt with three-

8

---

dimensional isothermal flow of incompressible, high-viscous Newtonian fluids with moving

interfaces. Turng and Peic (2002) integrated a CAE simulation tool (C-MOLD) with a

process optimization program (OPTIMUS) to determine the optimal design and process

variables for injection moulding. A polynomial type of Response Surface Methodology

(RSM) was used in OPTIMUS to determine the relationship between the process design

objectives and the settings of design and process parameters. Gerber, Dubay, & Healy (2006)

combined Computational Fluid Dynamics (CFD) with Model Predictive Control (MPC) to

improve and control polymer melt temperature within the barrel and nozzle region,

considering a three-heater zone interaction. They used CFX-TASCflow for the CFD

simulations model that acts as a slave to the MPC optimisation algorithm. Deng, Zheng & Lu

(2008) integrated the CAE (Moldflow) software with Particle Swarm Optimisation (PSO)

algorithm to optimise the multi-class design variables of the injection moulding such as part

thickness, process parameters and gate location, intending to achieve the multi-response

quality requirements: warpage, weld lines, air-traps and so on. As PSO was designed to

search Pareto optimality (Loera et al. 2008), their experiment on plastic AC power outlet got

the complete optimality, without normalisation of objective functions and specification of the

weight factors. The different levels of successes in the above reviewed papers show the

strength of integrating different techniques: computational, numerical, mathematical to

approach the optimum process design variables.

2.4. Conclusion and discussions

The literature shows that optimisation of process control parameter is important to achieve

product quality requirements in order to improve the efficiency of plastic injection moulding

process. Although there are a number of studies that focused on the impact of process

parameters, few of them focused on combination of process and design factors. Moreover,

most papers used only optimisation methods in searching optimal solution. The literature

suggests that few researches carried out with integration of computer simulation and

optimisation methods are more efficient to attain the global optimum. Therefore, there is a

need to develop this assessment that can help injection moulding industry worldwide from

certain aspects of the production efficiency. For that reason, this research will effort to

develop an integration system that can optimise the impacts of process control factors, and

the research findings are likely to be useful for further studies on the same issue.

9

---

3. Methodology

This project is attempting to develop a computer integrated optimisation system (CIOS) for

injection moulding process, targeting high product quality as well as production efficiency.

The CIOS integrates computer simulation with optimisation methods using C++

programming language to achieve global optimal solution of the control parameter settings

for MIMO plastic injection moulding.

3.1. Sampling control factors

In sampling the data for experiment, Stratified Sampling Design (Leedy & Ormrod 2005) will

be used, setting certain criteria among the population of each group. In this study, two levels

of data from each group will be used as a sample that represents all the characteristics of

population of each group. To avoid bias, the sampling procedure is carefully planned for

control factors as follows:

To demonstrate the effectiveness of the proposed approach, four criteria for quality responses

are selected as the milestones of the project. The first two dimensional properties are length

and weight, then the latter are warpage and shrinkage belonged to surface properties. The

quality responses of length and weight have "nominal the best" characteristic. Length has a

specification with limited tolerance whereas weight does not have a certain target value. For

the time being, the quality responses of warpage and shrinkage have "the smaller the better"

characteristic. After setting the objectives for optimisation purpose, process control factors

are determined as the input data. From the literature review, four process factors dominant in

the injection moulding process and influential to the quality responses are selected. They are

injection velocity that influences the mechanical and dimensional stability of the product,

melt temperature that has strong relation with the polymer viscosity, filling velocity, filling

pressure, and cavity pressure-time profile, mould temperature that is critical for cycle time

and warpage formation, and packing pressure that affects the product shrinkage and warpage.

While proper gate location can improve the warpage and production efficiency, it is taken

into account as a process design variable. The formulation of process control factors with two

levels can be seen in Table.1.

10

---

Table.1.Process control factors and levels

Factor

Level 1

Level 2

A: Injection velocity (mm/sec)

Xa1

Xa2

B: Melt temperature (°C) Xb1

Xb2

C: Mould temperature (°C) Xc1

Xc2

D: Packing pressure (MPa)

Xd1

Xd2

E: Gate location

Xe1

Xe2

Error E1

E2

3.2. Data collection and data analysis

This research will be implemented through combination of qualitative and quantitative

approach in continue mode for system building and conducting experiment, respectively.

Secondary data will be used due to its benefits such as easy to collect, less time-consuming

for data collection, analysis and interpretation (Blaxter, Hughes & Tight 2001). The data

required for system phenomenon will be collected from the literature: technical reports,

recent and latest researches. As the proposed system uses new techniques, numerous form of

data will be collected and examined from various angles to discover the problems existing

within the phenomenon. To evaluate the effectiveness of the system and verify the validity of

the theories, a case study will be followed based on practical experiment.

The data from the experiment results will be collected using statistical spreadsheet via

longitudinal study. Then the sorted data will be set into tables and charts to analysis and

interpret the significance of data whereas cause and effect relationship of inputs and outputs

will be discovered by studying data correlation. Leedy and Ormrod′s (2005) randomised two-

factor design will be used for experiment design.

11

---

After collecting the data, this research will use absolute mean deviation (AMD) to compare

the performance of length quality. In the meantime, sample standard deviation will be used to

compare the performance of four quality responses and sample mean to compare the

performance of warpage and shrinkage quality response. The comparison of the quality

response results will be tabulated as shown in Appendix A.

Furthermore, data analysis will be followed up with null hypothesis test and t-test to confirm

the hypotheses previously set in the project.

3.3. Framework of proposed approach

The study pays high attention to the filling stage of the plastic injection moulding process,

considering multiple responses. The continuous interaction of the flow pattern, temperature

and pressure profiles as well as control parameters makes this stage the most complicated

phase out of four. Besides, the nature of plastic material is more viscous and compressible in

molten stage so the melt (fluid) and its flow will be considered as non-Newtonian

compressible fluid under non-isothermal flow, taking into account of the effects of heat and

pressure. After determining the objectives and constraints, for process optimisation, a new

integration system is proposed. In this approach, VR technique presented by Zhou, Shi, & Ma

(2009) will be used for simulation purpose to analyse the variables and highlight the

problematic area. Then, the statistical results from simulation will be imported to coupled

optimisation model that includes DOE, SA and PSO. The schematic diagram of the flow of

proposed approach can be seen in Figure.1.

12

---

Figure.1. Flow diagram of proposed CIOS

3.4. Virtual Simulation Model (VSM)

The main purpose of employing VSM is to facilitate visualisation and optimisation of the

injection moulding process based on VR technology. "VR is the use of computer generated

virtual environments and the associated hardware to provide the user with the illusion of

physical presence within that environment which allows the designer to `virtually

manufacture′ the product while designing it" (Ma et al. 2007, p. 1093). This system includes

three main parts: virtual environment, pre-processing and post-processing as shown in

Figure.2.

13

---

Figure.2. Structure of VSM (Zhou, Shi & Ma 2009, p. 299)

In application of this model, necessary information will be firstly supplied for creating

prototypes of plastic product and mould design, which are inputs of the virtual system and

will be done in commercial 3D CAD (Computer Aided Design) software. It can be seen that

CAD system is used as an external task to create models of the products and mould design.

3.4.1. Virtual environment

The virtual environment is used for visualisation of mono-mode and stereo-mode of

prototypes. The latter mode can provide 3D stereoscopic view by means of an emitter and a

CrystalEyes Workstation that is a wireless set of liquid crystal shutter eyewear for

Stereo3DTM imaging. The shutter eyewear generates a stereoscopic feeling by synchronizing

with the display device, showing the images to the left eye and right eye alternatively via a

switch. The former mode requires no emitter and glasses as it does not provide stereoscopic

display.

3.4.2. Pre-processing

With regard to the pre-processing, preparation work will be done in each component namely

Model Loader (ML), Moulding Simulator (MS) and Motion Definer (MD). ML is used to

read and manage the product part and mould design, and to display them in the virtual

environment. MS is applied for an interface of process control parameter setting and

14

---

presentation of simulation results based on FEA and CFD. MD is for grouping all mould

parts and generation of the movement pathways for moulding simulation.

3.4.3. Post-processing

In post-processing phase, FEA results visualiser is employed to validate the moulding process

in terms of numerical results for flow patterns, process and design parameters, and quality

related factors and so on. By analysing the results, the possible faults and inadequacies can be

evidently highlighted. Meanwhile, the effects of the control parameters such as melt

temperature, gate location on the mouldability and product quality can be studied by

changing them interactively. The motion visualiser is used to display the mould motion and

the machine operations. The moulding visualiser is to combine FEA results, moulding

machine motion and mould assembly so the moulding cycle can be viewed intuitively.

3.4.4. Application of VSM

Fundamental steps of the application of VSM are demonstrated as follows. The process starts

with creation of product prototype and mould design using CAD software. A tetrahedral

finite element mesh of the product is exported as STL files. ML reads the files and creates

models, and generates a scheme of the injection machine and mould design based on the

CAD models. While all mould parts are discretized by triangular finite elements, a hybrid

finite-element/finite-difference/control-volume numerical solution is used to simulate the

mould filling. Then the data of simulation are sorted. The numerical results can be evaluated

with FEA results visualiser; and melt flow during filling stage can be viewed. To simulate the

true process, the user must correctly define the processing conditions. In order to optimise the

moulding process parameters, the simulation results will be exported to the coupled

optimisation system.

In this project, a mathematical and numerical model will be used to simulate the filling stage

of injection moulding based on three-dimensional model. While this model is dealing with

three-dimensional flow, 3D control volume method will be employed to track the flow front.

To calculate the flow pattern, pressure and temperature profiles, the finite-element/finite-

difference/control-volume method will be used to solve the momentum equation, continuity

equation and energy equation.

15

---

3.5. Coupled Optimisation Model (COM)

This system couples two types of optimisation algorithms that include DOE, and SAPSO-

based ANN, which have different natures in searching the optimal solution. The structure of

COM can be seen in Figure.3.

Simulation Results

DOE

Arranged & Optimised Results

SAPSO-based ANN

Global Optimum

Figure.3. Flow diagram of COM

3.5.1. DOE

To identify the process condition and product components that influence product quality and

productivity, DOE will be used in this study to find cause and effect relationship between the

output and input experimental factors in a process. The DOE procedure includes four steps as

following (Park & Ahn 2004):

·

Project

definition:

identify the objective of the project and find the scope of the

problem.

·

Screening

: reduction of the number of variables by identifying the key variables that

affect product quality.

·

Optimization

: determination of the optimal values for various experimental factors.

·

Verification

: performing a follow-up experiment at the predicted best processing

conditions to confirm the optimization results.

16

---

DOE approach can be divided into a full factorial design and a fractional factorial design.

- Full factorial design:

k

N

=

m

(

n

)

full

l

- Fractional factorial design:

1

N

m

(

n

) -

fractional

=

k

l

In this study, k=5, nl=2 Nfull=25=32. In this five-factor experiment, there are 5 main effects

(A,B,C,D, and E), 10 two-factor interactions (AB, AC, AD,...), 10 three-factor interactions

(ABC, ABD, ADC,...), 5 four-factor interactions (ABCD, ACDE,...) and 1 five-factor

interaction. However, higher ­ order interaction effects (interaction effects involving three or

more factors) are very seldom significant. Therefore, the fractional experiment is designed to

be able to consider main effects and two-way interactions. It is efficiently used in the

screening DOE procedures when there are a large number of factors.

Alias Relationship for Fractional 25 Designs can be seen in Appendix B and 25-1 Design in

Appendix C (Yang & El-Haik 2003).

The results of DOE can be analyzed by evaluating the main effects for all factors and

interactions for all of the two-way combinations. The main effect of the

j

th factor (

Ej

) and the

interaction between the

j

th and

k

th factor

Ijk

) are calculated as follows (Park & Ahn 2004):

Nl R

ij

i

i

=

E

1

j

=

NRi

i

=1

N l

(

l

)

R

ij ik

i

i

=

I

1

jk

=

NRi

i

=1

17

---

Where

N

is the total number of experiments,

Ri

the response variable for the

i

th combination,

lij

equals -1 for the lower level, +1 to the upper level of the

j

th factor and is subjected to

orthogonal condition as follows (Yang & El-Haik 2003):

Nl

0 (j=1,2,...,N)

ij

=

i

=1

The results of the experimental designs are analysed by using analysis of variance (ANOVA).

The ANOVA is utilised to investigate the relationship between a response variable and one or

more independent variables. If the difference between the averages of the levels is greater

than what could reasonably be expected from the variation that occurs within the level, it can

be determined. The test for ANOVA generally uses a ratio of:

n

average

(

SS

g

)

1

SS

between

between

F

=

=

(

n

- )1

j

average

(

SS

)

n

-1 =1

SS

within

g

j

within

where SSbetween and SSwithin denote the square sum variation between groups, and the square

sum variation within a group, respectively. Each term can be expressed as follows:

ng

ng ni

2

SS

=

n

(

x

-

x

′ )

2

SS

= (

x

-

x

)

between

j

j

within

ij

j

j

1

=

,

j

1

=

i

1

=

When applying the DOE, the best factor level settings and optimal output performance level

are identified. On the other hand, to identify the optimal process condition, we can use

Minitab software or Excel to analyse the DOE.

3.5.2. SAPSO-based ANN

At the latter phase of optimisation process, the data proceeded by DOE will be solved by the

application of SAPSO-based ANN method. This is a hybrid solution initiated by the three

original algorithms SA, PSO and ANN. Accordingly, all standard steps of a sample PSO

algorithm, which is known as one of the most productive computational intelligence

technique, presented by Da and Xiurun (2004) must be followed:

18

---

1. Initialise a number of particles with positions chosen randomly and velocities on d

dimensions in a particular space.

2. For each particle, specify the fitness function to evaluate the desired optimisation of d

variables:

np no

(

y

-

o

2

i

ij

)

fitness

i

=1

i

=

= -

1

nonp

The larger the fitness, the better set of weights and thresholds, which can be defiend as

follow:

n

n

f

w x

1/ 1 exp

w x

i i

- =

+

-

i i

-

i

=1

i

=1

3. Compare the finess evaluation of each particle with particle′s pid, if the current value

is better than pid, then set the value and location of pid equal to those of the current

value.

4. Compare the finess evaluation with the all population′s pid, if the current value is

better than pid, then set pid to the array of index and value of the current particle.

5. Change the velocity and positon of the particle according to the two following

equations:

vi(t+1)=wvi(t)+c1r1(pid ­ xi(t)) + c2r2(pgd ­ xi(t))

xi(t+1)=xi(t) + vi(t+1)

6. return to step 2 and repeat until a pre ­ specified fitness function or a maximum

number of iteration is met.

19

---

Concurrently, SA which is another intelligent algorithm, integrated into PSO method to

explore the global optimum. Particularly, the flow of process is shown as follow (Da &

Xiurun 2004):

Accept p

id = pgd with

p

No

id > pgd

pgd

-

pid

prob

= 1- exp-

Yes

temp

Accept pid = pgd

with

prob

= 1

Figure.4. SAPSO logical sequence

The SAPSO method above is then employed to train a three ­ layer artificial neural network

to generate a comprehensive SAPSO-based ANN method.

Overall, it takes approximately three minutes to complete a sample circle of SAPSO-based

ANN experiment with the computer equipped a configuration as (Da & Xiurun 2004).

4. Requirements of the project

4.1. Timeframe

This research has been planned to carry out within 1 year by the nominated 4 students at the

University of South Australia (UniSA). The detail timeframe for this study can be seen in

Appendix D.

4.2. Supervision

The supervision of this project will be acquired from 2 supervisors; one is Dr. Ke Xing and

another one is Prof. Lee Luong from the School of advanced Manufacturing and Mechanical

Engineering, UniSA.

20

---

4.3. Facilities

To perform the study the required facilities will be provided by UniSA, including computer

system installed with all required software, permission to have access to database for further

literature and workshop during the experiment period, support for statistical analysis from

School of Mathematics or Language and Teaching Unit (LTU) and so forth.

4.4. Estimated cost

The total cost for this project is estimated at $10,000 for 2 supervisors, trips to the industries

and other expenses.

5. Significance of the project

This research aims to produce a comprehensive package for optimising the injection

moulding process. Chen et al. (2009) claim that determining the optimal process setting is

important because of potential influences on process efficiency, quality and the cost of

products. The proposed approach couples computer simulations with different optimisation

methods so it will be an effective tool to verify the optimal parameters for the entire process

under multi-response considerations. Therefore, this study will propose a new efficient

methodology with high flexibility for injection moulding to replace high-cost and time-

consuming approaches. In addition, the use of this proposed integration system would

provide detail information about the whole operation process in training.

6. Conclusion and future work

This research integrates the latest optimisation techniques: VR, DOE and SAPSO-based

ANN to develop the most effective optimisation system for injection moulding process. As a

result, the final outcome can be a fruitful comprehensive system applicable for a wide variety

of products with high complexity 3D design in moulding industry. The experiment results are

then compared with other approaches for official testification and self-evaluation before real-

world application or manufacturing environment. However, as VR technique is still under

development stage, this study may challenge with shortcomings and lack of technical support

in building the system. Besides, the system is not included collision detection to find the

21

---

possible defects of the mould design. In the fast-paced of industry development, there is no

limit of improvement in any area. Therefore, it is necessary to do further researches that are

more intensive to response the highly changing global market place. Future work in this

study area should take into account cost estimation of moulding process and waste reduction

to carry out the optimisation process in an economical pathway.

22

---

7. References

Bikas, A, Nikos Pantelelis, N & Kanarachos, A 2002, `Computational tools for the optimal

design of the injection moulding process′,

Journal of Materials Processing Technology

, vol.

122, no. 1, pp. 112-126.

Blaxter, L, Hughes, C & Tight, M 2001,

How to research

, 2nd edn, Open University press,

Philadelphia, USA.

Chang, RY & Yang, WH 2001, `Numerical simulation of mold filling in injection molding

using a three-dimensional finite volume approach′,

International Journal for Numerical in

Fluids

, vol. 37, pp. 125-148.

Chen, WC, Fu, GL, Tai, PH, Deng, WJ & Fan, YC 2007, `ANN and GA-based process

parameter optimisation for MIMO plastic injection moulding′,

Proceedings of the 6th

International Conference

, Machine Learning and Cybernetics, Hong Kong, pp. 1909-1917.

Chen, WC, Wang, MW, Fu, GL & Chen, CT 2008, `Optimization of plastic injection

molding process via Taguchi′s parameter design method, BPNN, AND DFP′,

Proceedings of

the 7th International Conference,

Machine Learning and Cybernetics, Kunming, pp. 3315-

3321.

Chen, WC, Fu, GL, Tai, PH & Deng, WJ 2009, `Process parameter optimization for MIMO

plastic injection molding via soft computing′,

Expert Systems with Applications

, vol. 36, no.

2, pp. 1114-1122.

Chiang, KT & Chang, FP 2006, `Application of grey-fuzzy logic on the optimal process

design of an injection-molded part with a thin shell feature′,

International Communications

in Heat and Mass Transfer

, vol. 33, pp. 94-101.

Chiu, CC, Su, CT, Yang, GH, Huang, JS, Chen, SC, & Chen, NT 1997, `Selection of optimal

parameters in gas assisted injection molding using a neural network model and the Taguchi

method′,

International Journal of Quality Science

, vol. 2, no. 2, pp. 106-120.

Deng, WJ 2008, `Integrating MRSN Ratio and ANP to Optimize Process Parameter of

Multiple-response Injection Molding Process′,

IEEE Service Operations and Logistics, and

Informatics

, vol. 2, pp. 2736-2740.

23

---

Deng, YM, Zheng, D & Lu, XJ 2008, `Injection moulding optimisation of multi-class design

variables using a PSO algorithm′,

International Journal of Advanced Manufacturing

Technology

, vol. 39, no. 7-8, pp. 690-698.

Dong, C 2006, `An equivalent medium method for the vacuum assisted resin transfer

molding process simulation′,

Journal of composite materials

, vol. 40, no. 13, pp.1193 ­ 1213.

Gerber, AG, Dubay, R & Healy, A 2006, `CFD-based predictive control of melt temperature

in plastic injection molding′,

Applied Mathematical Modelling

, vol. 30, no. 9, pp. 884-903.

Kim, S, Turng, L 2004, `Developments of three-dimensional computer-aided engineering

simulation for injection moulding′,

Modelling and simulation in materials science and

engineering

, vol. 12, no. 3, pp. S151 ­ S173.

Kurtaran, H & Erzurumlu, T 2006, `Efficient warpage optimization of thin shell plastic parts

using response surface methodology and genetic algorithm′,

The International Journal of

Advanced Manufacturing Technology,

vol. 27, no. 5-6, pp. 468-472.

Lam, YC, Deng, YM & Au, CK 2006, `A GA/gradient hybrid approach for injection

moulding conditions optimization′,

Engineering with Computers

, vol. 21, no. 3, pp. 193-202.

Lam, YC, Zhai, LY, Tai, K & Fok, SC 2004, `An evolutionary approach for cooling system

optimization in plastic injection moulding′,

International Journal of Production Research

,

vol. 42, no.10, pp. 2047-2061.

Lee KS & Lin JC

2006, `Design of the runner and gating system parameters for a multi-

cavity injection mould using FEM and neural network′,

International Journal of Advanced

Manufacturing Technology

, vol. 27, pp. 1089-1096.

Leedy, PD & Ormrod, JE 2005,

Practical research,

Pearson Merrill Prentice Hall, 8th edn,

Columbus, Ohio, USA.

Liao, SJ, Chang, DY, Chen, HJ, Tsou, LS, Ho, JR, Yau, HT, Hsieh, WH, Wang, JT &

Su,

YC 2004, `Optimal Process Conditions of Shrinkage and Warpage of Thin-Wall Parts′,

Polymer Engineering and Science,

vol. 44, no. 5, pp. 917-928.

24

---

Llado´, J & Sa´nchez

,

B 2008, `Influence of injection parameters on the formation of blush in

injection moulding of PVC′,

Journal of Materials Processing Technology

, vol. 204, pp. 1-7.

Loera, VG, Castro, JM, Diaz, JM, Mondrago´n

,

OLC & Cabrera-R´ios, M 2008, `Setting the

Processing Parameters in Injection Molding Through Multiple-Criteria Optimization: A Case

Study′,

IEEE Transactions on Systems, Man, and Cybernetics-Part C: Applications and

Reviews,

vol. 38, no. 5, pp. 710-715.

Manzione, LT (ed.) 1987,

Applications of Computer Aided Engineering in Injection

Moulding,

Macmillan, New York.

Ozcelik, B & Erzurumlu, T 2006, `Comparison of the warpage optimization in the plastic

injection molding using ANOVA, neural network model and genetic algorithm′,

Journal of

Materials Processing Technology,

vol. 171, pp. 437-445.

Oktem, H, Erzurumlu, T & Uzman, I 2007, `Application of Taguchi optimization technique

in determining plastic injection molding process parameters for a thin-shell part′,

Material

and Design

, vol

.

28, no. 4, pp. 1271-1278.

Park, K and Ahn, JH 2004, `Design of experiment considering two-way interactions and its

application to injection molding processes with numerical analysis′,

Journal of Materials

Processing Technology

, vol. 146, no. 2, pp. 221-227.

Rosato, DV & DV Rosato 1995,

Injection Molding Handbook: The Complete Molding

Operation Technology, Performance, Economics, 2nd edn,

Chapman & Hall, Massachusetts.

Sercer, M, Godec, D, Bujanic, B 2007, `Application of Moldex3D for thin-wall injection

moulding simulation′,

AIP conference proceedings

, vol. 908, no. 1, pp. 1067 ­ 1072.

Shen, C, Wang, L & Li, Q 2007, `Optimization of injection molding process parameters

using combination of artificial neural network and genetic algorithm method′,

Journal of

Materials Processing Technology,

vol. 183, no. 2-3, pp. 412-418.

Shen C, Wang, L & Li Q 2007, `Numerical Simulation of Compressible Flow with Phase

Change of Filling Stage in Injection Molding′,

Journal of Reinforced Plastics and Composite,

vol. 26, no. 4, pp. 353-372.

25

---

Shojaei, A 2006, `Numerical simulation of three-dimensional flow and analysis of filling

process in compression resin transfer moulding′,

Composites Part A: Applied Science and

Manufacturing,

vol. 37, pp. 1434-1450.

Su, CT & Chang, HH 2000, `Optimization of parameter design: an intelligent approach using

neural network and simulated annealing′,

International Journal of Systems Science,

vol. 31,

no. 12, pp. 1543-1549.

Subramanian, NR, Tingyu, L & Seng, YA 2005, `Optimizing warpage analysis for an optical

housing′,

Mechatronics

, vol. 15, pp. 111-127.

Tseng, HY 2006, `Welding parameters optimization for economic design using neural

approximation and genetic algorithm′,

International Journal Advanced Manufacturing

Technology

, vol. 27, no. 9, pp. 897-901.

Turng, LS & Peic, M 2002, `Computer aided process and design optimization for injection

moulding′,

Proceedings of the Institution of Mechanical Engineers, Part B: Journal of

Engineering Manufacture,

vol. 216, no. 12, pp. 1523-1532.

Wu, CH & Liang, WJ 2005, `Effects of geometry and injection-molding parameters on weld-

line strength′,

Polymer Engineering and Science

, vol. 45, no. 7, pp. 1021-1030.

Yang, K & El-Haik, B 2003,

Design for six sigma: a roadmap for product development

,

McGraw-Hill, New York.

Zhai, M, Lam, YC & Au, CK 2006, `Runner sizing and weld line positioning for plastics

injection moulding with multiple gates′,

Engineering with Computers

, vol. 21, pp. 218-224.

Zhou, H, Shi, S & Ma, B 2009, `A virtual injection molding system based on numerical

simulation,

International Journal Advanced Manufacturing Technology

, vol. 40, pp. 297-306.

Zhou, J & Turng, LS 2007, `Process Optimization of Injection Molding Using an Adaptive

Surrogate Model with Gaussian Process Approach′,

Polymer Engineering and Science

, vol.

47, pp. 684-694.

26

---

Appendix A

Table of quality response comparison of different approach

AMD

Standard Deviation

Mean

Length Weight Length

Warpage Shrinkage Warpage Shrinkage

Other

approach

Proposed

approach

Improvement

27

---

Appendix B

Alias Relationship for Fractional 25 Designs

I = ABCDE

A = BCDE AB = CDE AC = BDE AD = BCE AE = BCD

B = ACDE BC = ADE BD = ACE BE = ACD

C = ABDE CD = ABE CE = ABD

D = ABCE DE = ABC

E = ABCD

28

---

Appendix C

25-1 Design

Run

Factors

number

A B C D E AB AC AD AE BC BD BE CD CE DE I=ABCDE

1

-1 -1 -1 -1 1

1

1

1

-1

1

1

-1

1

-1

-1

1

2

1

-1 -1 -1 -1

-1

-1

-1

-1

1

1

1

1

1

1

1

3

-1 1

-1 -1 -1

-1

1

1

1

-1

-1

-1

1

1

1

1

4

1

1

-1 -1 1

1

-1

-1

1

-1

-1

1

1

-1

-1

1

5

-1 -1 1

-1 -1

1

-1

1

1

-1

1

1

-1

-1

1

1

6

1

-1 1

-1 1

-1

1

-1

1

-1

1

-1

-1

1

-1

1

7

-1 1

1

-1 1

-1

-1

1

-1

1

-1

1

-1

1

-1

1

8

1

1

1

-1 -1

1

1

-1

-1

1

-1

-1

-1

-1

1

1

9

-1 -1 -1 1

-1

1

1

-1

1

1

-1

1

-1

1

-1

1

10

1

-1 -1 1

1

-1

-1

1

1

1

-1

-1

-1

-1

1

1

11

-1 1

-1 1

1

-1

1

-1

-1

-1

1

1

-1

-1

1

1

12

1

1

-1 1

-1

1

-1

1

-1

-1

1

-1

-1

1

-1

1

13

-1 -1 1

1

1

1

-1

-1

-1

-1

-1

-1

1

1

1

1

14

1

-1 1

1

-1

-1

1

1

-1

-1

-1

1

1

-1

-1

1

15

-1 1

1

1

-1

-1

-1

-1

1

1

1

-1

1

-1

-1

1

16

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

1

29

---

Appendix D

Timeframe of the project

30

---