See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/265857192

Six Sigma methods applied in an injection moulding company

Article in International Journal of Lean Six Sigma · May 2014

DOI: 10.1108/IJLSS-07-2013-0037

CITATIONS READS

25 6,900

3 authors:

Werner Timans Kees Ahaus

Stenden University Emmen University of Groningen
6 PUBLICATIONS 412 CITATIONS 43 PUBLICATIONS 1,526 CITATIONS

SEE PROFILE SEE PROFILE

Jiju Antony
Heriot-Watt University
492 PUBLICATIONS 25,341 CITATIONS

SEE PROFILE

All content following this page was uploaded by Werner Timans on 01 January 2015.

The user has requested enhancement of the downloaded file.

 The current issue and full text archive of this journal is available at
www.emeraldinsight.com/2040-4166.htm

Six Sigma methods applied in an Six Sigma

methods
injection moulding company
Werner Timans
Department of Mechanical Engineering, Stenden University of Applied
Science, Emmen, The Netherlands 149
Kees Ahaus Received 4 July 2013
Faculty of Economics and Business, University of Groningen, Groningen, Revised 29 August 2013
The Netherlands, and Accepted 15 September 2013

Jiju Antony
School of Management and Languages, Heriot-Watt University,
Edinburgh, Scotland, UK

Abstract
Purpose – The purpose of this paper is to provide a demonstration of the application of techniques for
robust optimization for improvement of the injection moulding processes in an injection moulding small
and medium sized enterprise (SME).
Design/methodology/approach – A critical to quality characteristic (CtQ) which is connected to
assembly problems is the subject of investigation. The CtQ is not directly measurable. The variation in
a dimension of a product, which is correlated to the CtQ, is studied using design of experiments (DoE)
and Taguchi methods. A two-cavity mould is used in the injection moulding process. To evaluate the
robustness of the process using signal-to-noise analysis, the data were transformed to compensate for
the systematic differences between the mould cavities.
Findings – The initial results showed that finding optimal process parameter settings commonly
valid for both cavities was impossible. After a modification of the mould, the experiments were rerun
and optimal settings could be found.
Practical implications – Applying DoE techniques in small and medium-sized injection moulding
companies is far from common practice. This case study demonstrates a method to apply DoE with five
process parameters which can serve as a standard method to prepare production when a new mould is
used for the first time.
Originality/value – The originality is connected to the combination of the applied methods and, in the
context of the case study, carried out in an SME unfamiliar with the power of the applied methods. The
value of the paper is to demonstrate the power of the most powerful technique in quality engineering to
improve an injection moulding process within the context of SMEs. The authors would accentuate the
point that the true power becomes visible when this powerful technique is introduced into an
organization with very little understanding of the technique. In addition, the case study is valuable to
practitioners because it proposes a new scientific and systematic approach to understand and optimize
the start-up of the moulding process.
Keywords SME, Lean Six Sigma, DoE, Injection moulding
Paper type Case study
International Journal of Lean Six
Sigma
Vol. 5 No. 2, 2014
1. Introduction pp. 149-167
Injection moulding is a cost-effective production process for producing complex plastic © Emerald Group Publishing Limited
2040-4166
parts in large quantities. A hot melt of thermoplastic polymer is forced into a DOI 10.1108/IJLSS-07-2013-0037
IJLSS mould-cavity at a lower temperature, where the hot melt solidifies. After solidification,
the mould is opened and the product is removed from the mould cavity. The process is
5,2 regulated by a number of process parameters. The process looks rather simple at first
sight, but predicting the quality of the final part is very complex, especially when the
demands with respect to specified dimensional, shape and surface properties are high.
Shrinking of the product after ejection from the mould is a very well-known source of
150 quality problems, especially for complex products with low wall thicknesses and
asymmetric shapes.
In The Netherlands, the vast majority of injection moulding companies are small and
medium sized enterprises (SMEs) with ⬍ 100 employees. In general, these companies
are familiar with state-of-the-art technology with regard to the mould design, including
computer aided design (CAD), melt-flow and finite elements (FE) simulation methods.
When a new mould has been mounted for the first time on the injection moulding
machine, experienced engineers and shop floor employees determine initial machine
set-points using information from simulation studies, material specifications and
machine signals. Customers express demands not only with regard to the quality of the
final products but increasingly also to the level of control of the related process
parameters. To meet these demands, the application of tools available in the Lean Six
Sigma (LSS) toolbox becomes increasingly important.
Implementation of LSS methods in manufacturing SMEs has been a subject of earlier
studies (Antony et al., 2005, 2008; Timans et al., 2012). Case studies in which tools are
applied which are included the LSS toolbox are available (Dodd et al., 2002; Liu et al.,
2002; Lin and Chananda, 2003-2004; Lee et al., 2006; Oktem et al., 2007; Tang et al., 2007;
Lo et al., 2009). In these case studies, both Taguchi and classical design of experiments
(DoE) methods have been applied, but none of the case studies address SME-
implementation issues as a main topic of study.
The SME context makes it very important to lay emphasis on clarifying methods
and tools and on the need for standardization of the application, to make it easily
repeatable in new projects. In this study, we focus on the optimization of the
injection moulding process once a new mould has been mounted on the injection
moulding machine for the first time. The determination of set-points for critical
process parameters is mainly based on experience of engineers and shop floor
employees. Therefore, we propose to implement a standard routine project focused
on improving the level of control of actions to be taken to prepare for normal
production. The effect of this routine should lead to shortening of time between the
first mount of the mould on the machine and production release. The Lean Six Sigma –
DMAIC cycle (Define, Measure, Analyse, Improve, and Control) can be used to structure the
project. Many researchers have described DMAIC project stages and appropriate tools for
use in the different DMAIC phases (Timans et al., 2009). In this study, we refer to DMAIC
project stages in a generic form, as described by De Koning and De Mast (2006) and outlined
in Figure 1. SMEs active in polymer injection moulding are, in general, familiar at a basic
level with the application of tools from the LSS toolbox, especially with basic control
charting techniques and process capability indicators, like potential process capability (Cp)
and process capability (Cpk). Supporting software is often provided by manufacturers of
injection moulding equipment. Measurement system analysis is recognized as very
important, but standardized methods to evaluate measurement systems have not been
widely implemented. Advanced statistical methods such as DoE are seldom applied within
 Define Six Sigma
Problem selecon and benefit analysis. methods
Definion of the CtQ or CtQs (Crical to
Quality characteriscs). A CtQ is a key
measurable characterisc of a product
or process.
151
Measure
Translaon of the problem into a
measurable form, and measurement of
the current situaon.

Analyse
Idenficaon of influence factors and
causes that determine CtQ behaviors.

Improve
Design and implement adjustments to
the process to improve the performance
of the CtQs.

Control
Adjustment of the process management Figure 1.
and control system in order that Flowchart of the DMAIC
improvements are sustained. steps

injection moulding SMEs. A study on the implementation of Lean Six Sigma (Timans et al.,
2012) revealed that manufacturing SMEs in The Netherlands recognized not only the
importance of DoE but also that DoE techniques were seldom used in practice. For injection
moulding companies in particular, the application of DoE techniques can be effective in
searching for optimal process settings when a new mould is mounted on an injection
moulding machine for the first time.
The product under study is to be produced in a company with about 20 employees, so the
SME context is evident. For the injection moulding company, it is a pilot study for gaining
experience in the application of methods which it intends to apply as standard methods in
preparing production runs in the future. The engineers should be able to understand the
rationale of the method, and the experiments should be completed in a brief time frame.
In the following sections, the case study will be presented first. Initially, the approach
was rather straightforward, but during the investigations some unexpected problems
arose, making it necessary to retrace one’s steps and take some unanticipated measures
to solve the problems. As such, this case study illustrates the iterative nature of the
DMAIC project stages, as has been described earlier (Pande et al., 2000, p. 239). The
subsequent discussion section focuses on the case study and on issues connected with
carrying out such projects within the injection moulding industry, paying attention to
the specific technological aspects of this industry and the awareness that many injection
IJLSS moulding companies are SMEs. Recommendations will be proposed for research to
improve the methods described in the case study further.
5,2
2. Case study
The product studied, illustrated in Figure 2, is a plastic housing for a pressure measuring
device. This is produced by a company in Emmen, The Netherlands. The material specified
152 is a 35 per cent glass-reinforced, heat stabilized, lubricated high-performance polyamide.
The customer delivered fully specified CAD drawings of the product, and responsibility for
the mould design was transferred to the company. The mould design was carried out in
cooperation with an external company that is specialized in mould design and
manufacturing. The mould has two cavities such that every cycle of the injection moulding
process delivers two products. A flow study was carried out by an external specialist to
analyze the moulding process regarding filling behaviour and warping risks. This analysis
delivered recommendations on gate position and gate dimensions, wall thicknesses at
critical spots, venting of the mould to prevent air-traps and cooling and ejection system.
Production would be carried out on an Arburg injection moulding machine.

2.1 The DMAIC steps

2.1.1 Define. The critical to quality (CtQ) characteristic in this study is the parallelism of the
two pins which are visible in Figure 2. For the customer, the parallelism of the pins is very
important because on these pins, the measuring device is mounted in the assembly process.
The parallelism of the pins, therefore, is regarded as critical by the customer. Significant
deviations from the ideal parallelism of the pins will cause problems in the assembly process.
Measuring this parallelism directly is very hard and would need specialist equipments and
skills unavailable within the company. Therefore, a product dimension, which in this paper
is referred to as “height” (Figure 2), was selected as a measurable characteristic which is
correlated to the parallelism of the pins. Antony (2003) highlights this particular aspect in his
book stating that experimenters should seek correlated characteristics if some CtQ
characteristics are difficult to measure. Due to the asymmetric structure of the product, some
warp is evoked when the product shrinks after ejection from the mould. This warp causes
variation in both the parallelism of the pins and in the height. During shrinking, the top
surface of the chamber (Figure 2) will be distorted and will get a slightly hollow shape, and

pins height

Figure 2.
Chamber top surface
The product Tower
the tower will bend inwards. The stronger the tower is bended, the stronger the chamber’s Six Sigma
top surface will be deformed into a hollow shape. The deformation of the top surface of the
chamber causes deviations in the parallelism of the pins. By controlling the injection
methods
moulding process in such a way that the height is stabilized, the parallelism of the pins is
stabilized too.
Given the final mould design and material specifications, the following project goals
were specified in the define phase: 153
• to quantify the contributions of the process parameters to variations in the height; and
• to divide these parameters into those which mainly affect product to product variation
of the height and those which mainly affect the mean level, followed by the
determination of optimal set points for the process parameters.

Four team members were directly involved: two experienced operators, the company’s
technical managing director and an external expert. In this pilot study, no financial targets
were set: the main goal of this project was to ensure a stable and robust production process,
easily controllable through the important process control parameters.
2.1.2 Measure. The mould was designed for a completely new product. The new
mould was tested by an experienced operator, who tried to operate the process with
initial set-points for the process parameters based on experience. During the initial tests,
it was not possible to find acceptable set-points, in particular because trapped air caused
burn spots in the polymer material at a specific location on the product. The mould
manufacturer therefore modified the mould, especially by widening the channels
through which air could escape during injection of the polymer. After this minor
modification, a first series of products was made. During this production run, the
operator was able to produce a series of products that appeared acceptable regarding
visible surface properties.
To validate the measurements of the height, the measurement system was analyzed.
Fifteen products were measured three times to quantify the contribution to variance
caused by the measurement process as compared to the total variance. The total
variance can be split up as follows:

␴total
2
⫽ ␴process
2
⫹ ␴meas
2

The variance component due to the measurement system was estimated to be 1.27
per cent of the total variance. The ␴meas was, therefore, estimated to be 11.3 per cent
of ␴total. General industrial guidelines for measurement system analysis prescribe
that ␴meas should preferably be ⬍ 10 per cent of ␴total, and that contributions up to
30 per cent could be acceptable depending on the application (AIAG, 2010, MSA
Manual,). In this case, the contribution of the measurement system is close to 10 per
cent. Based on these data, we qualified the performance of the measurement system
to be acceptable as yet.
2.1.3 Analyze. The members of the team discussed factors that should be included in
the research. Based on material specifications, flow-analysis and experience, the team
selected five process parameters which should be included in the experiments:
(1) injection velocity (Vinj);
(2) holding pressure (Phold);
IJLSS (3) melt temperature (Tmelt);
5,2 (4) mould temperature (Tmold); and
(5) cooling time (t-cool).

The team was aware that, next to controllable factors, noise factors could also influence
the process performance. These noise factors could be environmental (for instance the
154 temperature of the environment), could be related to small variations in machine
settings (small variations in controlled set-points over successive machine cycles) or to
variations in material properties. The team considered specifically the risk of variations
due to inconsistencies in the material properties and judged these risks to be very small
with the material being used. A larger risk was expected due to differences related to the
two cavities in the mould, and therefore it was decided to pay specific attention to these
differences in the study. In the brainstorming session, risks were also discussed related
to the perceived influences of interaction effects. These discussions did not deliver clear
arguments on which interactions should be included or excluded in the research.
Therefore, the team decided that a design set-up should be selected in such a way that all
two-factor interaction effects could be estimated separately.
Preliminary trials were carried out to explore the experimentation window. The limits of
the process parameter settings should be such that all the products would come out of the
machine completely filled and also could easily be ejected from the mould. The team also
discussed the nature of the relationships between the measurement data and the process
control parameters within the process window, and argued that these relationships would be
at least monotonous and probably close to linear, and therefore decided to vary the control
parameters at two levels within the process window. Based on the preliminary trials, the
team set limits for the variation of the five control parameters (see Table I):
The injection velocity is not expressed in physical units, but as a percentage of its
maximum value. In the injection moulding machine used, this is controlled by a
regulating valve in the injection system.
2.1.4 Improve. The safest design for the experiments would be a full factorial design
with five control parameters. A full factorial design with five factors, with all factors
being set to two levels, would need 25 ⫽ 32 experiments. Using this design, and
assuming linearity, all main factor effects, all two-factor interaction effects and all
higher-order interaction effects could be estimated. However, it seemed fair to assume
that, in our case study, higher-order interaction effects would be small compared to the
main effects and two-factor interaction effects, and therefore we concluded that a
fractional factorial design would be adequate (see for instance Montgomery, 2005).
Given that it was not possible to exclude any two-factor interactions based on technical
arguments, the decision was taken to use a 25-1 design, in which all main effects and

Control parameter Low level High level

Vinj 40 per cent 60 per cent

Phold 500 bar 700 bar
Table I. Tmelt 320°C 330°C
Process window for Tmold 140°C 160°C
performance tests t-cool 13 s 17 s
two-factor interaction effects could be estimated from 16 experiments at corner points. Six Sigma
To be able to test for curvature, it was decided to add some centre-point runs. These
centre-point runs were planned at the start, midway and at the end of the experiments.
methods
Table II summarizes the overall test design.
The design was not randomized because of practical considerations. Injection
moulding processes running with fixed process parameters are very stable. The
time needed to carry out all of the 19 experiments would only take about three hours, 155
and trends in the measurement results due to slowly changing uncontrolled
influences were not expected to occur within a timeframe of three hours. Changing
Tmold settings would take some time because, after a change in Tmold, the injection
moulding machine needs some time to stabilize. Tmold is controlled by an external
temperature regulating unit and, after a change in the Tmold set-point, it takes some
time to reach a new thermal equilibrium. Hence, the test schedule was organized to
minimize the number of times Tmold would have to be altered. Using the
centre-point settings for the first experiment, at the halfway stage, and at the end
enabled us to test our assumption that randomization was unnecessary.
Each trial was carried out by allowing the machine to run for five cycles and so
produce ten products, five from each cavity. After the runs, the products were stored for
24 hours to allow them to cool to room temperature and to shrink to their final
dimensions. The products were then measured, and the measurement values were
stored in a Minitab spread-sheet.
2.1.5 Analysis of data from Cavity 1 and Cavity 2. In the analysis, we first
concentrated on the main effects. Figures 3 and 4 present Pareto charts of the main

Vinj Phold Tmelt Tmold t-cool

0 0 0 0 0
⫺1 ⫺1 ⫺1 ⫺1 1
1 ⫺1 ⫺1 ⫺1 ⫺1
⫺1 1 ⫺1 ⫺1 ⫺1
1 1 ⫺1 ⫺1 1
⫺1 ⫺1 1 ⫺1 ⫺1
1 ⫺1 1 ⫺1 1
⫺1 1 1 ⫺1 1
1 1 1 ⫺1 ⫺1
0 0 0 0 0
⫺1 ⫺1 ⫺1 1 ⫺1
1 ⫺1 ⫺1 1 1
⫺1 1 ⫺1 1 1
1 1 ⫺1 1 ⫺1
⫺1 ⫺1 1 1 1
1 ⫺1 1 1 ⫺1
⫺1 1 1 1 ⫺1 Table II.
1 1 1 1 1 Design array for the
0 0 0 0 0 experiments
IJLSS
5,2 Pareto Chart of the Standardized Effects
(response is Mean Height1, Alpha = 0,05)
2,179

Tmold
156
Phold
Term

Tmelt

t-cool

Vinj

Figure 3. 0 1 2 3 4
Pareto of effects based on Standardized Effect
data from Cavity 1

Pareto Chart of the Standardized Effects

(response is Mean-Height2, Alpha = 0,05)
2,179

Phold

t-cool
Term

Tmelt

Tmold

Vinj

Figure 4.
0 1 2 3 4 5 6 7
Pareto of effects based on
Standardized Effect
data from Cavity 2
effects. The residuals are approximately normally distributed, and Figure 5 illustrates Six Sigma
this for Cavity 1.
Approximately 80 per cent of the variation is caused by the main effects. The total
methods
variation in the data is considerably larger for Cavity 2 than for Cavity 1. An F-test
revealed that the variances of the data from the two cavities are significantly different
with a confidence level ⬎ 99 per cent.
To compare the data means of the two cavities, a two-sample paired t-test was carried 157
out. The paired test was appropriate because the samples were produced pairwise in a
two-cavity mould, and are therefore were not independent from each other. The estimate
of the difference was 22.7 ␮m (␮2 ⬎ ␮1). The 95 per cent confidence interval of ␮2-␮1
ranged from 17.0 to 28.4 ␮m.
Comparing the results of the two cavities, the most salient differences are visible in
the Pareto charts. For both cavities, the injection velocity does not have a significant
effect (␣ ⫽ 0.05). For Cavity 1, Tmold is the most important factor, whereas this is not
significant for Cavity 2. This means that it will be impossible to optimize the process for
both cavities at the same time. This conclusion had certainly not been anticipated and
greatly worried the management team. It was concluded from studying the construction
of the mould that the main cause was probably the layout of the cooling channels inside
the mould. The two cavities were cooled by an external temperature regulating unit. The
cavities were cooled sequentially: the cooling channel first flowed around the first cavity
and then on to the second cavity before returning to the external tempering unit. A
consequence of this is that the temperature of the cooling fluid increases when it reaches
the second cavity, so the cooling of the second cavity will be less effective. This cooling
method is not uncommon in injection moulding practice, but in our case study, it could very
well be the main cause of the different outcomes for both cavities. It was decided to return the

Normal Probability Plot

(response is Mean Height1)
99

95
90

80
70
Percent

60
50
40
30
20

10

Figure 5.
1
-3 -2 -1 0 1 2 3 Normal plot of residuals
Standardized Residual based on data from
Cavity 1
IJLSS mould to its manufacturer to modify the layout of the cooling channels in such a way that
both cavities would be directly connected to the tempering unit. In this way, in both cavities,
5,2 the temperature of the cooling fluid would be the same at the start of the cooling process.
After the return of the modified mould, the experiments would be repeated.
As the system stood, Vinj seemed to be the only parameter that was not significant
for either cavity, although this could change with the modified mould. Overall, it was
158 believed that the balance among the influences of the process parameters could change
after modifying the layout of the cooling channels.
2.1.6 Analysis of data from Cavity 1 and Cavity 2 (modified mould). Some weeks later,
once the mould was modified and returned, the set of experiments listed in Table II was
repeated. The new data were collected in the same way as before. In the Tables III
and IV, ANOVA results are presented. Figures 6 and 7 present Pareto charts for both
cavities. The residuals are again approximately normally distributed for both cavities.
Figure 8 shows a main effect plot for Cavity 1 based on the new data. In the ANOVA tables
(based on means per experiments), all insignificant effects (␣ ⫽ 0.05) have been pooled into
the residual error. The Lack of Fit contains the insignificant contributions of Tmelt and of all
two-factor interactions. The curvature is clearly insignificant, confirming our initial
expectation that within the process window (see Table I), a linear model would be sufficient.
Over 85 per cent of the variation is caused by the main effects alone. The total
variation of the data is still somewhat larger for Cavity 2 than for Cavity 1, but the
difference is much smaller than with the original mould. An F-test showed that the
variances of the data are not significantly different for the two cavities. The most
significant control parameters Tmold and t-cool have comparable influences with
both cavities (see Figures 6 and 7). For both cavities, these two control parameters
determine about 75 per cent of the total variation. Tmelt is insignificant for both
cavities.
The data from both cavities are not independent from each other because in each
process cycle, two products are produced, one in each cavity. The difference between the
means was again tested by applying a two-sample paired t-test. The estimate of the
difference was 29.7 ␮m (␮2 ⬎ ␮1). The 95 per cent confidence interval of ␮2-␮1 covered
the range from 28.3 to 31.2 ␮m. The confidence interval has narrowed considerably
compared to the earlier test results, primarily due to the decreased variation in the data

ANOVA
Source df Seq SS Adj SS Adj MS F p

Main effects 4 3076.24 3076.24 769.06 47.91 0.000

Vinj 1 262.44 262.44 262.44 16.35 0.001
Phold 1 207.36 207.36 207.36 12.92 0.003
Tmold 1 1310.44 1310.44 1310.44 81.63 0.000
t-coo 1 1296.00 1296.00 1296.00 80.73 0.000
Residual error 14 224.74 224.74 16.05
Curvature 1 13.75 13.75 13.75 0.85 0.374
Lack of Fit 11 163.60 163.60 14.87 0.63 0.753
Pure error 2 47.39 47.39 23.69
Table III. Total 18 3300.98
ANOVA of Height1, based
on new data from Cavity 1 Notes: S ⫽ 4.00661; R-Sq ⫽ 93.19 per cent; R-Sq(adj) ⫽ 91.25 per cent
 ANOVA
Six Sigma
Source df Seq SS Adj SS Adj MS F p methods
Main effects 4 3516.93 3516.93 879.23 28.56 0.000
Vinj 1 174.24 174.24 174.24 5.66 0.032
Phold 1 408.04 408.04 408.04 13.25 0.003
Tmold 1 1730.56 1730.56 1730.56 56.21 0.000
t-cool 1 1204.09 1204.09 1204.09 39.11 0.000
159
Residual error 14 431.01 431.01 30.79
Curvature 1 60.66 60.66 60.66 2.13 0.168
Lack of Fit 11 335.07 335.07 30.46 1.73 0.423
Pure error 2 35.28 35.28 17.64
Total 18 3947.94 Table IV.
ANOVA of Height2, based
Notes: S ⫽ 5.54853; R-Sq ⫽ 89.08 per cent; R-Sq(adj) ⫽ 85.96 per cent on new data from Cavity 2

Pareto Chart of the Standardized Effects

(response is mean Height1_new, Alpha = 0,05)
2,179

Tmold

t-cool
Term

Vinj

Phold

Tmelt
Figure 6.
Pareto of main effects
0 1 2 3 4 5 6 7 8 9
based on new data from
Standardized Effect
Cavity 1

from Cavity 2. From a process-control perspective, it would be appropriate to analyze all

the data together. To do this correctly, the data should be corrected for any systematic
differences between the data from both cavities. Therefore, 30 ␮m (rounded value of 29.7
␮m) was subtracted from the Cavity 2 data and the complete dataset (called
height-overall) was analyzed. The results are presented in Table V and illustrated in
Figures 9 and 10.
In this ANOVA, the Lack of Fit contains the main factor Tmelt and two-factor
interactions between control factors.
IJLSS
5,2 Pareto Chart of the Standardized Effects
(response is mean Height2_new, Alpha = 0,05)
2,179

Tmold
160
t-cool
Term

Phold

Vinj

Tmelt
Figure 7.
Pareto of main effects
0 1 2 3 4 5 6 7 8 9
based on new data from
Standardized Effect
Cavity 2

Main Effects Plot (data means) for Height1_new

Vinj Phold Point Ty pe
54905
Corner
Center
54900

54895
Mean of Height1_new

54890

54885
-1 0 1 -1 0 1
Tmold t-cool
54905

54900

54895

54890
Figure 8.
Main effect plots based on 54885
-1 0 1 -1 0 1
new data from Cavity 1
Figure 9 shows that the residuals follow an approximately normal distribution, providing Six Sigma
confidence that the transformation of the data for Cavity 2 was appropriate. The Pareto
values shown in Figure 10 are in line with the Pareto graphs of the distinct cavities. The
methods
estimated standard deviation (S) and R-Sq values of the overall height (Table V) are similar
with the estimates from the data for the individual cavities (Tables III and IV).
The second research goal formulated in the Define phase was to divide the
parameters into those which mainly affect the process stability and those which mainly 161
affect the mean height. To achieve this, a Taguchi signal-to-noise analysis was carried
out on the height-overall data. Many literature sources are available on the application of

ANOVA for mean Height-overall

Source df Seq SS Adj SS Adj MS F p

Main effects 4 3278.21 3278.21 819.55 50.94 0.000

Vinj 1 216.09 216.09 216.09 13.43 0.003
Phold 1 299.29 299.29 299.29 18.60 0.001
Tmold 1 1513.21 1513.21 1513.21 94.06 0.000
t-cool 1 1249.62 1249.62 1249.62 77.67 0.000
Residual error 14 225.24 225.24 16.09
Curvature 1 33.04 33.04 33.04 2.24 0.159
Lack of Fit 11 189.47 189.47 17.22 12.63 0.076
Pure error 2 2.73 2.73 1.36
Total 18 3503.45 Table V.
ANOVA of combined data
Notes: S ⫽ 4.01105; R-Sq ⫽ 93.57 per cent; R-Sq(adj) ⫽ 91.73 per cent from both cavities

Normal Probability Plot

(response is mean Height-overall)
99

95
90

80
70
Percent

60
50
40
30
20

10

Figure 9.
1
-3 -2 -1 0 1 2 3 Normal probability plot
Standardized Residual based on new data from
both cavities
IJLSS
5,2 Pareto Chart of the Standardized Effects
(response is mean Height-overall, Alpha = 0,05)
2,14

162 Tmold

t-cool
Term

Phold

Vinj

Figure 10.
Pareto of main effects 0 2 4 6 8 10
based on new data from Standardized Effect
both cavities

signal-to-noise analysis in relation to robustness (Ross, 1996; Phadke, 1989) . For each
experimental run, a signal-to-noise value was calculated using the variance-only
formula for signal to noise (S/N) nominal is best: S/N ⫽ ⫺10log (␴2). In this formula, ␴ is
estimated from S, the sample’s standard deviation calculated from the ten data points for
each of the experiments. The alternative Taguchi S/N nominal is the best formula – a
better choice when the variance is related to the mean level – S/N⫽ ⫺10log (␴2/␮2)
which was also tested, but a comparison showed that both transformations led to the
same conclusions. From this result, we infer that the influence of the mean experimental
levels is very low in the S/N analysis. The analysis revealed that mould temperature,
Tmold, and the holding pressure, Phold, were the factors with the strongest influences
on the S/N levels. Both factors should be set to their higher levels to achieve the highest
S/N value, corresponding to the lowest S (and the most consistent product). In general
high S/N values correspond to good process stability. Figure 11 shows that the effect of
mould temperature on S/N is very high when the holding pressure is set at the higher
level, but it is more modest when the holding pressure is set at the lower level. This
means that setting both the holding pressure and the mould temperature at high levels
will improve the stability of the process. An analysis of the standard deviations of the
experiments leads to the same conclusions.
The cooling time, t-cool, is insignificant in the S/N analysis but was a very significant
factor in the Pareto analysis, as shown in Figure 10. As such, t-cool could be used to
control the mean level.
An important concern was what was causing the difference between the means of
product heights of the products coming out of the two cavities. The difference
between the means was 29.7 ␮m (␮2 ⬎ ␮1), with a 95 per cent confidence interval
(28.3 to 31.2 ␮m). The question was raised if the difference was caused by
 Six Sigma
Interaction Plot (data means) for S/N methods
Phold
-1
-10 1

163
-11
Mean

-12

-13

-14
Figure 11.
-1 1
S/N interaction plot of
Tmold
Tmold*Phold

differences in process parameter values (pressure, temperature) inside the two

mould cavities or by differences in the corresponding physical dimensions of the
cavities. The difference between the two cavity depths was measured using a digital
height gauge instrument and appeared to be 30 ␮m. Therefore, we conclude that the
difference between the means of the product heights from the two cavities was
mainly due to differences in mould cavity depths and not due to differences in
process parameters inside the cavities.
2.1.7 Control. The experiments were not carried out in a laboratory environment, but
on a production machine situated in the production hall alongside other machines
producing a wide range of products. Therefore, additional effects of scaling up to
production are not anticipated. The case study was set up to prove that the company
was able to produce the product on a continuous and consistent basis. The
recommended process settings have been implemented and documented in the process
parameter sheets which are used by employees during production.
We recall that the CtQ characteristic was the parallelism of the two pins, as depicted in
Figure 2. Bad parallelism causes problems during production because then the assembly of
the measuring device on these pins is troubled. Because the process runs with the important
process settings set on their optimal values, several production orders have been delivered to
the customer without receiving any customer complaints.

3. Discussion, conclusions and outline of further study

3.1 The case study
In the define phase, the following goals were set:
• to quantify the contributions of the process parameters to the variation in height; and
IJLSS • to divide these parameters into those which mainly affect product to product variation
of the height and those which mainly affect the mean level, followed by the
5,2 determination of optimal set points for the process parameters.

With regards to the first goal, the Pareto plot of the main effects in Figure 10 presents the
contributions of the parameters to the variation of the height. The validity of the
164 analysis is sufficiently justified by the ANOVA results shown in Table V and the normal
probability plot in Figure 9. The second goal was addressed through a Taguchi S/N
analysis, and it was concluded that high values for Phold and Tmold, shown in Figure
11, enhance the stability of the process.
This case study demonstrates the iterative nature of the DMAIC project phases,
especially with regards to the Analyze and Improve phases. In the Analyze phase,
the focus is on determining which parameters are important and which interactions
between the process parameters could be important. The experimental results
showed that it was not possible initially to find optimal settings that were equally
valid for both cavities. Only after modifying the mould cooling system it was
possible to optimize the process. Looking back, this demonstrates that the
separation between the Analyze and Improve phases is somewhat artificial. As
anticipated in the introduction, it is not always realistic to straightforwardly follow
the DMAIC project phases. Sometimes it is necessary to step back to a previous
phase, reaching the best results through an iteration of the phases. Recognizing this
during this study was an important lesson.
As to the main goal of our research, providing a standardized approach to improve
the optimization of the injection moulding process between first mount of the mould
on the injection moulding machine and production release, this case study demonstrates
the methods that we have proposed. The applied DoE array itself is very useful as a
standard to be used in similar projects, because it delivers reliable information on main
effects and two-factor interaction effects when no more than five process parameters are
included in the experiments, which will in general be sufficient. In this particular case,
curvature does not seem to be a major problem, but in new cases, it would be wise to
investigate curvature again, especially when a process window is used with wider
ranges from the lowest to the highest parameter settings.

3.2 Implementing LSS in an injection moulding SME

Lean Six Sigma has been developed, since about 2000, as an effective merger of lean
manufacturing and Six Sigma (George, 2002; Snee and Hoerl, 2007; Snee, 2010). The LSS
toolbox provides a diversity of tools, some of which are easy to use and others that are
less easy to use and need a thorough education and training program before application
is feasible. The toolbox contains tools with a lean background as well as tools with a Six
Sigma background. In this case study, primarily Six Sigma tools were applied, but in the
context of injection moulding, lean tools are very useful as well.
Based on the results of this study, we recommend that if small injection moulding
companies aspire to grow into organizations applying LSS, at least one engineer should
have sufficient knowledge on the use of relevant statistical tools, including Gage
R&R and DoE. Statistical techniques are only useful when applied correctly.
Attaching importance to the understanding of the technicalities of statistical
methodology by management is not common practice (Sheil and Hale, 2012).
Preparing for the production of newly developed products, ones that have not been Six Sigma
produced before, is a critical issue on the way towards production release.
Production preparation involving new moulds could be carried out as best practice
methods
LSS projects, embedded in standardized routines.

3.3 Further study

In this study, it was noted that while the value of applying DoE in an LSS program may 165
be evident, the question remains as to how such a tool may be introduced effectively in
such a program in an SME. The application of two-level designs has been proved to be
the right way in this particular case study, but further study is needed to be able to prove
if it is the right way in all cases. New case studies can be carried out using both two-level
designs with centre-point measurements and three-level designs to find out which
designs are the best choices in particular cases to serve as standards in the context of
polymer injection moulding. Moreover, in a two-level factorial design, we choose the
best level from the two values for a process parameter. However, the true optimal
value could be either less than or more than the chosen value for that process parameter.
The authors would, therefore, would like to explore the potential of utilizing a response
surface methodology in the future to determine the optimal region. How to educate
engineers and operators at levels appropriate for their daily practice in applying
statistical techniques in the injection moulding context will be an important challenge in
the near future.

References
AIAG (2010), Measurement System Analysis, AIAG Reference manual, 4th ed, Automotive
Industry Action Group, available at: www.aiag.org/source/Orders/
prodDetail.cfm?productDetail⫽MSA-4.
Antony, J. (2003), Design of Experiments for Engineers and Scientists, Butterworth-Heinemann,
Oxford.
Antony, J., Kumar, M. and Labib, A. (2008), “Gearing Six Sigma into UK manufacturing
SMEs: results from a pilot study”, Journal of the Operational Research Society, Vol. 59
No. 4 , pp. 482-493.
Antony, J., Kumar, M. and Madu, C.N. (2005), “Six Sigma in small- and medium-sized UK
manufacturing enterprises: some empirical observations”, International Journal of Quality
and Reliability Management, Vol. 22 No. 8, pp. 860-874.
De Koning, H. and De Mast, J. (2006), “A rational reconstruction of Six-Sigma’s breakthrough
cookbook”, International Journal of Quality and Reliability Management, Vol. 23 No. 7,
pp. 766-787.
Dodd, K., Ney, K., Neaves, S. and Rain, M. (2002), “Improvement in preform weights from a
48-cavity PET injection molder”, Quality Engineering, Vol. 14 No. 3, pp. 501-511.
George, M.L. (2002), Lean Six Sigma – Combining Six Sigma Quality with Lean Speed,
McGraw-Hill, New York, NY.
Lee, L.S. and Lin, J.C. (2006), “Design of the runner and gating system parameters or a multi-cavity
injection mould using FEM and neural network”, International Journal of Advanced
Manufacturing Technology, Vol. 27, pp. 1089-1096.
Lin, T. and Chananda, B. (2003-2004), “Quality Improvement of an injection molded product using
Design of Experiments: a case Study”, Quality Engineering, Vol. 16 No. 1, pp. 99-104.
IJLSS Liu, S.J. and Chang, K.H. (2002), “An empirical study of gas penetration in full-shot gas-assisted
injection moulded parts”, Journal of Engineering Manufacture, Vol. 216 No. 12,
5,2 pp. 1549-1559.
Lo, W.C., Tsai, K.M. and Hsieh, C.Y. (2009), “Six sigma approach to improve surface precision of
optical lenses in the injection-moulding process”, International Journal of Advanced
Manufacturing Technology, Vol. 41, pp. 885-896.
166 Montgomery, D.C. (2005), Design and Analysis of Experiments, 5th ed Wiley, Singapore.
Oktem, H., Erzurumlu, T. and Uzman, I. (2007), “Application of Taguchi optimization technique in
determining plastic injection molding process parameters for a thin-shell part”, Materials &
Design, Vol. 28, pp. 1271-1278.
Pande, P.S., Neuman, R.P. and Cavanagh, R.R. (2000), The Six-Sigma Way – How GE, Motorola,
and Other Top Companies Are Honing Their Performance, McGraw-Hill, New York, NY.
Phadke, M.S. (1989), Quality Engineering using Robust Design, Prentice-Hall International, Inc.
Ross, P.J. (1996), Taguchi Techniques for Quality Engineering, McGraw-Hill.
Sheil, J. and Hale, D. (2012), “The ‘democratisation’ of statistics: parastatisticians in
manufacturing”, Quality & Reliability Engineering International, Vol. 28, pp. 524-534.
Snee, R.D. (2010), “Lean Six Sigma— getting better all the time”, International Journal of Lean Six
Sigma, Vol. 1 No. 1, pp. 9-29.
Snee, R.D. and Hoerl, R.W. (2007), “Integrating lean and Six Sigma – a holistic approach”, Six
Sigma Forum Magazine, pp. 15-21.
Tang, S.H., Tan, Y.J., Sapuan, S.M., Sulaiman, S., Ismail, N. and Samin, R. (2007), “The use of
Taguchi method in the design of plastic injection mould for reducing warpage”, Journal of
Materials Processing Technology, Vol. 182, pp. 418-426.
Timans, W., Antony, J., Ahaus, K. and Van Solingen, R. (2012), “Implementation of Lean Six Sigma
in small- and medium-sized manufacturing enterprises in the Netherlands”, Journal of the
Operational. Research Society, Vol. 63, pp. 339-353.
Timans, J.W.J., Ahaus, C.T.B. and Van Solingen, R. (2009), “A Delphi study on Six Sigma tools and
techniques”, International Journal of Six Sigma and Competitive Advantage, Vol. 5 No. 3,
pp. 205-221.

About the authors

Werner Timans teaches Quality Engineering and Quality Management at the Department of
Mechanical Engineering of the Stenden University of Applied Science in Emmen, The
Netherlands, and works as an independent consultant. Teaching robust engineering to students
and professional engineers, as well as the deployment of quality and reliability tools and
techniques in engineering and manufacturing organizations, has been his main interest
throughout the past 25 years. He holds an MSc degree in Experimental Physics from the
University of Groningen. His research focuses on Lean Six Sigma implementation in
manufacturing industries, especially in companies developing and manufacturing polymer
products. Werner Timans is the corresponding author and can be contacted at: jwj.
timans@home.nl
Kees Ahaus is a part-time Professor of Healthcare Management at the Faculty of Economics
and Business of the University of Groningen, The Netherlands, and is the Managing Director of
TNO Management Consultants, a subsidiary of The Netherlands Organization for Applied
Scientific Research (TNO). He studied Management Science at the Eindhoven University of
Technology (MSc) and gained his PhD at the Faculty of Management and Organization of the
University of Groningen. He has written several international papers and a number of books on
quality management.
 Jiju Antony is a Professor of Quality Management in the School of Management and Six Sigma
Languages at the Heriot-Watt University. He received his BE in Mechanical Engineering from
South India, MSc in Quality and Reliability Engineering from the National University of Ireland methods
and PhD in Quality Management from the University of Portsmouth, UK. He has published over
250 journal and conference papers and six textbooks in the areas of design of experiments, quality
management, Taguchi methods and Six Sigma. He is the Editor of the International Journal of
Lean Six Sigma and a certified Lean Six Sigma Master Black Belt. He is a Fellow of the Royal
Statistical Society, Fellow of the Institute of Operations Management, Fellow of the Chartered 167
Quality Institute, Fellow of the Institute of Six Sigma Professionals and has been recently elected
to the International Academy of Quality.

To purchase reprints of this article please e-mail: reprints@emeraldinsight.com

Or visit our web site for further details: www.emeraldinsight.com/reprints

View publication stats