experimental design for process optimization
TRANSCRIPT
An application ofexperimental designfor processoptimisation
Hefin Rowlands,
Jiju Antony and
Graeme Knowles
Introduction
Of key importance in any step towards
improvement in operation and efficiency is
the move from a reactive approach to a
proactive approach (Kolarik and Pan, 1991).
The traditional reactive approach is about
detecting and correcting problems that
already exist. It is a historic approach and is
always focussed on past events. It manifests
itself as an inspection based philosophy where
quality is inspected into products at the
expense of high rework. Under this approach
no effort is put in to understand and solve the
cause of the rework or rejects since the
company is always under pressure to deliver
usually by fire fighting the problems. It can be
very difficult to break out of this scenario.
A proactive approach places emphasis on
measurement, analysis, prediction and
prevention. A greater emphasis is also needed
at the design stage in order to prevent defects
and errors appearing later in the product's life
cycle. To this end, quality must be designed
into products and processes rather than
inspected into them. An important and well-
recognised tool to achieve this goal is the
Taguchi method (Ross, 1988). The Taguchi
approach is aimed at minimising variation in
product and process performance and thereby
achieves continuous quality improvement of
products and processes.
Higher education has a role to play to
educate industry in the application of
powerful problem solving techniques such as
Taguchi methods. In other words, it is
important to have a proper communication
between the industrial and academic world
for tackling new challenging problems in
industry. This paper briefly outlines the
results of Taguchi case studies carried out by
part-time engineering students at the
University of Wales College Newport
(UWCN). The results of the study have
provided a greater stimulus among the
engineering fraternity for the wider
application of such applied statistical methods
in industry. Typical applications of the
Taguchi method include:. minimising the defects on a steel strip that
has been electrolytically coated with a tin
based coating;. to optimise the production process of
retaining a metal ring in a plastic body in
a braking system;. to investigate an over-adjustment
problem in a braking system;
The authors
Hefin Rowlands is Head of the Department of
Engineering, University of Wales College Newport,
Newport, UK.
Jiju Antony and Graeme Knowles are senior teaching
fellows at the International Manufacturing Centre,
University of Warwick, Coventry, UK
Keywords
Optimization, Taguchi methods, Design of experiments,
United Kingdom
Abstract
Dr Taguchi is a Japanese engineer and an international
quality consultant who has made breakthrough
improvements in product and process quality through the
use of statistical design of experiments (SDOE). The
Taguchi method became popular in the West in the 1980s
as a means to design robust products and processes.
Although many companies and industries have used the
method with success, the real benefits of the approach
were not realised and fully understood in many cases.
This lack of success could be attributed to a number of
factors, but mainly because the experiments were treated
in isolation and not integrated into a continuous
improvement strategy. This paper briefly presents the
results of the application of the Taguchi methodology in
the UK industry. The paper also illustrates the application
of the Taguchi method for optimising the production
process of retaining a metal ring in a plastic body in a
braking system.
Electronic access
The research register for this journal is available at
http://www.mcbup.com/research_registers/
quality.asp
The current issue and full text archive of this journal is
available at
http://www.emerald-library.com
Techniques
78
The TQM Magazine
Volume 12 . Number 2 . 2000 . pp. 78±83
# MCB University Press . ISSN 0954-478X
. to improve the weld strength of an
electrolytic welding process in the joining
of steel coils;. optimisation of a spot welding process;. to improve the resistance of conductive
plastic tracks.
Overview of the Taguchi method
In today's modern global market, quality is a
key issue for companies wishing to keep their
customers and remain competitive in
business. Quality can no longer be simply the
result of an inspection process, but needs to
be a company-wide management philosophy.
Quality improvement programmes are very
much part of the strategic planning process of
successful companies (McKeown, 1992).
Alongside the strategic planning issues are the
importance of design and the idea of
designing quality into products and processes.
By investing more effort in design, quality can
be designed into a product, thus reducing the
need to rely on inspection to ensure quality.
However, the task of optimising the design
becomes increasingly difficult as products
become more complex. A method which has
received much attention, both positive and
negative, over recent years as a quality
improvement tool is the Taguchi method.
The Taguchi method is based on statistical
design of experiments and is applied at the
parameter design stage to establish optimum
process settings or design parameters. The
following are the objectives of Taguchi's
parameter design:. making products and processes
insensitive to environmental variations
(e.g. humidity, ambient temperature,
dust, electrical supply voltage);. making product and processes insensitive
to manufacturing variations or
imperfections;. making products insensitive to product
deterioration (reliability degradation, tool
wear, etc.); and. making products insensitive to unit-to-
unit variations (component-to-
component variation, shift-to-shift
variations, machine-to-machine
variations, material-to-material variation
and so on).
Taguchi's philosophy of quality improvement
is to place effort into reducing variation in
products and processes at source. Rather than
reduce variation in individual components by
specifying tighter tolerances (tolerance
design), Taguchi's method addresses the
issue by careful selection of design parameters
(called factors). Reduction in variation in the
final product is achievable without the
additional cost of specifying tighter tolerance
components. This approach of parameter
design results in a more robust design that is
capable of withstanding variations from
unwanted sources such as raw materials,
components, manufacturing processes and
the environment.
The Taguchi philosophy and its associated
experimental design method has been
extensively used in the manufacturing
environment to improve production
processes, for example a metal injection
moulding process (Fox and Lee, 1990) and a
plasma deposition process in device
fabrication (Logothetis et al., 1990). In such
environments, careful planning of the
experiment is important if the full benefits of
the experimental methods are to be realised
(Coleman and Montgomery, 1993). Other
examples of manufacturing related
applications of the Taguchi method include
scheduling (Dooley and Mahmoodi, 1992)
and optimisation of a robot's performance
capability for continuous path operation (Wu
et al., 1991).
Despite the successful applications of the
Taguchi method, a wider use of the approach
and its associated techniques is only possible
by gaining a better understanding of the
method and its analysis. The successes and
failings of the Taguchi approach to parameter
design have been widely discussed (Nair,
1992; Lochner, 1991; Pignatiello and
Ramberg, 1991; Antony, 1996). In summary,
Taguchi's main successes have been to
emphasise the importance of quality in design
and to simplify the use of experimental design
as a general purpose tool for quality
engineers. Amongst the many criticisms of the
Taguchi method is the use of the signal-to-
noise (S/N) ratio as a performance measure
statistic. S/N ratio measures the functional
robustness of products and processes. The S/
N ratios have been criticised as providing
misleading results in certain cases. Although
the classical experimental design has a much
wider appeal than the Taguchi method, the
Taguchi method does provide the practical
engineer with a useful starting point for
quality improvement. This is fundamentally
because the former is more focused on the
79
An application of experimental design for process optimization
Hefin Rowlands, Jiju Antony and Graeme Knowles
The TQM Magazine
Volume 12 . Number 2 . 2000 . 78±83
statistical aspects whereas the latter is
primarily focused on the engineering aspects
of quality. The beauty of Taguchi method lies
in the fact that it integrates statistical methods
into the powerful engineering process.
Potential applications of Taguchimethods in industry
Taguchi methods have intensive applications
in many manufacturing companies. The
following section illustrates the applications of
Taguchi methods in various manufacturing
and service industry sectors. The
manufacturing sectors are classified into
plastics, automotive, metal fabrication,
process and electronics and semi-conductors.
Plastics. Process type: injection moulding process.. Nature of the problem: high scrap and
rework rate due to excessive process
variability.. Size of the experiment: eight trials or runs.. Benefits: zero defects were achieved.
Annual savings were estimated to be
above £40,000.
Automotive. Product type: diesel injector.. Nature of the problem: high rework rate.. Size of the experiment: sixteen trials or
runs.. Benefits: annual savings were estimated to
be over £10,000.
Metal fabrication. Type of process: welding.. Nature of the problem: low welding
strength of tin coated wires to a
connector.. Size of the experiment: sixteen trials or
runs.. Benefits: process capability index
increased from 0.50 to 2.5. Annual
savings were estimated to be £16,000.
Process. Type of process: chemical process.. Nature of the problem: low process yield.. Size of the experiment: eight trials.. Benefits: process yield was improved by
over 10 per cent.
Electronics and semi-conductors. Type of process: wire bonding process.. Nature of the problem: low wire pull
strength and therefore large customer
returns were experienced by the
company.. Size of the experiment: 16 trials.. Benefits: the average pull strength has
increased by 30 per cent and therefore
customer returns have decreased from 18
per cent to nearly 2 per cent. Annual
savings were estimated to be over
£30,000.
Applications in service industry. Minimising the time to respond to
customer complaints.. Minimising errors on service orders. Reducing the service delivery time to
customers.. Reducing the length of stay in an
emergency room in hospitals and health
care institutions.. Comparing competitive strategies of
launching new products.
Getting into Taguchi
In order to apply Taguchi methods in
industry, one may require planning,
engineering, communication, statistical and
teamwork skills. Moreover, right people and
right environment are crucial for the effective
application of Taguchi methods for tackling
process and product quality problems. The
participation and commitment of top
management are also vital for the successful
implementation. The following key points
must be taken into account when introducing
Taguchi methodology into design and
production:. Do you get excessive variability in your
processes?. Do you understand your product and
processes?. Is your process performance on target?. Is your product performance robust
under various environmental conditions?. Do you have to set up tolerances on the
critical parts to minimise variability?
Role of Taguchi methods in total qualitymanagement
There are many definitions of quality but the
definition of quality proposed by Dr Taguchi
(Kolarik, 1995) is more relevant in terms of
working towards target performance of
product/process. Indeed working towards the
target performance reflects the continuous
improvement attitude. Although not explicitly
80
An application of experimental design for process optimization
Hefin Rowlands, Jiju Antony and Graeme Knowles
The TQM Magazine
Volume 12 . Number 2 . 2000 . 78±83
stated, Deming's definition implies that the
needs of the customer may change. Also with
the view that new customers may be added in
the future, this suggests a dynamic definition
of quality. Quality issues do not stay still and
therefore we must be prepared to adapt our
ideas and views on quality in order to meet
the ever-changing needs of our customers.
This idea is consistent with the approach of
continuous improvement.
The ethos of total quality management is
continuous improvement. In order to improve
our product/process quality, we need to
measure appropriate quality characteristic(s),
which are most critical to our customers. In
otherwords, we cannot simply manage any
process without measurement. The role of
Taguchi methods in TQM is to identify and
optimise the critical quality characteristics
which affect the final product/process
performance.
Taguchi case study
This section details one of the Taguchi
experiments that the students have
implemented at their workplace. In many
cases this would have been the first
application of such a technique in the
company. The results of the study have
provided a greater stimulus in terms of the
wider application of the Taguchi methods in
other core processes. The objective of the
study was to optimise the production process
of retaining a metal ring in a plastic body in a
braking system by a hot forming method.
The production process consisted of a
heated die, which was forced down by air
pressure onto a valve body forming a plastic
lip into which a retaining metal ring was
inserted. Although the process was fairly
straightforward it was felt that the maximum
strength of the product was not being
achieved. A test rig was designed to simulate
the production process and to enable a series
of experiments to be performed. A Taguchi
style experiment was carried out to identify
the process variables that would provide the
greatest and consistent pull-out strength. This
was tested on a standard tensometer.
A brainstorming session consisting of chief
development engineers, senior design
engineers, plastic engineers, identified a list of
factors which were thought to affect the pull-
out strength. These factors were then
categorised as control or noise factors as
shown below:. Control factors: die temperature, hold
time, batch number, maximum force
during hot forming and force application
rate.. Noise factors: dimensional variation of
valve body, depth of stake, material
variation and rate of hot forming.. Control/noise factor: strain rate during pull-
out.
All five control factors were used for the
experiment. The strain rate pull-out was used
as a noise factor and controlled at two levels
for the experiment to simulate a varying load
on the product when in use. Due to the
difficulties and expense of manufacturing
valve bodies to different dimensions and with
variations in material, it was decided not to
include these in the experiment.
The levels for each factor were selected
systematically by the engineers based on their
knowledge and experience of the process. The
levels chosen are shown in Table I. As a
reference for the results of the experiment,
based on the experience of the engineers, the
following levels were considered to give the
best performance: A2, B1, C1, D1, E1.
The experiment was carried out using a
modified L8 orthogonal array (Taguchi and
Konishi, 1987) to account for the four levels
of factor A. The S/N calculations are based on
a larger the better S/N ratio.
Table II illustrates the average pull-out
values and average S/N ratio values
corresponding to each level of the chosen
factors.
Figure 1 illustrates the mean response
graph which simply plots the average pull-out
strength values at each level of the selected
factors. Figure 2 illustrates the S/N ratio plot
which basically shows the average S/N ratio
values at each factor level.
In order to determine which of the effects
are statistically significant, it was decided to
perform the Analysis of Variance (ANOVA).
ANOVA is a powerful technique which
Table I Factor levels
Factor Level 1 Level 2 Level 3 Level 4
A Die temperature (deg. C) 180 200 220 240
B Hold time (sec) 5 15
C Batch no. 1 2
D Maximum force (KN) 6 7
E Force application rate (KN/sec) 5 1
81
An application of experimental design for process optimization
Hefin Rowlands, Jiju Antony and Graeme Knowles
The TQM Magazine
Volume 12 . Number 2 . 2000 . 78±83
sub-divides the total variation in the data into
useful and meaningful components of
variation. The results of the ANOVA are
shown in Tables III and IV respectively.
In order to determine the significant effects,
the calculated F-ratios are compared with the
tabled F-ratios. For the S/N analysis, from F
tables, F0.05,1,2 = 18.51, F0.05,3,2 = 19.16,
F0.10,1,2 = 8.53 and F0.10,3,2 = 9.16. This
indicates that only factor B is statistically
significant at the 90 per cent confidence level.
For the mean analysis, F0.05,1,17 = 4.45,
F0.05,3,17 = 3.20, F0.10,1,17 = 3.03 and
F0.10,3,17 = 2.44. This clearly indicates that
factors A, B, C and D have significant effect
on the mean pull-out strength.
Determination of optimal factorsettings
The optimal settings are those which provide
the best process/product performance based
on the obtained data from the experiment. In
this case, the objective was to maximise the
pull-out strength with minimum variation. As
only factor B has significant effect on the S/N
ratio, we have chosen level 2 of factor B, as it
provides a higher S/N ratio. It is important to
note that the higher the S/N ratio, the greater
the process robustness and product
performance. Having analysed the S/N ratio,
the next step was to determine the factor
settings that yielded the maximum pull-out
strength. In this case, we have again chosen
those factor settings with higher mean pull-
out strength values. The analysis of mean
pull-out strength values has provided the
following factor settings:. Factor A ± level 4;. Factor B ± level 2;. Factor C ± level 2; and. Factor D ± level 2.
The predicted average pull-out strength
obtained based on the optimal factor settings
is approximately 4.5KN. A confirmation run
gave results close to the prediction. This
shows a significant improvement on the pull-
out strength compared to the average pull-out
strength value of 3.26, i.e. an increase of
about 38 per cent.
Table II Response table ± mean pull-out strength and mean S/N ratio
Factor levels Average (KN) Average S/N ratio
A1 2.72 8.44
A2 2.93 9.27
A3 3.29 10.30
A4 4.12 12.26
B1 2.92 9.09
B2 3.60 11.04
C1 3.15 9.65
C2 3.38 10.49
D1 3.15 9.68
D2 3.38 10.45
E1 3.26 10.02
E2 3.27 10.12
Figure 1 Mean reponse plot for the experiment
Figure 2 Mean S/N ratio plot for the experiment
Table III Results of ANOVA on the raw data
Factor Sum of squares
Degree of
freedom Mean square F-ratio
Per cent
contribution
B 2.77 1 2.77 87.08 25.35
A 6.87 3 2.29 71.88 62.63
C 0.34 1 0.34 10.55 2.81
D 0.29 1 0.29 9.25 2.43
Pooled error 0.54 17 0.03 ± ±
Total 10.82 23 0.47 ± 100
82
An application of experimental design for process optimization
Hefin Rowlands, Jiju Antony and Graeme Knowles
The TQM Magazine
Volume 12 . Number 2 . 2000 . 78±83
Conclusions
This paper has illustrated a practical
application of Taguchi method in industry.
The study has shown a significant
improvement (approximately 38 per cent) in
pull-out strength and thereby encouraged the
company in a wider application of the method
in other processes. The main feature, which
resulted in the success of this study, was the
careful experimental planning. A wide range
of personnel was involved in the
brainstorming exercise to identify the most
appropriate parameters for the experiment.
The potential benefit of the Taguchi method
to industry is vast. The paper demonstrates
the potential applications of Taguchi methods
in both manufacturing and service industry
sectors. It is important to bear in mind that
the successful application of Taguchi
methods requires planning, engineering,
statistical, communication and teamwork
skills. Management commitment and active
participation are crucial for the
implementation of such methods for tackling
product and process quality problems. The
real benefits are achievable when the method
is used in conjunction with other tools and
techniques and integrated into the continuous
improvement programme of the company.
For companies which have not yet started on
this path, now is the time to start. It is hoped
that this paper will stimulate more companies
to take up the challenge and implement a
programme of Taguchi experiment as part of
their continuous improvement programme.
References
Antony, J. (1996), `̀ Likes and dislikes of Taguchimethods'', Journal of Productivity, Vol. 37 No. 3,October-December, pp. 477-81.
Coleman, D.E. and Montgomery, D.C. (1993), `̀ A
systematic approach to planning for a designed
industrial experiment'', Technometrics, Vol. 35
No. 1, pp. 1-27.Dooley, K.J. and Mahmoodi, F. (1992), `̀ Identification
of robust scheduling heuristics: application of
Taguchi methods in simulation studies'', Computers
and Industrial Engineering, Vol. 22 No. 4,
pp. 359-68.Fox, R.T. and Lee, D. (1990), `̀ Optimization of metal
injection moulding: experimental design'', The
International Journal of Powder Metallurgy, Vol. 26
No. 3, pp. 233-43.Kolarik, W.J. (1995), Creating Quality, Concepts, Systems,
Strategies and Tools, McGraw Hill, New York, NY.Kolarik, W.J. and Pan, J.N. (1991), `̀ Proactive quality:
concept, strategy and tools'', Proceedings of the
International Industrial Engineering Conference,
pp. 411-20.Lochner, R.H. (1991), `̀ Pros and cons of Taguchi'', Quality
Engineering, Vol. 3 No. 4, pp. 537-49.Logothetis, N., Atkinson, C.J., Salmon, J.P. and Best, K.F.
(1990), `̀ Development of newly installed
processes'', International Journal of Advanced
Manufacturing Technology, Vol. 5, pp 256-74.McKeown, P. (1992), `̀ Implementing quality improvement
programmes'', Robotics & Computer
Integrated Manufacturing, Vol. 9 No. 4/5,
pp. 311-20.Nair, V.N. (Ed.) (1992), `̀ Taguchi's parameter design: a
panel discussion'', Technometrics, Vol. 34 No. 2,
pp. 127-61.Pignatiello, J.J. (Jr) and Ramberg, J.S. (1991), `̀ Top ten
triumphs and tragedies of Genichi Taguchi'', Quality
Engineering, Vol. 4 No. 2, pp. 211-25.Ross, P.J. (1988), Taguchi Techniques for Quality
Engineering, McGraw Hill, New York, NY.Taguchi, G. and Konishi, S. (1987), Orthogonal Arrays and
Linear Graphs, ASI Press.Wu, C.M., Black, J.T and Jiang, B.C. (1991), `̀ Using
Taguchi methods to determine optimize robot
process capability for path following'', Robotics &
Computer-Integrated Manufacturing, Vol. 8 No. 1,
pp. 9-25.
Table IV Results of ANOVA on the S/N ratio data
Factor Sum of squares
Degree of
freedom Mean square F-ratio
Per cent
contribution
B 7.64 1 7.64 12.52 26.56
A 16.28 3 5.43 8.90 54.41
D 1.42 1 1.42 2.33 3.04
Pooled error 1.22 2 0.61 ± 15.99
Total 26.56 7 3.8 ± 100
Commentary
A detailed exploration of the Taguchi method and its role in process optimization.
83
An application of experimental design for process optimization
Hefin Rowlands, Jiju Antony and Graeme Knowles
The TQM Magazine
Volume 12 . Number 2 . 2000 . 78±83