Introduction
Teaching the Taguchi
method to industrial
engineers
Jiju Antony and
Frenie Jiju Antony
The authors
Jiju Antony is at the International Manufacturing Centre,
Department of Engineering, University of Warwick,
Coventry, UK.
Frenie Jiju Antony is at the School of Management
Studies, Cochin University of Science and Technology,
Kerala, India.
Keywords
Taguchi methods, Statistical process control,
Design of experiments
Abstract
The Taguchi method (Tm) is a powerful problem solving
technique for improving process performance, yield and
productivity. It reduces scrap rates, rework costs and
manufacturing costs due to excessive variability in
processes. However, its application by industrial engineers
in the UK is limited, in part due to the inadequate
statistical education of engineers. This paper presents a
simple experiment which can be used in the classroom to
teach engineers the basics of the technique and illustrates
simple analytical and graphical tools which promote rapid
understanding of the results of the experiment.
Electronic access
The research register for this journal is available at
http://www.mcbup.com/research_registers
The current issue and full text archive of this journal is
available at
http://www.emerald-library.com/ft
Work Study
Volume 50 . Number 4 . 2001 . pp. 141±149
# MCB University Press . ISSN 0043-8022
141
Dr Genichi Taguchi is a Japanese quality
management consultant who has developed
and promoted a philosophy and methodology
for continuous quality improvement in
products and processes. Within this
philosophy, Taguchi shows how the statistical
design of experiments (SDOE or DOE) can
help industrial engineers design and
manufacture products that are both of high
quality and low cost. His approach is
primarily focused on eliminating the causes of
poor quality and on making product
performance insensitive to variation. DOE is a
powerful statistical technique for determining
the optimal factor settings of a process and
thereby achieving improved process
performance, reduced process variability and
improved manufacturability of products and
processes.
Taguchi (1986) advocates the use of
orthogonal array designs to assign the factors
chosen for the experiment. The most
commonly used orthogonal array designs are
L8 (i.e. eight experimental trials), L16 and
L18. The power of the Taguchi method is
that it integrates statistical methods into the
engineering process. Bendell et al. (1989) and
Rowlands et al. (2000) report success of the
Taguchi method in the automotive, plastics,
semiconductors, metal fabrication and
foundry industries. However, Antony (1996)
suggests that the application of the Taguchi
method in the UK manufacturing and service
industries is limited and often applied
incorrectly. Moreover, a typical remark is ‘‘I
can do the text book and class room
examples, but I am not comfortable and
confident in applying the concepts and
principles of DOE in my work area’’.
According to Antony et al. (Antony et al.
1996a; 1998a; 1998b; 1999; Antony, 1998)
the following issues are key to this lack of or
improper application of experimental design
techniques based on the Taguchi method:
The word ‘‘statistics’’ invokes fear in
many industrial engineers. Many
engineers in the UK leave universities
without a complete understanding of the
power of statistics and are therefore likely
to avoid the use of statistical techniques
in their subsequent careers.
Few graduating engineers have been
exposed to applied statistical quality
techniques such as DOE, robust design,
 Teaching the Taguchi method to industrial engineers Work Study
Jiju Antony and Frenie Jiju Antony Volume 50 . Number 4 . 2001 . 141±149

etc. This is another symptom of the plastics, automotive, process, metal

statistical education of the engineering
fraternity.
Engineers consistently avoid the use of
applied statistical techniques in tackling
process optimisation and quality control
problems. Where techniques are in use,
e.g. the use of control charts for process
analysis and monitoring, there often
appears to be a lack of a full
understanding of the basic and
fundamental principles behind their
application (Morrison, 1997).
Many textbooks and courses on DOE
primarily focus on the statistical analysis of
the problem under study. However, this is
but one component of DOE which involves
planning, design, execution, analysis and
interpretation of results.
A lack of communication between the
academic and industrial worlds, and
between functional specialists restricts the
application of the Taguchi method
(Tm)and DOE (Antony et al., 1998a). It
is important, though too rare, that
quality, manufacturing, process, design
and operational departments
communicate and work effectively with
one another.
fabrication, food and electronics and semi-
conductor sectors (Rowlands et al., 2000).
Typical applications in service industry
The use of Tm in service industries is not
often reported. This may be because :
service performance is often more
difficult to measure;
the performance of a service process
depends a great deal on the behaviour
and attitude of the service provider and it
varies with time; and
the identification and measurement of
control factors and their influence on
performance characteristic(s) is often
difficult.
However, there clearly are possible applications
of Tm in the service sector. Examples include:
reducing the time taken to respond to
customer complaints;
reducing errors on service orders; and
reducing the length of stay in an
emergency room in hospital.
If the use of Tm is to become more prevalent,
ways must be found to teach engineers (and
others) effectively how to apply it successfully.

Potential applications and benefits of

using the Taguchi method
The Taguchi method has wide application in
manufacturing organisations. Table I
illustrates the application of Tm in the
Steps in performing a Taguchi
experiment
The process of performing a Taguchi
experiment follows a number of distinct steps:

Table I Typical applications of Tm in manufacturing

Process/product Nature of problem
Injection moulding High scrap rate due to
process excessive process variability
Diesel injector High rework rate
Welding process Low weld strength
Chemical process Low process yield
Biscuit Excessive variability in
biscuit length
Wire-bonding process Low wire pull strength
Experiment size Benefits
8 trials Annual savings were
estimated to be over
£40,000
16 trials Annual savings were
estimated to be over
£10,000
16 trials Annual savings were
estimated to be over
£16,000
8 trials Process yield was improved
by over 10 per cent
16 trials Biscuit length variability was
reduced by over 25 per cent
16 trials Annual savings were over
£30,000

142
 Teaching the Taguchi method to industrial engineers
Jiju Antony and Frenie Jiju Antony
Step1: formulation of the problem – the
success of any experiment is dependent
on a full understanding of the nature of
the problem.
Step 2: identification of the output
performance characteristics most relevant
to the problem.
Step 3: identification of control factors,
noise factors and signal factors (if any).
Control factors are those which can be
controlled under normal production
conditions. Noise factors are those which
are either too difficult or too expensive to
control under normal production
conditions. Signal factors are those which
affect the mean performance of the
process.
Step 4: selection of factor levels, possible
interactions and the degrees of freedom
associated with each factor and the
interaction effects.
Step 5: design of an appropriate
orthogonal array (OA).
Step 6: preparation of the experiment.
Step 7: running of the experiment with
appropriate data collection.
Step 8: statistical analysis and
interpretation of experimental results.
Step 9: undertaking a confirmatory run of
the experiment.
Paper helicopter experiment
In many academic institutions within the UK,
the focus of engineering statistics is on the
theory of probability (for example, card
shuffling, dice rolling, etc.), the mathematical
aspects of probability and probability
distributions (e.g. normal, exponential,
binomial, Poisson, log-normal, etc.),
hypothesis tests, etc. Quality improvement
techniques (DOE, Tm, SPC, etc.) are often
not covered. Understandably, graduates are
not confident about using such techniques at
their place of work.
As part of an exercise to increase the
awareness of Tm amongst industrial
engineers, the authors used a simple paper
helicopter experiment, readily used in
academic institutions. Due to a limited
amount of time, one member from each
group in the class was involved with the
experimental work. However, the students
were all asked to analyse and interpret the
data (on an individual basis). The results of
143
Work Study
Volume 50 . Number 4 . 2001 . 141±149
the analysis were discussed in the classroom
as part of the process of gaining an
understanding of experimental objectives and
process.
The paper helicopter experiment is quite
well known among engineers and statisticians
in both the academic and industrial worlds.
Many industrial training programmes on Tm
use it in some form. However they often focus
on the design and analysis of the experiment
without providing guidance to engineers on
the interpretation of results from the analysis.
Moreover, many courses do not cover the
importance of careful experimental planning
for the success of any industrially designed
experiment.
The purpose of this experiment was to
provide undergraduate engineering students
with an understanding of the role of
Taguchi’s ‘‘parameter design’’ (sometimes
called ‘‘robust design’’) in tackling both
product and process quality-related problems
in real-life situations. Parameter design is a
well established methodology for improving
product and process quality at minimal cost
by reducing the effect of undesirable external
influences which cause variation in product or
process performance (Phadke, 1989).
The objective of the exercise was to identify
the optimal settings of control factors which
would maximise the flight time of paper
helicopters (with minimum variation). Here
control factors refer to those which can be
easily controlled and varied by the designer or
operator in normal production conditions. A
brainstorming session by a group of students
identified six control factors which were
thought to influence the time of flight (refer to
Table II). Brainstorming should be
considered an integral part of the Taguchi
methodology – it is a useful technique in
identifying the most influential factors in an
experiment.
In order to simplify the experiment, each
factor was studied at two levels. The ‘‘level’’
of a factor here refers to the specified value of
Table II Control factors and their range of settings for the experiment
Control factor Labels Level 1 Level 2
Paper type A Regular Bond
Body length B 8cm 12cm
Wing length C 8cm 12cm
Body width D 2cm 3cm
Number of clips E 1 2
Wing shape F Flat Angled
 Teaching the Taguchi method to industrial engineers
Jiju Antony and Frenie Jiju Antony
a setting. For example, in the experiment,
body width was studied at 2cm and 3cm.
Factors at three (and higher) levels make
analysis more complicated – and are therefore
not used in awareness-raising sessions.
Having identified the control factors, it is
important to list the interactions which are to
be studied for the experiment. Interaction
exists when the effect of one factor is not the
same at different levels of the other factor. An
effect refers to the change in response due to
the change in level of a factor (Antony et al.,
1998b). Consider, for example, the factors
wing length and body length of the paper
helicopter. Assume each factor was kept at
two-levels for the study. Time of flight is the
response (or quality characteristic) of interest.
Interaction between wing length and body
length exists when the effect of wing length on
time of flight at two different levels of body
length is different.
For this experiment, three interactions were
identified (from the brainstorming session) as
being of interest:
(1) body length £ wing length (B £ C or
BC);
(2) body length £ body width (B £ D or
BD); and
(3) paper type £ body length (A £ B or AB).
The following noise factors were identified (as
having some impact on the flight time but
being difficult to control):
operator-to-operator variation;
draughts;
reaction time; and
ground surface.
One aim was to determine the control factor
settings which would best dampen the effect
of these noise factors. According to Taguchi,
there is an optimal combination of factor
settings which counters the effects of noise. In
order to minimise the effect of these noise
factors, the same student was responsible for
all timings – reducing the effects of variable
reaction times when hitting the stopwatch
upon release of the helicopter and its hitting
the ground.
Figure 1 illustrates a template for the model
of a paper helicopter which can be made from
an A4 size paper. It forms the basis of a simple
experiment, requiring only simple items such
as paper, scissors and paper clips. It takes
about six hours to design the experiment,
collect the data and then perform the
statistical analysis (with the ‘‘experiment’’
144
Work Study
Volume 50 . Number 4 . 2001 . 141±149
Figure 1 Template for paper helicopter design
itself taking about 90 minutes). In this case,
the statistical analysis was executed as a
homework assignment though the results
were discussed in the classroom in detail.
Choice of orthogonal array design
The choice of a suitable orthogonal array
(OA) design is critical for the success of an
experiment and depends on the total degrees
of freedom required to study the main and
interaction effects, the goal of the experiment,
resources and budget available and time
constraints. Orthogonal arrays allow one to
compute the main and interaction effects via a
minimum number of experimental trials
(Ross, 1988). ‘‘Degrees of freedom’’ refers to
the number of fair and independent
comparisons that can be made from a set of
observations. In the context of SDOE, the
number of degrees of freedom is one less than
the number of levels associated with the
factor. In other words, the number of degrees
of freedom associated with a factor at p-levels
is (p-1). As the number of degrees of freedom
associated with a factor at two levels is unity,
in the present example the number of degrees
of freedom for studying the six main effects is
equal to six. The number of degrees of
freedom associated with an interaction is the
product of the number of degrees of freedom
associated with each main effect involved in
the interaction (Antony, 1998). In this simple
case, the number of degrees of freedom for
studying the three interaction effects is equal
to three. Therefore the total degrees of
freedom is equal to nine (i.e. 6 + 3). It is
important to notice that the number of
 Teaching the Taguchi method to industrial engineers Work Study
Jiju Antony and Frenie Jiju Antony Volume 50 . Number 4 . 2001 . 141±149

experimental trials must be greater than the
total degrees of freedom required for studying
the effects. The standard OAs for factors with
two levels are L4 L8 , L1 6 , L32 and so on. Here
the notation ‘‘L’’ implies that the information
is based on the Latin square arrangement of
factors. A Latin square arrangement is a
square matrix arrangement of factors with
separable factor effects. Here the numbers 4,
8, 12, 16, etc. denote the number of
experimental trials. For the helicopter
experiment, as the total degrees of freedom is
equal to nine, the closest number of
experimental trials that can be employed for
the experiment is 16 (i.e. L1 6 OA). Having
identified the most suitable OA, the next step
was to assign the main and interaction effects
to various columns of the array. A standard
L16 OA (see Appendix) contains 15 columns
for either studying 15 main effects, or a
combination of main and interaction effects
so that the degrees of freedom will add up to
15. In the present example, there are only six
main and three interaction effects. This
means that only nine columns out of 15 are
used. For example, factor D (refer to Table
III) was assigned to column 1 and factor C to
column 2. Column 3 is empty (see Table III)
as the interaction between these factors was of
no interest in this experiment. Using the
standard linear graphs and OA (Ross, 1988),
the remaining factors and interactions were
assigned to the columns of an L1 6 in the
following manner.
Column 1 – body width (D), column 2 –
wing length (C), column 4 – body length (B),
column 5 – body width £ body length (B £
D), column 6 – wing length £ body length (B
£ C), column 7 – wing shape (F), column 8 –
paper type (A), column 12 – body length £
paper type (AB) and column 14 – number of
clips (E).
The experimental layout showing all the
factors and interactions along with the flight
times (measured in seconds) is shown in
Table III. As each factor was studied at two
levels, coded level 1 represents the low level of
a factor setting and level 2 represents the high
level setting. Each experiment was replicated
in order to capture variation in results due to
uncontrolled noise.
Statistical analysis and interpretation of
results
In Taguchi’s parameter design, the basic
objective is to identify the conditions which
optimise process/product performance. In
arriving at this optimal set of conditions,
Taguchi advocates the use of signal-to-noise
ratio (SNR) – the need is to maximise the
performance of a system or product by
minimising the effect of noise while
maximising the mean performance. The SNR
is treated as a response (output) of the
experiment, which is a measure of variation
when uncontrolled noise factors are present in

Table III Experimental layout

Column no. 1 2 4 5 6 7 8 12 14
Factors/interactions D C B BD BC F A AB E Flight time
Trial no.
1 1 1 1 1 1 1 1 1 1 2.76, 2.83
2 1 1 1 1 1 1 2 2 2 2.20, 2,13
3 1 1 2 2 2 2 1 2 2 1.93, 2.30
4 1 1 2 2 2 2 2 1 1 2.19, 2.10
5 1 2 1 1 2 2 1 1 2 2.40, 2.50
6 1 2 1 1 2 2 2 2 1 2.82, 2.31
7 1 2 2 2 1 1 1 2 1 3.39, 3.01
8 1 2 2 2 1 1 2 1 2 2.62, 2.39
9 2 1 1 2 1 2 1 1 1 2.46, 2.12
10 2 1 1 2 1 2 2 2 2 2.08, 1.90
11 2 1 2 1 2 1 1 2 2 2.14, 2.29
12 2 1 2 1 2 1 2 1 1 2.05, 2.12
13 2 2 1 2 2 1 1 1 2 2.96, 2.70
14 2 2 1 2 2 1 2 2 1 2.47, 2.60
15 2 2 2 1 1 2 1 2 1 2.62, 2.91
16 2 2 2 1 1 2 2 1 2 2.32, 2.41

145
 Teaching the Taguchi method to industrial engineers Work Study
Jiju Antony and Frenie Jiju Antony Volume 50 . Number 4 . 2001 . 141±149

the system (Antony et al., 1999). Taguchi has
developed and defined over 60 different
SNRs for engineering applications of
parameter design. For the present study, as
the objective was to maximise time of flight it
was decided to select the SNR related to
larger-the-better (LTB) quality
characteristics. This is generally used for
quality characteristics such as strength, fuel
efficiency, process yield, life of a component
and so on. For LTB quality characteristics,
the SNR is given by the following equation:
£1 1¤
SNR ˆ ¡10log ¤ …1 †
n y2i
where n = number of values at each trial
condition (i.e., 2 from Table II) and yi = each
observed value.
Table IV illustrates the SNR values (based
on equation 1) corresponding to each trial
condition.
Table V illustrates the average SNR values
(SNR) at low (level 1) and high (level 2) levels
and the effect of each main and interaction
effect on the SNR.
Sample calculation for factor ‘‘C’’:
Average SNR at level 1 of factor ‘‘C’’ =
SNRC 2 = 1/8 * [8.93 + 6.71 + 6.41 + 6.62
+7.12 + 5.95 + 6.89 + 6.38]
= 6.88
Similarly, average SNR at level 2 of factor
‘‘C’’ = SNRC 2 = 1/8 * [7.78 + 8.05 + 10.06 +
7.95 + 9.01 + 8.07 + 8.80 + 7.47]
= 8.40
Effect = SNRC 2 - SNRC 1
= 8.40 - 6.88 = 1.52
Table IV SNR table
Trial number SNR Trial number
1 8.93 9 7.12
2 6.71 10
3 6.41 11
4 6.62 12
5 7.78 13
6 8.05 14
7 10.06 15
8 7.95 16
The other main and interaction effects were
calculated in a similar manner (see Table V).
Having obtained the average SNR values,
the next step is the identification of significant
main and interaction effects which influence
the SNR. To achieve this, a powerful
graphical tool called half-normal probability
plots (HNPP) is useful.
A half-normal probability plot (HNPP) is
obtained by plotting the absolute values of the
effects (both main and/or interaction effects)
along the X-axis and the per cent probability
along the Y-axis. The per cent probability
can be obtained by using the following
equation:
…i ¡ 0:5†
Pi ˆ £ 100 …2 †
n
where: n = number of estimated effects
(n = 15) and i is the rank of the estimated
effect when arranged in the ascending order of
magnitude (e.g. for factor C, i = 15).
Figure 2 illustrates the HNPP of the factor
and interaction effects for the helicopter
experiment. The computer software package
‘‘Design-ease’’ was used to construct the plot.
Those effects which are active and real will
fall off the straight line, whereas the inactive
and insignificant effects will fall along the
straight line (Daniel, 1959). The figure
reveals that main effects A, C, E and F are
statistically significant. i.e. paper type, wing
length, number of clips and wing shape are
statistically significant. In order to support
and justify this claim, another graphical tool
(main effects plot) is used. This shows the
average SNR values at low and high level
settings of each factor. Figure 3 illustrates the
main effects plot for the paper helicopter
SNR experiment (using the values from Table V).
This graphical aid provides non-statisticians
with a better picture of the importance of the
5.95 effects of the chosen control factors. The
6.89 slope of the line is an indication of the
6.38 importance of a main or interaction effect.
9.01 The figure shows that the most dominant
8.07 factor is the wing length, followed by paper
8.80 type, wing shape and number of clips. As each
7.47 factor was chosen at two levels, the effect of

Table V Average SNR table

Factors or interactions D C B BD BC F A AB E
SNR1 7.81 6.88 7.70 7.63 7.87 8.00 8.12 7.66 8.00
SNR2 7.46 8.40 7.57 7.65 7.40 7.27 7.15 7.62 7.28
Effect estimate ±0.35 1.52 ±0.13 0.02 ±0.47 ±0.73 ±0.97 ±0.04 ±0.72

146
 Teaching the Taguchi method to industrial engineers
Jiju Antony and Frenie Jiju Antony
Figure 2 Half-normal plot of effects
Figure 3 Main effects plot of the control factors
each factor must be assumed to be linear. If
non-linear effects are to be studied, it is
necessary to choose more than two levels for
each factor. However, it is good practice to
start off an experiment with two levels and
then perform smaller sequential experiments
at higher levels to gain a better understanding
of the nature of the process.
For this experiment, none of the interaction
effects is significant. Consider, for example,
the interaction between the body length and
body width. In order to compute this
interaction, the first step is to compute the
average SNR values at each of the four
combinations of the factor levels. Table VI
shows the average SNR values for these four
combinations.
147
Work Study
Volume 50 . Number 4 . 2001 . 141±149
An interaction plot is useful in providing a
rapid understanding of the nature of
interactions (Schmidt and Launsby, 1992).
Interaction plots are constructed by plotting
the average response values (in this case, SNR
values) at each factor level combination.
Parallel lines are an indication of the absence
of interaction between the factors, whereas
non-parallel lines are an indication of the
presence of interaction between the factors.
Figure 4 shows that the effect of body width
on the flight time at both levels of body length
is the same. In other words, the effect of body
width on the flight time is the same,
irrespective of the level of body length. This
implies the absence of interaction between
these two factors.
Determination of the optimal control
factor settings
The selection of optimal settings depends on
the objective of the experiment or the nature
of the problem under study. For the
helicopter example, the objective was to
maximise the flight time. In Taguchi
experiments, the objective is to identify the
factor settings which yield the highest SNR –
these settings will generally produce a
consistent and reliable product. Moreover,
the process which produces the product will
Table VI Average SNR values
Body length Body width Average SNR
1 1 7.87
1 2 7.54
2 1 7.76
2 2 7.39
Figure 4 Interaction plot between body length and body width
 Teaching the Taguchi method to industrial engineers
Jiju Antony and Frenie Jiju Antony
be insensitive to various sources of
uncontrollable variation. For the paper
helicopter experiment, the optimal control
factor settings based on the highest SNR have
been determined. These are shown in Table
VII. In order to decide which level is better for
maximising flight time, the SNR values at
both low (level 1) and high (level 2) levels of
each factor are compared.
Once the optimal settings are established, it
is useful to undertake a confirmation trial
before onward actions are undertaken
(Antony, 1996). Three helicopters were made
using the optimal factor settings and the
average flight time was recorded as 3.56
seconds. This shows an improvement of
above 30 per cent on the average flight time
using the range of variable settings. The
results also reveal that flight time increases for
larger wing length and smaller body length.
Summary and conclusions
The experiment was carried out with the aim
of optimising the flight time of a paper
helicopter. In order to study the effect of
variables and the possible interactions
between them in a minimum number of trials,
the Taguchi approach to experimental design
was adopted. As the experiment itself was
simple, the students found it to be a clear
illustration of the process of :
defining the problem;
identifying the control variables and
possible interactions;
defining the required levels for each
variable/factor;
determining the response of interest;
selecting the most suitable orthogonal
array;
performing the experiment;
undertaking the analysis; and
interpreting the results to obtain a better
understanding of the situation under
review.
Table VII Optimal control factor settings
Control factors Optimum level
Paper type Regular (level 1)
Body length 8cm (level 1)
Wing length 12cm (level 2)
Body width 2cm (level 1)
Number of clips
1 (level 1)
Wing shape Flat (level 1)
148
Work Study
Volume 50 . Number 4 . 2001 . 141±149
The Taguchi method is a powerful
approach to address process variability and
optimisation problems. However, the
application of SDOE and Tm by the
engineering fraternity in UK organisations
is limited due, in part, to a shortage of skills
in problem solving and inadequate
statistical knowledge. This paper
demonstrates a simple means of introducing
students to this powerful tool. The
approach uses a simple paper helicopter
experiment. For simplicity, all control
parameters were studied at two levels. This
mirrors actual practice – in most
optimisation problems, factors at two levels
are the most widely used (Gunst and
Mason, 1991; Lucas, 1992). The paper
helicopter experiment is quite old and has
been widely used by many statisticians for
teaching purposes. However, this approach
has focused on minimal statistical jargon
and number crunching and on the use of
modern graphical tools to achieve a rapid
understanding of the results from the
statistical analysis. The authors strongly
believe that the experiment provides a
simple and beneficial way to help engineers
approach experimental design in a way that
ensures it is transferrable to their own work
environment.
References
Antony, J. (1996), ``A strategic methodology to the use of
advanced statistical quality control techniques’’,
PhD thesis.
Antony, J. (1998), ``Some key things industrial engineers
should know about experimental design’’, Logistics
Information Management, 1998, Vol. 11 No. 6,
pp. 386-92.
Antony, J. et al. (1996), ``Optimisation of core tube life
using Taguchi experimental design methodology’’,
Journal of Quality World (Technical Supplement),
IQA, March, pp. 42-50.
Antony, J. et al. (1998a), ``A strategic methodology to the
use of advanced statistical quality improvement
techniques’’, The TQM Magazine (The International
Bi-Monthly for TQM), Vol. 10 No. 3, pp. 169-176.
Antony, J. et al. (1998b), ``Key interactions’’, Journal of
Manufacturing Engineer, IEE, Vol. 77 No. 3,
pp. 136-8.
Antony, J. et al. (1999), Experimental Quality ± A Strategic
Approach to Achieve and Improve Quality, Kluwer
Academic Publishers, Dordrecht, December.
Bendell, A. (Ed.) (1989), Taguchi Methods Applications in
World Industry, IFS Publications, Bedford.
Daniel, C. (1959), ``Use of half-normal plots in interpreting
factorial two level experiments’’, Technometrics,
Vol. 1 No. 4, pp. 53-70.
 Teaching the Taguchi method to industrial engineers Work Study
Jiju Antony and Frenie Jiju Antony Volume 50 . Number 4 . 2001 . 141±149

Gunst, R.F. and Mason, R.L. (1991), How to Construct
Fractional Factorial Experiments, ASQC Statistics
Division, ASQC Press, Milwaukee, MI.
Lucas, J.M. (1992), ``Split plotting and randomisation in
industrial experiments’’, ASQC Quality Congress
Transactions, Nashville, TN, pp. 374-82.
Morrison, J.M. (1997), ``Statistical engineering ± the key
to quality’’, Engineering Science and Education
Journal, pp. 123-7.
Phadke, M.S. (1989), Quality Engineering using Robust
Design, Prentice-Hall International, Englewood
Cliffs, NJ.
Ross, P.J., (1988), Taguchi Techniques for Quality
Engineering, McGraw-Hill Publishers, New York, NY.
Rowlands, H., Antony, J. and Knowles, G. (2000), ``An
application of experimental design for process
optimisation’’, The TQM Magazine, Vol. 12 No.2,
pp. 78-83.
Schmidt, S.R and Launsby, R.G. (1992), Understanding
Industrial Designed Experiments, Air Academy
Press, Washington, DC.
Taguchi, G. (1986), Introduction to Quality
Engineering, Asian Productivity Organisation,
Tokyo.

Appendix
Table AI Coded design matrix of an L16 (21 5 ) orthogonal array
Column
Trial 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2
3 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2
4 1 1 1 2 2 2 2 2 2 2 2 1 1 1 1
5 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2
6 1 2 2 1 1 2 2 2 2 1 1 2 2 1 1
7 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1
8 1 2 2 2 2 1 1 2 2 1 1 1 1 2 2
9 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2
10 2 1 2 1 2 1 2 2 1 2 1 2 1 2 1
11 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1
12 2 1 2 2 1 2 1 2 1 2 1 1 2 1 2
13 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1
14 2 2 1 1 2 2 1 2 1 1 2 2 1 2 1
15 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2
16 2 2 1 2 1 1 2 2 1 1 2 1 2 2 1

149