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

net/publication/228957785

Teaching experimental design techniques to industrial engineers

Article · January 1998

CITATIONS READS

25 3,599

2 authors:

Jiju Antony Nick Capon

Heriot-Watt University University of Portsmouth
428 PUBLICATIONS 19,759 CITATIONS 8 PUBLICATIONS 133 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Changing Patterns of Lean Six Sigma research View project

Lean Six Sigma View project

All content following this page was uploaded by Jiju Antony on 12 November 2015.

The user has requested enhancement of the downloaded file.

Int. J. Engng Ed. Vol. 14, No. 5, pp. 335±343, 1998 0949-149X/91 $3.00+0.00
Printed in Great Britain. # 1998 TEMPUS Publications.

Teaching Experimental Design Techniques

to Industrial Engineers*
JIJU ANTONY and NICK CAPON
Portsmouth Business School, University of Portsmouth, Milton Site, Portsmouth, PO4 8JF, UK.
E-mail: jiju.antony@port.ac.uk
Experimental Design (ED) is a powerful technique for understanding a process, studying the
impact of potential variables or factors affecting the process and thereby providing spontaneous
insight for continuous quality improvement possibilities. ED has proved to be very effective for
improving the process yield, process capability, process performance and reducing process
variability. However research has shown that the application of this powerful technique by the
engineering fraternity in manufacturing companies is limited due to lack of skills and expertise in
manufacturing, lack of statistical knowledge required by industrial engineers and so on. This paper
illustrates some of the recent research findings on the problems and gaps in the state-of-the-art in
ED. In order to bridge the gap in the statistical knowledge required by engineers, the article
presents a paper helicopter experiment which can be easily carried out in a class-room to teach
experimental design techniques. The results of the experiment have provided a greater stimulus for
the wider application of ED by industrial engineers in real-life situations for tackling quality
problems.

INTRODUCTION common remark made by many engineers is `I

can do the text book and class room examples
EXPERIMENTAL DESIGN (ED) is a strategy of but I am not comfortable while applying the
planning, conducting, analysing and interpreting concepts and principles of ED in my work area'.
experiments so that sound and valid conclusions These findings point to the following issues [6]:
can be drawn efficiently, effectively and eco-
nomically. It provides the experimenters a greater . Statistical education for engineers at university
understanding and power over the experimental level is generally inadequate. The courses cur-
process. ED has seen an increased application over rently available in engineering statistics often
fifteen years, as both manufacturing and service tend to concentrate on the theory of probability,
industries have attempted to refine and improve card shuffling, probability distributions and the
product, process and service quality. Experimental more mathematical aspects of the subject rather
Design technology is not new to industrial and than the techniques which are more practically
manufacturing engineers in today's modern busi- useful to the engineering fraternity. Thus many
ness environment. ED was developed in the early engineers would deem statistics as useless in
1920s by Sir Ronald Fisher at the Rothamsted their late careers in industries.
Agricultural Field Research Station in London, . The lack of communication between the indus-
England. After World War II, English prac- trial and the academic worlds restricts the appli-
titioners of experimental design brought it to cation of ED in many manufacturing and service
the US, where the chemical process industry was industries.
among the first to apply it [1]. . Lack of skills and expertise required by engi-
A number of successful applications of ED for neers in manufacturing, especially in problem
improving process performance, reducing process formulation and definition.
variability, improving process yield etc. have been . The existing methodologies in ED provide no
reported by many manufacturers over the last insight into problem analysis and classification.
fifteen years [2, 4]. Research has shown that the Thus many engineers experience difficulties in
application of ED techniques by the engineering analysing a particular process quality problem
fraternity in both manufacturing and service and then converting the engineering problem
industries is limited and when applied they are into statistical terms from which appropriate
often performed incorrectly [5]. In other words, solutions can be chosen.
there is a cognitive gap in the knowledge of . Current software systems and expert systems in
statistics required by engineers in using ED as ED often tend to concentrate on data analysis
a problem-solving tool. Moreover, the most and do not properly address interpretation of
data. Thus many industrial engineers having
* Accepted 15 August 1998. performed the statistical analysis would not

335
336 J. Antony and N. Capon

know what to do with the results without The following steps are useful while one may be
assistance from statistical consultants in the performing an industrial experiment;
field.
1. Definition of the objective of the experiment.
2. Selection of the response or output.
3. Selection of the process variables or design
BENEFITS OF EXPERIMENTAL DESIGN parameters (control factors), noise factors and
the interactions among the process variables of
ED enables industrial engineers to study the interest. (Noise factors are those which cannot
effects of several variables affecting the response be controlled during actual production condi-
or output of a certain process [7]. ED methods tions, but may have strong influence on the
have wide potential application in the engineering response variability. The purpose of an experi-
design and development stages. It is the strategy of menter is to reduce the effect of these undesir-
the management in today's competitive world able noise factors by determining the best factor
market to develop products and processes insensi- level combinations of the control factors or
tive to various sources of variation using ED. The design parameters. For example, in an injection
potential applications of ED in industries are: moulding process, humidity and ambient tem-
. reducing product and process design and perature are typical noise factors.)
4. Determination of factor levels and range of
development time;
. studying the behaviour of a process over a wide factor settings.
5. Choice of appropriate experimental design.
range of operating conditions;
. minimising the effect of variations in manu- 6. Experimental planning.
7. Experimental execution.
facturing conditions;
. understanding the process under study and 8. Experimental data analysis and interpretation.
thereby improving its performance;
. increasing process productivity by reducing PAPER HELICOPTER EXPERIMENT
scrap, rework etc.;
. improving the process yield and stability of an The following section describes the application
on-going manufacturing process; of ED for optimising the time of flight of a paper
. making products insensitive to environmental helicopter which can be made from A4-size paper.
variations such as relative humidity, vibration, The experiment was carried out by the first author
shock and so on; in a class-room for a post-graduate course in
. studying the relationship between a set of quality management at University of Portsmouth.
independent process variables (i.e., process The experiment requires paper, scissors, ruler,
parameters) and the output (i.e., response). paper clips and a measuring tape. It would take

Fig. 1. Model of a paper helicopter design.

 Teaching Experimental Design Techniques to Industrial Engineers 337

about 5 ±6 hours to design the experiment, collect experiment. The level of a parameter is the speci-
the data and then perform appropriate statistical fied value of a setting. For example, in the above
analysis. The model of a paper helicopter design is injection moulding experiment, 2108C and 2508C
shown in Fig. 1. are the low and high levels of mould temperature.
The objective of the experiment was to deter- It was decided to set each parameter at two levels
mine the optimal settings of the design parameters or values as this forms the building block for
which would maximise the time of flight. For the studying parameters at three and higher levels.
application of ED in solving process and product Design or process parameters at three levels are
quality problems, it is essential that the objective of more complicated to teach in the first place and
the experiment must be specified clear, brief and moreover the authors strongly believe that it might
concise. Having defined the objective of the experi- turn off engineers from learning ED any further
ment, the possible parameters which might influ- [9]. It is usually best to experiment with the largest
ence the time of flight were determined through a range feasible to observe the effect of a design
thorough brainstorming session. These parameters parameter on the output or response. Here effect
were then classified into control parameters and refers to the change in average response when a
noise parameters. Control parameters are those design (or process) parameter goes from a low level
which can be controlled easily by the operator to a high level.
during the experiment. For example, shrinkage of Table 1 illustrates the list of control parameters
parts in an injection moulding process is quite and their selected ranges for the experiment. In the
critical, as it badly affects the final assembly. The context of ED, a `response' is the quantity an
control factors which might have an impact on the experimenter wants to measure during the experi-
parts shrinkage are screw speed, mould tempera- ment in order to judge the performance of the
ture, cycle time and mould pressure. Noise para- product. In this case the response or performance
meters are those which are hard to control or monitored is the time of flight measured in
expensive to control by the operator during the seconds. Note that selection of an appropriate
experiment [8]. For example in the above process, response for any industrial experiment is critical
relative humidity is a noise factor. for its success [10]. For teaching purposes, it is
The following control parameters were selected good practice to choose continuous responses (e.g.
for the paper helicopter experiment: surface roughness, strength, efficiency, life, etc.)
. than those which are attributes (e.g. taste, colour,
paper type
. appearance etc.).
wing length
. body width
. Interactions of interest
body length
. Two factors, say, X and Y are said to interact
number of clips attached
. with each other if the effect of control parameter X
wing shape.
on the response (or output) is different at different
The two noise factors which could not be directly levels of control parameter Y or vice versa [11]. If
controlled during the experiment were: the effect of control parameter X on the response is
. draft the same at all levels of control parameter Y, then
. operator. the interaction between the control parameters is
said to be zero.
In order to minimise the effect of these noise For industrial experiments with two control
factors on the time of flight, extra caution was parameters X and Y considered at two levels
taken during the experiment. For instance, the (referred to as 2-level parameters), the interaction
experiment was conducted in a closed room in effect can be computed by the equation:
order to dampen the effect of draft. The same
person (i.e., operator) was responsible to minimise Interaction effect ˆ 12 Effect of control parameter
the reaction time of hitting the stopwatch when the
helicopter is released and when it hits the floor. X at high level of Y
Once the design parameters are selected, the Effect of control parameter
next stage is to determine the number of levels in
which the parameters should be studied for the X at low level of Y †

Table 1. List of control parameters for the experiment

Control parameters Parameter labels Low level 1† High level ‡1†

Paper type A Normal Bond

Wing length B 80 mm 130 mm
Body width C 20 mm 35 mm
Body length D 80 mm 130 mm
No. of clips E 1 2
Wing shape F Flat Angled 45 up
338 J. Antony and N. Capon

or all the control parameters (or factors) in the

experiment at their respective levels simultaneously
Interaction effect ˆ 12 Effect of control parameter [13]. A factorial experiment can be either full
factorial or fractional factorial. A full factorial
Y at high level of X experiment (usually represented by 2k ) is preferred
Effect of control parameter when both main and interaction effects are to be
evaluated independently. The standard number of
Y at low level of X † experimental trials or runs for a factorial experi-
ment are 4, 8, 16, 32, 64 and so on. This can be
easily derived using the formula: Number of
Choice of experimental design and design matrix experimental trials ˆ 2 k where k is the number of
for the experiment control parameters at 2-level.
The choice of experimental design depends on However, in many situations this type of experi-
the number of degrees of freedom associated with ment may be not feasible and practical due to time
main and interaction effects and cost and time and cost constraints. Under such circumstances,
constraints. Here the degrees of freedom is the fractional factorial experiments provide a reason-
number of independent and fair comparisons that able alternative that still allows the experimenters
can be made from a set of observations [12]. For to evaluate main effects [14]. Fractional factorial
example, if a control parameter is set as 2-level, experiment will consume only a fraction of the full
then only one fair and independent comparison factorial experiment and is generally represented
between the levels (i.e., low and high) can be made. by 2 k l † , where 12 l yields the fraction. For example,
For a control parameter at 3-level, the number of 2 5 2† implies that the experimenter wishes to study
fair and independent comparisons that can be five control parameters in eight experimental trials.
made among the levels is two. Therefore, the This is a 14 of a full factorial experiment (i.e., 25 ).
number of degrees of freedom associated with a For more information on fractional factorial
control parameter at p levels is p 1†. The experiments, the readers are advised to consult
number of degrees of freedom associated with an [15].
interaction is the product of the number of degrees For the helicopter experiment, as the total
of freedom associated with each main effect number of degrees of freedom is equal to nine,
involved in the interaction. For example, if a the closest number of experimental trials that can
control parameter X is at 2-level and another be employed for the experiment is 16 (i.e., 2 6 2†
control parameter Y at 3-level, then the number fractional factorial). This implies that only a
of degrees of freedom associated with their quarter replicate of a full factorial experiment is
interaction is 2 (i.e., 1  2). needed for the study. The experiment was
For the helicopter experiment, as we are inter- performed based on the design matrix as shown
ested in studying six main effects and three inter- in Table 2. Each trial was randomised to minimise
action effects, the total number of degrees of the effect of noise. Randomisation is a method of
freedom is equal to nine ( i.e., 6 ‡ 3). It is impor- safeguarding the experiment from systematic bias
tant to meet the criterion that the number of which causes variation in response or output.
experimental trials required for a certain experi- The design matrix displays all the control
ment must be greater than the number degrees of parameter settings for the experiment. Having
freedom associated with the main and interaction constructed the design matrix, the flight times
effects to be studied for the experiment. A factorial were recorded (see Table 2) corresponding to
experiment is an experiment where one may vary each trial condition.

Table 2. Design matrix for the helicopter experiment: ( ) represents the experimental trials in random order

Trial no./run A D B C E F Time of flight (s)

1 (6) Normal 80 80 20 1 Flat 2.49

2 (9) Bond 80 80 20 2 Flat 1.80
3 (11) Normal 130 80 20 2 Angled 1.82
4 (15) Bond 130 80 20 1 Angled 1.99
5 (12) Normal 80 130 20 2 Angled 2.11
6 (2) Bond 80 130 20 1 Angled 1.96
7 (16) Normal 30 130 20 1 Flat 3.19
8 (14) Bond 130 130 20 2 Flat 2.27
9 (10) Normal 80 80 35 1 Angled 2.12
10 (1) Bond 80 80 35 2 Angled 1.58
11 (7) Normal 130 80 35 2 Flat 2.15
12 (3) Bond 130 80 35 1 Flat 2.05
13 (8) Normal 80 130 35 2 Flat 2.60
14 (4) Bond 80 130 35 1 Flat 2.09
15 (5) Normal 130 130 35 1 Angled 2.63
16 (13) Bond 130 130 35 2 Angled 2.18
 Teaching Experimental Design Techniques to Industrial Engineers 339

Table 3. Coded design matrix for the helicopter experiment

Trial no./run A D B C E F Time of flight (s)

1 (6) 1 1 1 1 1 1 2.49
2 (9) ‡1 1 1 1 ‡1 1 1.80
3 (11) 1 ‡1 1 1 ‡1 ‡1 1.82
4 (15) ‡1 ‡1 1 1 1 ‡1 1.99
5 (12) 1 1 ‡1 1 ‡1 ‡1 2.11
6 (2) ‡1 1 ‡1 1 1 ‡1 1.96
7 (16) 1 ‡1 ‡1 1 1 1 3.19
8 (14) ‡1 ‡1 ‡1 1 ‡1 1 2.27
9 (10) 1 1 1 ‡1 1 ‡1 2.12
10 (1) ‡1 1 1 ‡1 ‡1 ‡1 1.58
11 (7) 1 ‡1 1 ‡1 ‡1 1 2.15
12 (3) ‡1 ‡1 1 ‡1 1 1 2.05
13 (8) 1 1 ‡1 ‡1 ‡1 1 2.60
14 (4) ‡1 1 ‡1 ‡1 1 1 2.09
15 (5) 1 ‡1 ‡1 ‡1 1 ‡1 2.63
16 (13) ‡1 ‡1 ‡1 ‡1 ‡1 ‡1 2.18

STATISTICAL ANALYSIS AND estimated. Table 4 illustrates the estimated main

INTERPRETATION OF RESULTS effects for the experiment.
In order to assist people with limited mathema-
Having obtained the response values, the next tical skills, the authors recommend a graphical plot
step is to analyse and interpret the results so that of main effects. The notion behind the use of this
necessary actions can be taken accordingly. The graphical representation is to provide novices a
first step in the analysis involves the computation better picture on the importance of the effects of
of both main and interaction effects. For computa- the chosen control parameters. Figure 2 illustrates
tion purposes, the low and high levels in the design the main effects plot of the control parameters.
matrix (refer to Table 2) must be replaced by 1 The main effects plot shows that the most
and ‡1 respectively. For example, control para- dominant control parameter on the time of flight
meter `A' column must be replaced by 1, ‡1, 1, is paper type, followed by wing length, wing shape
‡1 and so on. The same procedure must be applied and number of clips. The control parameter body
to other control parameters. The resulting coded width has no impact on the time of flight.
design matrix for the experiment is shown in Table
3. This table is used for all computations of main Calculation of interaction effects
and interaction effects. For the helicopter experiment, we were inter-
ested to study the following three interactions:
Calculation of main effects . Wing length Body width (B  C)
The effect of a main/interaction effect is the . Wing length Body length (B  D)
difference in the average response at low and . Paper type Number of clips (A  E)
high levels. For example, the effect of control
parameter A can be calculated as follows: Consider the interaction between the control para-
. Average time of flight at high level of A ˆ 1:99 meters B and D. In order to compute the inter-
. Average time of flight at low level of A ˆ 2:39 action, we must obtain the average time of flight at
. Effect of control parameter A ˆ 2:39 1:99 ˆ each level combination of these control para-
0:40 meters. There are all together four combinations
of levels between these two parameters:
A negative sign indicates that the slope of the line B 1 D 1 , B 1 D‡1 , B‡1 D 1 and B‡1 D‡1 . The
connecting the low and high values is negative. In average time of flight at these combinations are
other words, the average time of flight at the low illustrated in Table 5.
level is higher than that at the high level. Similarly,
the effects of other control parameters can be Interaction effect B  D†

Table 4. Table for main effects for the experiment ˆ 12 Effect of B at high level of D

Control Average at Average at Effect Effect of B at low level of D†

parameters high level low level
ˆ 12 ‰ 2:568 2:003† 2:190 1:998†Š
A 1.99 2.39 0.40
D 2.29 2.09 0.20 ˆ 0:187
B 2.38 2.00 0.38
C 2.20 2.18 0.02 An alternative approach for computing the inter-
E 2.06 2.32 0.26
F 2.04 2.34 0.30 action effect between B and D can be achieved by
multiplying the coded levels of B and D in each
340 J. Antony and N. Capon

Fig. 2. Main effects plot of the control parameters.

row of Table 3. Having obtained the levels for the The main effects plot and interaction plot
product (B  D), we will then calculate the average however does not tell us which of the main and/
flight times corresponding to low and high levels of or interaction effects are statistically significant.
(B  D). The interaction effect between B and D Under such circumstances, it is good practice to
can now be estimated in a similar manner to the employ normal probability plots [16]. For normal
main effects: probability plots, the main and interaction effects
of control parameters should be plotted against
Interaction effect B  D† cumulative probability (%). Inactive main and
ˆ Average flight time at high level of B  D† interaction effects tend to fall roughly along a
straight line whereas active effects tend to appear
Average flight time at low level of B  D† as extreme points falling off each end of a straight
line. These active effects are judged to be statis-
Similarly, the interactions between B and C , i.e., tically significant. Figure 4 shows a normal
(B  C), and the interaction between A and E, probability plot of effects (both main and inter-
i.e., (A  E) can be computed. The results are action) of control parameters at 99% confidence
summarised in Table 6. level (or 1% significance level). Here significance
level is the risk of saying that a factor is
Interaction plot significant when in fact it is not. In other words,
This is a very powerful graphical tool for inter- it is the probability of the observed significant
preting the interaction effects. It provides a better effect (either main or interaction) being due to
and rapid understanding of the nature of inter- pure chance. For experimental design problems,
actions among the control parameters under we generally consider both 5% and 1% significance
consideration. Non-parallel lines in the interaction levels. If measures the significance level, then
plot connotes the existence of interaction among 1 † measures our confidence for an effect to be
the parameters, whereas parallel lines indicates the statistically significant [17]. The graph (Fig. 4)
non-existence of interaction among the parameters shows that main effects A, B, E and F are
for investigation. Consider the interaction between statistically significant. The interaction between B
wing length (B) and body length (D) for the above and D was not statistically significant at 1%
experiment. The interaction plot is shown in Fig. 3. significance level (or 99% confidence level)
As the lines are non-parallel, there is an interaction though it appeared to be important in the
between the control parameters B and D. interaction graph (refer to Fig. 3).

Table 5. Average response table

Table 6. Table of interaction effects
B D Average time of flight
Interaction effects Estimate of the effect
1 1 1.998
1 ‡1 2.003 BD 0.187
‡1 1 2.190 BC 0.021
‡1 ‡1 2.568 AE 0.186
 Teaching Experimental Design Techniques to Industrial Engineers 341

Fig. 3. Interaction plot between parameters B and D.

Fig. 4. Normal probability plot of effects.

Determination of optimal control parameter were kept at high levels. A confirmatory experi-
settings ment was carried out to verify the results from the
Having identified the significant control para- analysis. Five helicopters were made based on the
meters, the next step is to determine the optimal optimal combination of control parameter levels.
settings of these parameters that will maximise the The average flight time was estimated to be 3.26 s.
flight time. In order to arrive at the optimal
condition, the mean time of flight at each level Table 7. Final optimal control parameter
of these parameters was analysed. As none of the settings
interaction effects were statistically significant,
the main concern was the average flight times at Control parameters Optimum level
the low and high level of the main effects (refer to Paper type Normal (low level)
Table 4). From Table 4, the final optimal settings Wing length 130 mm (high level)
of control parameters was derived (see Table 7). It Body width 20 mm (low level)
is quite interesting to notice that the optimal Body length 130 mm (high level)
control parameter settings is one which correspond Body length 130 mm (high level)
Number of clips 1 (low level)
to trial condition 7 (see Table 2). The time of flight Wing shape Flat (low level)
was maximum when wing length and body length
342 J. Antony and N. Capon

Significance of the work experimental design could be applied in reducing

The purpose of this section is to bring the the cycle time for new products, maximising the
importance of teaching experimental design response for a certain advertisement and com-
techniques to engineers with limited statistical paring competitive strategies and decision-making
skills for tackling quality engineering problems in processes.
many organisations. It is quite important to note
that this experiment is quite old in its nature and
has already been widely used for some time by CONCLUSIONS
many statisticians for teaching purposes. Never-
theless the focus here was to minimise the statis- Experimental design is a very powerful problem-
tical jargon associated with the technique and solving technique that assists industrial engineers
bring modern graphical tools for better and rapid for tackling quality control problems effectively
understanding of the results to non-statisticians. and economically. Research has shown that the
This experiment was conducted in a class-room application of ED by industrial engineers is limited
to help students come to grips with experimental due to lack of skills in manufacturing and lack of
design. As it is an extremely simple experiment, the statistical knowledge. The paper illustrates the
concept and the objective of the experiment was cognitive gap in the knowledge required by indus-
quite straight forward. The students of the class trial engineers for understanding the potential
found this particular experiment interesting benefits of this powerful-problem solving tech-
specifically in terms of selecting the appropriate nique. The purpose of this paper is to bridge this
experimental design, conducting the experiment gap by illustrating a simple experiment which can
and interpreting the results of the experiment. be easily carried out in a room to teach experi-
Many students were quite astounded with the use mental design techniques to engineers. In order to
of graphical tools and its reduced involvement of keep the experiment simple, all the control para-
number crunching. The experiment helped them to meters were studied at 2-level. The authors believe
understand the importance of brainstorming for that control parameters at 2-level will form a firm
identifying the key variables and understanding foundation for studying the parameters at 3-level
the concept and nature of interactions among the and higher. The results of the experiment has
variables. The authors strongly believe that the provided a greater stimulus for the wider appli-
experiment provided a simple and beneficial way cation of ED by industrial engineers in real-life
to help students view on experimental design in a situations.
more approachable manner which consequently
led to their eagerness to apply it in their own AcknowledgementsÐThe authors would like to thank two
work environment. As many of the students had anonymous referees for their valuable and useful comments
marketing background, they commented that on an earlier draft of this paper.

REFERENCES
1. D. C. Montgomery, The use of statistical process control and design of experiments in product and
process improvement, IEE Transactions, 24, 5, (1992) pp. 4±17.
2. M. B. Sirvanci and M. Durmaz, Variation reduction by the use of designed experiments, Quality
Engineering, 5, 4, (1993) pp. 611±618.
3. J. Antony and M. Kaye, An application of Taguchi's parameter design methodology for process
improvement, J. Quality World Technical Supplement, (1996) pp. 35±41.
4. J. S. Ramberg, et al. Improvements on Taguchi methods for semiconductor design/
manufacture, at Semiconductor Manufacturing Conference, November 1989, pp. MS89-798-1 to
MS89-798-16.
5. J. Antony and M. Kaye, Experimental quality, J. Manufacturing Engineer, IEE, 74, 4, (1995)
pp. 178±181.
6. J. Antony et al., Use your reason to understand statistics, Quality World (Technical Paper), 23, 10,
(1997) pp. 850±853.
7. D. C. Montgomery, Experiment design and product and process development, Manufacturing
Engineering, (1988) pp. 57±63.
8. G. Taguchi, Introduction to Quality Engineering, Asian Productivity Organisation, Tokyo, Japan
(1986).
9. J. Antony and M. Kaye, Experimental quality: a strategic approach to achieve and improve
quality. Unpublished work.
10. R. M. Miesel, A planning guide for more successful experiments, ASQC Annual QC Transactions,
(1991) pp. 174±179.
11. M. J. Mazu, Interpreting a significant interaction effectively, ASQC Congress Transactions, (1990)
pp. 115±120.
12. N. Belavendram, Quality By Design, Prentice-Hall, UK (1995).
13. G. W. William, Experimental design: robustness and power issues, ASQC Quality Congress
Transactions, (1990) pp. 1051±1056.
14. N. L. Frigon, Staying on target with design of experiments, Quality Digest, (Dec. 1994)
pp. 65±69.
 Teaching Experimental Design Techniques to Industrial Engineers 343

15. G. E. P. Box, W. G. Hunter, W. S. Hunter, Statistics for Experimenters, John Wiley and Sons, New
York (1978).
16. H. C. Benski, Use of a normality test to identify significant effects in factorial designs, J. Quality
Technology, 21, 3, (1989) pp. 174±178.
17. P. J. Ross, Taguchi Techniques for Quality Engineering, McGraw-Hill, New York (1988).

Jiju Antony is a Research Fellow with the Portsmouth Business School, University of
Portsmouth. He received a BE in Mechanical Engineering from the University of Kerela,
South India, M.Eng.Sc. in Quality and Reliability Engineering from the National
University of Ireland and a Ph.D. in Quality Control for Manufacturing from the
University of Portsmouth. He has published over 25 refereed papers in the areas of
reliability engineering, design of experiments, Taguchi methods for improving process
quality, robust technology development and problem solving using quality tools and
techniques. Dr Antony has applied design of experiments and Taguchi methods in more
than five manufacturing companies.

Nick Capon is a Senior Lecturer with the Portsmouth Business School. His research
interests are in the areas of operations management and quality management, with current
research into business process re-engineering and risk management. He has an industrial
background as a manufacturing manager, including experience in materials management
and systems analysis in aerospace engineering and electronics. He has published over 10
articles.

View publication stats