Quality Engineering and Taguchi Methods: A Perspective
Quality Engineering and Taguchi Methods: A Perspective
A Perspective
Robust product design and parameter design-methods to develop products that will perform well regardless of changes in uncontrollable envtronmental conditions or that are insensitive to component vanatlon-are key
concepts in the work of Or. Taguchi. We should encourage. design ~nd
manufacturing engineers to apply these useful ideas. But In deslgnlnQ. experiments and analyzing data - key aspects of the practical Implementation
- better and simpler methods are available and should be preferred over
Taguchi's less intuitive and more cumbersome approaches.
Smen Bisgaard
uring the last few years we have
witnessed a surge in interest in
quality improvement. Quality improvement, or as it used to be
called, quality control, is a field with
a loog history that continues to
grow. New contributions are made
every year. In the 1960s and 70s
there was a slump in the Western
development partly because of misunderstanding of the enormous economic benefits of quality control.
During that period, quality control
was mostly perceived as a set of
techniques for sorting and final inspection of products. Thus, quality
control was seen as an added expense.
This view is rapidly changing.
Many companies are now developing a much more visionary
company-wide quality improvement
strategy. Perhaps the most important development is that in some
companies, management is coming
to the fundamental conceptual understanding that quality does not
necessarily come at a higher cost.
By using quality improvement concepts, quality engineering methods,
and quality management principles,
it is possible to develop products
and processes of higher quality at
lower overall cost. In fact, quality
management is an integrating concept for cost-effective, rational, cooperative manufacturing of highquality products and for delivering
high-quality services.
Fall 1989
I>
13
14
Target
Fall 1989
70 7
year
graph
cause & effect diagI1ll11
Pareto chart
histogxam
control chan
scatter diagxam
process capability
analysis of variance
statistical test & estimation
optimization & EVOP
time series
'81 82 83 84 ., 80 81
)
29 16 19 18 19 15 26 36 26 20 40 40 44 59 70 88 55 78
UI6~1819~DY~~~~G~~4OD.
6 7 10 9 12 2 10 22
5
2 5 4 5 16 12
7 4 I 8 5 5 8 12
3
7
3 4 12
I
2 4
nomogI1ll11S
design of cxpcrimcnt
correlation & re,eression
4 75 10 77 7.
(Fu
18
10
2
I
I
17 21
14 15
8 IS
IS 13
2 -
18
11
11
6
I
20 23 33
17 22 28
20 23 26
6 2 13
141
- -' (General)
36 34
25 37
21 23
13 IS
2 7
I
8 14 13 IS 14 14 27 12 14 27 29 44 36 59 49
5
9 4 3 13 6 9 9 10 11 16 17 10 20
3
3 1 1
2 9 4 4 6 2 5 4 7
1
3 3 6 3 621 5 I I I 3
1
I
III
12112
2
1
cumulative method
sampling inspection
- -
221
I 1
34 40
33
20
24
-
36
13
22
3
45 43
18 18
5 10
3 9
1
(Multivariate Analysis)
3 2 3 3 5 2 4 3 5 12 4 5 17 12 14 23 13 8
mUltiple regression
.
1 I
1 6 I 6 6 3 4 7 8 7 4 2
prinCIpal componep.t analysIs .
12-1-2211411342
discriminant and cluster analysts ------45278
quantification theory
---1---1
oth""
(Miscellaneous)
FfA
FMEA
Weibull disttibution
distribution theory
sensoI)' test
relation chaIt
PERT
simulation
computer technic & BOPS
oth""
No. or Dresentanons
3
I
3 5 1 4
4 2
3
5 1 I 4
6 6 30 29 42 18 14 43 32 40
- 11 8 14 10 18
2
6 5 3 10 - II 12 9 4 4
1324321143723
5 3 5 7 1 4
2 5 3 531
I 424
2 2 3 153
4 4 5 7 6 5 4 IS 7 6
2 2 4 2 1 1 132
213
1 3 123
2 6 6 1 10
2 1 1
2 4 2 I 2 4 - 12 6 9
126 60 65 94 110 97 95 114 107 130 138 148 178 172 202 210 183 175
many outstanding papers on the application of statistical methods to industrial problems and in particular
on the use of experimental design.
During those early years, quality experts and statisticians in the
United States and Great Britain cooperated closely. For example, the
Massachusetts Institute of Technology (MIT) sponsored an Industrial
Statistics Conference in 1938.
Among the speakers were Dr. Waiter A. Shewhart" from Bell Labs,
and Leslie E. Simon,12 then captain
of the U.S. Army Ordinance Department, and later general and author
of the book An Engineer's Manual
of Statistical Methods'" From
academia came the statistician S. S.
Wilks," Princeton University, who
was very active in the area of engineering applications of statistics and
later co-author with J. S. Hunter and
I. Guttman of the book Introductory
Engineering Statistics,15 and H. A.
I>
15
16
improvements or enhancements
over other methods well-known to
statisticians.
Robust Product Design
Taguchi's important and useful
contribution is rather his engineering
idea of robust product design and
parameter design. A robust car, for
exampie, is a car that not only starts
and runs well under ideal environmental conditions, but performs well
even when the temperature is 110F
and the humidity is 90 percent, or
when the temperature is -30F and
the humidity is 0 percent. Thus robust products are insensitive to
changes in uncontrollable conditions
or other kinds of variability.
We also use the term "robust"
about products that are insensitive
to production variation from nominal
values of the parts from which they
are assembled. This is important
because when products are produced under mass production conditions, the components invariably differ from their specified nominal
values. If assembied products can
be made robust (insensitive) to component variation, production tolerances can be opened up, and that
often means great savings in production costs.
Taguchi's method for making
products more robust is called Parameter DeSign. This method is
based on the idea that the design
parameters of a product can be manipulated by experimental methods
with the objective of finding product
design configurations that are less
sensitive to variation. Taguchi's bestknown example concerns the manufacture of tiles. Rather than buy a
new expensive kiln that would have
less temperature variation, the Ina
Tile Company of Japan experimentally found that by adding an inexpensive lime additive to the clay,
fewer tiles would break due to temperature variation in the old kiln. Thus
a simple cheap solution was found
Whereby the tiles were made robust
to temperature variation.
Robust products and parameter
design are important and very use
ful ideas. They should be included
as standard concepts used by all
Target
Fall 1989
(DOE) has had no impact at all unless modified for the Taguch/ way of
thinking.
One of the tenets for quality
professionals is, "Let us talk with
data."27 In fact, the "27th Quality
Control Conference for Foremen"
held recently in Japan had as its
motto "Make judgment according to
facts and act on the facts." Taguchi
definitely has been able to create an
unprecedented interest in the United
States in design of experiments for
quality improvement. For this interest, we should be very grateful. But
to claim that traditional design of experiments has had no impact is at
best uninformed.
An overwhelming collection of
published examples of successful
applications of designed expenments for quality improvement can
be found in the journals Industrial
Quality Control, Applied Statistics,
Suppiement to the Journal of the
Royal Statisticai Society, Proceedings of the American Society for
Quality Control, Electrical Manufacturing, Technometrics, Journal of the
American Statistical Association,
Food Technology, Tooling and Production, Biometrics, Journal of the
American Society of Naval Eng/neers Journal of the Institute of Petroleu'm, The Industrial Chemist,
American Statistician, Chemical Engineering, Journal of the Society of
Plastic Engineering, Journal of
American Leather Chemists Association American Society for Testing
Mat~ria/s Bulletin, Electrical Engineering, Analytic Chemistry, Rubber
Age, Tappi, Bell Systems Technical
Journal, Blast Furnace and Steel
Plant, Industrial and Engineering
Chemistry, American Society of Mechanicai Engineers Transactions,
Journal of Quality Technology, and
many others. The published examples are, of course, just the tip of
the iceberg. It seems hard to believe
that people over a period of more
than 40 years would continue to
publish examples that had "no im..
pact at all."
So why did Western statisticians in the past not succeed in get-
I>
17
18
Target
Fall 1989
17
tics,
L:
Author:
Dr. S0ren Bisgaard is a professor in the
Department of Industrial Engineering,
University of Wisconsin-Madison, and
he is affiliated with the university's Center for Quality and ProductiVity Improvement.
19