Caso Métodos Taguchi

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Teaching Taguchi’s Approach to

Parameter Design

G
ENICHI TAGUCHI ’ S APPROACH TO cation limits are equal in status to those with the
ensuring quality is based on build- desired target value. In contrast, Taguchi’s approach
ing robust product and process encourages manufacturers to produce goods with
designs. While the Taguchi the least possible variation, with the goal of being
approach has been successfully on target 100% of the time—which ultimately leads
applied in industry, these applications are often to more satisfied customers.
too complicated to be replicated in a classroom What poor quality really means to society. The
setting.1 Some attempts have been made to fundamental idea in Taguchi’s approach is that
develop realistic cases for classroom discussion, each time a poor-quality product is shipped, there
but they are usually too discipline-specific and is a loss imparted to society.3 The effect of the
hence require certain technical expertise to poor quality is assigned a value through the use of
This easily understand.2 There is an exercise, however, that a loss function model. According to this model,
instructors can use to illustrate Taguchi’s the loss due to a product’s variation in perfor-
understood approach. The application—designing and mance is proportional to the square of the perfor-
building paper airplanes—is generic enough to mance characteristic’s deviation from its target
application be useful in any type of class in any discipline. value. The resulting graph of the loss function vs.
can be used The only prerequisite is knowledge of basic sta- quality characteristic is a parabola, with zero loss
tistics. when the quality characteristic is equal to the tar-
in training The paper airplane exercise was initially get value. The focus of Taguchi’s approach is to
designed for a graduate course in quality at strive for zero loss.
classes North Carolina Agricultural and Technical State How to produce high-quality goods. Instead of
University in Greensboro. The exercise requires controlling process quality by using control charts
everywhere. only minimal resources: paper, scissors, and a or screening out defects by inspecting products, the
measuring tape. It takes about five hours to per-
form the experiment and collect the data.
Before students conduct the experiment,
however, the instructor has to provide a
background in the Taguchi approach.
by Teaching the basic concepts
Sanjiv Sarin
When covering the fundamental
concepts of the Taguchi approach, the
instructor needs to cover:
Why the current mind-set needs to be
changed. Too often, manufacturers are pre-
occupied with the so-called goalpost mentali-
ty in which the most important goal is to stay
within a customer’s specification limits. (The
lower and upper specification limits are analo-
gous to the two uprights of a goalpost.) The goal-
post approach does not encourage manufacturers
to produce goods that are on target because prod-
ucts that are close to the upper and lower specifi-

102 /
Quality Progress May 1997
Table 1. Controllable and Uncontrollable Factors
T h e Ta g u c h i a p p r o a c h i s b a s e d
Controllable factors
on the premise that engineers,
Factor Level 1 Level 2 Level 3
not statisticians, will be per-
Paper weight (A) One sheet Two sheets Three sheets
forming the experiments.
Design (B) Design 1 Design 2 Design 3

Paper width (C) 4 in. 6 in. 8 in.


Taguchi approach emphasizes building quality into the design of
products and their processes. To do that, a product’s designer Paper length (D) 6 in. 8 in. 10 in.
needs to judiciously select those parameters that define the prod-
uct and reduce variability in the performance of those parame- Uncontrollable factors
ters. This is referred to as on-target performance, a
concept that is reinforced by associating a value to quality Factor Level 1 Level 2
through the use of the loss function. Launch height (E) Standing on ground Standing on chair
The process of selecting a product’s important parameters is
called parameter design.4 Parameter design involves finding the Launch angle (F) Horizontal to ground 45° above horizontal
optimal settings of controllable factors in a product’s design so
Ground surface (G) Concrete Finished wood
that the final design is robust in the face of a variety of uncon-
trollable factors. Controllable factors are those design parame-
ters that need to be optimized. These factors should be speci-
fied on production drawings or specification sheets. For exam-
ple, the controllable factors in the design of a computer moni- Table 2. L4 Orthogonal Array for Uncontrollable Factors
tor might include the screen’s dimensions, curvature, and
weight. Uncontrollable, or noise, factors can also influence a Run Factor E Factor F Factor G
product’s performance, but they are not under the designer’s
1 1 1 1
control. In the case of the computer monitor, the noise factors
might include input voltage variation, ambient lighting condi- 2 1 2 2
tions, and vibrations when the monitor is in use at the cus- 3 2 1 2
tomer’s home.
4 2 2 1
Parameter design involves assigning controllable
factors to an inner array and the uncontrollable factors to an
outer array. The inner and outer arrays specify combinations of
settings of the factor levels so that each factor appears at each
level an equal number of times. Furthermore, for a given level Table 3. L9 Orthogonal Array for Controllable Factors
of a factor, all other factors vary across all their levels in a bal-
anced manner. These arrays are called orthogonal arrays. Run A B C D
Depending on the number of factors and the number of levels
1 1 1 1 1
of each factor, an appropriate orthogonal array can be selected.
Orthogonal arrays are generated from full-factorial design 2 1 2 2 2
matrices by deliberately confounding some interaction effects 3 1 3 3 3
with each other or with main effects. Collections of standard
orthogonal arrays are available in several books.5 4 2 1 2 3
Taguchi’s approach vs. design-of-experiments (DOE) 5 2 2 3 1
methodology. There has been some debate about the novelty of
6 2 3 1 2
Taguchi’s approach with respect to the well-established DOE
methodology. While DOE is concerned with research and 7 3 1 3 2
knowledge building, Taguchi’s approach is driven by practical- 8 3 2 1 3
ity. It is based on the premise that engineers, not statisticians,
9 3 3 2 1
will be performing the experiments, so an approach that is
expedient and easy to apply is of value.

/
Quality Progress May 1997 103
varying designs, weights, and paper sizes.
Table 4. Inner L9 and Outer L4 Array and the Data Collected Students design and perform an experi-
ment to determine the paper airplane’s
Factor Distance of flight (inches) parameter design. In other words, students
determine at which levels to set the con-
E1 E1 E2 E2
Treatment F1 F2 F1 F2 Signal-to- trollable factors so that the uncontrollable
condition A B C D G1 G2 G2 G1 noise ratio X value factors have minimal effect on the plane’s
performance.
1 1 1 1 1 49 44 12 38 26.7 35.8
The response being studied is the dis-
2 1 2 2 2 91 42 44 38 33.2 53.8 tance traveled (in inches) by each airplane
3 1 3 3 3 59 48 39 67 34.0 53.3 from the point it is launched to the point it
comes to rest (including sliding distance
4 2 1 2 3 116 89 48 88 37.2 85.3
on the ground). Averages and signal-to-
5 2 2 3 1 32 38 56 108 32.8 58.5 noise ratios are used to measure the out-
6 2 3 1 2 24 55 39 46 31.0 41.0
come. The desired outcome is the maxi-
mum distance traveled by the airplane.
7 3 1 3 2 42 76 122 41 34.6 70.3 Here is how the graduate students at
8 3 2 1 3 50 73 47 65 34.9 58.8 North Carolina Agricultural and Technical
State University conducted this experiment
9 3 3 2 1 76 34 80 37 33.2 56.8
and the results they obtained:
To begin, the students used investigation
techniques, such as the cause-and-effect
diagram, to develop a list of important fac-
tors to consider. Then they decided whether those factors were
Table 5. Average Signal-to-Noise (S/N) Ratios controllable (i.e., those that affect the airplane’s performance
and are under the designer’s control) or uncontrollable (i.e.,
Level Average Level Average Level Average those that can influence the airplane’s performance but are not
1 S/N ratio 2 S/N ratio 3 S/N ratio Change
under the designer’s control).
A1 31.3 A2 33.7 A3 34.2 2.9 The factors that the students decided were under the airplane
B1 32.8 B2 33.6 B3 33.2 0.8
designer’s control were:
• The weight of the airplane (factor A). The weight was altered
C1 30.9 C2 34.5 C3 33.8 3.6 by the number of sheets of paper used to make the airplane.
D1 30.9 D2 32.9 D3 35.4 4.5 Planes made with one, two, and three sheets of paper were
used to provide three different weights.
• Design of the airplane (factor B). The airplane’s design was
altered by the way in which it was folded. Three different
Table 6. X Values designs were used. (In this experiment, the designer can
make three different versions of the paper airplane following
any designs he or she wants, as long as these three designs
Level X Level X Level X
1 value 2 value 3 value Change are used for all of the airplanes. The purpose of this experi-
ment is not to test certain airplane designs, but rather to
A1 47.6 A2 61.6 A3 62.0 14.4
determine at which levels to set the controllable factors so
B1 63.8 B2 57.0 B3 50.4 13.4 that the uncontrollable factors have minimal effect on per-
C1 45.2 C2 65.3 C3 60.7 20.1
formance.)
• The width of the paper (factor C). The width of the paper
D1 50.4 D2 55.0 D3 65.8 15.4 was altered by cutting it into three different widths: 4 in., 6
in., and 8 in.
• The length of the paper (factor D). The length of the paper
was altered by cutting it into three different lengths: 6 in., 8
Clearly, Taguchi’s approach is based on traditional DOE, but in., and 10 in.
it goes a step further. It seeks to develop products and process The factors that the students decided were not under the air-
designs that are insensitive to noise factors and that are on tar- plane designer’s control were:
get with minimum variability. • Launch height (factor E). The airplanes were launched at
two different heights: while standing on the ground and
Teaching the mechanics of parameter design while standing on a chair.
Scissors, a measuring tape, and 62 sheets of standard 8.5- • Launch angle (factor F). The airplanes were launched at two
in.-by-11-in. photocopying paper are needed to conduct the different angles: horizontal to the ground and 45 above hor-
following exercise. It involves making paper airplanes with izontal.

104 /
Quality Progress May 1997
Figure 1. Effect of Paper Weight (Factor A) Figure 2. Effect of Design (Factor B)
Average S/N ratio Average S/N ratio
X value X value
70 70
60 60
50 50
40 40
30 30
20 20
10 10
0 0
One Sheet Two Sheets Three Sheets Design 1 Design 2 Design 3
A1 A2 A3 B1 B2 B3

Figure 3. Effect of Paper Width (Factor C) Figure 4. Effect of Paper Length (Factor D)
Average S/N ratio Average S/N ratio
X value X value
70 70
60 60
50 50
40 40
30 30
20 20
10 10
0 0
4-in. width 6-in. width 8-in. width 6-in. length 8-in. length 10-in. length
C1 C2 C3 D1 D2 D3

• Ground surface (factor G). The airplanes were launched on launched four times following the prescribed noise factor con-
two different surfaces: concrete and finished wood. ditions. For example, one flight of the TC1 plane involved
These controllable and uncontrollable factors are summarized launching it horizontally (F1) while standing on a concrete
in Table 1. floor (E1 and G1, respectively). Since there were four combi-
Since there were three uncontrollable factors, each with two nations of noise factors for each treatment condition, a total of
levels, a level 4 (L4) orthogonal array was used to denote the 36 (4  9) experimental runs were conducted.
noise factors (see Table 2). In run 1, each factor was at level 1. All launches were conducted by the same person in a closed
Run 2 involved setting factor E at level 1 (i.e., E1), while set- room or enclosed hallway with no air currents. When launch-
ting factors F and G at level 2 (i.e., F2 and G2, respectively). ing the airplane, the student had his elbow touching his torso,
Runs 3 and 4 had similar variations. It is worth noting that only using only the forearm, wrist, and hand to send the plane into
four experimental runs were need- flight. The student used all
ed to compute the effects of the of his force in all of the
four factors. Clearly, this is a frac- throws, but his elbow
tional factorial experiment; a full A ll launches were conducted remained touching his torso
factorial experiment would have b y t h e s a m e p e r s o n i n a c l o s e d at all times. The nose of the
required eight runs (23). plane marked its final resting
An L9 orthogonal array was r o o m o r e n c l o s e d h a l l w a y w i t h point.
used for the controllable factors, In addition to showing the
since there were four, each with n o a i r c u r r e n t s . inner-outer array configura-
three levels (see Table 3). The tion, Table 4 shows the data
array featured nine runs, which that were collected by the
again shows that this is a fractional factorial experiment—oth- students. For example, for TC1, the airplane flew 49 in. when
erwise 81 runs would have been needed (34). noise factors E, F, and G were at level 1. Table 4 also contains:
Next, students combined the L9 and L4 orthogonal arrays in • Signal-to-noise (S/N) ratios. In the Taguchi approach, the
an inner-outer array configuration. Based on the configuration S/N ratio is defined as S/N = log10 [(1/y2)/n], where y is
in Figure 4, students built nine planes following each treatment the response and n is the number of observations for a treat-
condition in the inner array. For example, treatment condition 1 ment condition. For the airplane experiment, n = 4 since
(TC1) involved building a plane following design 1 (B1) using there were four observations for a given control factor com-
one sheet of paper (A1) that was 4 in. wide and 6 in. long (C1 bination.
and D1, respectively). Each of the nine planes was then • Average (X) values. The average response is the mean of the

/
Quality Progress May 1997 105
four observations for each treatment condition. to parameter design. It can easily be conducted in a classroom
The S/N ratios and X values were analyzed further by the environment using minimal apparatus. Although the exercise
students. Table 5 contains a response table for the S/N ratios was initially prepared for use in a college classroom, it can be
in Table 4. Average S/N ratios were calculated for each level used in any training setting.
of each controllable factor by calculating the mean of the S/N Learning about the Taguchi approach discourages reliance
ratios of the corresponding treatment conditions. For exam- on the goalpost mentality. Instead, students learn to use classic
ple, Table 4 contains three treatment conditions for A1: TC1, DOE techniques to find ways to reduce the variability in the
TC2, and TC3. So, working from the data in Table 4, the performance of a product when it is in the customers’ hands. It
average S/N ratio for A1 is (26.7 + is this measure of quality
33.2 + 34)/3 = 31.3. The “Change” that ultimately determines
column in Table 5 notes the differ- customer satisfaction.
ence between the largest and the T h e Ta g u c h i a p p r o a c h f o c u s e s
smallest average S/N values in the
corresponding row. first on selecting levels of con- Acknowledgment
A similar response table was con- t r o l l a b l e f a c t o r s t h a t r e d u c e The author would like to
structed for the X values (see Table 6). thank the students of the
The averages of the X ratios—com- v a r i a b i l i t y ; b r i n g i n g t h e p r o c e s s INEN 718 class at North
monly known as X—were calculated Carolina Agricultural and
for each level of each controllable fac- o n t a r g e t i s t h e s e c o n d a r y Technical State University for
tor by calculating the mean of the X o b j e c t i v e . the data used in this paper.
values of the corresponding treatment
conditions. For example, Table 4 con-
tains three treatment conditions for B1: TC1, TC4, and TC7. References
So, working from the data in Table 4, the X value for B1 is 1. For example, see M.S. Phadke, R.N. Kacker, D.V. Speeney, and
(35.8 + 85.3 + 70.3)/3 = 63.8. The “Change” column in Table 6 M.J. Greico, “Off-Line Quality Control in Integrated Circuit
notes the difference between the largest and the smallest X val- Fabrication Using Experimental Design,” Bell System Technical
ues in the corresponding row. Journal, Vol. 62, No. 5, 1983, pp. 1,273-1,309, or B. Schmidt, H.
The results in Tables 5 and 6 are presented graphically in Dierl, and H. Kaltenbach, “Optimization of IC Chip Bonding Process
Figures 1 through 4. The students’ evaluation of these figures via Taguchi Methods,” in Proceedings of the ASI Fifth Symposium on
shed light on the appropriate settings for the four controllable Taguchi Methods (Dearborn, MI: American Supplier Institute, 1987),
factors: pp. 475-485.
2. For example, see M.C. Liu, “A Class Project of Quality
• Factor A. For the factor of paper weight, a plane built with
Engineering,” or M. Lulu, “Integrating Quality and Design: A Case in
three sheets of paper had the best average S/N ratio and X Circuit Design,” in Proceedings of the Third Industrial Engineering
value. Thus, level 3 was deemed the appropriate setting. Research Conference (IERC) (St. Louis, MO: IERC, 1993), pp. 327-
• Factor B. For the factor of design, a plane built using design 331 and pp. 343-347, respectively.
1 was superior based on the X value. Based on the average 3. G. Taguchi, E.A. Essayed, and T. Hsiang, Quality Engineering in
S/N ratio, however, the difference between design 1 and Production Systems (New York, NY: McGraw-Hill Book Co., 1989).
design 2 was marginal. When there is disagreement between 4. D.M. Byrne and S. Taguchi, “The Taguchi Approach to
the level that optimizes the X value and the one that maxi- Parameter Design,” Quality Progress, December 1987, pp. 19-26.
mizes the average S/N ratio, it is usually recommended that 5. For example, see A. Park, Taguchi Methods: Introduction to
the average S/N ratio be given priority. This is due to the fact Quality Engineering (Dearborn, MI: American Supplier Institute,
1991).
that the Taguchi approach focuses first on selecting levels of
controllable factors that reduce variability; bringing the Sanjiv Sarin is a professor at North Carolina Agricultural and
process on target is the secondary objective. Thus, according Technical State University in Greensboro. He received a doctorate in
to the Taguchi approach, design 2 should be recommended. industrial engineering from the State University of New York in
But as a matter of practicality, design 1 was chosen since it Buffalo. Sarin is an ASQC certified quality auditor.
results in significant improvement in the X value while sac-
rificing the S/N value to a relatively lesser degree.
• Factor C. For the factor of paper width, both the average
S/N ratio and the X value indicated that level 2 (6-in. width)
was the best. What did you think about this article?
• Factor D. For the factor of paper length, level 3 (10-in. Quality Progress needs your
length) was ideal according to both the average S/N ratio Excellent Circle #325
feedback. On the postage-paid
and the X value. reader service card inserted toward Good Circle #326
the back of this magazine, please
The ultimate judge circle the number that corresponds Fair Circle #327
with your opinion of the preceding Poor Circle #328
Making paper airplanes is a simple experiment that can help article.
students understand the basic principles of Taguchi’s approach

106 /
Quality Progress May 1997

You might also like