Layout of

36 Orthogonal Arrays
Using Linear Graphs

36.1. Introduction 597

36.2. Linear Graphs of Orthogonal Array L8 597
Linear Graph (1) 598
Linear Graph (2) 598
36.3. Multilevel Arrangement 599
36.4. Dummy Treatment 605
36.5. Combination Design 606
Reference 608

36.1. Introduction
Linear graphs have been developed by Taguchi for easy assignment of experi-
ments. These graphs can be used to assign interactions between factors in an array
to calculate interactions in the design of experiments. But in quality engineering,
linear graphs are not used for assigning interactions but for special cases such as
multilevel assignment. This chapter is based on Genichi Taguchi, Design of Experi-
ments. Tokyo: Japanese Standards Association, 1973.

36.2. Linear Graphs of Orthogonal Array L8

Before explaining the modifications of orthogonal arrays for experimental layout,
it is necessary to know the column or columns where an interaction between col-
umns is confounded. To simplify explanation, orthogonal array L8 is used. Two
linear graphs can be drawn for the L8 table as shown in Figure 36.1.

Linear graph (1) in Figure 36.1 means that the interaction between columns 1 Linear Graph (1)
and 2 comes out to column 3, and the interaction between columns 1 and 4 comes
out to column 5. Column 7 is shown as an independent point apart from the
triangle.

Taguchi’s Quality Engineering Handbook. Genichi Taguchi, Subir Chowdhury and Yuin Wu 597
Copyright © 2005 Genichi Taguchi, Subir Chowdhury, Yuin Wu.
598 36. Layout of Orthogonal Arrays Using Linear Graphs

Figure 36.1
Linear graphs of
orthogonal array L8

If there are four two-level factors, A, B, C, and D, and if A is assigned to column

1, B to column 2, C to column 4, and D to column 7, then interaction A ⫻ B is
obtained from column 3, A ⫻ C from column 5, and B ⫻ C from column 6. Such
assignment problems are easily solved using linear graph (1).

Linear Graph (2) Linear graph (2) is used for an experiment where the interactions between one
particular factor and some other factors are important. This is illustrated in the
following experiment, with two types of raw materials, A1 and A2; two annealing
methods, B1 and B 2; two temperatures, C1 and C2; and two treating times, D1 and
D2.
It is expected that since B1 and B 2 use different types of furnaces, their oper-
ations are quite different and the optimum temperature or optimum treating time
might not be the same for B1 as for B 2. On the other hand, it is expected that a
better raw material must always be better for any annealing method. Accordingly,
only the main effect is cited for factor A. Thus, we need to get the information of
main effects A, B, C, D and interactions B ⫻ C and B ⫻ D.
These requirements are shown in Figure 36.2. Such a layout is easily arranged
using two lines out of the three radial lines of linear graph (2) shown in Figure
36.1.
The remaining columns, 6 and 7, are removed from the form, as shown in
Figure 36.3, with the independent points as indicated. An independent point is
used for the assignment of a main effect; factor A is assigned to any one of the
remaining two points, since only the main effect is required for A. The layout is
shown in Table 36.1.

Figure 36.2
Information required for
a linear graph
36.3. Multilevel Arrangement 599

Figure 36.3
Layout of the
experiment

Such a layout can also be arranged from linear graph (1). Remove column 6,
which is the line connected by columns 2 and 4, indicate it as an independent
point like column 7, then assign factor A, as shown in Figure 36.4.

36.3. Multilevel Arrangement

At the planning stage of an experiment, we sometimes want to include some mul-
tilevel factors. That is to arrange a four- or eight-level column in two-level series
orthogonal arrays, or a nine-level column in three-level series orthogonal arrays.
Using linear graphs, a four- or eight-level factor can be assigned from a two-level
series orthogonal array.
Next, the method of arranging a four-level factor from orthogonal array L8 is
explained. Assume that A is a four-level factor, and B, C ,and D are two-level factors.

Table 36.1
Layout of experiment
B C BC D BD A e
Column 1 2 3 4 5 6 7

1 1 1 1 1 1 1 1
2 1 1 1 2 2 2 2
3 1 2 2 1 1 2 2
4 1 2 2 2 2 1 1
5 2 1 2 1 2 1 2
6 2 1 2 2 1 2 1
7 2 2 1 1 2 2 1
8 2 2 1 2 1 1 2
600 36. Layout of Orthogonal Arrays Using Linear Graphs

Figure 36.4
Layout using linear
graph (1)

To obtain these main effects from an L8 array, it is necessary to prepare a four-

level column in the array. For instance, line 3 with two points 1 and 2 in linear
graph (1) is selected. Then three columns are removed from the linear graph and
rearranged using a four-level column, as shown in Table 36.2. That is, from the
two columns shown as the two points at both ends of the line (or any two of the
three columns), four combinations (11, 12, 21, and 22) are obtained; these com-
binations are then substituted for 1, 2, 3, and 4, respectively, to form a four-level
column. Then columns 1, 2, and 3 are removed from the table, as shown on the
right side of Table 36.2.

Table 36.2
Preparation of orthogonal array L8(4 ⴛ 24) from L8 (27)
L8(27) L8(4 ⴛ 24)
No. 1 2 3 4 5 6 7 123 4 5 6 7

1 1 1 1(1) 1 1 1 1 1 1 1 1 1
2 1 1 1(1) 2 2 2 2 1 2 2 2 2
3 1 2 2(2) 1 1 2 2 2 1 1 2 2
4 1 2 2(2) 2 2 1 1 2 2 2 1 1
5 2 1 2(3) 1 2 1 2 3 1 2 1 2
6 2 1 2(3) 2 1 2 1 3 2 1 2 1
7 2 2 1(4) 1 2 2 1 4 1 2 2 1
8 2 2 1(4) 2 1 1 2 4 2 1 1 2
36.3. Multilevel Arrangement 601

It is seen from the example above that for a four-level column,

a line with points at each end is necessary, as shown in Figure 1o B C D E
36.5. The figure shows a four-level column of three degrees of
freedom replaced by three columns of one degree of freedom. 3 o o o o
When a four-level column is arranged in a two-level series
orthogonal array, the four-level column is substituted for the 2o 4 5 6 7
three columns of the array; the three columns consist of any two
columns and the column of interaction between the two col- A
umns. These three columns are similar to those illustrated in the Figure 36.5
linear graph in Figure 36.5. Notice the graph has a line with two Layout with a four-level factor
points at each end. To do this, pick up a line and the two points
at the ends of the line, select any two points (normally select the two points at the
ends) out of the three columns to form a new four-level column. Letting the four
combinations of the two columns, 11, 12, 21, and 22, correspond to 1, 2, 3, and
4, respectively, a four-level column is then formed. Delete the three columns de-
scribed above and then put the four-level column in the table, as shown in Table
36.2. To determine the degrees of freedom, the three columns with one degree
of freedom each are replaced by a column with three degrees of freedom.
Next, the formation of an eight-level column is illustrated. For this purpose, a
closed triangle (plus the line to represent the interaction between a point at the
top and the baseline in the triangle) from a two-level series linear graph is used.
These seven columns (i.e., three apexes, three bases, and one perpendicular line)
with seven degrees of freedom in total are replaced by a eight-level column with
seven degrees of freedom.
Figure 36.6 shows a triangle chosen from a linear graph of the L16 array. From
apex 1, draw a perpendicular line to base 6; the line is found to be column 7 from
Appendix A. There are eight combinations of 1 and 2 formed by the three columns
(three apexes 1, 2, and 4). The combinations are 111, 112, 121, 122, 211, 212,
221, and 222. As in the case of the four-level arrangement, it is necessary to erase
column 1 through 7 from the orthogonal array after inserting the eight-level col-
umn. Thus, the L16(215) array is rearranged to be an L16(8 ⫻ 28) table. The new
orthogonal array is shown in Table 36.3.

Figure 36.6
Linear graph from an
L16 array
602 36. Layout of Orthogonal Arrays Using Linear Graphs

Table 36.3
Orthogonal array L16(8 ⴛ 28) from L16(215)
Column 1 2 4 A 1 2 4 A

1 1 1 1 2 1 1 5
1 1 2 2 2 1 2 6
1 2 1 3 2 2 1 7
1 2 2 4 2 2 2 8

Col. 1–7 8 9 10 11 12 13 14 15
No. 1 2 3 4 5 6 7 8 9

1 1 1 1 1 1 1 1 1 1
2 1 2 2 2 2 2 2 2 2
3 2 1 1 1 1 2 2 2 2
4 2 2 2 2 2 1 1 1 1
5 3 1 1 2 2 1 1 2 2
6 3 2 2 1 1 2 2 1 1
7 4 1 1 2 2 2 2 1 1
8 4 2 2 1 1 1 1 2 2
9 5 1 2 1 2 1 2 1 2
10 5 2 1 2 1 2 1 2 1
11 6 1 2 1 2 2 1 2 1
12 6 2 1 2 1 1 2 1 2
13 7 1 2 2 1 1 2 2 1
14 7 2 1 1 2 2 1 1 2
15 8 1 2 2 1 2 1 1 2
16 8 2 1 1 2 1 2 2 1

❒ Example

To obtain information concerning the tires for passenger cars, five two-level factors
from the manufacturing process, A, B, C, D, and E, are cited. It is judged that factor
A is most important, followed by B and C. Accordingly, it is desirable to obtain the
interactions A ⫻ B and A ⫻ C. For the life test of tires, four passenger cars, R1,
R2, R3, and R4, are available and there are no driving restrictions during the testing.
36.3. Multilevel Arrangement 603

In other words, once the tires for the test are put on the cars, there is no restriction
on the roads driven or the distance traveled, and the cars are allowed to run as
usual. As a result, the driving conditions among these four cars must differ
considerably.
On the other hand, it is known that the different positions of tires of a car cause
different conditions; the condition of the rear tires is worse than the front and there
is also a difference between the two front tires since a car is not symmetrical, a
driver does not sit in the middle, the road conditions for both sides are not the
same, and so on.
For these reasons, we must consider the difference among the four positions in
a car: V1: front right; V2: front left; V3: rear right; and V4: rear left. Such influences
should not interfere with the effects of the five factors and interactions A ⫻ B and
A ⫻ C.
The real purpose of the experiment is to obtain the information concerning fac-
tors A, B, C, D, and E, and not for R and V. The character of factors A, B, C, D,
and E is different from that of factors R and V.
The purpose of factors A, B, C, D, and E, called control factors, is the selection
of the optimum levels. There are four cars, R1, R2, R3, and R4. For each level of R,
the number of tests is limited to four (four wheels). The factor that has restricted
the number of tests for each level, in this case four, is called a block factor. From
a technical viewpoint, the significance of the difference between levels in a block
factor is not important. Therefore, the effect of such a factor is not useful for ad-
justment of the levels of other factors. The purpose of citing a block factor is merely
to avoid its effect being mixed with control factors.
The factor, like the position of tire, V, is called an indicative factor. It is mean-
ingless to find the best level among the levels of an indicative factor.
From the explanation of block factor or indicative factor, the information of R
(cars) and V (positions of tires) are not needed; these factors are cited so to avoid
mixing the control factors (A, B, C, D, and E).
This experiment is assigned as follows. Factors and degrees of freedom are listed
in Table 36.4. All factorial effects belong to either two levels or four levels; the total
degrees of freedom is 13; hence L16 (with 15 total degrees of freedom) is probably
an appropriate one. (The selection of either a two- or three-level series orthogonal
array depends on the majority of the number of levels of factors.)
R and V are both four-level factors. As described in Chapter 35, a four-level
factor has to replace three two-level columns, which consist of a line and two points
at the ends. Accordingly, the linear graph required in this experiment is shown in
Figure 36.7.
There are six standard types of linear graphs given for an L16 array [1]. But
normally, we cannot find an exactly identical linear graph to the one that is required
in a particular experiment. What we must do is to examine the standard types of
linear graphs, compare them with the linear graph required for our purpose, and
select the one that might be most easily modified. Here, type 3 is selected, for
example, and modified to get the linear graph we require. Figure 36.8 shows
type 3.
604 36. Layout of Orthogonal Arrays Using Linear Graphs

Table 36.4
Distribution of degrees of freedom
Factorial Degrees of Factorial Degrees of
Effect Freedom Effect Freedom

A 1 A⫻B 1
B 1 A⫻C 1
C 1 R 3
D 1 V 3
E 1

First, factorial effects A, B, C, A ⫻ B, and A ⫻ C are assigned to the figure on

the right side. Next, R and V are assigned as a line to the figure on the left side.
To the remaining columns, factors D, E, and error are assigned. The layout is shown
in Figure 36.9 and Table 36.5.
From the table, it is known that in experiment 10, the tire is manufactured under
condition A1B1C2D2E1; the tire thus manufactured is put on V2, the front left wheel
of car R3. This is the case when only one tire is manufactured for each experiment
number.
When four cars are used, two replications are performed for each experiment.
When eight cars are used, the half portion, or 16 tires, are tested by cars R1, R2,
R3, and R4, as shown in Figure 36.9. Then another half portion is tested using cars
R5, R6, R7, and R8, which replace R1, R2, R3, and R4 as follows: R5 for R1, R7 for
R3, R6 for R2, and R8 for R4.

Figure 36.7
Linear graph for
experiment
36.4. Dummy Treatment 605

Figure 36.8
Type 3 linear graph

36.4. Dummy Treatment

To put a two-level factor A in a three-level series table, A is formally treated as a
three-level factor: that is, to provide three levels for A and run one of the two
actual levels twice. Usually, we select one level that is probably more important
than the other, and let this level replicate. For instance, A1 ⫽ A1, A2 ⫽ A2, and
A3 ⫽ A1. A3 is formally treated as the third level of A, but actually it is A1.
When a factor such as A is assigned to column 3 of an L9(34) array, the layout
would look like the one shown in Table 36.6. In the table, 3 in the third column
of the original L9 array is rewritten as 1⬘ to show that it is the dummy level of A.
The main effect A is calculated as

(A1 ⫹ A1⬘)2 A2 (A ⫹ A⬘1 ⫹ A2)2

SA ⫽ ⫹ 2⫺ 1 (36.1)
6r 3r 9r

or

(A1 ⫹ A⬘1 ⫺ 2)2

SA ⫽ (36.2)
18r

where r is the number of replication in orthogonal array L 9.

Figure 36.9
Layout of experiment
606 36. Layout of Orthogonal Arrays Using Linear Graphs

Table 36.5
Layout of experiment
No. A B C D E AⴛC AⴛB Car Position e e

1 1 1 1 1 1 1 1 R1 V1 1 1
2 2 1 1 1 2 2 2 R1 V2 2 2
3 1 2 2 1 1 2 2 R2 V2 1 1
4 2 2 2 1 2 1 1 R2 V2 1 1
5 2 1 1 2 1 2 2 R2 V3 1 1
6 1 1 1 2 2 1 1 R2 V4 2 2
7 2 2 2 2 1 1 1 R1 V3 2 2
8 1 2 2 2 2 2 2 R1 V4 1 1
9 2 1 2 2 2 1 2 R3 V1 1 2
10 1 1 2 2 1 2 1 R3 V2 2 1
11 2 2 1 2 2 2 1 R4 V1 2 1
12 1 2 1 2 1 1 2 R4 V2 1 2
13 1 1 2 1 2 2 1 R4 V3 1 2
14 2 1 2 1 1 1 2 R4 V4 2 1
15 1 2 1 1 2 1 2 R3 V3 2 1
16 2 2 1 1 1 2 1 R3 V4 1 2
Column 11 4 5 3 9 14 15 1 2 1 1
6 8 2 3
7 10

36.5. Combination Design

There is a technique to put two-level factors in a three-level series orthogonal array.
When there are two factors with two levels, and if their interaction is not required,
the combination design is used instead of the dummy treatment.
A combined factor (AB) with three levels is formed as follows:
(AB)1 ⫽ A1B1

(AB)2 ⫽ A2B1

(AB)3 ⫽ A1B 2
The main effect, A, is obtained from the difference of (AB)1 and (AB)2. Similarly,
the main effect, B, is obtained from (AB)1 and (AB)3.
When there are 20 two-level factors, 10 three-level factors are formed by each
of the two factors; these combined factors are assigned to an L27 array. Because
36.5. Combination Design 607

Table 36.6
Assignment of A to column 3
Level
No. 1 2 3 4

1 1 1 1 1
2 1 2 2 2
3 1 3 1⬘ 3
4 2 1 2 3
5 2 2 1⬘ 1
6 2 3 1 2
7 3 1 1⬘ 2
8 3 2 1 3
9 3 3 2 1

there are so many two-level factors, it is better to assign these factors to an L32(231)
array. But for an experiment such as the 210 ⫻ 36 type, it is more practical to
combine each two for the 10 two-level factors to get five three-level combined
factors and assign the other six three-level factors to an L27 array.
Also, if it is necessary to obtain the interactions between combined factor AB
and the three-level factor C, AB and C are assigned to the two dots of a line to
obtain interactions A ⫻ C and B ⫻ C.
A three-level factor (A) and a two-level factor (B) are combined to form a
combined four-level factor:

(AB)1 ⫽ A1B1

(AB)2 ⫽ A2B1

(AB)3 ⫽ A3B1

(AB)4 ⫽ A1B 2

The variation of a combined three-level factor (AB) by two factors (A, B) of two
levels is calculated as

1
S(AB) ⫽ [(AB)12 ⫹ (AB)22 ⫹ (AB)32] ⫺ CF (36.3)
r

where r is the number of units of A1B1, A2B1, and A1B 2.

However, the following relationship does not hold, since A and B are not
orthogonal:

S(AB) ⫽ SA ⫹ SB (36.4)
608 36. Layout of Orthogonal Arrays Using Linear Graphs

If we want to separate A and B, calculate

[(AB)1 ⫺ (AB)2]2
SA ⫽ (36.5)
2r
[(AB)1 ⫺ (AB)3]2
SB ⫽ (36.6)
2r
and decompose S(AB) into
S(AB) ⫽ SA ⫹ Sres (36.7)
or
S(AB) ⫽ SB ⫹ S⬘res (36.8)
In the analysis of variance table, S(AB) is listed inside the table, while SA and SB are
listed outside.

Reference
1. ASI, 1987. Orthogonal Arrays and Linear Graphs. Livonia, Michigan: American Supplier
Institute.