QUALITY ENGINEERING

UNIT II. EXPERIMENTS WITH ORTHOGONAL ARRANGEMENTS

GOALS):

Apply the appropriate techniques to develop the design of experiments (DDE) method.

TOPICS:

2.1 Planning and Conducting Experiments.

2.2 The design (2 3 ).
2.3 Definition of Orthogonality.
2.4 The Orthogonal Arrangement L 4 (2 3 ).
2.5 Analysis of variance in orthogonal arrangements.
2.6 Reasons to use Orthogonal arrangements.
2.7 Other arrangements for factors at two levels.
2.8 Linear Graphs.
2.9 Orthogonal arrangements for three-level factors.
2.10 Methods to modify orthogonal arrangements.

2.1. Planning and conducting experiments

An experiment is a procedure through which an attempt is made to verify (confirm or verify)

one or several hypotheses related to a certain phenomenon , through the manipulation and
study of the correlations of the variables that are presumably its cause .

Obviously, experimentation must be carefully planned (designed) to provide the desired

information. Such planning must consider two important aspects related to all
experimentation:
1. Experimentation is usually expensive. The ability to experiment is limited by the cost in time
and resources (personnel, starting products, etc...). Therefore, an optimal organization of
experimentation should contemplate the smallest number of experiments that allows obtaining
the information sought.
2. The observed result of an experiment (y) has uncertainty:
y=h+e
Where h is the “true” (unknown) result of the experiment and e is a random contribution,
which varies each time the experiment is repeated. Therefore, Statistics, a discipline that
provides the tools to work in environments of uncertainty, plays a fundamental role in the
design of experiments and in the evaluation of experimental results.
The analysis of the experimental results will allow conclusions to be drawn about the system
under study and future actions to be decided. Both because of the importance of the
decisions that can be made and because of the high cost of experimentation, it does not
seem appropriate to leave the choice of experiments and the evaluation of the results to the
mere intuition of the experimenter. It seems more reasonable to use a mathematical and
ITESRC _____________________________________________________________ 1
 QUALITY ENGINEERING

statistical methodology that indicates how to plan (design, organize) the sequence of
experiments in an optimal way, so that both the cost of experimentation and the influence of
experimental error on the information sought are minimized. Such planning and analysis is the
main objective of the Statistical Design of Experiments.

What is Statistical Design of Experiments?

The Statistical Design of Experiments (SED), also called experimental design, could be
defined as a methodology based on mathematical and statistical tools whose objective is to
help the experimenter:
1. Select the optimal experimental strategy that allows obtaining the information sought with
the minimum cost.
2. Evaluate the experimental results obtained, guaranteeing maximum reliability in the
conclusions obtained.
3. In what situations is the Statistical Design of Experiments applied?
The situations in which the DEE can be applied are very numerous. In general, it is applied to
systems like the one shown in Figure 1, in which one or more dependent experimental
variables or responses (y) are observed whose value depends on the values of one or more
controllable independent variables (x) called factors. . The responses may also be influenced
by other variables that are not controlled by the experimenter. The relationship between x and
y does not have to be known.

Input variables output variable

X1, x2, ….,xn and

Uncontrolled variables

Examples of experimental systems are:

 A chemical reaction, whose yield (y) can be a function of, among others, the reaction
time (x1), the reaction temperature (x2) and the type of catalyst (x3) used. Other
variables that may influence are, for example, the purity of the reagents, the
cleanliness of the material, the speed of stirring,...
 A chromatographic separation, where the separation time depends on the pH and the
percentage of organic modifier of the mobile phase.
 A food, produced by mixing different proportions (x) of its ingredients, which gives rise
to different smells and flavors (y)

These are just some examples of the wide field of application of the DEE, which covers, in
industry, from the laboratory to the production process. By facilitating faster and lower cost
product development, the DEE plays a fundamental role from an industrial point of view and
provides an important competitive advantage for the company that uses it.

Although the DEE can be applied when a single factor is studied (such as in choosing the
most appropriate standards to construct a calibration line), its advantages are undoubtedly
better appreciated when more than one factor must be studied.
ITESRC _____________________________________________________________ 2
 QUALITY ENGINEERING

Experimentation on systems like the one described in the previous Figure usually pursues
one of the following objectives:

 Obtain initial knowledge about a new system under study. What factor values can the
research focus on?
 Determine the influence of the factors on the observed responses. Of all the factors
that affect the process, which ones have the most influence? How do they interact with
each other?
 Optimize responses. Which factor values provide the highest quality responses?
 Determine the robustness of the system. How do uncontrolled variations in factor
values affect the response?

The application of the design of experiments requires considering the following stages that will
be discussed below:

1. Understand the problem and clearly define the objective.

2. Identify the factors that could potentially influence the objective function, and the values
that these can take. The necessary information will be searched among these values.
3. Establish an experimental strategy, called an experiment plan.
4. Carry out the experiments with the values of the factors decided in point 3 to obtain the
values of the studied responses.
5. Answer the questions posed, either directly or using a mathematical model. If necessary,
return to stage 1.

2.2. The design ( 2 3 ).

The design of experiments is a statistical technique that helps us identify which factors or
variables affect the behavior of a production process.

The experiment is a structured and coherent set of tests that are analyzed to gain an
understanding about the process.

The process of planning, executing and analyzing the experiment in such a way that valid and
useful inferences are achieved is what is called: DESIGN OF EXPERIMENTS

The conventional method in experimentation is to vary one factor at a time, keeping all others
constant to find the best values or levels for each factor. This causes the number of experiments
to be small, but suffers from several problems:

1. Each comparison is based on a limited amount of data and is therefore low in reliability.
2. The fact that changing one factor can influence the effect of another factor is not
accepted.

ITESRC _____________________________________________________________ 3
 QUALITY ENGINEERING

3. Keeping other factors constant is not always practical and if it were, the variability of
the system would be artificially reduced. This can be achieved in laboratory conditions
but not in production environment.
4. Unstructured experimentation is encouraged with a consequent increase in time and
cost.

DESIGN 2 3

Suppose that three factors A, B and C are being studied, each with two levels. This design is
known as a factorial design, 2 3 and the eight treatment combinations can be represented
graphically by a cube.

2k
There are actually three different notations that are widely used for runs in design :

a. The first is the " +,- " notation, called " geometric ".
b. The second consists of the use of lowercase letters to identify treatment
combinations .
c. In the third, the digits 1 and 0 are used to denote the high and low levels of the factor ,
respectively.

k
A complete replica of such a design requires 2 x 2 x... to be collected. x 2 = 2
observations and is known as the 2 k general design.

2.3. Definition of orthogonality

Orthogonality of factors: two factors FI and FJ with I and J levels, respectively, are orthogonal
if in each level i of FI the number of observations of the J levels of FJ are in the same
proportions. This property allows us to separate the simple effects of the factors under study.

The orthogonal array is an engineering tool that simplifies and in some cases eliminates
much of the statistical design efforts. It is a way to examine many factors simultaneously at
low cost. Dr. Taguchi recommends the use of orthogonal arrays to make matrices that
contain the controls and noise factors in the design of experiments. It has simplified the use
of this type of design by incorporating orthogonal arrays and linear graphs, finally, in contrast
to traditional approaches as noise equivalents: as long as the interactions are relatively
smooth, the main effects analyst will provide us with optimal conditions. and good
reproducibility in an experiment.

Orthogonal arrangements are tools that allow the engineer to evaluate what
how robust are the process and product designs with respect to the
noise factors.

ITESRC _____________________________________________________________ 4
 QUALITY ENGINEERING

Dr. Taguchi's method for designing experiments uses techniques that involve low costs and
are applicable to the problems and requirements of modern industry. The purpose of product
design is to find that combination of factors that provides the most stable and reliable
performance at the lowest manufacturing price.

Dr. Genichi Taguchi's quality engineering system is one of the greatest engineering
achievements of the 20th century. Dr. Taguchi's philosophical work began to take shape in
the early 1950s when he was recruited to help improve the Japanese telephone system that
had been designed for World War II. Taguchi employed design experiments using especially
a table known as "orthogonal arrays" to address design processes. Orthogonal arrays are a
special set of Latin charts, constructed by Taguchi to plan product design experiments.

Taguchi orthogonal array analysis is used to produce the best parameters for optimal
process design, with the minimum number of experiments (tests). The results obtained for
the orthogonal arrangements are analyzed to obtain the following objectives:

 Estimate the contribution of individual factors that influence quality at the product
design stage.
 Gain the best condition for a process or a product, so that good quality characteristics
can be sustained.

The design of experiments with the Taguchi method uses matrix arrangements called
orthogonal arrays. These arrangements are tables of numbers that help us analyze
appropriate combinations of factors and levels for each experiment.

The characteristics of these arrangements are:

1. Each level appears an equal number of times.

2. Each level of each factor appears with each level of any other.
3. They are efficient and reliable.
4. They greatly reduce the number of experiments.
5. Some can be used for particular factor and level experiments.

Orthogonal arrangements and their advantage

The advantage of orthogonal arrangements is that they can be applied to experimental

design involving a large number of factors.

Orthogonal arrangements and their disadvantage

The disadvantage of the orthogonal arrangement is that it can only be applied in the initial
stage of product or process system design. An orthogonal arrangement ensures that the
effect of "B" on "A1" is the same effect of "B" on "A2". This way you can be sure that you are
making comparisons between effects of levels of a factor.

Orthogonal arrangement representing La(b)C where:

ITESRC _____________________________________________________________ 5
 QUALITY ENGINEERING

L = Indicates that it is an orthogonal arrangement

a = Number of experimental runs
b = Number of levels for each factor
c = Number of columns or factors of an orthogonal arrangement.

2.4. The orthogonal arrangement L 8 (2 3 ).

The following table shows an orthogonal arrangement L 4 ( 2 3 )

4= No. of experiments or runs

2= levels
3= factors

Exp. No. TO b c Results

1 1 1 1 and 1
2 1 2 2 and 2
3 2 1 2 and 3
4 2 2 1 and 4

In the literature. Orthogonal arrangements are constituted differently. Instead of numbers 1 s

and 2 8s like those in the previous table, “+” and “-” signs are used as shown in the following
table:

Exp. No. TO b c Results

1 - - - and 1
2 - + + and 2
3 + - + and 3
4 + + - and 4

In the linear equation of the form:

L = C 1 Y 1 + C 2 Y 2 + ........+ C n Y n

Where the C j are constants

If Σ C j = 0 the equation is said to be a “CONTRAST”

If you have two contrasts:

L 1 = A 1 Y 1 + A 2 Y 2 +.........+ A n Y n and

L 2 = B 1 Y 1 + B 2 Y 2 +.........+ B n Y n and if

ΣAiBi=0
ITESRC _____________________________________________________________ 6
 QUALITY ENGINEERING

So the equations are orthogonal

An orthogonal arrangement then represents a set of linearly independent equations.

In the following Table shows the 128 combinations of a total factorial design and the 8
experiments of an L 8 orthogonal arrangement (2 3 ).

Orthogonal arrangement: L 8 ( 2 7 ).

8= No. of experiments or runs

2= levels
7= factors

ORTHOGONAL ARRANGEMENT L 8 (2) 7

1 2 3 4 5 6 7
AN
na TO b c d F g RESULTS
D
1 1 1 1 1 1 1 1 Y1 LEVEL 1
2 1 1 1 2 2 2 2 Y2 LEVEL 1
3 1 2 2 1 1 2 2 Y3 LEVEL 1
4 1 2 2 2 2 1 1 Y4 LEVEL 1
5 2 1 2 1 2 1 2 Y5 LEVEL 2
6 2 1 2 2 1 2 1 Y6 LEVEL 2
7 2 2 1 1 2 2 1 Y7 LEVEL 2
8 2 2 1 2 1 1 2 Y8 LEVEL 2

Annotations

1 Taguchi methods optimize to find the best values and robustness.

2 In a system with 15 design parameters with two levels, there are 32768! combinations.
3 In a case like this, either the method is applied: “choose and have faith” or.
4 Orthogonal arrangements are used where only 16 of the 32768 combinations have to be
analyzed.
5 Orthogonal arrays were developed by R. Fisher in 1920.
6 In orthogonal arrangements there is a high probability of finding the optimum.

2.5. Analysis of variance in orthogonal arrangements

ITESRC _____________________________________________________________ 7
 QUALITY ENGINEERING

The following example illustrates the use of the Taguchi method to optimize a heat treatment
process.

Example: Hardening of injection nozzles

Injection nozzles are carburized to achieve a specific hardness and case depth in the valve
seat area.

Suppose that it is required to optimize the heat treatment process in such a way that the
distortion of the; nozzle body while maintaining nominal case depth and hardness.

The chosen factors and their levels associated with carburization and the tempering process
are listed below:

FACTORS LEVELS
A: PACKAGING DENSITY 3 baskets 1 basket

B: OVEN TEMPERATURE 870°C 840°C

C: BAKE TIME 8 hrs. 6 hrs.

D: LOAD SIZE 100 Kg. 350kg

E: DEGREE OF OIL High low

AGITATION
F: ELEVATION/DIFFUSION 60% 50%
RATIO OF CO 2
G: OIL TEMPERATURE 150°C 80°C

The above factors and their levels have been assigned in an orthogonal arrangement L 8 .

The answer (quality characteristics) is distortion in microns (um).

L8 TO b c d AN F g and
D microns Level 1 result
1 1 1 1 1 1 1 1 42
2 1 1 1 2 2 2 2 72
3 1 2 2 1 1 2 2 28 Ỹ box
4 1 2 2 2 2 1 1 41
Random mind
5 2 1 2 1 2 1 2 84 Ỹ A 2 level 2
6 2 1 2 2 1 2 T = 424 63
1 / 8 = 53
7 2 2 1 1 2 2 1 35
Below is the 8 2 2 1 2 1 1 2 59 response table for
the average T = 424 distortion values.

ITESRC _____________________________________________________________ 8
 QUALITY ENGINEERING

LEVEL 2 – LEVEL 1 = DIF.

Difference largest absolute value = range (order from largest to smallest)

TO b c d AND F g
LEVEL 1 45.75 65.25 52.00 47.25 48.00 56.50 45.25
LEVEL 2 60.25 40.75 54.00 58.75 58.00 49.50 60.75
DIF. 14.50 -24.5 2.00 11.50 10.00 -7.00 15.50
RANGE 3 1 7 4 5 6 2

EXAMPLE:

A 1 = 42 + 72 + 28 + 41 / 4 = 45.75

The range is progressive numbers according to the order of absolute differences.

The average value of the values is given by:

T = 45 .75 + 60.25 / 2 = 53
This value is used for comparison purposes and must be equal to any pair of values in the
table.

As the quality characteristics (distortion) are of the “lower the better” type, the optimal
solution obtained from the response table is the combination:

A1B2C1D1E1F2G1

1.5 PREDICTIONS OF OPTIMAL CONDITIONS

Since the optimal combination (A 1 B 2 C 1 D 1 E 1 F 2 G 1) was not experimented, we

must use the following prediction formula to predict the optimal value of the nozzle distortion:

Prediction = (Ā – T) + (B - T) +.....+ T

A 0 = optimal value

Where A 0 , B 0 ,... are the optimal values taken from the solution in the answer
table and T is the average of the values:

T = 42 + 72 + 28 +41 +84 +63 +35 + 59/ 8 = 53

Then the distortion prediction is:

_ __ _ __ _ _ _ _ _ _ _ _ _ _ _ _ _
U = ( Ā1 – T) + (B2 – T) + ( C1 – T) +( D1 – T) + ( E1 – T) + ( F1 – T) + ( G1 – T) +T

____ _ _ _ __ _ _ _ __ ____ _

ITESRC _____________________________________________________________ 9
 QUALITY ENGINEERING

U = Ā 1 B 2 C 1 D 1 E 1 F 1 G 1 - 6T prediction

U = sum of the optimal values minus (n- 1) times the average value with n = No. of
factors
Therefore :
U = 45.75 +40.75 + 52.00 + 47.25 + 48.00 +49.50 + 45.24 – (6 * 53) = 10.50 um.

The experiment can be carried out as a confirmatory test

NOTE

Only data were analyzed for distortion

It is necessary to repeat the analysis for hardness to verify that the optimal
conditions for distortion ensure sufficient depth for the case.

Another way to determine the prediction consists of choosing from the orthogonal
arrangement the combination closest to the optimal one and subtracting from the
value of the distortion of this combination the differences in the changes in the
levels of the factors that are different. This is:

Taking experiment No. 1: A 1 B 1 C 1 D 1 E 1 F 1 G 1  42 um.

Change from B 1 to B 2 = 42 – 24.5 = 17.5

Change from F 1 to F 2 = 17.5 – 7 = 10.5 um.

Observation:

Although the above is a quick way to calculate the prediction, it is not applicable in
all cases.
To do this, the following conditions must be met:

 The array must be completely filled (no empty columns).

 If there is any iteration, the selected experiment must have the factors
interacted at the same level as the optimal one.
 The additivity condition must be met with the factors.

Example 2

In a panel forming process, an undesirable characteristic is the emission of formaldehyde in

the final product. It is desired that this emission be as minimal as possible. It is currently
estimated at 0.45 ppm. (Parts per million).
It is believed that five factors may be affecting the emission, these are: type of resin, solution
concentration, pressing cycle time, humidity and pressure.

ITESRC _____________________________________________________________ 10
 QUALITY ENGINEERING

If you want to analyze the effect of these factors, it is necessary to vary them, that is, test
them under different values each. Each of these values is called a level. At least two different
levels or values are required for each factor. We arbitrarily call one of them low level or level
“1”, the other high level or level “2”.

Factor Descripción Nivel I Nivel 2

A Tipo de resina Tipo I Tipo II
B Concentración 5% 10%
C Tiempo de ciclo de prensado 10 seg 15 seg
D Humedad 3% 5%
E Presión 800 psi. 900 psi.

In this case we are interested in analyzing the effect of 5 effects or factors at two levels each,
therefore, an L8 orthogonal arrangement will be used. This implies that 8 tests or
experimental conditions will be executed. On the other hand, there are 7 columns, each
column can be assigned or associated with a factor. If in particular, we assign the factors in
order to the first five Columns, leaving the last two columns free, the arrangement is:

Note that in the empty columns, 6 and 7, the letter e 1 , e 2 respectively has been written to
indicate that the natural variation or random error will be evaluated in them.
If no factor is assigned, it is expected that natural variation will manifest itself there. Yi results
are shown in ppm.

The analysis of results can be carried out in two different ways. One of them through a series
of graphs, the other through analysis of variance. In this example, first the use of analysis of
variance is shown, then the use of graphs is shown.

Variance analysis

1) As a first step, the totals of the response variable or readings are obtained for each of the
factor levels.

To calculate the totals for each level of factor A, we note that the first four tests of the array
were carried out with the factor at its level 1 (Resin type I) and the next four at its level 2
(resin type II).

The totals are therefore:

A 1 = total of the readings that were taken with factor A at its level 1
ITESRC _____________________________________________________________ 11
 QUALITY ENGINEERING

= 0.49+0.42+0.38+0.30=1.59

A 2 = total of the readings that were taken with factor A at its level 2
= 0.21+0.24+0.32+0.28= 1.05

For factor D, tests 1,3,5 and 7 were carried out at level 1 (5% humidity), therefore the totals
are:

D 1 = Total of the readings that were taken with the D factor at its level 1
= 0.49+0.38+0.21+0.32= 1.40

D 2 = Total of the readings that were taken with the D factor at its level 2
= 0.42+0.30+0.24+0.28= 1.24

In summary we have:

Factor TO b c d AND and and

Level 1 1.59 1.36 1.51 1.40 1.39 1.28 1.35
Level 2 1.05 1.28 1.13 1.24 1.25 1.36 1.29
2.64 2.64 2.64 2.64 2.64 2.64 2.64

Note that the sum of the two levels must always give the total of the eight readings 2.64.

2) Next, a quantity is obtained that we will call the sum of squares, which is calculated as
follows:

Sum of the squares of the factor x= SS X= (Total level 2 – Total level 1) 2 / n

Where “n” represents the total number of readings that were taken.

So for example, for factor A, we will have to given that n=8 

SSA= (A 2 –A 1 ) 2 / 8= (1.59-1.05) 2 / 8=0.03645 with 1 g .1

For factor B we have

SSB= (B 2 –B 1 ) 2 / 8= (1.28-1.36) 2 / 8= 0.00080 with 1 g.1

Similarly
SSC= (C 2 –C 1 ) 2 / 8= (1.13-1.51) 2 / 8= 0.01805 with 1 g.1
SSD= (D 2 –D 1 ) 2 / 8= (1.24-1.40) 2 / 8= 0.00320 with 1 g.1
SSE= (E 2 –E 1 ) 2 / 8= (1.25-1.39) 2 / 8= 0.00245 with 1 g.1
SSe= 0.00080 with 1 g.1
SSe= 0.00045 with 1 g.1

ITESRC _____________________________________________________________ 12
 QUALITY ENGINEERING

The sum of squares of the columns where no factor was assigned (SSe) are taken as error
estimates and added.

SSe= 0.00080+0.00045= 0.00125 with 2 g.1

3) An ANOVA table is constructed, this is :

Effect H.H G.L. V Fexp

TO 0.03645 1 0.03645 58.32
b 0.00080 1 0.00080 1.28
c 0.01805 1 0.01805 28.88
d 0.00320 1 0.00320 5.12
AND 0.00245 1 0.00245 3.92
Mistake 0.00125 2 0.000625
Total 0.0622 7

Under the SS column you have the sums of squares. Under the Gl column (degrees of
freedom), we will have the number of columns that were used to evaluate the factor, in this
case, it can only be one for each factor and more than one only in the case of error.

Column V is obtained by dividing the number under the SS column by the number in the
GL column.

Thus, for example, for factor A we have

SSA= 0.03645, GL of A=1

V= SSA/GL= 0.03645/1= 0.03645

Finally, the value of Fexp is obtained by dividing the value of V of each factor by the value of
V for estimating the error.

Fexp of A= V(A) / V(error)= 0.03645/0.000625=58.32

4) We obtain the following conclusions:

All those factors that have an Fexp value greater than 2 are considered to affect the
response variable, formaldehyde emission in this case. These are called significant factors.

In this example, factors A, C, D and E, type of resin, cycle time, humidity and pressure
respectively, are significant.

It is customary that those effects that were not significant are considered as random error, in
order to obtain a better estimate (with a greater number of degrees of freedom).
ITESRC _____________________________________________________________ 13
 QUALITY ENGINEERING

In this case for example, a better estimate of SSe is:

SSe= SSB + SSe= 0.00080+0.00125= 0.00205
With 1 + 2 = 3 degrees of freedom and (Ve)= (SSe)/3= 0.00205/3= 0.00068
Estimates obtained in this way are usually written in parentheses.
The ANOVA table is now
Effect H.H G.1 V Fexp
TO 0.03645 1 0.03645 53.60
c 0.01805 1 0.01805 26.54
d 0.00320 1 0.00320 4.71
AND 0.00245 1 0.00245 3.60
Mistake 0.00205 3 0.00068
Total 0.0622 7

It remains for us to decide at what level each significant factor will have to be set, and what
we can expect. To make this decision, it is very helpful to obtain the averages of the readings
taken at each level for each of the significant factors.

The average formaldehyde emission for each level is obtained by dividing each of the
totals by 4, (c/total is the sum of four readings).
 A 1 = A 1 /4= 1.59/4= 0.3975
 A 2 = A 2 /4= 1.05/4= 0.2625

The rest of the averages are:

Factor Level 1 Level 2
TO  A 1 = 0.3975  A 2 = 0.2625
b  B 1 = 0.3400  B 2 = 0.3200
c  C 1 = 0.3775  C 2 = 0.2825
d  D 1 = 0.3500  D 2 = 0.3100
AND  E 1 = 0.3475  E 2 = 0.3125

The general average denoted as Y is:

Y= (0.49+0.42+0.38+0.30+0.21+0.24+0.32+0.28)/8=T/n= 2.64/8= 0.33

The factors A, C, D and E that affect formaldehyde emission must be set at the level that
minimizes the emission, that is, at the level that obtains the lowest average, in this example;
A 2 , C 2 , D 2 and E 2 ; type II resin, 15 seconds pressing time, 5% humidity and 900 psi.

ITESRC _____________________________________________________________ 14
 QUALITY ENGINEERING

Factor B plays an extremely important role here. Since it does not affect the emission of
formaldehyde, within the analyzed range, it is used to reduce production costs. This is done
by setting it to its most economical level. What will be the expected emission level under the
proposed new emissions?
To answer this question, for each significant effect a subtraction is calculated, which we will
call the effect of each factor with respect to the general average, in this case the effect is:

EF A = (average under the proposed condition of the general average factor)

= A 2 – Y= 0.2625-0.3300= -0.0675 (A was set at its level 2)
EF C = C 2 – Y= 0.2825-0.3300= -0.0475
EF D = D 2 – Y= 0.3100-0.3300=-0.0200
EF E = E 2 – Y= 0.3125-0.3300= -0.0175

Finally, the expected result under conditions A2, C2, D2, E2, which we will call Yest. It is
calculated by adding to the general average AND all the effects of the significant factors.

Yest= Y + EF A + EF C +EF D +EF E= 0.3300-0.0675-0.0475-0.0200-0.0175=0.1775

Analysis using graphs

There is an alternative to ANOVA analysis, this is a series of graphs shown below.

1) First, the averages are obtained at each level, for each of the factors, including the empty
columns.
To do this, we find the totals for each level and divide by the number of readings with
which each total was obtained. For our example, we already have the totals at each level in
the previous section. The averages are:

Factor TO b c d AND and and

Level 1 0.3975 0.3400 0.3775 0.3500 0.3475 0.3200 0.3325
Level 2 0.2625 0.3200 0.2825 0.3100 0.3125 0.3400 0.3225
Global average Y= T/n= 2.64/8 = 0.33

Note that for each factor, one of the averages is greater and the other is less than the
global average. This must always happen.

2) Calculate the difference between the average levels for each factor, and order them from
highest to lowest in absolute value.

This is for example for factor A

A 1 – A 2 = 0.3975 – 0.2625= 0.1350; for the rest we have:

ITESRC _____________________________________________________________ 15
 QUALITY ENGINEERING

Factor TO b c OF and and

Difference 0.1350 0.0200 0.0950 0.0400 0.0350 0.0200 0.0100

In the ANOVA table we find the results obtained previously:

No. TO b c d AND e1 e2 Yi
1 1 1 1 1 1 1 1 0.49
2 1 1 1 2 2 2 2 0.42
3 1 2 2 1 1 2 2 0.38
4 1 2 2 2 2 1 1 0.30
5 2 1 2 1 2 1 2 0.21
6 2 1 2 2 1 2 1 0.24
7 2 2 1 1 2 2 1 0.32
8 2 2 1 2 1 1 2 0.28
T1 1.59 1.36 1.51 1.40 1.39 1.28 1.35 Tot
T2 1.05 1.28 1.13 1.24 1.25 1.36 1.29 2.64
SS 0.03645 0.00080 0.01805 0.00320 0.00245 0.00080 0.00045 Ve
gl 1 1 1 1 1 2
V 0.03645 0.00080 0.01805 0.00320 0.00245 .00062F 58.32
1.28 28.88 5.12 3.92
Sg Yeah No Yeah Yeah Yeah
P 1 0.3975 0.3400 0.3775 0.3500 0.3475 AND
P2 0.2625 0.3200 0.2825 0.3100 0.3125 0.33
Neither 2 - 2 2 2
Eff -0.0675 -0.0475 -0.0200 -0.0175

Yest. = Y + Ef A 2 + Ef C 2 + Ef D 2 + Ef E 2
T1 = Total readings at level 1
T2 = Total readings at level 2

ITESRC _____________________________________________________________ 16
 QUALITY ENGINEERING

n = Total number of readings

H.H = (T 2 - T 1 ) 2 /n
gl = Degrees of freedom (columns)
V = SS/df
F = V/Ve
Sg = Significant effect?
P1 = Average level 1
Q2 = Average level 2
Neither = Selected level
Eff = Effect of the variable
AND = Average of all data
Yes = Estimated value of the variable under the proposed conditions

Ordering from greatest to least absolute value (ignoring the sign), we have:

Factor TO c d AND b and and

Difference 0.1350 0.0950 0.0400 0.0350 0.0200 0.0200 0.0100

It can be seen that the order in which the previous data were left is also the order from
highest to lowest Fexp obtained with the ANOVA.

Following the previous order, a graph is obtained as shown below:

.40

.35

.33

.30

.25

A1A2C1C2D1D2E1E2B1B2e1e2e1e2

Using this graph, the effect of each factor can be evaluated. The higher the line of each factor,
or the more vertical it is, the greater the effect of this factor.

We observe a group of inclined lines, followed by a group of lines that suddenly “lie down”
or become horizontal. It is expected that the lines that have empty columns or random errors
will be practically horizontal.

Note that the conclusions reached in this example are similar to those of the ANOVA, that
is, significant factors A, C, D and e, likewise the recommended levels can be quickly

ITESRC _____________________________________________________________ 17
 QUALITY ENGINEERING

identified, if we wish to reduce the response variable, we take the lowest level, in this case A
2 , C 2 , D 2 and E 2 , that is, the points below the global average line.

In conclusion, the graphic method can be used for exposition or presentation purposes and
the ANOVA for the purposes of making a more objective decision.

INTERACTIONS

An interaction is said to occur when a change in the level of one factor changes the
influence of another factor on the response.

It is then said that the two factors in question INTERACT.

An example of interaction occurs in the use of a catalyst. A factor may not have an effect until
the catalyst is added.

Consider the case of baking a cake in the oven.

Two of the control factors are: temperature and baking time.

If the quality characteristic is the height that the cake rises.

CONTROL FACTORS LEVELS

A: BAKE TEMPERATURE LOW high
B: BAKE TIME 2 HR. 1 hr.

Quality characteristics Y = cake height

The following graphs show the main effects of the factors:

AND
AND

At 1 A2B1B2

Analyzing the graphs. The optimal conditions seem to be A 2 B 2

That is: high temperature for 2 hours.

However, if we study the interaction we will see the effect shown in the following figure:

ITESRC _____________________________________________________________ 18
 QUALITY ENGINEERING

B1

B2

At 1 To 2

The combination of high temperature and long cooking time will cause the cake to burn.

THEREFORE TEMPERATURE AND TIME INTERACT.

The following graphs show when there is a strong interaction, a weak interaction or when
there is no interaction.

B1

B2

At 1 To 2

If the lines are parallel there is no interaction.

The following equality must be fulfilled
A 1 B 1 + A 2 B 2 = A 1 B 2+ A 2 B 1
(A 1 B 1 * A 2 B 2 ) = ( A 1 B 2 * A 2 B 1 ) = 0

(A*B) 1 (A*B) 2

Remember that the notation A*B does not indicate multiplication but rather interaction of
factor A with factor B if A*B shows an effect then there is an INTERACTION.

ITESRC _____________________________________________________________ 19
 QUALITY ENGINEERING

To analyze two factors and their interaction, an orthogonal arrangement L 4 can be used as
shown in the following table:

L4 b A*B
TO
1 2 3
1 1 1 1
2 1 2 2
3 2 1 2
4 2 2 1

Note that in the previous example, in the interaction between two two-level factors it was
necessary to make a comparison to identify, this is a DEGREE OF FREEDOM.

In general, the degrees of freedom necessary to study an interaction are equal to the product
of the degrees of freedom of each factor:

Interaction DOF = DOF (A) * DOF (B)

* = if indicates multiplication

Example 1

If you have a three-level factor and another two-level factor, the degrees of freedom of
the interaction will be:

Three-level factor A 3 _____________________________ DOF (A) = 3 – 1 = 2

Two-level factor B 2 _____________________________ DOF (B) = 2 – 1 = 1

AND
DOF (A * B) = 2*1 = 2
In this case, they are necessary
Two comparisons B1
B2

A|A2A3
Example 2
If there are two three-level factors, the degrees of freedom of the interaction will be:

ITESRC _____________________________________________________________ 20
 QUALITY ENGINEERING

Three-level factor A 3 _____________________________ DOF (A 3 ) = 3 – 1 = 2

Two-level factor B 3 _____________________________ DOF (B 3 ) = 3 – 1 = 2

DOF (A * B) = DOF (A 3 ) * DOF (B 3 ) = ( 2 )( 2 ) = 4

B1
IN THIS CASE 4 ARE REQUIRED
COMPARISONS.
B2
B3

A1A2A3

Main effects of factors

The effect is defined as the change in the response produced by the change in the level of
the factor.

This is often called the “main effect” because it refers to the effects of first interest in the
experiment.

For example, consider the data shown in the following table:

Factor B
20 30
Factor A

40 52

The main effect of factor A can be seen as the difference between the average response of
the first level of A and the average response of the second level of A, that is:

A = 40+52 / 2 – 20 + 30 / 2 = 21
That is, by increasing factor A from level 1 to level 2, it causes an average increase in the
response of 21 units. In a similar way. The main effect of B will be:
B = 30 + 52 / 2 – 20 + 40 / 2 = 11

If the factors have more than two levels, the above must be modified.

In some experiments, we may find that the difference in response between levels of one
factor is not the same at all levels of the other factors. When this occurs, there is interaction
between the factors. For example, consider the data in the following table:

Factor B

ITESRC _____________________________________________________________ 21
 QUALITY ENGINEERING

20 30

Factor 50 12
A

At the first level of B the effect of A is:

A = 50 – 20 = 30

At the second level of B the effect of A is:

A = 12 – 40 = - 28

Since the effect of A depends on the chosen level of B, then there is interaction between A
and B.

The previous concepts are illustrated in the following graphs.

B2

B1
B1 B2
B2

B1 B1 B2

At 1 To 2
At 1 To 2

Graph for security data. Board

Graph for the data lines (not parallel) if there are
First table (almost parallel lines) interaction
no interaction!

INTERACTION AND CONFUSION

It may happen that some interaction may be confused in the experiment:

For example, suppose that the effects of three control factors A, B, and C were studied using
the experimental program in Table L 8 below.

Door No. b c empty Data

TO 4,5,6,
and 7
1 1 1 1 1 5
2 1 1 1 2 5
3 1 2 2 1 9

ITESRC _____________________________________________________________ 22
 QUALITY ENGINEERING

4 1 2 2 2 9
5 2 1 2 1 9
6 2 1 2 2 9
7 2 2 1 1 13
8 2 2 1 2 13
72

_
a
T _ _ n Answer table
O b c d
Level 7 7 9 9
1
Level 11 11 9 9
2 Note
Diff. 4 4 0 0
1. Δ C = 0 so C in this experimental program has no effect.

2.
Δ (empty column) = 0, then this test is considered to indicate that the experimental
program was developed correctly

On the other hand, another experimental program was developed with the same
equipment according to the following arrangement:

Test No. TO b and c Data

1 1 1 1 1 5
2 1 1 1 2 6
3 1 2 2 1 8
4 1 2 2 2 9
5 2 1 2 1 8
6 2 1 2 2 9
7 2 2 1 1 13
8 2 2 1 2 14

_ _ _ _
TO b and c
Level 1 7 7 9.5 8.5
Level 2 11 11 8.5 9.5
Diff. 4 4 -1 1

At 1 To 2
B1 5.5 8.5
ITESRC _____________________________________________________________ 23
 QUALITY ENGINEERING

B2 8.5 13.5

Note that: 5.5 = 5 + 6 / 2

Data where levels A and B equal 1 have been averaged.

In this experiment:

Since ΔC = 0 C if it has effect

As Δ (empty columns) ∙≠ 0 this column does have an effect

Therefore the interaction A * B has been CONFUSED with the effect of C in the first
experimental program canceling each other, while in the second the effects have been
separated.

EXAMPLE OF ANALYSIS WITH INTERACTION

Assume that the quality objective for a brake adjuster is to minimize the deflection of
the adjuster when applying the brake.

Y = brake adjuster deflection, “LOWEST BEST” type characteristics.

The control factors are:

CONTROL FACTORS LEVELS

A: MATERIAL At 1 To 2
B: SELECTION AREA B1 B2
C: LENGTH C1 C2
D: HEAT TREATMENT D1 D2
E: TORQUE ADJUSTMENT E1 E2

Interactions. . . A*D and D*E

1. CONSTRUCT THE LINEAR GRAPH.

2. PLAN THE EXPERIMENTAL PROGRAM.
3. DETERMINE THE ANSWER TABLE.
4. DRAW INTERACTION DIAGRAMS.
5. PERFORM ANOVA
6. RECOMMENDATIONS.
7. DEFLECTION PREDICTION.
ITESRC _____________________________________________________________ 24
 QUALITY ENGINEERING

Solution
The degrees of freedom are: DOF(A) = 1
DOF(B) = 1
DOF(C) = 1
DOF(D) = 1
DOF(E) = 1
DOF ( A*D ) = 1
DOF ( A*E ) = 1
TOTAL DOF = 7

USING THE TRIANGULAR TABLE, THE LINEAR GRAPH IS AS FOLLOWS:

A (1) B(5) C(7)

A*D(3)

E(4)
D(2) OF 6)

The orthogonal arrangement and data are shown below:

Exp. 1 At 1 D2 A*D E4 B5
OF 6 C 7 DATA TOTAL
3

1 1 1 1 1 1 1 1 50 56
2 1 1 1 2 2 2 2 81 75
3 1 2 2 1 1 2 2 48 50
4 1 2 2 2 2 1 1 92 106
5 2 1 2 1 2 1 2 112 120
6 2 1 2 2 1 2 1 64 55
7 2 2 1 1 2 2 1 42 37
8 2 2 1 2 1 1 2 72 68

THE ANSWER TABLE IS:

TO b c d AND A*D OF
LEVEL 69.75 57.88 62.75 76.63 64.38 60.13 84.50
1
LEVEL 71.25 83.13 78.25 64.38 76.63 80.88 56.50
2
DIF. 1.50 25.25 15.50 -12.25 12.25 20.75 -28.00
RANGE 7 2 4 5 5 3 1

In DYE factors the range is the same because the absolute difference is the same.

ITESRC _____________________________________________________________ 25
 QUALITY ENGINEERING

In the response table the value of 69.75 was calculated as follows:

69.75 = 50 + 56 +81+ 75 +48 +50 +92 + 106 / 8 = Ā 1 = 106 + 156 + 98 + 198/ 8

The interaction diagrams are as follows:

At 1 To 2
D1 65.50 87.75
D2 74.00 54.75

D1

D2

For example:

____
A 1 D 1 = 50 + 56 + 81 +75 / 4 = 65.50

SIMILARLY:
___
A 1 D 2 = 74.00
___
A 2 D 1 = 87.75
___
A 2 D2 = 54.75

FOR D*E INTARACTION

E1 E2
D1 84.50 68.75
D2 44.25 84.50

ITESRC _____________________________________________________________ 26
 QUALITY ENGINEERING

D1

D2

____
D 1 E 1 = 50 + 56 + 112 + 120 / 4 = 84.50
____
D 1 E 2 = 68.75
___
D 2 E 1 = 44.25
___
D 2 E 2 = 84.50

THE RECOMMENDED CONCLUSION IN THIS EXAMPLE IS:

A2B1C1D2E1

Note that from the interaction tables we have:

A 2 D 2 = 54.75 IS THE LOWEST VALUE

E 1 D 2 = 44.25 IS THE LOWEST VALUE

The prediction of the average deflection with A 2 B 1 C 1 D 2 E 1 is :

Û = ( A 2 – T ) + ( B 1 – T) + ( C 1 – T) + ( D 2 – T) + ( E 1 – T) + T + A 2 D 2 – (A 2 – T ) – (D 2 –
T ) – T+ D 2 E 1 – ( D 2 – T ) – ( E 1 – T ) – T

Û = A 2 + B 1 + C 1 + D 2 + E 1 + A 2 D 2 - A 2 - D 2 + D 2 E 1 – 2T

Û = B 1 + C 1 + A 2 D 2 + D 2 E 1 - D 2 - 2T

ITESRC _____________________________________________________________ 27
 QUALITY ENGINEERING

Û = 57.88 + 62.75 + 54.75 +44.25

64.38 – 2 ( 70.25 )= 14.25

If we analyze experiment No. 7 of the orthogonal arrangement, we will notice that it

differs from the optimal combination only in the factor B (it is B 2 and not B 1 )

The prediction can be determined from this combination:

Exp. No. 7: A 2 B 1 C 1 D 2 E 1

42+37/2 = 25.25 = 14.25

Average in difference of

exp. No. 7 B 2 to B 1

Nomenclature of orthogonal arrays

L a (b c ) . . . . . . . . . . .. . ... .. . . . . L = Latin square (although it really isn't)

a = Number of rows or combinations of the array

b = No. levels of the factors

c = No. of columns or factors

Degrees of freedom = a – 1

Support material:
 Book: Six Sigma, methodology and techniques. Author: Escalante

 Quality Engineering support.htm

 ..\..\DisTaguchi unit 2.doc

 DOE TAGUCHI U2.ppt (support presentation for minitab use)

 Software: minitab

2.6. Reasons to use orthogonal arrays

ITESRC _____________________________________________________________ 28
 QUALITY ENGINEERING

An orthogonal arrangement can be used for simplicity and in some cases eliminates much of
the statistical design effort. It is a way to examine many factors simultaneously at low cost.
Dr. Taguchi recommends the use of orthogonal arrays to make matrices that contain the
controls and noise factors in the design of experiments.
Orthogonal arrays are tools that allow the engineer to evaluate how robust the process and
product designs are with respect to noise factors.

Dr. Taguchi's method for designing experiments uses techniques that involve low costs and
are applicable to the problems and requirements of modern industry. The purpose of product
design is to find that combination of factors that provides the most stable and reliable
performance at the lowest manufacturing price.

Taguchi orthogonal array analysis is used to produce the best parameters for optimal
process design, with the minimum number of experiments (tests). The results obtained for
the orthogonal arrangements are analyzed to obtain the following objectives:

a. Estimate the contribution of individual factors that influence quality at the product
design stage.
b. Gain the best condition for a process or a product, so that good quality characteristics
can be sustained.

2.7. Other arrangements for factors at two levels

ITESRC _____________________________________________________________ 29
 QUALITY ENGINEERING

TWO LEVEL SERIES

1.- L 4 (2 3 ) 4 rows
3 columns
It is the smallest arrangement (total factorial)

2.- L 8 (2 7 ) seven two-level factors

3.- L 12 (2 11 ) eleven two-level factors

There is no linear graph or triangular table. This is
Because the interaction effect between any
Part of columns is spread throughout the array and not in any column.
This is a good arrangement when little is known about interactions.

4.- L 12 (2 11 ) fifteen two-level factors

5.- L 32 ( 2 31 ) thirty-one two-level factor

6.- L 64 (2 63 ) sixty-three two-level factors

Below are some comments on other arrangements.

 The L12 arrangement is a special case. It can be seen in the appendix that no linear
graphs or interaction matrix are shown, this is because it is designed to analyze only up to
eleven individual factors without interactions. With this arrangement interactions cannot
be analyzed.

ITESRC _____________________________________________________________ 30
 QUALITY ENGINEERING

The interactions in an L12 array are distributed evenly across all columns. The advantage of
this is that it allows you to investigate 11 factors without worrying about their interactions. The
L12 array in general has good reproducibility of conclusions.

Something similar can be said about the L18 arrangement

 For an L16 arrangement there is a wide variety of possible linear graphs; the appendix
shows the six most used ones with three variants each.
 For an L32 arrangement, 13 different graphs are shown in the appendix, among the
several possible ones that exist.
 In any case, you can try to build more graphs according to your needs, always respecting
the interaction matrix.
 In the line graphs attached in the appendix, it is observed that the vertices are
represented with different symbols, specifically with o,  and  . The reason and its
meaning is as follows:

Taguchi suggests that the tests or runs be carried out in the order indicated by the rows of
the orthogonal arrangement, that is, first the conditions indicated by row 1, followed by those
in row 2 and so on.

Support material:
Book: Six Sigma, methodology and techniques. Author: Escalante

2.8. Line graphs

These types of linear graphs are used to facilitate the assignment of factors and interactions
in an orthogonal arrangement.

Each point represents a column to assign factors. The lines connecting the dots represent
the columns that should be assigned to the interactions between the connecting factors.

ORTHOGONALITY

An orthogonal design allows the levels of the factors under various conditions to be
compared with the same efficiency.

Degrees of freedom of a factor

ITESRC _____________________________________________________________ 31
 QUALITY ENGINEERING

Degrees of freedom are a measure of the amount of information that can be obtained. To
know the effect of a given factor in an experiment, the performance of the product is
compared by putting the factor at various levels when conducting an experiment; The more
levels a factor has, the more degrees of freedom it will have, and therefore the more
information to be obtained.

Interaction between factors.

An interaction exists when the effect of one factor depends on the level at which another
factor is found. The changes of a factor B will be graphed to determine if there is an
interaction.

These plots facilitate the assignment of factors and interactions to an orthogonal

arrangement.

Its interpretation is the following:

 The numbers represent the column number in the orthogonal array.

 The points represent the factors.

 The line segments joining two points represent the interaction between those
two factors.

There are graphic representations of triangular tables for each of the orthogonal
arrangements.

Line graphs consist of nodes and lines that connect them.

Each node represents a factor and the line that joins them to an interaction.

Line graphs become more useful as the complexity of the experiment increases and are
used in most advanced design techniques.

The linear graph for an interaction of the factors is shown in the following graph:

A node for factor A

AA*BB A node for factor B
A connection line for
The interaction.

Example:

Let A, B, C, and F be 2-level factors and A*B, A*D be interactions.

The degrees of freedom for this example are:
ITESRC _____________________________________________________________ 32
 QUALITY ENGINEERING

FACTOR GDL B(2)

TO 1 A*B (3) C(6) E(7)
b 1
c 1 A(1)
d 1 A*D(5)
AND 1 D(4)
A*B 1*1
A*B 1*1
7
Support material:
 Book: Six Sigma, methodology and techniques. Author: Escalante

2.9. Orthogonal arrangements for three-level factors

The series of three-level arrangements allows three factors to be investigated.

The L9 (34) provides information on four factors at three levels, using nine experimental
conditions. An L9 has eight degrees of freedom. Its use allows eight orthogonal comparisons
to be made.

The eight degrees of freedom can be decomposed into two degrees of freedom per column.

One column is required for each factor. The first step is to form cumulative categories from
the initial categories, so that cumulative category 1 is equal to initial category 1, cumulative
category 2 is equal to initial categories 1 plus 2.

EXAMPLE:
(i) = (1)
(ii) = (1) + (2)
(iii) = (1) + (2) + (3) etc.

The second step is to know the proportion that each accumulated category has.

EXAMPLE:
P1 = 25/90
P11 = 49/90
P111 = 65/90

ITESRC _____________________________________________________________ 33
 QUALITY ENGINEERING

PIV = 90/90

Each category is assigned a weight, according to the formula...

Wj = 1/(Pj*(1-Pj))

For each category the correlation factor is calculated, with the following formula:

CFj = (i¨2)/n

SUM OF SQUARE OF FACTORS

It is obtained by the sum of squares of each class, multiplied by its weight according to its
formula.

Ssa = (Ssa class 1)*W1+(Ssaclass 11)*W11 + (Ssa class 111)*W111........ + (Ssa class
n)*Wn
Sstotal = (total number of data) * (number of categories - 1).

DEGREES OF FREEDOM

The degrees of freedom are calculated based on the degrees of freedom of a factor for
variables, multiplied by the number of accumulated categories minus one.

EXAMPLE:

In a problem that has, four factors with three levels we obtain the following...

2*(4-a) = 6

We obtain that this example has 6 degrees of freedom.

The error
The error can be obtained by subtracting the total sum of squares from each factor.

Ss error = Ss total -Ssa -Ssb -Ssc -Ssd

With all this data, an Anova Table is constructed to take a

decision.

Conclusion

Analysis of variance offers more than a shortcut method for obtaining the error sum of
squares. It is reliable and highly accurate to obtain the correct option and reduce cost for
many experiments.

THREE LEVEL SERIES

1.- L 9 (3 4 ) four three-level factors

ITESRC _____________________________________________________________ 34
 QUALITY ENGINEERING

2.- L 27 (3 13 ) thirteen factors of three levels

3.- L 81 (3 40 ) forty three-level factors

NOTE

A three-level factor requires two degrees of freedom, so each column has two degrees
of freedom.

The interaction between two three-level factors requires 2*2 = 4 degrees of freedom = 2
columns.
The interactions in the triangular table have two columns assigned and both should be used
to study the interaction.

MIXED SERIES.

1.- L 18 (2 1 , 3 7 ) a factor of two levels and seven of three

Degrees of freedom:
2 level factor. . . . . . . . . . DOF = 1
7 factors of 3 levels. . . . . . . DOF = 14= 2*7
15 DOF

But in an L 18 arrangement there are 17 DOF

The remaining DOFs are used to study the interaction between the first two
columns of the array.
The factor of two novels and the first factor of three.
The remaining interactions are distributed throughout the array as in the L 12
array.

2.- L 36 (2 11 , 3 12 ) eleven two-level factors and twelve three-level factors.

3.- L 36 (2 3 , 3 13 ) three two-level factors of three includes interactions like L 8 .

4.- L 54 (2 1 , 3 26 ) a factor of two levels and twenty-six of three

Support material:
 Book: Six Sigma, methodology and techniques. Author: Escalante

ITESRC _____________________________________________________________ 35
 QUALITY ENGINEERING

2.10. Methods to modify orthogonal arrangements

Preset based on modifications to the linear graphs.

OBJECTIVE: That these graphs can be used for a particular problem. Standard linear
graphs can be modified to be used in a specific experiment, using triangular matrices, and
the orthogonality between the columns is maintained.

EXAMPLE: An L16 can be modified, remaining as follows:

FACTORS: A, B, C, D, E, F, G, H, I, J.

INTERACTIONS: AxB, CxD, ExF, FxG, AND GxH.

Solution:

1. The linear graph is drawn according to factors and interactions:

The standard linear graph is selected from which the linear graph of the previous
point can be obtained.

2. Reviewing the triangular matrix, it is determined how the linear graph can be
modified if the interaction between columns 6 and 11 is needed, then it is eliminated
and column 13 is left free, placing the interaction between the columns and it.

Orthogonal arrangement solution

The degrees of freedom are used to select a smaller orthogonal array that is appropriate.
The required degrees of freedom are first established based on the number of factors and
the number of levels of each of these, as well as the interactions that are desired to be
investigated. Example:

If you have 5 factors (A, B, C, D, E) with two levels each, and you also need to investigate
the AXB and AXC interactions, it is necessary to use an orthogonal arrangement L8, which
has 7 degrees of freedom.

Example:

ITESRC _____________________________________________________________ 36
 QUALITY ENGINEERING

Degrees of freedom of the factors (5X1) = 5

Degrees of Freedom of AXB (1X1) = 1
AXC degrees of freedom (1X1) = 1
7 gl

Assignment of factors to an orthogonal array

Procedures:

1. The appropriate orthogonal arrangement is selected; For this, the necessary degrees of
freedom due to the factors and interactions are taken into account.

2. The linear graph is drawn according to the desired interactions.

3. One of the standard line graphs of the chosen arrangement is selected, from which the
line graph from the previous step can be obtained.

4. The factors and interactions are assigned according to the columns that correspond to
them according to the resulting linear graph.

EXAMPLE: Newton's Law of Motion F = MA. Nature knows this relationship but we do not.
How can we get the relationship quickly without getting caught up in years of theoretical
research like Newton did? A DDE provided for two inputs, such as mass (M) and
acceleration (A), each at 2 levels, can be established as a process of 4 experimental trials (or
trials) in low and high positions as shown in the table .

REHEARSAL M TO
1 LOW LOW
2 LOW HIGH
3 HIGH LOW
4 HIGH HIGH

A set of experimental conditions is used to interrogate nature about its behavior under
certain circumstances. The specific set of conditions (or design matrix) is constructed to
analyze the response, with the purpose of constructing a model that approximates the model.
real finding in nature (ie, approximately equal to perfect knowledge by nature).
ITESRC _____________________________________________________________ 37
 QUALITY ENGINEERING

The example in the table, in DDE, is known as a full factorial design of 2 factors at 2 levels.
Various types of experimental designs are available to satisfy the specific objective you have,
Ie linear modeling, non-linear modeling, selection, etc.
Assume that our experiment generated an average set of results for force for each
experimental trial, as shown in the table.
The low and high set points for the table (5, 10) and acceleration (100, 200) are chosen
based on ranges of interest to the experimenter.

REHEARSA
TICKETS AVG DEPARTURES
L
M TO F
1 5 100 500
2 5 200 1000
3 10 100 1000
4 10 200 2000

To maximize the amount of knowledge acquired, the modeling analysis of a DDE is

conducted with standardized input values, where low values are coded as (1) and high
values are coded as (+ 1). These coded values standardize the scale and units of the input
variables. The result is a matrix with coded entries as shown in table 2.

REHEARSAL M TO M*A F
1 -1 -1 1 500
2 -1 1 -1 1000
3 1 -1 -1 1000
4 1 1 1 2000

Note, in the table, the addition of a column M*A which is generated by the product of
the comitted columns M and A and which was used to analyze the interactive or
combined effect of M and A. Columns M, A and M*A represent the three effects to be
evaluated, ie the linear effects of M and A and the effect of the interaction of M with A
(M*A). The analysis is conducted as shown in the table .

REHEARSAL M TO M*A F
1 -1 -1 1 500
2 -1 1 -1 1000
3 1 -1 -1 1000
4 1 1 1 2000
AVG(+1) 1500 1500 1250 1125
ITESRC _____________________________________________________________ 38
 QUALITY ENGINEERING

AVG(-1) 750 750 1000

AVG(+1)-AVG(-1) 750 750 250

To obtain the numbers in the shaded sections, you must do the following for each of the
columns M, A, and M*A: find the average of the output F when the column values are -1, find
the average of the output F when the column values are +1, and then find the difference of
the two averages (∆). For example, when the columns of effect M are A - 1 (Trials 1
and 2) the output is 500 for trial 1 and 1000 for trial 2, which averages 750.

The model generated by the DDE is built using a least squares regression design which can
be simplified for 2-level designs as shown below:

F is the average predicted force

Mc is the variable coded for mass
Ac is the variable coded for acceleration
FE is the grand experimental mean
∆M is the size of the linear effect for M, etc.

So our current model becomes:

F=1125+(750/2)*Mc+(750/2)*Ac+(250/2)*Mc*Ac

Remember that our prediction model is for hardcoded values (I. E -1, +1) of M and A, and
we can transform it into a model with current values of Ma and Aa using the following
relationship:

Ma=((Ma+Mb)/2)+((Ma+Mb)/2)*Mc

Where:
MA Current fixed value of mass
MC Fixed Mass Coded Value
MA Current experimental value set high
MB Current experimental value set to low
For our example:

Ma=7.5+2.5 Mc

So:
The calculations previously described are a simple way to go from encoded to actual values
and vice versa, as shown in the combined scale in the figure.

ITESRC _____________________________________________________________ 39
 QUALITY ENGINEERING

Support material:

Book: Six Sigma, methodology and techniques. Author: Escalante

Book: Probability and Statistics for engineers. Author: Walpole. Myers.

http://www.scielo.cl/scielo.php?pid=s0717-75182004000200006&script=sci_arttext

Quality Engineering support.htm

experimental design planning.pdf

DOE TAGUCHI U2.ppt (support presentation for minitab use)

http://www.slideshare.net/ldelatorre2010/ingenieria-de-calidad

unit 2 orthogonal arrangements.doc

Software: Minitab

Complementary exercises: Exercises u2.

Practice to be carried out: practice Quality Engineering Unit 2

ITESRC _____________________________________________________________ 40
QUALITY ENGINEERING

ITESRC _____________________________________________________________ 41