What is a Taguchi design (also called an orthogonal array)?

A Taguchi design is a designed experiment that lets you choose a product or process that
functions more consistently in the operating environment. Taguchi designs recognize that not
all factors that cause variability can be controlled. These uncontrollable factors are called noise
factors. Taguchi designs try to identify controllable factors (control factors) that minimize the
effect of the noise factors. During experimentation, you manipulate noise factors to force
variability to occur and then determine optimal control factor settings that make the process or
product robust, or resistant to variation from the noise factors. A process designed with this goal
will produce more consistent output. A product designed with this goal will deliver more
consistent performance regardless of the environment in which it is used.

A well-known example of Taguchi designs is from the Ina Tile Company of Japan in the 1950s.
The company was manufacturing too many tiles outside specified dimensions. A quality team
discovered that the temperature in the kiln used to bake the tiles varied, causing nonuniform tile
dimension. They could not eliminate the temperature variation because building a new kiln was
too costly. Thus, temperature was a noise factor. Using Taguchi designed experiments, the team
found that by increasing the clay's lime content, a control factor, the tiles became more
resistant, or robust, to the temperature variation in the kiln, letting them manufacture more
uniform tiles.

Taguchi designs use orthogonal arrays, which estimate the effects of factors on the response
mean and variation. An orthogonal array means the design is balanced so that factor levels are
weighted equally. Because of this, each factor can be assessed independently of all the other
factors, so the effect of one factor does not affect the estimation of a different factor. This can
reduce the time and cost associated with the experiment when fractionated designs are used.

Orthogonal array designs concentrate primarily on main effects. Some of the arrays offered in
Minitab's catalog let a few selected interactions to be studied.

You can also add a signal factor to the Taguchi design in order to create a dynamic response
experiment. A dynamic response experiment is used to improve the functional relationship
between a signal and an output response.

Session window output for a Taguchi design

Minitab calculates response tables, linear model results, and generates main effects and
interaction plots for:
  signal-to-noise ratios (S/N ratios, which provide a measure of robustness) vs. the control
factors

 means (static design) or slopes (Taguchi dynamic design) vs. the control factors

 standard deviations vs. the control factors

 natural log of the standard deviations vs. the control factors

Use the results and plots to determine what factors and interactions are important and assess
how they affect responses. To get a complete understanding of factor effects, you should usually
assess signal-to-noise ratios, means (static design), slopes (Taguchi dynamic design), and
standard deviations. Ensure that you choose a signal-to-noise ratio that is appropriate for the
type of data you have and your goal for optimizing the response.

NOTE
If you suspect curvature in your model, select a design - such as 3-level designs - that lets you
detect curvature in the response surface.
Return to top

A comparison of Taguchi static designs and Taguchi dynamic

designs

Minitab provides two types of Taguchi designs that let you choose a product or process that
functions more consistently in the operating environment. Both designs try to identify control
factors that minimize the effect of the noise factors on the product or service.

Static response

In a static response design, the quality characteristic of interest has a fixed level.

Dynamic response
In a dynamic response design, the quality characteristic operates along a range of values
and the goal is to improve the relationship between a signal factor and an output
response.

For example, the amount of deceleration is a measure of brake performance. The signal
factor is the degree of depression on the brake pedal. As the driver pushes down on the
brake pedal, deceleration increases. The degree of pedal depression has a significant
effect on deceleration. Because no optimal setting for pedal depression exists, it is not
logical to test it as a control factor. Instead, engineers want to design a brake system that
 produces the most efficient and least variable amount of deceleration through the range
of brake pedal depression.

Return to top

Conduct a Taguchi designed experiment

Conducting a Taguchi designed experiment can have the following steps:

1. Before you start using Minitab, you need to choose control factors for the inner
array and noise factors for the outer array. Control factors are factors you can
control to optimize the process. Noise factors are factors that can affect the
performance of a system but are not in control during the intended use of the
product.

Engineering knowledge should guide the selection of control factors and responses.
You should also scale control factors and responses so that interactions are unlikely.
When interactions between control factors are likely or not well understood, you
should choose a design that is capable of estimating those interactions. Minitab can
help you design a Taguchi experiment that does not confound interactions of
interest with each other or with main effects.

Noise factors for the outer array should also be carefully selected and might require
preliminary experimentation. The noise levels selected should represent the range
of conditions under which the response variable should remain robust.

Note

While you cannot control noise factors during the process or product use, you need
to be able to control noise factors for experimentation purposes.
2. Go to Stat > DOE > Taguchi > Create Taguchi Design to generate a Taguchi design
(orthogonal array). Each column in the orthogonal array represents a specific factor
with two or more levels. Each row represents a run; the cell values identify the factor
settings for the run. By default, Minitab's orthogonal array designs use the integers
1, 2, 3, to represent factor levels. If you enter factor levels, the integers 1, 2, 3, will
be the coded levels for the design.

You can also use Stat > DOE > Taguchi > Define Custom Taguchi Design to create a
design from data that you already have in the worksheet. Define Custom Taguchi
Design lets you specify which columns are your factors and signal factors. You can
then easily analyze the design and generate plots.
 3. After you create the design, you can use Stat > DOE > Modify Design to rename
the factors, change the factor levels, add a signal factor to a static design, ignore an
existing signal factor (treat the design as static), and add new levels to an existing
signal factor.

4. After you create the design, you can use Stat > DOE > Display Design to change
the units (coded or uncoded) in which Minitab expresses the factors in the
worksheet.

5. Conduct the experiment and collect the response data. The experiment is done by
running the complete set of noise factor settings at each combination of control
factor settings (at each run). The response data from each run of the noise factors in
the outer array are usually aligned in a row, beside the factor settings for that run of
the control factors in the inner array.

6. Use Stat > DOE > Taguchi > Analyze Taguchi Design to analyze the experimental
data.

Note

You should analyze each response variable separately with Taguchi designs.
Although Taguchi analysis accepts multiple response columns, these responses
should be the same variable measured under different noise factor conditions.

7. Use Stat > DOE > Taguchi > Predict Taguchi Results to predict signal to noise
ratios and response characteristics for selected new factor settings.

Return to top
Example of a Taguchi design

The following table displays the L8 (27) Taguchi design (orthogonal array). L8 means 8 runs.
27 means 7 factors with 2 levels each. If the full factorial design were used, it would have
27 = 128 runs. The L8 (27) array requires only 8 runs - a fraction of the full factorial design.
This array is orthogonal; factor levels are weighted equally across the entire design. The
table columns represent the control factors, the table rows represent the runs (combination
of factor levels), and each table cell represents the factor level for that run.

A B C D E F G

1 1 1 1 1 1 1 1

2 1 1 1 2 2 2 2
 A B C D E F G

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

L8 (27) Taguchi Design

In this example, levels 1 and 2 occur 4 times in each factor in the array. If you compare the
levels in factor A with the levels in factor B, you will see that B1 and B2 each occur 2 times in
conjunction with A1 and 2 times in conjunction with A2. Each pair of factors is balanced in this
approach, letting factors to be assessed independently.
Return to top

How does Minitab choose the default Taguchi design?

For 2-level designs based on L8 (3 or 4 factors), L16 (3-8 factors), and L32 (3-16 factors)
arrays, Minitab will choose a full factorial design if possible. If a full factorial design is not
possible, then Minitab will choose a Resolution IV design.

For all other designs, the default designs in Minitab are based on the catalog of designs by
Taguchi and Konishi.

Minitab takes a straightforward approach in determining the default columns that are used
in any of the various orthogonal designs. Say you are creating a Taguchi design with k
factors. Minitab takes the first k of columns of the orthogonal array.

Interactions and interaction tables in Taguchi

designs
Taguchi designs are primarily intended to study main effects of factors. Occasionally, you might
want to study some of the 2-way interactions. Some of the Taguchi designs (orthogonal arrays)
let you study a limited number of 2-way interactions. This usually requires that you leave some
columns out of the array by not assigning factors to them. Some of the array columns are
confounded with interactions between other array columns. Confounding means that the factor
effect is blended with the interaction effect, thus they cannot be assessed separately.

You can choose to have Minitab automatically assign factors to array columns in a way that
avoids confounding. Or, if you know exactly what design you want and know the columns of the
entire array that correspond to the design, you can assign factors to array columns yourself.
Assigning factors to columns of the array does not change how the design looks in the
worksheet. For example, if you assigned Factor A to Column 3 of the array and Factor B to
Column 2 of the array, Factor A would still show up in Column 1 in the worksheet and Factor B
would still show up in Column 2 in the worksheet.

What are interaction tables?

Interaction tables show confounded columns, which can help you assign factors to array
columns. The columns and rows represent the column numbers of the Taguchi design
(orthogonal array). Each table cell contains the interactions confounded for the two columns of
the orthogonal array.The following are interaction tables for each array.

For example, the entry in cell (1, 2) is 3. This means that the interaction between columns 1 and
2 is confounded with column 3. Thus, if you assigned factors A, B, and C to columns 1, 2, and 3,
you could not study the AB interaction independently of factor C. If you suspect that there is a
substantial interaction between A and B, you should not assign any factors to column 3.
Similarly, the column 1 and 3 interaction is confounded with column 2, and the column 2 and 3
interaction is confounded with column 1.

Interaction table for the L8(27) array

1 2 3 4 5 6 7

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

3 7 6 5 4
4 1 2 3

5 3 2
6 1
NOTE
The L12 (211) array does not have an interaction table. It is a specially designed array, in that
interactions are distributed more or less uniformly to all columns. There is also no linear graph
for this array. It should not be used to analyze interactions. The advantage of this design is its
capability to investigate 11 main effects, making it a highly recommended array.

Interaction table for the L16(215) array

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

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

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

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

11 7 6 5 4

12 1 2 3
13 3 2
14 1

What is the signal-to-noise ratio in a Taguchi

design?
In Taguchi designs, a measure of robustness used to identify control factors that reduce
variability in a product or process by minimizing the effects of uncontrollable factors (noise
factors). Control factors are those design and process parameters that can be controlled. Noise
factors cannot be controlled during production or product use, but can be controlled during
experimentation. In a Taguchi designed experiment, you manipulate noise factors to force
variability to occur and from the results, identify optimal control factor settings that make the
process or product robust, or resistant to variation from the noise factors. Higher values of the
signal-to-noise ratio (S/N) identify control factor settings that minimize the effects of the noise
factors.
Taguchi experiments often use a 2-step optimization process. In step 1 use the signal-to-noise
ratio to identify those control factors that reduce variability. In step 2, identify control factors
that move the mean to target and have a small or no effect on the signal-to-noise ratio.

The signal-to-noise ratio measures how the response varies relative to the nominal or target
value under different noise conditions. You can choose from different signal-to-noise ratios,
depending on the goal of your experiment. For static designs, Minitab offers four signal-to-
noise ratios:

Signal-
Goal of the Data
to-noise Signal-to-noise ratio formulas
experiment characteristics
ratio

Larger is Maximize the Positive S/N = -10 *log(Σ(1/Y2)/n)

better response

Nominal Target the Positive, zero, or S/N = -10 *log(σ2)

is best response and negative
you want to
base the signal-
to-noise ratio
on standard
deviations only

Nominal Target the Non-negative with

is best response and an "absolute zero" The adjusted formula is:
(default) you want to in which the
base the signal- standard deviation
to-noise ratio is zero when the
on means and mean is zero
standard
deviations
Signal-
Goal of the Data
to-noise Signal-to-noise ratio formulas
experiment characteristics
ratio

Smaller Minimize the Non-negative with S/N = -10 *log(Σ(Y2)/n))

is better response a target value of
zero

For Taguchi dynamic designs, Minitab provides one signal-to-noise ratio (and an adjusted
formula), which is closely related to the nominal-is-best S/N ratio for static designs.

NOTE
The Nominal is Best (default) signal-to-noise ratio is useful for analyzing or identifying scaling
factors, which are factors in which the mean and standard deviation vary proportionally. Scaling
factors can be used to adjust the mean on target without affecting signal-to-noise ratios.

How to calculate the signal to noise ratios in the

response table for Taguchi design
The Response Table contains a row for the average signal to noise ratio for each factor level,
Delta, and Rank. The table contains a column for each factor.

Delta
Delta is the difference between the maximum and minimum average signal to noise
ratios for the factor.

Rank
The Rank is the rank of each Delta, where Rank 1 is the largest Delta.

To get the standard deviation for each factor level, consider the following example. You
have a Taguchi design where the inner array has 2 factors (A and B), stored in C1 and C2,
respectively, and the outer array has two responses, stored in C3 and C4.

Part 1

1. Choose Calc > Row Statistics.

2. Choose Mean.
 3. In Input variables, enter C3 C4.

4. In Store result in, enter C6.

5. Click OK.

Part 2

1. Choose Calc > Row Statistics.

2. Choose Standard deviation.

3. In Input variables, enter C3 C4.

4. In Store result in, enter C7.

5. Click OK.

Part 3

1. Choose Calc > Calculator.

2. In Store result in variable, enter C8.

3. In Expression, enter: 10 * LOGT(C6**2 / C7**2).

4. Click OK.

NOTE
You can also get these signal to noise ratios by
choosing Stat > DOE > Taguchi > Analyze Taguchi Design, clicking Storage,
checking Signal to Noise ratios, and clicking OK twice.

Part 4

1. Choose Stat > Basic Statistics > Store Descriptive Statistics.

2. In Variables, enter C8.

3. In By variables, enter C1.

4. Click Statistics.

5. Check Mean.

6. Click OK twice.
The last column in the worksheet (named Mean1 if there was not already a column with this
name before performing these steps) contains the signal to noise ratios that are displayed
in the Response Table for factor A.

Repeat Part 4, entering C2 in step 3 to get the signal to noise ratios for factor B.