Robust Design Modules

Design Of Experiments Using
Taguchi Approach

Author
Ranjit K. Roy, Ph.D., P.E.
Why?
Solve Problems
Optimize products
Where?
Manufacturing
Development
Design

Cover Pages and Table of Contents Page ii

Table of Contents

Section Headings Page

Module-1: Overview and Approach (10 Pages)
1.1 What is The Taguchi Approach and Who is Taguchi? 1-1
1.2 New Philosophy and Attitude Toward Quality 1-2
1.3 The New Discipline 1-4
1.4 Simpler and Standardized Experiment Design Technique 1-6
1.5 Seminar Objectives 1-7
Review Questions 1-9

Module-2: Experiments Using Standard Orthogonal Arrays (30 pages)
2.1 Design of Experiment Basics 2-1
2.2 Experiments with 2-level Factors 2-4
2.3 Designing Experiments with Seven 2-level Factors 2-6
2.4 Full Factorial Design with Seven 2-level Factors 2-7
2.5 Tools for Experiment Designs 2-8
2.6 Application Steps 2-11
2.7 Procedure for Experiment Planning (Brainstorming) 2-13
2.8 Results with Multiple Criteria of Evaluations 2-14
2.9 Common Terms and their Definitions 2-15
2.10 Analysis of Results 2-17
2.11 Experiment Designs with Larger Number of Factors 2-21
2.12 Basic Experiment Design and Analysis Strategy 2-22
2.13 Accuracy of Orthogonal Array Experiments 2-23
2.14 Learning Check List 2-24

Module-3: Interaction Studies (19 Pages)
3.1 Experiments Designed to Study Interactions 3-1
3.2 Guidelines for Interaction Studies 3-5
3.3 Steps in Interaction Analysis 3-6

Module-4: Experiment design with Mixed Level Factors (20 Pages)
4.1 Modification of Standard Orthogonal Arrays 4-1
4.2 Upgrading Column Levels 4-2
4.3 Downgrading Columns 4-6
4.4 Incompatible Factor Levels 4-10
4.5 Combination Design (Special Technique) 4-11
Course RECAP 4-13

Cover Pages and Table of Contents Page iii

Table of Contents(Continued)

Module-5: Robust Design Strategy (18 Pages)
5.1 Ambitious Business Goals 5-1
5.2 Mechanics of Outer Array Designs 5-3
5.3 Benefits of Outer Array Design 5-5
5.4 Analysis of Repeated Results 5-6
5.5 Definition of MSD for the three QC's 5-8
5.6 Experiment Design and Analysis Strategies 5-12
5.7 Experiment Design Tips 5-13

Module 6: Standard and S/N Analyses with ANOVA (34 Pages)
6.1 Analysis Guidelines 6-1
6.2 Test for Presence of Interaction 6-5
6.3 ANOVA 6-6
6.4 Pooling Technique 6-13
6.5 Confidence Interval of Main Effects 6-15
6.6 Discussion on Some Key Issues 6-17
6.7 Example 8B: Standard Analysis with Multiple Runs 6-19
6.8 Example 9A: S/N Analysis - Nominal is the Best 6-22
6.9 Example 9B: S/N Analysis - Smaller is Better 6-24
6.10 Example 9C: S/N Analysis - Bigger is better 6-26
6.11 Understanding and Interpreting Error Terms 6-29

Module-7: Loss Function (11 Pages)
7.1 Taguchi Loss Function 7-1
7.2 Manufacturer and Supplier Tolerances 7-5

Module-8: Application Steps (12 Pages)
8.1 Description of Application Steps 8-2
8.2 Brainstorming for Design of Experiments(DOE) 8-3
8.3 Overall Evaluation Criteria 8-5
8.4 Taguchi Vs. Classical DOE 8-8

Module-9: Robust Design for Dynamic Systems (Optional, 25 Pages) 9-1

Reference Materials (Appendix, 26 Pages)
Cover Pages and Table of Contents Page iv

Quick References

TOPICS PAGE

o Module 1: Review Questions 1-15
Module 2: Review Questions 2-23

o OA and Triangular Table 2-8
o F-Table for 95% Confidence 6-14
o Examples (Design and Analysis) 2-11, 4-2, 5-8, 5-14, 5-16, 6-2
6-22, 6-25, 6-28, 6-30
o Example (Complete with QC) 8-14
o F-Tables Section C A-2
o Common Orthogonal Arrays A-5
o Linear Graphs A-16
o Triangular Tables A-13
o Glossary of Terms A-17
o References A-19
o Application Steps Guide A-28
o Seminar Evaluation A-33

General Reference
Taguchi, Genichi: System of Experimental Design, UNIPUB Kraus Intl. Publications, White Plains,
New York, 1987
Roy, Ranjit: A Primer on the Taguchi Method, The Society of Manufacturing Engineers, One SME Drive, Dearborn,
Michigan, USA.

INTERNET: For general subject references (Taguchi + Seminar + Software + Consulting + Case Studies +
Application Tips), try search engines like Yahoo, Lycos, Webcrawler, etc. For Nutek products, services, and
application examples, visit:

http://www.rkroy.com http://www.wwnet.com/~rkroy
http://www.wwnet.com/~rkroy/wp-inb.html http://www.wwnet.com/~rkroy/wp-inc.html
http://www.wwnet.com/~rkroy/wp-ind.html http://www.wwnet.com/~rkroy/wp-ine.html

Module - 1

Overview and Approach
There are a number of statistical techniques available for engineering and scientific studies.
Taguchi has prescribed a standardized way to utilize the Design of Experiments (DOE)
technique to enhance the quality of products and processes. In this regard it is important to
understand his definition of quality, the method by which quality can be measured, and the
necessary discipline for most application benefits. This module presents an overview of
Taguchis quality improvement methodologies.

1.1 What is The Taguchi Approach and Who is Taguchi?
Design of Experiments (DOE) using Taguchi Approach is a standardized form of
experimental design technique (referred as classical DOE) introduced by R. A. Fisher in
England in the early 1920s. As a researcher in Japanese Electronic Control Laboratory, in the
late 1940s, Dr. Genichi Taguchi devoted much of his quality improvement effort on
simplifying and standardizing the application of the DOE technique. Although Taguchi
successfully applied the technique in many companies throughout the world, it was introduced
to USA and other western countries only in the early 1980s.

How can we improve quality in all phases of design and manufacturing?

What can we do to affect quality in Design and Production?

Design of Experiment using the Taguchi Approach is just one of the many tools.

Why you should apply?
- Improve consistency of performance and save cost
- Build insensitivity (Robustness) towards the uncontrollable factors
- Solve production and manufacturing problems

Where you should apply?
- Optimize Designs using analytical simulation studies
- Select better alternative in Development and Testing
- Optimize manufacturing Process Designs
- Determine the best Assembly Method
- Solve manufacturing and production Problems
What's new in the Taguchi approach?
- New Philosophy
* Timing for quality activity. Building quality into design
* Estimating the cost of lack of quality
Module 1: Overview and Approach Page 1 - 2
* General definition of quality

- New Discipline
* How to work as a team and decide things together by consensus
* How to be proactive and objectively plan experiments

- Simpler and Standardized experimental technique

1.2 New Philosophy and Attitude Toward Quality
Traditionally, quality activities took place only at the production end. Dr. Genichi Taguchi
proposed that a better way to assure quality is to build it in the product by designing quality
into the product. In general, he emphasized that the return on investment is much more when
quality was addressed in engineering stages before production. A few popular techniques
used in different phases of engineering activities are listed below:

Design : Design Reviews, DOE, Taguchi methods, etc.
Development: DOE, Taguchi Methods, Growth Modeling, etc.
Test & Validation: Reliability, Taguchi Methods
Production: SPC, DOE, Taguchi Methods, Loss Function

How do you improve quality in Design, Development, and Validation and Production?

The Taguchi approach presents an attractive option in all activities mentioned above.

The Biggest Bang For Your Buck

A B C D
Design Development Test & Validation Production

Return on investment is much higher when applied in concept design. The return is
immediately realized in problem solving applications.

Realistic Expectation Leads to Satisfactory Results
- Most applications happens to be in the manufacturing and problem solving
- Applications in design are slow but yield better returns
- No matter what the activities, DOE generally is effective

How do we measure the cost of quality?
* By measuring quality not by Rejection at production, but in
terms of deviation from the target.
* By using the Loss function formulation to quantify the cost of quality.

New Defination Of Quality : Consistency of performance

How do we improve consistency of performance?
* Minimize variation around the target by reducing the distance of the mean to the
target(m) and by reducing the standard deviation ( ).

$Loss
Target y
L=K(y -Y
o
)
Y
o
2

Target
Mean
What you are now
What you want
to become
Target
m

To improve quality(Q)
Q = + m
1.3 The New Discipline
The Taguchi method is most effective when experiments are planned as a team and all
decisions are made by consensus. The Taguchi approach demands a new way of
working together as a group while attempting to apply the technique in the industrial
applications. The major difference can be understood by comparing the new method
with the old approach.

Typical Old Approach
(Series Process)

Some thinking
Some more thinking
Try this
Try that

There are five basic steps in application of Taguchi experimental design technique to
a project. The brainstorming is a necessary first step in the application process.

Do Lots Of Thinking
(Brainstorming)
Trial#1
Trial#2
Analyze Results
I. Planning
II. Designing
III. Doing
IV. Analyzing
V. Confirming
Application Steps
Run confirmation
Experiments
Conduct Experiments
- What are we after?
- How do we measure results?
- etc.

New Approach
(A Parallel Process)

1.4 Simpler and Standardized Experiment Design Technique

* Simpler (canned) experiment design methods
* Robust design technique
* Standardized analysis procedure
* Clear guidelines for interpretation of results
* Special data transformation (S/N ratios) for achieving reduced variation.

Key Points

Do it up front. Apply quality improvement tools as far up in the design as possible.
Measure quality in terms of variation around the target. Quantify ill effects of
poor quality to the society.
Incorporate the new discipline of working together in project teams and determine all
project related matters by the group consensus.
Use Taguchi's Off-line quality engineering concepts in three phases of
engineering and production (Off-Line Quality Control)

- System Design (basic research)
- Parameter Design (common for industrial applications)
- Tolerance Design (usually follows parameter design)

Parameter Design is a special form of experimental design technique which was
introduced by R. A. Fisher in England in the early 1920's. Parameter design as proposed
by Dr. Genichi Taguchi is the subject of this seminar.


1.5 Seminar Objectives

What Will The Course Cover?

What Will You Learn?

How To Design Experiments Using Taguchi Approach.
- Use Standard Orthogonal Array (OA) For Simple Design
- Handle Interaction
- Handle Mixed Levels
- Includes Noise Factors/Outer Array (Robust Design)

Steps In Analysis Of Main Effects And Determintion of Optimum Condition.
- Main effect studies
- Interaction analysis
- Analysis of Variance (ANOVA)
- Signal to Noise ratio (S/N)
- Dynamic Characteristics

Learn to Quantify Improvements Expected From Improved Designs in Terms of
Dollars. Apply Taguchi's loss function to compute
$ LOSS.

Learn to Brainstorm for Taguchi Experiments.

Determine evaluation criteria, factors, levels, interactions, noise factors, etc. by group
consensus.

What This Seminar Will Not Do

This seminar is not intended to teach Statistical Science or attempt to cover general
philosophy of quality improvement.

1.6 Basic Taguchi Experiment Design and Analysis Strategies

* Experiment designs using Std. Orthogonal Arrays
* Main effect studies and optimum condition
* Experiment designs to include interactions
* Noise factors and ROBUST design
* Analysis Of Variance (ANOVA)
* Overall Evaluation
Criteria (OEC)
* S/N ratio analysis
* Problem solving
Loss
Function
* Mixed level factor designs
Dynamic
Characteristics.

Module - 2

Experiment Designs Using Standard Orthogonal Arrays
Modern Industrial environments pose experiments of numerous kinds. Some have few factors,
some have many, while there are others that demand factors to have mixed levels. A vast
majority of the experiments, however, fall in the category where all factors possess the same
number of levels. In Taguchi approach a fixed number of orthogonal arrays are utilized to
handle many common experimental situations.

Topics Covered:
Basic Experiment Design Techniques.
Experiments with standard orthogonal arrays.
Standard analysis of experimental results.

2.1 Design of Experiments Basics

DOE is an experimental strategy in which effects of multiple factors are studied
simultaneously by running tests at various levels of the factors. What levels should we
take, how to combine them, and how many experiments should we run, are subjects of
discussions in DOE.

Factors are variables (also think of as ingredients or parameters) that have direct
influence of the performance of the product or process under investigation. Factors are of
two types:

Discrete - assumes known values or status for the level.
Example: Container, Vendor, Type of materials, etc.

Continuous - can assume any workable value for the factor levels.
Example: Temperature, Pressure, Thickness, etc.

Levels are the values or descriptions that define the condition of the factor held while
performing the experiments.

Examples: Type of Container, Supplier, Material, etc. for discrete factor
200 Deg., 15 Seconds, etc. when the factors are of continuous type.

To study influence of a factor, we must run experiments with two or more levels of the
factors. Two is minimum number of levels required to make comparison of the
performance and thereby determine the influence. Why not test at more levels? When
Module 2: Experiment Designs Using Standard Orthogonal Arrays Page 2 - 2
should you consider testing at more than two levels? Results of tests with two levels
produce only two data points. Two data points when joined together represent influence
that behave in a straight line, whether the actual behavior is linear or not. So what if the
actual behavior is non-linear? We can only detect that in the results when there are more
than two data points from tests that have more than two levels. Thus, if non-linear
behavior is suspected, we should consider testing at more than two levels of the factor.

Response with Two Levels
*
*
A A
1 2
*
*
A A
1
Response with Three Levels
A
2
3
*
Factor A Factor A

While studying the influence of a factor, if we decide to test it at two levels, only
two tests are required, where as, if three levels are included, then three tests will have
to be performed.

EXAMPLE: Baking Processes at two and three Temperature Settings..

What about influences of other factors? What if we want to study a number of factors
together. How many tests do we need to run?

Consider two factors, A and B, at two levels each. They can be tested at 4
combinations.

4 Experiments: A
1
B
1
A
1
B
2
A
2
B
1
A
2
B
2

A
1
A
2
A => A
1
A
2

B
1
* * B => B
1
B
2

B
2
* *

Likewise three factors A, B & C tested at 2-levels each
Requires 8 experiments

Factors: A: A
1
, A
2
B: B
1
, B
2
C: C
1
, C
2

8 Experiments: A
1
B
1
C
1
A
1
B
1
C
2
A
1
B
2
C
1
A
1
B
2
C
2

A
2
B
1
C
1
A
2
B
1
C
2
A
2
B
2
C
1
A
2
B
2
C
2

Which can be written in notation form as: (use 1 for level 1, etc.)

The total number of possible combinations (known as the full factorial) from a given
number of factors all at 2-level can be calculated using the following formulas.

Of course the full factorial experiments are always too many to do.

How to do the least number of experiments to get the most information? How do
you select which ones to do?

To design common industrial experiments, Taguchi constructed a set of special
orthogonal arrays.

An L-4 array is used to design an experiment to study three 2-level factors

The word "DESIGN" means determining the number of experiments to be performed
and the manner in which they should be carried out, i.e., number and the factor level
combinations. Taguchi has constructed a number of orthogonal arrays to accomplish
experiments design. Each array can be used to suit a number of experimental
situations. The smallest among the orthogonal array is an L-4 constructed to
accommodate three two level factors.

A B C
1 1 1
1 1 2
1 2 1
etc.
Experimental Condition Full Factorial

3 Factors at 2 level, 2
3
= 8
4
= 16
7
= 128
15
= 32,768

2.2 Experiments with 2-Level Factors

Consider that there are three factors A, B and C each at two levels. An experiment to
study these factors will be accomplished by using an L-4 array as shown below. L-4
is smallest of many arrays developed by Taguchi to design experiments of various
sizes.

Key observations: First row has all 1's. There is no row that has all 2's. All columns
are balanced and maintain an order.

The columns of the array are ORTHOGONAL or balanced. This means that there are
equal number of levels in a column. The columns are also balanced between any two.
This means that the level combinations exist in equal numbers.

Within column 1, there are two 1's and two 2's. Between column 1 and 2, there is one
each of 1 1, 1 2, 2 1 and 2 2 combinations.

Factors A, B And C All at 2-level produces 8 possible combinations (full factorial)

Taguchis Orthogonal array selects 4 out of the 8

EXPT #1 A
1
B
1
C
1

EXPT #2 A
1
B
2
C
2

EXPT #3 A
2
B
1
C
2

EXPT #4 A
2
B
2
C
1

Expected performance from experiments designed using orthogonal arrays

Orthogonal arrays are used to design experiments and describe trial conditions. Experiments design using
orthogonal arrays yield results that are more reproducible. .

An experiment designed to study three 2-level factors requires an L-4 array which prescribes
4 trial conditions. The number of experiments for seven 2-level factors which require an L-8
array is eight.
L-4 (2
3
) Orthogonal Array

Trial # A B C
1 1 1 1
2 1 2 2
3 2 1 2
4 2 2 1

Array Descriptions:
1. Numbers in array represent the
levels of the factors
2. Rows represents trial conditions
3. Columns indicate factors that can
be accommodated
3. Columns of an OA are
orthogonal
4. Each array can be used for many
experimental situations

One factor at a time experiments with three 2-level factors require the same number of
experiments as described by the scheme below (One Factor at a Time).

Factor effects can be easily obtained by subtracting the result of Test #1 from another as
shown below.

Effect of A = Y
2
-Y
1,
Effect of B = Y
3
-Y
1
Effect of C = Y
4
-Y
1

Characteristics of orthogonal array designs are:
a. Levels appear in equal numbers.
b. Combination of A
1
B
1
, A
1
B
2
, etc. appear in equal numbers.
c. Effect of factor A can be separated from the effects of B and C.

In another arrangement of one factor at a time experiment, seven 2-level factors can be studied
by 8 experiments obtained from the following arrangement.

Experiment design with seven 2-level factors using an orthogonal array (L-8) which requires
the same number of experiments. L-8 OA design is recommended in spite of the same size of
experiments for a number of reasons.

- Use average effect for basis of conclusion.
- Higher reproducibility.
- Optimum based on robust design.

Expt#A B C D E F G
1 1 1 1 1 1 1 1 Eight experiments are
2 2 1 1 1 1 1 1 required to study seven
3 2 2 1 1 1 1 1 2-level factors by changing
4 2 2 2 1 1 1 1 one factor at a time.
5 2 2 2 2 1 1 1
6 2 2 2 2 2 1 1
7 2 2 2 2 2 2 1
8 2 2 2 2 2 2 2

Four experiments are required to
study three 2-level factors by
changing one at a time.

Test# A B C RESULT
1. 1 1 1 Y
1

2. 2 1 1 Y
2

3. 1 2 1 Y
3

4 1 1 2 Y
4

2.3 Designing Experiments With Seven 2-Level Factor

Experiments with seven 2-level factors are designed using L-8 arrays. An L-8 array has seven
2-level columns. The factors A, B, C, D, ... G can be assigned arbitrarily to the seven columns
as shown.

Notes:____________________________________________________________

Historical Development of OA
Orthogonal arrays were first conceived by Euler (Euler's Greco-Latin Squares). OA's are used for expressing
functions and assigning experiments. (Reference: System of Experimental Design by G. Taguchi, pp. 165, also
1021 - 1026 ).

Latin Squares of dimension n x n are denoted by L
1
, L
2
. . . L
n-1
. 3 x 3

L
1
L
2
_
1 2 3 1 3 2
2 3 1 2 1 3
3 1 2 3 2 1

Orthogonal arrays used by Taguchi are constructed by combining the Latin Squares

The development of orthogonal arrays dates back to times before Taguchi. Theories and procedures of
orthogonal arrays are beyond the scope of this seminar.

L
8
Orthogonal Array

COL>> 1 2 3 4 5 6 7
TRIAL# A B C D E F G
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

2.4 Full Factorial Design With Seven 2-Level Factors

If Full Factorial experiments with seven 2-level factors are desired, it can be laid out using
the following scheme. With factors arranged as shown, 16x8 = 128 squares represent the
description of the 128 trial conditions representing the full factorial experiments. The spaces
marked as Tr#1, Tr#2, etc. indicate the 8 trial conditions of the experiment designed using the
Taguchi orthogonal array. Note that the best condition is one among the 120 conditions which
may or may not be one of the 8 conditions described by the orthogonal array design.

Full Factorial A1 A2
Experiments B1 B2 B1 B2
C1 C2 C1 C2 C1 C2 C1 C2
G1 Tr#1

E1 F1 G2

G1

F2 G2 Tr#3

D1 G1

E2 F1 G2
Tr#5

G1 ??

Tr#7
F2 G2

G1

E1 F1 G2

Tr#8
G1
Tr#6

F2 G2

D2 G1 Tr#4

E2 F1 G2

G1

F2 G2 Tr#2

Tr#1, Tr#2, etc. are location/description of the trial conditions defined by the orthogonal array.
?? is the location/combination of the optimum condition to be determined.

2.5 Tools For Experiment Designs

Triangular Table for 2-level Orthogonal 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
(15)
ETC...

L-4 (2
3
) Orthogonal Array
Trial #\ A B C
1 1 1 1
2 1 2 2
3 2 1 2
4 2 2 1
L
8
(2
7
) Array
COL.>>
TRIAL# 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


L
12

Column =>
Cond. 1 2 3 4 5 6 7 8 9 10 11
1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 2 2 2 2 2 2
3 1 1 2 2 2 1 1 1 2 2 2
4 1 2 1 2 2 1 2 2 1 1 2
5 1 2 2 1 2 2 1 2 1 2 1
6 1 2 2 2 1 2 2 1 2 1 1
7 2 1 2 2 1 1 2 2 1 2 1
8 2 1 2 1 2 2 2 1 1 1 2
9 2 1 1 2 2 2 1 2 2 1 1
10 2 2 2 1 1 1 1 2 2 1 2
11 2 2 1 2 1 2 1 1 1 2 2
12 2 2 1 1 2 1 2 1 2 2 1

Note:
The L-12 is a special array designed to investigate main effects of 11 2-level factors.
This array is not recommended for analyzing interactions

L
16

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

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


Standard Notations for Orthogonal Arrays

Symbol L
n
(l
m
)

Where n = Number of experiments
L = Number of levels
m = Number of factors

L
8
(2
7
), 8 = Number of experiments
2 = Number of levels
7 = Number of factors

Common Orthogonal Arrays

( ) ( ) ( ) ( ) ( ) L L L L L
4
3
8
7
12
11
16
15
32
31
2 2 2 2 2 , , , , ,
==> (2-level arrays)

L L L
9
4
18
1 7
27
13
3 2 3 3 , , , , ==> (3-level arrays)

L L
16
5
32
1 9
4 2 4 , , , ==> (4-level arrays)

Examples and use of L
54
(2
1
, 3
25
), L
81
(3
40
), etc.
are beyond the scope of this seminar)

Note: Arrays L
18
(2
1
3
7
), L
32
(2
1
4
9
), and L
54
(2
1
3
25
) are for mixed level factors.
2.6 Application Steps
(Brainstorm)
- What are evaluation criteria
- How do we measure them
- How do we combine the criteria
- What are the factors
- What are the levels
etc.
Design
Perform
Analyze
Confirm

Example 1: Plastic Molding Process

A: Injection Pressure A
1
= 250 psi A
2
= 350 psi
B: Mold Temperature B
1
= 150 deg. B
2
= 200 deg.
C: Set Time C
1
= 6 sec. C
2
= 9 sec.

Experiment planning(Brainstorming): It is the first step in the application process.
Before an experiment can be planned and carried out, there are questions related to the
project of the nature shown below must be resolved (Evaluation Criteria (Table of overall
evaluation criteria), Factors & Levels, Interactions, Noise factors, etc. )

- How would we evaluate results after the experiments are carried out?
- How do you decide what are the factors?
- How do you determine which factors to include in the study?
- What are the levels of the factors and how are they established?
- How do you plan to measure the performances?
- What are the criteria of evaluation?
- How do you plan to combine them if there are multiple criteria?

Answers to these and many other questions regarding the experiments are resolved
in the experiment planning (Brainstorming) session.

NOTE: Any time the project team involves more than two people, formal brain -storming
should be carried out. The following are some of the desirable characteristics of an
effective brainstorming session.
- all who have first hand knowledge of the project are invited
- it is facilitated by a person who is not involved in the project
- enough time allowed for discussions of related items
- all in attendance are allotted one vote (one person one vote) regardless
of their responsibilities or position
- all decisions in the sessions are made by consensus
- all in attendance understand and participate in all discussions
etc.

2.7 Procedure for Experiment Planning (Brainstorming)

Experimental designs produce most benefits when they are planned properly. For proper planning it is necessary that a
special session is dedicated to discuss various aspects of the project with the project team and that all decisions are made
by the group consensus. The planning session should be arranged by the leader, and when possible, have someone who is
not involved in the project facilitate the session.

Topics of Discussions
a) Project Objectives ( 2 - 4 hours)
- What are we after? How many objectives do we wish to satisfy?
- How do we measure the objectives?
- What are the criteria of evaluation and their quality characteristic?
- When there are more than one criterion, would we have a need to combine them?
- How are the different evaluation criteria weighted?
- What is the quality characteristic for the Overall Evaluation Criteria(OEC)?

b) Factors (same as Variables, Parameters, or Input, 1 - 2 hours)
- What are all the possible factors?
- Which ones are more important than others (pareto diagram)?
- How many factors can we include in the study?

c) Levels of the Factors ( 1/2 hours)
- How are the levels for the factors selected? How many levels?
- What is the trade off between levels and factors?

d) Interactions (between two 2-level factors, 1/2 hours)
- Which are the factors most likely to interact?
- How many interactions can be included?
- Should we include an interaction or an additional factor?
- Can we afford to study the interactions?

e) Noise Factors and Robust Design Strategy ( 1/2 - 1 hour)
- What factors are likely to influence the objective function, but are not controllable?
- How can the product under study be made insensitive to the noise factors?
- What are the uncontrollable or Noise factors
- Is it possible to conduct experiments by exposing them to the simulated Noise
conditions?
f) Experiment and Analysis Tasks Distribution (1/2 hours)
- What steps are to be followed in combining all the quality criteria into an OEC?
- What to do with the factors not included in the study?
- How to simulate the experiments to represent the customer/field applications?
- How many repetitions and in what order will the experiments be run?
. Who will do what and when? Who will analyze the data?


2.8 Results with Multiple Criteria of Evaluation

Evaluation Criteria Table
Criteria description Worst
reading
Best
reading
Quality
Characteristics
Relative
Weighting
Sample 1
Readings
Tensile strength 12000 15000 Bigger 55 % 12652
Rupture Strain 0.10 0.30 Nominal 30 % 0.207
Brinnel Hardness 60 45 Smaller 15 % 58
O E C = > 30.0
Before evaluations from different criteria can be combined, the following three conditions must be met.
- Units of measurements must be the same (This is generally done by expressing the evaluation in term of a
fraction/ratio of the highest magnitude of evaluations)
- Quality characteristics(QC) for all must be either Bigger or Smaller (For Nominal characteristic, Deviation
calculated subtracting the Target value from the evaluation
is used. Deviation always bears the Smaller QC which can then be converted to Bigger QC by subtracting the
ratio from 1.0)
- Each criteria must be included with appropriate weighting (This can be reflected by multiplying the
contribution each evaluation makes by the Relative weighting )

Overall Evaluation Criteria(OEC)
For Sample 1.
(Readings 12652, .207, and 58 as shown in the table above)

|12652-12000| |.207-.3| |58-45|
(OEC)
1
= ( ----------------- ) x 55 + ( 1 - ---------- ) x 30 + (1 - ---------- ) x 15
|15000-12000| |.30-.10| |60-45|

= 11.95 + 16.05 + 2.0 = 30.0

Likewise OEC for the other three samples for the L-4 experiments (used later in the example)
are computed as: (OEC)
2
= 25, (OEC)
3
= 34
(OEC)
4
= 27 (Which are results used later in the example)

Note: There can be only one OEC (or RESULT) for a single sample and that the OEC calculated for each sample
becomes the sample RESULT, which are then be used to carry out the analysis and to determine the optimum
(Qualitek-4, OEC Option)

The experiment design is accomplished by using an L
4
array.

The experiment requires 4 separate trials as follows.

Trial #1: Injection Pressure = 250 psi
Mold Temperature = 150 deg.
Set Time = 6 sec.

Trial. #2: Injection Pressure = 250 psi
Mold Temperature = 200 deg.
Set Time = 9 sec. etc.

2.9 Common Terms and their Definitions

Factors: Any item that has direct influence on the performance. It is synonymous with PARAMETER,
VARIABLE, INPUT, INGREDIENTS, etc.
Level: Level indicates the value of the factor used while conducting the experiment.
Experiment: Experiment refers to the total tests described by the orthogonal array. There may be one or more
experiments in a TAGUCHI CASE STUDY. An experiment includes a number of TRIAL CONDITIONS as
dictated by the array used to design the experiment.
Trial Condition: Trial conditions are the individual experiments within an experiment.
Data Collection Table: This is a necessary table for multiple criteria and contains sample observations which
are used to combine into sample OEC/result.

Factors A B C
1 1 1 1
2 1 2 2
3 2 1 2
4 2 2 1

RESULTS: In most text books on design of experiments, the performance of the product or process under study,
is termed as the response. Since the performance may be evaluated by multiple evaluation criteria, the single
quantity which represent the combined effects of all evaluations(OEC) represents the performance of the test
sample and is called the result..
Repetition and Replication refer to the manner in which the experiments are carried out.
REPETITION: In this method the trial condition is selected randomly, then all samples in the trial are carried
out in sequence.
REPLICATIONS: This is the most random way of carrying out the trial conditions. The order of running the
test is selected by randomly selecting the sample to be tested from among the total samples.

Process Diagram - Function of a product or a process can be viewed in terms of a SYSTEM which require an
INPUT to produce OUTPUT making use of many FACTORS. Such schematics representing the functions of a
system is popularly known as the Process Diagram.

- When the input does not change during the investigation, it is called a STATIC SYSTEM

- When the input is variable, it is called a DYNAMIC SYSTEM

The function of the system under study (product or process) can be viewed in terms of its Process Diagram which
reflects OUTPUT as a result of INPUT and other INFLUENCING FACTORS to the system.

Process Diagram
System/Process
(Pound Cake Baking Process)

Mixing , Kneading, and
allowing time for baking.

Result ,
Response, Quality
Characteristic, or
Overall Evaluation
Criteria(OEC)

- Evaluation
(Readings)

- Observations
Input to the
Process

* Heat/Electricity
Noise Factors

* Oven Type
* Kitchen Temp
* Humidity
Control Factors
and Levels

* Sugar
* Butter
* Flour
* Milk
Input
Output
What to do after Experiments are Designed?

* Decide how to carry out the experiments
-Single or Multiple Samples
- If multiple runs: Repeat or Replicate.
* Collect and Transformation data to OEC or Results

Results: If performance is measured in terms of multiple criteria of evaluation, they must be
combined into an overall evaluation criteria (OEC) before it can be used as a sample result.

2.10 Analysis of Results

Calculate factor averages and determine
* Optimum Condition
* Nature of Influence of Factors
* Expected Result At Optimum Condition

For additional information about the relative influences of the factors and their significance,
calculate ANOVA statistics and complete ANOVA table.

In this example, upon completion of the trials, the performances are evaluated using
the three evaluation criteria as discussed earlier. The evaluations for each sample
are then combined into OEC's which are taken as the sample results as shown below.

Calculation of Factor Average Effects

Trial/Col. A B C Results
1 1 1 1 30 (Y
1
) <== Evaluations which make
1 1 2 2 25 (Y
2
) 30 are shown earlier.
1 2 1 2 34 (Y
3
)
1 2 2 1 27 (Y
4
)

_
Grand avg. of performances, T = ( 30 + 25 + 34 + 27 ) / 4 = 29

Sum of all results containing the effects of the factor
Average effect of a factor = ---------------------------------------------------------
Number of results included in the sum


Plot of Factor Average Effects (Main Effects)

26.0
32.0
30.0
28.0
34.0
Grand Avg.
A1 A2 B1 B2 C1 C2
27.5
30.5
32
26
28.5
29.5
Factor Levels
Main Effects
(A2 - T)
(B1 - T)
(C1 - T)

Main effect, Factorial effect, or Column effect refers to the trend of change of the
average effect of the factor assigned to the column. Main effect is generally expressed by
the difference of the average effects at the two levels(for 2-level factor) or by plotting the
average effect. Expressed numerically.

Factor averages
__
A
1
= (Y
1
+ Y
2
) / 2 = (30 + 25 )/ 2 = 27.5
__
A
2
= (Y
3
+ Y
4
) / 2 = (34 + 27 )/ 2 = 30.5
__
B
1
= (Y
1
+ Y
3
) / 2 = (30 + 34 )/ 2 = 32.0
__
B
2
= (Y
2
+ Y
4
) / 2 = (25 + 27 )/ 2 = 26.0
__
C
1
= (Y
1
+ Y
4
) / 2 = (30 + 27 )/ 2 = 28.5
__
C
2
= (Y
2
+ Y
3
) / 2 = (25 + 34 )/ 2 = 29.5

__ __
Main effect of factor A = (A
2
- A
1
) = (30.5 - 27.5) or show graph
(Always draw graph if the factor has 3 or more levels)

Quality Characteristics
To determine which among the levels of the factor is most desirable for achieving project
objectives, it is necessary that we establish the Quality Characteristic (QC) applicable.
The QC appropriate is by the OEC equation in cases where there are multiple objectives
and the results are obtained by combining the individual evaluations.

The Condition which is likely to produce the most desirable results is obtained from
the plot of average effect.

Based on QC = Bigger is Better,

Optimum Condition = A
2
B
1
C
2

Factor Influence A(up) B(down) C(up)

Expected Performance
Performance at the optimum condition is estimated by adding the amount each selected
average effect is deviant from the grand average to the grand average.

How much do factors influence?

From main effects study we know that C, for example, doesnt have as much
influence as B or A, but what exactly is its influence?

Answers to these questions are obtained by performing Analysis of Variance (ANOVA).

Quality Characteristics Example
1. Nominal Is Best 12 volt for a 12 volt battery
2. Smaller is Better surface roughness
3. Bigger Is Better strength

_ _ _ _ _ _ _
Y
opt
= T + (A
2
- T) + (B
1
- T) + (C
2
- T)

= 29 + (30.5 - 29) + (32 - 29) + (29.5 - 29)
= 34

Additional information we want to obtain are :
Relative Influence of factor A = ?
" " B = ?
" " C = ?

Significance of influence of a factor, Confidence Interval (C.I.) of main effects,
Confidence Interval of performance at the optimum condition, etc.

Analysis Of Variance (ANOVA), which is a common statistical technique used to
study the relative influences of multiple variables, is performed to obtain the above
information.

Common Terms in ANOVA
S - Sum of Squares S' - Pure Sum Of Squares
f - Degree of Freedom C.F.- Correction Factor
V - Variance (Mean Square) T - Total Of All Observations
F - Variance Ratio e - Error.

The detailed description of ANOVA and the calculations involved will be discussed
later in this seminar. For now, some familiarity with the use of ANOVA results will be
our immediate objective.

ANOVA Table
The last column in the ANOVA table indicates the relative influence of the column
effects (factor or interaction assigned to the column). The numbers in % are
determined by taking ratio of the column variation(S) to the total variation. The nature
of the relationships among the different statistical items will be discussed later in this
seminar.

From the last column of ANOVA table, significance of the factor influences can be
determined by testing for significance. A small percentage of influence to the variation of the
results means that the factor tolerance can be relaxed, while the tolerances for factors that
show higher percentages of influence, may have to be tightened or watched carefully.
The sum of all percentage influence always add up to 100%. The last row of the table
indicates the influence of All Other Factor/Experiential Error is obtained by taking away all
factor influences from 100%.

ANOVA Screen (Qualitek-4 software)

Why should we perform ANOVA?
ANOVA offers the following statics:

* Relative influence of factors and interactions
* A level of confidence on the estimated performance at optimum condition and main
effects.
* Significance of factor and interaction influence

2.11 Experiment Designs with Larger Number of Factors

There are many other applications of an L-8 array which will be discussed in the later
sections. For now consider the following application situations.

How do we assign factors to columns?
What if we do not have enough factors?

L
8
(2
7
) Array
Col.>>
Trial# 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

Use An L
8
For

4 Factors At 2-Level
or 5 Factors At 2-Level
- Number of Trial Conditions Remain the Same
- Unused Columns Stay Empty
(Empty Columns Reflect Error Term In ANOVA)

Experiment Designs with 3-Level Factors

4 Factors at 3 levels - use L
9
(3
4
)

Up To 7 Factors At 3 Levels And 1 Factor At 2 Levels

- Use L
18
(2
1
, 3
7
)

Designs with Larger Number of 2-Level Factors

Up To 15 Factors at 2 Levels - Use L
16
(2
15
)

Experiment Designs With 4-Level Factors

Up To 5 Factors at 4 Levels - Use L
16
(4
5
)

2.12 Basic Experiment Design and Analysis Strategy

- Use Standard Array When Possible
* L-4 for 3 2-Level Factors
* L-8 for up to 7 2-Level Factors
* L-9 for 4 3-Level Factors
etc.

- Assign factors to columns arbitrarily. Leave unused columns empty.

- Run Experiments by selecting trials at random.

- Analyze Results by calculation of average column effects which yield:

* Factor influence
* Optimum condition
* Estimated performance at optimum

Note: A large number of experiments can be designed using standard orthogonal arrays. There is
also much to be gained by doing simpler experiments such as the ones discussed so far. For starting
experiments, standard arrays are strongly recommended.

2.13 Accuracy of Orthogonal Array Experiments
(An Empirical Verification )

Assume that a process performs in such a way that its response can be represented by a
simple analytical expression in terms of the three major influencing factors A, B, and C.
For purposes of investigations, these factors are each assigned two levels expressed in
numerical terms as shown below.

Response function Y = 3 x A - 10 x B + 5 x C (Y represent the process behavior)
where A1 = 10 & A2 = 20, B1 = .5 & B2 = .2 and C1 = 1 & C2 = 4

Method 1: Taguchi Experiment

T = (30 + 48 + 75 + 63)/4 = 54

Averages: A1 = (30+48)/2 = 39 A2 = (75+63)/2 = 69
B1 = (30+75)/2 = 52.5 B2 = (48+63)/2 = 55.5
C1 = (30+63)/2 = 46.5 C2 = (48+75)/2 = 61.5

Optimum Combination => A2 B2 C2
Y
OPT
= T + (A2 - T) + (B2 - T) + (C2 -T)
= 54 + (69 - 54) + (55.5-54) + (61.5 - 54) = 78

Method 2: Full Factorial Experiment

Maximum Value = 78 which checks with the optimum above obtained from orthogonal array experiment
shown above. Since the system represented by the equation in terms of factors A, B and C is assumed to
behave in a linear manner, the orthogonal array results, which is based on linear model, produces 100%
accuracy. (Note that the highest value, which happens to be the eighth combination(Y8), is not necessarily
the last combination.
L-4 Experiment Design
Trial A B C Results
1 1 1 1 Y1
2 1 2 2 Y2
3 2 1 2 Y3
4 2 2 1 Y4

Expected Results from the 4 experiments

Y1 = 3xA1 - 10xB1 + 5xC1 = 30
Y2 = 3xA1 - 10xB2 + 5xC2 = 48
Y3 = 3xA2 - 10xB1 + 5xC2 = 75
Y4 = 3xA2 - 10xB2 + 5xC1 = 63

Y1 = 3xA1 - 10xB1 + 5xC1 = 30 Y2 = 3xA1 - 10xB1 + 5xC2 = 45
Y3 = 3xA1 - 10xB2 + 5xC1 = 33 Y4 = 3xA1 - 10xB2 + 5xC2 = 48
Y5 = 3xA2 - 10xB1 + 5xC1 = 60 Y6 = 3xA2 - 10xB1 + 5xC2 = 75
Y7 = 3xA2 - 10xB2 + 5xC1 = 63 Y8 = 3xA2 - 10xB2 + 5xC2 = 78


2.14 Learning Check List

Use this page to keep track of questions whose answers you already know and list additional
questions you may want to ask the instructor. ( Do I Know ):
How to measure and quantify quality
improvement

Why should you consider combining multiple
criteria into a single index
What are major steps in application of
DOE/Taguchi

What are the three adjustments you must do
before combining evaluations from multiple
criteria
What are items of discussions in planning
session

How are the relative weightings of different
criteria of evaluations determined
What are continuous and discrete factors

In case of multiple objectives, in what way can
the Overall Evaluation Criteria(OEC) can add
value to your analysis

How are factors determined and when are they
discussed in the planning session.

What does Quality Characteristic(QC) refers to
and how is it determined
What considerations dictate how many factors
can be included in the experiment

How is QC for the OEC determined
How are the number of levels and their values
determined

What is the function and benefits of OUTER
ARRAY design
What does the term DESIGN mean and how is
an experiment designed

What is the advantage of analysis using S/N
ratios when there are multiple samples per trial
What are the desirable order of running the
experiments and why

What design strategy should you follow when
faced with a lrge number of factors and their
interaction

How many simple arithmatic calculations are
needed for an experiment with a given number of
factors and levels

It is possible to determine presence of all
possible two-factor interactions even if you did not
reserve any columns for them
How are the factor average effects calculated

In case of ROBUST design with noise factors, it
is possible to determine the average effects of the
noise factors
What are the three kinds of information the
factor average effects can produce

The common purposes of ANOVA are to
determine the relative influences of factors to the
variation of results and the Confidence
Interval(C.I)
How is the optimum condition determined

When would you know that you have confirmed
the prediction made by experimental results

What is Main Effect/Factorial Effect/Column
Effect

What are interactions and how many different
factor interactions are possible

What is the true meaning of estimate of
performance at the optimum condition

Module - 3

Interaction Studies
Interactions among factors under study is quite common. Any one factor may interact with
any or all of the other factors creating the possibility of presence of a larger number(N x {N-
1}/2) of interactions. Determining the scope of experiment by balancing the number of factors
and interactions to be included in the study, requires clear understanding of the interaction
effects. This module provides a detail procedure for detection and analysis of interaction
between two 2-level factors.

Topic Objectives:

* Design Experiments to include interaction studies.

* Test for existence and relative influence of interactions.

* Determine optimum condition and estimate optimum performance when
interaction is significant.

3.1 Experiments Designed to Study Interactions
(Between two 2-level factors)

- Interaction is dependence of one factor on another.

- When influence of one factor depends on the presence of another factor, the factors are
considered to interact with each other.

Consider the effect of temperature on human comfort at two levels of humidity.
Module 3: Interaction Studies Page 3 - 2
Confounding Effect of Interaction Between
Two 2-level Factors

Research with orthogonal arrays showed that when factors a (2-level) and b (2-level) are
placed in columns 1 and 2, their interaction effect, if any, is mixed with effects of a
factor assigned to column 3.

Notations AxB Represents Interaction Between Factors A and B
.
Notation 1x2 => 3 means that the interaction effect of two factors assigned to column 1 and
column 2 will be mixed with effect of factor placed in column 3. Thus if a factor is indeed
present in column 3 and the factors in column 1 and 2 do indeed have interaction, then the
effect determined for factor in column 3 will not be true. On the other hand if we wish to
determine the effect of interaction between factors in column 1 and 2, then we may want to
leave column 3 empty and not assign any factor.

The three columns (col# 1, 2 and 3) which contains the interacting factors and their
interaction effect, form an interacting group.

The columns in an interacting group are commutative, i.e.

1 x 2 => 3, 1 x 3 => 2 and 2 x 3 =>1

Consider an experiment designed to study factors A, B, C etc. If AxB interaction exists,
then the effect of factor C assigned to column 3 will be contaminated by the presence of
AxB. That is, the calculated effect of C, will not be accurate. Thus if interaction between
AxB is suspected, no factor should be assigned to column 3, but reserved for interaction
effect AxB.

T
1
T
2
H
H
1
2
Interaction between Temperature and Humidity

How to determine interactions between factors which are assigned to any two arbitrary
columns?

1 x 4 => ? 5 x 6 => ? etc.

Taguchi provided the Triangular Table which contains information with respect to
interaction between factors assigned to any two columns.

The triangular table is used to determine interaction between factors. There are separate
triangular tables for two, three and four level factors. The discussions in this seminar will
be limited to interactions between two 2-level factors. Use of the 3-level or 4-level
triangular table and the corresponding study of the interactions will be left to the
attendees for future study.

To study interaction between two 2-level factors, we need to reserve one 2-level column
as indicated by the triangular table. Interaction column assignments obtained from the
Triangular Table can be represented in graphical form which are known as linear graphs.

How to read the triangular table
From the table shown in the next page.
3 x 4 => 7, 2 x 6 => 8, 1 x 4 => 5, 3 x 5 => 6, 2 x 7 => 5, 1 x 5 => 4
3 x 4 => 7, 4 x 7 => 3, 3 x 7 => 4, 5 x 4 => 1, 1 x 5 => 4, 1 x 4 => 5, etc.

Triangular Table for
Two-Level Orthogonal Arrays

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
(15)
ETC...

Example 2: Cake Baking Experiment
(Designs with interaction between two 2-level factors)

A: Egg (2-level) interaction:
Egg x Milk (AxC)
B: Butter (2-level) interaction:
Butter x Milk (BxC)
C: Milk (2-level)
D: Flour (2-level)
E: Sugar (2-level)

Design Consideration
5 Factors at 2-levels requires 5 columns. two interactions between two 2-level factors
require 2 additional columns.

An L
8
has 7 2-level columns. It might work! (We say might, as we do not know yet
whether the array will satisfy all design requirements).

Linear graphs for interaction design
Linear graphs are graphical representations of the interaction readings from the Triangular
Table.

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

3.2 Guidelines for Interaction Studies

Design Strategy:
Assign interacting factors first. Then reserve the column for their interaction effect. Next,
consider other interacting factors if any. Assign the remaining factors to the available
column at random.

- Use triangular table and/or linear graph
- Start with A & C (treat interacting factors first)
- Assign A to col. 1 and C to col. 2
- Assign B to col. 4
- Assign interaction B x C = 4 x 2 to col. 6
- Assign D to col. 5 & E to col. 7

Now, try to design by assigning c to column 4.
- C to col. 4 - AxC to col. 6 - D to col. 3
- A to col. 2 - BxC to col. 5 - E to col. 7
- B to col. 1

If we want to study both AxC and BxD, could we use an L-8 Array?

Review Triangular table and confirm that an L-8 array:
- has only one independent interacting group of columns
- can be used to study more than one interaction only if there are common factors
(AxB BxC CxA or AxB AxC AxD will work but AxB CxD will not)

Notes on interaction effects
Interaction between two factors represents a situation whereby the condition of one factor
modifies the response due to the condition of the other factors. The interaction column
identification and development of interaction table, are not covered in this seminar. Please
refer to the text book SYSTEM OF EXPERIMENTAL DESIGN by Taguchi.

L
8
(2
7
) Array

A C AxC B D BxC E
TRL# 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

There are many ways to
accomplish the design. The
factor assignment shown
below represents one of the
many valid designs.

3.3 Steps in Interaction Analysis

- Treat interaction column as any other factor

- Compute main effects (say AxC) as if it were another factor. (use this for ANOVA)

- Determine importance from plot of interaction using the factor columns.

- If unimportant, ignore and determine optimum condition as done before.

- If important, include interaction incalculation of optimum condition.

Col:> A C AxC B D BxC E
Expt # 1 2 3 4 5 6 7 Results
1 1 1 1 1 1 1 1 66
2 1 1 1 2 2 2 2 75
3 1 2 2 1 1 2 2 54
4 1 2 2 2 2 1 1 62
5 2 1 2 1 2 1 2 52
6 2 1 2 2 1 2 1 82
7 2 2 1 1 2 2 1 52
8 2 2 1 2 1 1 2 78
Total = 521
Average Column Effects
__ __
A
1
= (66 + 75 + 54 + 62)/4 = 64.25 A
2
= (52 + 82 + 52 + 78)/4 = 66.00
__ __
B
1
= (66 + 54 + 52 + 52)/4 = 56.00 B
2
= (75 + 62 + 82 + 78)/4 = 74.25
__ __
C
1
= (66 + 75 + 52 + 82)/4 = 68.75 C
2
= (54 + 62 + 52 + 78)/4 = 61.50
__ __
D
1
= (66 + 54 + 82 + 78)/4 = 70.00 D
2
= (75 + 62 + 52 + 52)/4 = 60.25
_____ ____
(AxC)
1
= (66 + 75 + 52 + 78)/4 = 67.75 (AxC)
2
= (54 + 62 + 52 + 82)/4 = 62.50
_____ _____
(BxC)
1
= (66 + 62 + 52 + 78)/4 = 64.50 (BxC)
2
= (75 + 54 + 82 + 52)/4 = 65.75
__ __
E
1
= (66 + 62 + 82 + 52)/4 = 65.50 E
2
= (75 + 54 + 52 + 78)/4 = 64.75

Optimum condition (without interaction)

From preliminary observation:
A
2
C
1
B
2
D
1
E
1

Because the QC = Bigger is better, the optimum condition is obtained as

A2 C1 B2 D1 E1

This is only the preliminary observation as the interaction effects are not yet analyzed.

56
60
64
68
72
76
Plot of Factor Average Effects (Main Effects)
A1 A2 C1 C2 B1 B2 D1 D2 E1 E2

Average Column Effects are Collected in Table Form Generally Called the
Table Of Main Effects

COLUMN FACTORS LEVEL-1 LEVEL-2 DIFFE.(2-1)

1 A:EGG 64.25 66.00 1.75
2 C:MILK 68.75 61.50 -7.25
3 AxC 67.75 62.50 -5.25
4 B:BUTTER 56.00 74.25 18.25
5 D:FLOUR 70.00 60.25 -9.75
6 BxC 64.50 65.75 1.25
7 E:SUGAR 65.50 64.75 -0.75

Grand average is calculated by averaging results of all test samples.
__
Grand Average T = (66 + 75 + ... + 78)/8 = 65.125


The interaction plots shown above comes from the columns which are reserved for the
interaction. The slopes of the lines indicate the relative influence of the interaction
effects in qualitative terms. It cannot, however be used to determine whether interaction
is indeed present and if present, whether it is significant.

Test for presence of Interaction:
Whether interaction is present or not, can be found by test of presence of interaction.

Test for significance of interaction:
Whether the interaction is significant or not is found by test of significance. This test can
only be done by using ANOVA. It also requires the interaction column effects calculated
and plotted as shown above.

Strategy:
First test to see if interaction exists. if interaction is present then proceed to determine if
it is significant.

Situation Action

* Interaction absent No action needed
* Interaction present but No action needed
not significant
* Interaction present and Action Needed
is significant * Modify optimum design
* Revise optimum performance

To carry out the first test of interaction, i.e., to see if interaction really is present, we need to
make some extra calculations for average effects of the two interacting factors together.
Since we are dealing with factors that are at 2 levels, there are 4 such combinations are
possible.
56
60
64
68
72
76
Plot of Interaction Column Effects
(AxC)1 (AxC)2 (BxC)1 (BxC)2


(A
1
C
1
) (A
1
C
2
) (A
2
C
1
) and (A
2
C
2
)

These quantities are used to plot a pair of lines which become the instrument for test for
presence of interaction.

Notice that quantities like A
1
C
1
A
1
C
2
etc. which are calculated making use of the columns
where the factors assigned, are different from (AxC)
1
, (AxC)
2,
etc. which are obtained
directly from the column reserved for the interaction.

COL:>> A C AxC B D BxC E
EXPT # 1 2 3 4 5 6 7 RESULTS
1 1 1 1 1 1 1 1 66
2 1 1 1 2 2 2 2 75
3 1 2 2 1 1 2 2 54
4 1 2 2 2 2 1 1 62
5 2 1 2 1 2 1 2 52
6 2 1 2 2 1 2 1 82
7 2 2 1 1 2 2 1 52
8 2 2 1 2 1 1 2 78
TOTAL = 521
Average effect of A
1
C
1
is found by averaging results which contain of both A
1
& C
1
.
_____
(A
1
C
1
) = (66 + 75)/2 = 70.50 (first two trial results only)

Similarly
_____ _____
(A
1
C
2
) = (54 + 62)/2 = 58.00 (A
2
C
1
) = (52 + 82)/2 = 67.00 and
_____
(A
2
C
2
) = (52 + 78)/2 = 65.00

At a fixed A = A
1
average effect of C is obtained by the difference of
____ ____
(A
1
C
1
) = 70.50 and (A
1
C
2
) = 58.00

Likewise, for A = A
2
the same can be obtained by the difference of
____ _____
(A
2
C
1
) = 67.00 and (A
2
C
2
) = 65.00


The presence of interaction can be easily tested by plotting the combined factor averages

Key Observation:
If the lines are non-parallel (need not be intersecting), then interaction is present.

1. A and C interaction is strongly present. B and C interaction is negligible.
2. Among the four combinations of A and C used for the plots, condition
A
1
C
1
represent the highest value (desired for BIGGER IS BETTER QC)

52
56
60
64
68
72
C1 C2
C1 C2
A1
A2
70.5
65.0
67.0
58.0
78.5
76 B2
B1
70.0
59.0
53.0
Plots for Test of Presence of Interaction

Severity Index: The strength of presence of interaction can be measured in terms of a numerical
quantity which measures the angle between the two lines. The Severity Index is formed such that it is
100% when the lines are perpendicular and 0% when the lines are parallel. For interaction between A
and C, the generalized formula for the Severity Index is as shown below.

ABS [ (A1C2 -A1C1) - (A2C2 - A2C1) ] x 100
Interaction Severity Index(SI) = %
2 x ABS[ Highest - Lowest ]

Similarly for BxC interaction, the average effects may be calculated as
_____ _____
(B
1
C
1
) = 59.00 (B
1
C
2
) = 53.00
_____ ____
(B
2
C
1
) = 78.50 and (B
2
C
2
) = 70.00

Example:
For AxC, SI = abs[ (58-70.5) - (65-67) ]x100 / [ 2 x (70.5 - 58 ) ] = 10.5x100/25 = 42 %
For BxC, SI = abs[ (53-59) - (70-78.5) ]x100 / [ 2 x (78.5 - 53 ) ] = 2.5x100/51 = 4.9 %

Conclusion: Factor levels A
1
C
1
must be included in the description of the optimum
condition.

Leading Questions:
* Is interaction present?
* If interaction is present, what are its influences?
* How significant are interactions? (determined by ANOVA, not done at this time)
* How do we find out what are best factor levels when interaction is present?

Combined factor averages used for test of presence of interactions such as A
1
C
1
,
A
1
C
2
, B
1
C
1
, etc. are calculated using the column where factors A and B are assigned.
The calculations do not make use of the columns reserved for AxC or BxC. Thus, it is
possible to test for interaction between any two factors (say D and E) even if we didn't
reserve a special column for their interaction effect (DxE). The above observation
naturally leads to the following conclusions:

- It is possible to examine whether two factors interact or not even if we didn't think
about it or made any special provision to study them.
- Thus if seven 2-level factors are studied using 7 columns of an L-8, it is possible to
test for presence of interaction between any two factors.

Then why did we sacrifice a column for AxC, BxC, etc. in our experiment design?
- Special columns for interaction effect are reserved to carry out the complete study.
- Without a separate column for interaction, although presence of interaction
can be tested, the relative significance cannot be determined.
- Test for presence of interaction yields only qualitative (subjective)
information, whereas, the test of significance for the column effects
(in this case the column of AxC, BxC, etc.) yield influence of the
interactions in a quantitative manner.

Actions to Follow:
If interaction exists, then we need to determine its influence on:
- optimum condition
- performance at optimum condition
Find New Factor Levels at Optimum
Based on QC = BIGGER IS BETTER, find the highest values from the graph, i.e.
A
1
C
1
(70.5) And B
2
C
1
(78.5). Therefore, the levels A
1
C
1
B
2
must be included in the
optimum condition. Accordingly the estimate of performance at optimum condition
must also be revised.

As a rule, the decision about adjusting optimum factor levels is made only if the
interaction influences are determined to be significant. Of course only those interaction
that had column reserved for its influence, can be tested for significance using the
ANOVA table information. As shown in the ANOVA table below, the interaction
column effect corresponding to column 6, which represent interaction effect between
factors in column 2 and column 4, is negligible. Thus even though the test for presence
of interaction was positive, there is no action necessary due to this interaction.

(QUALITEK-4 ANOVA Screen)

Performance at the Optimum Condition

Optimum Condition Is A
2
C
1
B
2
D
1
E
1
(Without Interaction)
_ _ _ _ _ _ _ _ _ _ _
Y
OPT.
= T + (A
2
- T) + (C
1
- T) + (B
2
- T) + (D
1
- T) + (E
1
- T)
_ _ _ _ _ _
= T + (66 - T) + (68.75 - T) + (74.25 - T) + (70 - T) + (65.5 - T)

= 65.125 + 0.875 + 3.625 + 9.125 + 4.875 + 0.375

= 65.125 + 18.875 = 84.00

When interaction effects of factors A and C are considered, the revised optimum factor
levels become: A
1
C
1
B
2
D
1
E
1
(with interaction)
(Since from test of presence of interaction AxC,
A
1
C
1 are chosen as the desired levels)

In the above expression the level of the interaction column AxC is determined from the
average effects for the interaction column in the same manner as the level for the factor is
done. [Optimum expression used by QUALITEK-4 software]

Alternative Method: In this expression, the terms (A
2
- T) and (C
1
- T) in the original
expression for optimum performance, are replaced by the term (A
1
C
1
- T).

Note: In cases where two pairs of interactions (AxC & BxC) produce different levels of
the same factor (say C) select the optimum that produces a conservative estimate.
- Select the lower calculated value if QC is Bigger is Better
- Select the higher calculated value if QC is Smaller is Better

Optimum Condition and Performance
Y T AC T B T D T E T
OPT.
= + + + +
1 1 2 1 1

= + + + + T T T T T 70 5 74 25 70 65 5 . . .
= + + + + = 6 5 1 2 5 5 3 7 5 9 1 2 5 4 8 7 5 0 3 7 5 8 4 8 7 5 . . . . . .
_ _ _ _ _ ___ _ _ _ _ _ _ _
Y
OPT.
= T + (A
1
- T) + (C
1
- T) + ([AxC]
1
- T ) + (B
2
- T) + (D
1
- T) + (E
1
- T)

( ) ( ) ( ) = + + + + T T T T 64 25 68 75 67 75 . . . ....

( ) = + + + + + + = 65125 875 3625 2 625 9125 4 875 375 84 875 . . . . . . . .


Module - 4

Experiment Designs With Mixed Level Factors
Experiment designs with all factors in the same level are easier to accomplish. Generally, one
of the standard arrays can be used as is. Frequently, however, factors that are included in the
study have different levels(2, 3 or 4 levels), particularly if many of the factors are of discrete
type. This module describes procedures of modifying standard orthogonal arrays to
accommodate factors at different levels.

Topics and objectives:

* Determine degrees of freedom (DOF).
* Upgrade column levels.
* Downgrade column levels.
* Learn the combination design technique.

4.1 Modification of Standard Orthogonal Arrays

Example 3: Generator Noise Study

1 Factor(X) at 4-Levels and
4 Factors (A, B, C and D) at 2-Levels

OA selection: L-8 has enough columns but are all at 2-level. L-16(4
5
) is the smallest 4-
level array and has room for five 4-level factors. We only need one 4-level column.

Degrees of freedom, (DOF) considerations:

Definitions of DOF -
DOF of DOF of a factor = number of level - 1
DOF of a column = number of level - 1
DOF of an array = total number of column DOF
DOF an experiment = total number of results - 1

DOF = no. of levels - 1 (for factors and columns)

Total DOF for the experiment = 3 + 1 + 1 + 1 + 1 = 7
Module 4: Experiment Designs With Mixed Level Factors Page 4 - 2

An L
8
has 7 DOF (7 factors at 2-levels each).
L
8
might work! but need one 4-level column.

4.2 Upgrading Column Levels

A 4-level column possesses 3- DOF
A 2-level column has 1- DOF

Therefore, to upgrade 2-level columns into a 4-level column, we will require 3 columns of
the L
8

Procedure for column modifications
* select a set of interacting columns (say 1, 2 & 3).
* discard one (3). using the two, form a new column
(follow a consistent rule to form a new column).
* replace the two columns with the new column,
place it at column 1.

Assume conversion rules: 1 1 = > 1 1 2 = > 2 2 1 = > 3 2 2 = > 4
(It is immaterial whether we replace 1 1 by 1 or 4. But it is important to follow the
established rules for all trial conditions.)

L-8 Array
COL >>
EXPT # 1 2 3 4 5 6 7
1 1 1>1 1 1 1 1 1
2 1 1>1 1 2 2 2 2
3 1 2>2 2 1 1 2 2
4 1 2>2 2 2 2 1 1
5 2 1>3 2 1 2 1 2
6 2 1>3 2 2 1 2 1
7 2 2>4 1 1 2 2 1
8 2 2>4 1 2 1 1 2

To prepare a new column, examine
columns 1 and column 2 . If the level
numbers involve 1 and 1, replace them
with a 1. If it is 1 and 2, replace them
with a 2. Continue the process for all 8
rows. Stick to the rules established.

The rules of column combination, 1
and 1 > 1, etc. may be established in any
order desired, as long as there are only 4
new numbers for the possible 4
combinations between two 2-level
columns.


Design
* Assign the 4-level factor to column 1.
X: Casement Structure

* Assign the remaining four 2-level factors to columns 4, 5, 6 & 7:
A: Air Gap B: Impregnation C:Contact Brush D:Stator Structure

Description of 8 separate trial conditions

Trial # 1
Casement Structure = Present Design........... Level 1
Air Gap = Present Gap...............Level 1
Impregnation = Present Type..............Level 1
Contact Brush = Type 1.......................Level 1
Stator Structure = Present Design.......... Level 1

Trial # 2
Casement Structure = Present Design........... Level 1
Air Gap = Increased Gap............Level 2
Impregnation = Harder Type...............Level 2
Contact Brush = Type 2........................Level 2
Stator Structure = Epoxy Coated.............Level 2

Trial # 3
Casement Structure = Textured....................Level 2
Air Gap = Present Gap.............. Level 1
Impregnation = Present type.............. Level 1
Contact Brush = Type 2.......................Level 2
Stator Structure = Epoxy Coated ..........Level 2
etc.
MODIFIED ARRAY
WITH FACTOR ASSIGNMENTS
COL >> X A B C D
EXPT # NEW COL 4 5 6 7
1 1 1 1 1 1
2 1 2 2 2 2
3 2 1 1 2 2
4 2 2 2 1 1
5 3 1 2 1 2
6 3 2 1 2 1
7 4 1 2 2 1
8 4 2 1 1 2

There are four steps in analysis of results

Step 1: Preparation and arrangements of results

Preparation
- Not needed when the performance is evaluated
in terms of a single criteria

- When there are multiple criteria of evaluations, they
must be combined into an overall evaluation
criteria (OEC, single number).

Note: That for analysis purposes, there can be only one result for each of the samples tested.
One column of results imply that there were only one sample tested per trial.

Arrangements
- results of one sample per trial are placed in the respective trial/rows.

Step 2: Calculation of column average effects
- Average column effects are calculated by using the modified OA and the experimental
results following the procedure used earlier and are arranged in a table of main effects.

The Main Effects
COL#/FACTORS LEVEL 1 LEVEL 2 LEVEL 3 LEVEL 4 (L2-L1)
1 CASEMENT STRUC 56.0000 72.5000 66.5000 69.5000 16.5000
4 AIR GAP 63.2500 69.0000 0.0000 0.0000 5.7500
5 IMPREGNATION 64.7500 67.5000 0.0000 0.0000 2.7500
6 CONTACT BRUSH 66.7500 65.5000 0.0000 0.0000 1.2500
7 STATOR STRUCTU 63.7500 68.5000 0.0000 0.0000 4.7500
QUALITY CHARACTERISTIC: SMALLER IS BETTER

MODIFIED ARRAY
WITH FACTOR ASSIGNMENTS
COL >> X A B C D
EXPT # NEW COL 4 5 6 1 Results
1 1 1 1 1 1 50
2 1 2 2 2 2 62
3 2 1 1 2 2 70
4 2 2 2 1 1 75
5 3 1 2 1 2 68
6 3 2 1 2 1 65
7 4 1 2 2 1 65
8 4 2 1 1 2 74

Step 3: Analysis of variance analysis (ANOVA)
- ANOVA is a relatively complex calculation process and will be discussed later (Section
B). ANOVA allows us to examine the relative influence of the column effects and select
the significant factors for optimum design.

Note: More elements of ANOVA are known to us now. The terms like S, V, F and S' are still
not known to us yet.

Step 4: Determine Optimum Condition and Performance.

- From the main effects table, using the QC, the optimum factor levels are selected and
the individual factor contributions are determined. The column effects (factor and
interaction) deemed insignificant in ANOVA are not included in the optimum table.

The expected result at the optimum condition represents the average performance of a
number of samples tested at the optimum condition. No sample may perform exactly as the
optimum value, but all samples are expected to perform near the expected value (bounded by
the Confidence Interval).

OPTIMUM TABLE: ESTIMATED PERFORMANCE AT OPTIMUM CONDITION

COLUMN#/FACTORS DESC. LEVEL DESC. LEVEL# CONTRIBUTION
1 CASEMENT STR PRESENT 1 -10.1250
4 AIR GAP PRESENT 1 -2.8750
5 IMPREGNATION SOFTER 1 -1.3750
7 STATOR STRUC STD. DSGN 1 -2.3750
TOTAL CONTRIBUTION FROM ALL FACTORS .... -16.7500
CURRENT GRAND AVERAGE OF PERFORMANCE ... 66.1250
EXPECTED RESULT AT OPTIMUM CONDITION .. 49.3750

A N O V A Table
COL.#/FACTOR f S V F S' P(%)
1 CASEMENT STR 3 309.3750 103.1250 33.0000 300.0000 68.3565
4 AIR GAP 1 66.1250 66.1250 21.1599 63.0000 14.3548
5 IMPREGNATION 1 15.1250 15.1250 4.8400 12.0000 2.7342
6 CONTACT BRUS(1)(3.1250) POOLED
7 STATOR STRUC 1 45.1250 45.1250 14.4399 42.0000 9.5699
OTHER/ERROR 1 3.1250 3.125 4.9843
TOTAL: 7 438.8750 100.00%


4.3 Downgrading Columns
(Changing a 3-Level column to 2 or 4-Level to 3)

Suppose that we have three 3-level factors and one 2-level factor.

No standard array is available. an l
9
has four 3-level columns.

The process of down-grading a column involves taking the standard column and simply
changing one level to another.

The process is commonly known as Dummy Treatment

For the case above start with an l-9 array
- Select a column, say col. 3
- Replace level 3 with 1
- Note the number changed by marking with a (')
- Select the substituting level as one where performance is known to be less stable.

Note:
1. When a column is dummy treated, by replacing 3 with 1', as done for the third column above, the column
looses balance within itself as well as with other column. Even though one or more columns of the orthogonal
array is no longer balanced, the design still is valid and ready to yield the desired conclusions upon completion of
analysis.
2. The level substituted for the highest column level may be any of the lower ones. However, level where
performance is expected to exhibit larger variations, should be selected when possible.
3. In the similar manner a 4-level column can be reduced to a 3-level column by replacing 4 with 1 or any other
desired level.

L
9
Array
B C A D
EXPT# 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

(' )indicates new
modified levels 1'
instead of 3.

Any column can be
selected for Dummy
Treatment.

Example 4: Lost Foam Casting Process

Process factors: 6 factors at 2 levels
2 factors at 3 levels
1 factor at 4 levels

Minimum DOF requirement:

DOF for 6 2-level factors 6 x (2 - 1) = 6
Total DOF = 13

An L
16
(2
15
) has 15 DOF. It might work!

An L-16 has 15 2-level columns. Our factors require 3 columns that are of levels higher
than 2.
We know how to prepare a 4-level column from 3 2-level columns and how to
downgrade a 4-level column into a 3-level column. But we do not have a technique to
prepare 2-level columns directly into a 3-level one, which we need for this experiment.

Column Modification Strategy
First upgrade 3 columns into 4-level columns, then, dummy treat two of the 3 4-level
columns to 3-level columns. To upgrade three columns to 4-level columns We need
Three Interacting Sets (see Triangular Table)

1 2 3 4 8 12 7 9 14

(These sets are only one selection among many other possible interacting groups of
columns)

Change all three sets first into 4-level columns. then dummy treat any two of these 4-level
columns to two 3-level columns. assign the two 3-level factors to these two columns.

Assign the 4-level factor to the 4-level column prepared from the set 7 9 14.


Column Modification Procedure (L-16 Array)

Upgrading 2-level columns to form 3-level & 4-level columns.
1 2 & 3 to form A 4 8 & 12 to form A 7 9 & 14 to form A
3-level col. new 1" 3-level col. new 4" 4-level col. new 7"

1 2 3 NEW 1 4 8 12 NEW 4 7 9 14 NEW 7
1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 2 2 2 1 2 2 2
1 1 1 1 2 1 2 3 2 1 2 3
1 1 1 1 2 2 1 4=1' 2 2 1 4
1 2 2 2 1 1 1 1 2 1 2 3
1 2 2 2 1 2 1 2 2 2 1 4
1 2 2 2 2 1 1 3 1 1 1 1
1 2 2 2 2 2 1 4=1' 1 2 2 2
2 1 2 3 1 1 1 1 2 2 1 4
2 1 2 3 1 2 1 2 2 1 2 3
2 1 2 3 2 1 1 3 1 2 2 2
2 1 2 3 2 2 1 4=1' 1 1 1 1
2 2 2 4=1' 1 1 1 1 1 2 2 2
2 2 2 4=1' 1 2 1 2 1 1 1 1
2 2 2 4=1' 2 1 1 3 2 2 1 4
2 2 2 4=1' 2 2 1 4=1' 2 1 2 3

Column modification rules: 1 1 > 1 1 2 > 2 2 1 > 3 2 2 > 4

Original L
16

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

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


When the orthogonal array is modified, the factors can then be assigned to the appropriate
columns. While the 3-level and 4-level factors are assigned to the columns prepared to
accommodate them, if there are interactions between two level factors, there assignments must
carefully follow the rules of interaction design discussed earlier.

Experiment Design Table

Col.# FACTORS LEVEL-1 LEVEL-2 LEVEL-3 LEVEL-4
1 I:Coating Type Type 1 Type 2 Type 3 ---------
2 Unused/Upgraded M/U --------- -------- ---------
3 Unused/Upgraded M/U --------- --------- ---------
4 G:Sand Compaction Plant X Plant Y Plant Z ---------
5 A:Metal Head Low high --------- ---------
6 B:Sand Supplier Supply 1 Supply 2 --------- ---------
7 H:Gating Type Plant X Plant Y Plant Z Plant W
8 Unused/Upgraded M/U
10 C:Sand Perm 200 Perm 300 Perm --------- ---------
11 D:Metal Temperature 1430 F 1460 F --------- ---------
13 E:Quench Type 450 F 725 F --------- ---------
15 F:Gas Level None High --------- ---------
Characteristic: The Bigger The Better (as selected for analysis)

Modified (L-16)
Expt\Col:1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
expt 1 1 0 0 1 1 1 1 0 0 1 1 0 1 0 1
expt 2 1 0 0 2 1 1 2 0 0 2 2 0 2 0 2
expt 3 1 0 0 3 2 2 3 0 0 1 1 0 2 0 2
expt 4 1 0 0 1 2 2 4 0 0 2 2 0 1 0 1
expt 5 2 0 0 1 1 2 3 0 0 2 2 0 1 0 2
expt 6 2 0 0 2 1 2 4 0 0 1 1 0 2 0 1
expt 7 2 0 0 3 2 1 1 0 0 2 2 0 2 0 1
expt 8 2 0 0 1 2 1 2 0 0 1 1 0 1 0 2
expt 9 3 0 0 1 2 1 4 0 0 1 2 0 2 0 2
expt 10 3 0 0 2 2 1 3 0 0 2 1 0 1 0 1
expt 11 3 0 0 3 1 2 2 0 0 1 2 0 1 0 1
expt 12 3 0 0 1 1 2 1 0 0 2 1 0 2 0 2
expt 13 1 0 0 1 2 2 2 0 0 2 1 0 2 0 1
expt 14 1 0 0 2 2 2 1 0 0 1 2 0 1 0 2
expt 15 1 0 0 3 1 1 4 0 0 2 1 0 1 0 2
expt 16 1 0 0 1 1 1 3 0 0 1 2 0 2 0 1

Module 4: Mixed Level Factor Designs Page 4 - 10

4.4 Incompatible Factor Levels

Experiments designed using orthogonal arrays require that all levels of all factors go with
all levels of the other factors included in the experiment.

The trial conditions described by the rows of level in the OA are arranged such that any
level of a factor, say a
1
, must exists with all levels of other factors such as B
1
, B
2
, C
1
,
C
2
, etc.
For example, A
1
must go with B
1
in Trial #1 and with B
2
in Trial #2. In some situations
trials may be incompatible and may not be carried out.

HOW to avoid incompatibility

* Combine factors only when they do not interact.
* It is always possible to combine two 2-level factors into one 4- level factor.

It is always possible to combine two 2-level factors into one 4-level factor. Thus X and Y
can be combined to create a new factor at 4-level.

New factor (XY) obtained by combining X and Y

(XY) > (XY)
1
= X
1
Y
1 ,
(XY)
2
= X
1
Y
2,
(XY)
3
= X
2
Y
1 ,
(XY)
4
= X
2
Y
2

4.5 Combination Design (A Special Technique)
(Fitting Two 2-level factors into one 3-level column)

Consider: 3 3-Level Factors (A,B,C)
2 2-Level Factors (X,Y)

L
4
(2
3
) Array
Trial# 1 2 3
1 1 1 1
2 1 2 2
3 2 1 2
4 2 2 1

/ X
1
/ Y
1

FACTOR X =| FACTOR Y =|
\ X
2
\ Y
2

Consider Three 2- Level
FactorS A, B, & C Studied
Using an L-4 ARRAY.


L
9
OA has 4 3-level columns. in the combination design, one of the four 3-level columns
can be used to accommodate the two 2-level factors.

If we were to look for a standard array to handle the five factors, we will need to go for an
L-16. Use of L-9 will result in 7 less experiments.

Assign A, B & C TO Columns 1, 2 & 3. Assign X & Y TO Column 4

Since factors X and Y can only be assigned to one column, they are combined to create a
new factor (XY) which may be assumed to have four levels as shown below.

X
1
Y
1
= (XY)
1
X
1
Y
2
= (XY)
3

X
2
Y
1
= (XY)
2
X
2
Y
2
= (XY)
4

Select any three levels, say (XY)
1
(XY)
2
(XY)
3.
Ignore the remaining level, Level
(XY)
4 in this case.

Assign (XY)
1
(XY)
2
(XY)
3
to the 3 levels of Col. 4

* Level 1 of column 4 will mean (XY)
1
i.e. X
1
Y
1

* Trial condition 1 will then be (A
1
B
1
C
1
X
1
Y
1
)

Once the experiments are carried out and the results collected, the average effects of
factors assigned to various columns can be calculated. Thus the three average effects of
the fictitious factor (XY), i.e., (XY)
1
, (XY)
2
and (XY)
3
are obtained.

Analysis of Results:

FACTORS A B C (XY)
EXPT.# 1 2 3 4
1 1 1 1 1
2 1 2 2 2
3 1 3 3 3
4 2 1 2 3
5 2 2 3 1
6 2 3 1 2
7 3 1 3 2
8 3 2 1 3
9 3 3 2 1


Main effect of X AND Y are computed from average effects of (XY).

To establish main effects of two factors, we generally need 4 data points.

We only have 3 levels in (X,Y). yet main effects of X and Y individually can be
extracted from 3 levels of (X,Y) as follows.

Assumption: There is no interaction between X and Y. This assumption is necessary since the
conclusions about the main effects are drawn based on a fixed value of the other factor. The main effect of
X above is only valid when Y = Y1.

___ ____
MAIN EFFECT OF X = (XY)
2
- (XY)
1

____ ___
= X
2
Y
1
- X
1
Y
1
AT FIXED Y = Y
1

___ ___
MAIN EFFECT OF Y = (XY)
3
- (XY)
1

____ ____
= X
1
Y
2
- X
1
Y
1
AT FIXED X = X
1


Course Recap

Design

- Simple cases, standard OA
- Interaction
- Mixed levels
- Combination design
- Robust/noise factor design ...yet to be discussed

Analysis

- Main effects
* Influence
* Optimum
* Estimated performance
- Analysis of variance ...yet to be discussed
* % contribution
* Confidence interval

Loss Function ...yet to be discussed

- How much $ saving
- Tolerances

Brainstorming ...more discussion to come


4-1: Design experiments to study the following situations with factors at mixed levels.
Select the appropriate orthogonal array and indicate the column assignments.
a. Three two level factors and one 4-level factor
b. Four 2-level factors and one 3-level factor
c. Two 3-level factors and one 2-level factor
d. Seven 3-level factors and one 2-level factor
e. Four 4-level factors and three 2-level factors
f. Three 4-level factors, two 3-level factors and five 2-level factors.

4-2: Design an experiment to economically study three 3-level factors (A, B and C) and
two 2-level factors (D and E). Assume that interaction between factors D and E is absent.

4-3: For the following experiment design.

a. Determine the trial # for the condition described below. (Check the correct answer)
Casement Structure = Bolted
Air Gap = Increased Gap (a) Trial #2
Impregnation = Present Type (b) Trial #6
Contact Brush = Type 2 (c) Trial #7
Stator Structure = Present Design (d) Trial #5
Review Questions

COL.# FACTORS LEVEL-1 LEVEL-2 LEVEL-3 LEVEL-4
1 CASEMENT STRUCTURE PRESENT TEXTURED BOLTED PRES.FIT
2 M/U
3 M/U
4 AIR GAP PRESENT INCREASE
5 IMPREGNATION PRESENT HARDER
6 CONTACT BRUSH TYPE 1 TYPE 2
7 STATOR STRUCTURE PRESENT EPOXY

#\COL: 1 2 3 4 5 6 7
TRIAL 1 1 0 0 1 1 1 1
TRIAL 2 1 0 0 2 2 2 2
TRIAL 3 2 0 0 1 1 2 2
TRIAL 4 2 0 0 2 2 1 1
TRIAL 5 3 0 0 1 2 1 2
TRIAL 6 3 0 0 2 1 2 1
TRIAL 7 4 0 0 1 2 2 1
TRIAL 8 4 0 0 2 1 1 2


b. Describe trial condition 4 (Spell out the factor levels appropriately)
Trial # 4 Casement structure =
Air Gap =
Impregnation =
Contact Brush =
Stator Structure =

4-4: To accommodate a 4-level factor in an L-8, which 3 columns will you work
with? Check all correct answers.
a.( ) Any three interacting group of columns
b.( ) Columns 1, 2 and 3 c.( ) Columns 2, 6 and 4
d.( ) Columns 1, 5 and 4 e.( ) Columns 1, 6 and 7

4-5: Which OA will you use to design an experiment with two 4-level factors and
one 2-level factor? ( ) L
8
( ) L
12
( ) L
16

4-6: To investigate a process with two factors at 4-levels, two factors at 3-levels
and 3 factors at 2-levels, an experimenter proposed using an L-16 in the
following way

Use columns 1, 2 and 3 for 1st 4-level factor
Use columns 4, 8 and 12 for 2nd 4-level factor
Use columns 5, 15 and 10 for 1st 3-level factor (with dummy treatment)
Use columns 7, 14 and 9 for 2nd 3-level factor (with dummy treatment)
Use columns 6, 11 and 13 for the three 2-level factors
Is this design correct? ( ) Yes ( ) No

4-7: An experimenter used an L-9 OA to study three 3-level factors (A, B, C) and
two 2-level factors (X and Y). Using the principles of combination design,
three combinations of X and Y (i.e. X
1
Y
1
, X
1
Y
2
, AND X
2
Y
1
) were assigned
to the three levels of column 1. Using the results as shown below, determine:

L-9 Array
Factors(XY) A B C
Trial 1 2 3 4 Results
1 1 1 1 1 60
2 1 2 2 2 50
3 1 3 3 3 40
4 2 1 2 3 55
5 2 2 3 1 45
6 2 3 1 2 35
7 3 1 3 2 65
8 3 2 1 3 60
9 3 3 2 1 55


____ ____ ____
a.) Average effect X
1
Y
1
= X
1
Y
2
= X
2
Y
1
=

b.) Main Effect of X c.) Main effect of Y.

4-8. [Concept: Degrees of Freedom ]

Below is an experiment designed to study four 3-level factors (A, B, C, and D). Three samples
per trial condition was tested. Determine:

i) Degrees of Freedom (DOF) of factor A. Ans. _______

ii) Total DOF of all the factors. Ans. _______

iii) DOF of the fourth column of the array. Ans. _______

iv) DOF of the array used for the experiment. Ans. _______

v) Total DOF of the experiment. Ans. _______

vi) Extra DOF of the experiment (over what the factors need). Ans. _______

vii) Identify the orthogonal array that has enough DOF to satisfy the needs of

Three 3-level factors
One 4-level factor
and Two 2-level factors

Ans. Total DOF needed _________ Array _________

Trial\Factor A D B C Results Average

1 | 1 1 1 1 | 3 4 5 4
2 | 1 2 2 2 | 7 6 5 6
3 | 1 3 3 3 | 4 5 3 4
4 | 2 1 2 3 | 6 7 5
5 | 2 2 3 1 | 8 9 7
6 | 2 3 1 2 | 9 10 8
7 | 3 1 3 2 | 6 8 7
8 | 3 2 1 3 | 5 7 6
9 | 3 3 2 1 | 5 6 7

4-9. [Concept: Array Modifications ]
In the experiment shown below, the 3-level factor D is assigned to column 4. All other factors
in the experiment are at 2-level. Modify the orthogonal array to correctly design your
experiment and determine.

i) Which columns would you combine to form a 3-level column? Ans. ________

ii) Indicate the level numbers for modified column 6: Ans._________

iii) Is it possible to determine if interaction between A and B exists? Ans. ________

iv) For Smaller is Better quality characteristic, plot the main effect and identify the level of
factor D which is desirable for the optimum condition.

L-8 A B D C E
Trial\Col# 1 2 3 4 5 6 7 Results

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

__
D1 =
__
D2 =
__
D3 =
D1 D2 D3
Main Effect of factor D

4-10. [Concept: Combination Design ]

An experiment to study three 3-level factors(A, B, and C) and two 2-level factors (X and Y)
was designed using an L-9 array. The factors X and Y were combined to form a 4-level factor
and assigned to column 2. Assuming that the first three levels of XY was used for the
experiment, determine:

I) Main effect of X. (Assume that there is no interaction between X and Y)

ii) Optimum condition when the quality characteristic is Smaller is better.

Trial\Factor A (XY) B C Results Average

1 | 1 1 1 1 | 3 4 5 4
2 | 1 2 2 2 | 7 6 5 6
3 | 1 3 3 3 | 4 5 3 4
4 | 2 1 2 3 | 6 7 5
5 | 2 2 3 1 | 8 9 7
6 | 2 3 1 2 | 9 10 8
7 | 3 1 3 2 | 6 8 7
8 | 3 2 1 3 | 5 7 6
9 | 3 3 2 1 | 5 6 7
____
X1Y1 = (4 + 6 + 4 ) / 3 = 4.66
____
X1Y2 =
____
X2Y1 =
X1 X2
MAIN EFFECT OF FACTOR X
__
A1 =
__
A2 =
__
A3 =
__
B1 =
__
B2 =
__
B3 =
__
C1 =
__
C2 =
__
C3 = Optimum Condition: _________

4-11. [Concept: Validation of results ]

Experiment design (L-8 array):
col. 1 A:Chemical ratio A1 A2 (all factors at two levels)
col. 2 H: Flow rate H1 etc.
col. 3 Interaction AxH
col. 4 Interaction AxD
col. 5 D:Pour weight
col. 6 B:Mold temperature
col. 7 C:Lid close time

Results of three samples tested in each trial: QC = Bigger is better

Average

Calculated values:
Grand Avg. =167.29
Levels: 2 1 2 1 1
Yopt = 232.83 w/o Int.
= 258 (with one Int.)

C.I. from ANOVA: +/- 9.36
Confirmation tests with five samples produced:

255 248 264 268 258 ==> 1293 / 5 = 258.6 (Average)

This exercise should be done as a group.
Consider that for the situation described in Exercise 2.5 was addressed with an experiment using an L-8
array. Three samples were tested for each trial condition, and the results analyzed. To confirm the estimate of
performance, five additional samples were tested at the optimum condition. The factor descriptions and the
results are as shown below.


Determine: (perform standard analysis)

I. Optimum condition without interaction consideration and Yopt.
II. Which one of the two interactions is stronger.
III. Determine Optimum condition and Yopt with interaction effect of II
IV. Determine if there is interaction between factors BxC. Is it as strong as the two
interactions which are included in the experiment.
V. Compare the confirmation test results with the calculated optimum performance and the
C.I. What course of action will you follow and why (ignore interaction and accept results,
refine prediction of optimum with interaction, repeat experiment with new interactions,
brainstorm for more factors, go for robust design, etc.)?

4-12: An L-8 OA was used to study one 4-level factor (A, col. 1) and four 2-level factors (B,
C, D and E cols. 4, 5, 6 and 7). One sample at each trial condition produced the following
results:

Test Results: 71 68 70 75 68 65 65 and 74

If SMALLER value of the result is desired, determine:

a) Optimum condition
b) Performance at optimum condition
c) Main effect of factor D

Ans: A
3
B
1
C
2
D
2
E
1
, 62, (67.00 - 72.00)

4-13: An experimenter used an L-9 OA to study three 3-level factors (A, B, and C in
cols. 1, 2, and 3) and two 2-level factors (X and Y in col. 4 as X1Y1 X1Y2 and X2Y1).
The results of the nine trials are as follows:

Test Results: 42 45 38 56 36 52 64 68 54

If LARGER value of the result is desired, determine: a) The optimum condition, b) Main
effect of factor X and c) Optimum performance.

Ans. A3 B1 C1 X2 Y1 (Level 3 of XY), (54.00 - 44.00), 72.33

Module - 5

Robust Design Strategies
Variation in performance occurs mainly due to control factor and noise factors
(uncontrollable). While DOE can identify the influential control factors which can indeed be
adjusted to improve the consistency in performance, for many system, the uncontrollable
factors cause most of the variation. In such situations, Taguchi, in his ROBUST DESIGN
strategy, proposes to minimize the influence of uncontrollable factors by adjusting the levels
of the controllable factors. In this approach, the desired design is sought not by selecting the
best performance under ideal condition, instead by looking for a design that produce
consistent performance having been exposed to the influence of the uncontrollable factors.

Topics and objectives:

* Understand the necessity of multiple runs.
* Follow the experimental strategy for robust product/process.
* Include noise factors using outer array design.
* Develop criteria for analyzing multiple results:
- MSD
- S/N Ratios

5.1 Ambitious Business Goals

Make product robust (rugged, insensitive) such that they work, all the time, the way
that they are supposed to. Try to achieve:

- Consistent performance
- Robust design
- Performance that is unaffected by influence of uncontrollable factors.

Why does the performance vary?
- Noise factors

How do we minimize the influence of noise factors?
- not by controlling them, but by adjusting
The Controllable Factors
Module 5: Robust Design Strategies Page 5 - 2

What is the design strategy:
- Robust design strategy
- Outer array design with noise factors

Why do we need to repeat experiments(trials)
When performance varies, we need to test more samples to get a representative
performance reading.

How to Repeat?

Just repeat or repeat with a purpose?

What is the purpose?
Our purpose is to reduce variation of performance around the target.

What causes variations? Variation in performance is caused by factors we cannot control,
do not want to control or are not aware of their influence. Such factors are called "Noise
Factors".

What do we want to do about the influence of the noise factors? Contrary to the
traditional approach to determine the causes of variation and try to control them, in
ROBUST DESIGN strategy, the approach is to reduce the influence of the noise factors
by simply controlling the controllable factors.

What can you do to reduce the influence of the noise factors?

Design experiments with controllable factors as usual. Identify the applicable noise
factors for the subject project and determine their combinations by using the appropriate
orthogonal array. Repeat the trial conditions by exposing them to the influence of the
noise factors.

Robust Design Strategy

- Identify noise factors

- Control them during tests in laboratory environment and work with at least two
levels.

- Combine the noise factors using the orthogonal array (outer array) to produce a
number of noise conditions.

- Repeat samples in the same trial conditions exposing each to the noise conditions.

Consider that we have three 2-level noise factors in the experiment with the cake
baking process.

Noise Factors: Oven type (N
1
)
Humidity (N
2
)
Room temperature (N
3
)

These 3 noise factors, at 2-level each, can be combined to produce 8 conditions.

Or

Alternatively we can use Taguchi L
4
OA to produce 4 conditions.

Objectives:
determine the optimum condition by selecting the controllable factor levels such that
variations due to uncontrollable factors is minimized.

" Reduce variability without actually
removing the cause of variation "

Approach:
Run multiple experiments (2 or more samples/Trial).

5.2 Mechanics of the Outer Array Designs

Outer Array design incorporates both control and noise factors. control factors and inner
array define the trial conditions, while the noise factors being assigned to outer array
define the conditions to which the trial conditions are exposed to while performing the
experiments.


Symbolic Result Notations:
The design requires (8 x 4 =) 32 samples tested and the results arranged as shown above.
For example, R
73
represents the result obtained by testing trial number 7 under the noise
condition 3 i.e. Oven at level 2, Temperature at level 1 and Humidity at level 2.
(Subscripts of all results , R, are not shown)

Advantages: Information about noise factors, interaction between control and noise
factors, number repetitions, noise conditions, etc.

Disadvantages: Higher level of disciplines, costlier, etc.

Order of Sophistication in Experiment Design
(From most desirable to least desirable)

1. Formal treatment of noise factors by outer array design
2. Repeating experiments with "RANDOM NOISE"
3. Run multiple samples per trial (Simply repeat)
4. Run one sample per trial (Poor man's experiment)

Experiment Design With Noise Factors

Tr#
1
2
3
8
1 2 3 4 5 6 7
L
8
Control factors
Noise Factors
L
4
Outer Array
Inner Array
R e s u l t s
1 2 3 4
1
2
3
1 1 2 2
1 2 1 2
1 2 2 1
R R R R
R R R R
R R R R
R R R R
. . . . . . .
. . .. . . .. .. . . .. .
73
11
21
Oven type
Temperature
Humidity
*


5.3 Benefits of Outer Array designs

* Optimum condition determined using outer array is least sensitive to the variation of
noise factors. (robust design)

* Number repetition and the conditions of the noise factor levels are discretely
determined by the size of the outer array.

* Influence of noise factor can be easily calculated in the same manner as control
factors (main effect of noise factors).

* Interaction between control factors and noise factors can also be determined if
desired.

Example 5: Robust Bearing Design Study

Designs Factors and Their Levels
4 factors at 2-levels each
3 interactions and 3 noise factors at 2-levels each

Inner Array(L-8) With Control Factors

Col# Factor Description Level-1 Level-2
1 Collar Rotor 0.020 0.070
2 Shaft To Bearing 0.015 0.040
3 Interaction 1x2 N/A
4 Finger To Drive 0.150 0.20
7 Rotor Chuck 0.025 0.040

Outer Array(L-4) And The Noise Factors

Col# Factor Dec. Level-1 Level-2
1 Temperature 70 F 150 F
2 Pressure 200 350
3 Fuel Type Type A Type B


General Guidelines for Repeating Experiments

* Whenever experimenting with actual hardware, run multiple samples per trial.

* When results under the same trial condition do not repeat, repeat samples.

* Experiments with analytical simulation need not be repeated.

* When possible, identify noise factors and include outer array in your design.

- If outer array isn't possible, run experiments
under random noise conditions.

* Determine number of repetitions arbitrarily based on expected variability and cast of
extra sample.

Note: Noise condition is considered random when the noise factors are identified and the
experiments are carried out at their varying noise levels.

5.4 Analysis of Repeated Results

Comparison of Multiple Data

8 9 10 => Average 9
7 9 11 => Average 9

This two set would look the same if we only compared the averages. Nature of
distribution of data represented by Standard Deviation, Scatter, etc. will be required to
accurately compare the two data sets.

Trl# A B C R1 R2 R3 R4 AVG.
1 20 30 22 26 24.5
2 30 34 23 27 26.0
3 32 34 26 28 30.0
4 36 32 30 30 32.0


Observations:

- Need a single quantity to compare each trial condition

- Average alone does not tell the whole story

- Require a quantity that includes both average and standard deviation

Conclusion:

- Need to devise a new yardstick of measurements which will be simple, yet include all
the desirable characteristics.

MSD And S/N Ratio

Mean Squared Deviation (MSD)
Reducing variation around the target is the objective of taguchi experiments. MSD
measures variation around the target and is also a function of Average and Standard
Deviation.

Lower (Y
avg
. - Y
0
) And STD. Deviation are satisfied by minimizing MSD.

MSD = A measure of Deviation of result from the target.
y
Y Y
o
_
Target
Avg.


generally speaking MSD = (Y
i
- Y
o
)
2
/n

Where Y
o
= Target Value.

It can be shown that

MSD = (STD. DEV.)
2
+ (Y
avg
. - Y
o
)
2

Which means that MSD can be minimized by lowering standard deviation and /or
reducing the distance of the average to the target.

5.5 Definition of MSD for the Three QCs

MSD is a quantity of which we always seek a smaller value. to maintain this
characteristic common for all three QC., the definitions of MSD are modified slightly
for the other two QC's. accordingly.

Nominal: MSD = [(Y
1
- Y
o
)
2
+.(Y
2
- Y
o
)
2
+ (Y
3
- Y
o
)
2
+. .]/n

Smaller: MSD = [( Y
1
2
+ .Y
2
2
+Y
3
2
+ . .)]/n

Bigger: MSD = [(1 / Y
1
2
+ 1 / Y
2
2
+ Y
3
2
+ ...... )] / n

Recommended Yardstick for Analysis

For convenience of handling a wide range of results and to increase its linear behavior,
MSD is transformed to S/N. when results are made to influence the performance in a
linear manner, the estimate of performance (Yopt), which uses a linear predictor
model, becomes more capable of performance more accurately.

S/N = - 10 LOG
10
(MSD)

S/N is called the Signal to Noise ratio of the result.


S/N ratio and MSD are defined such that, regardless of the quality characteristic,
i.e. bigger, smaller or nominal,

* Smaller value for MSD and
* Larger value for S/N ratio is desired

Example 6: Cam Lifter Design Study

Sample calculation of S/N:

1st. ROW: MSD = (2
2
+ 3
2
+ 4
2
)/3 = 9.666

S/N = -10 LOG
10
(MSD) = -9.86

2nd. ROW: MSD = (4
2
+ 5
2
+ 3
2
)/3 = 16.666

S/N = -10 LOG
10
(MSD) = -12.218

etc.

If we were to compare results of trials 3 and 4 to determine which one is better, we can
now easily do that by comparing the S/N ratios.

Based on averages, condition results of trials 3 & 4 are equal. Based on S/N ratios,
condition 3 is better, since comparing

-14.1 > -14.4 ( -14.1 is bigger than -14.4 )

Expt. A B C R1 R2 R3 S/N Avg.
1 1 1 1 2 3 4 - 9.86 3
2 1 2 2 4 5 3 -12.2 4
3 2 1 2 4 5 6 -14.1 5
4 2 2 1 3 5 7 -14.4 5

The Main Effects
Col# Factors Level-1 Level-2 (L2-L1)
1 Spring Rate -11.04 -14.26 -3.23
2 Cam Profile -11.97 -13.32 -1.35
3 Wt of Push Rod -12.14 -13.16 -1.02

An experiment with
three 2-level factors (A,
B and C) and 3 samples
per trial yielded the
following results (QC =
Smaller is Better)

Module 5: Robust Design Strategies Design of Experiments Using Taguchi Approach Page 5 - 10

A1 A2 B1 B2 C1 C2
T
-15
-14
-13
-11
-12
Plot of Factor Average Effects

Optimum Condition : A
1
B
1
C
1

Note that optimum condition is selected based on the higher values of S/N. while performing S/N
analysis, higher values will always be selected regardless of quality characteristic (in this case, smaller
the better).

Example 7: Engine Idle Stability Study
Three Factors at 3-levels each
Three Repetitions ( normal operating noise)

Design Factors and their Levels
Col. # Factors level-1 level-2 level-3
1 Indexing -30 deg o deg +50 deg
2 Overlap area -30% 0 % +30 %
3 Spark advance 20 deg 30 deg 40 deg
4 Unused/upgraded M/U
Characteristic: Smaller is better (as selected for analysis)

Estimate of Performance at the Optimum Condition
(QC: smaller is better)
Factor description level description Level# Contribution
Spring rate Current design 1 1.6105
Cam profile Type 1 1 0.6729
Wt. Of push rod Lighter 1 0.5100
Contribution from all factors (total)......................... 2.79336
Current grand average of performance........................ -12.64613
Expected result at optimum condition......................... -9.85276


TRIAL RESULTS

Trial# R(1) R(2) R(3) S/N Ratios
1 20 25 26 -------- -------- -27.5358
2 34 36 26 -------- -------- -30.1815
3 45 34 26 -------- -------- -31.0913
4 13 23 22 -------- -------- -25.9550
5 36 45 35 -------- -------- -31.8051
6 23 25 34 -------- -------- -28.8649
7 35 45 53 -------- -------- -33.0528
8 56 46 75 -------- -------- -35.5939
9 35 46 53 -------- -------- -33.1175
L
9
(3
4
)
COL==>
COND 1 2 3 4
1 1 1 1 1
2 1 2 2 2
3 1 3 3 3
4 2 1 2 3
5 2 2 3 1
6 2 3 1 2
7 3 1 3 2
8 3 2 1 3
9 3 3 2 1

Main Effects (Qualitek-4 Software Output)

Col#/ Factors Level 1 Level 2 Level 3 Level 4 (L2-L1)
1 Indexing -29.6028 -28.8749 -33.9214 0.7278
2 Overlap Area -28.8478 -32.5268 -31.0245 -3.6789
3 Spark Advance -30.6648 -29.7513 -31.9830 0.9135
Quality Characteristic: The Smaller The Better
Data Type : S/N Ratio

A N O V A Table Data Type: S/N Ratio
Col.#/Factor F S V F S' P(%)
1 Indexing 2 44.6445 22.3222 846.5925 44.5917 61.2715
2 Overlap 2 20.5283 10.2641 389.2777 20.4755 28.1345
3 Spark Adv. 2 7.5517 3.7758 143.2037 7.4990 10.3040
Other/Error 2 0.0527 0.026 0.2898
Total: 8 72.7773 100.00%


The expected result (-25.874) shown above represents the S/N ratio of the results of multiple
samples tested at the Optimum Condition. To get an estimate of the performance expressed in
the measured units, the S/N must be transformed back as follows:

MSD = 10
- [(S/N)/10]
(Since S/N is now known)
= 10
- [(-25.874)/10]
= 749.376

But MSD = [ Y
1
2
+ Y
2
2
+ Y
3
2
+ .. .. Y
n
2
]/n (For Smaller is better)
= [ Y
expected
]
2
/ 1 ( assuming same result for all samples)

Or Y
expected
= [ MSD ]
1/2
= [ 749.376 ]
.50
= 27.374 (in terms of the original units)

5.6 Experiment design and Analysis Strategies
The experiment you design will follow one of this paths illustrated in the following diagram
depending on the complexities of your experiment. No matter your control factor design
(INNER ARRAY), you should always attempt to identify and formally treat the noise factors
by using ROBUST DESIGN principles outlined earlier. An of course, regardless of your
design, run your experimental conditions in RANDOM ORDER when possible.

Optimum Table ( Estimated Performance At Optimum Condition)

Col #/ Factors Level Desc. Level# Contribution
1 Indexing O Deg 2 1.9247
2 Overlap Area -30% 1 1.9518
3 Spark Advance 30 Deg 2 1.0484
Total Contribution From All Factors .... 4.9251
Current Grand Average Of Performance ... -30.7998
Expected Result At Optimum Condition ... -25.8747


5.7 Experiment Design Tips

Experiment designs with all factors at one level are likely to be accomplished with standard
orthogonal arrays. These are simpler kinds of designs and do not require any modifications.
On the other hand, if factors are of mixed levels, and there are interactions included in the
experiment, modifications of the array will be necessary. Of course, no matter how your inner
array design is, presence of noise factors in your system requires consideration of formal
treatment of the noise effects by use of the appropriate outer array.

When possible, and if variation is a concern, each trial condition should be repeated multiple
times.

EXPERIMENT DESIGN
ROADMAP
Designs Using
Standard Arrays
Mixed Level &
Interaction Designs
Assigns Factors
Arbitrarily
Modify Columns and
Assign Interacting
Factors Properly
Consider Noise Factors
Determine Repetitions
Run Experiments
in Random Order


Experiment Design Tips

Experiment Requirements Design Practices
# 2-L
fact
ors
3-L
Fact
ors
4-L
Fact
ors
Interacti
ons
Orthogonal array and column assignments
(Solutions are not necessarily unique)

1

2 - 3

-

-

-

L-4, factors assigned to columns arbitrarily
2 2 - - 1 - AxB L-4, factors A in col. 1,B in col. 2 and interaction
AxB in col. 3
3 4 - 7 - - - L-8, factors cols. 1, 2, 4 & 6.
Remaining columns left empty.
4 4 - 1 - L-8, 4-L factor in col. 1, 2-L factors in cols. 4, 5, 6
& 7
5 1 - 4 1 - - L-8, 3-L factor in col. 1, 2-L factors in cols. 4, 5, 6
& 7 as appropriate
6 1 - 4 - 1 - L-8, 3-L factor in col. 1, 2-L factors in cols. 4, 5, 6
& 7 as appropriate
7 3 - 6 - - 1 - AxB L-8, factor A in col.1, B in col. 2 and interaction
AxB in col. 3. Other 2-level factors in the
remaining column.
8 3 - 5 - - 2- AxB
BxC
L-8, Factors A in col. 1, B in col. 2 and C in col.
4. Interactions AxB in col. 3 and BxC in col. 6
9 3 - 4 - - 3 - AxB,
BxC and
CxA
L-8, Factors A in col. 1, B in col. 2 and C in col. 4.
Interactions AxB in col. 3, BxC in col. 6, and CxA
in col. 5.
10 4 - - 3 - AxB,
AxC and
AxD
L-8, Factors A in col. 1, B in col. 2, C in col. 4 and
Din col. 7. Interactions AxB in col. 3 and AxC in
col. 5 and AxD in col. 6
11 4 - 5 - - 2 - AxB
CxD
L-16, factor A in col. 1, B in col. 2 and int. AxB in
col. 3. Factors C in col. 4, D in col. 8 and int. CxD
in col. 12.
12 8 -
11
- - Present
but
ignored.
L-12, assign factors to columns arbitrarily
13 12 -
15
- - - L-16, assign factors to columns arbitrarily
14 - 2 -
4
- - L-9, factors assigned arbitrarily

15 1- 2 2 - - L-9, Dummy treat columns for 2-level factors.

Similarly hundreds of such common experiment designs can be conceived and proposed for everyday use by
experimenters. A large set of such designs are available in the web site:

http://www.wwnet.com/~rkroy/wp-tip.html


5-1: Design experiments to suit the following experimental study objectives. Determine the
appropriate INNER and OUTER arrays.

a. Four 2-level factors and three 2-level noise factors.
b. Five 2-level factors and three 3-level noise factors.
c. Ten 2-level factors and five 2-level noise factors.

5-2: Check the answers that most closely match yours, in the following situations.

a. Why do we need to consider running multiple samples for each trial condition?
Ans: To [ ] obtain better representative performance [ ] reduce trials [ ]
reduce experimental error.
b. While repeating experiments, what was the objective/purpose Taguchi wanted
to satisfy?
Ans: [ ] study more factors [ ] learn about noise factors [ ] design robustness.

5-3: Answer the following questions as they relate to projects you are involved in.

a. What are noise factors? Give an example of noise factors related to a project
you know of. What does robust design mean as applied to your project.
b. Can averages be used to compare two data sets? Discuss the limitations.
c. Why is MSD preferred as a better representative of a data set over the average?
d. What are the advantages of transforming MSD to S/N ratio.
e. If set A: has MSD = 5 and set B: has MSD = 6, Would S/N for set B: be
higher than set A:

5-4: An L-16 OA was used to design an experiment to study fifteen 2-level factors. What
is the degree of freedom of an error factor when:

a.) Each trial condition is tested once.
b.) Each trial condition is repeated 3 times and standard analysis is carried out.
c.) Each trial condition is repeated 5 times and the S/N ratio of the result is
used for analysis.

5-5: What does zero error term (f
e
= 0, S
e
= 0) mean? Check all appropriate boxes.

a. ( ) It indicates a poorly run experiment.
b. ( ) It represents a very well run experiment.
c. ( ) It does not mean that there is no experimental error. It simply means that the
information concerning error sum of squares can not be specifically determined.

REVIEW QUESTIONS

5-6: What are noise factors? Check all appropriate boxes.

a. ( ) Factors which influence the performance but cannot be controlled in
field application/production.
b. ( ) Factors that are difficult to control.
c. ( ) Control factors which are not included in the experiment.
d. ( ) Factors which have strong influence on the outcome.
e. ( ) Environmental factors like temperature, humidity, etc., only.

5-7: A group of process engineers involved in a soldering process optimization study
selected the following factors for investigation:
3 4-level factors
1 2-level factor
3 noise factors to be evaluated at 2-levels each

For evaluation of performance, each soldered sample was to be evaluated by 4 separate
criterion (solder overflow, flexibility, resistance and shape). Determine the following for
the experiment:

a.) OA for inner array
b.) OA for outer array
c.) Total number of samples required
d.) Number of trial conditions
e.) Number of repetitions
f.) Total number of observations/evaluations
g.) Evaluations per result

5-8: Compare S/N ratios of the following two sets of data and determine which set is
more desirable. Consider the Target/nominal value = 12.

SET 1: 11, 9, 12, 10 AND 9. SET 2: 10, 12, 8, 14 AND 16.

Ans: Comparing the S/N ratios, Since -6.627 > -9.038 set 1 is the desired set.
(Please Show All Calculations)

5-9: In an experiment, the estimate of performance at optimum condition is expressed as S/N
= 22.5. If the quality characteristic is "Bigger is Better", Determine the expected performance
in terms of the measured value.


5-10: [Concept: Outer array Design for Noise factors ]

i) What is the smallest size of the outer array? Ans. _______

ii) Which orthogonal array will you use to formally include five 2-level noise factors? Ans.________

iii) For an experiment designed using an L-16 inner array and L-9 outer array, how many samples will you need
to complete the experiment? Ans. ________

iv) An experiment was designed to study seven 2-level factors and three 2-level noise factors as shown below.
Using descriptions of the control factors and the noise factors, determine the following items:

(a) Describe the condition(levels) of the noise for the second sample in trial # 1.

Ans. Toll Holder ______ Coolant ________ Operator __________

(b) Calculate the average effect of Tool Holder type A. Ans. _________

Noise Factors Level 1 Level 2
X: Operator Average Above average
Y: Coolant Oil Water Base
Z: Tool Holder Type A Type B

1 2 3 4
L-8 A B C D E F G
Trial\Col# 1 2 3 4 5 6 7 Results

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

X 1 2 2 1

Y 1 2 1 2

Z 1 1 2 2

Outer Array
Inner Array
Control Factors
Noise Factors
(X, Y, and Z)

5-11 [Concept: Analysis Using S/N Ratio ]

Determine the performance at the optimum condition in terms of S/N ratio of the results when the quality
characteristic is smaller is better.

Trial\Column A B C Results S/N

1 1 1 1 4 6 5 7 -14.98
2 1 2 2 7 8 6 7 -16.94
3 2 1 2 5 6 4 3 -13.32
4 2 2 1 3 2 4 2 - 9.16

trial 1:

MSD = ( 4
2
+ 6
2
+ 5
2
+ 7
2
) / 4 = 31.5
S/N = - 10 Log (MSD) = - 10 Log(31.5) = -14.98

trial 2:
MSD = ( 7
2
+ 8
2
+ 6
2
+ 7
2
) / 4 = S/N = -16.94

trial 3: MSD = ( 5
2
+ 6
2
+ 4
2
+ 3
2
) / 4 = S/N = -13.32

trial 4: MSD = ( 3
2
+ 2
2
+ 4
2
+ 2
2
) / 4 = S/N = -9.16

_
Grand Average T = ( -14.98 - 16.94 - 13.32 - 9.16 )/4 = -13.60

__
A1 = (-14.98 - 16.94)/2 = -15.96

__
A2 = (-13.32 - 9.16)/2 = -11.24

__
B1 = (-14.98 - 13.32)/2 = -14.15

__
B2 = (-16.94 - 9.16)/2 = -13.05

__
C1 = (-14.98 - 9.16)/2 = -12.07

__
C2 = (-16.94 -13.32)/2 =

Optimum Condition : A2 B2 C1

Yopt = -13.06 + (- 11.24 + 13.60) + ( - 13.05 + 13.6) + ( -12.07 + 13.6)

= - 13.06 + 2.36 + 0.55 + 1.53

=

Module - 6

Standard and S/N Analyses with ANOVA
To demonstrate how the results of experiments are analyzed, including the detailed steps in
ANOVA, a number of hypothetical examples are treated in this section. Analysis for single
and multiple sample results by standard method(using averages) are demonstrated first.
Treatment of results of multiple samples using S/N ratios, and under the three different
quality characteristics, are shown next. Practitioners of DOE who wish to manually analyze
results and develop finer understanding of statistical implications, may find this section
helpful.

Topic Objectives:

* Perform standard analysis with single and multiple runs.

* Carry out analysis using S/N ratios.

* Calculate elements of ANOVA for relative percentage influences.

* Revise percentage influence by "Pooling" insignificant factors.

* Determine confidence interval for optimum performance.

6.1 Analysis Guidelines
Depending on the number of samples tested per trial condition, the method of analysis varies.
If your results are from more than two samples per trial(two or more columns of results),
whether or not an outer array is included in the experiments, analysis using S/N ratio is
recommended. If standard analysis is pursued for repeated sample results, be aware that the
conclusions with regard to factor influences and the optimum condition may vary. Trust the
S/N analysis results more.

Module 6: Standard and S/N Analyses with ANOVA Page 6 - 2

Example 8A: Standard analysis with single run

ANALYSIS STRATEGY
Single sample
per trial
Multiple samples
per trial
Perform
Perform
Standard Analysis
S/N Analysis
With or without
Noise Factors and
Outer Array
QC
SMALLER is better
BIGGER is better
NOMINAL is the best

Experiment With Five 2-Level Factors
Col. # Factor Names Level 1 Level 2
1 Factor A A1 A2
2 Factor C C1 C2
3 Interaction Axc N/A
4 Factor B B1 B2
5 Factor D D1 D2
6 Interaction Bxc N/A
7 Factor E E1 E2
Characteristic: Smaller Is Better.


Level Totals and their Average Effects

A
1
= (y
1
+ y
2
+ y
3
+ y
4
) = 42 + 50 + 36 + 45 = 173

A
2
= (y
5
+ y
6
+ y
7
+ y
8
) = 35 + 55 + 30 + 54 = 174.

C
1
= (y
1
+ y
2
+ y
5
+ y
6
) = 42 + 50 + 35 + 55 = 182. or C
1
182 4 45 50 = = / .

C
2
= (y
3
+ y
4
+ y
7
+ y
8
) = 36 + 45 + 30 + 54 = 165 or C
2
165 4 41 25 = = / .

Factor and Interaction Column Effects
L-8 Orthogonal Array

FACTORS: A C AxC B D BxC E
Trial #\ Cols:1 2 3 4 5 6 7 RESULTS(y)
Trial # 1 1 1 1 1 1 1 1 42.00
Trial # 2 1 1 1 2 2 2 2 50.00
Trial # 3 1 2 2 1 1 2 2 36.00
Trial # 4 1 2 2 2 2 1 1 45.00
Trial # 5 2 1 2 1 2 1 2 35.00
Trial # 6 2 1 2 2 1 2 1 55.00
Trial # 7 2 2 1 1 2 2 1 30.00
Trial # 8 2 2 1 2 1 1 2 54.00

_
or A
1
= 173/4 = 43.25
_
or A
2
= 174/4 = 43.50
_
Similarly B
1
= 143 B
1
= 35.75
_
B
2
= 204 B
2
= 51.00
_
D
1
= 187 D
1
= 46.75
_
D
2
= 160 D
2
= 40.00
_
E
1
= 172 E
1
= 43.00

____
(AxC)
1
= (y
1
+ y
2
+ y
7
+ y
8
) = 176 (AxC)
1
= 44.00
____
(AxC)
2
= (y
3
+ y
4
+ y
5
+ y
6
) = 171 (AxC)
2
= 42.75
____
(BxC)
1
= (y
1
+ y
4
+ y
5
+ y
8
) = 176 (BxC)
1
= 44.00
____
(BxC)
2
= (y
2
+ y
3
+ y
6
+ y
7
) = 171 (BxC)
2
= 42.75


Average factor effects can now be plotted and preliminary determination of OPTIMUM CONDITION
can be made.

A1 A2 D1 D2 C1 C2 B1 B2 E1 E2
30
40
50
T= 43.375
_
Plot of Factor Average Effects

For smaller is better characteristic, without considering interaction, the optimum condition
becomes:

A
1
C
2
B
1
D
2
E
1


6.2 Test for Presence of Interaction

Interaction effects of all factors(if any) are mixed with factors which are assigned to the columns. If no
factors are assigned to the designated column, the average effects found in the column (column 3 for
AxC and column 6 for BxC) captures the effect of interaction the relative significance of which is
determined by ANOVA as will be discussed later. For now, whether factor A interacts with factor C or
factor B with factor C, will be found by test of presence of interaction.

Combined factor averages for interaction plots for AxC and BxC

C1 C2
30
40
50
Plots of Interactions
C1 C2
52.5
49.5
38.5
33
B2
46.0
B1
45.0
40.5
42.5
A1
A2

____
A
1
C
1
= y
1
+ y
2
= 42 + 50 = 92 or A
1
C
1
= 92/2 = 46.00
____
A
1
C
2
= y
3
+ y
4
= 36 + 45 = 81 A
1
C
2
= 81/2 = 40.50
____
A
2
C
1
= y
5
+ y
6
= 35 + 55 = 90 A
2
C
1
= 90/2 = 45.00
____
A
2
C
2
= y
7
+ y
8
= 30 + 54 = 84 A
2
C
2
= 84/2 = 42.00

____
B
1
C
1
= y
1
+ y
5
= 42 + 35 = 77 B
1
C
1
= 77/2 = 38.50
____
B
1
C
2
= y
3
+ y
7
= 36 + 30 = 66 B
1
C
2
= 66/2 = 33.00
____
B
2
C
1
= y
2
+ y
6
= 50 + 55 = 105 B
2
C
1
= 105/2= 52.52
____
B
2
C
2
= y
4
+ y
8
= 45 + 54 = 99 B
2
C
2
= 99/2 = 49.50


For plotting purposes, either factor can be chosen as the x-axis. The shape of the plot,
however, may look different, but the conclusions derived will not.

Observations:
Based on the plot above, B and C interact slightly and A and C interact more. ignore BxC,
consider AxC. for the "Smaller is better " QC, A
1
C
2
(40.5) produce the lowest average value.
therefore, A
1
C
2
must be included in the optimum condition. in this case, A
1
C
2
is already
included.

A
1
C
2
B
1
D
2
E
1

Even though interaction is present, how significant is it compared to the factors? Generally
optimum condition is modified only if the interaction is determined to be significant.
Interactions for which a column has been reserved, can be tested for significance from anova.

6.3 ANOVA (Analysis of Variance)

The main objective of ANOVA is to determine relative influence of the factors and
interactions to the variation of the results. The variation of the results is determined by
calculating the deviation of the results from the target.

As defined earlier

Mean Squared Deviation, MSD = ( y
i
- y
o
)
2
/ n , i = 1, 2, .........n

The Total deviation TSD = ( y
i
- y
o
)
2
, i = 1, 2, .........n

Therefore TSD = N x MSD

Again

( y
i
- y
o
)
2
= [ ( y
i
- y
avg
) + ( y
avg
- y
o
) ]
2
, i = 1, 2, .........n
0 Y
0
Y
avg
Yi

Y
i


= [ ( y
i
- y
avg
)
2
+ 2 ( y
i
- y
avg
) ( y
avg
- y
0
) + ( y
avg
- y
o
)
2
]

= n x [ ( y
i
- y
avg
)
2
/n + 0 + ( y
avg
- y
o
)
2

or
( y
i
- y
o
)
2
= n x
2
+ n x m
2

where m = ( y
avg
- y
o
) and

By definition, Variance
2
= [ ( y
i
- y
avg
)
2
/n or [ ( y
i
- y
avg
)
2
= n
2

The variation around the mean of the data is called the total sum of squares, S
T

therefore,

S
T
= [ ( y
i
- y
avg
)
2
/n

that is n
2

= ( y
i
- y
o
)
2
- n m
2

When y
o
= 0, that is the target is at the origin, then m = y
avg
and

S
T

= y
i
2
- n x y
avg

2

= y
i
2
- n x ( T/n )
2

Variation due to
individual data
about the mean of
the data
( S )

Variation of the
mean from the
target
Total variation
about the target
value

ANOVA Statistics

STEP 1. Total of all results:

T = 42 + 50 + 36 + 45 + 35 + 55 + 30 + 54 = 347

STEP 2. Correction factor: C.F = T
2
/ n = 347
2
/ 8 = 15051.125

NOTE: n = TOTAL # OF EXPERIMENTS. i.e. = 8

STEP 3. Total sum of squares as the basis for comparing variations around the mean

Step 4. Factor sum of squares:

S
A
= A
1
2
/N
A1
+ A
2
2
/N
A2
- C.F

= 173
2
/4 + 174
2
/4 - 15051.125 = 0.125

S
B
= B
1
2
/N
B1
+ B
1
2
/N
B2
- C.F

= 143
2
/4 + 204
2
/4 - 15051.125 = 465.125

Similarly
S
C
= 182
2
4 + 165
2
/4 - 15051.125 = 36.125

S
D
= 187
2
/4 + 160
2
/4 - 15051.125 = 91.125

S
E
= 172
2
/4 + 175
2
/4 - 15051.125 = 1.125

n __
S
T
= ( Y
i
+ Y ]
2 Which can be reduced to the following form
i=1

8
S
T
= Y
i
2
- C.F
i=1

= ( 42
2
+ 50
2
+ 36
2
+ ..... + 54
2
) - 15051.125 = 599.88


S
AxC
= (AxC)
1
2
/N
(AxC)1
+ (AxC)
2
2
/N
(AxC)2
- C.F

= 176
2 /
4 + 171
2
/4 - 15051.125 = 3.125

S
BxC
= 176
2
/4 + 171
2
/4 - 15051.125 = 3.125

S
e
= S
T
- ( S
A
+ S
B
+ S
C
+ S
D
+ S
E
+S
AxC
+ S
BxC
)

= 599.88 - 599.88 = 0

The errors sums of squares, S
e,
will always equal to zero when all columns of the array are
occupied and all experiment degrees of freedom of the experiment are used up.

Notations:
N
A1
= Total # of experiments in which factor A
1
is present
N
B1
= Total # of experiments in which factor B
1
is present
A
1
= Total of results (y
i
) which include factor A
1

B
1
= Total of results (y
i
) which include factor B
1

etc.

These short formulas for sums of squares used above apply only for 2 - level factors.

SPECIAL CASE: For 2-level factors only

(A
1
- A
2
)
2
(173 - 174)
2

S
A
= = = 0.125
(N
A1
+ N
A2
) (4 + 4)

Similarly
(143 - 204)
2
S
B
= = 465.125
(4 + 4 )

(182 - 165)
2

S
C
= = 36.125
(4 + 4)

S
D
= 91.125, S
E
= 1.125

S
AxC
= 3.125, S
BxC
= 3.125


Step 5. Experiment Degrees of Freedom

f
T
= n - 1 = 8 - 1 = 7 (Total number of results - 1)

Step 6. factor degrees of freedom

f
A
= (# levels of factor A) - 1 = 2 - 1 = 1

f
B
= (# levels of factor B) - 1 = 2 - 1 = 1

Similarly, f
C
= 1, f
D
= 1, f
E
= 1

f
AxC
= f
A
x f
C
= 1 x 1 = 1 f
BxC
= f
B
x f
C
= 1 x 1 = 1

DOF of error term, f
e
= f
T
- (f
A
+ f
B
+ f
C
+ f
D
+ f
E
+ f
AxC
+ f
BxC
) = 0

(If Error DOF is zero, the error sums of squares must also equal to zero)

Step 7. Mean Squares (Variance)

V
A
= S
A
/f
A
= 0.125 / 1 = 0.125 V
B
= S
B
/f
B
= 465.125 / 1 = 465.125

V
C
= S
C
/f
C
= 36.125 / 1 = 36.125 V
D
= S
D
/f
D
= 91.125 / 1 = 91.125

V
E
= S
E
/f
E
= 1.125 / 1 = 1.125 V
AxC
= S
AxC
/f
A
= 3.125 / 1 = 3.125

Variance for the error term, V
e
= S
e
/f
e
= 0/0 = INDETERMINATE.

When the variance of error term (V
e
) is zero , variance ratios and pure sum of squares (S') of
the column effects cannot be calculated at this time.

Step 8. Percent contribution

P
A
= S
A
/S
T
= 0.125/ 599.88 = 0.02% P
B
= S
B
/S
T
= 465.125/ 599.88 = 77.54%

P
C
= S
C
/S
T
= 36.125/ 599.88 = 6.02% P
D
= S
D
/S
T
= 91.125/ 599.88 = 15.20%

P
E
= S
E
/S
T
= 1.125/ 599.88 = 0.19% P
AxC
= S
AxC
/S
T
= 3.125/ 599.88 = 0.52%

P
BxC
= S
BxC
/S
T
= 3.125/ 599.88 = 0.52%

Pe = 0 Since the error sums of squares is absent.

A N O V A Table

Pooling:
Pooling is a common practice of disregarding the column influence(P%) which are
considered insignificant. Whether a column effect is insignificant or not, is determined by
TEST FOR SIGNIFICANCE. But the test for significance can only be performed when the
error DOF is non-zero. POOLING however, should always be attempted even when
significance test cannot be performed. A common practice is to start pooling with the
weakest factors.

Pool factors A, E, AxC and BxC as their contributions are judged to be negligible (less than
1%)

Step 1. Sum of squares of error

S
e
= S
T
- (S
B
+ S
C
+ S
D
) = 599.88 - (592.4) = 7.5 (neglecting all terms pooled)

Note: S
e
will now be equal to (S
A
+ S
E
+ S
AxC
+ S
BxC
) i.e. Sum of all factors pooled.

Step 2. Degree of freedom of error term

f
e
= f
T
- (f
B
+ f
C
+ f
D
) = 7 - 3 = 4

Step 3. Variance of error term

V
e
= S
e
/f
e
= 7.5 / 4 = 1.875


Step 4. F-Ratios for significant factors

F
C
= V
C
/ V
e
= 36.125 / 1.875 = 19.267

F
B
= V
B
/ V
e
= 465.125 / 1.875 = 248.067

F
D
= V
D
/ V
e
= 91.125 / 1.875 = 48.600

Step 5: Pure sum of squares(S') for significant factors

S
C
' = S
C
- (V
e
x f
C
) = 36.125 - (1.875 x 1) = 34.25

S
B
' = S
B
- (V
e
x f
B
) = 465.125 - (1.875 x 1) = 463.25

S
D
' = S
D
- (V
e
x f
D
) = 91.125 - (1.875 x 1) = 89.25

Step 6: Percent contribution

P
C
= S
C
'/ S
T
= 34.25/599.88 = 5.71%

P
B
= S
B
'/ S
T
= 463.25/599.88 = 77.22%

P
D
= S
D
'/ S
T
= 89.25/599.88 = 14.88%

P
e
= 100% - (P
C
+ P
B
+ P
D
) = 2.19%

Revised A N O V A Table


6.4 Pooling Technique
To examine the method for pooling a factor, consider factor C which has 5.71% influence and may be
considered for pooling. If factor C were arbitrarily pooled, the ANOVA terms will be revised as
follows.

S
e
= S
T
- (S
B
+ S
D
) = 599.88 - (556.25) = 43.625

f
e
= f
T
- (f
B
+ f
D
) = 7 - 2 = 5

V
e
= S
e
/ f
e
= 43.625 / 5 = 8.725

F
B
= V
B
/ V
e
= 465.125 / 8.725 = 53.309

F
D
= V
D
/ V
e
= 91.125 / 8.725 = 10.444

S
B
' = S
B
- (V
e
x f
B
) = 465.125 - (8.725 x 1) = 456.40

S
D
' = S
D
- (V
e
x f
D
) = 91.125 - (8.725 x 1) = 82.40

P
B
= S
B
'/ S
T
= (456.40 * 100)/599.88 = 76.08

P
D
= S
D
'/ S
T
= ( 82.40 * 100)/599.88 = 13.74

P
e
= 100% - (P
B
+ P
D
) = 10.18

Is factor C significant? Should we pool factor C ?

To study the possibility of pooling, is to perform test of significance for factor C. The
significance test is carried out by comparing the experiment F-ratio with the standard table
value for a desired Confidence Level (subjective: 90, 95 or 99% commonly used).

From the ANOVA table F
C
= 19.267

From The F-TABLE, Find F Value At n
1
= DOF OF Factor C = 1
n
2
= DOF Of Error Term = 4 At A Confidence Level (say 95%) .

F (table) = 7.7086

Since F
C
Is Greater Than the F(table), Factor C should not be pooled.

Recommendations: Create non-zero DOF for the error term by pooling the weaker
factors. Test for significance first, then pool. Attempt to pool until the DOF for the error term
is about half the experiment DOF.


The F-ratios are contained in F-Table developed by R. A Fisher and are available in most
common books on statistical science. There are tables for most commonly used confidence
levels (90%, 95%, 99%, etc. are common). F-Ratios are arranged in a two dimensional table
corresponding to a degree freedom for the factor (horizontally toward right) and the degree of
freedom for the error term (vertically down). In some text, the F-Tables are specified by level
of significance instead of the confidence level. The level of significance (symbol: Greek letter
Alpha) is complementary to the level of confidence. That is to say 95% confidence level is
equivalent to 5% level of significance.

Sample Table Readings at 95% confidence level:
F (3, 8 ) = 4.0661(3rd column, 8th row) F (2,5) = 5.7862 ( 2nd column ,5th row)

To perform test of significance at other confidence levels, the corresponding F-Table must be
secured first. F-Table at other confidence levels may be found in the reference section of this
seminar handout.

Qualitek-4
Pooling
Method

F-TABLE (95%)
n
2\
n
1
1 2 3 4 5 6 7 ..etc.
1 161.45 199.50 215.71 224.58 230.16 233.99 236.77
2 18.513 19.000 19.164 19.247 19.296 19.330 19.353
3 10.128 9.5521 9.2766 9.1172 9.0135 8.9406 8.8868
4 7.7086 6.9443 6.5914 6.3883 6.2560 6.1631 6.0942
5 6.6079 5.7862 5.4095 5.1922 4.3874 4.2839 4.2066
6 5.9874 5.1433 4.7571 4.5337 4.3874 4.2839 4.2066
7 5.5914 4.7374 4.3468 4.1203 3.9725 3.8660 3.7870
8 5.3277 434590 4.0661 3.8378 3.6875 3.5806 3.5005
9 5.1174 4.2565 3.7626 3.6331 3.4817 3.3738 3.2927
10 4.9646 4.1028 3.7083 3.4780 3.3258 3.2172 3.1355
11 4.8443 3.9823 3.5874 3.3567 3.2039 3.0946 3.0123
12 4.7472 3.8853 3.4903 3.2592 3.1059 2.9961 2.9134
13 4.6672 3.8056 3.4105 3.1791 3.0254 2.9153 2.8321
etc. .. ..

6.5 Confidence Interval of Main Effects

Confidence intervals are calculated after pooling and only for factors not pooled.

For factor C at level C
1

n
2
= 4 N
e
= 8/(1+1) = 4 F(1,4) = 7.7086 AT 95% Confidence Level

Which gives C.I. = +/- 1.9034 AT 95% Confidence Level.

i.e the main effect of factor C at level C
1
will be 45.5 +/- 1.9034
in 95 out of every 100 experiments conducted.

C1 C2
30
40
50
45.5
41.25
Main Effect With Confidence Interval
Factor C

_______________
The confidence interval (C.I.) = +/- \/ (F(1,n
2
) x V
e
/N
e
)

where f(1,n
2
) = F-value from f table at dof 1 & n
2
at a required confidence level.
n
2
= degrees of freedom of the error term.
V
e
= variance of error term (from anova)

n
e
= effective # of replications

Total # of results ( or # of S/N ratios)

DOF of mean (=1 always) + DOF of the factor
=

Estimated Performance at Optimum Condition

Since factors B, B and D are considered significant, the performance at optimum condition
will be estimated using only these three factors.

Note: The optimum condition for smaller is better quality characteristic is B
1
D
2
C
2
.

Optimum Condition and Confidence Interval (C.I)

_ _ _ _ _ _ _
Y
opt
= T + (B
1
- T) + (D
2
- T) + (C
2
- T)

= 43.375 + (41.25 - 43.375) + (40 - 43.375) + (33 - 43.375) = 30.25

__
Grand average of performance: T = 347 / 8 = 43.375

C.I. Formula:
______________
THE CONFIDENCE INTERVAL (C.I.) = + \/ [F( 1,n
2
) x V
e
/N
e
]

Where F(1,n
2
) = F - Value from the F-Table for factor DOF & Error DOF
at the desired Confidence Level
V
e
= Variance of the Error Term(from ANOVA)
N
e
= Effective Number of Replication

Total number of Results or S/N
=
DOF of Mean(always=1) + DOF of all factors included
estimating the Mean performance at Optimum Condition.


The confidence interval applicable to the estimated performance at optimum condition,
is calculated as follows.

n
2
= 4 N
e
= 8/(1+3) = 2 F(1,4) = 7.7086 at 95% confidence level

C.I. = +/- 2.2039 at 95% confidence level

The confidence interval calculated earlier assumes a large number of repetitions. when only a
finite number of repetitions are planned, the formula for confidence interval is slightly
modified.

Where F(n
1
,n
2
) = Computed value of F
with n
1
= 1, n
2
= error DOF, at a desired Confidence Level

V
e
= Error Variance N
e
= Effective number of replications.

N
r
= Number of repetitions

With F = 4.0674 n
1
= 1 n
2
= 5 V
e
= 8.725

N
e
= 2.666667 N
r
= 3

Confidence interval: C.I. = 30.25 +/- 5.0092 (at 90% Confidence level)

6.6 Discussions on Some Key Issues

When should we accept the ANOVA results (%)?

Test for significance provides the criteria for making such decisions. The F-tests are used to
statistically determine whether the constituents (factor sums of square) of the total sums of
squares are significant with respect to the components in the error variance. Generally, the
F-test is made with the use of an F-table (named after the originator of design of experiments,
R. A. Fisher). The F-tables contain reference values for F(
alpha)
(i, j), where i, j are the
degrees of freedom of the numerator and denominator of the respective F-ratios and (alpha),
is the level of significance. If (alpha) - .05, the confidence level = 1 - (alpha) = .95
______________________________________
Confidence Interval: C.I. = +/- \/ (( F(n
1
,n
2
) x V
e
x(N
e
+ N
r
)) / N
e
*N
r
)


For example, for factor A:
if F
A
= V
A
/ V
e
> table value

if affirms the existence of influence of factor A. Put another way, if

F
A
= V
A
/ V
e
obtained from the actual data is larger than the F-table value, an incorrect
conclusion may result a maximum of 5 (for (alpha) = .05) times out of 100. The number 5%,
is called the level of significance. (5% level of significance = 100 - 5 = 95% level of
confidence).

Generally, factors which are determined insignificant, are pooled. However, do not ignore
(pool) those insignificant factors whose percent contributions are as large as its influence on
results may be large.
Error Term (Unaccounted factors)

What does absence of error DOF and percent influence mean?
When degrees of freedom of error is absent, f
e
= 0 (which results in P
e
= 0), does not mean
that there is no error, but that the information concerning factors contributing to the error
term cannot be specifically determined. In these instances, factors with smaller factorial
effects are often pooled together to produce an estimate of error. Generally, pool until f
e
=
1/2 f
T

Error term - What is it?
Error sums of squares reflect variation in results due to factors other that those included in
the experiment.
Sources:
- Experimental set up
- Noise factors
- Control factors not included in the study.

All other conditions being the same, the magnitude of error variance will also depend on the
quality characteristic used. For example, durability life, instead of deflection, is used as the
QC for comparing two mechanical parts, the results are expected to show much higher error
contribution.

What to make of high values of error term (80%, 90%, etc.)

A high value of error term is not alone a reflection of the "quality" of the experiment. If P
e
is
high, it does not mean the experiment is bad or that the factorial effects are not reliable.
Instead such a conclusion must await test of significance.


Why should we pool?
Pooling enlarges the error DOF and its contribution. When the error DOF is greater, the
confidence level on the factors is lower. That makes the results look bad. So, why pool?

* To make a conservative estimate. The higher probability allows us to identify true
contributors.
* To reduce the chance of identifying less significant factors as being significant.

When you must repeat?
Trials must be repeated if the results are expected to vary from sample to sample. Remember
to always repeat when:

* Cost of repetition/sample is small
* Variability in results from sample to sample is more than that between two
trial runs.

How to determine the number of repetitions:

* If noise factors are included in the design as factors in the outer array, the size of
the outer array determines the number of repetitions.

* If noise factor is random or not included in the design, the number of
repetitions is determined subjectively.

- Select 3 or more

- Go for higher number if cost of an additional sample is small

- Choose a smaller number if cost of an additional sample is substantial

6.7 Example 8B: Standard analysis with multiple runs

QC= SMALLER IS BETTER (Same design as Ex. 8A)
Trial \ Results R1 R2 R3 Avg
1 38.00 42.00 46.00 42.00
2 45.00 50.00 55.00 50.00
3 38.00 36.00 34.00 36.00
4 55.00 45.00 35.00 45.00
5 30.00 35.00 40.00 35.00
6 65.00 55.00 45.00 55.00
7 40.00 30.00 20.00 30.00
8 58.00 54.00 50.00 54.00


Standard analysis is carried out the average values of the trial results. the variation of
the results in the same trial does not influence the computation of main effects.

The Main Effects

ANOVA

DOF = 8 x 3 - 1 = 23

B
1
= (38.0 + 42.0 + 46.0) + (38.0 + 36.0 + 34.0)
+ (30.0 + 35.0 + 40.0) + (40.0 + 30.0 + 20.0) = 429

Note: B
1
is obtained from results of trial conditions 1, 3, 5 & 7.

B
2
= (45.0 + 50.0 + 55.0) + (55.0 + 45.0 + 35.0)
+ (65.0 + 55.0 + 45.0)+ (58.0 + 54.0 + 50.0) = 612

The formula used below to calculate sum of squares in only for 2-level factors.

V
B
= S
B
/ f
B
= 1395.375 / 1 = 1395.375

F
B
= V
B
/ V
e
= 1395.375/39.72 = 35.126

( B
1
- B
2
)
2
(429 - 612)
2

S
B
= = = 1395.375
(N
B1
+ N
B2
)
2
24


S
B
' = S
B
- V
e
x f
B
= 1395.275 - 39.72 = 1355.65

P
B
= 100 x S
B
'/S
T
= 100 x 1355.65/2571.63 = 52.70%

Note : V
e
& V
T
used are from ANOVA table shown on next page.

A N O V A TABLE (Qualitek-4 screen)

Optimum Condition And Performance


6.8 Example 9A: S/N analysis - Nominal is the best (same design as ex. 8a)

Target value, Y
o
= 40.

For trial # 1,

MSD = [ ( y
1
- y
o
)
2
+ (y
2
- y
o
)
2
...... ]/n

MSD = [ ( 38 - 40 )
2
+ ( 42 - 40 )
2
+ ( 46 - 40 )
2
] /3

= 14.667

S/N = - 10 LOG
10
(MSD) = -10 LOG
10
(14.667) = -11.67

Note: In case of analysis using S/N, regardless of the quality characteristic, optimum condition is determined based on the
higher value of main effect.

ANOVA Calculations

Trial \ Rep. R1 R2 R3 S/N
1 38.00 42.00 46.00 -11.67
2 45.00 50.00 55.00 -20.67
3 38.00 36.00 34.00 -12.72
4 55.00 45.00 35.00 -19.63
5 30.00 35.00 40.00 -16.21
6 65.00 55.00 45.00 -24.65
7 40.00 30.00 20.00 -22.22
8 58.00 54.00 50.00 -23.16

The Main Effects
Col.# FACTOR NAMES LEVEL 1 LEVEL 2 (L2 - L1)
1 FACTOR A -16.17 -21.55 -5.39
2 FACTOR C -18.29 -19.43 -1.14
3 INTERACTION AxC -19.43 -18.29 -1.13
4 FACTOR B -15.70 -22.02 -6.33
5 FACTOR D -18.04 -19.68 -1.64
6 INTERACTION BxC -17.66 -20.06 -2.41
7 FACTOR E -19.54 -18.18 1.35

The Main Effects
_
A
1
= ( -11.67 + (-20.67) + (-12.72) + (-19.63)) / 4 = -16.17
_
A
2
= ( -16.21 + (-24.65) + (-22.22) + (-23.16)) / 4 = -21.55, etc.


f
T
= 8 - 1 = 7 ( Each S/N ratio counted as 1 DOF regardless
of number of repetitions)

C.F = T
2
/n = [(-11.67) + (-20.67) + (-12.72) + (-19.63)
+ (-16.21) + (-24.65) + (-22.22) + (-23.16)] /8 = 2847.4831

Note : 1. The original results are not considered.
2. T - is calculated from summing the S/N column.
3. n - the total number of S/N values.

S
T
= [(-11.67)
2
+ (-20.67)
2
+ (-12.72)
2
+ (-19.63)
2

+ (-16.21)
2
+ (-24.65)
2
+ (-22.22)
2
+ (-23.16)
2
] - C.F = 163.77

S
A
= A
1
2
/ n
A1
+ A
2
2
/ n
A2
- C.F (General Formula)

= 1046 + 1859.33 - 2847.4831

= 58.063

V
A
= S
A
/f
A
= 57.85 /1 = 58.063 S
A
' = S
A
- f
A
x V
e
= 57.85 - 1 x 3.53 = 54.32

P
A
= 100 x S
A
'/ S
T
= 100 x 54.32/ 163.77 = 33.2%

Note : Value of V
e
above is taken from ANOVA Table shown below.

A N O V A Table
Cols: FACTORS f S V F S' P
1 FACTOR A 1 58.063 58.063 16.44 54.53 33.30
2 FACTOR C (1) (2.56) POOLED
3 INT. AxC (1) (2.56) POOLED
4 FACTOR B 1 80.029 80.029 22.659 76.50 46.71
5 FACTOR D (1) (5.33) POOLED
6 INT. BxC 1 11.548 11.548 3.270 8.02 4.87
7 FACTOR E (1) (3.680) POOLED
ALL OTHER/ERROR 4 14.130 3.53 15.10
TOTAL: 7 163.77 100.00%


Note : 1. The Expected result at optimum condition is the S/N Ratio
of the results.

2. Since the S/N = -10 log
10
(MSD)

MSD = 10
- (S/N /10 )
= 15.142

i.e. The results at optimum condition are expected to produce
________
Y = 40 +/- \/(15.142)

= 40 +/- 3.891[6-28]

6. 9 Example 9B: S/N Analysis - Smaller is better
(Same experiment as example 8A)

MSD = ( Y
1
2
+ Y
2
2
+ Y
3
2
+ ......) /n

For Trial # 1

MSD = ( 38
2
+ 42
2
+ 46
2
) / 3

= 1774.666

S/N = -10 LOG
10
(MSD)

= -32.50

Trial \ Rep. R1 R2 R3 S/N
1 38.00 42.00 46.00 -32.50
2 45.00 50.00 55.00 -34.01
3 38.00 36.00 34.00 -31.14
4 55.00 45.00 35.00 -33.21
5 30.00 35.00 40.00 -30.95
6 65.00 55.00 20.00 -29.86
8 58.00 54.00 50.00 -34.67


FACTOR DESCRIPTION LEVEL DESCRIPTION LEVEL # CONTRIBUTION
FACTOR A A1 1 2.6940
FACTOR B B1 1 3.1628
INTERACT 2 2 x 4 1 1.2014
TOTAL CONTRIBUTION FROM ALL FACTORS ... 7.0583
CURRENT GRAND AVERAGE OF PERFORMANCE ... -18.8604
EXPECTED RESULT AT OPTIMUM CONDITION ... -11.8021


A N O V A Table
Cols: FACTORS f S V F S' P
1 FACTOR A (1) (0.03) POOLED
2 FACTOR C 1 1.518 1.518 16.23 1.42 5.85
3 INT. AxC (1) (0.09) POOLED
4 FACTOR B 1 19.097 19.097 204.230 19.00 78.04
5 FACTOR D 1 3.361 3.361 35.948 3.26 13.42
6 INT. BxC (1) (0.25) POOLED
7 FACTOR E (1) (0.01) POOLED
OTHER/ERROR 4 0.37 0.09 2.69
TOTAL: 7 24.35 100.00%

Main Effects
Col.# FACTOR NAMES LEVEL 1 LEVEL 2 (L2-L1)
1 FACTOR A -32.71 -32.59 0.12
2 FACTOR C -33.09 -32.21 0.87
3 INT. AxC -32.75 -32.55 0.20
4 FACTOR B -31.10 -34.19 -3.09
5 FACTOR D -33.30 -32.00 1.29
6 INT. BxC -32.82 -32.47 0.35
7 FACTOR E -32.61 -32.69 -0.08


FACTOR DESCRIPTION LEVEL DESC. LEVEL # CONTRIBUTION
FACTOR C C2 2 0.4356
FACTOR B B1 1 1.5450
FACTOR D D2 2 0.6482
TOTAL CONTRIBUTION FROM ALL FACTORS .... 2.6288
CURRENT GRAND AVERAGE OF PERFORMANCE ....-32.6497
EXPECTED RESULT AT OPTIMUM CONDITION ... -30.0208

Calculation of Main Effects
_
A
1
= ( -32.50 - 34.01 - 31.14 - 33.21 ) /4 = -32.71
_
A
2
= ( -30.95 - 34.91 - 29.86 - 34.67 ) /4 = -32.59 etc.


Expected value in terms of the original units of measurements:

S/N = -30.020
or MSD = 10
- [ - 30.020/10 ]
= 1004.8

But MSD = ( Y
1
2
+ Y
2
2
+ Y
3
2
+ ......) /n = [ Y
expected
]
2
/ 1

or
Y
expected
= [MSD]
1/2
= [1004.8 ]
0.5
= 31.698

6.10 Example 9C: S/N analysis - Bigger is better (Same example 8B)

MSD = [ 1/y
1
2
+ 1/y
2
2
+ ........ ] / n

For Trial # 1
MSD = ( 1/38
2
+ 1/42
2
+ 1/46
2
) / 3 = 5.773 x 10

S/N = - 10 Log
10
(MSD) = 32.38

Results and the S/N ratios


ANOVA


Main Effects

Confidence Interval at Expected Performance

The expected limits in the original units of measurement can be calculated as follows:

S/N = 34.2975, 35.349, and 36.4005

When
S/N = 35.349
Yexp. = Sqr. Root { 1/( 10
- [S/N)/10]
)} = Sqr. Root ( 1/ 10
-3.5349
) = 58.54

S/N = 34.2975
- [S/N)/10]
)} = Sqr. Root ( 1/ 10
-3.42975
) = 51.86

S/N = 36.4005
- [S/N)/10]
)} = Sqr. Root ( 1/ 10
-3.64005
) = 66.07

Expected Range of Results (at 90% C. I.)

Lower limit: S/N = 34.2975 Result = 51.86
Mean Value: S/N = 35.349 Result = 58.54
Upper Limit: S/N = 36.4005 Result = 66.07


6.11 Understanding and Interpreting Error Terms

When influence of error term is high, it implies that effect of :
1. Factors not included in the experiment
2. Noise factors
3. Experimental/setup error is high

Separating the individual contribution from the error factor which has the combined
effects of the above three, is a difficult task and require further investigations. The
original results of the experiments contain much of the clues.

I. High DOF
High% Influence
expect results to vary appreciably
at the same trial condition.

II. Low/High DOF
Low % Influence
smaller sample to sample variation

III. Low DOF
high % influence (wide variation)

Trials
Array
Runs

Runs
Trials
Array

Trials
Array
Runs

All Other/Error Term in ANOVA

DOF % INFLUENCE IMPLICATIONS
ZERO Must be zero No conclusion can be drawn
NON ZERO LOW Less influence from unknown causes
NON ZERO HIGH Need to focus on variability


Key Decisions

1. Is the experiment satisfactory?
When the experiments are carried out and the results analyzed, determine the OPTIMUM
condition, the performance at this condition, and the CONFIDENCE INTERVAL(C.I.).
Run a few samples at the optimum condition (Confirmation tests). If the results fall
within the confidence interval, you have good results.

The confirmation of the results is independent of the error term in your experiment.
Larger error term(% influence) will produce wider C.I., accounting for larger variation in
the expected results at the optimum condition.

What if the confirmation is unsatisfactory?
Most likely cause for this are the interactions among the factors included in the
experiments. If columns have been reserved for interactions, examine the significance of
interactions and adjust factor levels accordingly. If interaction actions were not
considered, you can still test for presence of interaction. Select factor levels based on
severity of presence of interactions of a few pairs. If this does not work, re-run
experiments considering few suspected interactions.

What to do about Low or high error term?
- If error % is small (smaller than the least significant factor), it means that the factors selected for study
indeed are the important factors.

- If error % is large (as big or bigger than the most significant factor), it does not mean the experiment is
inconclusive, it calls for more analysis. Pay special attention to the noise factors and go for robust design.

When should you consider the need to repeat the experiments from the start?
suspicion that the control factors not included in the experiment may cause such high values.


6-1: What does it mean when ANOVA indicates a high percentage of influence
(P
e
= 30%, 60% or 80%; fe > 0)? Check all appropriate answers.
a. ( ) It shows the experiment was poorly conducted and should be repeated.
b. ( ) It does not necessarily indicate that the experiment was bad.
c. ( ) It indicates that the performance of the product/process is largely
influenced by factors not included in the study.
d. ( ) It means that the results at the same trial condition varies significantly
when repeated.
e. ( ) It means that the results vary extensively from trial condition to trial
condition.

6-2: How do you determine when and what factors to `pool'? Check all correct answers.
a. ( ) Pool all factors which fail the test for significance (say 95% confidence).
b. ( ) Start pooling with the factor that has the least % influence (P
e
).
c. ( ) There is no general % guideline for pooling. It depends on several factors.
d. ( ) You should always attempt to pool if error DOF = 0.
e. ( ) In general, try to pool until the error DOF is about half the total DOF.

6-3: Why do we perform ANOVA? Check all appropriate answers. To determine:
a. ( ) Optimum performance.
b. ( ) Relative influence of individual factor on the variation of results.
c. ( ) Significance of influence of individual factor.
d. ( ) Best design condition.
e. ( ) Confidence level of expected performance at optimum.
f. ( ) Confidence level of Main Effect of a factor.
g. ( ) Performance at conditions other than those covered by the experiment.

6-4: When should you perform analysis using S/N?

6-5: If you have multiple samples (results) per trial condition, could you perform standard
analysis?

6-6: If you analyze the experiment using both S/N and Std. analysis, should you arrive at
the same conclusions?

6-7: The three basic steps in analysis are :
* Average column effects
* ANOVA
* Optimum performance calculations

Which steps must you follow to determine optimum condition?
Review Questions

6-8: Refer to the ANOVA table below to answer the following questions:

a. What is the total degrees of freedom for this experiment? ___________.

b. What influence do factors other than those studied
have on the final result? _______________________.

c. If you wanted to study only those factors that contribute more than 10% to the
outcome, which factors would you pool?_____________________________

d. Determine the effective number of replication, N
e
for factor EGG.

e. Calculate the Confidence interval(C.I.) at 95% confidence level for factor EGG.
Use 95% F-Table for F ratio number.

f. If factors CHIPS and SUGAR were pooled, what will be the degrees of freedom
of the error term?

g. Is this ANOVA due to standard analysis or S/N analysis of results?

h. If S/N analysis is performed, what will be the degrees of freedom of the error
before any factor is pooled.

i. What does the percentage influence of the error term indicate? Discuss how this
quantity can influence your follow up experimental strategies.

j. If the trial conditions were repeated 5 times (instead of 2) and the results analyzed by
standard analysis, what will be the degrees of freedom of the error term?

ANOVA
Factor Desc. f S V F P
EGG 1 33.063 33.063 25.190 35.11
CHIPS 1 1.563 1.563 1.190 0.28
SUGAR 1 0.563 0.563 0.429 0.00
BUTTER 1 14.063 14.063 10.714 14.10
FLOUR 1 5.063 5.063 3.857 4.15
TIME 1 18.063 18.063 13.762 18.52
NUTS 1 7.563 7.563 5.762 6.91
ALL OTHER/ ERROR 8 10.500 1.310 20.94 20.94
TOTAL 15 90.440 100.00%
(Experiment used an L-8 array and 2 samples/trial conditions)


6-9. [Concept: ANOVA Statistics ]

Calculate the ANOVA statistics for the experimental results shown below.

i) Relative influence of factor A

Trial\Column A B C Results
1 1 1 1 12
2 1 2 2 10
3 2 1 2 15
4 2 2 1 16
Average Effects
L1 L2
A 11 15.5
B 13.5 12.5
C 14 12.5
_
T = ( 12 + 10 + 15 + 16 ) = 53 or T = 13.25 C.F. = 53 x 53 /4 =702.25

A
1
= (12 + 10 ) = 22 A
2
= (15 +16) = 31

S
T
= ( 12
2
+ 10
2
+ 15
2
+ 16
2
) - 702.25 = 22.75

S
A
= ( 22 - 31 )
2
/( 2 + 2) = 20.25 DOF of A, f
A
= 2 - 1 = 1

V
A
= P
A
= = 89.01 %

Col Factor f S V F S P(%)
1 A 1 20.25 20.25 ---- 20.25 89.01
2 B 1 0.25 0.25 ---- 0.25 1.09
3 C 1 2.25 2.25 ---- 2.25 9.89
Error factor 0 -----
Total 3 22.75 100
ANOVA

6-10. [Concept: Confidence Interval ]

For an experiment whose POOLED ANOVA is as shown above the
Optimum Condition is A2 B1(Bigger is Better QC) and Y opt = 16.25.

i) Determine Effective number of Replication Ans. ________(1.33)

ii) If F(1.1) = 1.60 at 60% confidence level,
determine the Confidence Interval. _______________
C. I. = \/ (1.6 x )/1.33 = +/- 0.548

Col Factor f S V F S P(%)
1 A 1 20.25 20.25 81 20 87.91
2 B (1) POOLED
3 C 1 2.25 2.25 9 2.0 8.79
Error factor 1 .25 .25 3.30
Total 3 22.75 100
POOLED ANOVA

Module - 7

Loss Function
Taguchis Loss Function is a mathematical formulation which attempts to quantify effect of
poor quality in terms of monetary units. In his formulation, lack of quality is designated by
variation around the target value. Since S/N ratio represents the status in respect to the
variation, using the Loss Function, any improvement in S/N, like in the performance at the
Optimum condition, can be translated in to loss in dollar amounts. The difference between the
loss before and after the experiments can be shown to produce savings. This module describes
how the Loss Function is used to quantify savings in dollars due to improvement in the design.

Topics objectives:

* Understand the need for loss functions.

* Define customer tolerance.

* Calculate $ loss per product at a design condition.

* Determine manufacturer and supplier tolerance.

7.1 Taguchi Loss Function
(An analytical model for quantification of design improvement)

Comparison of Old and New ways
Situation Old way New Way
* Loss when Y is at target zero$ zero$
* Loss when Y = LCL+ or UCL- zero$ cost of rejection
* Loss when Y = LCL+ or UCL- zero$ cost of rejection

Module 7: Loss Function Page 7 - 2

Sony TV Example

Questions were raised by studying Sony TV production and disproportional amount of
warranty returns.

- There are warranty returns even if all parts are made within specifications
- How to compare two batches, both made within the limits
- Loss can't be a step function, but a continuous one.
y
Target Y
o
$Loss
Lower Limit
Upper Limit
Old Concept
No Loss
$Loss
< = y = >
Target Y
o
< = y = >
$Loss
No Loss at Target
Loss begins as soon as
we deviate from the target
New Concept
LCL
UCL
Loss = Cost of Rejection


Loss is expressed as additional cost of production (overhead) for a single product.

L = K (y - y
o
)
2
$ Per unit product

For multiple product the loss equation becomes:

L = K (MSD) $/Unit product

Where
L = loss/unit product in $
k = a constant
y
o
= target value
y = performance measure

Direct calculation of loss from cost of rejection (y outside limits) assuming

Monthly production = 2,000 units

Monthly rejection = 50 per unit

Cost of rejection = $8 per unit

Therefore:

Loss = (50 x $ 8 )/2000 = $ 0.50/unit


Example 10: Savings Due to Improved Design

Loss Before Experiment/Current Design

L = K (MSD) = 163.265 x .0475) = 7.75 NOTE: S/N = 13.23307
OR MSD = .0475
Loss After Experiment/New Design

L = K (MSD) = 163.265 x .0145) = 2.36 NOTE: S/N = 18.38631
OR MSD = .0145

Savings Per Month = 7.75 - 2.36 = $ 5.39/PER PRODUCT

or

Total Savings Per Month = 1500 x 5.39 = $ 8085

Find Monthly Savings

Loss Function
L = K (Y-Y
o
)
2
For a single unit.

Thus

20 = K (Y
o
+/- TOL - Y
o
)
2

(since, when y = y
o
+/- TOL, all
parts are rejected and the loss
equals cost of rejecting a single
unit)
or 20 = K (TOL)
2

or
K = 20/(TOL)
2

= 20/(.35)
2
= 163.265

Also,
L = K (MSD)
for the population of parts


7.2 Manufacturer and Supplier Tolerances (A special application of loss function)

Example 11:
Nominal value of quality characteristic (y) = 100 lb.
Tolerance of y = +/-15 lb.
Cost to repair a nonfunctional unit by customer = $ 40
Cost to repair a nonfunctional unit by manufacturer = $ 15
Cost to repair a nonfunctional unit by supplier = $ 5

Find the required tolerances for manufacturer and supplier

Solution:

LOSS L = K (Y - Y
o
)
2
OR L = K (TOL)
2

Terefore, K = L/(TOL)
2
= 40/(15)
2
= .1778

Manufacturer Tolerance
_____ ________
TOL = +/- \/(L/K) = +/- \/(15/.1778) = +/- 9.18

Supplier Tolerance
_____ ________
TOL = +/- \/ (L/K) = +/- \/ (5/.1778) = +/- 5.3

Target
+/- 15
+/- 9.18
+/- 5.3


7.3 Variation Reduction Diagram

From a given set of data, statistics such as average, standard deviation, MSD, S/N, Cpk, Cp
and the normal distribution can be plotted. Of course for Cp and Cpk calculations,
specifications limits must be known. Conversely, if S/N ratio is known, average and Standard
deviation can be calculated. Once the average and the standard deviation is known, the normal
distribution can be plotted using the Gaussian equation.

When S/N is given, MSD can be calculated

MSD = 10
-[(S/N)/10]

also, since MSD =
2
+ (a - y
o
)
2
a = sample average and y
o
= Target
value

Standard deviation, can also be calculated when a and y
o
are known or assumed.

Example:
Data: 8.9 8.6 9.1 9.3 8.3 9.3 9.6 9.7 9.8 7.9 8.5 and 8.75

Calculated statistics with Target y
o
= 9 and Number of samples N = 12

Average. a = 8.98 = 0.59 MSD = .32 and S/N = 4.92

Note that used expressing MSD =
2
+ ( a - yo )
2
is calculated by dividing
eviation squares by N and not (N-1) as done in convensional standard deviation
caculation. Thus, to equate MSD with the RHS of the equation, must be modified as

msd
=
conventional
{(N-1)/N}
0.5
= 0.59 {11/12}
0.5
= 0.565

or (
msd
)
2
+ (a - y
o
)
2
= 0.31909 + .0004 = 0.32

which is what the MSD is calculated to be.
Normal Distribution
Among several distribution equations, most widely used is the Normal Distribution which was discovered
by Gauss, Laplace and Demoivre, independently. But it is generally identified as Gaussian distribution or
the Gaussian error law .

where x is the random variable, f(x) probability of x ( - <= x <= + ) and
= average(Mean) of population, = standard deviation of population

1
2

e

( x - )
2
2
2
f(x) =

How to Plot Distribution of Current Design

(a) If the current production is in place and samples are available, then collect sample data
and calculate average and standard deviation from it. Plot the normal distribution and mark
target value if applicable.

or

(b) If this is a new product and current production samples are not available, calculate
average and standard deviation using results of all trial conditions. Plot normal distribution
using the calculated average and standard deviation, and mark target value if applicable.

How to Plot the Estimated Performance at Improved Design

(a) If confirmation tests have been already done and the results are available, then calculate
the average and the standard deviation from the confirmations sample results. Plot normal
distribution (superimpose over the current distribution) using the average and the standard
deviation just calculated.

or

(b) If DOE results are analyzed but confirmation tests are not done, then calculate the
improved standard deviation using the estimated S/N at the optimum condition, average S/N
from all trial conditions, and the standard deviation of the current design. You will need to
assume, in this case, that the average performance is on target such that (a - y
o
) = 0.

Since MSD = 10
-[(S/N)/10]

and MSD =
2
+ (a - y
o
)
2
=
2
when average is on target

improved
=

current
{ (MSD)
improved
/(MSD)
current
}
0.5

improved
=

current
10

(Above equation should apply to all three quality characteristics)
[(S/N)
current
- (S/N)
imprved
]
20


Suppose that the improved S/N = 6.50,

then using

current
= 0.59 and (S/N)
current
= 4.92 the improved standard deviation becomes

improved
= 0.59 x 10
(4.92 - 6.50)/20
= 0.49

Thus the improved distribution will be a narrower (since .49 <.59) distribution superimposed
over the plot for the current design.

Number of samples
N = 12

Current Design
a = 8.98
= 0.59
MSD = .32 and
S/N = 4.92

Improved Design
S/N = 6.50
a = 9.0
= 0.49
99.73 %
95.45 %
68.27 %
1 -1 2 -2 3
-3

Area under Normal Distribution

7-1. Match the symbols of this formula to their proper definitions. L(y) = k(Y - Y
O
)
2

_______ Measured value of quality characteristic (a) L

(b) (y)
_______ Target value of quality characteristic
(c) k

_______ Constant which is dependent (d) Y
upon the cost structure of
manufacturing operation or (e) Yo
organization.

7-2. Consider a typical project in your own area and see if you can apply Taguchi's Loss
Function to quantify the savings that may result form an improved design. Discuss
with your group how you would determine the following items:

a.) Current status of design (performance) in terms of S/N ratio.
b.) Target value in case of "nominal is best" performance.
c.) Customer tolerance.
d.) Cost of rejection per unit.
e.) Rate of production per month.
f.) Expected performance (S/N ratio) of proposed design.

7-3. Taguchi's loss equation helps you calculate the $ loss associated with a production
process. Discuss how can you use it to determine expected savings.

7-4. For the given manufacturing conditions, calculate the $ savings/month.

Target = 5.00, Customer tolerance = .05, Cost of rejection = $21.00
Production = 18000/month(Rolled steel rods)
S/N before = 32.00 S/N after = 39.00

Hints: K (constant) = 8400
Before experiment: Loss = $ 5.3/unit
After experiment: Loss = $ 1.05/unit
Savings..................... = $ 76374

REVIEW QUESTIONS

7-5. [Concept: Loss Function ]
i) For the given production parameters, determine the savings from the improved design
condition.

Target/Nominal value = 9 Volts
Tolerance = 0.25 Volts
Cost of rejection at production = $ 0.15
Production per month = 450,000 units

S/N Before the experiment = 19.0
S/N in the improved design = 21.25

Partial Solution:

L = K (y - Yo )
2

when y = Yo +/- Tol,

then K = $ 0.15 / (().25)
2
or K = 2.40

Thus L = 2.40 (MSD)

Before Experiment, L = 2.40 x = $0.03/unit

After Experiment. L = 2.40 x = $ 0.017/unit

Therefore, Savings = ( .03 - .017 ) x 450,000 = $ 5.850 / month

ii) Calculate the savings resulting from the new process design which is expected to produce
S/N = -12.60. The current performance is at 20,000 units per month at a cost of $ 65 per unit.
The reject rate is 400 pieces per month and the current S/N is estimated to be at -15.0.
Compute the saving expected from the new design.

Partial solution:

Current Loss = 400 x $ 65 / 20000 = $ 1.30

L = K (MSD), MSD = 10 -(SN/10) or K = 1.30 /31.62 = 0.0411

Therefore, New Loss L = 0.0411 (New MSD) = = $.75 /unit

Thus, Savings = (1.30 -0.75 ) x 20,000 = $ 11,000 / month

$ Loss
9 Volts
$ .03
$ .017

7.6: An experiment used an L-9 to study three 3-level factors (A, B and C in cols. 1, 2 and
3) and one 2-level factor (D in col. 4 with level 3 changed to 1). Each of the 9 trials were
repeated 3 times and the following results were recorded. The TARGET value of the
performance is 50 (Nominal is the best).

Trial# Results
1....... 59 58 60 Analyze results using S/N ratio
2....... 48 49 52 and determine:
3....... 45 44 41 a) S/N of trial# 1
4....... 78 71 65 b) Effect of factor D, Good/Bad
5....... 67 69 65 c) Optimum condition
6....... 72 67 69 d) Optimum performance
7....... 45 54 59 1. in terms of S/N
8....... 55 56 57 ii. in terms of measured units
9....... 65 66 68 e) Status (S/N) of current design
f) Savings expected from new design
g) Difference in Cpk (variation reduction)

CURRENT PRODUCTION AND COST DATA
Performance: 52 54 45 43 57 51 53 44 34 39 and 62
(Based on inspection of current production parts. calculate S/N for current status).

Target = 50, Tolerance = 15 ( UCL = 75 and LCL = 35)

Monthly production ..... = 20,000 units
Cost of repair/rejection by Mfgr = $ 6.5/unit
Cost of repair/return by customer = $ 10/unit
Cost of repair at supplier's plant = $ 3/unit

Ans: a) -19.13 b) Good c) A1B2C2D2 d) S/N = - 4.78, 50 +/- 1.73 e) -18.099
f) $24,617 g) Will depend on the current S/N, UCL, LCL etc.

Module - 9

Robust Design
for
Dynamic Systems

Experiment Design and Analysis using
Dynamic Characteristics

by

Ranjit K. Roy, Ph.D.,P.E.

Module - 9: Robust Design for Dynamic Systems Page 9 - 2

9.1 Introduction

The word robust means insensitive. Insensitive to what? It means insensitive or immune to the
influence of uncontrollable (noise) factors. When a product or process is robust, its response is
less influenced by the uncontrollable factors. A robust design has less variation.

In robust design method, the attempt is to determine a design that is most insensitive to
variation. More specifically, robust design refers to the technique of advance experimental
design approach proposed by Dr. Genichi Taguchi in which the system design is optimized
with customer satisfaction as the primary objective.

Design of Experiment (DOE)
DOE is a statistical technique used to study multiple variables simultaneously. It is used
mostly in studying product or process parameters through experimental investigation. DOE
allows one to logically organize the experimental strategy and draw objective conclusions
about the factors involved.

DOE Using Taguchi Approach
It is a special form of DOE practiced and popularized by Dr. Genichi Taguchi of Japan. The
Taguchi approach is a relatively simpler and standardized type of DOE. In the Taguchi
approach, the goal is to determine optimum design condition, not by seeking the best
performance, but by evaluating the design that produce the minimum variation. This approach
further employs a formal way to study the effect of uncontrollable factors and relate the effect
of variation around the target, quantitatively, in terms of dollars(Loss Function).

Robust Design Method - a Different Perspective
Robust design requires a different approach to designing products and developing
technologies. The differences are pronounced in all aspects of experimental studies.

I. What are we after?
What are we to study?

Conventional approach:
Whats wrong and how can we fix it?

Robust design approach:
What is the product or process expected to do?
What is the intended function?
How can we design it to do the intended job the best way?

II. What do we want to investigate?
What to include in the study?
What to do about he influence of factors we cannot control?


Study control factors and control or eliminate uncontrollable factors.

Control the controllable factors such that the system is insensitive to the variation of
the uncontrollable factors.
When the system is insensitive, it exhibits reduced variation about the desired (ideal)
performance (function). The ideal function for the system is determined by studying the
expected performance before conducting the experiments.

III. What are the desired objectives?
What are the evaluation criteria?
How do we measure each of the evaluation criteria?

Measuring symptoms of poor quality and making decision based on it.

Measuring how well the performance input is transformed into the intended function.

9.2 Why Robust Design?
It is a more formal approach than conventional practice.

Conventional practice:

DESIGN ===> BUILD ====> TEST

Taguchi quality engineering approach:

Robust design method is applied in the parameter design phase of the Taguchis quality
engineering recommendation shown above.

Concept
Design
Parameter
Design
Tolerance
Design

1. Identify project team
2. Determine evaluation criteria
& Ideal function
3. Identify control and signal
factors
4. Identify noise factors and
select strategy
Expt 1
Expt. 2 Expt. 3 Etc.
Carry out experiments
Analyze results
Run confirmation experiments
I. Planning
II. Designing
III. Doing
IV. Analyzing
AnalyzingAnaly
V. Confirming
Parameter Design

What would customer want to do?
What does customer feel or see the effect as?

Examples:

* Slow down vehicle * Vehicle slows down

* Turn vehicle to the right * Vehicle turns to the right

How is the intent being fulfilled?
How is the intent measured?
How is the result evaluated?

The intent is an action on the part of customer. The measured value of the intent in
quantitative term, is called the SIGNAL.

Example: Amount steering wheel is turned, Force applied on the brake to slow down
vehicle, etc.

The measured value of the result is generally referred to as the RESPONSE.

Example: The amount vehicle turned, Deceleration of the vehicle when the brake is applied.,
etc.

Voice of Customer
Customer
Intent
Perceived
Results

* Push brake pedal * Fluid pressure * Vehicle slows down
activates brake pads

* Turn steering wheel * Steering linkages * Vehicle turns
turn wheels
Voice of the Customer
Customer
Intent
Signal
Perceived Result
Energy
Transformation
Model of the System Response

6.3 Ideal Function

Ideal function is the mathematical relationship of transformation between response and the
signal.

Voice of the Customer
Customer
Intent
Signal
Perceived
Energy
Transformatn.
on
Sources of Variation in Results
Variations in
Control factors
Customer usage Wear &
Tear
Environmental
Conditions
* Intended
Functions
* Other than
Intended
Functions
Transfer System
Ideal Function
Signal (M) Response (y)
M
y
Transfer
Function

In mathematical terms, the transfer function is the constant of proportionality between
response, y, and the signal, M.

y (proportional to ) M

or y = M (linear relationship)

where
= Constant of proportionality which establishes the constant relationship between the
signal and the response.

What Does an Ideal Function Do?

It defines a mathematical model of the system performance based on physics. The engineer
can use this model to evaluate the system performance by comparing the closeness of the data
to the model and adjust the control factors that produce results closest to the ideal function.

Note:
In contrast to the traditional approach where system performance is represented by a math
model (analytical simulation, FEM model, distribution equation, etc.), the robust design
approach starts with idealized behavior representation and achieve that by adjusting the
system parameters.

M y
Transfer
Function
Ideal Function, y = M

(Engineering responsibility)

y
M
y = M
y
M
y = M
y
M
y = M
y
M
y = M
y
M
y = M
y
M
y = M
Steering wheel angle
Amount
wheel
turned
Water faucet opening
Water
flow
rate
Fuel
gauge
reading
Amount of gas in the tank
Idle
engine
RPM
Acceleration pedal depression
Air flow
rate
Fan speed
Cooking
time
Energy level in microwave
Examples of Ideal Functions

9.4 Desirable Performance From Robust Design of Dynamic Systems

The most desirable ideal function is the one which is linear and the variation is minimum at
any point on the line.

What if the ideal function is not linear?

If the ideal function is not linear, then the function can be transformed such that the plot
becomes linear. Generally this can be accomplished by plotting the ideal function in a Log or
semi-Log paper.
y
M
y
M
y
M
y
M
Nonlinear
High variation
Nonlinear
Low variation
Linear
Large variation
Linear
Low variation
Least Desirable
Most Desirable

Suppose that

y = M
n

which is polynomial of nth order. By taking Log of both sides,

Log (y) = Log( ) + n Log (M)

or Y = C + n M ( a linear equation)

where Y = Log (y) and M = Log (M)

9.5 Robust Design Strategy

* Define ideal function

* Determine control and noise factors for the system

* Design and run experiments

* Compare how close the results are to the ideal function

* Consider closeness to the Ideal Function as the criteria to determine the best
control factor settings.

Experiment Layout Strategy

* Select inner array based on control factors

* Run experiments at various levels of signal and with different noise factor setting at each
trial condition. Repeat samples at each combination of signal and noise (cell).

The number of test samples required per trial condition depends on the signal & noise
combination, and the number of repetitions for each combination. If three samples are tested
for each combination of two levels of signal and two levels of noise, then there will be 12
samples necessary for each trial condition, as shown below.


Analysis Approach

* Determine how all data for each trial condition deviate from the ideal function. Express the
variation of the data in terms of a single number (S/N ratio). Use these numbers to carry out
analysis of the experiments and determine the optimum condition.

Ideal
M1 M2 M3
M4
Respons
e
*
*
*
*
*
*
*

*

*
*
*

*
*
*

*
*
*

*
*

*
Signal Factor
Inner
Array
Signal M1 Signal M2
N1
N2
N1 N2
Results

Static and Dynamic Systems

DOE application projects are of two types; Static and Dynamic. In Static system, response
does not change with time, nor is it dependent on a specific factor. Such systems always
function at about the same performance level. Dynamic systems, on the other hand, have
variable performance depending on a specific factor, called the Signal Factor.

Examples of Dynamic Systems

1. Brake system
2. Steering control mechanism
3. Water faucet control system
4. Variable speed electric fan
5. Diving board Structure design

Ideal Function - How should the system behave?

Response Control - What is desirable behavior?

Linear Nonlinear Response

Small variations

Large variations

Signal

Goal - Design the system such that the response follow a straight line with the least amount of
variation around it.

Analogies:

Robust Static System - performs like a golfer who has a consistent putting record.

Robust Dynamic System - is like a good biker who always steadily rides on a straight line up
and down the hill.

Desirable
Desirable
??
?? most undesirable
Signal factor

Dynamic Characteristic

Reference texts:
1. Taguchi Methods by Glen S. Peace, Addison Wesley Publishing Company, Inc. NY, 1992,
Pages 338-363
2. Quality Engineering Using Robust Design by Madhav S. Phadke, PTR Prentice Hall,
Englewood , Cliff, NJ. Pages 114 - 117.
3. Taguchis Quality Engineering Philosophy and Methodology by Shin Taguchi, American
Supplier Institute, Inc. Allen Park, MI.
4. Robust Technology Development by Yuin Wu, American Supplier Institute

What is Dynamic Characteristic?
A system is considered to exhibit dynamic response characteristic when the strength
of a particular factor has a direct effect on response under investigation. Such
a factor with a direct influence on the result is called a SIGNAL factor.

SIGNAL FACTOR - is an input to the system. Value/level may change

CONTROL FACTOR - is also an input to the system. Values/level is fixed to
the optimum level for the best performance.

NOISE FACTOR - is an uncontrollable factor. Its level is random during
actual performance.

STATIC SYSTEM GOAL - is to determine combination of control factor levels
which produces the best performance when exposed to
the influence of the varying levels of noise factors.

DYNAMIC SYSTEM GOAL - is to find the combination of control factor levels which
produces different levels of performances in direct proportion to the signal factor, but
produces minimum variation due to the noise factors at each level of the signal.

Example: Fabric dyeing process
Control factor: Types of dyes, Temperature, pH number, etc.
Signal factor: Quantity of dye
Noise factor: Amount of starch

Conducting Experiments with Dynamic Characteristics
When carrying out experiments, proper order and sequence of samples tested under each trial
condition, must be maintained. The number of samples required for each trial condition, will
depend on the number of levels of signal factor, noise conditions and repetitions for each cell
(a fixed condition of noise and signal factor). To design experiments for a dynamic system
and enter the results using QUALITEK-4 software, follow these steps:

Step 1. Design experiment with control factors using option F1 from
main menu ( manual or automatic design).

Step 2. Print description of trial conditions using option F5.

Step 3. Select option F8, the F5 to describe(*) DYNAMIC CHARACTERISTICS
and print description of test conditions (noise and signal).

* You will need to describe signal and noise factors and their levels.
You will also have to decide on the number of levels of signal and
noise factors. BUT MOST IMPORTANTLY, you will have to choose
the nature of the ideal function (Straight line representing the
behavior Response vs. Signal) applicable to your system.

Step 4. Strictly follow the prescribed test conditions.

Step 5. Enter results in the order and locations (run#) prescribed in step 4 using option F1
from analysis menu.

Signal-To-Noise Ratio Equations (alternate dynamic characteristic equations)
Signal factor may not always be clearly defined or known. For common
industrial experiments, one or more attributes may be applicable:

* true value known
* interval between factor levels known
* factor level ratios known
* factor level values vague

Depending on the circumstances of the input signal values and the resulting response data,
different signal-to-noise (S/N) ratio equations are available.


Zero Point proportional - Select this type of equation when response line
passes through the origin. The signal may be known, unknown or vague.

Response (y)

Zero Point proportional

Signal Strengths(M)

Reference Point proportional - this type of equation should be the choice response line does
not pass through the origin but through a known value of the signal or when signal values are
wide apart or far away from origin.

Response (y)

Reference Point proportional

Ref. Point

Signal Strengths(M)

When the signal values are known, zero point or reference point
proportional should be considered first. If neither is appropriate,
the linear equation should be used.

Linear Equation - is based on the least squares fit equation and should be
used where neither zero and reference point proportional equation are not
appropriate. Use it when signal values are close together and response does
not pass through the origin.


Response (y)

Linear Equation

Signal Strengths(M)

When in Doubt - plot the response as a function of the signal factor
values on a linear graph and examine the magnitude of y-intercept. If it passes
through origin, use ZERO POINT. If the intercept is not through origin
but the line passes through a fixed point, use REF.POINT. In all other
situation use LINEAR EQUATION.

Zero Point Proportional: y = M

r = r
o
( M
1
2
+ M
2
2
+ ........ M
k
2
)

Sum of response under each signal level (I=1, 2, .... k)
y
i
= y
ij

j = 1,2.. r
o

Where
y = response of the system (QC) , M = Signal factor strength
= slope of the (ideal) response line, k = number of signal levels
r
o
= number of samples tested at each signal level
M
1,
M
2,
etc, are strength of the signal levels
y
11,
y
12,
etc, are response from sample 1 and 2 under signal level 1.

y
1,
y
2,
etc., are totals of responses under signal levels 1, 2, .. k, etc.

Slope = ( M
1
y
1
+ M
2
y
2
+........M
k
y
k
) / r and S
t
=Y
11
2
+ Y
12
2
+ ..+ Y
2
kro

Variation caused by linear effect:

S
= 1/r (M
1
y
1
+ M
2
y
2
+ ........M
k
y
k
)
2
= r
2

Ref. Point Proportion: y - y
s
= (M - M
s
)

y
s
= 1/r
o
(y
1
+ y
2
+ ...........y
ro
)

where r
o
= no. samples for each signal level
M
s
= reference signal strength

r = r
o
[(M
1
- M
s
)
2
+ (M
2
- M
s
)
2
+.....(M
k
- M
s
)
2
]

Adjusted sum of response for each Signal factor strength
ro
y
i
= (y
j
- y
s
)

j = 1
= [y
1
(M
1
- M
s
) + y
2
(M
2
-M
s
) + . . . + y
k
(M
k
-M
s
)] / r

k ro
S
t
= (y
ij
- y
s
)
2
i =1 j=1

S
= [y
1
(M
1
- M
s
) + y
2
(M
2
- M
s
) +..... y
k
(M
k
- M
s
) ]
2
/ r

: Variation associated with error or non linearity

S
e
= S
t
- S

Error Variance: V
e
= S
e
/ (Kr
o
- 1)
S/N: = 10 log 1 (S
- Ve)
r Ve


S
e
= S
t
- S

Error Variance: V
e
= S
e
/ (Kr
o
- 1)
S/N: = 10 log 1 (S
- Ve)
r Ve


where
_ _
m = y, e = error
_
M = (M
1
+ M
2
+ ..... + M
k
) / k
_ _ _
r = r
o
[(M
1
-M)
2
+ (M
2
- M)
2
+ .. (M
k
- M )
2

y
i
= y
ij

j = 1,2.. r
o

__ __ __
= 1/r [ y
1
(M
1
- M) + y
2
(M
2
-M) + y
K
(M
K
-M) ]

k ro
S
T
= y
2
11
+ y
2
12
+ y
2
kro
- ( y
ij
)
2
/k.r
o
I=1 j=1
__ __ __
S

= [y
1
(M
1
- M) + y
2
(M
2
-M) + y
K
(M
K
-M) ]
2
/ r.


S
e
= S
t
- S

Error Variance: V
e
= S
e
/ (Kr
o
- 2)

S/N: = 10 log 1 (S
- Ve)
r Ve

_ _
Linear Equation: y = m + ( M - M ) + e


9.5 Examples of Dynamic Systems

Discharge of water from water faucet: The rate of discharge is a function of valve opening.
The position of the valve is the signal factor. A well designed valve system will discharge in
direct proportion of the opening.

Design of a golf club: The club must be designed such that the distance is strictly a function
of the angle of the swing, which is the signal factor.

Example calculations:

Case of LINEAR EQUATION (Expt. file: ASI400.QT4)
The results of samples tested for trial#1 of an experiment with dynamic
characteristic. There are three signal levels, two noise levels, and
two repetitions per cell.

M1 M2 M3
Noise 1 Noise 2 Noise 1 Noise 2 Noise 1 Noise 2
____|_________________|_____________________|_________________
Trl#1| 5.2 5.6 5.9 5.8 | 12.3 12.1 12.4 12.5 | 22.4 22.6 22.5 22.2

Signal strengths: M1 = 1/3, M2 = 1, M3 = 3

CALCULATIONS FOR S/N:

Mavg = (1/3 + 1 + 3 ) / 3 = 1.444

ro = 4 (2 smple/cell x 2 noise levels)

r = 4 [(1/3 - 1.444)
2
+ (1 - 1.444)
2
+ (3 - 1.444)
2
] ... (L)

= 4 ( 1.2343 + 0.1971 + 2.421 )

= 15.41

y1 = 5.2 + 5.6 + 5.9 + 5.8 = 22.5

y2 = 12.3 + 12.1 + 12.4 + 12.5 = 49.3

y3 = 22.4 + 22.6 + 22.5 + 22.2 = 89.7


Beta = (1/r)[22.5 x (1/3-1.444) + 49.3 x (1-1.444) + 89.7 x (3-1.444)]
= (1/15.41) [ -24.9975 - 21.692 + 139.5732 ]
= 92.8842/15.4101
= 6.01

S
beta
= r x Beta^2 = 15.4101 x 6.0274^2 = 556.82

y
avg
= [5.2 + 5.6 + ...... + 22.2]/12 = 161.5/12 = 13.46

S
t
= (5.2 - yavg)
2
+ (5.6 - yavg)
2
+ .....+ (22.2 - yavg)
2

= 68.23 + 61.78 + 57.15 + 58.67 + 1.346 + 1.85 + 1.123 + .921
+ 79.92 + 83.54 + 81.72 + 76.387

= 572.65

S
e
= St - Sb = 572.65 - 556.82 = 15.83

V
e
= Se / ( 12 - 2 ) = 15.83 /10 = 1.583

Eta = 10 Log (S
beta
- V
e
) / (r x V
e
) ... for all Eqns.

= 10 Log [(556.82 - 1.583)/(15.41*1.583)]

= 10 Log(22.76)

= 13.572 (S/N for the trial# 1 results )

Likewise, S/N for all other trial conditions in the experiments are calculated. The analysis of
results is then carried out just the same way as done for the static S/N case. The optimum
condition determined by selecting the higher average factor effects as DYNAMIC
CHARACTERISTIC is nothing but a special case of NOMINAL IS THE BEST quality
characteristic. The optimum so determined, represents the control factor level combination
which is expected to produce linear response with the least variation along the straight line.
The exact value of the response, of course, is dependent on the strength of the signal and may
be estimated by the ideal function of the S/N equation.


Sample screens from Qualitek-4 Software
(Experiment: AS-400.QT4)

Description of Signal and Noise factors

System description and trial statitics


System trial response

Sytem configuration(P-Diagram)


9.6 Theory and Derivation of formulae (Reference 2. Page 114)

Notations: Equations apply to one row of results under dynamic characteristic (any trial)
K = number of signal levels
r
o
= number of results under each signal level (noise level x repetitions per cell)

When both the signal factor(M) and the quality characteristic(y) are continuous, the quality loss(Q) can be
expressed as

Q
kr
y M
o
ij i
j
r
i
k o
=
=

1
1 1
2

The quality loss comprises of two components: a) Deviation from linearity (error e), b) The slope(Beta) being
other than one. The slope can be estimated by minimization of the loss (estimated by least squares criterion).

y M e
ij i ij
= +

d
d
y M
ij i
j
r
i
k o
=
= =

- )
2
1 1
0

that is
y M M
ij
j
r
i
k o
=
= =

1
1 1
1
0 - )

which yields

=
= =
= =

y M
M
ij
j
r
i
k
i
j
r
i
k
o
o
1
1 1
2
1 1
- )
]

The quality loss after adjustment is given by

Q
K
kr
v M
a
o
ij i
j
r
i
k o
=
= =

- )
1 1
2

=
= =

K
kr
y
M
o
ij
j
r
i
k o
1
1 1
2


=
= =

K
kr
y M
o
ij
j
r
i
k o

2 1
1 1
2
1
- )

=

K
kr
kr
o
o
e
1
2
2
p _

= K
e
2
2

where the error variance(square of Sigma) is given by

e
o
ij i
j
r
i
k
kr
y M
o
2
1 1
2
1
1
=

= =

p _
- )

Minimizing the quality loss is equivalent to maximizing S/N given by

S/N =

= 10
10
2
2
log
e

Note:
Slope (Beta) is the change in y produced by unit change in signal strength(M). Thus the square of slope quantifies
effect of signal. The denominator(square of Sigma) represents the effect of noise. Hence the quantity (Eta) is
called S/N ratio.

In case where there is a target, we must make adjustments to both slope and the intercepts. This leads to response
equation of Reference Point Proportional type as shown below

y a M
o
= +
0

Module 8: Appendix (Reference Materials) Page A - 1

Reference Materials

Appendix

Contents Pages
Table of F Ratios(F-Tables)
90% and 95% Confidence Level A-2
99% and 97.5% Confidence Level A-3
List of common orthogonal arrays A-4
L
4
(2
3
) A-5

L
8
(2
7
) A-5

L
12
(2
11
) A-6

L
16
(2
15
) A-6

L
32
(2
31
) A-7

L
9
(3
4
) A-8

L
18
(2
1
x3
7
) A-8

L
27
(3
13
) A-9

L
16
(4
5
) A-10

L
32
(2
1
x4
9
) A-11

Triangular Tables
2-level arrays A-12
how to read the triangular table A-13
linear graphs for 2-level arrays A-13
3-level arrays A-14
4-level arrays A-15

Glossary of terms A-16
Mathematical relationships A-18
References A-21
Practice session using qt4 A-22
Class project outline A-24
Report cover A-25
Variation Reduction Plot A-26
Program Evaluation A-27

7.2 Table of F-ratios (F-Tables)

F-Table (90%)
n
2\
n
1
1 2 3 4 5 6 7
1 39.864 493500 53.593 55.833 57.241 58.204 58.906
2 8.5263 9.0000 9.1618 9.2434 9.2926 9.3255 9.3491
3 5.5393 5.4624 5.3908 5.3427 5.3092 5.2847 5.2662
4 4.5448 4.3246 4.1908 4.1073 4.0506 4.0098 3.9790
5 4.0604 3.7797 3.6195 3.5202 3.4530 3.4045 3.3679
6 3.7760 3.4633 3.2888 3.1808 3.1075 3.0546 3.0145
7 3.5894 3.2574 3.0741 2.9605 2.8833 2.8274 2.7849
8 3.4579 3.1131 2.6238 2.8064 2.7265 2.6683 2.6241
9 3.3603 3.0065 2.8129 2.6927 2.6106 2.5509 2.5053
10 3.2850 2.9245 2.7277 2.6053 2.5216 2.4606 2.4140
11 3.2252 2.8595 2.6602 2.5362 2.4512 2.3891 2.3416
12 3.1765 2.8068 2.6055 2.4801 2.3940 2.3310 2.2828
13 3.1362 2.7632 2.5603 2.4337 2.3467 2.2830 2.2341
14 3.1022 2.7265 2.5222 2.3947 2.3069 2.2426 2.1931
15 3.0732 2.6952 2.4898 2.3614 2.2730 2.2081 2.1582
ETC.

F-Table (95%)

n
2\
n
1
1 2 3 4 5 6 7
1 161.45 199.50 215.71 224.58 230.16 233.99 236.77
2 18.513 19.000 19.164 19.247 19.296 19.330 19.353
3 10.128 9.5521 9.2766 9.1172 9.0135 8.9406 8.8868
4 7.7086 6.9443 6.5914 6.3883 6.2560 6.1631 6.0942
5 6.6079 5.7862 5.4095 5.1922 4.3874 4.2839 4.2066
6 5.9874 5.1433 4.7571 4.5337 4.3874 4.2839 4.2066
7 5.5914 4.7374 4.3468 4.1203 3.9725 3.8660 3.7870
8 5.3277 4.4590 4.0661 3.8378 3.6875 3.5806 3.5005
9 5.1174 4.2565 3.7626 3.6331 3.4817 3.3738 3.2927
10 4.9646 4.1028 3.7083 3.4780 3.3258 3.2172 3.1355
11 4.8443 3.9823 3.5874 3.3567 3.2039 3.0946 3.0123
12 4.7472 3.8853 3.4903 3.2592 3.1059 2.9961 2.9134
13 4.6672 3.8056 3.4105 3.1791 3.0254 2.9153 2.8321
14 4.6001 3.7389 3.3439 3.1122 2.9582 2.8477 2.7642
15 4.5431 3.6823 3.2847 3.0556 2.9013 2.7905 2.7066
ETC.

F-Table (99%)
n
2\
n
1
1 2 3 4 5 6 7
1 4052.2 4999.5 5403.3 5624.6 5763.7 5859.0 5928.3
2 98.503 99.000 99.166 99.249 99.299 99.332 99.356
3 34.116 30.817 29.457 28.710 28.237 27.911 27.672
4 21.198 18.000 16.694 15.977 15.522 15.207 14.976
5 16.258 13.274 12.060 11.392 10.967 10.672 10.456
6 13.745 10.925 9.7795 9.1483 8.7459 8.4661 8.2600
7 12.246 9.5466 8.4513 7.8467 7.4604 7.1914 6.9928
8 11.259 8.6491 7.5910 7.0060 6.6318 6.3707 6.1776
9 10.561 8.0215 6.9919 6.4221 6.0569 5.8018 5.6129
10 10.044 7.5594 6.5523 5.9943 5.6363 5.3858 5.2001
11 9.6460 7.2057 6.2167 5.6683 5.3160 5.0692 4.8861
12 9.3302 6.9266 5.9526 5.4119 5.0643 4.8206 4.6395
13 9.0738 6.7010 5.7394 5.2053 4.8616 4.6204 4.4410
14 8.8616 6.5149 5.5639 50354 4.6950 4.4558 4.2779
15 8.6831 6.3589 5.4170 4.8932 4.5556 4.3183 4.1415
ETC.

FOR COMPLETE F-TABLE INFORMATION, PLEASE CONSULT ANY TEXT ON THE SUBJECT. A LIST OF
TEXTS CAN BE FOUND IN THE REFERENCE PAGE.

F-Table (97.5%)
n
2\
n
1
1 2 3 4 5 6 7
1 647.79 799.50 864.16 899.58 921.85 937.11 948.22
2 38.506 39.000 39.165 39.248 39.298 39.331 39.355
3 17.443 16.044 15.439 15.101 14.885 14.735 14.624
4 12.218 10.649 9.9792 9.6045 9.3645 9.1973 9.0741
5 10.007 8.4336 7.7636 7.3879 7.1464 6.9777 6.8531
6 8.8131 7.2598 6.5988 6.2272 5.9876 5.8197 5.6955
7 8.0727 6.5415 5.8898 5.5226 5.2852 5.1186 4.9949
8 7.5709 6.0595 5.4160 5.0526 4.8173 4.6517 4.5286
9 7.2093 5.7147 5.0781 4.7181 4.4844 4.3197 4.1971
10 6.9367 5.4564 4.8256 4.4683 4.2361 4.0721 3.9498
11 6.7241 5.2559 4.6300 4.2751 4.0440 3.8807 3.7586
12 6.5538 5.0959 4.4742 4.1212 3.8911 3.7283 3.6065
13 6.4143 4.9653 4.3472 3.9959 3.7667 3.6043 3.4827
14 6.2979 4.8567 4.2417 3.8919 3.6634 3.5014 3.3799
15 6.1995 4.7650 4.1528 3.8043 3.5764 3.4147 3.2194
ETC.

Common Orthogonal Arrays

TYPE OF NUMBER OF
ARRAY FACTORS LEVELS

L
4
(2
3
) 3 2

L
8
(2
7
) 7 2

L
12
(2
11
) 11 2

L
16
(2
15
) 15 2

L
32
(2
31
) 31 2

L
9
(2
4
) 4 3

L
18
(2
1
,3
7
) 1 2
and 7 3

L
27
(3
13
) 13 3

L
54
(2
1
,3
25
)-modified 1 2
and 25 3

L
81
(3
40
)-modified 40 3

L
16
(4
5
)-modified 5 4

L
32
(2
1
,4
9
)-modified 1 2
and 9 4

L
64
(4
21
) 21 4

etc.


2-Level Orthogonal Arrays

Interactions
(Linear Graphs)

1 3 2
1x2 =>3

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

L
4
(2
3
) Array

Trial#\ 1 2 3
1 1 1 1
2 1 2 2
3 2 1 1
4 2 2 1
L
8
(2
7
) Array
COL.>>
Trial# 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


2- Level Orthogonal Arrays(Contd.)

L
16

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

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

5 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2
6 1 2 2 1 1 2 2 2 2 1 1 2 2 1 1
7 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1
8 1 2 2 2 2 1 1 2 2 1 1 1 1 2 2

9 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2
10 2 1 2 1 2 1 2 2 1 2 1 2 1 2 1
11 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1
12 2 1 2 2 1 2 1 2 1 2 1 1 2 1 2

13 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1
14 2 2 1 1 2 2 1 2 1 1 2 2 1 1 2
15 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2
16 2 2 1 2 1 1 2 2 1 1 2 1 2 2 1

L
12

Column =>
Cond. 1 2 3 4 5 6 7 8 9 10 11
1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 2 2 2 2 2 2
3 1 1 2 2 2 1 1 1 2 2 2
4 1 2 1 2 2 1 2 2 1 1 2
5 1 2 2 1 2 2 1 2 1 2 1
6 1 2 2 2 1 2 2 1 2 1 1
7 2 1 2 2 1 1 2 2 1 2 1
8 2 1 2 1 2 2 2 1 1 1 2
9 2 1 1 2 2 2 1 2 2 1 1
10 2 2 2 1 1 1 1 2 2 1 2
11 2 2 1 2 1 2 1 1 1 2 2
12 2 2 1 1 2 1 2 1 2 2 1
NOTE:
The L-12 is a special array designed to investigate main effects of 11 2-level factors.
THIS ARRAY IS NOT RECOMMENDED FOR ANALYZING INTERACTIONS


2- Level Orthogonal Arrays(Contd.)
(2-LEVEL, 31 FACTORS)

L
32

(2
31
)
Col => 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 3 3
Cond1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
3 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2
4 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1
5 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2
6 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1
7 1 1 1 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 1 1 1 1
8 1 1 1 2 2 2 2 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 1 1 1 1 2 2 2 2

9 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2
10 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1
11 1 2 2 1 1 2 2 2 2 1 1 2 2 1 1 1 1 2 2 1 1 2 2 2 2 1 1 2 2 1 1
12 1 2 2 1 1 2 2 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 1 1 2 2 1 1 2 2
13 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1
14 1 2 2 2 2 1 1 1 1 2 2 2 2 1 1 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2
15 1 2 2 2 2 1 1 2 2 1 1 1 1 2 2 1 1 2 2 2 2 1 1 2 2 1 1 1 1 2 2
16 1 2 2 2 2 1 1 2 2 1 1 1 1 2 2 2 2 1 1 1 1 2 2 1 1 2 2 2 2 1 1

17 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2
18 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1
19 2 1 2 1 2 1 2 2 1 2 1 2 1 2 1 1 2 1 2 1 2 1 2 2 1 2 1 2 1 2 1
20 2 1 2 1 2 1 2 2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1 1 2 1 2 1 2 1 2
21 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1
22 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1 2 1 2 1 1 2 1 2 2 1 2 1 1 2 1 2
23 2 1 2 2 1 2 1 2 1 2 1 1 2 1 2 1 2 1 2 2 1 2 1 2 1 2 1 1 2 1 2
24 2 1 2 2 1 2 1 2 1 2 1 1 2 1 2 2 1 2 1 1 2 1 2 1 2 1 2 2 1 2 1

25 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1
26 2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2
27 2 2 1 1 2 2 1 2 1 1 2 2 1 1 2 1 2 2 1 1 2 2 1 2 1 1 2 2 1 1 2
28 2 2 1 1 2 2 1 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1 2 1 2 2 1 1 2 2 1
29 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2
30 2 2 1 2 1 1 2 1 2 2 1 2 1 1 2 2 1 1 2 1 2 2 1 2 1 1 2 1 2 2 1
31 2 2 1 2 1 1 2 2 1 1 2 1 2 2 1 1 2 2 1 2 1 1 2 2 1 1 2 1 2 2 1
32 2 2 1 2 1 1 2 2 1 1 2 1 2 2 1 2 1 1 2 1 2 2 1 1 2 2 1 2 1 1 2


3- Level Orthogonal Arrays

L
9
(3
4
)
COL==>
COND 1 2 3 4
1 1 1 1 1
2 1 2 2 2
3 1 3 3 3
4 2 1 2 3
5 2 2 3 1
6 2 3 1 2
7 3 1 3 2
8 3 2 1 3
9 3 3 2 1

L
18
( 2
1
3
7
)

Col==>
Trial 1 2 3 4 5 6 7 8
1 1 1 1 1 1 1 1 1
2 1 1 2 2 2 2 2 2
3 1 1 3 3 3 3 3 3
4 1 2 1 1 2 2 3 3
5 1 2 2 2 3 3 1 1
6 1 2 3 3 1 1 2 2
7 1 3 1 2 1 3 2 3
8 1 3 2 3 2 1 3 1
9 1 3 3 1 3 2 1 2

10 2 1 1 3 3 2 2 1
11 2 1 2 1 1 3 3 2
12 2 1 3 2 2 1 1 3
13 2 2 1 2 3 1 3 2
14 2 2 2 3 1 2 1 3
15 2 2 3 1 2 3 2 1
16 2 3 1 3 2 3 1 2
17 2 3 2 1 3 1 2 3
18 2 3 3 2 1 2 3 1


3- Level Orthogonal Arrays(contd.)

L

27
( 3
13
)
Column =>
Cond. 1 2 3 4 5 6 7 8 9 10 11 12 13
1 1 1 1 1 1 1 1 1 1 1 1 1 1
2 1 1 1 1 2 2 2 2 2 2 2 2 2
3 1 1 1 1 3 3 3 3 3 3 3 3 3
4 1 2 2 2 1 1 1 2 2 2 3 3 3
5 1 2 2 2 2 2 2 3 3 3 1 1 1
6 1 2 2 2 3 3 3 1 1 1 2 2 2
7 1 3 3 3 1 1 1 3 3 3 2 2 2
8 1 3 3 3 2 2 2 1 1 1 3 3 3
9 1 3 3 3 3 3 3 2 2 2 1 1 1

10 2 1 2 3 1 2 3 1 2 3 1 2 3
11 2 1 2 3 2 3 1 2 3 1 2 3 1
12 2 1 2 3 3 1 2 3 1 2 3 1 2
13 2 2 3 1 1 2 3 2 3 1 3 1 2
14 2 2 3 1 2 3 1 3 1 2 1 2 3
15 2 2 3 1 3 1 2 1 2 3 2 3 1
16 2 3 1 2 1 2 3 3 1 2 2 3 1
17 2 3 1 2 2 3 1 1 2 3 3 1 2
18 2 3 1 2 3 1 2 2 3 1 1 2 3

19 3 1 3 2 1 3 2 1 3 2 1 3 2
20 3 1 3 2 2 1 3 2 1 3 2 1 3
21 3 1 3 2 3 2 1 3 2 1 3 2 1
22 3 2 1 3 1 3 2 2 1 3 3 2 1
23 3 2 1 3 2 1 3 3 2 1 1 3 2
24 3 2 1 3 3 2 1 1 3 2 2 1 3
25 3 3 2 1 1 3 2 3 2 1 2 1 3
26 3 3 2 1 2 1 3 1 3 2 3 2 1
27 3 3 2 1 3 2 1 2 1 3 1 3 2


This array is called the modified L-16 array which is made by combining the 5
interacting groups in the original 16 2-level columns.

1
2
3, 4, 5
Linear Graph of L
16

To study interaction between two 4-level factors
we must set aside three 4-level columns.
1 2 3 4 5
Col. =>
Trial
1 1 1 1 1 1
L
16
( 4 )
5
2 1 2 2 2 2
3 1 3 3 3 3
4 1 4 4 4 4
5 2 1 2 3 4
6 2 2 1 4 3
7 2 3 4 1 2
8 2 4 3 2 1
9 3 1 3 4 2
10 3 2 4 3 1
11 3 3 1 2 4
12 3 4 2 1 3
13 4 1 4 2 3
14 4 2 3 1 4
15 4 3 2 4 1
16 4 4 1 3 2


4- Level Orthogonal Arrays(contd.)
(2-Level and 4-Level)

L
32
(2
1
x4
9
)

Trial\Column==> 1 2 3 4 5 6 7 8 9 10
1 1 1 1 1 1 1 1 1 1 1
2 1 1 2 2 2 2 2 2 2 2
3 1 1 3 3 3 3 3 3 3 3
4 1 1 4 4 4 4 4 4 4 4
5 1 2 1 1 2 2 3 3 4 4
6 1 2 2 2 1 1 4 4 3 3
7 1 2 3 3 4 4 1 1 2 2
8 1 2 4 4 3 3 2 2 1 1

9 1 3 1 2 3 4 1 2 3 4
10 1 3 2 1 4 3 2 1 4 3
11 1 3 3 4 1 2 3 4 1 2
12 1 3 4 3 2 1 4 3 2 1
13 1 4 1 2 4 3 3 4 2 1
14 1 4 2 1 3 4 4 3 1 2
15 1 4 3 4 2 1 1 2 4 3
16 1 4 4 3 1 2 2 1 3 4

17 2 1 1 4 1 4 2 3 2 3
18 2 1 2 3 2 3 1 4 1 4
19 2 1 3 2 3 2 4 1 4 1
20 2 1 4 1 4 1 3 2 3 2
21 2 2 1 4 2 3 4 1 3 2
22 2 2 2 3 1 4 3 2 4 1
23 2 2 3 2 4 1 2 3 1 4
24 2 2 4 1 3 2 1 4 2 3

25 2 3 1 3 3 1 2 4 4 2
26 2 3 2 4 4 2 1 3 3 1
27 2 3 3 1 1 3 4 2 2 4
28 2 3 4 2 2 4 3 1 1 3
29 2 4 1 3 4 2 4 2 1 3
30 2 4 2 4 3 1 3 1 2 4
31 2 4 3 1 2 4 2 4 3 1
32 2 4 4 2 1 3 1 3 4 2


How to read the triangular table
To find the interaction column between factor a placed in column 4 and factor b placed in
column 7, look for the number at the intersection of the horizontal line through (4) and
the vertical line through (7), which is 3.

4 x 7 => 3

A x B => AxB

Likewise 1 x 2 => 3

3 x 5 => 6

Etc.

The set of three columns (4, 7, 3), (1, 2, 3), etc. are called interacting groups of columns.

Triangular Table for

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
(15)
ETC...

Linear Graphs for 2- Level Orthogonal Arrays

1
3
2
1
2
3
5 4
7
6

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

1
2 4
3
6
5
7

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

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


DOF for A x B => 2 x 2 = 4

where a & b are two 3-level factors.

Interaction effects between two 3-level factors are mixed with two 3-level columns.

( 1 ) x ( 2 ) => Cols.: 3 and 4

2 DOF X 2 DOF => 2 (col. 3) + 2 (col. 4) = 4 DOF

Triangular Table for 3-Level Orthogonal Arrays

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

(1) 3 2 2 6 5 5 9 8 8 12 11 11
4 4 3 7 7 6 10 10 9 13 13 12
(2) 1 1 8 9 10 5 6 7 5 6 7
4 3 11 12 13 11 12 13 8 9 10
(3) 1 9 10 8 7 5 6 6 7 5
2 13 11 12 12 13 11 10 8 9
(4) 10 8 9 6 7 5 7 5 6
12 13 11 13 11 12 9 10 8
(5) 1 1 2 3 4 2 4 3
7 6 11 13 12 8 10 9
(6) 1 4 2 3 3 2 4
5 13 12 11 10 9 8
(7) 3 4 2 4 3 2
12 11 13 9 8 10
(8) 1 1 2 3 4
10 9 5 7 6
(9) 1 4 2 3
8 7 6 5
(10) 3 4 2
6 5 7
(11) 1 1
13 12
(12) 11
ETC.


DOF FOR A x B => 3 x 3 = 9

where a and b are two 4-level factors.
Interaction effects between two 4-level columns shows up in three 4-level columns.

( 1 ) x ( 2 ) => COLS.: 3, 4 and 5

3 DOF x 3 DOF => 3 (col. 3) + 3 (col. 4) + 3 (col. 5) => 9 DOF
Triangular Table for 4-Level Orthogonal Arrays

1 2 3 4 5 6 7 8 9 10 11
(1) 3 2 2 2 7 6 6 6 11 10
4 4 3 3 8 8 7 7 12 12
5 5 6 4 9 9 9 9 13 13
(2) 1 1 1 10 11 12 13 6 7
4 3 3 14 15 19 17 14 15
5 5 4 18 19 20 21 18 19
(3) 1 1 11 10 13 12 7 6
2 2 16 17 14 15 17 18
5 4 21 20 19 18 20 21
(4) 1 12 13 10 11 8 9
2 17 16 15 14 15 14
3 19 18 21 20 21 20
(5) 13 12 11 10 9 8
15 14 17 16 16 17
20 21 18 19 19 18
(6) 1 1 1 2 3
8 7 7 14 16
9 9 8 18 21
(7) 1 1 3 2
6 6 17 15
9 8 20 19
(8) 1 4 5
6 15 17
(9) 5 4
16 14
19 20
(10) 1

Glossary of Terms
ANOVA: Analysis of Variance is a table of information which displays relative influence
of factor or interaction assigned to the column of the orthogonal array.

Controllable factors: A design variable that is considered to influence the response and
is included in the experiment. Its levels can be controlled at experimenter's will.

Error: The amount of variation in the response caused by factors other than controllable
factors included in the experiment.

Histogram: A graphical representation of the sample data using classes on the Horizontal
Axis and frequency on the Vertical Axis.

Interaction: Two factors are said to have interaction with each other if influence of one
depends on the value of the other.

Linear Graph: A Graphical representation of relative column locations of factors and
their interactions. These were developed by Dr. TAGUCHI to assist in assigning different
factors to columns of the Orthogonal Array.

Loss Function: A mathematical expression proposed by Dr. TAGUCHI to quantitatively
determine the harm caused by the lack of quality in the product. This harm caused by the
product is viewed as a loss to the society and is expressed as a direct function of mean
square deviation from the target value.

Noise Factors: Those factors that have influence over the response but cannot be
controlled in actual applications. They are of three kinds.

Outer Noise - Environmental Conditions vs. Humidity
Temperature, Operators, etc.
Inner Noise - Deterioration of Machines, tools and parts.

Between Product Noise - Variation from piece to piece.

Off-Line Quality Control: The Quality enhancement efforts in activities before
production. These are activities such as upstream planning, R & D, systems design,
parameter design, Tolerance design and loss function, etc.

Orthogonal Array: A set of tables containing information on how to determine the least
number of experiments and their conditions. The word orthogonal means balanced.

Quality Characteristic: The yardstick which measures the performance of a product or a
process under study. For a plastic molding process, this could be the strength of the molded
piece. If we are after baking the best cake, this could be a combination of taste, shape and
moistness.


Response: A Quantitative value of the measured quality
characteristic. e.g. stiffness, weight, flatness, etc.

Robustness: Describes a condition in which a product/process is least influenced by variation
of individual factors. To become robust is to become less sensitive to variations.

S/N Ratio: Stands for signal to noise ratios. i.e., the ratio of the power of the signals to the
power of noise (error). A high S/N ratio will mean that there is high sensitivity with the least
error of measurement. In Taguchi Analysis using S/N ratios, a higher value is always desirable
regardless of the quality characteristic.

Signal Factor: A factor that influences the average value, but not the variability in response.

System Design : Design of a product or a process using special Taguchi Techniques.

Target Value: A value that a product is expected to possess. Most often this value is different
from what a single unit actually does. For a 9 volt transistor battery the target value is 9 volts.

Tolerance Design: This is a sophisticated version of parameter design used to optimize
tolerance, reduce cost and increase customer satisfaction.

Variables, Factors, or Parameters: These words are used synonymously to indicate the
controllable factors in an experiment. In case of a plastic molding experiment, molding
temperature, injection pressure, set time, etc. are factors.


Useful Mathematical Relationships

ANOVA

or ( y
i
- y
o
)
2
= n
T
2
+ n m
2
..... (2)

where m = ( y
avg
- y
o
) and by definition,

Variance
T
2
= [ ( y
i
- y
avg
)
2
/n or [ ( y
i
- y
avg
)
2
= n
T
2
..... (3)

The variation around the mean of the data is called the total sum of squares, S
T

therefore (by definition of total sums of squares),

S
T
= ( y
i
- y
avg
)
2
..... (4)

that is n
T
2

= ( y
i
- y
o
)
2
- n m
2
(since S
T
= n
T
2
) ..... (5)

When y
o
= 0, that is the target is at the origin, then m = y
avg
and the total sums of
squares

S
T

= y
i
2
- n y
avg

2

= y
i
2
- n ( T/n )
2
..... (6)
Symbols and Notations
y
avg
= Average of result
c
= Standard deviation(classical definition/sample)
y
min
= Minimum result
T
= Standard deviation(Taguchi/population)
y
max
= Maximum result S
t
= Total sums of squares
y
o
= Target/Nominal MSD = Mean Squared Deviation m= ( y
avg
- y
o
)
y

= Result S/N = Signal to noise ratio T = Total of all results
n

= Number of results C
pk
= Capability index = Summation sign

Variation due to
individual data
about the mean of
the data ( S
T
)

Variation of the
mean from the
target
Total variation
about the target
value
( y
i
- y
o
)
2
= [ ( y
i
- y
avg
) + ( y
avg
- y
o
) ]
2
, i = 1, 2, .........n ..... (1)
= [ ( y
i
- y
avg
)
2
+ 2 ( y
i
- y
avg
) ( y
avg
- y
0
) + ( y
avg
- y
o
)
2
]

= n x [ ( y
i
- y
avg
)
2
/n + 0 + ( y
avg
- y
o
)
2


Mean Squared Deviation

By definition MSD = ( y
i
- y
o
)
2
/n and S
T
= ( y
i
- y
avg
)
2

Therefore, MSD = ( y
i
- y
o
)
2
/n or MSD = S
T
/ n ..... (7)

or MSD = [n
T
2

+ n m
2
] / n (from Eqn. 5)

MSD =
T
2
+ (y
avg
- y
0
)
2
for QC = Nominal ......... (8)

Similarly (derivations not shown)

MSD =
T
2
+ (y
avg
- 0)
2
for QC = Smaller

and MSD = [ 1 + 3
T
2
/ (y
avg
)
2
] / (y
avg
)
2
for QC = Bigger
Variances (Standard vs. Taguchi)

T
2
= ( y
i
- y
avg
)
2
/n and
c
2
= [ ( y
i
- y
avg
)
2
/ (n - 1)

therefore,
c
2
=
T
2
n / (n-1) ..... (9)

or MSD =
c
2
n / (n-1) + m
2
..... (10)

Signal to Noise Ratio

S/N = - 10 Log (MSD) ..... (11)

= - 10 Log (
T
2
+ m
2
)

= - 10 Log (
c
2
n / (n-1) + m
2
)

or MSD = 10
-[S/N]/10
..... (12)

Capability Index

Cpk = [ y
avg
- y
min
] / (3
c
) for y
avg
< y
o
..... (13)

and Cpk = [ y
max
- y
avg
] / (3
c
) for y
avg
< y
o


L
1
, (S/N)
1
, L
2
, and (S/N)
2
are Loss and S/N before and after experiments, then savings can
be calculated as

Substituting L = L
1
and MSD = MSD
1

in the above, the constant K = L
1
/ (MSD)
1
( condition before experiment)

Therefore, L = L
1
[MSD] / (MSD)
1

Substituting L = L
2
and MSD = MSD
2
in the above equation

then

L
2
= L
1
[(MSD)
2
] /(MSD)
1
..... (15)

or L
2
= L
1
[10
- (S/N)2
/ 10 ] / [10
- (S/N)1 / 10
]

= L
1
10
- [ (S/N)2 - (S/N)1 ]/10

Savings in Terms of % of Current Loss
When current loss is unknown, the savings expected from the improved design can be
expressed as % of loss before experiment (L
1
) as

Savings per unit of product = 100 x ( L
1
- L
2
) / L
1
..... (16)

= 100 x ( L
1
- L
1
10
+ [ (S/N)1 - (S/N)2 ]/10
) / L
1

= [ 1 - 10
+ [ (S/N)1 - (S/N)2]/10
] x 100 % of L
1

where
(S/N)1 = S/N ratio of the current and (S/N)2 = S/N ratio of the improved design

Also, since MSD is proportional to square of the standard deviation, the improved S/N can
be estimated when the expected performance at optimum, and the average performance and
S/N at cured condition are known.

(S/N)2 = (S/N)1 + 20 Log - e [Y
average
./Y
expected
.] / Log - e [10]
Loss Function

For multiple parts, L = K [MSD], where K is a constant ..... (14)

= K [
T
2
+ m
2
]

= K [
c
2
n / (n-1) + m
2
]


References

1. Patrick M. Burgman. Design of Experiments - The Taguchi Way. Manufacturing
Engineering. May 1985, PP 44-46

2. Yuin Wu and Dr. Willie Hobbs Moore. 1986. Quality Engineering Product and Process
Optimization. Dearborn, Michigan. American Supplier Institute.

3. Ronald L. Iman and W.J Conover. 1983. A Modern Approach to Statistics. John Wiley
& Sons.

4. Yuin Wu. 1986. Orthogonal Arrays and Linear Graphs. Dearborn, Michigan. American
Supplier Institute.

5. Burton Gunter. 1987. A Perspective on the Taguchi Methods. Quality Progress.

6. Lawrence P. Sullivan. June 1987. A Power of the Taguchi Methods. Quality Progress.

7. Philip J. Ross. 1988. Taguchi Techniques for Quality Engineering, McGraw Hill
Book Company. New York, NY.

8. Thomas B. Baker and Don P. Causing. March 1984. Quality Engineering by
Design - The Taguchi Method. 40th Annual ASQC Conference.

9. Jim Quinlan. 1985. Product Improvement By Application of Taguchi Methods,
Flex Products, Inc., Midvale, Ohio. Winner of Taguchi Applications Award by
American Supplier Institute.

10. Genichi Taguchi. 1987. System of Experimental Design, UNIPUB, Kraus International
Publications, New York.

11. Ranjit K. Roy, 1996, QUALITEK-4 (for Windows): Software for Automatic Design
of Experiment Using Taguchi Approach, IBM or Compatible computer, NUTEK Inc. 30600
Telegraph Road. Suite 2230, Birmingham, Michigan 48025. (Fax. 1-248-642-4609). Free
DEMO from http://www.rkroy.com

12. Ranjit K. Roy. 1990. A Primer on the Taguchi method, Society of Manufacturing
Engineers, Dearborn, Michigan, USA. ISBN: 0-87263-468-X Fax: 1-313-240-8252 or 1-
313-271-2861


Practice Session Using Qualitek-4 Software

1. Run Qualitek-4(QT4) program from your computer and click OK past the Registration
Screen to the experiment configuration screen.
2. Notice that experiment file PISTON.Q4W is already loaded in memory. If not,
OPEN file PISTON.Q4W.

Analyze Results of PISTON and other demonstration experiments

3. Click Analysis menu and select Standard Analysis. Since PISTON has three
samples/results per trial, QT4 will remind you to perform S/N. Click OK to proceed. In the
next screen check Bigger is Better Quality Characteristic and click OK.

4. Review results and click on Graph if you wish to vie it. Click OK when done.

5. While in the Main Effect screen, click Plot to view plots of Main Effect. Once you are in
the graphics screen, click the << or >> button at the bottom of the graph to display other factor
plots. Click OK to return to Main Effect screen.

6. Click on Interaction button, to display interaction plots between any two factors. Select the
Automatic option if you want QT4 to calculate interaction between all possible factors, even
though you may not have thought about all the interactions or included them in your study.
QT4 calculates N (N-1)/2 possible pairs of interactions for N factors and ranks them by
Severity of their presence (Severity Index, 0 - 100%). Review the list to note the impotent
pairs of interaction for future reference. Click OK or Cancel to return to Main Effect screen
and click OK to proceed to ANOVA screen.

7. In ANOVA screen, try POOLING the factor with least amount of influence(last column).
To POOL double click on the factor description. At prompt, review % Confidence Level and
click OK. Notice how the ANOVA table gets updated. You can also try one or more of the
following tasks.
a) Pool another factor with the least influence
b) Reset ANOVA
c) Pool all factors with less than 98.6% confidence level.
d) Plot bar graph , pie chart, and Re-start ANOVA
e)Review Pie diagram

Click OK or ANOVA button to move to the ANOVA screen

8. At the Optimum screen:
a) Read expected response
b) Convert this result to a real number when S/N analysis is used.
c) Plot graph of factor contribution
d) Find response for all factors at level 2
e) Go back to ANOVA and pool all factors below 10% influence
f) Determine 90% confidence interval of the optimum response
g) Go to optimum screen

Click OK to return to the main screen. From the file menu select Open and select any other
experiment file from the list of over 50 files. Analyze results by following steps 3 -8.

Experiment Designs

9. Prepare a list of factors and levels for a practice experiment(say 5 factors at 2-level each).
Click on the DESIGN menu and select Manual Design. Check L-8 for your design. When in
design screen, describe factors and their levels by assigning to any rows (which are columns
of the array). Since you will be using five of the seven available columns, click on the
UNUSED button to designate the two columns as unused. Click OK to move to the
Orthogonal Array screen. Click OK to experiment file screen. Supply only the first eight
characters of your file name. The file extension .Q4W is automatically added.

10. After design is completed, you may review the trial condition by clicking on REVIEW
menu and selecting Trial Condition. Should you want to carry out these experiments, you may
print some or all the trial conditions.

Result Entry

11. Once the experiments are carried out, you will need to enter the results by clicking on
RESULT menu and then, selecting Enter Result option. If your result includes multiple
evaluation criteria, you will sleet Multiple Criteria(OEC) option. Enter results/evaluation
criteria as applicable. When done entering results, perform analysis following steps 3 - 8 as
described above.

Exercises (Solve problems from Review Questions using Qualitek-4 software):

(a) Prob. 2- 5 at Page 2-30 (b) Prob. 3- 6 at Page 3-19

(c) Prob. 4-12 at Page 4-20 (d) Prob. 5-11 at Page 5-18

(e) Prob. 7-6 at Page 7-11 (S/N analysis, variation reduction, and savings)

(f) Prob. 8-7 at Page 8-10 (Prepare OEC from readings in each trial condition/ Outer array)

Capturing, Pasting, and Cropping QT4 screens for reports and presentations
In addition to the standard print output, you can capture any screens and conveniently size and
paste it into your WORD document. (1) Be in the screen of your choice (2) Press
Alt+PrintScreen together (3) Open WORD document (4) Select EDIT and PASTE to place the
screen on WORD document. (5) Select picture(screen), then from INSERT menu select
Frame. (6) To crop picture, select and place the mouse on corner or middle posts. Then press
Shift key. Notice the mouse cursor shape changes. Press right button on mouse and drag to
desired location and release.

Class Project Application Guidelines
(for 4-day seminar/workshop)

Team objectives: Learn how to apply Taguchi experimental design and analyze results by
working with fellow team members. While working together as a team, each member of the
group should assume a role related to the selected project and brainstorm for the following
items.
1. Select a project, describe the objective and
a. Establish what you are after (better looks, performance, improve quality, etc.)
b. Decide how you will evaluate what you are after and how many
criteria of evaluations will you have.
c. When you have more than one criterion for evaluations, what
are the relative weightings for each and how will you combine them
into single quantifiable number.

2. Determine Factors, Factor Levels, Interactions, Noise Factors, Number of Samples, etc.
3. Design experiments and describe Trial conditions and Noise conditions.
4. Carry out experiments or assume results (as if the experiments were done).
5. In case of actual experiments, record observations (evaluation criteria) for each test sample.
6. Assign tasks among team members and present your CASE STUDY REPORT to the class.
Make copies for all attendees and the instructor.

MINIMUM REPORT CONTENTS (about 5 pages)
* Proj. title, date and name of participants
* A brief description of the project and objectives(about 100 words)
* Evaluation criteria table containing criteria description, QC, weighting, etc.(if more
than one criteria)
* Factor, levels and column assignments
* Results, optimum, % influences, C.I..
* $ Savings, etc. (optional items) and Variation Reduction diagram
* Outline actions based on findings.

PROJECT SCOPES:
Your experiment should include:
* One 3 or 4-level factors OR interaction(s) between two 2-level factors
* Explanation about the noise factors and how they are handled
* One or more measurable criteria of evaluations

PRESENTATION ITEMS:
Make copies of your report for all attendees and the instructor. Based on the results (assumed
or actual), tell the class what your findings are:
(a) how each factor behave
(b) what are significant factors
(c) what is the optimum condition and how much improvement is expected from the
optimum design
(d) what is the potential $ savings from the new design
(f) How much was improvement in variation(Cpk, Cp, etc.) etc.

Class Project Report Date:________
(Cover Sheet)

Project Title _________________________________ Location _____________________

Participants: 1.______________________ 2.____________________________

3._____________________ 4.____________________________

CRITERIA DES. Worst Value Best Value QC Rel. Weighting
-----------------------------------------------------------------------------------------------------------
1.

2.

3.

4.
etc.
-----------------------------------------------------------------------------------------------------------

OEC/RESULT = ( ) x + ( ) x + ( ) x + ( ) x

Example:

FACTORS Level 1 Level 2 Level 3 Level 4
---------------------------------------------------------------------------------------------------
1.

2.

3.

4.

etc.

Noise factors and Outer array :

NOTE: Show Design, results and analysis. Prepare report and make copies for all seminar attendees.


Variation Reduction Plot (Calculation of cost savings and capabilities indices)

S/N = - 10 Log (MSD) or MSD = 10
-[S/N]/10

MSD =
2
+ (y
avg
- y
0
)
2
for QC = Nominal
=
2
+ (y
avg
- 0)
2
for QC = Smaller
= [ 1 + 3
2
/ (y
avg
)
2
] / (y
avg
)
2
for QC = Bigger

When Average and S/N of current results are known, MSD and Std. Dev. can be calculated from above. If
the S/N at improved condition is available, the Std. Dev. can be estimated (Std. Dev. proportional to Square root
of MSD).
Since only an estimate of ( S/N) at improved condition is generally available at the completion of a designed
experiment, Std. Deviation, can only be approximated with the assumption that the average remain unchanged.

2
=
1

(MSD2 / MSD 1 )
0.5
( subscript: 1 - current and 2 - improved conditions)

When Standard Deviations in the current and improved condition are known, the corresponding Cp and
Cpk values can be calculated.

Cp = (UCL - LCL) / (6 )
Cpk = [ y
avg
- y
min
] / (3 ) for y
avg
< y
o

= [ y
max
- y
avg
] / (3 ) for y
avg
< y
o

Savings = 10
+ [ (S/N)1 - (S/N)2]/10
] x 100 as % of Loss at the current condition


(This page is intentionally left blank)

Program Evaluations

Program Title_____________________________________ ________________________________________________________
Program completion date

Instructor__________________________ Training Location/host.____________________________________

We appreciate your comments and suggestions. Please take a moment to let us know how we can improve and serve you better. (Please use the
following numbers for evaluation purposes).

[6] Excellent [5] Very Good [4] Satisfactory [3] Poor [2] Unsatisfactory [1] Needs Improvement

1. Overall Reaction to Program: 6 5 4 3 2 1 (Circle one number)

Comments:___________________________________________________________________________________________________

____________________________________________________________________________________________________________

2. Reaction to Instructor: 6 5 4 3 2 1 (Circle one number)

Comments:___________________________________________________________________________________________________

____________________________________________________________________________________________________________

____________________________________________________________________________________________________________

____________________________________________________________________________________________________________

3. What should be added and/or deleted to improve this program? (Please be specific)

____________________________________________________________________________________________________________

4. What portions of this program do you feel will be most helpful at your work? How?

____________________________________________________________________________________________________________

____________________________________________________________________________________________________________

5. Do you feel the training provided you enough understanding of the technique for you to be able to start applying it to your own projects if
opportunities were available?
[ ] Yes [ ] No [ ] May be

6. Would you consider/recommend us for training at your facility? [ ] yes [ ] No [ ] May be

Who should we contact for training at your facility __________________________________ Ph: _____________________

7. Would you recommend that others attend this seminar? [ ]YES [ ]NO

8. (Optional)
Your Name_____________________________________ Your Ph#/E-mail:___________________________________________
THANK YOU for taking the time to complete this evaluation.

(This page should not be duplicated or distributed)

Instruction for Duplication and Distribution of Seminar Notes

Preparing Hard Copy From WORD Document:

1. Unzipp the zipped file(s). The expanded files should constitute Module00.doc,
Module1a.doc, .... Module9a.doc and Module9b.doc.

2. Use MS WORD 6.0 or higher to view each module and check the format (once in a while
WORD in another system may shift the format and insert blank pages. There should be no
blank pages in the section without any note explaining the purpose.)

3. Print document starting from first file: Module00.doc.

4. Check for and missing pages or quality of printed pages. Watch for any DRWAING
BOXES (highlighted box) falling on the page breaks, overlapping graphics, etc.

5. Put all documents together and reproduce following the guidelines below.

Reproduction and Distribution Guidelines

1. Use paper for 3-ring binders (3-hole punched, preferably 20lb, 24lb or 65lb bond paper)

2. Copy double sided

3. Optionally, copy front and back page on card stock

4. Put each complete set in 3/4 inch binder (3 rings)

For any question, send E-mail to: rkroy@wwnet.com
Appendix: Exercises and Workshop Guidelines Page E - 1

Single Point Learning Exercises:
Please do this and other exercises with your partner. When completed, confirm your results
by sharing with other groups in the class.

Exercise 1: [Reference Module 2, Review questions 2-8 and 2-9, page 2-26]

Concept Experiment design and analysis using standard orthogonal arrays

In an effort to study the production problem (high reject rate) experienced in a machining
process, three factors among several possible causes were selected for a quick study. An L-4
experiment was designed to study the three 2-level factors as shown below. A large number of
samples were tested in each trial conditions and average performances were calculated in two
in tow measure objectives as shown below (Evaluation Criteria: surface finish and
Capability).

(a) Describe (recipe) the factor levels used to conduct 3
rd
experiment.

(b) Determine the Optimum Condition for the best surface finish.

(c) Determine the Optimum Condition for the best Capability.

(d) If experiment # 1 is considered the current performance, estimate the % improvement of
surface finish expected from the new process settings.

(Key answers: 1 2 2, 16.24 4.25 = 12, 1 1 2, 1.254 + .091 = 1.356)
Results
Trial#\ A B C Surface Finish (QC=S) Capability
(QC=B)
1 1 1 1 17 micron 1.26
2 1 2 2 12 1.32
3 2 1 1 16 0.28
4 2 2 1 20 1.16
Factors Level I Level II
A: Tool Type High Carbon Carbide Tip
B: Cuttine Speed 1500 rpm 2000 rpm
C: Feed Rate 2 mm/sec. 5 mm/sec.

Exercise 2: [Reference Module 3, Review questions 3-8 .. 3-11, page 3-16]

Concept Experiment design and analysis to study Interactions

Having completed the experiment described in Exercise 1, The project team decided to
expand the scope in their repeat experiment to include four Factors and three Interactions
among these factors. The experiment design and the results are as shown below.

(a) Describe the condition of the eighth experiment with result 14.
(b) Determine the more severe of the three Interactions included in the study (AxB, BxC, and
CxA)
Hint: Calculate A1B1, A1B2, A2B1, A2B2, etc. For each pair, the angle between the line is indicated by
difference between the end points differences, i.e., (A1B2 A1B1) (A2B2-A1B2). The angle between the lines
indicates the strength of interaction.

(c) Is the interaction between factors C and D present? If present, how does it compare with
the three Interactions included in the experiment.
Hint: It is possible to test for the presence of Interaction, even though no column is reserved for it. Calculate
C1D1, C2D2, etc.

(d) Determine the Optimum Condition without interaction effect.
(e) Prescribe the Optimum Condition when modified for interactions AxB, BxC, and CxA.
(f) How would your experiment design change if you wanted to study AxB and CxD
interactions.

[Key answers: AxB > BxC = CxD > CxA, 2 1 2 2, No change, 11.5 5.5 = 17 )

Factors Level- I Level- II
D: Tool Holder Current Design New Design

Interactions: AxB, BxC, and CxA
Trial# A B AxB C CxA BxC D Results(QC= Bigger)
1 1 1 1 1 1 1 1 8 units
2 1 1 1 2 2 2 2 12
3 1 2 2 1 1 2 2 10
4 1 2 2 2 2 1 1 11
5 2 1 2 1 2 1 2 15
6 2 1 2 2 1 2 1 13
7 2 2 1 1 2 2 1 9
8 2 2 1 2 1 1 2 14


Exercise 3: [Reference Module 4, Review questions 4-1, 4-2 and 4-12, page 2-26]

Concept Array modification to accommodate mixed level factors.

In the machining process optimization effort described in Exercise 2, the project team decided
to study four 2-level factors and one 3-level factor as described below. The modified array,
factor assignment, and the results are as shown.

(a) Determine the Main Effect of factor E (Plot)
(b) Describe the Optimum Condition (factor level described)
(c) Estimate the performance at the optimum condition.

[Key answers: 49, 57, 67, 1 1 1 2 1 55.75 20.75 = 35) )

Factors Level I Level II Level III
D: Tool Holder Current Design New Design
E: Part Complexity Simple Common Complex

Interactions: Not studied
Trial# E A B C D Results(QC= Smaller)
1 1 0 0 1 1 1 1 36 units
2 1 0 0 2 2 2 2 60
3 2 0 0 1 1 2 2 48
4 2 0 0 2 2 1 1 66
5 3 0 0 1 2 1 2 72
6 3 0 0 2 1 2 1 62
7 1 0 0 1 2 2 1 48
8 1 0 0 2 1 1 2 54


Exercise 4: [Reference Module 4, pages 4-10 . . 4-12]

Concept Combination Design (Special case of design)

A group of automotive transmission engineers ran an experiment to study the Effect of Drive
Shaft Parameters on Vibration. To expedite the study time, they studied three 3-level factors
and two 2-level factors using an L-9 array. The Factor descriptions, and the results are as
shown below.

(a) Describe the condition of the
9
th
experiment

Hint: Assume that the three levels: D1E1,
D1E2, and D2E1 of the four levels of (DE)
were assigned to column 1 and tested.

(b) Determine the Main Effects of factors
D and E which are both assigned to
Column 1.

(c) Optimum Condition

[Key answers: D2 E1 A3 B2 C1, (27.66 26.66), (18.66 26.66), A3 B1 C2 E1 D2 ]

Factors Level I Level II Level III
A: SY PD/OD Runout Lower Current Higher
B: U-Joint Angle Less than 0 deg 0 degree More than 0 deg.
C: Tube Runout Below Spec. At Spec. Above Spec.
D: Mount Compliant Rigid
E: Speed 30 MPH 70 MPH

Interactions: Not studied

Results
Expt# DE A B C (QC= Smaller)
1 1 1 1 1 29
2 1 2 2 2 24
3 1 3 3 3 27
4 2 1 2 3 32
5 2 2 3 1 30
6 2 3 1 2 21
7 3 1 3 2 18
8 3 2 1 3 22
9 3 3 2 1 16


Concept Robust Design using Noise Factors and Outer Array

Process scientists involved in Photolithography Process Optimization Study, carried out an
experiment with five 2-level Control Factors and three Noise Factors. The experiment design
and the results (Pattern size/consistency) obtained are as shown.

(a) Describe the recipe for the test that produce second result (25*) of the third trial
condition. Indicate the noise condition to which this experiment was exposed.
(b) Calculate the Main Effect of the Noise Factor: Developer Type (Z)

[Key answers: ____, Main effect of Z = 14.437 15.625)

Control Factors Level I Level II
A: Concentration 3.5% 5%
B: Exposure time 8 Seconds 15 Seconds
C: Development Time 60 Minutes 90 Minutes
D: Light Intensity Low High
E: Spin Speed 2000 rpm 4000rpm

Interactions: AxB, and BxC
Noise Factors Level I Level II
X: Wafer Size 1 mm Thick mm Thick
Y: Room Condition Air Conditioned No A/C
Z: Developer Type Fresh 2 Hour Old

Trial# A B AxB C D BxC E
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

Results (QC=Smaller)
(1) (2) (3) (4)
10 7 9 7
20 17 16 15
27 25* 30 23
32 34 33 30
2 5 2 4
3 5 4 2
26 12 10 16
17 15 7 16

Z 1 2 2 1

Y 1 2 1 2

X 1 1 2 2

Exercise 6: [Reference Module 5, pages 5-8 . . 5-10]
Concept Analyses using S/N ratios of results

For the Robust design of the Photolithography Process Optimization Study described in
Exercise 5, perform the Signal to Noise (S/N) ratio analysis and determine the following:

(a) Calculate the S/N ratios for results of trial # 7 and 8.

(b) Determine the Optimum Condition (Ignore AxB and BxC)

(c) Estimate the Performance at Optimum Condition in terms of S/N and in original units of
the results (Ignore AxB and BxC).

[Key answers: -24.68, -23.11, 2 1 1 1 1, -21.46 + 11.406 = -10.054 S/N => 3.182]

Trial# A B AxB C D BxC E
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

Results (QC=Smaller) S/N Ratios
(1) (2) (3) (4)
10 7 9 7 -18.44
20 17 16 15 -24.66
27 25* 30 23 -28.43
32 34 33 30 -30.18
2 5 2 4 -10.88
3 5 4 2 -11.30
26 12 10 16 _____
17 15 7 16 _____


Exercise 7: [Reference Module 6, Pages 6-8 . . 6-15]
Concept Computation of ANOVA Terms

Determine the percent influence of the factors included in the study (shown below) to the
variation of the results (One 4-level factor A and three 2-level factors B, C, and D ).

Calculate:

(a) S
T
= ?

Hint: C.F. = 190.125

(b) S
A
= 49/2 + 225/2 + ..

Hint:
A1 =7, A2=15, A3=10, A4=7
DOF = 3, Error S
e
= 0.125

V
A
= S
A
/DOF

F
A
= V
A
/ V
e

S
A
=

(c) P
A
=

Compute ANOVA terms for other factors as time permits.

[Key answers: CF= 190.125, ST = 28.875, SA=21.375, VA=7.125, FA=57, SA=21, PA=72.72]

Trial# A B C D Results
1 1 1 1 1 4
2 1 2 2 2 3
3 2 1 1 2 8
4 2 2 2 1 7
5 3 1 2 1 4
6 3 2 1 2 6
7 4 1 2 2 2
8 4 2 1 1 5
Total = 39

Concept Loss Function and Variation Reduction

In a study involving design parameters for automobile power window operations, the
optimum design condition was found to produce the average performance on the target and
improve the S/N ratios from 12.749 (at current) to -7.5. The time to close the window in the
current design (for 10 samples) is as shown below.

(a) Assuming that
Tolerance on closure time = 12 (+/-)
Cost of rejection / rework = $ 30/part
Production per month = 15,000 units
(USL = 6 LSL = 34)

Determine the savings expected when the
improved design is implemented.

(b) Calculate the standard deviation of results
expected from the improved design and
plot the graphs that display how variation is reduced. Assume that the distribution average
is on the target and that the standard deviation is directly proportional to the squared root
of MSD. (Standard deviation calculated using n instead of n-1 as the divisor)

MSD ( Std. Dev.)
2

(c) Calculate the expected value of the process capability, Cpk from the improved design.

[Key answers: $41,149/month, Std. Dev: 3.735 to 2.043, Cpk: 1.052 to 2.283]

Current
Performance
18 Sec. (total for 4 windows)
26 n = 10
25 Target = 20
16 Avg. = 22.2
22 Std. Dev.= 3.735
19
24 S/N = -12.742
26
27
19

Exercise 9: [Reference Module 9, Robust designs with Dynamic Characteristics]
Concept Experiment for Robust Design with Dynamic Characteristics

Engineers in a cold forming company planned an experiment to study one of the inherent
problems with cold heading process: the breakdown/wear of tooling dimensions due to
contact with the raw materials being formed. The study was initiated to find a suitable
combinaton of the process parameters whereby certain raw material coatings can improve tool
life.

Using Qualitek-4 software (Downloaded DEMO version from: http://www.rkroy.com or regular
program), design the experiment and analyze the results to determine the optimum condition.

[Key answers: Optimum cond 1 1 1 1, Perf. S/N 1.79 + 3.342 = 5.132]
Factors Level I Level II
A: Raw material type Type 1 Type 2
B: Material strength Lower Higher
C: Application method Method 1 Method 2
D: Lubricants Low Viscousity Higher Viscousity
Interactions: AxB, BxC, and CxA

Signal Factor: Material Size (3 levels: 0.25, 0.5, and 1.25)
Noise Factor: Cleaning Process (2 levels: Intermitten and Continuous)
Symptom: Excessive tool wear Intent: achieve longer tool life
Response: Tool Life Customer perceived result: Increased tool life
Trial# A B AxB C CxA BxC D Results(QC= Zero Point Prop. DC)
1 1 1 1 1 1 1 1 Tool Life
2 1 1 1 2 2 2 2
3 1 2 2 1 1 2 2 24 results/trial condition
4 1 2 2 2 2 1 1 4 samples for each cell.
5 2 1 2 1 2 1 2
6 2 1 2 2 1 2 1
7 2 2 1 1 2 2 1 (see below)
8 2 2 1 2 1 1 2

Trial# Results:(8 Trial 24 results/trial)
1 20,80,13,25,66,73,23,81,52,30,46,23,21,90,20,88,12,95,87,19,65,87,87,55
2 44,63,69,86,18,2,59,44,92,74,40,75,48,59,89,58,62,34,86,69,12,76,10,88
3 69,61,85,27,18,95,48,64,85,53,62,8,38,99,94,24,41,1,10,76,60,22,74,33
4 46,73,16,74,95,20,35,23,11,21,23,78,96,81,82,43,55,3,16,22,90,88,28,16
5 23,85,60,93,97,98,85,62,96,70,31,36,11,93,85,5,30,56,42,59,45,50,89,90
6 63,95,19,41,77,2,93,92,44,54,34,54,91,82,69,55,47,13,28,23,40,75,36,86
7 52,47,87,2,93,94,11,76,3,26,60,44,96,58,51,20,88,7,98,20,96,85,20,58
8 99,94,90,54,23,55,79,43,66,86,15,80,82,43,51,19,26,76,40,11,50,40,36,10


Exercise 10: Group Project [Reference Module 8, Pages 8-1 . . .8-9]
Concept Apply DOE/Taguchi technique to your own application project

1. Form team and select a project
2. Brainstorm to identify Evaluation Criteria, Factors, Levels, Interaction, Noise Factors, etc.
3. Design and describe experiment
4. Carry out or generate hypothetical results
5. Perform analysis (Assume data to be real if hypothetical results are accepted)
6. List observations and propose new design
7. Prepare REPORT containing:

8. Presentation to the class
Make copy of your report for all in the class + instructor.
Present your experimental findings and recommendations to the class

Use Qualitek-4 software for design and analysis.
Use software provided by the instructor or download from http://www.rkroy.com
Get help on experiment designs from http://www.rkroy.com/wp-tip.html

Report Content
1. Project Title
- include names of team members (participants in the study)
2. Brief Description of the project function and the purpose of the study
- Indicate reasons for the study and the benefits derived
3. Evaluation Criteria
- discuss how different objectives were measured
4. Factors and Levels
- explain how factors were selected from a larger list
- discuss the rationale for number of levels and their values
5. Noise Factors and Interactions, if any
- discuss how interaction(if included) were selected
- justify use of noise factors in the design (if included)
6. Orthogonal Array and the design
7. Main Effects indicate trend of influence of factors and interactions
8. ANOVA list factors with higher relative influence to the variation
9. Optimum Condition and Performance (convert if in S/N)
- indicate any factor level adjusted for interaction
10. Confidence Interval (C.I.)
11. Expected Savings from the new design
12. Conclusions and Recommendations
- based on the results, what do prpose for further study
- explain when would you know that the experiment is satisfactory



Program Evaluations

Program Title_____________________________________ ______________________________________________________
Program completion date

Instructor Ranjit K. Roy & Training Location/host.____________________________________

We appreciate your comments and suggestions. Please take a moment to let us know how we can improve and serve you better. (Please use the
following numbers for evaluation purposes).

[6] Excellent [5] Very Good [4] Satisfactory [3] Poor [2] Unsatisfactory [1] Needs Improvement

1. Overall Reaction to Program: 6 5 4 3 2 1 (Circle one number)

Comments:___________________________________________________________________________________________________

____________________________________________________________________________________________________________

2. Reaction to Instructor: 6 5 4 3 2 1 (Circle one number)

Comments:___________________________________________________________________________________________________

_____________________________________________________________________________________________________________

_____________________________________________________________________________________________________________

_____________________________________________________________________________________________________________

3. What should be added and/or deleted to improve this program? (Please be specific)

_____________________________________________________________________________________________________________

4. What portions of this program do you feel will be most helpful at your work? How?

_____________________________________________________________________________________________________________

_____________________________________________________________________________________________________________

5. Do you feel the training provided you enough understanding of the technique for you to be able to start applying it to your own projects if
opportunities were available?
[ ] Yes [ ] No [ ] May be

6. Would you consider/recommend us for training at your facility? [ ] yes [ ] No [ ] May be

Who should we contact for training at your facility __________________________________ Ph: _____________________

7. Would you recommend that others attend this seminar? [ ]YES [ ]NO

8. (Optional)
Your Name_____________________________________ Your Ph#/ & E-mail:_________________________________________
THANK YOU for taking the time to complete this evaluation.

Robust Design Modules

Transféré par

Informations du document

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Robust Design Modules

Transféré par

Droits d'auteur :

Formats disponibles

Design Of Experiments Using

Vous aimerez peut-être aussi