SlideShare a Scribd company logo
1 of 8
14.5 ORTHOGONAL ARRAYS
                         (Updated Spring 2003)

Orthogonal Arrays (often referred to Taguchi Methods) are often employed
in industrial experiments to study the effect of several control factors.

Popularized by G. Taguchi. Other Taguchi contributions include:
ā€¢ Model of the Engineering Design Process
ā€¢ Robust Design Principle
ā€¢ Efforts to push quality upstream into the engineering design process

An orthogonal array is a type of experiment where the columns for the
independent variables are ā€œorthogonalā€ to one another.

Benefits:

1. Conclusions valid over the entire region spanned by the control factors
   and their settings
2. Large saving in the experimental effort
3. Analysis is easy

To define an orthogonal array, one must identify:
1. Number of factors to be studied
2. Levels for each factor
3. The specific 2-factor interactions to be estimated
4. The special difficulties that would be encountered in running the
   experiment

We know that with two-level full factorial experiments, we can estimate
variable interactions. When two-level fractional factorial designs are used,
we begin to confound our interactions, and often lose the ability to obtain
unconfused estimates of main and interaction effects. We have also seen
that if the generators are choosen carefully then knowledge of lower order
interactions can be obtained under that assumption that higher order
interactions are negligible.

Orthogonal arrays are highly fractionated factorial designs. The
information they provide is a function of two things
ā€¢ the nature of confounding
ā€¢ assumptions about the physical system.

Before proceeding further with a closer look, letā€™s look at an example
where orthogonal arrays have been employed.

Example taken from students of Alice Agogino at UC-Berkeley



                                           A
Airplane Taguchi Experiment

This experiment has 4 variables at 3 different settings. A full factorial
experiment would require 34 = 81 experiments. We conducted a Taguchi
experiment with a L9(34) orthogonal array (9 tests, 4 variables, 3 levels).
The experiment design is shown below.

    Experi-       Weight       Stabilizer       Noise         Wing
     ment          A               B             C             D
      #
       1            A1             B1            C1             D1
       2            A1             B2            C2             D2
       3            A1             B3            C3             D3
       4            A2             B1            C2             D3
       5            A2             B2            C3             D1
       6            A2             B3            C1             D2
       7            A3             B1            C3             D2
       8            A3             B2            C1             D3
       9            A3             B3            C2             D1




                                            B
The students that performed this experiment suggest that the S/N ratio
graph should be critically examined to select the desired variable levels.

Variable levels: A2 / B1 / C3 / D1

PROBLEM: Traditionally those that employ orthogonal arrays adopt a
ā€œpick the winnerā€ approach. Donā€™t take advantage of the full information
provided by the experiment. No model, no residuals, no sequential
approach to experimentation.



                                           C
With Orthogonal Arrays / Taguchi Methods, the approach to design of
experiments, the philosophy towards interaction effects is different. Unless
an interaction effect is explicity identified a-priori and assigned to column
in the orthogonal array, it is assumed to be negligible. Thus, unless
assumed otherwise, interactions at all levels are assumed to be negligible.

Taguchi encourages users to select variables that do not interact with one
another. A VERY INTERESTING IDEA! However, this idea may be hard
to realize in practice.

Orthogonal arrays are frequently listed in handbooks/manuals. Users often
extract the listed designs and use them blindly without thinking. NOTE OF
CAUTION: choose orthogonal arrays wisely -- select the array with
highest possible resolution so as to minimize the effect of any erroneous
assumptions we make regarding effects being negligible.

Example

An arc welding experiment performed by national railway cooperation of
Japan in 1959. Nine variables were studied as shown in the table:


                  Factor                 Lower level       Higher Level

 A - Kind of welding tool                      J100             B17
 B - Drying method of rods                 no drying          one-day
 C - Welded material                           SS41            SB35
 D - Thickness of welded material            8 mm              12 mm
 E - Angle of welding Device                70 deg.           60 deg.
 F - Opening of Welding Device              1.5 mm             3 mm
 G - Current                                150 A              130 A
 H - Welding methods                       weaving             single
 I - Preheating                          no preheating     150 deg. pre-
                                                             heating


Nine variables and four two factor interactions were studied using a
L16(215) orthogonal array as shown

16 tests (2 levels for each main variable) -- 15 columns (factors) studied


                                           D
Test    A   G     AG     H    AH    GH    B    D    E   F    I   e   e   C   AC

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


The columns labeled ā€œeā€ may be used for error estimation


Letā€™s take a hard look at this experiment using what we know about
fractional factorial designs.

This may be written as 29-5 fractional factorial as follows




                                           E
D          H       G        A          B       E         F   I   C
  Test
             1          2       3        4          5       6         7   8   9

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


This design is a 2III9-5 fractional factorial design with the following
generators.

I = 2345,    I = -146       I = -137     I = 1348       I = -12349

Defining relations

I = 2345 = -146 = -137 = 1348 = -12349
  = -12356 = -12457 = 1258 = -159 = 3467 = -368 = 2369 = -478 = 2479
  = -289 = 2567 = -24568 = 4569 = -23578 = 3579 = -34589 = 1678
  = -12679 = 124689 = 123789 = 12345678 = -135689 = 125689 = 145789
  = -2346789 = -56789

Assuming third and higher order interactions to be negligible the


                                              F
confounding pattern is as follows:

lI est I(mean)                         l8 est 23 + 45 + 69
l1 est 1 - 46 - 37 - 59                l9 est 24 + 35 + 79
l2 est 2 - 89                          l10 est 34 + 25 + 18 + 67
l3 est 3 - 17 - 68                     l11 est -27 - 49 - 56
l4 est 4 - 16 - 78                     l12 est -26 -39 -57 (error)
l5 est 12 + 58(error)                  l13 est 8 - 36 - 47 - 29
l6 est -7 + 13 + 48                    l14 est 5 - 19
l7 est -6 + 14 + 38                    l15 est -9 + 15 +28

What happened to our standard procedure?? We should be using
lI l1 l2 l3 l4 l12 l13 l14 l23 l24 l34 l123 l124 l134 l234 l1234

Clearly the above design is a resolution III design.

It is not possible to have a 16-run two-level fractional factorial design with
nine variables and resolution greater than III. However, we could have
chosen better gereators. We can pick generators in such a way so that the
confounding is better than what we obtain ā€œby defaultā€.


As an alternative, consider a design with the following generators

I = 1235,     I = 2346     I = 1347      I = 1248      I = 12349

The confounding relations are as follows

lI est I(mean)                         l8 est 23 + 15 + 46 + 78
l1 est 1 + 69                          l9 est 24 + 36 + 18 + 57
l2 est 2 + 79                          l10 est 34 + 26 + 17 + 58
l3 est 3 + 89                          l11 est 5 + 49
l4 est 4 + 59                          l12 est 8 + 39
l5 est 12 + 35 + 48 + 67               l13 est 7 + 29
l6 est 13 + 25 + 47 + 68               l14 est 6 + 19
l7 est 14 + 37 + 28 + 56               l15 est 9 + 45 + 16 + 27 + 38

The main effects are still confounded with at least one two factor
interaction. But 8 out of the 9 main effects are confounded with only one 2
factor interaction.




                                            G
Another point to note is that all the 2 factor interactions (associated with
the main effects) have variable 9 in them

Hence if we know that there is some variable that is least likely to be
important -- or that does not interact with the other variables -- then we can
assign it to be variable 9.

CONCLUSION:
ā€¢ Idea of publishing a standard set of designs (orthogonal arrays is a great
  idea) -- Taguchi brought the field of DOE to the masses
ā€¢ Select variables that donā€™t interact -- interesting!! Assuming that inter-
  actions do not exist doesnā€™t mean they arenā€™t there!
ā€¢ As a precaution select design generators that give best design design
  resolution and that provide desirable confounding structure.
ā€¢ Other??




                                            H

More Related Content

What's hot (8)

Plcyoth
PlcyothPlcyoth
Plcyoth
Ā 
Ejercicio 211 del libro de Baldor
Ejercicio 211 del libro de BaldorEjercicio 211 del libro de Baldor
Ejercicio 211 del libro de Baldor
Ā 
Configuracion wds
Configuracion wdsConfiguracion wds
Configuracion wds
Ā 
Trial penang 2014 spm matematik tambahan k1 k2 skema [scan]
Trial penang 2014 spm matematik tambahan k1 k2 skema [scan]Trial penang 2014 spm matematik tambahan k1 k2 skema [scan]
Trial penang 2014 spm matematik tambahan k1 k2 skema [scan]
Ā 
Function problem p
Function problem pFunction problem p
Function problem p
Ā 
C1 january 2012_mark_scheme
C1 january 2012_mark_schemeC1 january 2012_mark_scheme
C1 january 2012_mark_scheme
Ā 
UGC NET COMPUTER SCIENCE JUNE 2010 PAPER-II
UGC NET COMPUTER SCIENCE JUNE 2010 PAPER-IIUGC NET COMPUTER SCIENCE JUNE 2010 PAPER-II
UGC NET COMPUTER SCIENCE JUNE 2010 PAPER-II
Ā 
Limits At Infinity Part2 12.17.07
Limits At Infinity Part2 12.17.07Limits At Infinity Part2 12.17.07
Limits At Infinity Part2 12.17.07
Ā 

Similar to Orth Arrays

Cluto presentation
Cluto presentationCluto presentation
Cluto presentation
Roseline Antai
Ā 
Ejercicios asignados a yonathan david diaz granados
Ejercicios asignados a yonathan david diaz granadosEjercicios asignados a yonathan david diaz granados
Ejercicios asignados a yonathan david diaz granados
YonathanDavidDiazGra
Ā 
2.2 add real numbers day 1-2
2.2 add real numbers   day 1-22.2 add real numbers   day 1-2
2.2 add real numbers day 1-2
bweldon
Ā 

Similar to Orth Arrays (20)

Absolute Value Notes
Absolute Value NotesAbsolute Value Notes
Absolute Value Notes
Ā 
Absolute Value
Absolute ValueAbsolute Value
Absolute Value
Ā 
Graph functions
Graph functionsGraph functions
Graph functions
Ā 
Prelim exam algebra
Prelim exam   algebraPrelim exam   algebra
Prelim exam algebra
Ā 
Inventory Model with Different Deterioration Rates with Stock and Price Depen...
Inventory Model with Different Deterioration Rates with Stock and Price Depen...Inventory Model with Different Deterioration Rates with Stock and Price Depen...
Inventory Model with Different Deterioration Rates with Stock and Price Depen...
Ā 
Ejercicio 211 del libro de baldor
Ejercicio 211 del libro de baldorEjercicio 211 del libro de baldor
Ejercicio 211 del libro de baldor
Ā 
zav
zavzav
zav
Ā 
2c astable monostable
2c astable monostable2c astable monostable
2c astable monostable
Ā 
Combined Line - Double Bubbles - Pie chart.
Combined Line - Double Bubbles - Pie chart.Combined Line - Double Bubbles - Pie chart.
Combined Line - Double Bubbles - Pie chart.
Ā 
ejercicio 211 del libro de Baldor
ejercicio 211 del libro de Baldorejercicio 211 del libro de Baldor
ejercicio 211 del libro de Baldor
Ā 
Chapter 7: Matrix Multiplication
Chapter 7: Matrix MultiplicationChapter 7: Matrix Multiplication
Chapter 7: Matrix Multiplication
Ā 
Stack application
Stack applicationStack application
Stack application
Ā 
EJES Y ARBOLES
EJES Y ARBOLESEJES Y ARBOLES
EJES Y ARBOLES
Ā 
Algebra and Trigonometry 9th Edition Larson Solutions Manual
Algebra and Trigonometry 9th Edition Larson Solutions ManualAlgebra and Trigonometry 9th Edition Larson Solutions Manual
Algebra and Trigonometry 9th Edition Larson Solutions Manual
Ā 
Cluto presentation
Cluto presentationCluto presentation
Cluto presentation
Ā 
LAMP_TRAINING_SESSION_6
LAMP_TRAINING_SESSION_6LAMP_TRAINING_SESSION_6
LAMP_TRAINING_SESSION_6
Ā 
Puanumrahdalimon
PuanumrahdalimonPuanumrahdalimon
Puanumrahdalimon
Ā 
Chapter 2
Chapter 2Chapter 2
Chapter 2
Ā 
Ejercicios asignados a yonathan david diaz granados
Ejercicios asignados a yonathan david diaz granadosEjercicios asignados a yonathan david diaz granados
Ejercicios asignados a yonathan david diaz granados
Ā 
2.2 add real numbers day 1-2
2.2 add real numbers   day 1-22.2 add real numbers   day 1-2
2.2 add real numbers day 1-2
Ā 

More from nazeer pasha

Tc Checklist
Tc ChecklistTc Checklist
Tc Checklist
nazeer pasha
Ā 
Software Testing Guide
Software Testing GuideSoftware Testing Guide
Software Testing Guide
nazeer pasha
Ā 
Cstp Certification Compare
Cstp Certification CompareCstp Certification Compare
Cstp Certification Compare
nazeer pasha
Ā 
Blackboxtesting 02 An Example Test Series
Blackboxtesting 02 An Example Test SeriesBlackboxtesting 02 An Example Test Series
Blackboxtesting 02 An Example Test Series
nazeer pasha
Ā 
Exploratory Testing
Exploratory TestingExploratory Testing
Exploratory Testing
nazeer pasha
Ā 
Chanakya Niti
Chanakya NitiChanakya Niti
Chanakya Niti
nazeer pasha
Ā 
Unit Testing
Unit TestingUnit Testing
Unit Testing
nazeer pasha
Ā 
Testing Types And Models
Testing Types And ModelsTesting Types And Models
Testing Types And Models
nazeer pasha
Ā 
Testing Framework
Testing FrameworkTesting Framework
Testing Framework
nazeer pasha
Ā 

More from nazeer pasha (20)

Linux
LinuxLinux
Linux
Ā 
Tomcat Configuration (1)
Tomcat Configuration (1)Tomcat Configuration (1)
Tomcat Configuration (1)
Ā 
Test Techniques
Test TechniquesTest Techniques
Test Techniques
Ā 
Testing Types Presentation
Testing Types PresentationTesting Types Presentation
Testing Types Presentation
Ā 
Good Ppt On Risk
Good Ppt On RiskGood Ppt On Risk
Good Ppt On Risk
Ā 
Bug Advocacy
Bug AdvocacyBug Advocacy
Bug Advocacy
Ā 
Doe Taguchi Basic Manual1
Doe Taguchi Basic Manual1Doe Taguchi Basic Manual1
Doe Taguchi Basic Manual1
Ā 
Teaching Testing Qw%202001
Teaching Testing Qw%202001Teaching Testing Qw%202001
Teaching Testing Qw%202001
Ā 
Testing
TestingTesting
Testing
Ā 
Tc Checklist
Tc ChecklistTc Checklist
Tc Checklist
Ā 
Software Testing Guide
Software Testing GuideSoftware Testing Guide
Software Testing Guide
Ā 
Cstp Certification Compare
Cstp Certification CompareCstp Certification Compare
Cstp Certification Compare
Ā 
Blackboxtesting 02 An Example Test Series
Blackboxtesting 02 An Example Test SeriesBlackboxtesting 02 An Example Test Series
Blackboxtesting 02 An Example Test Series
Ā 
Exploratory Testing
Exploratory TestingExploratory Testing
Exploratory Testing
Ā 
Chanakya Niti
Chanakya NitiChanakya Niti
Chanakya Niti
Ā 
Unit Testing
Unit TestingUnit Testing
Unit Testing
Ā 
Testing
TestingTesting
Testing
Ā 
Testing Types And Models
Testing Types And ModelsTesting Types And Models
Testing Types And Models
Ā 
Swtesting
SwtestingSwtesting
Swtesting
Ā 
Testing Framework
Testing FrameworkTesting Framework
Testing Framework
Ā 

Recently uploaded

+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
?#DUbAI#??##{{(ā˜Žļø+971_581248768%)**%*]'#abortion pills for sale in dubai@
Ā 

Recently uploaded (20)

The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024The 7 Things I Know About Cyber Security After 25 Years | April 2024
The 7 Things I Know About Cyber Security After 25 Years | April 2024
Ā 
Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024Tata AIG General Insurance Company - Insurer Innovation Award 2024
Tata AIG General Insurance Company - Insurer Innovation Award 2024
Ā 
Partners Life - Insurer Innovation Award 2024
Partners Life - Insurer Innovation Award 2024Partners Life - Insurer Innovation Award 2024
Partners Life - Insurer Innovation Award 2024
Ā 
Strategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a FresherStrategies for Landing an Oracle DBA Job as a Fresher
Strategies for Landing an Oracle DBA Job as a Fresher
Ā 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdf
Ā 
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
Ā 
Understanding Discord NSFW Servers A Guide for Responsible Users.pdf
Understanding Discord NSFW Servers A Guide for Responsible Users.pdfUnderstanding Discord NSFW Servers A Guide for Responsible Users.pdf
Understanding Discord NSFW Servers A Guide for Responsible Users.pdf
Ā 
Developing An App To Navigate The Roads of Brazil
Developing An App To Navigate The Roads of BrazilDeveloping An App To Navigate The Roads of Brazil
Developing An App To Navigate The Roads of Brazil
Ā 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected Worker
Ā 
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, AdobeApidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Apidays New York 2024 - Scaling API-first by Ian Reasor and Radu Cotescu, Adobe
Ā 
Handwritten Text Recognition for manuscripts and early printed texts
Handwritten Text Recognition for manuscripts and early printed textsHandwritten Text Recognition for manuscripts and early printed texts
Handwritten Text Recognition for manuscripts and early printed texts
Ā 
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data DiscoveryTrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
TrustArc Webinar - Unlock the Power of AI-Driven Data Discovery
Ā 
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Ā 
A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?A Year of the Servo Reboot: Where Are We Now?
A Year of the Servo Reboot: Where Are We Now?
Ā 
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...Workshop - Best of Both Worlds_ Combine  KG and Vector search for  enhanced R...
Workshop - Best of Both Worlds_ Combine KG and Vector search for enhanced R...
Ā 
Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...Apidays New York 2024 - The value of a flexible API Management solution for O...
Apidays New York 2024 - The value of a flexible API Management solution for O...
Ā 
Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)Powerful Google developer tools for immediate impact! (2023-24 C)
Powerful Google developer tools for immediate impact! (2023-24 C)
Ā 
AWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of TerraformAWS Community Day CPH - Three problems of Terraform
AWS Community Day CPH - Three problems of Terraform
Ā 
GenAI Risks & Security Meetup 01052024.pdf
GenAI Risks & Security Meetup 01052024.pdfGenAI Risks & Security Meetup 01052024.pdf
GenAI Risks & Security Meetup 01052024.pdf
Ā 
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Apidays Singapore 2024 - Building Digital Trust in a Digital Economy by Veron...
Ā 

Orth Arrays

  • 1. 14.5 ORTHOGONAL ARRAYS (Updated Spring 2003) Orthogonal Arrays (often referred to Taguchi Methods) are often employed in industrial experiments to study the effect of several control factors. Popularized by G. Taguchi. Other Taguchi contributions include: ā€¢ Model of the Engineering Design Process ā€¢ Robust Design Principle ā€¢ Efforts to push quality upstream into the engineering design process An orthogonal array is a type of experiment where the columns for the independent variables are ā€œorthogonalā€ to one another. Benefits: 1. Conclusions valid over the entire region spanned by the control factors and their settings 2. Large saving in the experimental effort 3. Analysis is easy To define an orthogonal array, one must identify: 1. Number of factors to be studied 2. Levels for each factor 3. The specific 2-factor interactions to be estimated 4. The special difficulties that would be encountered in running the experiment We know that with two-level full factorial experiments, we can estimate variable interactions. When two-level fractional factorial designs are used, we begin to confound our interactions, and often lose the ability to obtain unconfused estimates of main and interaction effects. We have also seen that if the generators are choosen carefully then knowledge of lower order interactions can be obtained under that assumption that higher order interactions are negligible. Orthogonal arrays are highly fractionated factorial designs. The information they provide is a function of two things ā€¢ the nature of confounding ā€¢ assumptions about the physical system. Before proceeding further with a closer look, letā€™s look at an example where orthogonal arrays have been employed. Example taken from students of Alice Agogino at UC-Berkeley A
  • 2. Airplane Taguchi Experiment This experiment has 4 variables at 3 different settings. A full factorial experiment would require 34 = 81 experiments. We conducted a Taguchi experiment with a L9(34) orthogonal array (9 tests, 4 variables, 3 levels). The experiment design is shown below. Experi- Weight Stabilizer Noise Wing ment A B C D # 1 A1 B1 C1 D1 2 A1 B2 C2 D2 3 A1 B3 C3 D3 4 A2 B1 C2 D3 5 A2 B2 C3 D1 6 A2 B3 C1 D2 7 A3 B1 C3 D2 8 A3 B2 C1 D3 9 A3 B3 C2 D1 B
  • 3. The students that performed this experiment suggest that the S/N ratio graph should be critically examined to select the desired variable levels. Variable levels: A2 / B1 / C3 / D1 PROBLEM: Traditionally those that employ orthogonal arrays adopt a ā€œpick the winnerā€ approach. Donā€™t take advantage of the full information provided by the experiment. No model, no residuals, no sequential approach to experimentation. C
  • 4. With Orthogonal Arrays / Taguchi Methods, the approach to design of experiments, the philosophy towards interaction effects is different. Unless an interaction effect is explicity identified a-priori and assigned to column in the orthogonal array, it is assumed to be negligible. Thus, unless assumed otherwise, interactions at all levels are assumed to be negligible. Taguchi encourages users to select variables that do not interact with one another. A VERY INTERESTING IDEA! However, this idea may be hard to realize in practice. Orthogonal arrays are frequently listed in handbooks/manuals. Users often extract the listed designs and use them blindly without thinking. NOTE OF CAUTION: choose orthogonal arrays wisely -- select the array with highest possible resolution so as to minimize the effect of any erroneous assumptions we make regarding effects being negligible. Example An arc welding experiment performed by national railway cooperation of Japan in 1959. Nine variables were studied as shown in the table: Factor Lower level Higher Level A - Kind of welding tool J100 B17 B - Drying method of rods no drying one-day C - Welded material SS41 SB35 D - Thickness of welded material 8 mm 12 mm E - Angle of welding Device 70 deg. 60 deg. F - Opening of Welding Device 1.5 mm 3 mm G - Current 150 A 130 A H - Welding methods weaving single I - Preheating no preheating 150 deg. pre- heating Nine variables and four two factor interactions were studied using a L16(215) orthogonal array as shown 16 tests (2 levels for each main variable) -- 15 columns (factors) studied D
  • 5. Test A G AG H AH GH B D E F I e e C AC 1 - - - - - - - - - - - - - - - 2 - - - - - - - + + + + + + + + 3 - - - + + + + - - - - + + + + 4 - - - + + + + + + + + - - - - 5 - + + - - + + - - + + - - + + 6 - + + - - + + + + - - + + - - 7 - + + + + - - - - + + + + - - 8 - + + + + - - + + - - - - + + 9 + - + - + - + - + - + - + - + 10 + - + - + - + + - + - + - + - 11 + - + + - + - - + - + + - + - 12 + - + + - + - + - + - - + - + 13 + + - - + + - - + + - - + + - 14 + + - - + + - + - - + + - - + 15 + + - + - - + - + + - + - - + 16 + + - + - - + + - - + - + + - The columns labeled ā€œeā€ may be used for error estimation Letā€™s take a hard look at this experiment using what we know about fractional factorial designs. This may be written as 29-5 fractional factorial as follows E
  • 6. D H G A B E F I C Test 1 2 3 4 5 6 7 8 9 1 - - - - - - - - - 2 + - - - - + + + + 3 - + - - + - - - - 4 + + - - + + + + - 5 - - + - + - + + + 6 + - + - + + - - - 7 - + + - - - + + - 8 + + + - - + - - + 9 - - - + + + - + + 10 + - - + + - + - - 11 - + - + - + - + - 12 + + - + - - + - + 13 - - + + - + + - - 14 + - + + - - - + + 15 - + + + + + + - + 16 + - + + + + - + - This design is a 2III9-5 fractional factorial design with the following generators. I = 2345, I = -146 I = -137 I = 1348 I = -12349 Defining relations I = 2345 = -146 = -137 = 1348 = -12349 = -12356 = -12457 = 1258 = -159 = 3467 = -368 = 2369 = -478 = 2479 = -289 = 2567 = -24568 = 4569 = -23578 = 3579 = -34589 = 1678 = -12679 = 124689 = 123789 = 12345678 = -135689 = 125689 = 145789 = -2346789 = -56789 Assuming third and higher order interactions to be negligible the F
  • 7. confounding pattern is as follows: lI est I(mean) l8 est 23 + 45 + 69 l1 est 1 - 46 - 37 - 59 l9 est 24 + 35 + 79 l2 est 2 - 89 l10 est 34 + 25 + 18 + 67 l3 est 3 - 17 - 68 l11 est -27 - 49 - 56 l4 est 4 - 16 - 78 l12 est -26 -39 -57 (error) l5 est 12 + 58(error) l13 est 8 - 36 - 47 - 29 l6 est -7 + 13 + 48 l14 est 5 - 19 l7 est -6 + 14 + 38 l15 est -9 + 15 +28 What happened to our standard procedure?? We should be using lI l1 l2 l3 l4 l12 l13 l14 l23 l24 l34 l123 l124 l134 l234 l1234 Clearly the above design is a resolution III design. It is not possible to have a 16-run two-level fractional factorial design with nine variables and resolution greater than III. However, we could have chosen better gereators. We can pick generators in such a way so that the confounding is better than what we obtain ā€œby defaultā€. As an alternative, consider a design with the following generators I = 1235, I = 2346 I = 1347 I = 1248 I = 12349 The confounding relations are as follows lI est I(mean) l8 est 23 + 15 + 46 + 78 l1 est 1 + 69 l9 est 24 + 36 + 18 + 57 l2 est 2 + 79 l10 est 34 + 26 + 17 + 58 l3 est 3 + 89 l11 est 5 + 49 l4 est 4 + 59 l12 est 8 + 39 l5 est 12 + 35 + 48 + 67 l13 est 7 + 29 l6 est 13 + 25 + 47 + 68 l14 est 6 + 19 l7 est 14 + 37 + 28 + 56 l15 est 9 + 45 + 16 + 27 + 38 The main effects are still confounded with at least one two factor interaction. But 8 out of the 9 main effects are confounded with only one 2 factor interaction. G
  • 8. Another point to note is that all the 2 factor interactions (associated with the main effects) have variable 9 in them Hence if we know that there is some variable that is least likely to be important -- or that does not interact with the other variables -- then we can assign it to be variable 9. CONCLUSION: ā€¢ Idea of publishing a standard set of designs (orthogonal arrays is a great idea) -- Taguchi brought the field of DOE to the masses ā€¢ Select variables that donā€™t interact -- interesting!! Assuming that inter- actions do not exist doesnā€™t mean they arenā€™t there! ā€¢ As a precaution select design generators that give best design design resolution and that provide desirable confounding structure. ā€¢ Other?? H