Design Of Experiment

(Full Factorial Design)

1
 Design Of Experiment (Full Factorial Design)

Learning Objective
 Basic Concept of Design Of Experiment
 Factor, level, response, main effect, interaction
 Randomization, replication, standard order, random
order, curvature, response surface, etc.
 Center point, reduced model, residual analysis
 Design, conduct and analyze a full 2 2 factorial
experiment
 Use Minitab to create, analyze and interpret
various DOE summary and plots.
 Design, conduct and analyze a full 2 3 factorial
experiment

Seagate Confidential 2 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Design Of Experiment – (DOE)

In Statistics, an experiment refers to any
process that generates a set of data.

An experimental design involves making

deliberate changes of the inputs to a process
(or product) and observing the corresponding
changes in the outputs.

We then draw some conclusion (or inference)

about which variables are responsible for the
observed changes in the outputs.

Seagate Confidential 3 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

DOE – Applications
 Product Design
 identification of key design variables
 design optimization
 preliminary assessment of product variability

 Process Design
 identification of key process variables
 assessment of effects of uncontrolled variables
 process optimization
 process capability studies

 Process Operation
 determination of component tolerances
 on-line process monitoring
 on-line process optimization
Seagate Confidential 4 Supplier Six Sigma Modular Training
Let’s begin with Basic Concept…

5
 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

Warm-up Example
Everyday when you wake up in the morning, after all the
wash up, you need to drink a cup of coffee. What does it
take to make a cup of good coffee? What are the factors that
affect the quality of the drink?
Noise
Factors (uncontrolled variables)
 ~ NID(0 , ²)
(controlled variables) Response
 Coffee power
Coffee Wow, good coffee!
 Water ^
Y = Y+
Temperature Making
 Sweetener
Process

Seagate Confidential 6 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

Relationship between Factors and Levels
[ k Factors ] and [ n Levels ]
 How many Factors are
1. Coffee power there ? What is the
I. Imported Brand
numbers of level for each
II. Local Brand
factors?
2. Water Temperature [Factor]
I. Boiling Hot. [Level]
II. Hot
 What other factor can
3. Sweetener
affect the taste of the
I. Milk
Coffee ?
II. Sugar

Seagate Confidential 7 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

[ 3 Factors ] with [ 2 Levels ] 23
1. Coffee power  How about the
I. Imported Brand combined action of the
II. Local Brand coffee power and water
temperature? Will that
2. Water Temperature joint influence make the
I. Boiling Hot. coffee taste better than
Hot
II.
expected?
We shall explore this
3. Sweetener
later.
I. Milk
II. Sugar

Seagate Confidential 8 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

Another Example - Real World Situation
In a casting process, two KPIVs, Temperature of the mold and
Pressure force of the material flow are suspected to influence
hardness of the cast product. You decide to investigate this
relationship using Temperature: 80C and 100C, and Pressure:
1 ton and 2 tons.
How many factors are there ? What is the number of level?
What is your the response?
Noise
Factors (uncontrolled variables)  ~ NID(0 , ²)
Response
(controlled variables)
 Temperature Hardness
Processes
80 100 ^
Y = Y+
 Pressure
1 ton 2 tons
Seagate Confidential 9 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

Why DOE !

We have learnt many Statistical methods, (T-test,

F-test, ANOVA….etc). How does DOE come in ?

DOE helps us to Plan, Study, Run experiments

with all possible treatment settings at the same
time, with randomization and replication to handle
noise variables.

Seagate Confidential 10 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

DOE – General Model

Noise
Factors (uncontrolled variables)
 ~ NID(0 , ²)
(controlled variables) Response
X1
Product
X2 ^
Y = Y+
or
:
Process
X4

Seagate Confidential 11 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

Factorial Designs
 When several factors are potentially important, the best
strategy of experimentation is to design a factorial experiment.
 A factorial experiment is an experiment in which factors are
varied together at the same time.
 Factorial experiment are the only way to investigate interaction
effects.
Noise
Factors (uncontrolled variables)
 ~ NID(0 , ²)
(controlled variables) Response
X1
Product
X2 ^
Y = Y+
or
:
Process
X4
Seagate Confidential 12 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

2k Full Factorial Design
 The simplest factorial experiment consists of two factors A and
B, each at two levels. We usually denote these levels as low (-
1) and high (+1) level of the factor.
 The 22 design can be represented as a square with 2 2 = 4 runs
(or treatment combinations). Each corner denotes a particular
treatment combination.
( - , +) Factor B ( + , +) The 22 factorial design
+1
Treatment A B
1 - -
Factor A
-1 +1
2 + -
3 - +
-1 4 + +
( - , -) ( + , -)
Seagate Confidential 13 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

Example 1
 Design a full factorial experiment to study the effect of
(A) Heat treatment of the bearing race, (B) Osculation of
ball and (C) Cage Design, to the Ball Bearing Life.

Hints:
R
R One Factor Two Factors
rr A BC
_ _ _

Osculation= r/R + + _

_
+
+ +

Seagate Confidential 14 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE

2k Full Factorial Design
 Two level Design
Number of Number of
Factors Runs A AB ABC
1 2 _ _ _ _ _ _
2 4
+ + _
+ _ _
3 8
4 16
_
+ _
+ _
5 32 + + + + _
6 64
_ _
+
7 128 + _ +
8 256 _
+ +
9 512 + + +
10 1024

Seagate Confidential 15 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE- Flow Chart

State practical problem in “English”
Establish objective

Translate into Statistical Design of Experiments

Identify factors, levels & response

Create Factorial Design

Select type of Design: Full or Fraction, Factors Response
Factors, center points, replicates, blocks

Run the Experiment State practical

implications in “English”
Data Tabulation

Analyze Factorial Design Interpretation, Conclusion

- Pareto chart, Factorial plot With Engg. Knowledge & Statistics
Seagate Confidential 16 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

State Practical Problem

State practical problem in “English”  Step 1: Practical Problem:
Establish objective
 A process engineer is tasked to improve
the hardness of their product. There are
Translate into Statistical Design of Experiments two important KPIVs identified; Mold
Temperature and Pressure force of the
Identify Factors, levels & the Response

material flow.
Create Factorial Design  His objective is to study the effect of the
Select type of Design Full or Fraction,
Factors,center points, replicates,blocks Mold Temperature and Pressure of the
material flow to the Hardness of the
Casting.
Run the Experiment
Noise
Date Tabulation

Factors (uncontrolled variables)

 ~ NID(0 , ²)
(controlled variables)
Analyze Factorial Design
-Pareto chart, Factorial plot Response
Temperature
Product Hardness
Interpretation, Conclusion
With Engg. Knowledge & Statistics
or Y = Y+
^
State practical implications in “English”
Pressure Process
Seagate Confidential 17 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Translate into Statistical DOE

State practical problem in “English”
Establish objective

 Step 2: Translate into statistical DOE

Translate into Statistical Design of Experiments
Identify Factors, levels & the Response
 Identify the factors, levels and
response.
Create Factorial Design
Select type of Design Full or Fraction,
Factors,center points, replicates,blocks

Factors
Noise
(controlled variables)(uncontrolled variables)  ~
Run the Experiment
Date Tabulation
NID(0 , ²)

Temperature Response
Analyze Factorial Design (measured
-Pareto chart, Factorial plot 80 100 Product outcome)
or
Pressure Hardness
Interpretation, Conclusion Process
1 ton 2 tons
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 18 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

State Practical Problem

 Step 2: Translate into statistical DOE
Identify the Factors, the levels and the response.

H O :  Temp  0 vs H1 :  Temp  0

H O :  pressure  0 vs H1 :  pressure  0 
  0  1 (Temperatur e)
  2 (Pr essure )
H O :  Temp X pressure  0 vs H1 :  Temp X pressure  0
 12 (Temp X Pr ess)
Noise
(uncontrolled variables)
Factors  ~ NID(0 , ²)
(controlled variables)
Response
Temperature
Product Hardness
80 100
or
Pressure ^
Process Y = Y+
1 ton 2 tons
Seagate Confidential 19 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Create Factorial Design

State practical problem in “English”
Establish objective
 Step 3: Create Factorial Design
 Stat > DOE>Factorial> Create Factorial
Translate into Statistical Design of Experiments
Identify Factors, levels & the Response
Design

Create Factorial Design

Select type of Design Full or Fraction,
Factors,center points, replicates,blocks

Choose
Run the Experiment
Data Tabulation

Choose “2”
Analyze Factorial Design
-Pareto chart, Factorial plot

Interpretation, Conclusion Choose “Design”

With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 20 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Create Factorial Design

1 2

Key-in Factors

Minitab will return this!

Did you get the same? 21
Seagate Confidential Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Create Factorial Design

 Stat > DOE>Display Design
 Change original level values to coded value

Seagate Confidential 22 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Run Experiment and Data Tabulation

State practical problem in “English”
Establish objective
 Step 4: Run Experiment & Date
Tabulation
 Create a Hardness column
Translate into Statistical Design of Experiments
Identify Factors, levels & the Response  Follow the Run Order to run your
experiment and collect data.
Create Factorial Design
 Input your Response[Hardness] data in to
Select type of Design Full or Fraction,
the Hardness column.
Factors,center points, replicates,blocks

Run the Experiment

Data Tabulation

Analyze Factorial Design

-Pareto chart, Factorial plot

Interpretation, Conclusion
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 23 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design - Graphs

State practical problem in “English”
Establish objective

 Step 5: Analyze Factorial Design

 Stat>DOE>Factorial>Analyze Factorial
Design
Translate into Statistical Design of Experiments
Identify Factors, levels & the Response

Create Factorial Design

Select type of Design Full or Fraction,
 Beside the ANOVA Table, the best
Factors,center points, replicates,blocks
way to analyze the Factorial Design is
visual.
Run the Experiment  In Analyze Factorial Design, Graphs
Data Tabulation

provide the following effect plots:

 Normal
Analyze Factorial Design
-Pareto chart, Factorial plot  Pareto
 In Factorial Plot, following plots:
Interpretation, Conclusion  Main Effect Plot
With Engg. Knowledge & Statistics
 Interaction Plot
State practical implications in “English”
 Cube Plot
Seagate Confidential 24 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Graphs

Step 5: Analyze Factorial Design
State practical problem in “English”
Establish objective

 Stat>DOE>Factorial>Analyze Factorial
Design
Translate into Statistical Design of Experiments
Identify Factors, levels & the Response
 Select the Hardness in the Response
 Select the Terms, Graphs

Create Factorial Design

Select type of Design Full or Fraction,

2
Factors,center points, replicates,blocks

1
Run the Experiment
Data Tabulation
3
Analyze Factorial Design
-Pareto chart, Factorial plot

Interpretation, Conclusion
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 25 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design-Term

State practical problem in “English”
Establish objective
 Step 5-1: Analyze Factorial Design
 Stat>DOE>Factorial>Analyze Factorial Design
Translate into Statistical Design of Experiments
 Term
Identify Factors, levels & the Response  Select all the terms by using >>, then Click OK

Create Factorial Design

Select type of Design Full or Fraction,
Factors,center points, replicates,blocks

Run the Experiment 4

Data Tabulation

Analyze Factorial Design

-Pareto chart, Factorial plot


  0  1 (Temperatur e)
Interpretation, Conclusion
  2 (Pr essure )
With Engg. Knowledge & Statistics

 12 (Temp X Pr ess)

State practical implications in “English”

Seagate Confidential 26 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design - Graphs

State practical problem in “English”
Establish objective
 Step 5-2: Analyze Factorial Design
 Stat>DOE>Analyze Factorial Design
 Graph

Translate into Statistical Design of Experiments

 Check the Box on the Normal, Pareto in the
Identify Factors, levels & the Response Effect Plot.
 Set Alpha at 10% = 0.1 then click OK

Create Factorial Design

Select type of Design Full or Fraction,
Factors,center points, replicates,blocks

Run the Experiment

Data Tabulation

Analyze Factorial Design

-Pareto chart, Factorial plot

Interpretation, Conclusion
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 27 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Pareto & Normal Graphs

State practical problem in “English”
Establish objective
 No line on the plot Pareto Chart of the Effects
(response is Hardness, Alpha = .10)

defining statistical A: Temp

B: Pressure

Translate into Statistical Design of Experiments

Identify Factors, levels & the Response
significance AB

BUT the AB
B


Create Factorial Design
Select type of Design Full or Fraction, interaction is the A

Factors,center points, replicates,blocks

largest contributor,
then pressure & 0 5 10 15 20 25

Run the Experiment

temperature Normal Probability Plot of the Effects
(response is Hardness, Alpha = .10)
Data Tabulation
A: Temp
B: Pressure

 No labels on the 0.5

Normal Score
Analyze Factorial Design
-Pareto chart, Factorial plot plot defining 0.0

statistical -0.5

Interpretation, Conclusion
significance
-20 -10 0
With Engg. Knowledge & Statistics
Effect

State practical implications in “English”

Seagate Confidential 28 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE- Main Effect

 The Main Effect of a factor is defined as Pressure 1 ton Pressure 2tons
the average change in the Output variable
produced by a change in the levels of a
factor. Consider the table below where we Temp 80 20 30
have two factors, temperature and
pressure. The output variable is 40 52
Hardness Temp 100

 The change in hardness across Pressures

(the Main Effect for Temp) is defined as:
 Likewise, the Main Effect for Pressure is
defined as:
40  52 20  30 As Temperature increases from level 1
Temp    21
2 2 to level 2, hardness increases by 21 points.

30  52 20  40 Hardness increases by 11 points as the

Press    11 Pressure was adjusted from level 1 to level 2
2 2
Seagate Confidential 29 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE- Main Effect

30 52
Pressure
B
[ -, + ] [+] [ +, + ]

Temperature
A
[-] [+]

[ -, - ] [-] [ +, - ]
20 40

Seagate Confidential 30 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Main Effects Plots

State practical problem in “English”
Establish objective  Step 5-3: Analyze Factorial Design
 Stat>DOE>Factorial>Factorial Plots
Translate into Statistical Design of Experiments  Check the Box on the Main Effect Plot.
Identify Factors, levels & the Response  Click Setup, select the response, term selection

Create Factorial Design

Select type of Design Full or Fraction,
Factors,center points, replicates,blocks

Run the Experiment

Data Tabulation

Analyze Factorial Design

-Pareto chart, Factorial plot

Interpretation, Conclusion
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 31 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Main Effects Plots

Main Effects Plot (data means) for Hardness

What does this mean ?

-1 1 -1 1
34
Remember:
Hardness

32
30 The Main Effect of a factor is
28 defined as the average
26 change in the output variable
Temp Pressure produced by a change in the
levels of a factor.

As temperature goes from level one to level two

the average sample hardness increases from 30 to 31

Make an equivalent statement for pressure….

Seagate Confidential 32 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Interaction Plot

State practical problem in “English”  Step 5-3: Analyze Factorial Design
Establish objective
 Stat>DOE>Factorial>Factorial Plots
 Check the Box on the Interaction Plot.
Translate into Statistical Design of Experiments  Click Setup, select the response.
Identify Factors, levels & the Response
 Select the Factor to include in plots

Create Factorial Design

Select type of Design Full or Fraction,
Factors,center points, replicates,blocks

Run the Experiment

Data Tabulation

Analyze Factorial Design

-Pareto chart, Factorial plot

Interpretation, Conclusion
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 33 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE- Interaction

Pressure 1 ton Pressure 2 tons

In some experiments we find that the main effect between the
levels of one
factor is not the same for different levels of the other factors. In this
case we
have an Interaction between factors.
Temp 80 20 30
Consider the following
set of data:

Temp 100 40 52

At the first level of the Pressure Factor, the effect for Temperature
is:

and at the second level of the Pressure Factor, the effect for
Temperature is: Temp = 50 - 20 = 30

Since the effect of Temperature on hardnessdepends on the level

of Pressure, we
say there is an interaction between Temperature and Pressure. In
this case Temp = 12 - 40 = -28
the trend of increasing hardness with temperature is even reversed.

Seagate Confidential 34 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Basic Concept Of DOE- Interaction

30 52
Pressure
B
[ -, + ] [+] [ +, + ]

Temperature
A
[-] [+]

[ -, - ] [-] [ +, - ]
20 40

Seagate Confidential 35 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Graphical Output - Interaction Plot

Interaction Plot (data means) for Hardness

Temp What does this mean ?

50 -1
+1 Remember
40

In some experiments we find

Mean

30 that the main effect between the

levels of one factor is not the
20 same for different levels of the
other factors. In this case we
have an Interaction between
-1 +1 factors.
Press

In this case there is clearly an interaction between

temperature and pressure as far as Hardness is
concerned - recall what the Pareto chart showed.

Seagate Confidential 36 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Cube Plot

State practical problem in “English”  Step 5-4: Analyze Factorial Design
Establish objective
 Stat>DOE>Factorial>Factorial Plots
 Check the Box on the Cube Plot.
Translate into Statistical Design of Experiments  Click Setup, select the response.
Identify Factors, levels & the Response
 Select the Factor to include in plots

Create Factorial Design

Select type of Design Full or Fraction,
Factors,center points, replicates,blocks

Run the Experiment

Data Tabulation

Analyze Factorial Design

-Pareto chart, Factorial plot

Interpretation, Conclusion
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 37 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Graphical Output – Cube Plot

Cube Plot (data means) for Hardness

50 12 What does this mean ?

1 Remember
The 4 corner represents the
Pressure treatment combinations that
were used to run the
experiment.

20 40
-1 -1 Temp
1

Since there are only two factors, temperature and

pressure, Minitab will provide you a Square plot.

Seagate Confidential 38 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Window Session

State practical problem in “English”

Step 5-5: Analyze Factorial Design

Establish objective

 Stat>DOE>Analyze Factorial Design
Translate into Statistical Design of Experiments
Identify Factors, levels & the Response
 Minitab Window session will display:

Estimated Effects and Coefficients for Hardness (coded units)

Create Factorial Design
Select type of Design Full or Fraction, Term Effect Coef
Factors,center points, replicates,blocks Constant 30.50
Temperat 1.00 0.50
Pressure -9.00 -4.50
Temperat*Pressure -29.00 -14.50

Run the Experiment Analysis of Variance for Hardness (coded units)

Data Tabulation
Source DF Seq SS Adj SS Adj MS F P
Main Effects 2 82.00 82.00 41.00 * *
2-Way Interactions 1 841.00 841.00 841.00 * *
Analyze Factorial Design Residual Error 0 0.00 0.00 0.00
-Pareto chart, Factorial plot Total 3 923.00

Estimated Coefficients for Hardness using data in uncoded units

Interpretation, Conclusion Term Coef

With Engg. Knowledge & Statistics Constant -352.000
Temperat
Pressure
4.40000
252.000
Why is the P-value missing ?
State practical implications in “English” Temperat*Pressure -2.90000

Seagate Confidential 39 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Re-visit the MiniTab Session Window

 What we did wrong on the analysis ?
 What went wrong with the way we run the experiment ?
Estimated Effects and Coefficients for Hardness (coded units)

Term Effect Coef

Constant 30.50
Temperat 1.00 0.50
Pressure -9.00 -4.50
Temperat*Pressure -29.00 -14.50

Analysis of Variance for Hardness (coded units)

Source DF Seq SS Adj SS Adj MS F P

Main Effects 2 82.00 82.00 41.00 * *
2-Way Interactions 1 841.00 841.00 841.00 * *
Residual Error 0 0.00 0.00 0.00
Total 3 923.00

Estimated Coefficients for Hardness using data in uncoded units

Term Coef
Constant -352.000
Temperat 4.40000
Pressure 252.000
Temperat*Pressure -2.90000

Seagate Confidential 40 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Repeats & Replicates

 Definitions
 Repeat - the same experimental run, performed more than
once, one after another
 Replicate - whole or part of the experimental design done
more than once, at different times and possibly in different
orders
 Why ?
 Repeats help signal-to-noise, especially if gages are only
poorly capable - they give a measure of the sample &
analysis variance
 Replicates are even more valuable - they give an estimate of
the total variability affecting the experiment
e.g. set-up differences and therefore an estimate of the
standard deviations of the effects of the variables

Seagate Confidential 41 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analysis with Replicates

 If we had run the earlier example (Hardness) with two
replicates we could have taken some further steps.
 Assume the data sets for the two replicated sets of
runs looked like this (they are also in

Pressure 1 ton Pressure 2 tons Pressure 1 ton Pressure 2 tons

Temp 80 20 30 Temp 80 21 38

Temp 100 40 52 Temp 100 48 9

Seagate Confidential 42 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Window Session

State practical problem in “English”
Establish objective
 Step 5-5: Analyze Factorial Design
 Stat>DOE>Analyze Factorial Design
 Now you have the p-value.
Translate into Statistical Design of Experiments
Identify Factors, levels & the Response
Estimated Effects and Coefficients for Hardness (coded units)

Term Effect Coef SE Coef T P

Create Factorial Design
Constant 29.88 0.4841 61.71 0.000
Temperat 0.25 0.12 0.4841 0.26 0.809
Select type of Design Full or Fraction,
Pressure -10.25 -5.12 0.4841 -10.59 0.000
Factors,center points, replicates,blocks
Temperat*Pressure -28.75 -14.38 0.4841 -29.69 0.000

Analysis of Variance for Hardness (coded units)

Source DF Seq SS Adj SS Adj MS F P

Run the Experiment
Main Effects 2 210.25 210.25 105.12 56.07 0.001
Data Tabulation
2-Way Interactions 1 1653.12 1653.12 1653.12 881.67 0.000
Residual Error 4 7.50 7.50 1.87
Pure Error 4 7.50 7.50 1.88
Total 7 1870.87
Analyze Factorial Design
-Pareto chart, Factorial plot Estimated Coefficients for Hardness using data in uncoded units

Term Coef
Constant -344.000
Interpretation, Conclusion
With Engg. Knowledge & Statistics
Temperat
Pressure
4.32500
248.500
Now we can assess
Temperat*Pressure -2.8750
statistical significance
State practical implications in “English”

Seagate Confidential 43 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Analyze Factorial Design- Window Session

State practical problem in “English”
Establish objective

 Step 6: Interpretation, Conclusion With

Translate into Statistical Design of Experiments Engineering Knowledge & Statistics
Identify Factors, levels & the Response

Create Factorial Design

Select type of Design Full or Fraction,
 Beside the ANOVA Table, visual
Factors,center points, replicates,blocks interpretation with the help of the
pareto graph, Main , Interaction plot
Run the Experiment
will enhance your understanding of
Data Tabulation the DOE results.
 Draw conclusion with the use of
Analyze Factorial Design
-Pareto chart, Factorial plot
Engineering knowledge and statistical
findings.
Interpretation, Conclusion
 Form Transfer function.
With Engg. Knowledge & Statistics

State practical implications in “English”

Seagate Confidential 44 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Identifying Significant Effects

Pareto Chart of the Standardized Effects
(response is Hardness, Alpha = .10)
On the Pareto plot
A: Temp
B: Pressure
this time we see the
AB
line defining the
significant effects to
B
its right

A
Normal Probability Plot of the Standardized Effects
(response is Hardness, Alpha = .10)

A: Temp
0 10 20 30 B: Pressure

0.5
Normal Score

On the normal plot 0.0 B

this time we see the -0.5

significant effects AB

given an annotation -30 -20 -10 0

Standardized Effect

Seagate Confidential 45 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Interpretation of the Replicated Experiment

Main Effects Plot (data means) for Hardness

The mean effects &

80 0
10 1 2 interaction plots look very
35.0 similar to the earlier case.
Hardness

32.5

30.0
More Degrees of Freedom
27.5 mean we can calculate P-
25.0 values.
Temperature Pressure Interaction Plot (data means) for Hardness

Temperature
50 80
100

The Temp term (P = 40

0.809) is not
Mean
30
significant, but
needs to be left in 20

the analysis 10
because of the 1
1 2
2

strong interaction. Pressure

Seagate Confidential 46 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

The Functional Relationship Y = f(X)

 From the Minitab tabular output of the non-replicated
experiment
Estimated Effects and Coefficients
for Hardness (coded units)

Term Effect Coef

Constant 30.50
Temp 1.00 0.50
Press -9.00 -4.50
Temp*Press -29.00 -14.50

we can create the equation:

 Hardness = 30.5
+ 0.50 * Temperature
- 4.50 * Pressure
- 14.50 * Temperature * Pressure
(with coded units +1 and -1 being used for each term)
Seagate Confidential 47 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Y = f(X)  Explanation
Where do these terms come from & what do they mean?
 30.5 - the constant term , is the average hardness across all the
experiment (I.e. when all factors are set to a coded level 0)
 0.5 * Temperature - from the main effects plot - if you move from
the mean level (coded as 0) to the +1 level of Temperature the
average hardness change is + 0.5 across the two levels of
Pressure
 Similarly for Pressure
 -14.5 * Temperature * Pressure - from the interaction plot if you
move from the mean level to the points where Pressure &
Temperature are both high (+1, +1) or both low (-1,-1) the
average hardness change is -14.5
 For example hardness for (+1, -1) (2,1 in uncoded units)
= 30.5 + 0.5 * +1 - 4.5 * -1 - 14.5 * +1 * -1
= 50 which is the observed value- so it works !
Seagate Confidential 48 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)

Randomization
 It is essential to randomize the runs in a DOE to
avoid being confused by effects from time or machine
sequences.
 This is done by default for you in Minitab when you
create an experimental design.

Seagate Confidential 49 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Center Points
 A two-level factorial experiment has one real shortcoming -
it cannot predict if the effects of the variables are linear
between the chosen levels.
 To examine this fully requires complex tools and multi-level
experiments, BUT there is a way to identify that a non-linear
relationship may exist
 This is done by the addition of “Center points” which have
other benefits as well
 They can often be set to existing process points where there is a lot
of other data to compare
 They can give an estimate of other variance sources (e.g.
sampling), even if no other repeats or replicates are done
 We will examine Center points fully in the section which
describes the 23 experiment

Seagate Confidential 50 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

3
2 Full Factorial Design Exercise
Typical Scenario
 Your company is preparing the pre-production of High Tech fluid
Dynamic Bearing motor.
 Your customer wants you to study the effect of oil viscosity,
temperature and a special additive on the motor wear before
qualifying your product.
 The factor levels that you are working on are:
 Viscosity [ 30, 40 ]
 Temperature [75, 100 ]
 Additive [ Formula A, Formula B]
 Since you process are setting up the line, you can afford to run 2
replicates only.

 As as the core team members, you are task to design

an experiment to run this experiment.
Seagate Confidential 51 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
DOE-Road Map for 2 Design
State problem in “English” State problem in “English”
Establish objective Study effect of ……on FDB Wear

Translate into Statistical Design of Experiments Translate into Statistical Design of Experiments
Identify Factors, levels & the response 23 design, Viscosity, Temp. Additive, WEAR

Create Factorial Design Create Factorial Design

Select type of Design Full or Fraction, Full Factorial Design, 23 design
Factors,center points, replicates,blocks 2 Replicates, no block

Run the Experiment Run the Experiment

Data Tabulation Data Tabulation

Analyze Factorial Design Basic Concept 23 Full Analyze Factorial Design

-Pareto chart, Factorial plot DOE-Road Map Factorial DOE -Pareto chart, Factorial plot

Interpretation, Conclusion With Interpretation, Conclusion With

Engineering Knowledge & Statistics Engineering Knowledge & Statistics

State practical implications State practical implications

in “English” in “English”
Seagate Confidential 52 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
2 Full Factorial Design Exercise
 Use the DOE-Road Map
 Create the Full Factorial Design.
 Stat > DOE>Factorial> Create Factorial Design

Seagate Confidential 53 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise
 Did you get the
same Output as
this ?

 23 Full Factorial
Design will result in
8 runs, because of
the 2 replicates,
Minitab will return
you; 16 runs
 Run order is same
as StdOrder ! Why !

Seagate Confidential 54 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Data Tabulation
 Your teams have
run the experiment
and collect the wear
on all the
treatments
 A Wear column was
created to record
the experiment
results.

 Open MiniTab file:

Wear.mpj
 Analyze the result.
 Stat >>DOE >>
Factorial >>
Analyze Factorial
Design
 Report your
observations
Seagate Confidential 55 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Normal Probability Plot
 Visual screen for the effect of factors.
 Viscosity has a ‘positive’ effect on motor wear.
 As you increase viscosity from 30 to 40, the average motor wear

increases.
 Temperature has ‘negative’ effect on motor wear.
 As you increase Temperature from 75 to 100, the average decreases.

Normal Probability Plot of the Standardized Effects

(response is Wear, Alpha = .10)

1.5 A: Viscosit
A B: Temp
C: Additive
1.0
AC
0.5
Normal Score

0.0

-0.5

-1.0

B
-1.5
-5 0 5
Standardized Effect

Seagate Confidential 56 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Pareto Chart
 Visual screen for the importance Factors in Pareto Chart
 Viscosity, Temperature, [Viscosity * Additive]
 These factors have a significant impact on FDB wear.

Pareto Chart of the Standardized Effects

(response is Wear, Alpha = .10)

A: Viscosit
B: Temp
A C: Additive

AC

BC

ABC

AB

0 1 2 3 4 5 6 7

Seagate Confidential 57 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Session Window [ANOVA]
Estimated Effects and Coefficients for Wear (coded units)  T-test also
confirm
Term Effect Coef SE Coef T P
Constant 3.8750 0.08478 45.71 0.000
that:
Viscosit 1.2000 0.6000 0.08478 7.08 0.000
 Viscosity
Temp -0.9000 -0.4500 0.08478 -5.31 0.001  Temp
Additive 0.1500 0.0750 0.08478 0.88 0.402  Viscosity
Viscosit*Temp -0.0500 -0.0250 0.08478 -0.29 0.776 * Additive
Viscosit*Additive 0.4000 0.2000 0.08478 2.36 0.046
 Are
Temp*Additive 0.2000 0.1000 0.08478 1.18 0.272
Viscosit*Temp*Additive 0.0500 0.0250 0.08478 0.29 0.776
Significant.

Analysis of Variance for Wear (coded units)

 Since there
is only 3
Source DF Seq SS Adj SS Adj MS F P significant
Main Effects 3 9.0900 9.09000 3.03000 26.35 0.000
factors, we
2-Way Interactions 3 0.8100 0.81000 0.27000 2.35 0.149
can
3-Way Interactions 1 0.0100 0.01000 0.01000 0.09 0.776
evaluate a
Residual Error 8 0.9200 0.92000 0.11500
reduce
Pure Error 8 0.9200 0.92000 0.11500
model.
Total 15 10.8300 58
Seagate Confidential Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Reduced Hierarchical Model
What is a hierarchical Model ?
 All main effect that are part of a significant interaction must
remain in the model.
 In fact, if you try to exclude Additive and include Viscosity*Additive,
MINITAB will prompt you to include the main effect for Additive. Try it !
Stat > DOE>Factorial> Create
Factorial Design
Select Wear as Response.
Click Terms.
Select the terms A, B and AC


 Because you are fitting what you

believe to be your final model,
create residual plots to assess the
model and the underlying
assumptions of independence,
randomness and normality.
Seagate Confidential 59 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Reduced Hierarchical Model
 Go back to the worksheet, follow these steps
 Stat > DOE>Factorial> Create Factorial Design
 Select Wear as Response.
 Click Terms.
 Select the terms A, B, C and AC

1
4

Seagate Confidential 60 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Reduced Hierarchical Model
 Stat > DOE>Factorial> Create Factorial Design
 Select Wear as Response.
 Click Graphs.
 Under Effects Plots, Uncheck Normal and Pareto.

 Under Residual Plots, Check Normal plot, Residuals versus Fits and Residual

versus Order

1 4
Seagate Confidential 61 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Assessing lack of fit
Assessing Lack of Fit
 The ANOVA table shows that the main effects(P=0.000) and the 2-way interaction
(P=0.028) are significant.
 It is important also to notice the lack of fittest. Does the model fit the data
adequately? The answer is Yes (P=0.679 >0.10)
Estimated Effects and Coefficients for Wear (coded units) Evaluate the
Term Effect Coef SE Coef T P reduced
Constant 3.8750 0.07906 49.02 0.000 model
Viscosit 1.2000 0.6000 0.07906 7.59 0.000  You see that
Temp -0.9000 -0.4500 0.07906 -5.69 0.000 Additive is
Additive 0.1500 0.0750 0.07906 0.95 0.363
now the only
non-
Viscosit*Additive 0.4000 0.2000 0.07906 2.53 0.028
significant
effect
Analysis of Variance for Wear (coded units)
Source DF Seq SS Adj SS Adj MS F P
Main Effects 3 9.0900 9.0900 3.03000 30.30 0.000
2-Way Interactions 1 0.6400 0.6400 0.64000 6.40 0.028
Residual Error 11 1.1000 1.1000 0.10000
Lack of Fit 3 0.1800 0.1800 0.06000 0.52 0.679
Pure Error 8 0.9200 0.9200 0.11500
Total 15 10.8300
Seagate Confidential 62 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Assessing lack of fit
What is Lack of Fit
 What lack of Fit appears in the ANOVA table, the Residual Error is
separated into the error given all terms in the model (Pure Error) and the
error contributed to residual error by terms removed from the model.
 Lack of fit considers the effect of removing higher order terms from the
model, as compared to all possible terms that could be in the model.

Estimated Effects and Coefficients for Wear (coded units)

Term Effect Coef SE Coef T P
 The SS and degrees of
Constant 3.8750 0.07906 49.02 0.000 freedom(DF) for lack of
Viscosit 1.2000 0.6000 0.07906 7.59 0.000
fit, come from the A*B
Temp -0.9000 -0.4500 0.07906 -5.69 0.000
Additive 0.1500 0.0750 0.07906 0.95 0.363
and B*C and A*B*C
Viscosit*Additive 0.4000 0.2000 0.07906 2.53 0.028 interactions.
 If sufficient degrees of
Analysis of Variance for Wear (coded units)
Source DF Seq SS Adj SS Adj MS F P freedom exist, lack of fit
Main Effects 3 9.0900 9.0900 3.03000 30.30 0.000 with always be shown for
2-Way Interactions
Residual Error 11
1 0.6400
1.1000
0.6400
1.1000
0.64000
0.10000
6.40 0.028
any model where all main
Lack of Fit 3 0.1800 0.1800 0.06000 0.52 0.679 effects remain.
Pure Error 8 0.9200 0.9200 0.11500
Total 15 10.8300

Seagate Confidential 63 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Residual Plots
Checking the assumptions about the residuals
Normal Probability Plot of the Residuals Residuals Versus the Order of the Data Residuals Versus the Fitted Values
(response is Wear) (response is Wear) (response is Wear)

2
0.5 0.5

0.4 0.4

1 0.3 0.3
Normal Score

0.2 0.2

Residual

Residual
0.1 0.1
0
0.0 0.0

-0.1 -0.1
-1
-0.2 -0.2

-0.3 -0.3

-2 -0.4 -0.4
-0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 2 4 6 8 10 12 14 16 3 4 5
Residual Observation Order Fitted Value

Residual plots
The residual plots show the model is adequate.
 The Normal Probability Plot of Residuals is close enough to a straight
line to assume that the errors are normal.
 The Residuals Versus Order Plot does not show any sign of trends in
the errors due to the order of the experiment runs.
 The random pattern in the Residuals Versus the Fitted Value plot,
indicates that the errors have constant variance

Seagate Confidential 64 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Factorial Plots
 Main Effect Plot
 Stat>DOE>Factorial>Factorial Plots
 Check the Box on the Main Effect Plot.
 Click Setup, select the response, term selection

Main Effects Plot (data means) for Wear

What can conclude from the
plot:
 Like the normal effect plot,
4.40
the main effect plot shows
4.15
the “positive” effect of
viscosity and the negative
Wear

3.90

3.65
effect of temperature.
3.40

Viscosity Temp Additive  In contrast, the plot shows

additive has very little
impact on average motor
wear

Seagate Confidential 65 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Factorial Plots
 Interaction Plot What can conclude from the plots:
 Stat>DOE>Factorial>Factorial Plots  Of primary interest in the interaction
 Check the Box on the Interaction plots, is the Viscosity*Additive term.
Plot. MINITAB plots the average response
 Click Setup, select the response, on the y-axis. Because you want to
term selection minimize motor wear, it is clear that
Interaction Plot (data means) for Wear
low viscosity (30) is better than high
viscosity(40), because the wear is
Viscosity
5
lower.
40 4  When you use Additive A, there is
30
3
smaller difference in mean motor
5 wear between viscosity level (30 and
Temp
4
40) than when you use Additive B.
100

75
 When Additive B is added, there is a
3
dramatic increase in motor wear for
Additive the B;40 combination as compared to
B;30, This is why the interaction is
significant.

 Parallel lines indicate no interaction between Viscosity and Temperature.

Although a slight interaction exists between. Temperature and Additive, the
effects plots and Session window tables show that it is not significant.
Seagate Confidential 66 Supplier Six Sigma Modular Training
 Design Of Experiment (Full Factorial Design)
3
2 Design Exercise- Factorial Plots
 Cube Plot What can conclude from the plots:
 Stat>DOE>Factorial>Factorial Plots  Use the cube plot to choose the best
 Check the Box on the Cube Plot. treatment. You are looking for the
 Click Setup, select the response, treatment combination that shows the
term selection lowest mean wear.
 The min. wear (2.8) is at Viscosity 30,
Cube Plot (data means) for Wear
Temperature 100 and Additive B.
 If change to Additive A, the Motor
2.8 4.4
wear only increase to 2.9. A
100
2.9 3.6 difference of 0.1 in motor wear
probably has no practical importance.
 Therefore, if you use a viscosity of 30
Temp 3.5 5.1
B

Additive

75
3.9 4.8
A and a temperature of 10OoC, you
30
Viscosity
40
have choice between Additive A or
B.

 In practice, this information could be used to make manufacturer’s

recommendations for oil weight and additives to optimize motor wear.

Seagate Confidential 67 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

End of Topic
What question do you have?

Seagate Confidential 68 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Reference Book
Author:
George EP Box
William G Hunter
J Stuart Hunter
Publisher:
John Wiley & Sons, Inc.
ISBN: 0-471-09315-7

Seagate Confidential 69 Supplier Six Sigma Modular Training

 Design Of Experiment (Full Factorial Design)

Reference Book

Author:
Stephen R Schmidt
Robert G Launsby
Publisher:
Air Academy Press
ISBN: 1-880156-05-9

Seagate Confidential 70 Supplier Six Sigma Modular Training