Professional Documents
Culture Documents
DOE Training
DOE Training
DOE
INTRODUCTION
Design of Experiments
•A systematic set of procedures and checks which permits one to
evaluate the effect of one or more factors without concern about
extraneous variables or subjective judgments.
•It begins with the statement of the experimental objective and ends
with the reporting of the results.
Objectives
Strategy of Experimentation
The purpose of an experiment is
Define the Problem to better understand the real
world, not to understand the
Establish the Objective experimental data
Select the Output (Response) William Diamond
IBM - Retired Statistician
Select the Input Factors (KPIVs)
Choose the Factor Levels
Select the Experiment Design
Collect the Data
Analyze the Data
Draw Conclusions
Achieve the Objective
Design Of Experiments:
Any testing for which the inputs are controlled and
the analysis has been planned
BK2-4DOE Special Cases 4
DOE - iNTRO (wec)
Strategy of Experimentation
Replication
Replication &
& Randomization
Randomization :: The
The R&R
R&R of
of DOE
DOE
Experimentation Sequence
Most
Mostoften
oftenused
usedto to RSM
RSMDesigns
Designs
study
studyhighly
highly To
Todefine
definethe
theoptimum
optimumoperating
operatingwindow
windowofof
complicated
complicatedeffects
effects
and the
thekey
keyvariables
variablesofofhigh
highimpact
andrelationships
relationships impact
involving
involvingtwotwooror
three
three variables. In
variables. In
general,
general,such
suchdesigns
designs
are
areemployed
employedto toyield Full
Full Factorial & 2 Designs
Factorial & 2
yield KK
aareasonably Designs
reasonablyprecise
precise
mathematical To
Toidentify
identifykey
keyvariables
variablesofofhigh
highimpact
mathematicalmodelmodel impact
of the phenomenon
of the phenomenon
under
underinvestigation.
investigation.
Fractional
FractionalFactorial
FactorialDesigns
Designs
Most
Mostoften
oftenlimited
limitedtototwo
twolevels
levels and
andare
are
principally
principallyused
usedforforthe
thestudy
studyofofMain
MainEffects.
Effects.
Experimentation Sequence
RSM
RSMDesigns
Designs
Used
Usedto tostudy
study To
Todefine
definethe
theoptimum
optimumoperating
operatingwindow
windowofof
relatively
relatively
uncomplicated the
thekey
keyvariables
variablesofofhigh
highimpact
impact
uncomplicated
effects
effectsand
and
interactions
interactions
resulting
resultingfrom
fromaa
relatively
relativelysmall
number
small Full
Full Factorial & 2 Designs
Factorial & 2 KK
Designs
numberof offactors.
factors.
Generally speaking, To
Toidentify
identifykey
keyvariables
variablesofofhigh
highimpact
impact
Generally speaking,
such
suchdesigns
designsareare
based
basedon onthe
thefull
full
factorial
factorialmodel
modelusing
using
two
twoor orthree
threelevels.
levels.
Fractional
FractionalFactorial
FactorialDesigns
Designs
Most
Mostoften
oftenlimited
limitedtototwo
twolevels
levels and
andare
are
principally
principallyused
usedforforthe
thestudy
studyofofMain
MainEffects.
Effects.
Experimentation Sequence
RSM
RSMDesigns
Designs
To
Todefine
definethe
theoptimum
optimumoperating
operatingwindow
windowofof
the
thekey
keyvariables
variablesofofhigh
highimpact
impact
Screening
Screening
experimentation
experimentationis is
used
usedtotodetermine Full
Full Factorial & 2 Designs
Factorial & 2
determine KK
which Designs
whichfewfewfactors
factors
from To
Toidentify
identifykey
keyvariables
variablesofofhigh
highimpact
fromaalarge
largehosthostofof impact
candidate
candidatefactors
factors
have
havethe
thegreatest
greatest
influence
influenceuponuponthe the
measured
measuredresponse.
response.
These
Theseareareusually
usually Fractional
FractionalFactorial
FactorialDesigns
Designs
done
doneatatthe
theearly
early Most
stages Mostoften
oftenlimited
limitedtototwo
twolevels
levels and
andare
are
stagesofofthethe
improvement principally
principallyused
usedforforthe
thestudy
studyofofMain
MainEffects.
improvementprocess.
process. Effects.
Advantages of 2K Factorials
2K Factorial Experiments
A 2k factorial is a designed experiment with k factors, each at
2 levels.
Factors - one of the controlled or uncontrolled inputs into a
process whose influence upon a response is being studied
in the experiment
Levels - the values at which you place the input factors
Example : 22 factorial
Temp Conc
-1 -1
1 -1
-1 1
1 1
Exercise
a b c d
2x2 Design -1 -1 -1 -1
1 -1 -1 -1
-1 1 -1 -1
1 1 -1 -1
-1 -1 1 -1
2x2x2
Design
1 -1 1 -1
Answer
-1 1 1 -1
1 1 1 -1
-1 -1 -1 1
1 -1 -1 1
-1 1 -1 1
1 1 -1 1
2x2x2x2 -1 -1 1 1
Design
1 -1 1 1
-1 1 1 1
1 1 1 1
Strategy of Experimentation
Define the Problem:
My golf score is too high.
Distance
Distance (KPOV)
(KPOV) is
is aa Variables
Variables Type
Type of
of Data
Data
BK2-4DOE Special Cases 16
DOE - iNTRO (wec)
Strategy of Experimentation
Strategy of Experimentation
Strategy of Experimentation
Select the Experiment Design
How Do I Do the Experiment? What Do I Test?
Strategy of Experimentation
Select the Experiment Design
How Do I Do the Experiment? What Do I Test?
Option 2 : Full Factorial
What Are All of the Possible combinations?
A. Ping Clubs Titleist Balls No Wind
B. Callaway Clubs Titleist Balls No Wind
C. Ping Clubs Pinnacle Balls No Wind
D. Callaway Clubs Pinnacle Balls No Wind
E. Ping Clubs Titleist Balls High Wind
F. Callaway Clubs Titleist Balls High Wind
G. Ping Clubs Pinnacle Balls High Wind
H. Callaway Clubs Pinnacle Balls High Wind
Let’s Look At It A Different Way
20
BK2-4DOE Special Cases
DOE - iNTRO (wec)
Titleist
E F High Wind
(yards) (yards)
Titleist Ball
High Wind
G H Ping Clubs
Pinnacle
(yards) (yards)
Titleist A B
(yards) (yards)
No Wind
Pinnacle C D
(yards) (yards)
Titleist
E F
(yards) (yards)
High Wind
G H
Pinnacle
(yards) (yards)
Strategy of Experimentation
Analyze the Data
Rules For Analyzing Variables Data
One Observation per Treatment More than One Observation per
Combination Treatment Combination
•Generate the Main Effects Plot & •Generate the Main Effects Plot &
Interactions Plot Interactions Plot
•Run the Anova Procedure •Run the Anova Procedure
(Stat>Anova>General Linear Model) (Stat>Anova>General Linear Model)
•CAN NOT analyze interactions •Test for Statistical Significance
•Test for Statistical Significance P< .05
P< .05 Starting with highest order interaction
•Run Diagnostics •Run Diagnostics
(Residuals & Fits) (Residuals & Fits)
•Loss of Information on Interactions •Able to statistically analyze interactions
BK2-4DOE Special Cases 23
DOE - iNTRO (wec)
Strategy of Experimentation
Analyze the Data
Strategy of Experimentation
Analyze the Data
Main Effects Plot :
•Graphs the change in the average response (KPOV) for a
single factor as it changes from one level to another
Resistan 95.1
94.3
93.5
Factor is important Temp Current
Strategy of Experimentation
Analyze the Data
Interactions :
•When the effect (KPOV) of varying the levels of one factor is
influenced by the level of another factor
95.5
Mean
•The greater the difference in slope, 94.5
92.5
-1 1
Example of a 23 Factorial
Temp -1 Temp 1
Con -1 Con 1 Con -1 Con 1
Cat -1 60 54 72 68
Cat 1 52 45 83 80
Example of a 23 Factorial
How do I do this in Minitab ?
The Design Matrix with results looks like:
Temp Conc Catalyst Viscosity
-1 -1 -1 60
1 -1 -1 72
-1 1 -1 54
1 1 -1 68
-1 -1 1 52
1 -1 1 83
-1 1 1 45
1 1 1 80
Strategy of Experimentation
Analyze the Data
Rules For Analyzing Variables Data
One Observation per Treatment More than One Observation per
Combination Treatment Combination
•Generate the Main Effects Plot & •Generate the Main Effects Plot &
Interactions Plot Interactions Plot
•Run the Anova Procedure •Run the Anova Procedure
(Stat>Anova>General Linear Model) (Stat>Anova>General Linear Model)
•CAN NOT analyze interactions •Test for Statistical Significance
•Test for Statistical Significance P< .05
P< .05 Starting with highest order interaction
•Run Diagnostics •Run Diagnostics
(Residuals & Fits) (Residuals & Fits)
•Loss of Information on Interactions •Able to statistically analyze interactions
BK2-4DOE Special Cases 31
DOE - iNTRO (wec)
Stat>ANOVA>Main Effects 75
70
Viscosit
65
60
55
Temp
Stat>ANOVA>Interaction Plot -1 -1
-1 -1
-1 -1
Conc 1 1
Cat
2. Run ANOVA
Stat>ANOVA>General Linear Model
General Linear Model
Factor Levels Values
Temp 2 -1 1
Conc 2 -1 1
Cat 2 -1 1
3. Run Diagnostics
Stat>ANOVA>Residuals Plot
Residual
0 0 X=0.000
-10
-5 -20 -3.0SL=-22.99
-30
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 1 2 3 4 5 6 7 8
Normal Score Observation Number
Residual
1 0
-5
0
-6 -4 -2 0 2 4 6 50 60 70 80
Residual Fit
Reduced Model
We found that Temp and Temp*Catalyst were important, so use the
Anova procedure to run only those terms. This has the effect of
giving us multiple observations so we can perform a more
sophisticated analysis.
Run Anova>General Linear Model
Analyze only Temp and Temp*Catalyst Interaction can be analyzed
because you are running a
Select fits and residuals in Storage reduced model
Diagnostics
Residual Model Diagnostics
Look for a Normal Plot of Residuals I Chart of Residuals
4 10
straight line 3
3.0SL=8.169
Residual
1
0 0 X=0.000 of control
-1
-2
conditions
-3 -3.0SL=-8.169
-4 -10
-1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 1 2 3 4 5 6 7 8
Normal Score Observation Number
Residual
1
1 0
-1
-2
-3
Look for a 0 -4
Look for
-3 -2 -1 0 1 2 3 4 50 60 70 80
symmetrical patterns
Residual Fit
distribution
Strategy of Experimentation
Analyze the Data
Rules For Analyzing Variables Data
•Run the Diagnostics (Residuals vs Fits)
Residual
0 0 X=0.000 conditions
-10
-10 -20
-3.0SL=-27.57
-30
-2 -1 0 1 2 0 5 10 15
Normal Score Observation Number
Look for a
Frequency
Residual
3
0 Look for
normal, 2
patterns
symmetric 1
-10
0
distribution -15 -10 -5 0 5 10 15 2990 3000 3010 3020 3030 3040
Residual Fit
Conclusion