design of engineering experiments the 2k-p fractional factorial designnoordin/s/aqe ch08 rev.pdf ·...
TRANSCRIPT
Design of Engineering Experiments The 2k-p Fractional Factorial Design
• Text reference, Chapter 8Text reference, Chapter 8• Motivation for fractional factorials is obvious; as the
number of factors becomes large enough to be “interesting” the size of the designs grows very quicklyinteresting , the size of the designs grows very quickly
• Emphasis is on factor screening; efficiently identify the factors with large effects
• There may be many variables (often because we don’t know much about the system)
• Almost always run as unreplicated factorials, but oftenAlmost always run as unreplicated factorials, but often with center points
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
1
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
2
Why do Fractional FactorialWhy do Fractional Factorial Designs Work?
• The sparsity of effects principle– There may be lots of factors, but few are important
i d i d b i ff l d– System is dominated by main effects, low-order interactions
• The projection propertyThe projection property– Every fractional factorial contains full factorials in
fewer factorsS i i i• Sequential experimentation– Can add runs to a fractional factorial to resolve
difficulties (or ambiguities) in interpretation
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
3
d cu t es (o a b gu t es) te p etat o
The One-Half Fraction of the 2k
• Section 8.2, page 290• Notation: because the design has 2k/2 runs, it’s referred to as a 2k-1
C id ll i l th 23 1• Consider a really simple case, the 23-1
• Note that I =ABC
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
4
The One-Half Fraction of the 23
For the principal fraction, notice that the contrast for estimating the main effect A is exactly the same as the contrast used for estimating the BCy ginteraction.
This phenomena is called aliasing and it occurs in all fractional designs
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
5
Aliases can be found directly from the columns in the table of + and - signs
Aliasing in the One-Half Fraction of the 23g
A = BC, B = AC, C = AB (or me = 2fi)
Aliases can be found from the defining relation I = ABCby multiplication:
AI = A(ABC) = A2BC = BC
BI =B(ABC) = AC
CI = C(ABC) = AB
Textbook notation for aliased effects:Textbook notation for aliased effects:
[ ] , [ ] , [ ]A A BC B B AC C C AB
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
6
The Alternate Fraction of the 23-1
• I = -ABC is the defining relation• Implies slightly different aliases: A = -BC,
B= -AC, and C = -AB• Both designs belong to the same family, defined
by I ABC
• Suppose that after running the principal fraction, the alternate fraction was also run
I ABC
the alternate fraction was also run• The two groups of runs can be combined to form a
full factorial – an example of sequential
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
7
experimentation
Design Resolution
• Resolution III Designs:– me = 2fi
l 3 1– example • Resolution IV Designs:
2fi = 2fi
3 12III
– 2fi = 2fi– example
• Resolution V Designs:
4 12IV
g– 2fi = 3fi– example 5 12V
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
8
Construction of a One half FractionConstruction of a One-half FractionThe basic design; the design generator
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
9
Projection of Fractional Factorialsj
Every fractional factorial contains full factorials incontains full factorials in fewer factors
The “flashlight” analogy
A one-half fraction will project into a full factorial in any k – 1 of the original factors
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
10
Example 8.1
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
11
Ex. 6.2 The Resin Plant Experiment
DOX 6E Montgomery 12
Example 8.1pInterpretation of results often relies on making some assumptionsOckham’s razorOckham s razorConfirmation experiments can be importantAdding the alternate fraction – see page 301g p g
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
13
Fig_08_04 Projection of the 2IV 4-1 design into a 23 design in A, C and D for Ex 8.1
DOX 6E Montgomery 15
DOX 6E Montgomery 16
Figure 8.5 A 25–1 Design for Example 8.2
DOX 6E Montgomery 17
DOX 6E Montgomery 18
The AC and AD interactions can be verified by inspection of the cube plot
DOX 6E Montgomery 19
verified by inspection of the cube plot
Possible Strategies forStrategies for
Follow-Up Experimentation
Following a Fractional
F t i l D iFactorial Design
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
20
Effect Alias Effect Alias estimate structure estimate structure
for principal for principal for alternate for principalfraction fraction fraction fraction
[i] [i] [i]' 1/2([i]+[i]') 1/2([i]-[i]')A 19 A+BCD 24.25 A-BCD 21.63 A -2.63 BCDB 1 5 B ACD 4 75 B ACD 3 13 B 1 63 ACDB 1.5 B+ACD 4.75 B-ACD 3.13 B -1.63 ACDC 14 C+ABD 5.75 C-ABD 9.88 C 4.13 ABDD 16.5 D+ABC 12.75 D-ABC 14.63 D 1.88 ABC
AB -1 AB+CD 1.25 AB-CD 0.13 AB -1.13 CDAC -18.5 AC+BD -17.75 AC-BD -18.13 AC -0.38 BD
DOX 6E Montgomery 21
AC 18.5 AC BD 17.75 AC BD 18.13 AC 0.38 BDAD 19 AD+BC 14.25 AD-BC 16.63 AD 2.38 BC
Confirmation experiment for this example:Confirmation experiment for this example: see page 302
U h d l di h bi i f iUse the model to predict the response at a test combination of interest in the design space – not one of the points in the current design.
Run this test combination – then compare predicted and observed.
For Example 8.1, consider the point +, +, -, +. The predicted response is
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
22
Actual response is 104.
The One-Quarter Fraction of the 26-2
Complete defining relation: I ABCE BCDF ADEFComplete defining relation: I = ABCE = BCDF = ADEF
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
23
The One-Quarter Fraction of the 2k
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
24
The One Quarter Fraction of the 26-2The One-Quarter Fraction of the 26 2
• Uses of the alternate fractions, E ABC F BCD
• Projection of the design into subsets of the original six variables
• Any subset of the original six variables that is not• Any subset of the original six variables that is not a word in the complete defining relation will result in a full factorial designg– Consider ABCD (full factorial)– Consider ABCE (replicated half fraction)
C id ABCF (f ll f t i l)Chapter 8 Design and Analysis of Experiments
7E 2009 Montgomery25
– Consider ABCF (full factorial)
A One-Quarter Fraction of the 26-2:A One-Quarter Fraction of the 2 :Example 8.4, Page 305
• Injection molding process with six factors• Design matrix, page 305g , p g• Calculation of effects, normal probability
plot of effectsp• Two factors (A, B) and the AB interaction
are importantp• Residual analysis indicates there are some
dispersion effects (see page 307) Chapter 8 Design and Analysis of Experiments
7E 2009 Montgomery26
p ( p g )
The General 2k-p FractionalThe General 2 p Fractional Factorial Design
S i 8 4 309• Section 8.4, page 309• 2k-1 = one-half fraction, 2k-2 = one-quarter fraction,
2k-3 = one-eighth fraction 2k-p = 1/ 2p fraction2 = one-eighth fraction, …, 2 p = 1/ 2p fraction• Add p columns to the basic design; select p
independent generatorsp g• Important to select generators so as to maximize
resolution, see Table 8.14• Projection – a design of resolution R contains full
factorials in any R – 1 of the factorsBl ki
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
27
• Blocking
The General 2k-p Design: Resolution may t b S ffi i tnot be Sufficient
• Minimum abberation designsg
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
28
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
29
table_08_15
fig_08_18
fig_08_19
fig_08_20
fig_08_21
fig_08_22
Resolution III Designs: Section 8 5Resolution III Designs: Section 8.5, page 320
• Designs with main effects aliased with two-factor interactions
• Used for screening (5 – 7 variables in 8 runs 9 - 15 variables in 16 runs forruns, 9 15 variables in 16 runs, for example)
• A saturated design has k = N 1 variables• A saturated design has k = N – 1 variables• See Table 8.19, page 320 for a 7 42III
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
39
Resolution III Designs
Saturated 2III7-4 design used for
studying 7 factors in 8 runsstudying 7 factors in 8 runs.
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
40
Saturated 2III7-4 design can be used to obtain resolutionIII g
III for studying fewer than 7 factors in 8 runs. Simplydrop any one on the column in the previous design
Resolution III Designs• Sequential assembly of fractions to separate aliased effects
(page 322)S it hi th i i one l id ti t f• Switching the signs in one column provides estimates of that factor and all of its two-factor interactions
• Switching the signs in all columns dealiases all main effects from their two-factor interaction alias chains –called a full fold-over
• Defining relation for a fold-over (page 325)g (p g )• Be careful – these rules only work for Resolution III
designs• There are other rules for Resolution IV designs and other• There are other rules for Resolution IV designs, and other
methods for adding runs to fractions to dealias effects of interest
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
42
• Example 8.7, eye focus time, page 323
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
43
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
44
R b th t th f ll f ld t h iRemember that the full fold-over technique illustrated in this example (running a “mirror image” design with all signs reversed) only works in adesign with all signs reversed) only works in a Resolution III design.
Defining relation for a fold over design see pageDefining relation for a fold-over design – see page 325.
Bl ki b i id i iBlocking can be an important consideration in a fold-over design – see page 325.
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
45
Plackett-Burman Designs
• These are a different class of resolution III design• The number of runs, N, need only be a multiple of , , y p
four• N = 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, …• The designs where N = 12, 20, 24, etc. are called
nongeometric PB designs• See text, page 326 for comments on construction
of Plackett-Burman designs
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
46
Plackett-Burman Designs
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
47
This is a non-regular design
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
48
Plackett-Burman Designs
Projection of the
g
Projection of the 12-run design into 3 and 4 factors
All PB designs have projectivity 3 (contrast with other(contrast with other resolution III fractions)
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
49
Plackett-Burman Designs
Th li t t i l i th PB d i• The alias structure is complex in the PB designs• For example, with N = 12 and k = 11, every main
effect is aliased with every 2FI not involving itselfeffect is aliased with every 2FI not involving itself• Every 2FI alias chain has 45 terms
P ti l li i t ti ll tl li t• Partial aliasing can potentially greatly complicate interpretation if there are several large interactions
• Use very very carefully but there are some• Use very, very carefully – but there are some excellent opportunities
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
50
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
51
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
52
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
53
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
54
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
55
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
56
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
57
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
58
R l i IV d V D i (P 322)Resolution IV and V Designs (Page 322)
A resolution IV design must have at least 2k runs.
“optimal” designs may often prove useful.
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
59
Sequential Experimentation with ResolutionSequential Experimentation with Resolution IV Designs – Page 339
We can’t use the full fold-over procedure given previously for Resolution III designs – it will result in replicating the runs in the
i i l d ioriginal design.
Switching the signs in a single column allows all of the two-factor interactions involving that column to be separated.
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
60
The spin coater experiment – page 340
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
61
[AB] = AB + CE
We need to dealias these interactionsinteractions
The fold-over design switches the signs in column A
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
62
The aliases from the complete design following the fold-over (32 runs) are as follows:
Finding the aliases is somewhat beyond the scope of this course (Chapter 10 provided details) but it can be determined using Design-Expert.
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
63
de e ed us g es g pe .
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
64
A full fold-over of a Resolution IV design is usually not necessary, and it’s potentially very inefficient.
In the spin coater example, there were seven degrees of freedom available to estimate two-factor interaction alias chains.
Af ddi h f ld (16 ) h l 12 d fAfter adding the fold-over (16 more runs), there are only 12 degrees of freedom available for estimating two-factor interactions (16 new runs yields only five more degrees of freedom).
A partial fold-over (semifold) may be a better choice of follow-up design. To construct a partial fold-over:
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
65
N tNot an orthogonal design – but that’s notthat s not such a big deal
l dCorrelated parameter estimates
Larger standard errors of regression model coefficients
ff t
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
66
or effects
There are still 12 degrees of freedom available to estimate
two-factor interactions
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
67
R l i V D i P 331Resolution V Designs – Page 331
We used a Resolution V design (a 25-2) in Example 8.2
Generally, these are large designs (at least 32 runs) for six or more factors
Irregular designs can be found using optimal design construction methods
JMP h ll biliJMP has excellent capability
Examples for k = 6 and 8 factors are illustrated in the book
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
68
Supersaturated Designs
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
69
Chapter 8 Design and Analysis of Experiments 7E 2009 Montgomery
70