design of experiments: part 1 - mit opencourseware · plan for today • discussion of the reading...
TRANSCRIPT
![Page 1: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/1.jpg)
Design of Experiments:Part 1
Dan FreyAssistant Professor of Mechanical Engineering and Engineering Systems
![Page 2: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/2.jpg)
Plan for Today
• Discussion of the reading assignment• History of DOE• Full factorial designs
– The design– The model– Analysis of the sum of squares– Hypothesis testing
• Other designs– Fractional factorial designs– Central composite designs
![Page 3: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/3.jpg)
Statistics as a Catalyst to Learning• Concerned improvement of a paper
helicopter• Screening experiment (16)• Steepest ascent (5)• Full factorial (16)• Sequentially assembled CCD (16+14=30)• Ridge exploration (16)• (16+5+30+16)*4 > 250 experiments• Resulted in a 2X increase in flight time vs
the starting point design
482 −IV
cut
fold
42
Box, G. E. P. and P. T. Y. Liu, 1999, “Statistics as a Catalyst to Learning by Scientific Method: Part 1”, Journal of Quality Technology, 31 (1): 1-15.
![Page 4: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/4.jpg)
Factors Considered Initially
Image removed due to copyright restrictions.TABLE 1:
in Box and Liu, 1999. Factor Levels Used in Design I: An Initial 2 S-4
IV Screening Experiment. and FIGURE 1: The Initial Helicopter Design
![Page 5: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/5.jpg)
Screening Design
• What is the objective of screening?• What is special about this matrix of 1s and -1s?
Image removed due to copyright restrictions.TABLE 2: in Box and Liu, 1999.Design I: Layout and Data for 28 -4
IV Screening D esign
![Page 6: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/6.jpg)
Effect Estimates
Image removed due to copyright restrictions.TABLE 3: in Box and Liu, 1999.Design I: Estimates for a 28 -4
IV Screening Design
![Page 7: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/7.jpg)
Normal Probability Plots
• What's the purpose of these graphs?
Image removed due to copyright restrictions.FIGURE 2: . in Box and Liu, 1999.Design I - Normal Plots for: (a) Location effects from y and (b) Dispersion Effects from 100 log(s)
![Page 8: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/8.jpg)
"Steepest" Ascent
• What does "steep" mean in this context?
Image removed due to copyright restrictions.FIGURE 4:
in Box and Liu, 1999.Data for 5 Helicopters on the path of Steepest Ascent Calculated
from Design 1
![Page 9: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/9.jpg)
Factors Re-Considered
Wing width wWing length l
Wing area A=lwWing aspect ratio Q=l/w
Image removed due to copyright restrictions.FIGURE 1: in Box and Liu, 1999.The Initial Helicopter Design
![Page 10: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/10.jpg)
Central Composite Design
AC
B
+
-
+
+
--
runs axial and
pointscenter with2n
Enables a model to be fit with all second order polynomial terms included (i.e. A2, AB, etc.)
by Box run 2
here shown 24
3
![Page 11: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/11.jpg)
Analysis of Variance
• What would you conclude about lack of fit?• What is being used as the denominator of F?
Image removed due to copyright restrictions. TABLE 10: in Box and Liu, 1999.Design III: Analysis of Variance for Completed Composite Design
![Page 12: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/12.jpg)
Thought Questions• If we “optimize” this thing, what does
that mean?• How were design parameters
chosen?• Were important ones missed? • What does Box say about variables
being recombined to make this process more efficient?
• Is it reasonable to run 248 experiments on a simple design? Under what circumstances?
• What are the key differences between the process described here and system design in industry?
cut
fold
![Page 13: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/13.jpg)
Plan for Today
• Discussion of the reading assignment• History of DOE• Full factorial designs
– The design– The model– Analysis of the sum of squares– Hypothesis testing
• Other designs– Fractional factorial designs– Central composite designs
![Page 14: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/14.jpg)
“An experiment is simply a question put to nature … The chief requirement is simplicity: only one question should be asked at a time.”
Fisher, R.A., 1921, “Studies in Crop Variation. I. An Examination of the Yield of Dressed Grain from Broadbalk,” Journal of Agricultural Science 11:107-135.
Russell, E. J., 1926, “Field experiments: How they are made and what they are,” Journal of the Ministry of Agriculture 32:989-1001.
![Page 15: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/15.jpg)
“To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment died of.”
- Fisher, R. A., Indian Statistical Congress, Sankhya, 1938.
![Page 16: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/16.jpg)
Fisher, R. A., 1926, “The Arrangement of Field Experiments,”Journal of the Ministry of Agriculture of Great Britain, 33: 503-513.
![Page 17: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/17.jpg)
Concept Question
[ ])1()()()()()()()(41
−−−−+++≡ bccbaacababcAA
B
C
+
-
+
+
--(1) (a)
(b)
(c)
(ab)(abc)(bc)
(ac)
Say the independent experimental error of observations(a), (ab), et cetera is σε.
We define the main effect estimate Α to be
εσσ 221 )1 =A
What is the standard deviation of the main effect estimate A?
εσσ41 )2 =A εσσ 8 )3 =A εσσ =A )4
![Page 18: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/18.jpg)
Response Surface Methodology
• A method to seek improvements in a system by sequential investigation and parameter design– Variable screening– Steepest ascent– Fitting polynomial models– Empirical optimization
Box, G. E. P. and Wilson, K. B. (1951), “On the Experimental Attainment of Optimum Conditions,” Journal of the Royal Statistical Society, B13, 1-38.
![Page 19: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/19.jpg)
Cross (or Product) Arrays
A B C D E F G1 -1 -1 -1 -1 -1 -1 -12 -1 -1 -1 +1 +1 +1 +13 -1 +1 +1 -1 -1 +1 +14 -1 +1 +1 +1 +1 -1 -15 +1 -1 +1 -1 +1 -1 +16 +1 -1 +1 +1 -1 +1 -17 +1 +1 -1 -1 +1 +1 -18 +1 +1 -1 +1 -1 -1 +1
Control Factors a -1 -1b -1c -1 +1 +1 -1
+1+1 -1+1 +1
132 −III
472 −III
1347 22 −− × IIIIII
Noise Factors
Taguchi, G., 1976, System of Experimental Design.
![Page 20: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/20.jpg)
“Robust Parameter Design … is a statistical / engineering methodology that aims at reducing the performance variation of a system (i.e. a product or process) by choosing the setting of its control factors to make it less sensitive to noise variation."
Robust Parameter Design
Wu, C. F. J. and M. Hamada, 2000, Experiments: Planning, Analysis, and Parameter Design Optimization, John Wiley & Sons, NY.
![Page 21: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/21.jpg)
George Box on Sequential Experimentation
“Because results are usually known quickly, the natural way to experiment is to use information from each group of runs to plan the next …”
“…Statistical training unduly emphasizes mathematics at the expense of science. This has resulted in undue emphasis on “one-shot” statistical procedures… examples are hypothesis testing and alphabetically optimal designs.”
Box, G. E. P. 1999, “Statistics as a Catalyst to Learning by Scientific Method: Part 2”, Journal of Quality Technology, 31 (1): 16-29.
![Page 22: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/22.jpg)
One way of thinking of the great advances of the science of experimentation in this century is as the final demise of the “one factor at a time” method, although it should be said that there are still organizations which have never heard of factorial experimentation and use up many man hours wandering a crooked path.
Logothetis, N., and Wynn, H.P., 1994, Quality Through Design: Experimental Design, Off-line Quality Control and Taguchi’s Contributions,
Clarendon Press, Oxford.
Majority View on “One at a Time”
![Page 23: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/23.jpg)
My Observations of Industry• Farming equipent company has reliability problems• Large blocks of robustness experiments had been
planned at outset of the design work• More than 50% were not finished• Reasons given
– Unforseen changes– Resource pressure– Satisficing “Well, in the third experiment, we
found a solution that met all our needs, so we cancelled the rest of the experiments and moved on to other tasks…”
![Page 24: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/24.jpg)
Plan for Today
• Discussion of the reading assignment• History of DOE• Full factorial designs
– The design– The model– Analysis of the sum of squares– Hypothesis testing
• Other designs– Fractional factorial designs– Central composite designs
![Page 25: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/25.jpg)
Basic Terms in Factorial DOE• Response – the output of the system you are measuring • Factor – an input variable that may affect the response • Level – a specific value a factor may take• Trial – a single instance of the setting of factors and the
measurement of the response• Replication – repeated instances of the setting of
factors and the measurement of the response• Effect – what happens to the response when factor
levels change• Interaction – joint effects of multiple factors
![Page 26: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/26.jpg)
Cuboidal Representation
A
B
C
+
-
+
+
--
This notation indicates
observations made with factors at
particular levels.
(1) (a)
(b)
(c)
(ab)(abc)(bc)
(ac)
Exhaustive search of the space of 3 discrete 2-level factors is the
full factorial 23 experimental design
![Page 27: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/27.jpg)
Tabular Representation
Trial A B C1 -1 -1 -12 -1 -1 -13 -1 +1 +14 -1 +1 +15 +1 -1 +16 +1 -1 +17 +1 +1 -18 +1 +1 -1
23 Design
A cube has eight vertices
![Page 28: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/28.jpg)
Three Level Factors
A
B
C
33 Design
8 vertices +12 edges +6 faces +1 center =27 points
![Page 29: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/29.jpg)
Creating and Randomizing Full Factorials in Matlab
1 12 13 14 11 22 23 24 21 32 33 34 3
X = fullfact([4 3]);r=rand(1,4*3);[B,INDEX] = sort(r);Xr(1:4*3,:)=X(INDEX,:);
3 21 13 12 22 11 24 31 33 34 14 22 3
X Xr
![Page 30: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/30.jpg)
Geometric Growth of Experimental Effort
2 3 4 5 6 7 80
50
100
150
2n
n2 3 4 5 6 7 8
0
1000
2000
3000
3n
n
![Page 31: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/31.jpg)
Calculating Main Effects
A
B
C
+
-
+
+
--
[ ])1()()()()()()()(41
−−−−+++≡ bccbaacababcA
(1) (a)
(b)
(c)
(ab)(abc)(bc)
(ac)
![Page 32: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/32.jpg)
Concept Test
A
B
C
+
-
+
+
--
[ ])1()()()()()()()(41
−−−−+++≡ bccbaacababcA
(1) (a)
(b)
(c)
(ab)(abc)(bc)
(ac)
If the standard deviation of(a), (ab), et cetera is σ, what is the standard deviation of the main effect estimate A?
1) σ 2) Less than σ 3) More than σ 4) Not enough info
![Page 33: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/33.jpg)
Factor Effect Plots
A
B
30 52
20 40- +-
+
1020304050
A- A+
B+B-
A- +
B
-
+
A0
0
![Page 34: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/34.jpg)
Calculating Interactions
A
B
C
+
-
+
+
--
[ ])()()()()1()()()(41 acbcabbacabcAC −−−−+++≡
(1) (a)
(b)
(c)
(ab)(abc)(bc)
(ac)
![Page 35: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/35.jpg)
Treatment Effects Model (Two Factors)
⎪⎩
⎪⎨
⎧
===
++++=nkbjai
y ijkijjiij
,,2 ,1 ,,2 ,1 ,,2 ,1
)(K
K
K
ετββτμ
0=∑ iτ If factor a has two levels 021 =+ττ
21 ττ −=
22A
=τ21A
−=τ
![Page 36: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/36.jpg)
Treatment Effects Model (Two Factors)
⎪⎩
⎪⎨
⎧
===
++++=nkbjai
y ijkijjiij
,,2 ,1 ,,2 ,1 ,,2 ,1
)(K
K
K
ετββτμ
0)()(11
== ∑∑==
b
jij
a
iij τβτβ a+b equations
but only a+b-1are independent
interactions – there are of ab these terms
(a-1)(b-1) DOF
this is not a product
![Page 37: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/37.jpg)
Concept Test
A
B
- +-
+ If there are no interactions in this system, then the factor effect plot from
this design could look like:
1020304050
A
B+B-
1020304050
A
B+
B-1020304050
A
B+
B-
1 2 3Hold up all cards that apply.
![Page 38: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/38.jpg)
Treatment Effects Model versus the Regression Model
• If the factors are two level factors• And they are coded as (-1,+1)• Then• And
εββββ ++++= 211222110 xxxxy
12 βτ = 11 βτ −=
1212)( βτβ =
ijkijjiijy ετββτμ ++++= )(
![Page 39: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/39.jpg)
Estimation of the Parameters β
εXβy +=
We wish to minimize the sum squared error
Assume the model equation
( ) ( )XβyXβyεε −−== TTL
To minimize, we take the derivative and set it
equal to zeroβXXyX
β β
ˆ22ˆ
TTL+−=
∂∂
The solution is ( ) yXXXβ TT 1ˆ −=
And we define the fitted model βXy ˆˆ =
Recall from the lecture on multiple regression
![Page 40: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/40.jpg)
Estimation of the Parameters β when X is a 2k design
The columns are orthogonal
( ) yXXXβ TT 1ˆ −=
( ) jiijT ≠= if 0XX
( ) jin kij
T == if 2XX
( ) IXX kT
n211
=− [ ]1yXT
A
B
C
+
-+
+
--
![Page 41: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/41.jpg)
Breakdown of Sum Squares
SS due to mean
∑∑∑= = =
a
i
b
j
n
kijky
1 1 1
2
2...yN=
∑∑∑= = =
−=a
i
b
j
n
kijkT yySS
1 1 1
2... )(
ESS
∑=
−=a
iiA yybnSS
1
2..... )(
“Grand Total Sum of Squares”
“Total Sum of Squares”
∑=
−=b
jjB yyanSS
1
2..... )(
∑∑= =
−−−=a
i
b
jjiijAB yyyynSS
1 1
2........ )(
![Page 42: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/42.jpg)
Breakdown of DOFabn
number of y values
1due to the mean
abn-1total sum of squares
a-1 for factor A
ab(n-1)for error
b-1 for factor B
(a-1)(b-1) for interaction AB
![Page 43: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/43.jpg)
Hypothesis Tests in Factorial Exp• Equality of treatment effects due to
factor A or due to factor B
• Test statistic
• Criterion for rejecting H0
iH
H
i
a
oneleast at for 0:
0:
1
210
≠
====
τ
τττ K
E
A
MSMSF =0
)1(,1,0 −−> nabaFF α
iH
H
i
a
oneleast at for 0:
0:
1
210
≠
====
β
βββ K
E
B
MSMSF =0
)1(,1,0 −−> nabbFF α
![Page 44: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/44.jpg)
Hypothesis Tests in Factorial Exp• Significance of AB interactions
• Test statistic
• Criterion for rejecting H0
0 oneleast at :
, allfor 0:
1
0
≠
=
ij
ij
H
jiH
τβ
τβ
E
AB
MSMSF =0
)1(),)(1(,0 −−−> nabbaFF α
![Page 45: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/45.jpg)
Example 5-1 – Battery Life FF= fullfact([3 3]);X=[FF; FF; FF; FF];Y=[130 150 138 34 136 174 20 25 96 155 188 110 40 122 120 70 70 104 74 159 168 80 106 150 82 58 82 180 126 160 75 115 139 58 45 60]';
[p,table,stats]=anovan(Y,{X(:,1),X(:,2)},'interaction');
hold off; hold onfor i=1:3; for j=1:3; intplt(i,j)=(1/4)*sum(Y.*(X(:,1)==j).*(X(:,2)==i)); endplot([15 70 125],intplt(:,i)); end
![Page 46: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/46.jpg)
Interaction plot
ANOVA table
![Page 47: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/47.jpg)
Regression – Battery Life FF= fullfact([3 3]);A=FF(:,1)-2; B=FF(:,2)-2; ones(1:3*3)=1;R=[ones' A B A.*A B.*B A.*B ];X=[R; R; R; R];Y=[130 150 138 34 136 174 20 25 96 155 188 110 40 122 120 70 70 104 74 159 168 80 106 150 82 58 82 180 126 160 75 115 139 58 45 60]';
[b,bint,r,rint,stats] = regress(Y,X,0.05);[t,m] = meshgrid(-1:.1:1,-1:.1:1);Yhat= b(1)+b(2)*t+b(3)*m+b(4)*t.*t+b(5)*m.*m+b(6)*t.*m;hold off; h=plot3(t,m,Yhat);hold on; scatter3(X(:,2),X(:,3),Y);
![Page 48: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/48.jpg)
Plan for Today
• Discussion of the reading assignment• History of DOE• Full factorial designs
– The design– The model– Analysis of the sum of squares– Hypothesis testing
• Fractional factorial designs
![Page 49: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/49.jpg)
Fractional Factorial ExperimentsCuboidal Representation
A
B
C
+
-
+
+
--
This is the 23-1 fractional factorial.
![Page 50: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/50.jpg)
Fractional Factorial ExperimentsTabular Representation
Trial A B C D E F G1 -1 -1 -1 -1 -1 -1 -12 -1 -1 -1 +1 +1 +1 +13 -1 +1 +1 -1 -1 +1 +14 -1 +1 +1 +1 +1 -1 -15 +1 -1 +1 -1 +1 -1 +16 +1 -1 +1 +1 -1 +1 -17 +1 +1 -1 -1 +1 +1 -18 +1 +1 -1 +1 -1 -1 +1
27-4 Design Resolution III.
FG=-A+1+1+1+1-1-1-1-1
Two-way interactions are aliased with main effects
![Page 51: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/51.jpg)
Fractional Factorial ExperimentsCuboidal Representation
A
B
C
+
-
+
+
--
[ ])()1()()(21 bcacabA −−+=
(1)
(ac)
(bc)
(ab)
![Page 52: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/52.jpg)
One at a Time Experiments
A
B
C
+
-
+
+
--
Provides resolution of individual factor effectsBut the effects may be biased
)1()( −≈ aA
(1) (a)
(b)
(c)
If the standard deviation of (a) and (1) is σ,what is the standard deviation of A?
![Page 53: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/53.jpg)
Efficiency• The variance for OFAT is
using 4 experiments• The standard deviation for 23-1 was
using 4 experiments• The inverse ratio of variance per unit is
considered a measure of relative efficiency
• The 23-1 is considered 2 times more efficient than the OFAT
σ2
σ
[ ] [ ] 244
222
=σσ
![Page 54: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/54.jpg)
Overview Research
Complex Systems
Methodology Validation
Concept Design
Outreach to K-12
Adaptive Experimentation and Robust Design
B D
AB AD
ABC ABDABCD
A
AC
main effects
two-factor interactions
BC BD CD
C
three-factor interactionsACD BCD
four-factor interactions
B D
AB AD
ABC ABDABCDABCD
A
AC
main effects
two-factor interactions
BC BD CD
C
three-factor interactionsACD BCD
four-factor interactions
( ) ∫ ∫∞ ∞
−
⎟⎠⎞
⎜⎝⎛ +−+
−+
−−⎟⎟
⎠
⎞⎜⎜⎝
⎛
∗∗
+−+
⎥⎦
⎤⎢⎣
⎡⎟⎟⎠
⎞⎜⎜⎝
⎛
⎟⎟⎠
⎞⎜⎜⎝
⎛>>>
012
222
21)2(221
21
1212
2
222
22
2
21
21
21)2(
21
210Pr dxdx
n
exerfn
xxx
INTMEINT
n
xxn
INTij
INTMEINT
ε
σσσσ
σσσσ
σπ
βββ
ε
PBS show “Design Squad”
![Page 55: Design of Experiments: Part 1 - MIT OpenCourseWare · Plan for Today • Discussion of the reading assignment • History of DOE • Full factorial designs – The design – The](https://reader031.vdocuments.net/reader031/viewer/2022031305/5bf0323909d3f2b7578b7d0d/html5/thumbnails/55.jpg)
Next Steps• Friday 27 April
– Recitation to support the term project• Monday 30 April
– Design of Experiments: Part 2• Wednesday 2 May
– Design of Computer Experiments• Friday 4 May
– Exam review• Monday 7 May – Frey at NSF• Wednesday 9 May – Exam #2