5/12/2015

Design of Engineering Experiments

Review

Summary
A complete replicate of a 2k design requires

2 2 ............... 2 2k

observations and is called a 2k factorial design

2
2 Factorial Design

ContrastA

Main Effect - A

Main / Interactio n Effect

Contrast
n 2k 1

Statistical Analysis
To determine which effects are significantly different from zero in a 2k experiment. Three
methods can be used:
1. By comparing the magnitude of an effect to its estimated error.
2. Using a regression model in which each effect is associated with a regression
coefficient.
3. ANOVA

5/12/2015

DOE 2k Factorial Design

An engineer suspects that the surface finish of metal parts is influenced by the type
of paint used and the drying time. She selects two drying times20 and 30
minutesand uses two types of paint. Three parts are tested with each combination
of paint type and drying time. The data are as follows:
Number of Factors (k): 2
Paint Type - A
Drying Time - B
Number of Levels for each factor: 2
Paint Type: Paint 1, Paint 2
Drying Time: 20 min, 30 min

Estimate the effects and develop

a regression model with the most
significant effects.

Response Variable: 1
Surface Finish
Number of Replicates (n): 3
Design: 22

DOE 2k Factorial Design

EFFECTS

REPLICATES

AB

(1)

74

64

50

92

86

68

78

85

92

ab

66

45

85

5/12/2015

DOE 2k Factorial Design

EFFECTS

REPLICATES

AB

Totals

(1)

74

64

50

188

92

86

68

246

78

85

92

255

ab

66

45

85

196

STEP # 1: Calculate the Main and the Interaction effects

Main Effect of A:

Contrast A (1) a b ab 188 246 255 196 0.166

32
n 2 k 1
n 2 k 1

Main Effect of B:

Contrast B
n2

k 1

(1) a b ab 188 246 255 196

2.833
32
n 2 k 1

Interaction Effect of AB:

Contrast AB
n2

k 1

(1) a b ab 188 246 255 196

19.5
32
n 2 k 1

DOE 2k Factorial Design

STEP # 2: Develop an Initial Regression Model

Y 0 1 X 1 2 X 2 12 X 1 X 2
0 ( Intercept ) :

Grand Average of all the 12 observations

0 ( Intercept ) 73.75

Effect A 0.166 0.083

2

12

Effect AB
2

Effect B
2

2.833
1.416
2

19.5
9.75
2

Initial Regression Model

Y 73.75 0.083 X 1 1.416 X 2 9.75 X 1 X 2

5/12/2015

DOE 2k Factorial Design

STEP # 3: Statistical Analysis
1.

By comparing the magnitude of an effect to its estimated error.

2.

ANOVA

3.

Using a regression model in which each effect is associated with a

regression coefficient.

DOE 2k Factorial Design

STEP # 3: Statistical Analysis
First Method: By comparing the magnitude of an effect to its estimated
error.
EFFECTS

REPLICATES

AB

Totals

Average

Variance

(1)

74

64

50

188

62.6

145.34

92

86

68

246

82

156

78

85

92

255

85

49

ab

66

45

85

196

65.3

400

Standard Error

145.34 156 49 400

187.58
4

seEffect

2
n2

k 2

187.58
7.90
31

5/12/2015

DOE 2k Factorial Design

STEP # 3: Statistical Analysis
First Method: By comparing the magnitude of an effect to its estimated
error.
2
187.58
seEffect

7.90
31
n 2k 2

t statistic
Effect

DOF 2 k n 1 8

Effect
se(effect )

Estimate of the
effect

Standard error
of the effect

t - statistic p - Value

-0.166

7.9

- 0.21

> 0.4

2.833

7.9

0.358

0.4 < p < 0.25

AB

-19.5

7.9

- 2.46

0.025 < p < 0.01

Hence, the most significant effect is the AB interaction effect

DOE 2k Factorial Design

STEP # 3: Statistical Analysis
Second Method: ANOVA

Determination of sum of squares

SS A

Contrast A 2 (1) a b ab2 188 246 255 1962

34
n 2k
n 2k

0.083

Contrast B 2 (1) a b ab2 188 246 255 1962 24.08

34
n 2k
n 2k
Contrast AB 2 (1) a b ab2 188 246 255 1962 1140.75

SS B
SS AB

n 2k

n 2k

34

SST ([Each observation] [Overall Mean])2 = 2667

SS E SST SS A SS B SS AB 2667 0.083 24.08 1140.75 1502.08

5/12/2015

DOE 2k Factorial Design

STEP # 3: Statistical Analysis
Second Method: ANOVA
Effect

Sum of
Squares

DOF

MS

f-statistic

p - value

0.083

0.083

0.0004

> 0.25

24.08

24.08

0.1281

> 0.25

AB

1140.75

1140.75

6.071

0.025 < p < 0.05

Error

1503.08

187.88

Total

2667

11

Degrees of freedom
SSA = (2-1) = 1
SSB = (2-1) = 1
SSAB = (a-1)(b-1) = 1
SST = n2k-1
SSE = 2k (n-1)

Hence, the most significant effect is

the AB interaction effect

DOE 2k Factorial Design

STEP # 3: Statistical Analysis
Third Method: Regression Analysis
Initial Regression Model
secoefficient

Y 73.75 0.083 X 1 1.416 X 2 9.75 X 1 X 2

se(effect ) 1 2
1 187.58

3.95
2
2 n 2 k 2 2 31

t statistic

Coefficient
se(coefficient )

DOF 2k n 1 8
Independent
variable

Coefficient Standard error of

Estimate
the Coefficient

tstatistic

p - Value

Intercept

73.75

3.95

18.67

p < 0.0005

-0.083

3.95

- 0.021

p > 0.4

1.416

3.95

0.358

0.25 < p < 0.4

AB

-9.75

3.95

2.468

0.01 < p < 0.025

Hence, the most significant effect is the AB interaction effect

5/12/2015

DOE 2k Factorial Design

STEP # 4: Refine the Model

Y 73.75 0.083 X1 1.416 X 2 9.75 X1 X 2

STEP # 5: Residual Analysis

DOE- Problems
A nickel-titanium alloy is used to make components for jet turbine aircraft engines. Cracking
is a potentially serious problem in the final part, as it can lead to non-recoverable failure. A
test is run at the parts producer to determine the effects of four factors on cracks. The four
factors are pouring temperature (A), titanium content (B), heat treatment method (C), and the
amount of grain refiner used (D). Two replicates of a 24 design are run, and the length of
crack (in m) induced in a sample coupon subjected to a standard test is measured. The
data are shown below:
Treatment

Replicate

Combination

Replicate
II

(1)

7.037

6.376

14.707

15.219

11.635

12.089

ab

17.273

17.815

10.403

10.151

ac

4.368

4.098

bc

9.360

9.253

abc

13.440

12.923

8.561

8.951

ad

16.867

17.052

bd

13.876

13.658

abd

19.824

19.639

cd

11.846

12.337

acd

6.125

5.904

bcd

11.190

10.935

abcd

15.653

15.053

1. Estimate the factor effects. Which

factors appear to be large?
2. Conduct an analysis of variance.
Do any of the factors affect
cracking? Use = 0.05.
3. Write down a regression model
that can be used to predict crack
length as a function of the
significant main effects and
interactions you have identified in
part (b).

5/12/2015

DOE- Problems
1. Estimate the factor effects. Which factors appear to be large?
Treatment

Replicate Replicate

Combination

II

+
+
+
+
+
+
+
+

+
+
+
+
+
+
+
+

+
+
+
+
+
+
+
+

+
+
+
+
+
+
+
+

-1
a
b
ab
c
ac
bc
abc
d
ad
bd
abd
cd
acd
bcd
abcd

7.037
14.707
11.635
17.273
10.403
4.368
9.36
13.44
8.561
16.867
13.876
19.824
11.846
6.125
11.19
15.653

6.376
15.219
12.089
17.815
10.151
4.098
9.253
12.923
8.951
17.052
13.658
19.639
12.337
5.904
10.935
15.053

Total

Average

13.413
29.926
23.724
35.088
20.554
8.466
18.613
26.363
17.512
33.919
27.534
39.463
24.183
12.029
22.125
30.706

6.7065
14.963
11.862
17.544
10.277
4.233
9.3065
13.1815
8.756
16.9595
13.767
19.7315
12.0915
6.0145
11.0625
15.353
11.98806

Variance

0.218461
0.131072
0.103058
0.146882
0.031752
0.03645
0.005724
0.133645
0.07605
0.017112
0.023762
0.017113
0.120541
0.024421
0.032512
0.18
0.08116

DOE- Problems
The table of + and - signs

Main / Interactio n Effect

Contrast
n 2k 1

5/12/2015

DOE- Problems
Term
A
B
C
D
AB
AC
AD
BC
BD
CD
ABC
ABD
ACD
BCD
ABCD

Effect
3.01888
3.97588
-3.59625
1.95775
1.93412
-4.00775
0.0765
0.096
0.04725
-0.076875
3.1375
0.098
0.019125
0.035625
0.014125

pouring temperature (A),

titanium content (B),
heat treatment method (C),
the amount of grain refiner used (D)

DOE Problems
2. Conduct an analysis of variance. Do any of the factors affect cracking?
Use = 0.05.

Statistical Analysis
Analysis of Variance - ANOVA
Determination of sum of squares

SS effect

Contrast

effect

n2

SST ([Each observation] [Overall Mean])2

SS E SST ( SS effects )

Degrees of freedom
SSA = (a-1)=(2-1) = 1
SSB = (b-1)=(2-1) = 1
SSAB = (a-1)(b-1) = 1
SST = n2k-1
SSE = 2k (n-1)

5/12/2015

DOE- Problems
2. Conduct an analysis of variance. Do any of the factors affect cracking?
Use = 0.05.
Effect
A
B
C
D
AB
AC
AD
BC
BD
CD
ABC
ABD
ACD
BCD
ABCD
Error
Total

Sum of
Squares

DF

Mean
Square

F-Statistic

P-Value

72.91

72.91

898.34

< 0.0001

126.46

126.46

1558.17

< 0.0001

103.46

103.46

1274.82

< 0.0001

30.66

30.66

377.8

< 0.0001

29.93

29.93

368.74

< 0.0001

128.5

128.5

1583.26

< 0.0001

0.047

0.047

0.58

0.4586

0.074

0.074

0.91

0.3547

0.018

0.018

0.22

0.6453

0.047

0.047

0.58

0.4564

78.75

78.75

970.33

< 0.0001

0.077

0.077

0.95

0.345

2.93E-03

2.93E-03

0.036

0.8518

0.01

0.01

0.13

0.7282

1.60E-03

1.60E-03

0.02

0.8902

1.3024

16

0.081

572.25

31

Effects, A, B, C,
D, AB, AC, ABC
are significant

DOE- Problems
3. Write down a regression model that can be used to predict crack length as
a function of the significant main effects and interactions you have
identified in part (b).
Based upon the ANOVA analysis the regression model can be written as:

y 11.99 1.51A 1.99B 1.80C 0.98D 0.97 AB 2.00 AC 1.57 ABC

where
Crack length (y)
pouring temperature (A),
titanium content (B),
heat treatment method (C),
the amount of grain refiner
used (D)

Intercept Grand average of all observations

Coeff

Effect
2

10

5/12/2015

DOE - Blocking & Confounding in the 2k design

Blocking a Replicated Design
If there are n replicates of the design, then each replicate is a block
Each replicate is run in one of the blocks (time periods, batches of

raw material, etc.)

Runs within the block are randomized

Consider the example where k = 2 factors, n = 3 replicates

n Block totals 2 Grand total 2

SS Blocks

2k
n2k
i 1

SST SSTreatments SSBlocks SSE

DOE- Problems
If BLOCKING is performed in the above experiment, what effect would that
have on the analysis of variance and the model:
A
+
+
+
+
+
+
+
+

B
+
+
+
+
+
+
+
+

C
+
+
+
+
+
+
+
+

Treatment
Replicate Replicate
D Combination
I
II
(1)
7.037
6.376
a
14.707
15.219
b
11.635
12.089
ab
17.273
17.815
c
10.403
10.151
ac
4.368
4.098
bc
9.360
9.253
abc
13.440
12.923
+
d
8.561
8.951
+
ad
16.867
17.052
+
bd
13.876
13.658
+
abd
19.824
19.639
+
cd
11.846
12.337
+
acd
6.125
5.904
+
bcd
11.190
10.935
+
abcd
15.653
15.053

n Block totals 2 Grand total 2

SS Blocks

2k
n2 k
i 1

SS Blocks 0.0158

11

5/12/2015

DOE- Problems
ANOVA ANALYSIS
Before Blocking
A
B
C
D
AB
AC
AD
BC
BD
CD
ABC
ABD
ACD
BCD
ABCD

Sum of
Squares
72.91
126.46
103.46
30.66
29.93
128.5
0.047
0.074
0.018
0.047
78.75
0.077
2.93E-03
0.01
1.60E-03

DF
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

Error
Total

1.3024
572.25

16
31

Effect

After Blocking
Mean
Square
72.91
126.46
103.46
30.66
29.93
128.5
0.047
0.074
0.018
0.047
78.75
0.077
2.93E-03
0.01
1.60E-03

FStatistic
898.34
1558.17
1274.82
377.8
368.74
1583.26
0.58
0.91
0.22
0.58
970.33
0.95
0.036
0.13
0.02

0.081

P-Value
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
< 0.0001
0.4586
0.3547
0.6453
0.4564
< 0.0001
0.345
0.8518
0.7282
0.8902

A
B
C
D
AB
AC
AD
BC
BD
CD
ABC
ABD
ACD
BCD
ABCD

Sum of
Squares
72.91
126.46
103.46
30.66
29.93
128.5
0.047
0.074
0.018
0.047
78.75
0.077
2.93E-03
0.01
1.60E-03

Blocks

0.0158

Error
Total

1.2866
572.25

16
31

Effect

DF
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

Mean
FSquare Statistic P-Value
911.375 < 0.0001
72.91
1580.75 < 0.0001
126.46
1293.25 < 0.0001
103.46
383.25
30.66
< 0.0001
374.125 < 0.0001
29.93
1606.25 < 0.0001
128.5
0.5875
0.047
0.4545
0.925
0.074
0.3505
0.225
0.018
0.6417
0.5875
0.047
0.4545
984.375 < 0.0001
78.75
0.9625
0.077
0.3412
2.93E-03 0.036625 0.8506
0.125
0.01
0.7283
0.02
1.60E-03
0.8902

0.080

The nuisance factor if there was ANY, did not effect the results
significantly.

DOE- Problems
If ONLY a single replicate is run for the above experiment, then conduct the
analysis using the confounding technique.

A
+
+
+
+
+
+
+
+

B
+
+
+
+
+
+
+
+

C
+
+
+
+
+
+
+
+

D
+
+
+
+
+
+
+
+

Treatment
Combination
(1)
a
b
ab
c
ac
bc
abc
d
ad
bd
abd
cd
acd
bcd
abcd

Replicate
I
7.037
14.707
11.635
17.273
10.403
4.368
9.360
13.440
8.561
16.867
13.876
19.824
11.846
6.125
11.190
15.653

12

5/12/2015

DOE - Blocking & Confounding in the 2k design

Confounding in Blocks
Confounding

is a design technique for arranging a complete

unreplicated factorial experiment in blocks, where the block size is
smaller than the number of treatment combinations in one replicate.

The usual rule is to pick an effect you are least interested in, and this is usually
the highest order interaction, as a means of specifying how to do blocking.

DOE - Blocking & Confounding in the 2k design

Confounding in Blocks 24 design
The table of + and - signs

13

5/12/2015

DOE- Problems
If ONLY a single replicate is run for the above experiment, then conduct the
analysis using the confounding technique.
A
+
+
+
+
+
+
+
+

B
+
+
+
+
+
+
+
+

C
+
+
+
+
+
+
+
+

D
+
+
+
+
+
+
+
+

Treatment
Combination
(1)
a
b
ab
c
ac
bc
abc
d
ad
bd
abd
cd
acd
bcd
abcd

Replicate
I
7.037
14.707
11.635
17.273
10.403
4.368
9.360
13.440
8.561
16.867
13.876
19.824
11.846
6.125
11.190
15.653

Block - 1

Block - 2

(1) = 7.037
ab = 17.273
ac = 4.368
bc = 9.360
ad = 16.867
bd = 13.876
cd = 11.846
abcd = 15.653

a = 14.707
b = 11.635
c
= 10.403
abc = 13.440
d
= 8.561
abd = 19.824
acd = 6.125
bcd = 11.190

n Block totals 2 Grand total 2

SS Blocks

2k
n2 k
i 1

DOE- Problems
ANOVA Analysis

14

5/12/2015

DOE- The 2k-p Fractional Factorial Design

The One-Half Fraction of the 2k design
Notation: because the design has 2k/2 runs, its referred to as a 2k-1
Consider a really simple case, the 2 3-1

Notice that the 23-1 design is formed by selecting only those treatment

combinations that have a plus in the ABC column.

DOE- The 2k-p Fractional Factorial Design

The One-Half Fraction of the 2k design Basic Definitions
ABC is called as the generator of this particular fraction.
Defining Relation: is the set of all columns that are equal to the identity

column I.
Hence, I = ABC which is called as the defining relation for this

example.
Two-factor interaction effects

Main Effects of the factors

Contrast
n 2 k 1
n = replicate

15

5/12/2015

DOE- The 2k-p Fractional Factorial Design

Construction of a One-half Fraction
The basic design; the design generator

Any Interaction effect could be used as a generator for the fractional factorial.
However, doing so, will result in loses of important information.
A = A . AB = B
Thus A and B effects get aliased.

DOE- Problems

A nickel-titanium alloy is used to

make components for jet turbine
aircraft engines.
Cracking is a
potentially serious problem in the
final part, as it can lead to nonrecoverable failure. A test is run at
the parts producer to determine the
effects of four factors on cracks.
The four factors are pouring
temperature (A), titanium content
(B), heat treatment method (C), and
the amount of grain refiner used (D).
If ONLY a single replicate is run
for the above experiment, then
conduct the analysis using the
fractional factorial design.

A
+
+
+
+
+
+
+
+

B
+
+
+
+
+
+
+
+

C
+
+
+
+
+
+
+
+

D
+
+
+
+
+
+
+
+

Treatment
Combination
(1)
a
b
ab
c
ac
bc
abc
d
ad
bd
abd
cd
acd
bcd
abcd

Replicate
I
7.037
14.707
11.635
17.273
10.403
4.368
9.360
13.440
8.561
16.867
13.876
19.824
11.846
6.125
11.190
15.653

16

5/12/2015

DOE Fractional factorial Design

First Technique: The table of + and - signs
Defining Relationship:

I ABCD

DOE Fractional factorial Design

First Technique: The table of + and - signs
Defining Relationship:

I ABCD

(1)

Crack
Length
7.037

ad

16.867

bd

13.876

cd

11.846

ab

17.273

ac

4.368

bc

9.36

abcd

15.653

Treatment

17

5/12/2015

DOE- Problems
Second Technique: From the basic design

Defining Relationship: I ABCD

Basic Design : 23
A
B
C
D = ABC Treatment
-1

-1

-1

-1

(1)

Crack
Length
7.037

-1

-1

ad

16.867

-1

-1

bd

13.876

-1

-1

cd

11.846

-1

-1

ab

17.273

-1

-1

ac

4.368

-1

-1

bc

9.36

abcd

15.653

Aliases

A BCD
AB CD

B ACD
C ABD
AC BD and AD BC

D ABC

DOE- Problems
Basic Design
A
-1
1
-1
-1
1
1
-1
1

B
-1
-1
1
-1
1
-1
1
1

C
-1
-1
-1
1
-1
1
1
1

D = ABC
-1
1
1
1
-1
-1
-1
1

Treatme
nt
(1)
ad
bd
cd
ab
ac
bc
abcd

Crack
Length
7.037
16.867
13.876
11.846
17.273
4.368
9.36
15.653
Contrasts
Effects

A BCD 3.01
B ACD 4.01
C ABD 3.45
D ABC 5.05

CA
-7.037
16.867
-13.876
-11.846
17.273
4.368
-9.36
15.653
12.042
3.0105

AB CD 1.83
AC BD 3.60
AD BC 0.388

CB
CC
CD
CAB
-7.037 -7.037 -7.037 7.037
-16.867 -16.867 16.867 -16.867
1 -13.876
13.876 -13.876 k13.876
-11.846 11.846 11.846 11.846
17.273 -17.273 -17.273 17.273
-4.368n =4.368
-4.368 -4.368
replicate
9.36
9.36
-9.36
-9.36
15.653 15.653 15.653 15.653
16.044 -13.826 20.204 7.338
4.011 -3.4565 5.051 1.8345

Contrast

n2

CAC
7.037
-16.867
13.876
-11.846
-17.273
4.368
-9.36
15.653
-14.412
-3.603

CAD
7.037
16.867
-13.876
-11.846
-17.273
-4.368
9.36
15.653
1.554
0.3885

1. Factors A, B, C and D are significant.

2. AC + BD also looks significant, but we
will have to run a few more
experiments to ascertain which of
them are significant.

18