Ch14 - Design of Experiments

7/8/2020
Design of Experiments Learning Objectives for Chapter 14

14
CHAPTER OUTLINE
with Several Factors
After careful study of this chapter, you should be able to do
14-1 Introduction 14-5.4 Addition of Center Points to the following:
14-2 Factorial Experiments a 2k Design 1. Design and conduct engineering experiments involving several
14-3 Two-Factor Factorial 14-6 Blocking & Confounding in the factors using the factorial design approach.
Applied Statistics and Probability for Experiments 2k Design 2. Know how to analyze and interpret main effects and interactions.
Engineers 14-3.1 Statistical Analysis of the 14-7 Fractional Replication of the 2k 3. Understand how the ANOVA is used to analyze the data from these
Fixed-Effects Model Design experiments.
14-3.2 Model Adequacy Checking 14-7.1 One-Half Fraction of the 2k 4. Assess model adequacy with residual plots.
Sixth Edition 14-3.3 One Observation per Cell Design 5. Know how to use the two-level series of factorial designs.
14-4 General Factorial Experiments 14-7.2 Smaller Fractions: The 2k−p 6. Understand how two-level factorial designs can be run in blocks.
Douglas C. Montgomery George C. Runger 14-5 2k Factorial Designs Fractional Factorial 7. Design and conduct two-level fractional factorial designs.
14-5.1 22 Design 14-8 Response Surface Methods and 8. Test for curvature in two-level factorial designs by using center
Chapter 14 14-5.2 2k Design for k ≥ 3 Factors Designs points.
14-5.3 Single Replicate of the 2k 9. Use response surface methodology for process optimization
Design of Experiments with Several Factors Design experiments.
Chapter 14 Title and Outline 2 Chapter 14 Learning Objectives 3

Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved.
14-1: Introduction
• An experiment is a test or series of tests.

• The design of an experiment plays a major role
in the eventual solution of the problem.
• In a factorial experimental design,
experimental trials (or runs) are performed at all
combinations of the factor levels.
• The analysis of variance (ANOVA) will be
used as one of the primary tools for statistical
data analysis.
Sec 14-1 Introduction 6

Fig_14-1 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Fig_14-2 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved.
14-2: Factorial Experiments 14-2: Factorial Experiments

Definition Definition
By a factorial experiment we mean that in each complete trial or replicate of the By a factorial experiment we mean that in each complete trial or replicate of the
experiment all possible combinations of the levels of the factors are investigated. experiment all possible combinations of the levels of the factors are investigated.
Table 14-2 A Factorial Experiment with Interaction

Table 14-1 A Factorial Experiment with Two Factors
Sec 14-2 Factorial Experiments 7 Sec 14-2 Factorial Experiments 8

Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Fig_14-5 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved.
1
7/8/2020
Fig_14-6 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Fig_14-7 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Fig_14-8 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved.
14-3: Two-Factor Factorial Experiments 14-3: Two-Factor Factorial Experiments

Table 14-3 Data Arrangement for a Two-Factor Factorial Design
Factor B The observations may be described by the

1 2  b Totals Averages linear statistical model:
y111, y112, y121, y122, y1b1, y1b2,
1 y1.. y1..
, y11n , y12n , y1bn
y211, y212, y221, y222, y2b1, y2b2, y2.. i = 1, 2,  , a
Factor A 2
, y21n , y22n , y2bn
y2..

Yijk =  + i +  j + ()ij + ijk  j = 1, 2,  , b (14-1)

k = 1, 2,  , n
a
ya11, ya12, ya21, ya22, yab1, yab2,
ya.. ya .. 
, ya1n , ya2n , yabn
Totals y.1. y.2. y.b. y
Averages y.1. y.2. y.b. y...
Sec 14-3 Two-Factor Factorial Experiments 14 Sec 14-3 Two-Factor Factorial Experiments 15
Fig_14-9 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved.
14-3: Two-Factor Factorial Experiments 14-3: Two-Factor Factorial Experiments 14-3: Two-Factor Factorial Experiments
14-3.1 Statistical Analysis of the Fixed-Effects Model 14-3.1 Statistical Analysis of the Fixed-Effects Model 14-3.1 Statistical Analysis of the Fixed-Effects Model
The hypotheses that we will test are as follows: The sum of squares identity for a two-factor ANOVA is
b n
yi .. =  
yi ..
yi .. = i = 1, 2, , a
   (yijk − y...)2 = bn  ( yi .. − y...)2 + an  (y. j. − y...)2
yijk a b n a b
j =1 k =1 bn 1. H0: 1 = 2 =  = a = 0 (no main effect of factor A)
i =1 j =1 k =1 i =1 j =1
H1: at least one i  0 (14-3)
 (yij. − yi .. − y. j. + y...)2 +    (yijk − yij.)2
a n y. j. a b a b n
y. j. =   yijk y . j. = j = 1, 2, , b +n
i =1 k =1
an 2. H0: 1 = 2 =  b = 0 (no main effect of factor B) (14-2) i =1 j =1 i =1 j =1 k =1
n
yij . H1: at least one j  0
yij .. = y yij . = i = 1, 2, , a
k =1
ijk
n j = 1, 2, , b or symbolically,
3. H0: ()11 = ()12 = = ()ab = 0 (no interaction)
y... = 
a b n y... H1: at least one ()ij  0 SST = SSA + SSB + SSAB + SSE
 yijk y... =
abn j = 1, 2, , a
(14-4)
i =1 j =1 k =1
Sec 14-3 Two-Factor Factorial Experiments 16 Sec 14-3 Two-Factor Factorial Experiments 17 Sec 14-3 Two-Factor Factorial Experiments 18
2
7/8/2020
To test H0: i = 0 use the ratio Definition
Table 14-4 ANOVA Table for a Two-Factor Factorial, Fixed-Effects Model
Computing formulas for the sums of squares in a two-factor analysis of variance.
MS A Source of Sum of Degrees of
F0 = y 2...
a b n
SST =   2
yijk − (14-5) Variation Squares Freedom
Mean Square F0
MS E i =1 j =1 k =1 abn
SS A
A treatments SSA a-1 MS A = MS A
a 2 2 a −1 MS E
y .. y ...
To test H0: j = 0 use the ratio SS A =  −
i
(14-6) B treatments SSB b-1 SS
MS B = B MS B
i =1 bn abn b −1 MS E
Interaction SSAB (a - 1)(b - 1) SS AB
MS B MS AB =
MS AB
F0 = SS B = 
b y 2. j .
−
y 2... (a − 1)(b − 1) MS E
MS E (14-7)
j =1 an abn Error SSE ab(n - 1) MS E =
SS E
ab(n − 1)
To test H0: ()ij = 0 use the ratio a b yij2 . y 2... Total SST abn - 1
SS AB =   − − SS A − SS B (14-8)
i =1 j =1 n abn
MS AB
F0 =
MS E SSE = SST − SSAB − SSA − SSB (14-9)
EXAMPLE 14-1 Aircraft Primer Paint Aircraft primer paints are applied Example 14-1 Example 14-1 a b n 2
y ...
to aluminum surfaces by two methods: dipping and spraying. The purpose SST =   y 2
ijk −
of the primer is to improve paint adhesion, and some parts can be primed Table 14-5 Adhesion Force Data for Example 14-1 i =1 j =1 k =1 abn
( 89.8 )
2
using either application method. The process engineering group = ( 4.0 ) + ( 4.5 ) +
2 2
+ ( 5.0 ) −
2
= 10.72
yi ..
responsible for this operation is interested in learning whether three Primer Type Dipping Spraying 18
different primers differ in their adhesion properties. A factorial experiment 1 4.0, 4.5, 4.3 12.8 5.4, 4.9, 5.6 15.9 28.7 a
yi2 .. y ...
2
was performed to investigate the effect of paint primer type and 2 5.6, 4.9, 5.4 15.9 5.8, 6.1, 6.3 18.2 34.1 SS types =  −
i =1 bn abn
application method on paint adhesion. For each combination of primer 3 3.8, 3.7, 4.0 11.5 5.5, 5.0, 5.0 15.5 27.0
( 28.7 ) + ( 34.1) + ( 27.0 ) (89.8 )
2 2 2 2
type and application method, three specimens were painted, then a finish y. j . 40.2 49.6 89.8 = y = − = 4.58
6 18
paint was applied, and the adhesion force was measured. The data from
the experiment are shown in Table 14-5. The circled numbers b y 2. j . 2
y ...
SSmethods =  −
in the cells are the cell totals yij. The sums of squares required to perform j =1 an abn
the ANOVA are computed as follows:
=
(40.2)2 + (49.6)2 − (89.8)2 = 4.91
9 18
Example 14-1 Example 14-1 Example 14-1
The ANOVA is summarized in Table 14-6. The experimenter has decided
a b yij2 . 2
y ... to use a = 0.05. Since f0.05,2,12 = 3.89 and f0.05,1,12 = 4.75, we conclude that
SSinteraction =   − − SS types − SSmethods Table 14-6 ANOVA for Example 14-1
i =1 j =1 n abn the main effects of primer type and application method affect adhesion
Source of Variation Sum of Degrees of Mean f0 P-Value
force. Furthermore, since 1.5 < f0.05,2,12, there is no indication of interaction
(12.8) + (15.9 ) + (11.5) + (15.9 ) + (18.2 ) + (15.5) (89.8)
2 2 2 2 2 2 2
Squares Freedom Square
= − − 4.58 − 4.91 = 0.24 between these factors. The last column of Table 14-6 shows the P-value Primer types 4.58 2 2.29 27.86 2.7  E-5
3 18 for each F-ratio. Notice that the P-values for the two test statistics for the
main effects are considerably less than 0.05, while the P-value for the test Application methods 4.91 1 4.91 59.70 4.7  E-6
and statistic for the interaction is greater than 0.05. Interaction 0.24 2 0.12 1.47 0.2621
Practical Interpretation: A graph of the cell adhesion force averages yij . Error 0.99 12 0.08
SSE = SST − SStypes − SSmethods − SSinteraction versus levels of primer type for each application method is shown in Fig. Total 10.72 17
= 10.72 − 4.58 − 4.91 − 0.24 = 0.99 14-10. The no-interaction conclusion is obvious in this graph, because the
two lines are nearly parallel. Furthermore, since a large response
indicates greater adhesion force, we conclude that spraying is the best
application method and that primer type 2 is most effective.
3
7/8/2020
14-3.1 Statistical Analysis of the Fixed-Effects Model 14-3.2 Model Adequacy Checking 14-3.2 Model Adequacy Checking
Example 14-1
Table 14-8 Residuals for the Aircraft Primer Experiment in Example 14-1
Figure 14-10 Graph of Application Method

average adhesion force Primer type Dipping Spraying
versus primer types for 0.10, -0.40, 0.30
Figure 14-11 Normal
1 -0.27, 0.23, 0.03
both application methods. probability plot of the
2 0.30, -0.40, 0.10 -0.27, 0.03, 0.23 residuals from the aircraft
primer paint experiment in
3 -0.03, -0.13, 0.17 0.33, -0.17, -0.17
Example 14-1.
14-3.2 Model Adequacy Checking 14-3.2 Model Adequacy Checking 14-3.2 Model Adequacy Checking
Figure 14-13 Plot of residuals from the aircraft primer paint experiment versus Figure 14-14 Plot of residuals from the aircraft primer paint experiment versus
Figure 14-12 Plot of residuals from the aircraft primer paint
application method. predicted values Y ijk .
experiment versus primer type.
14-4: General Factorial Experiments 14-4: General Factorial Experiments 14-4: General Factorial Experiments
Table 14-9 Analysis of Variance Table for the Three-Factor Fixed Effects Model EXAMPLE 14-2 Surface Roughness A mechanical engineer is
Model for a three-factor factorial experiment Source of Sum of studying the surface roughness of a part produced in a metal-cutting
Variation Squares Degrees of Freedom Mean Square Expected Mean Squares F0
operation. Three factors, feed rate (A), depth of cut (B), and tool angle
(C), are of interest. All three factors have been assigned two levels,
Yijkl =  + i +  j +  k + ()ij + ( )ik + ( ) jk
A SSA a-1 MSA
bcn  i2 MS A
2 +
B SSB b-1 MSB
a −1 MS E and two replicates of a factorial design are run. The coded data are
acn   2j MS B
2 + shown in Table 14-10.
 i = 1, 2, , a b −1 MS E
C SSC c-1 MSC abn   k2
 j = 1, 2, , b 2 + MSC
c −1 Table 14-10 Coded Surface Roughness Data for Example 14-2
 MS E
+ ( )ijk + ijkl 

(14-10) AB SSAB (a - 1)(b - 1) MSAB
2 +
cn  ()ij2 MS AB Depth of Cut (B)
k = 1, 2, , c
(a − 1) (b − 1) MS E
0.025 inch 0.040 inch
SSAC
bn  ( )ik
2 MS AC Feed Rate (A) Tool Angle (C) Tool Angle(C)
 l = 1, 2, , n
AC (a - 1)(c - 1) MSAC
 +
2
 (a − 1)(c − 1) MS E
BC SSBC (b - 1)(c - 1) MSBC 15° 25° 15° 25° yi 
an  () 2jk MS BC 9 11 9 10
2 +
SSABC (b − 1)(c − 1) MS E 20 inches per
ABC (a - 1)(b - 1)(c - 1) MSABC 7 10 11 8 75
minute
n   ( )ijk
2
MS ABC
Error SSE abc(n - 1) MSE 2 + 10 10 12 16
(a − 1) (b − 1) (c − 1) MS E
Total SST abcn - 1 30 inches per
12 13 15 14 102
minute
Sec 14-4 General Factorial Experiments 34 Sec 14-4 General Factorial Experiments 35 Sec 14-4 General Factorial Experiments 36
4
7/8/2020
14-4: General Factorial Experiments 14-4: General Factorial Experiments Design of Experiments – Definitions
Example 14-2
Example 14-2
• Balance study
The ANOVA is summarized in Table 14-11. Since manual ANOVA computations are tedious for three-factor
experiments, we have used Minitab for the solution of this problem.
The F-ratios for all three main effects and the interactions are formed by
Table 14-11 Minitab ANOVA for Example 14-2
dividing the mean square for the effect of interest by the error mean
ANOVA (Balanced Designs) square. Since the experimenter has selected a = 0.05, the critical value
for each of these F-ratios is f0.05,1,8 = 5.32. Alternately, we could use the
Factor Type Levels Values
Feed fixed 2 20 30 P-value approach. The P-values for all the test statistics are shown in the Xg
Depth fixed 2 0.025 0.040 last column of Table 14-11. Inspection of these Most likely, both feed rate
Angle fixed 2 15 25
and depth of cut are important process variables.
Analysis of Variance for Roughness Practical Interpretation: Further experiments might study the important
Source DF SS MS F P
factors in more detail to improve the surface roughness.
Feed 1 45.563 45.563 18.69 0.003 P-values is revealing. There is a strong main effect of feed rate, since the
Depth 1 10.563 10.563 4.33 0 071 F-ratio is well into the critical region. However, there is some indication of
Angle 1 3.063 3.063 1.26 0.295
Feed*Depth 1 7.563 7.563 3.10 0.116 an effect due to the depth of cut, since P = 0.0710 is not much greater Xg
Feed* Angle 1 0.062 0.062 0.03 0.877 than a = 0.05. The next largest effect is the AB or feed rate  depth of cut
Depth* Angle 1 1.563 1.563 0.64 0.446
Feed* Depth* Angle 1 5.062 5.062 2.08 0.188 interaction.
Error 8 19.500 2.437
Total 15 92.938
Sec 14-4 General Factorial Experiments 37 Sec 14-4 General Factorial Experiments 38
Design of Experiments - Definitions Design of Experiments Design of Experiments

• Some notions • Strategy 1 • Strategy 1
– Y1, Yi, I, ei – Experiment Result – E(M1)=E(Y2)-E(Y1)=E(2+e2)-E(1+e1)
• Y1 = 1+e1 • No item Y1 = E(2)+E(e2)-E(1)-E(e1)
• Yi = i+ei
• Object 1 Y2 = 2- 1
• Yi+1 = i+1+ei+1
• Object 2 Y3
• Yi+1-Yi= (i+1+ei+1) - (i+ei) [random item] – var(M1)=var(Y2-Y1)
• Object 3 Y4
• Esperance E(ei), Variance var(ei)
• M0= Y1 = var(2+e2) + var(1+e1)
• M1=Y2-Y1 = 22
– E(X)= E(X)
• M2=Y3-Y1
– E(X±Y)=E(X)±E(Y)
• M3=Y4-Y1
– E(X.Y)=E(X).E(Y) (if X and Y are independent!)
= E(X).E(Y)+cov(X,Y)
Design of Experiments Design of Experiments Design of Experiments

• Strategy 2 • Strategy 2 •Strategy 3
– Experiment Result –Experiment Result
• No item Y1 M0= Y1 E (M 1 ) =
1
E (Y2 ) + E (Y3 ) − E (Y1 ) − E (Y4 ) •Object 1,2,3 left Y1 M0-M1-M2-M3 = Y1
2
• Object 1+2 Y2 M0+M1+M2=Y2 •Object 2+3 vs 1 Y2 M0+M1-M2-M3=Y2
1
• Object 1+3 Y3 M0+M1+M3=Y3 = ( 2 +  3 − 1 −  4 ) •Object 1+3 vs 2 Y3 M0-M1+M2-M3=Y3
2
• Object 2+3 Y4 M0+M2+M3=Y4 •Object 1+2 vs 3 Y4 M0-M1-M2+M3=Y4
1
var(M 1 ) = var[ (−Y1 + Y2 + Y3 − Y4 )] (Y2 − Y1 )
M1 =
(− Y1 + Y2 + Y3 − Y4 ) 2 M1 =
2 = 2 2
M2 =
(− Y1 + Y2 − Y3 + Y4 )
M2 =
(Y3 − Y1 )
2 2
M3 =
(− Y1 − Y2 + Y3 + Y4 ) (Y − Y )
2 M3 = 4 1
Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. 2 Copyright © 2014 John Wiley & Sons, Inc. All rights reserved.
5
7/8/2020
Design of Experiments Design of Experiments Design of Experiments
• Strategy 3 •Strategy 4 • Strategy 4

–Experiment Result
•Object 1,2,3 right
1 +  2 − 3 −  4 
Y1 M0+M1+M2+M3 = Y1
E ( M 1 ) = E (Y2 ) − E (Y1 )
1 1
•Object 2+3 vs 1 Y2 M0+M1-M2-M3=Y2
E (M 1 ) =
2 4
•Object 1+3 vs 2 Y3 M0-M1+M2-M3=Y3
1 1 
= ( 2 − 1 )
2 •Object 1+2 vs 3 Y4 M0-M1-M2+M3=Y4 var(M 1 ) = var
4
Yi 


Y1 + Y2 + Y3 + Y4
M0 =
1 4 2
var(M 1 ) = var[ (−Y1 + Y2 )] (Y + Y − Y − Y )
M1 = 1 2 3 4 =
2 4 4
2 (Y − Y + Y − Y )
M2 = 1 2 3 4
= 4
2
M3 =
(Y1 − Y2 − Y3 + Y4 )
14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

Design of Experiments
14-5.1 22 Design 14-5.1 22 Design
• Remarks
– Choice of the experiment conditions The main effect of a factor A is estimated by
– Thinking before doing DoEs
– Experiment matrix a + ab b + (1) 1
Exp m1 m2 m3 Exp m1 m2 m3 A = y A+ − y A− = − = [a + ab − b − (1)] (14-11)
1 0 0 0 1 0 0 0 2n 2n 2n
D1 = 2 1 0 0 D2 = 2 1 1 0
4 2 3 0 1 0 3 1 0 1
var(M i )  (i = 1,2,...kobject ; N = k + 1)
N 4 0 0 1 4 0 1 1
Exp m1 m2 m3 Exp m1 m2 m3
 2
var(M i )  (i = 1,2,...kobject ; N = k + 1) 1 −1 −1 −1 1 1 1 1
N
D3 = 2 1 −1 −1 D4 = 2 1 −1 −1
3 −1 1 −1 3 −1 1 −1 (1), a, b, and ab also represent the totals of all n observations
4 −1 −1 1 4 −1 −1 1 Sec 14-5 2k Factorial Designs 50 Sec 14-5 2k Factorial Designs 51
14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5.1 22 Design 14-5.1 22 Design 14-5.1 22 Design
The quantities in brackets in Equations 14-11, 14-12, and 14-13

The main effect of a factor B is estimated by The AB interaction effect is estimated by are called contrasts.
For example, the A contrast is ContrastA = a + ab – b – (1)
b + ab a + (1) 1 ab + (1) a + b 1
B = yB + − yB − = − = [b + ab − a − (1)] (14-12) AB = − = [ab + (1) − a − b] (14-13) Table 14-12 Signs for Effects in the 22 Design
2n 2n 2n 2n 2n 2n
Factorial Effect
Treatment
I A B AB
Combination
(1) + − − +
a + + − −
b + − + −
ab + + + +
Sec 14-5 2k Factorial Designs 52 Sec 14-5 2k Factorial Designs 53 Sec 14-5 2k Factorial Designs 54
6
7/8/2020

14-5.1 22 Design 14-5.1 22 Design Example 14-3 Epitaxial Process An article in the AT&T Technical Journal
(March/April 1986, Vol. 65, pp. 39–50) describes the application of two-level
factorial designs to integrated circuit manufacturing. A basic processing
Contrasts are used in calculating both the effect estimates step in this industry is to grow an epitaxial layer on polished silicon wafers.
and the sums of squares for A, B, and the AB interaction. The wafers are mounted on a susceptor and positioned inside a bell jar.
The sums of squares formulas are Chemical vapors are introduced through nozzles near the top of the jar. The
susceptor is rotated, and heat is applied. These conditions are maintained
[a + ab − b − (1)]2 until the epitaxial layer is thick enough.
SS A =
4n Table 14-13 The 22 Design for the Epitaxial Process Experiment
[b + ab − a − (1)]2
SS B = Design Factors Thickness (m)
4n Treatment
[ab + (1) − a − b]2 Combination A B AB Thickness (m) Total Average
SS AB = (1) − − + 14.037 14.165 13.972 13.907 56.081 14.020
4n a + − − 14.821 14.757 14.843 14.878 59.299 14.825
b − + − 13.880 13.860 14.032 13.914 55.686 13.922
ab + + + 14.888 14.921 14.415 14.932 59.156 14.789

Table 14-13 presents the results of a 22 factorial design with n = 4 replicates using the The numerical estimates of the effects indicate that the effect of deposition time is
factors A = deposition time and B = arsenic flow rate. The two levels of deposition time large and has a positive direction (increasing deposition time increases thickness),
are − =short and + =long, and the two levels of arsenic flow rate are - =55% and + = 59%. since changing deposition time from low to high changes the mean epitaxial layer
The response variable is epitaxial layer thickness (mm). We may find the estimates of the
thickness by 0.836 mm. The effects of arsenic flow rate (B) and the AB interaction
effects using Equations 14-11, 14-12, and 14-13 as follows:
appear small. The importance of these effects may be confirmed with the analysis
1
A= [a + ab − b − (1)] of variance. The sums of squares for A, B, and AB are computed as follows:
2n
1 [a + ab − b − (1)]2 [6.688]2
= [59.299 + 59.156 − 55.686 − 56.081] = 0.836
2(4) SS A = = = 2.7956
16 16
1
B= [b + ab − a − (1)] [b + ab − a − (1)]2 [−0.538]2
2n SS B = = = 0.0181
1 16 16
= [55.686 + 59.156 − 59.299 − 56.081]
2(4) [ab + (1) − a − b]2 [0.252]2
= −0.067 SS AB = = = 0.0040
16 16
1
AB = [ab + (1) − a − b] (56.081 +  + 59.156) 2
2n SST = 14.037 2 +  + 14.9322 −
1 16
AB = [59.156 + 56.081 − 59.299 − 55.686]
2(4) = 3.0672
= 0.032

Residual Analysis Residual Analysis Residual Analysis
Figure 14-16 Normal

probability plot of residuals
for the epitaxial process Figure 14-17 Plot of Figure 14-18 Plot of
experiment. residuals versus residuals versus
deposition time. arsenic flow rate.
7
7/8/2020

Residual Analysis
14-5.2 2k Design for k  3 Factors 14-5.2 2k Design for k  3 Factors
Figure 14-19 The standard deviation of epitaxial layer thickness at the

four runs in the 22 design. Figure 14-20 The 23 design.

14-5.2 2k Design for k  3 Factors 14-5.2 2k Design for k  3 Factors 14-5.2 2k Design for k  3 Factors
The main effect of A is estimated by The main effect of C is estimated by
A = y A+ − y A− C = yC + − yC − Other two-factor interactions effects estimated by
1 1 1
= [a + ab + ac + abc − (1) − b − c − bc] = [c + ac + bc + abc − (1) − a − b − ab] AC = [(1) − a + b − ab − c + ac − bc + abc]
4n
4n 4n
1
BC = [(1) − a − b − ab − c − ac + bc + abc]
The main effect of B is estimated by The interaction effect of AB is estimated by 4n
The three-factor interaction effect, ABC, is estimated
B = yB + − yB − by
1
1
= [b + ab + bc + abc − (1) − a − c − ac]
AB = [abc − bc + ab − b − ac + c − a + (1)] 1
4n ABC = [abc − bc − ac + c − ab + b + a − (1)]
4n 4n

14-5.2 2k Design for k  3 Factors 14-5.2 2k Design for k  3 Factors 14-5.2 2k Design for k  3 Factors
Table 14-15 Algebraic Signs for Calculating Effects in the 2 3 Design Table 14-15 has several interesting properties:
Contrasts can be used to calculate several quantities:
Treatment Factorial Effect 1. Except for the identity column I, each column has an equal number of
Combination I A B AB C AC BC ABC plus and minus signs.
Contrast
(1) + − − + − + + − Effect = (14-14)
a + + − − − − + +
2. The sum of products of signs in any two columns is zero; that is, n2 k −1
the columns in the table are orthogonal.
b + − + − − + − +
ab + + + + − − − − 3. Multiplying any column by column I leaves the column
(Contrast)2
c + − − + + − − + unchanged; that is, I is an identity element. SS = (14-15)
ac + + − − + + − − n2k
4. The product of any two columns yields a column in the table, for
bc + − + − + − + −
example A  B = AB, and AB  ABC = A2B2C = C, since any
abc + + + + + + + + column multiplied by itself is the identity column.
8
7/8/2020

Example 14-4 Surface Roughness Consider the surface roughness Example 14-4 Example 14-4
experiment originally described in Example 14-2. This is a 23 factorial It is easy to verify that the other effects are
design in the factors feed rate (A), depth of cut (B), and tool angle (C), with The main effects may be estimated using Equations 14-11 through
n = 2 replicates. Table 14-16 presents the observed surface roughness 14-13. The effect of A, for example, is
B = 1.625
data. 1 C = 0.875
A= [a + ab + ac + abc − (1) − b − c − bc]
Table 14-16 Surface Roughness Data for Example 14-4 4n AB = 1.375
1 AC = 0.125
Design Factors = [22 + 27 + 23 + 40 − 16 − 20 − 21 − 18] BC = -0.625
Treatment 4(2)
A B C Surface Roughness Totals ABC = 1.125
Combinations
1
(1) -1 -1 -1 9, 7 16 = [27] = 3.375
a 1 -1 -1 10, 12 22 8 Examining the magnitude of the effects clearly shows that feed
b -1 1 -1 9 , 11 20
ab 12, 15 27
and the sum of squares for A is found using Equation 14-15: rate (factor A) is dominant, followed by depth of cut (B) and the
1 1 -1
c -1 -1 1 11, 10 21 AB interaction, although the interaction effect is relatively small.
ac 1 -1 1 10, 13 23 (Contrast A ) 2 (27) 2 The analysis of variance, summarized in Table 14-17, confirms
bc -1 1 1 10, 8 18 SS A = = = 45.5625 our interpretation of the effect estimates.
abc 1 1 1 16, 14 30 n2k 2(8)

Example 14-4 14-5.3 Single Replicate of the 2k Design
Residual Analysis
Example 14-5 Plasma Etch An article in Solid State Technology [“Orthogonal Design for
Process Optimization and Its Application in Plasma Etching” (May 1987, pp. 127–132)]
Table 14-17 Analysis of Variance for the Surface Roughness Experiment
describes the application of factorial designs in developing a nitride etch process on a
single-wafer plasma etcher. The process uses C2F6 as the reactant gas. It is possible to
Source of Degrees of
Variation Sum of Squares Freedom Mean Square f0 P-Value vary the gas flow, the power applied to the cathode, the pressure in the reactor chamber,
A 45.5625 1 45.5625 18.69 0.0025 and the spacing between the anode and the cathode (gap). Several response variables
Figure 14-22 Normal would usually be of interest in this process, but in this example we will concentrate on
B 10.5625 1 10.5625 4.33 0.0709 probability plot of residuals etch rate for silicon nitride.
C 3.0625 1 3.0625 1.26 0.2948 from the surface roughness
AB 7.5625 1 7.5625 3.10 0.1162 experiment. We will use a single replicate of a 2 4 design to investigate this process. Since it is
AC 0.0625 1 0.0625 0.03 0.8784 unlikely that the three- and four-factor interactions are significant, we will
BC 1.5625 1 1.5625 0.64 0.4548 tentatively plan to combine them as an estimate of error. The factor levels used in the
ABC 5.0625 1 5.0625 2.08 0.1875 design are shown below:
Design Factor
Error 19.5000 8 2.4375 Level Gap (cm) Pressure (mTorr) C2F6 Flow (SCCM) Power (w)
Total 92.9375 15 Low (−) 0.80 450 125 275
Hi g h ( + ) 1.20 550 200 325

14-5.3 Single Replicate of the 2k Design 14-5.3 Single Replicate of the 2k Design 14-5.3 Single Replicate of the 2k Design
Example 14-5 Example 14-5
Table 14-20 Contrast Constants for the 24 Design Table 14-19 presents the data from the 16 runs of the 2 4 design.
Table 14-20 is the table of plus and minus signs for the 2 4 design.
The signs in the columns of this table can be used to estimate the
factor effects. For example, the estimate of factor A is
1
A = [a + ab + ac + abc + ad + abd + acd
8
+ abcd − (1) − b − c − bc − d − bd − cd − bcd ]
1
= [669 + 650 + 642 + 635 + 749 + 868 + 860
8
+ 729 − 550 − 604 − 633 − 601 − 1037
− 1052 − 1075 − 1063]
= −101.625
9
7/8/2020

It is easy to verify (using R, for example) that the complete set of effect
estimates is
A = -101.625 AD = -153.625
B = -1.625 BD = -0.625 Figure 14-23 Normal
probability plot of effects
AB = -7.875 ABD = 4.125 from the plasma etch
experiment.
C = 7.375 CD = -2.125
AC = -24.875 ACD = 5.625
BC = -43.875 BCD = -25.375
ABC = -15.625 ABCD = -40.125
D = 306.125 Notice that in the analysis of variance we have pooled the three- and four-factor
interactions to form the error mean square.

Practical Interpretation: Since A = −101.625, the effect of increasing the

ANOVA table
gap between the cathode and anode is to decrease the etch rate.
However, D = 306.125; thus, applying higher power levels will increase
the etch rate. Figure 14-24 is a plot of the AD interaction. This plot
indicates that the effect of changing the gap width at low power settings is
small, but that increasing the gap at high power settings dramatically
reduces the etch rate. High etch rates are obtained at high power settings
and narrow gap widths.
The residuals from the experiment in Example 14-5 can be obtained from
the regression model
 101.625   306.125   153.625 

yˆ = 776.0625 −   x1 +   x4 −   x1x4
Figure 14-24 AD (gap-power) interaction from the plasma etch experiment.  2   2   2 

14-5.3 Single Replicate of the 2k Design 14-5.3 Single Replicate of the 2k Design 14-5.4 Additional Center Points to a 2k Design
Example 14-5 Example 14-5 A potential concern in the use of two-level factorial designs is the
assumption of the linearity in the factor effect. Adding center points to the 2k
The residuals from the experiment in Example 14-5 can be obtained from design will provide protection against curvature as well as allow an
the regression model independent estimate of error to be obtained. Figure 14-26 illustrates the
 101.625   306.125   153.625  situation.
yˆ = 776.0625 −   x1 +   x4 −   x1x4 Figure 14-25 Normal
 2   2   2  probability plot of
For A (-1) and D(-1) the predict value is: residuals from the plasma
etch experiment.
Four residuals at this treatment combination are:
10
7/8/2020

14-5.4 Additional Center Points to a 2k Design 14-5.4 Additional Center Points to a 2k Design 14-5.4 Additional Center Points to a 2k Design
Example Process Yield A chemical engineer is studying the percentage Example
A single-degree-of-freedom sum of squares for of conversion or yield of a process. There are two variables of interest,
reaction time and reaction temperature. Because she is uncertain about
curvature is given by: the assumption of linearity over the region of exploration, the engineer The mean square error is calculated from the center points as follows:
decides to conduct a 22 design (with a single replicate of each factorial
run) augmented with five center points. The design and the yield data are  ( yi − yC ) 2
shown in Fig. SS E Center points
MS E = =
nC − 1 nC − 1
5
Figure The 22 design with five  ( yi − 40.46) 2 0.1720
i =1
center points for the process = = = 0.0430
yield experiment in Example. 4 4
14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-6: Blocking and Confounding in the 2k Design
14-5.4 Additional Center Points to a 2k Design 14-5.4 Additional Center Points to a 2k Design
Example 14-6 Example
The average of the points in the factorial portion of the design is yF = 40.425 , and General method of constructing blocks
the average of the points at the center is yC = 40.46 . The difference Table Analysis of Variance
yF − yC = 40.425 − 40.46 = −0.035 appears to be small. The curvature sum of squares Sum of Degrees of Mean
employs a defining contrast
in the analysis of variance table is computed from Equation 14-16 as follows: Source of Variation Squares Freedom Square f0 P-Value
nF nC ( yF − yC )2
A (Time) 2.4025 1 2.4025 55.87 0.0017
SSCurvature = B (Temperature) 0.4225 1 0.4225 9.83 0.0350
nF + nC L = a1x1 + a2x2 +  + akxk
AB 0.0025 1 0.0025 0.06 0.8237
=
(4)(5)(− 0.035)2 = 0.0027 Curvature 0.0027 1 0.0027 0.06 0.8163
4+5 Error 0.1720 4 0.0430
Practical Interpretation: The analysis of variance indicates that both factors exhibit Total 3.0022 8
significant main effects, that there is no interaction, and that there is no evidence of
curvature in the response over the region of exploration. That is, the null
hypothesis H 0 :  j =1  jj = 0 cannot be rejected.
k
Sec 14-5 2k Factorial Designs 98 Sec 14-5 2k Factorial Designs 99 Sec 14-6 Blocking and Confounding in the 2k Design 100
14-6: Blocking and Confounding in the 2k Design 14-6: Blocking and Confounding in the 2k Design 14-6: Blocking and Confounding in the 2k Design
Sec 14-6 Blocking and Confounding in the 2k Design 101 Sec Figure
14-6 Blocking 23 design
The and Confounding in the
in two 2k Design
blocks 102 Sec 14-6 Blocking and Confounding in the 2k Design 103
11
7/8/2020
14-6: Blocking and Confounding in the 2k Design 14-6: Blocking and Confounding in the 2k Design 14-6: Blocking and Confounding in the 2k Design
Example 14-7 Missile Miss Distance An experiment is Example 14-7

performed to investigate the effect of four factors on the The experimental design and the resulting data are shown in
terminal miss distance of a shoulder-fired ground-to-air missile. Fig. 14-30. The effect estimates obtained from R are shown in
The four factors are target type (A), seeker type (B), target Table 14-23. A normal probability plot of the effects in Fig. 14-31
altitude (C), and target range (D). Each factor may be reveals that A (target type), D (target range), AD, and AC have
conveniently run at two levels, and the optical tracking system large effects. A confirming analysis of variance, pooling the
will allow terminal miss distance to be measured to the three-factor interactions as error, is shown in Table 14-24.
nearest foot. Two different operators or gunners are used in
the flight test and, since there may be differences between Practical Interpretation: Since the AC and AD interactions are
operators, the test engineers decided to conduct the 2 4 design significant, it is logical to conclude that A (target type), C (target
in two blocks with ABCD confounded. Thus, the defining altitude), and D (target range) all have important effects on the
contrast is miss distance and that there are interactions between target
type and altitude and target type and range. Notice that the
Figure 14-30 The 24 design in two blocks for Example 14-7. (a) Geometric L = X1 + X2 + X3 + X4 ABCD effect is treated as blocks in this analysis.
view. (b) Assignment of the 16 runs to two blocks.
Sec 14-6 Blocking and Confounding in the 2k Design 104 Sec 14-6 Blocking and Confounding in the 2k Design 105 Sec 14-6 Blocking and Confounding in the 2k Design 106
14-6: Blocking and Confounding in the 2k Design 14-6: Blocking and Confounding in the 2k Design 14-7: Fractional Replication of the 2k Design
Example 14-7 Example 14-7
Table 14-23 Minitab Effect Estimates for Example 14 -7 Table 14-24 Analysis of Variance for Example 14-7
14-7.1 One-Half Fraction of the 2k Design
Estimated Effects and Coefficients for Distance
Term Effect Coefficient Sum of Degrees of Mean Table 14-25 Plus and Minus Signs for the 23 Factorial Design
Constant 6.938 Source of Variation Squares Freedom Square f0 P-Value
Block 0.063 Blocks (ABCD) 0.0625 1 0.0625 0.06 —
A 2.625 1.312 Treatment Factorial Effect
A 27.5625 1 27.5625 25.94 0.0070
B 0.625 0.313 B 1.5625 1 1.5625 1.47 0.2920 Combination
I A B C AB AC BC ABC
C 0.875 0.438 C 3.0625 1 3.0625 2.88 0.1648 a + + − − − − + +
D 1.875 0.938 D 14.0625 1 14.0625 13.24 0.0220
AB -0.125 -0.063 AB 0.0625 1 0.0625 0.06 — b + − + − − + − +
AC -2.375 -1.187 AC 22.5625 1 22.5625 21.24 0.0100 c + − − + + − − +
AD 1.625 0.813 AD 10.5625 1 10.5625 9.94 0.0344 abc + + + + + + + +
BC -0.375 -0.188 BC 0.5625 1 0.5625 0.53 — ab + + + − + − − −
BD -0.375 -0.187 —
BD 0.5625 1 0.5625 0.53 ac + + − + − + − −
CD -0.125 -0.062
Figure 14-31 Normal probability plot of the CD 0.0625 1 0.0625 0.06 — bc + − + + − − + −
ABC -0.125 -0.063
effects for the Missile Miss Distance Error (ABC + ABD + ACD + BCD) 4.2500 4 1.0625
ABD 0.875 0.438 (1) + − − − + + + −
Experiment.
ACD -0.375 -0.187 Total 84.9375 15
BCD -0.375 -0.187
Sec 14-6 Blocking and Confounding in the 2k Design 107 Sec 14-6 Blocking and Confounding in the 2k Design 108 Sec 14-7 Fractional Replication of the 2k Design 109
14-7: Fractional Replication of the 2k Design 14-7: Fractional Replication of the 2k Design 14-7: Fractional Replication of the 2k Design
14-7.1 One-Half Fraction of the 2k Design 14-7.1 One-Half Fraction of the 2k Design 14-7.1 One-Half Fraction of the 2k Design
Main effects and two-factor interaction effects Linear combination of observations
aliases
Figure 14-32 The one-half fractions of the 23 design. (a) The principal fraction,
I = +ABC. (b) The alternate fraction, I = -ABC.
Sec 14-7 Fractional Replication of the 2k Design 110 Sec 14-7 Fractional Replication of the 2k Design 111 Sec 14-7 Fractional Replication of the 2k Design 112
12
7/8/2020
Example 14-8 Plasma Etch To illustrate the use of a one-half fraction, consider the
14-7.1 One-Half Fraction of the 2k Design plasma etch experiment described in Example 14-5. Suppose that we decide to use a Example 14-8
24-1 design with I = ABCD to investigate the four factors gap (A), pressure (B), C2F6
flow rate (C), and power setting (D). This design would be constructed by writing down
as the basic design a 2 3 in the factors A, B, and C and then setting the levels of the
fourth factor D = ABC. The design and the resulting etch rates are shown in Table 14-
26. The design is shown graphically in Fig. 14-33.
Table 14-26 The 24-1 Design with Defining Relation I = ABCD
Treatment
A B C D = ABC Combination Etch Rate
− − − − (1) 550
+ − − + ad 749
− + − + bd 1052
+ + − − ab 650
− − + + cd 1075
+ − + − ac 642
− + + − bc 601 Figure 14-33 The 24-1 design for the plasma etch experiment of
+ + + + abcd 729 Example 14-8.
In this design, the main effects are aliased with the three-factor interactions; note that the alias of A Clearly, A and  D are large, and if we believe that the three-factor interactions are
is
negligible, the main effects A (gap) and D (power setting) significantly affect etch rate.
A  I = A  ABCD or A = A2BCD = BCD Table 14-27 Effect Estimates from Minitab, Example 14-8
and similarly B = ACD, C = ABD, and D = ABC. The interactions are estimated by forming the AB, AC, and AD columns and adding them to
the table. For example, the signs in the AB column are +, −, −, +, +, −, −, +, and this Fractional Factorial Fit
The two-factor interactions arc aliased with each other. For example, the alias of AB is CD :
column produces the estimate Estimated Effects and Coefficients for Etch Rate
AB  I = AB  ABCD or AB = A2B2CD = CD  AB = AB + CD = 14 (550 − 749 − 1052 + 650 + 1075 Term Effect Coef
− 642 − 601 + 729) = −10 Constant 756.00
The other aliases are AC = BD and AD = BC. The estimates of the main effects and their aliases
are found using the four columns of signs in Table 14-26. For example, from column A we obtain Gap −127.00 −63.50
the estimated effect From the AC and AD columns we find Pressure 4.00 2.00
 A = A + BCD = 1 (−550 + 749 − 1052 + 650 − 1075
4  AC = AC + BD = −25.50 F 11.50 5.75
+ 642 − 601 + 729)
Power 290.50 145.25
= −127.00 and Gap*Pressure −10.00 −5.00
The other columns produce  AD = AD + BC = −197.50
 B = B + ACD = 4.00  C = C + ABD = 11.50 Gap*F −25.50 −12.75
The  AD estimate is large; the most straightforward interpretation of the results is that since Gap*Power −197.50 −98.75
and
 D = D + ABC = 290.50 A and D are large, this is the AD interaction. Thus, the results obtained from the 24−1
design agree with the full factorial results in Example 14-5.
Example 14-8 Projection of the 2k-1 Design Projection of the 2k-1 Design
Figure 14-34 Normal probability plot of the effects for the plasma etch
Figure 14-36 The 22 design obtained by dropping factors B and C from the plasma
experiment in Example 14-8.
etch experiment in Example 14-8.
Figure 14-35 Projection of a 23-1 design into three 22 designs.
13
7/8/2020
Design Resolution Example 14-9 Injection Molding Parts manufactured in an injection-
14-7.2 Smaller Fractions: The 2k-p Fractional molding process are showing excessive shrinkage, which is causing
1. Resolution III Designs. These are designs in which no main effects
are aliased with any other main effect, but main effects are aliased Factorial problems in assembly operations upstream from the injection-molding area.
with two-factor interactions and some two-factor interactions may be In an effort to reduce the shrinkage, a quality-improvement team has
aliased with each other. The 2 3−1 design with I = ABC is a resolution Table 14-28 Alias Structure for the 2 6IV− 2 Design with I = ABCE = BCDF = ADEF decided to use a designed experiment to study the injection-molding
III design. We usually employ a Roman numeral subscript to indicate process. The team investigates six factors—mold temperature (A), screw
A = BCE = DEF = ABCDF AB = CE = ACDF = BDEF
design resolution; thus, this one-half fraction is a2 3III−1 design. speed (B), holding time (C), cycle time (D), gate size (E), and holding
B = ACE = CDF = ABDEF AC = BE = ABDF = CDEF
pressure (F)—each at two levels, with the objective of learning how each
2. Resolution IV Designs. These are designs in which no main effect is C = ABE = BDF = ACDEF AD = EF = BCDE = ABCF
aliased with any other main effect or two-factor interactions, but two-factor factor affects shrinkage and obtaining preliminary information about how the
D = BCF = AEF = ABCDE AE = BC = DF = ABCDEF factors interact.
interactions are aliased with each other. The 2 4−1 design with I = ABCD used
in Example 14-8 is a resolution IV design (2 4IV−1 ) . E = ABC = ADF = BCDEF AF = DE = BCEF = ABCD
F = BCD = ADE = ABCEF BD = CF = ACDE = ABEF The team decides to use a 16-run two-level fractional factorial design for
3. Resolution V Designs. These are designs in which no main effect or ABD = CDE = ACF = BEF BF = CD = ACEF = ABDE these six factors. The design is constructed by writing down a 24 as the
two-factor interaction is aliased with any other main effect or two-factor basic design in the factors A, B, C and D and then setting E = ABC and
ACD = BDE = ABF = CEF
interaction, but two-factor interactions are aliased with three-factor
5−1
F = BCD as discussed above. Table 14-29 shows the design, along with the
interactions. The 25−1 design with I = ABCDE is a resolution V design(2 V ) . observed shrinkage (10) for the test part produced at each of the 16 runs in
the design.
Example 14-9 6−2
Table 14-29 A 2IV Design for the Injection-Molding Experiment Example 14-9 Example 14-9
Observed A normal probability plot of the effect estimates from this experiment is shown in
Shinkage Fig. 14-37. The only large effects are A (mold temperature), B (screw speed),
Run A B C D E = ABC F = BCD (10) and the AB interaction. In light of the alias relationship in Table 14-28, it seems
1 − − − − − − 6 reasonable to tentatively adopt these conclusions. The plot of the AB interaction
2 + − − − + − 10
in Fig. 14-38 shows that the process is insensitive to temperature if the screw
3 − + − − + + 32
4 + + − − − + 60 speed is at the low level but sensitive to temperature if the screw speed is at the Figure 14-37 Normal probability
5 − − + − + + 4 high level. With the screw speed at a low level, the process should produce an plot of effects for the injection-
6 + − + − - + 15 average shrinkage of around 10% regardless of the temperature level chosen. molding experiment in
7 − + + − − − 26 Example 14-9.
8 + + + − + − 60 Based on this initial analysis, the team decides to set both the mold temperature
9 − − − + - + 8 and the screw speed at the low level. This set of conditions should reduce the
10 + − − + + + 12 mean shrinkage of parts to around 10%. However, the variability in shrinkage
11 − + − + + − 34
from part to part is still a potential problem. In effect, the mean shrinkage can be
12 + + − + − − 60
13 − − + + + − 16 adequately reduced by the above modifications; however, the part-to-part
14 + − + + − − 5 variability in shrinkage over a production run could still cause problems in
15 − + + + − + 37 assembly. One way to address this issue is to see if any of the process factors
16 + + + + + + 52 affect the variability in parts shrinkage.
Example 14-9 Figure 14-39 presents the normal probability plot of the residuals. This plot appears Example 14-9
satisfactory. The plots of residuals versus each factor were then constructed. One of
these plots, that for residuals versus factor C (holding time), is shown in Fig. 14-40. The
plot reveals much less scatter in the residuals at the low holding time than at the high
holding time. These residuals were obtained in the usual way from a model for predicted
shrinkage
yˆ = ˆ 0 + ˆ 1x1 + ˆ 2 x2 + ˆ 12 x1x2
Figure 14-38 Plot of AB (mold
= 27.3125 + 6.9375x1 + 17.8125x2 + 5.9375x1x2 Figure 14-39 Normal probability plot
temperature-screw speed)
interaction for the injection- of residuals for the injection-molding
where x1, x2, and x1x2 are coded variables that correspond to the factors A and B and the experiment in Example 14-9.
molding experiment in
Example 14-9. AB interaction. The regression model used to produce the residuals essentially removes
the location effects of A, B, and AB from the data; the residuals therefore contain
information about unexplained variability.
Figure 14-40 indicates that there is a pattern in the variability and that the variability in the
shrinkage of parts may be smaller when the holding time is at the low level.
Practical Interpretation: Figure 14-41 shows the data from this experiment projected onto
a cube in the factors A, B, and C. The average observed shrinkage and the range of
observed shrinkage are shown at each corner of the cube.
14
7/8/2020
14-7: Fractional Replication of the 2k Design 14-7: Fractional Replication of the 2k Design 14-8: Response Surface Methods and Designs
Response surface methodology:
Example 14-9 Example 14-9 (RSM), is a collection of
mathematical and statistical
techniques that are useful for
modeling and analysis in
applications where a response of
interest is influenced by several
variables and the objective is to
Figure 14-40 Residuals versus optimize this response.
holding time (C) for the injection-
molding experiment in Example 14-9.
Figure 14-41 Average shrinkage and range of shrinkage in factors A, B, and

C for Example 14-9.
Sec 14-7 Fractional Replication of the 2k Design 131 Sec 14-7 Fractional Replication of the 2k Design 132 Sec 14-8 Response Surface Methods and Designs 133
14-8: Response Surface Methods and Designs 14-8: Response Surface Methods and Designs 14-8: Response Surface Methods and Designs
Method of Steepest Ascent
The first-order model
Y = 0 + 1x1 + 2x2 +  + kxk + Method of Steepest Ascent
Method of Steepest Ascent
The method of steepest ascent is a procedure for moving sequentially along the path of
k k steepest ascent, that is, in the direction of the maximum increase in the response. Of course,
The second-order model Y = 0 +  i xi +  ii xi2 +  ij xi x j +  if minimization is desired, we are talking about the method of steepest descent. The fitted
first-order model is k
i =1 i =1 i j ˆ +
yˆ =  0  ˆ i xi
i =1 (14-17)
and the first-order response surface, that is, the contours of , is a series of parallel lines such
as that shown in Fig. 14-14. The direction of steepest ascent is the direction in which
Figure 14-44 First-order
increases most rapidly. This direction is normal to the fitted response surface contours. We
usually take as the path of steepest descent the line through the center of the region of response surface and path of
interest and normal to the fitted surface contours. Thus, the steps along the path are steepest ascent.
proportional to the regression coefficients {ˆ i } . The experimenter determines the actual step
size based on process knowledge or other practical considerations.
Sec 14-8 Response Surface Methods and Designs 134 Sec 14-8 Response Surface Methods and Designs 135 Sec 14-8 Response Surface Methods and Designs 136
Example: Process Yield Steepest Ascent In the previous example (14-6) we described an
experiment on a chemical process in which two factors, reaction time (x1) and reaction
temperature (x2), affect the percent conversion or yield (Y). Figure shows the 22 design plus
five center points used in this study. The engineer found that both factors were important,
there was no interaction, and there was no curvature in the response surface. Therefore, the
first-order model
Y = 0 + 1x1 + 2x2 + 
should be appropriate. Now the effect estimate of time is 1.55 hours and the effect estimate
of temperature is 0.65°F, and since the regression coefficients ̂1 and ̂2 are one-half of the
corresponding effect estimates, the fitted first-order model is
yˆ = 40.44 + 0.775x1 + 0.325x2

Figure 14-45(a) and (b) show the contour plot and three-dimensional surface plot of this
model. Figure 14-45 also shows the relationship between the coded variables x1 and x2 (that
defined the high and low levels of the factors) and the original variables, time (in minutes) and
temperature (in °F).
15
7/8/2020
Example: How to improve the Yield ? Example Steepest ascent experiment
yˆ = 40.44 + 0.775x1 + 0.325x2
➔ We would move 0.775 units in the x1 direction for every 0.325 units in the x2.
➔ Steepest ascent slope: 0.325/0.775
0.325
➔ Basic step: 5min of reaction time ➔ ∆𝑥1 = 1; ∆𝑥2 = = 0.42
0.775
Response surface plots for the first-order model
Example Steepest ascent experiment Example Steepest ascent experiment Steepest ascent experiment algorithm:
Assume that the point x1 = x2 = · · · = xk = 0 is the base or origin point:
1. Choose a step size in one of the process variables, say Δxj. Usually, we would select the
variable we know the most about, or we would select the variable that has the largest
absolute regression coefficient |𝛽ĵ |.
2. The step size in the other variables is:
3. Convert the Δxi from coded variables to the natural variables.

Interaction & Quadratic term significant !
path of steepest ascent is proportional to the signs and magnitudes of the regression coefficients
in the fitted first-order model
First order model not adequate approximation
Curvature indicate we are near the optimum
Additional analysis to locate optimum !
Lack Of Fit test Lack Of Fit test Central Composite Design (CCD)
• A test that compares the deviation of actual points from the fitted surface,
relative to pure error. If a model has a significant lack of fit, it should be
• A test that compares the deviation of actual points from the fitted surface,
relative to pure error. If a model has a significant lack of fit, it should be
• Rotatable
investigated before being used for prediction. investigated before being used for prediction.
• The lack-of-fit test requires that we have true replicates on the response y • The lack-of-fit test requires that we have true replicates on the response y
for at least one set of levels on the regressors x1, x2,…, xk. These are not just for at least one set of levels on the regressors x1, x2,…, xk. These are not just
duplicate readings or measurements of y. duplicate readings or measurements of y.
Test statistic for LOF:
147 148 149

16
7/8/2020
Second-Order Response Surface: Second-Order Response Surface: Second-Order Response Surface:
Fitted second order model: Fitted second order model:
Stationary point :
b is a (k × 1) vector of the first-order regression coefficients

B is a (k × k) symmetric matrix whose main diagonal elements are the pure
quadratic coefficients (𝛽̂ii) and whose off-diagonal elements are one-half the mixed
quadratic coefficients (𝛽̂ij, i ≠ j).
Characterizing the Response Surface: Characterizing the Response Surface: Second-Order Response Surface: Example
Stationary Points
Canonical analysis:
the surface is steepest in the 𝑤i direction for which |𝜆i| is the greatest.
Second-Order Response Surface: Example

14-8: Response Surface Methods and Designs
Second-Order Response Surface: Example Second-Order Response Surface: Example
Sec 14-8 Response Surface Methods and Designs 156 157 158
17
7/8/2020
Second-Order Response Surface: Example Second-Order Response Surface: Example Second-Order Response Surface: Example
159 160 161

Second-Order Response Surface: Example Your turn:

• Canonical analysis
• Canonical form of the fitted model is:
stationary point is maximum

162 163 164
Important Terms & Concepts of Chapter 14
Analysis of variance (ANOVA) Normal probability plot of factor

Blocking & nuisance factors effects
Center points Optimization experiment
Central composite design Orthogonal design
Confounding Regression model
Contrast Residual analysis
Defining relation Resolution
Design matrix Response surface
Factorial experiment Screening experiment
Fractional factorial design Steepest ascent (or descent)
Generator 2k factorial design
Interaction Two-level factorial design
Main effect
165 Chapter 14 Summary 167
Copyright © 2014 John Wiley & Sons, Inc. All rights reserved. Copyright © 2014 John Wiley & Sons, Inc. All rights reserved.
18

Ch14 - Design of Experiments

Uploaded by

Copyright:

Available Formats

Ch14 - Design of Experiments

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Ch14 - Design of Experiments

Uploaded by

Copyright:

Available Formats

7/8/2020

Design of Experiments Learning Objectives for Chapter 14

Chapter 14 Title and Outline 2 Chapter 14 Learning Objectives 3

• An experiment is a test or series of tests.

Sec 14-1 Introduction 6

14-2: Factorial Experiments 14-2: Factorial Experiments

Table 14-2 A Factorial Experiment with Interaction

Sec 14-2 Factorial Experiments 7 Sec 14-2 Factorial Experiments 8

14-3: Two-Factor Factorial Experiments 14-3: Two-Factor Factorial Experiments

Factor B The observations may be described by the

Figure 14-10 Graph of Application Method

+ ( )ijk + ijkl 

Design of Experiments - Definitions Design of Experiments Design of Experiments

Design of Experiments Design of Experiments Design of Experiments

Design of Experiments Design of Experiments Design of Experiments

• Strategy 3 •Strategy 4 • Strategy 4

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

The quantities in brackets in Equations 14-11, 14-12, and 14-13

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

Figure 14-16 Normal

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

Figure 14-19 The standard deviation of epitaxial layer thickness at the

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

AC = -24.875 ACD = 5.625

BC = -43.875 BCD = -25.375

ABC = -15.625 ABCD = -40.125

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

Practical Interpretation: Since A = −101.625, the effect of increasing the

 101.625   306.125   153.625 

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

Four residuals at this treatment combination are:

14-5: 2k Factorial Designs 14-5: 2k Factorial Designs 14-5: 2k Factorial Designs

Example 14-7 Missile Miss Distance An experiment is Example 14-7

Figure 14-41 Average shrinkage and range of shrinkage in factors A, B, and

yˆ = 40.44 + 0.775x1 + 0.325x2

yˆ = 40.44 + 0.775x1 + 0.325x2

➔ Steepest ascent slope: 0.325/0.775

Response surface plots for the first-order model

Assume that the point x1 = x2 = · · · = xk = 0 is the base or origin point:

3. Convert the Δxi from coded variables to the natural variables.

Test statistic for LOF:

147 148 149

Fitted second order model: Fitted second order model:

b is a (k × 1) vector of the first-order regression coefficients

Second-Order Response Surface: Example

159 160 161

Second-Order Response Surface: Example Your turn:

• Canonical form of the fitted model is:

stationary point is maximum

Important Terms & Concepts of Chapter 14

Analysis of variance (ANOVA) Normal probability plot of factor