EXPERIMENTAL DESIGN

 Experiments are performed to investigate characteristics of a system.

 Responses or characteristics are the results evaluated from
experiments.

 The responses should be analyzed appropriately according to the
objectives of the experiments.

 The factors are the sources that affect the experiments.

 It may not be easy to identify all the factors and give right values
because there may be many factors and they may have distributions
and be affected by the variations of the environment.
3/30/2014 2
 Goal of experiments:

 Experiments help us in understanding the behavior of a
(mechanical) system


 Data collected by systematic variation of influencing factors
helps us to quantitatively describe the underlying phenomenon
or phenomena
3/30/2014 3
Primary Goals

 Determine the variables and their magnitude that influences the
response

 Determine the levels for these variables

 Determine how to manipulate these variables to control the
responses
3/30/2014 4
 SPC methods and experimental design techniques are powerful
tools for the improvement and optimization of a process or
System design.

 Experimental design can be used for

 Improve a process by increasing its performance and
eliminate troubles.

 Establish statistical control of a process variable ,ie identify
the variables to control the processes.






3/30/2014 5
Terminologies

 FACTOR

 It is a variable that is changed and observations are achieved.

Examples

 Time
 Temperature
 Operator
3/30/2014 6
2. LEVEL

 A value that is assigned to change the factor

Eg

 Two levels of factor could be

 110
o
c and 150
o
C
3/30/2014 7
3. TREATMENT CONDITION

 The set of conditions ( factors & levels) for a test in an
experiment.

4. REPLICATE

 A repeat of a treatment condition .
 It requires a change in the setup.

5.REPETITION

 Multiple results of a treatment condition


3/30/2014 8
6. RANDOMIZATION

 Treatment conditions are run in a chance in order to
prevent any build up in the results.

7. ORTHOGONAL ARRAY


 Simplified method of putting together the treatment
conditions so that the design is balanced and factors can be
analyzed singly or in combination

3/30/2014 9
8. INTERACTION

 Two or more factors that together produce a result different
than their separate effects.

3/30/2014 10
Analysis of Variance

 In general the samples are selected as the levels

 The specimen in the each level are known as observations or
replications

 The no of variables under the study at a time is the number of
factors
3/30/2014 11
 The observations of the experiment can be written in a
mathematical model as

1......a) (i level factor each at ns observatio of number n
levels factor of number a
1.....n) j ..a, 1......... (i levels factor within noise or error
1.......n) (i level factor each at ) effect (factor treatment to due difference f
ns observatio the all of meanvalue
1......n) level(i factor
th
i the at n observatio
th
j y
f y
i

ij
i
ij
ij i ij
=
= = e
=
=
=
e +
=
=
=
=
=
+ =
µ
µ
3/30/2014 12
 Data for factor level Experimental Design

Factor
Level
DATA
1 Y
11
Y
12
…….. Y
1n1
2 Y
21
Y
22
…….. Y
2
n
2
. . . …….. .
. . . ……… .
a Y
a1
Y
a2
Yan
a
3/30/2014 13
 In general the observations of the experiments are assumed to
be normally and independently distributed with the same
varience in each factor level .

 Thus the sum of the factor effects fi can be defined as


0
a
1 i
i =
¿
=
f
3/30/2014 14
 From the preceding equation the null statistical hypothesis for
the F test is stated as

0 f f f f : H a 3 2 1 0 = = = = = .........
3/30/2014 15
 If this hypothesis is false an alternative hypothesis is stated
as

i one least at for 0 f : H 1 0 =
3/30/2014 16
 Testing for the validity of the hypothesis is accomplished by
the analysis of variance (ANNOVA)

 The total variability in the data is given as

2
i
i
a
1 i
i
1 j i
i
ij
a
1 i
i
1 j
2
ij
N
T
n
A
n
n
A
y
n
N
T
y
|
.
|

\
|
÷ +
¿ ¿
(
¸
(

¸

|
.
|

\
|
÷ =
¿ ¿
|
.
|

\
|
= =
= =
÷
3/30/2014 17
 Where Total no of observations




 Total number of observations under the i
th
factor is given by


¿
=
=
a
1 i
i n N
¿
=
=
i
1 j
ij i
n
Y A
3/30/2014 18
Total of the observations under all factor levels are given by




 After expanding the square in the relationship for the total
variability and simplifying the equation, the total variability in the
data can be expressed as


¿
=
=
a
1 i
i A T
E F T SS SS SS + =
3/30/2014 19
 Where total sum of squares is given by



 Sum of squares due to factor



 Sum of squares due to errors is given by

N
T
n
Y SS
2
a
1 i
i
i j
ij
2
T ÷
¿ ¿
=
= =
N
T
n
A
SS
2
a
1 i i
2
i
F ÷
¿
=
=
F T E SS SS SS ÷ =
3/30/2014 20
 If there are large difference among the means at different levels
SS
F
will be large.

 The variability due to changing factor levels can be obtained by
comparing the magnitude of SS
F
and SS
E

 To facilitate this mean squares of the factor and the errors are
calculated by normalizing the sum of squares by their
respective number of degrees of freedom


E
E
E
F
F
F
V
SS
MS
V
SS
MS
=
=
3/30/2014 21
Where the degrees of freedom for the factor and error is given
by

a N V
1 a V
E
F
÷ =
÷ =
3/30/2014 22
 Analysis of Variance is a technique to determine the significant
factors based on the F test that is used to compare ratio of factor
with error variance











 It can be said that factor level means are different for a level α
test




E
F
MS
MS
F =
E 2
F 1
2 1
V V
and
V V

v , v Fαα F if
=
=
>
Where
3/30/2014 23
 Result for one factor Experimental design

Source Sum of
squares
Degree of
freedom
Mean
Square
F Fαv
1
v
2
Signifi
cance(
yes/No
)
Factor SS
F
V
F
MS
F
F
Error SS
E
V
E
MS
E
Total SS
T
3/30/2014 24
Consider the following data for four levels of a factor







Based on the following data use α= 0.05 to determine if
there is any significant difference in the data

Level Data
1 26 25 32 29 30
2 34 31 26
3 15 16 25 18 14
4 12 15 13 11
EXAMPLE PROBLEM
3/30/2014 25
SOLUTION
a=4
n
1
=5
n
2
=3
n
3
=5
n
4
=4
N= no of observations



= n
1
+n
2
+n
3
+n
4
= 4+5+3+5+4
= 17

¿
=
=
a
1 i
i n N
3/30/2014 26
 V
F
= a-1
= 4-1=3

 V
E
= N-a
=17-4 =13

 Total number of observations under the i
th
factor is given by



A
1
=142
A
2
=91
A
3
=88
A
4
=51

¿
=
=
i
1 j
ij i
n
y A
3/30/2014 27
Level Data Total S
2
1 26 25 32 29 30 142 28.4 8.3
2 34 31 26 91 30.3 16.33
3 15 16 25 18 14 88 17.4 21.30
4 12 15 13 11 51 12.8 2.92
y
3/30/2014 28
 Total of the observations under all factor levels are given by



=142+91+88+51
=372
Where total sum of squares is given by


¿
=
=
a
1 i
i A T
N
T
n
Y SS
2
a
1 i
i
i j
ij
2
T ÷
¿ ¿
=
= =
5 . 1020
17
372
9117
2
= ÷ =
3/30/2014 29
65 . 860
17
371
4
51
5
88
3
91
5
142
2 2 2 2 2
=
÷ + + + =
÷
¿
=
= N
T
n
A
SS
2
a
1 i
2
i
i
F
3/30/2014 30
85 . 159 = ÷ = F T E SS SS SS
30 . 12
13
85 . 159
88 . 286
3
65 . 860
= =
=
= =
=
E
E
E
F
F
F
V
SS
MS
V
SS
MS
3/30/2014 31
33 . 23 =
=
E
F
MS
MS
F
3/30/2014 32
Source Sum of
square
s
Degree
s of
freedo
m
Mean
Square
F Fαv
1
v
2
Signifi
cant
(Yes/N
o)
Factor 860.65 3 286.88 23.33 3.41 Yes
Error 159.85 13 12.3
Total 1020.5
Since F> Fαv
1
v
2
it can be stated that the factor level means are significant


3/30/2014 33
One factor at a Time
 Consider an experimental design with N
F
factors with two
levels for each factor.

 At treatment condition 1,all the factors are run at level 1 to
obtain the benchmark or reference value.



The next treatment condition is to run factor A at level 2
while other factors are kept at level 1

 For third treatment condition factor B is run at level 2 while
other factor are run at level 1

3/30/2014 34
 The change in response due to a change in the level of the factor
is known as the effect of factor

 The main effect of the factor is defined as the difference
between the average response at two levels .

 In order to determine the effect of each factor the response at
level 2 are compared to the reference value (level 1)

0 C C
0 B B
0 A A
Y Y e
Y Y e
Y Y e
÷ =
÷ =
÷ =
3/30/2014 35
Factor /Levels Response(Res
ults)
Treatmen
t
Conditio
n

A B C D …. N
F
1 1 1 1 1 …. 1 Y
0
2 2 1 1 1 ….. 1 Y
1
3 1 2 1 1 ….. 1 Y
2
4 1 1 2 1 ….. 1 Y
3
5 1 1 1 2 ….. 1 Y
4
.
.
n
tc
Y
NF
3/30/2014 36
Example problem
 An experiment involves maximizing a response variable.
The three factors are compared at two levels. The factors
and levels are given in the following table.

Factors Level 1 Level 2
Speed (m/s) 25 35
Pressure (Pa) 50 75
Time( min) 5 10
3/30/2014 37
 At treatment condition all the factors are run at level 1 to
obtain benchmark value.

 TC2 is to run the experiment is to run speed at level 2 while
pressure and time are kept at level 1

 TC3 is to run the experiment at pressure at level 2 and speed
and time at level 1

 TC4 is to run the experiment at speed ,pressure at level 1 and
time at level 2
3/30/2014 38
Treatment
condition
Factor Levels Response
Results
Speed Pressure Time
1 1 1 1 2.8
2 2 1 1 3.4
3 1 2 1 4.6
4 1 1 2 3.7
The effect of speed
=e
s
=Ys-Yo= 3.4-2.8=0.6

The effecr of Pressure=

e
p
=Yp-Yo=4.6-2.8=1.8

The effect of time =
e
T
=Yt-Yo=3.7-2.8=0.9
3/30/2014 39
 It appears that

 Pressure
 Time and Speed have a positive effect on the process

In other words increasing any of the factors increases the
response
3/30/2014 40
Orthogonal Design
 Orthogonality means the experimental design is balanced.



3/30/2014 41
Treatment
Condition
Factor Levels Response(R
esults
A B C
1 1 1 1 y
111
2 2 1 1 Y
211

3 1 2 1 Y
121

4 2 2 1 Y
221

5 1 1 2 Y
112

6 2 1 2 Y
212

7 1 2 2 Y
122

8 2 2 2 Y
222

Table : Orthogonal Experiment Design for three
factors with Two levels
3/30/2014 42
Treatment
Condition
Factor Levels Response(R
esults
A B C
1 1 1 1 y
111
2 2 1 1 Y
211

3 1 2 1 Y
121

4 2 2 1 Y
221

5 1 1 2 Y
112

6 2 1 2 Y
221

7 1 2 2 Y
122

8 2 2 2 Y
222

Treatment Conditions 1 ,2,3 & 5 are the same as treatment
condition 1,2,3 &4 in the previous for the one factor at a
time in the previous experiments 3/30/2014 43
 Here in this design there are

 Four 1 levels

 Four 2 levels

 Although different designs have different no of rows,
columns and levels there will be an equal number of
occurrences for each levels.

 The idea of balance provides statistically independent
results
3/30/2014 44
 Looking at the relationship between one column and the
other we find the that for each level within one column
,each level in any other column will occur in equal
number of times


 Therefore the four level 1s for factor A have two level 1 s
and two level 2s for factor B and two level 1s and two
level 2s for factor C
3/30/2014 45
Treatment
Condition
Factor Levels Response(R
esults
A B C
1 1 1 1 y
111
2 2 1 1 Y
211

3 1 2 1 Y
121

4 2 2 1 Y
221

5 1 1 2 Y
112

6 2 1 2 Y
221

7 1 2 2 Y
122

8 2 2 2 Y
222

3/30/2014 46
 The effect of factor A at each level is statistically independent
because the effects of factor B and factor C for each level are
averaged into both levels of the factor A


 Statistical independence is true for factor B and factor C
3/30/2014 47
 In order to find the main effects of each factor/level the results
must be averaged.
 The average of the response for factor A due to the level i(i=1,2)
is given by




 Similarly the average response for factor B due to level j (j-
1,2)is


 And so on for all the factors

¿
÷
= ijk....m
F
1
i Y
1) (N
2
A
¿
÷
= ijk....m
F
1
Y
1) (N
2
i B
3/30/2014 48
 While the number of treatment conditions increased by a
factor of two the number of values that were used to calculate
the level/Factor average increased by a factor of four.


 This improvement in the number of values used to determine
the average makes the orthogonal design very efficient in
addition to being balanced
3/30/2014 49
 The main effects of Factors A and B are then given by






 And so on for other factors

1 2 B
1 2 A
B B e
A A e
÷ =
÷ =
3/30/2014 50
Factors Level 1 Level 2
Speed (m/s) 25 35
Pressure (Pa) 50 75
Time( min) 5 10
3/30/2014 51
Treatment
Condition
Factor Levels Response(Re
sults
Speed Pressure Time
1 1(25m/s) 1(50 Pa) 1(5 min) 2.8

2 2(35m/s) 1(50 Pa)

1(5 min) 3.4
3 1(25m/s)

2(75 Pa) 1(5 min) 4.6


4 2(35m/s) 2(75 Pa) 1(5 min) 3.8
5 1(25m/s)

1(50 Pa) 2(10 min) 3.7
6 2(35m/s) 1(50 Pa) 2(10 min) 2.7
7 1(25m/s) 2(75 Pa) 2(10 min) 3.1
8 2(35m/s) 2(75 Pa) 2(10 min) 4.4

3/30/2014 52
 The average response for factors due to each level i
¿
÷
= ijk....m
F
1
i Y
1) (N
2
A
¿
÷
= ijk....m
F
1
Y
1) (N
2
i B
3/30/2014 53
475 . 3 ) 4 . 4 1 . 3 7 . 2 7 . 3 (
4
1
T
650 . 3 ) 8 . 3 6 . 4 4 . 3 8 . 2 (
4
1
T
975 . 3 ) 4 . 4 1 . 3 8 . 3 6 . 4 (
4
1
P
150 . 3 ) 7 . 2 7 . 3 4 . 3 8 . 2 (
4
1
P
575 . 3 ) 1 . 4 7 . 2 8 . 3 4 . 3 (
4
1
S
550 . 3 ) 1 . 3 7 . 3 6 . 4 8 . 2 (
4
1
S
2
1
2
1
2
1
= + + +
= + + +
= + + +
= + + +
= + + +
= + + +
=
=
=
=
=
=
175 . 0 650 . 3 475 . 3 T T e
825 . 0 150 . 3 975 . 3 P P e
025 . 0 550 . 3 575 . 3 1 S S e
1 2 T
1 2 P
2 S
÷ = ÷
= ÷ ÷ =
= ÷ =
= ÷ =
=
÷ =
3/30/2014 54
 Based on these calculations

 Increase in pressure (level 1 to level 2) has strong
effect on the process

 Where as small effects of speed and time are mostly
due to the natural variation in the process

 This information is better than that obtained from the
one factor at a time approach

3/30/2014 55
TWO FACTORS
3/30/2014 56
 Consider an experiment run N times with

 two factors A and B

3/30/2014 57
Factor
A
Levels
Factor B Levels
1 2 …….. b
1 Y
111,
Y
112
,
…..Y
11n
Y
121,
Y
122
,
…..Y
12n

Y
1b1
,Y
1b2
,…
..Y
1bn
2 Y
211
,Y
212
….Y
21n
Y
221
,Y
222
,
…Y
22n
………
.
Y
2b1
,Y
2b2
,…
..Y
2b3
.
.

a Y
a11
,Y
a12
,
…Y
a1n
Y
a21
,Y
a22

.Y
a2n
………
.
Y
ab1
,Y
ab2
,…
Y
abn
Data for Two factorial Experimental Design
3/30/2014 58
 These observations can be written in the mathematical
model as

1......n) k 1....b, j ....a), 1 (i levels and factors within noise or error ijk
1....b) (j B level factor each B)at effect (factor treatment to due difference j
1...a) (i A level factor each at A) effect (factor treatment to due difference i
th th th
ijk
where
ijk ij j i ijk
(fB)
(fA)
ns observatio the all for value mean
1.....n) K 1....b, j 1....a, (
B and A factors the of levels j the and i the at n observatio k Y
f(AB) (fB) (fA) μ Y
= = = =
= =
= =
e
=
= = =
=
e + + + + =
µ
3/30/2014 59
a= no of levels for factor A
b= no of levels for factor B
n = number of observations for each run (replicate)
The total number of observations is given by

N=abn

Total number of observations under i
th
and j
th
levels for factors A
and B


¿
=
=
n
1 k
ijk ij Y A
3/30/2014 60
 Total of the observations under the i
th
level for factor A is
given by




 Total of the observations under the j
th
level for factor B is
given by
¿
=
=
b
A A
1 j
ij Ai
¿
=
=
a
A A
1 i
ij Bj
3/30/2014 61
 Total of the observations under all the factor levels are
given by




 After expanding the square in the relationship for the total
variability in the data and simplifying the equation ,the total
variability in the data is expressed as

¿
=
¿
=
= =
b
1 j
Bj
a
1 i
Ai A A T
E AB B A T SS SS SS SS SS + + + =
3/30/2014 62
 Where the total sum of squares



 Sum of squares due to factor A



 Sum of squares due to factor B

N
T2
y SS
a
1 i
b
1 j
n
1 k
ijk
2
T ÷
¿ ¿ ¿
=
= = =
N
T
bn
A
SS
2
a
1 i
Ai
2
A ÷
¿
=
=
N
T
an
A
SS
2
b
1 j
B
2
B ÷ =
¿
=
j
3/30/2014 63
 Sum of squares due to the interaction between factors
A and B
AB B A T E
B A
2
a
1 i
b
1 j
2
AB
SS SS SS SS SS
SS SS
N
T
n
ij A
SS
÷ ÷ ÷ =
÷ ÷ ÷ =
¿ ¿
= =
3/30/2014 64
ab - N V
1) 1)(b (a V
1 b V
1 a V
V
SS
V
SS
MS
V
SS
MS
V
SS
MS
E
AB
B
A
E
E
AB
AB
AB
B
B
B
A
A
A
=
÷ ÷ =
÷ =
÷ =
=
=
=
=
E MS
3/30/2014 65
E 2
B 1 where
V2 V1, B
of value particular a for nt significa is A factor The
E 2
A 1 where
2 1 A
E
AB
AB
E
B
B
E
A
A
V V
V V
Fαα F If
V V
V V
v Fαα F If
MS
MS
F
MS
MS
F
MS
MS
F
=
=
=
=
>
>
=
=
=
o
3/30/2014 66
o of value particular a for nt significa is B and A of interation The
E 2
AB 1 where
V2 V1, AB
V V
V V
Fαα F If
=
=
>
 Result for Two factor Experimental design

Source Sum of
squares
Degree of
freedom
Mean
Square
F Fαv
1
v
2
Signifi
cance(
yes/No
)
Factor
A
SS
A
V
A
MS
A
F
A
Fαv
A
v
E

Factor
B
SS
B
V
B
MS
B
F
B
Fαv
B
v
E
Factor
AB
SS
AB
V
AB
MS
AB
F
AB
Fαv
AB
v
E
Error SS
E
V
E
MS
E
Total SS
T
3/30/2014 67
Example problem
3/30/2014 68
 Consider an experiment on the volatility of a fluid with two
factors concentration and temperature.


 Determine the significance of each factor for a level α=0.05
Factor A
levels
Factor B levels

1 2 3
1 41,38 43,49 46,47
2 42,43 46,45 48,49
3/30/2014 69
 Result for Two factor Experimental design

Source Sum of
squares
Degree of
freedom
Mean
Square
F Fαv
1
v
2
Signifi
cance(
yes/No
)
Factor
A
6.75

1 6.75

1.165 5.99 No
Factor
B
90.5

2
45.25
11.081
6
5.14 YES
Factor
AB
6.5 2
3.25
0.7959 5.14 No

Error 24.5

6 4.0833

Total 128.25

11
3/30/2014 70