IMPROVE

DOE

1
 Design Of Experiment (DOE)
¤ Aim of Experimentation ¤
▪ Discovery of new scientific principles or phenomena

▪ Establishing cause-and-effect relationships

▪ Seeking improved methods of processes

▪ Finding robust design for a product

▪ Finding optimum operating conditions of a process

▪ Comparing different processes , machines , materials , etc.

EXPERIMENTATION IS A VITAL FUNCTION

IN OUR UNENDING QUEST FOR PROGRESS.

2
 Importance of DOE

• Introduction to SPC would stabilize a process and reduce the

variability .
• But SPC alone will not allow the organizations to meet the
quality & cost challenges.
• DOE can be used in conjunction with SPC to minimize process
variation further .
• Solving chronic quality problems has always been a challenge.
An important part of many quality improvement and quality
development program is an empirical evaluation of the effect
of process variables upon process performance.

3
 Importance of DOE ( Contd.)

• The classical method of experimentation is to employ trial and

error methods or to restrict attention to one factor at a time ,
the other factors being kept constant.
• The main drawback of this approach is that it leads to
misleading and erroneous conclusions if “interaction “ effects
exist.
• The various techniques developed to aid the experimenter in
efficiently planning and conducting these trials constitute the
subject matter of Design of Experiments (DOE).

4
 Experimental Process ( Contd)

• “ Uncontrollable ” factors that affect the process cause

random disturbances.
• The experimentation process : A process in which the
experiment is performed at controllable values of the selected
independent variables Xi’s, and responses Yj’s are observed
while attempting to negate the effect of the uncontrollable
factors Zl’s.
• The responses are typically affected by error which can be
viewed as arising out of the combined random effects of all
uncontrollable variables.

5
 “Black Box” View of Experimental Process
Uncontrollable variables
Zl (l = 1,2,…,p )

Intput responses Output

variables
Xi ( i = 1,2, …, k ) Experimental Process Yj ( j = 1,2, …,m )
[ Independent [ Dependant
& controllable ] variables ]

Cause-and-effect function :
Yj = f (X1,X2,..Xk ; Z1,Z2,…Zp)

Yj = f ( X1,X2,…, Xk) +εj

6
 Cardinal Principles of Experimental Designs

1. REPLICATION ( Repetition of the basic experiment)

▪ To obtain an estimate of experimental error
▪ To increase sensitivity of comparisons
2. RANDOMISATION ( Randomly determining the sequences of trials,
assigning treatments to experimental units , taking measurements , etc, )
▪ To forestall or ‘average out’ the effects of unknown variations in raw materials ,
time trend , etc. over which the experimenter has no control
▪ To avoid any possible systematic bias so that a particular treatment is not
continually favored or handicapped
▪ To validate the statistical assumption that the observations ( or errors) are
independently distributed random variables
3. LOCAL CONTROL ( Technique of blocking or balancing the experimental
units)
▪ To ensure uniformity in the background conditions by isolating known
disturbing factors

7
 Terminology in DOE
• Factor : A variable or attribute which influences or is
suspected of influencing the characteristic being investigated
Examples:
Variable : Time , Temp , Pressure , Speed ,etc
Attribute : Method , Operator , Supplier , Position , etc.
• Level : Specific value of a factor
Examples :
Variable : Time 20 mins ( First level) , 40 mins ( second level)
Attribute : Method 1 ( first level ) , Method 2 ( Second level)

8
 Terminology in DOE ( Contd)

• Treatment : A set of levels of all factors employed in a given

experimental trial
Example : A trial using temp T1 , pressure P2 , M/C M1 would
constitute one treatment
• Experimental unit : Facility for conducting a trial.
Experimental units are allocated to different treatments.
• Response : Numerical result of a trial with a given treatment
Example : Yield , Purity , Surface finish , No. of defects , Cost ,
etc.

9
 Terminology in DOE ( Contd)

• Effect : Change in response due to change in levels of a factor

• Main Effect : Effect of a factor averaged over different levels of the other
factors
• Interaction : The difference in effect of one factor when a second factor
is changed from one level to another
• Experimental Error : Variation in response when the same treatment is
repeated
• Randomization : The process of assigning experimental units to
treatments in a purely chance manner , sequence of conducting the trials
by a chance process

10
 Some Standard Designs

• There are various standard statistical designs available for

experimentation , such as

Completely Randomized Design ,

Randomized Block Design ,
Latin Square Design ,
Full Factorial Design ,
Fractional Factorial Design , etc.

11
Completely Randomized Design ( One-way Classification):

The simplest design

~ For one factor with several levels
~ Allotment of various treatments at random to different
experimental units
~ Quite efficient when experimental units are
homogeneous
12
 Some Standard Designs ( contd.)
2) Randomized Block Design :
~ One factor (treatment) under investigation
with one extraneous factor (in a block)
~ Principle of Local Control is used
( by considering several internally
homogenous blocks).
~ Example :
• To evaluate 5 batches of a raw material w.r.t. a
quality characteristic by processing on three
machines. Each m/c constitutes a block , and
the 5 batches ( levels or treatments) of the M/C1 M/C 2 M/C 3
raw material to be processed in each m/c in a
random order to avoid any bias. Fresh D E A
randomization for each block.
A B E
• If A,B,C,D & E are the batches , the RB Design
may be as follows : C D B
No. of blocks (m/cs) B C D
= Common Replication number
No. of treatments ( five batches) E A C
= No. of units in a block
13
 Some Standard Designs ( contd.)
3) Latin Square design
~ One factor (treatment)
under investigation with two Machine
extraneous factors Operator

~ In the example for Randomized 1 2 3 4 5

Block design , suppose another
extraneous variable ‘operator or I A B C D E
analyst’ is considered.
Then the design shown here, II B C D E A
called Latin Square design,
may be used: III C D E A B
Here each treatment (batch of IV D E A B C
RM) occurs once and only once in
each row ( Operator) and in each V E A B C D
column( M/C).

14
 Example

• To study effects of Four Types of

Graphite Coaters on Light-box Readings

Day Graphite Type

M A K L Total
1 4.0 4.8 5.0 4.6 18.4
2 4.8 5.0 5.2 4.6 19.6
3 4.0 4.8 5.6 5.0 19.4
Total 12.8 14.6 15.8 14.2 57.4

Avg 4.3 4.9 5.3 4.7 4.8 ANOVA ( CR Design)

Source DF SS MS F Critical
value at
5 % level

Graphite 3 1.53 0.51 5.4 4.07

Error 8 0.75 0.094
Total 11 2.28
15
 Some Standard Designs ( contd.)
4) Factorial Design
~ We are often concerned with the individual
and joint effects of more than one variable or Expt. A B C
factor on a particular response. No.
~ The combinations of the levels of these
1 1 1 1
factors will now play the roles of the different
treatments. 2 1 1 2
~ Extraneous variables , if any, are handled as 3 1 2 1
before. 4 1 2 2
~ If all possible interactions are to be 5 2 1 1
examined , every possible combinations of
6 2 1 2
levels of various factors have
to be carried out in one block. 7 2 2 1
~ FF Design has a serious limitation for a 8 2 2 2
simultaneous study of a large number of
factors. E.g. a 6-factor experiment for 2 levels
for eachrequires 26 or 64 treatment Full Factorial (FF) Design
combinations or trials. Consider three factors A,B & C ,
each at 2-levels.
16
FACTORIAL EXPERIMENT

17
 Experiments with Multiple Factors

• Two approaches for experimentation with two or

more factors
– One-Factor-At- a-Time (OFAT)
• Only one factor is changed at a time rest are kept constant
• Inefficient (requires larger number of test runs)
• Fails to consider any possible interaction between factors
– Factorial Experiments
• Factors are varied together
• More efficient approach
• Can determine factor interaction(s)
• Main effects and interaction effects
18
 Interaction effect
An Example: Objective is to determine the effect of book the
students read and the study time per week on the marks they
score in the examination. Suppose there are two books (B1 and
B2) and study times are 10 and 20 hours. The average scores of
the class are as follows:

High 30 52
Study Hours Study Hours
10 (L) 20 (H)
B1 20 30
Book

B2 40 52
Low 20 40
B1 B2
Book
19
 High 30 52

Study Hours(B)
Low 20 40
B1 B2
Book (A)

Effect: The change in the mean response when the factor is changed from low to high

Determining Effects
52
- =21 Hours High 40
- =11 30
Hours Low

- =1
20
B1 B2
Book (A)
20
 Interaction effect

Let the new data is as below

High 40 12

Study Hours
Study Hours
10 (L) 20 (H)
B1 20 40
Book

B2 50 12 Low 20 50
B1 B2
Book

21
 High 40 12

Study Hours(B) Low 20 50

B1 B2
Book (A)

Determining Effects 50
Hours Low

- =1 40

Score
- =-9 Hours High
- =-29 20 12
B1 B2
Book (A)

22
Interaction effect: Average difference in effect of Factor 1 (Factor 2) for 2
levels of Factor 2 (Factor 1)

Effect of Book (Factor 1):

When Hours (Factor 2) High = 12-40 =-28

When Hours (Factor 2) Low = 50-20 =30

Effect of Hours (Factor 2):

When Book (Factor 1) High = 12-50 =-38

When Book (Factor 1) Low = 40-20 =20

Interaction Effect = ==-29

23
Data Table Two Factor Factorial Design
Factor B
Totals Means
1
2 … b
1

2
Factor A

..

Totals

Means …

24
 Statistical Model

= Overall mean which is common to all parameters

= Effect of factor A

= Effect of factor B

= Interaction Effect

= Random error arising due to e.g. nuisance variables,

heterogeneity of experimental units, measurement etc.

25
 Sum of Squares Computations
=-
= -
=
=--
Where = - Sum of squares of cell

=-- - totals of ab cells

26
 ANOVA Table
Source of Sum of Degrees of Mean Square
Variation Squares Freedom

Main Effects
A SSA a-1

B SSB b-1

Interaction
AB SSAB (a-1)(b-1)

Error SSE ab(n-1)

Total SST abn-1

27
 An example
Three missile system are tested (for their preference) for burning rates
of four propellant types. Each combination of missile and propellant
was burned twice (replication n=2).
Propeller Type (Factor B) Row
Totals
b1 b2 b3 b4

34.0 30.1 29.8 29.0 244.0

Missile Systems (Factor A)

a1 32.7 32.8 26.7 28.9

32.0 30.2 28.7 27.6 237.4

a2 33.2 29.8 28.1 27.8

28.4 27.3 29.7 28.8 228.8

a3 29.3 28.9 27.3 29.1

Column 189.6 179.1 170.3 171.2 710.2

Totals
28
 SS Calculations
• = = 91.68
• = 14.52
• = = 40.08
• = ( ) -14.52-40.08 = 22.17

• = 91.68-14.52-40.08-22.17= 14.91
𝑠 𝑠 subtotal

29
 ANOVA Table
Source of Sum of Degrees of Mean
Variation Squares Freedom Square (α=0.05)

Main Effects
A 14.52 2 7.26 5.85 3.8853

B 40.08 3 13.36 10.77 3.4903

Interaction Effects
AB 22.17 6 3.70 2.98 2.9961

Error 14.91 12 1.24

Total 91.68 23

30
31
 Nomenclature (Full) Factorial Designs
• We consider all possible combinations of factors
and their levels in full factorial experiments
• Based on number of factors and their levels
– Two level Designs
• Two factors each at two levels: 22
• Three factors each at two levels: 23
• K factors each at two levels: 2k
– Three level designs
• Two factors each at three levels: 32
• Three factors each at three levels: 33
• K factors each at three levels: 3k
32
 2 Factorial Designs
K

• These are special full factorial designs with K

factors each at 2 Levels
• Useful for screening experiments
– Out of large number of factors we are interested
in selecting few important factors
– More than two levels of the factors are not
required

33
 Special Factorial Designs ( 2 )2

Tabular display Design Matrix

Factor 1
- + Factor 1
Run
Letter
Factor 2 Response
- 1 2 Rep
Factor
2

+ 3 4 1 (1) – –

2 a + –
Geometric Display
3 4
3 b – +
+
4 ab + +
Factor 2

_ 1 2
_ +
Factor 1 34
 An example
Replicates
Factor 1
Run
Letter
Factor 2 Totals
Rep 1 2 3

1 (1) – – 28 25 27 80
2 a + – 36 32 32 100
3 b – + 18 19 23 60
4 ab + + 31 30 29 90

35
 Analysis of 22 Design
b=60 ab=90
+

Factor B
-
(1)=80 a=100
- Factor A +

Effects

- ===8.33 Values in
bracket
- ===-5 are called
Contrasts
- ===1.67

36
Effects and Sums of squares using Contrasts
Contrast A = ab+a-b-(1) = (90+100-60-80)= 50
Contrast B = ab+b-a-(1) = (90+60-80-100)= -30
Contrast AB = ab+(1)-a-b = (90+80-100-60)= 10

Effect = SS = Here, k=2 and n=3

EffectA = = 8.33 SSA = = 208.33

EffectB = = -5.00 SSB = = 75.00

EffectAB = = 1.67 SSAB = = 8.33

SSTOTAL = =323.00

37
 ANOVA Table

SOURC SS DOF MS Fcal

E
A 208.33 1 208.33 53.14

B 75.0 1 75.0 19.13

AB 8.33 1 8.33 2.13

ERROR 31.34 8 3.92

TOTAL 323.00 11

38
 23 Factorial Designs
Three factors each at two levels
Tabular display Geometric Display
F2
Low (-) High (+) 7 8
F3 F3
Low High Low High + 3 4
(-) (+) (-) (+)
5 6 +

Factor 2
1 5 3 7
Low (-)
1 5 3 7

3
F1 _ 1

or
2 _

ct
2 6 4 8

Fa
_ +
High (+) Factor 1
2 6 4 8

39
Design Matrix for 23 Design

RUN Treatment Effects Treatme

Combinations nt Totals
I A B C AB AC BC ABC

1 (1) + – – – + + + –
2 a + + – – – – + +
3 b + – + – – + – +
4 ab + + + – + – – –
5 c + – – + + – – +
6 ac + + – + – + –
7 bc + – + + – – + –
8
Design abc
Matrix for 23+can easily
+ + obtained
be + from+ Matrix+for 22 (yellow
+ +
colored)

Can be analyzed in the same way as that for 22 Designs 40

Why Fractional Factorial Designs
Number of Number of
Factors Treatment
Combinations or
runs
1 2
2 4
3 8
4 16
Number of
5 32 treatment
6 62 combinations
or trail runs
7 128 when number
8 256 of levels =2
9 512
10 1024

41
Main and Interactions terms in 2K Designs
Order of Interaction
# Factors Main
(K) Effects 2 3 4 5 6 7 8 9 10

2 2 1
3 3 3 1
4 4 6 4 1
5 5 10 10 5 1
6 6 15 20 15 6 1
7 7 21 35 35 21 7 1
8 8 28 56 70 56 28 8 1
9 9 36 84 126 126 84 36 9 1
10 10 45 120 210 252 210 120 45 10 1

42
Number of Runs in Fractional Designs

Fractional Design Fraction of runs of Full

Factorial Design (2k)
2k-1 1/2
2k-2 1/4
2k-3 1/8
2k-4 1/16

43
 Obtaining 23-1 from 23
Choosing appropriate rows from 23 Full Factorial design
RU Treatment Factor Effects

N Combinations I A B C AB AC BC ABC

1 (1) + – – – + + + –
2 a + + – – – – + +
3 b + – + – – + – +
4 ab + + + – + – – –
5 c + – – + + – – +
6 ac + + – + – + –
7 bc + – + + – – + –
8 abc + + + + + + + +
44
 Obtaining 2 from 2 Cont...
3-1 3

Considering Rows 1-2-3-4

RUN Treatment Effects

Combinations I A B C AB AC BC ABC

1 (1) + – – – + + + –
2 a + + – – – – + +
3 b + – + – – + – +
4 ab + + + – + – – –

All Cs are with negative sign hence effect of C cannot be calculated

45
 Obtaining 2 from 2 Cont...
3-1 3

Considering Rows 2-3-4-5

RU Treatment Factor Effects

N Combinations I A B C AB AC BC ABC

2 a + + – – – – + +
3 b + – + – – + – +
4 ab + + + – + – – –
5 c + – – + + – – +
Unequal number of positives and negatives in columns C, BC and ABC
hence effects of these cannot be estimated

46
 Obtaining 2 from 2 Cont...
3-1 3

Considering Rows 2-3-5-8

RU Treatment Factor Effects

N Combinations I A B C AB AC BC ABC

2 a + + – – – – + +
3 b + – + – – + – +
5 c + – – + + – – +
8 abc + + + + + + + +
No problem hence rows 2-3-5-8 can be considered for Half Fraction of 23

47
 Development of Orthogonal Array (OA) Designs

• The concept of Orthogonal Arrays was first introduced by Dr. C.R.Rao

( 1947) of Indian Statistical Institute.
• Dr. G. Taguchi ( 1959) of Japan prepared various OA tables and
introduced the concept of Linear Graph associated with each OA ,
and also developed various techniques for adapting them to meet
the requirements of different practical situations.
• These are Fractional Factorial Designs which enable estimation of
important effects , by deliberate sacrifice of information on some
unimportant effects of higher order.
• These methods have the advantage of being highly flexible and
readily amenable to allocation of different levels of factors when
these levels are not same in number for all the factors to be studied.
• These also help in cutting the size of experimentation to the bare
minimum.
• However , importance of interactions is largely ignored in Taguchi’s
methods.

48
 Orthogonal array
• An orthogonal array is a matrix of numbers arranged
in columns and row
• Each column represents a specific factor
• Each row represents the state of the factors in a
given experiment
• The levels of the various factors are balanced and can
be separated from the effects of the other factors
within experiment

49
 Orthogonal array
Expt.
▪ Main Effect plan , OA L8( 27 ) No. A B C D E F G
with 7 factors , each at 2 levels
▪ Eight experiments with each of the 1 1 1 1 1 1 1 1
seven columns consisting of four 1’s
and four 2’s. 2 1 1 1 2 2 2 2
▪ In any two columns , four kinds of
combinations in the horizontal 3 1 2 2 1 1 2 2
direction : 11,12, 21 and 22.
The two columns are said to be “balanced” 4 1 2 2 2 2 1 1
or “orthogonal”.
▪ It is a Full Factorial design for three 5 2 1 2 1 2 1 2
factors , each at two levels , from
which all the main effects and interaction 6 2 1 2 2 1 2 1
effects can be estimated.
7 2 2 1 1 2 2 1

8 2 2 1 2 1 1 2

50
Has a number of unique properties:
• Equal proportions of experiments
• Equal proportions of remaining factor levels
• Equal proportions of combinations of factor levels
is denoted as L 8 (27)
• Degree of freedom for factors =( number of levels –1)
• Degree of freedom of OA: (Number of experiment-1)

51
 O.A. Designs
• Some standard Orthogonal Arrays :
• 2-Level series : L4(23) , L8(27) , L12(211), L16(215),
L32(231), L64(263)
• 3-Level series : L9(34) , L27(313) , L81(340)
• 4-Level series : L16(45) , L64(421)
• 5-Level series : L25(56)
• Mixed series : L18(21 x 37) , L32(21 x 49),
L36(211 x 312) , L36(23 x 313) , L50(21 x 511) ,
L54(21 x 325)

52
 OA L8(27) Design

Expt Col 2 3 4 5 6 7
No. 1
• OA L8(27) Layout :
A B AB C AC BC ABC

• Linear Graph : 1 1 1 1 1 1 1 1
2 1 1 1 2 2 2 2
1 A 3 1 2 2 1 1 2 2
AB AC 4 1 2 2 2 2 1 1
3 5 5 2 1 2 1 2 1 2
6 2 1 2 2 1 2 1
7 2 2 1 1 2 2 1
2 4 7 8 2 2 1 2 1 1 2
B 6 C ABC
BC
53
 Selection of OA Design Layout – An Example
• Consider four factors A,B,C and D , each at 2 levels. Apart from main effects , interactions BxC and BxD are also of interest.

Step 1 . The required Linear Graph :

BxC

BxD
Step 2.
C

Step 3. L8 (27) may be tried

B
D A e

Effect A B C D BxC BxD Total

d.f. 1 1 1 1 1 1 6

54
 Example ( Contd.)
Step 4. The standard linear graph selected is :
2
3
1 5 4

7 6

Step 5. Use two lines out of three lines in linear graph in Step 4. The
remaining edge is deleted from the graph , and the associated numbers
are now indicated as independent points as shown below: Assign factors to
columns as shown.
2 C
3
1 5 4
B A 7 e
D 6

55
 OA Designing ( Contd.)

The OA Layout :

Expt Used for conducting Trials Used for estimation only

. No.
B (1) C (2) D (4) A (6) BxC (3) BxD (5) e (7)

1 1 1 1 1 1 1 1
2 1 1 2 2 1 2 2
3 1 2 1 2 2 1 2
4 1 2 2 1 2 2 1
5 2 1 1 1 2 2 2
6 2 1 2 2 2 1 1
7 2 2 1 2 1 2 1
8 2 2 2 1 1 1 2

56
 Principle of robustization
• Attempt to reduce loss due to noise factors
inherent in any system by considering control
factors
• Attempt to reduce σ2

57
 Robust design
• System design
– Require technical knowledge
– The process of examining competing technologies for
producing a product - Includes choices of technology
and process design
• Parameter design
– Involves determination of parameter values that are
least sensitive to noise
– Use of experimental design methods
• Tolerance design
– Improves the quality by narrowing tolerances on the
process or product parameters
• The focus of Taguchi is on Parameter design
58
 Steps in robust design of experiments
• Define the problem
• Determine the objective
• Define the quality characteristics
• Design the experiment
– Selection of factors and interaction
– Selection of number of levels of the factors, range of factor
levels
– Selection of Orthogonal Array (OA)
– Assignment of factors and interaction to OA column

59
• Perform the experiment
• Analyze the experiment
• Verifying the experiment result

60
 Assigning factors to an orthogonal array
• Count the total degree of freedom needed for the
experiment based on the number of factors and factor
levels
• Select an OA that has at least the required number of
degree of freedom
• Draw the required linear graph
• Select an appropriate standard linear graph
• Assign each main effect and interaction to
appropriate column

61
Ina Sito Tile experiment

62
 Ina sito Tile experiment
Factor Level 1 Level 2

A lime content A1 5% A2 1%
B Granularity B1 Coarse B2 fine
C Agalmatolite C1 43% C2 53%
D agalomatolite type D1 Current D2 Cheaper

E Charge quantity E1 1300 kg E2 1200kg

F waste return F1 0% F2 4%
G Felspar content G1 0% G2 5% 63
 Result of experiment
exp A B C D E F G Resul
t

1 1 1 1 1 1 1 1 16
2 1 1 1 2 2 2 2 17
3 1 2 2 1 1 2 2 12
4 1 2 2 2 2 1 1 6
5 2 1 2 1 2 1 2 6
6 2 1 2 2 1 2 1 68
7 2 2 1 1 2 2 1 42
8 2 2 1 2 1 1 2 26

64
 Response Table
A B C D E F G

Level 1 12.75 26.7 25.2 19 30.5 13.5 33

5 5

Level 2 35.5 21.5 23 29.2 17.7 34.7 15.2

Difference 22.2 5.2 2.2 10.2 12.7 21.2 17.7

Rank 1 6 7 5 4 2 3

Sample calculation
Level 1 average of A= (16+17+12+6)/4 = 12.75
Level 1 average of B= (16+17+6+68)/4= 26.75
Level 1 average of C= (16+17+42+26)/4= 25.25
65
Main Effect of Process Factors on % Defects in Tiles

Optimum setting for minimum defects

A1, B2, C2, D1, E2, F1, G2
66