You are on page 1of 38

TAGUCHI TECHNIQUE FOR

OPTIMIZATION
Dr. R. JEYAPAUL
Associate Professor
Department of Production Engineering
National Institute of Technology
Tiruchirappalli - 15
Factors

Input Process Output

Factors
Need for Optimization

Choose variable

Formulate Constraints

Formulate Objective function

Set up variable bounds

Choose an optimization algorithm

Obtain solution
TERMINOLOGY

DEPENDENT VARIABLE:
      It is an outcome or response of an experiment.
      It is also called as response variable.
      Criterion used is also a dependent variable.
 
INDEPENDENT VARIABLE (OR) FACTORS:
      Variables, which are varied in the experiment.
      Can be controlled at fixed levels.
      Can be varied or set at levels of our interest.
      Randomized.
      Can be qualitative (or) quantitative.
 
CONCEPTS OF TAGUCHI
PHILOSOPHY
 Quality should be designed into the product
and not inspected into it

 Quality is best achieved by minimizing the


deviation from a target. The product design
should be such that it is immune to
uncontrollable factors

 The cost of quality should be measured as a


function of deviation from the standard
TAGUCHI TECHNIQUE
 Off-line QA Techniques

 Ensures Quality of design of process and


products

 Robust design is the procedure used

 Makes use of “Orthogonal Arrays” for


designing experiments
Traditional Approaches
 One factor at a time
 Multiple factors, varying one factor at a
time
 Multiple factors, varying all factors at a
time
 Factorial design
 Taguchi methodology
CONVENTIONAL TEST STRATERGIES
1. One factor experiments
Determining the effect of one factor keeping all other factors constant

Trial Factor Level Test Result Average


1 A1 * * Y1
2 A2 * * Y2
2. Several Factors one at a time

Trial Factors Result Average

A B C

1 1 1 1 * * * Y1

2 2 1 1 * * * Y2

3 1 2 1 * * * Y3

4 1 1 2 * * * Y4
3. Several factors all at the same time

Trial Factor and Level Result Average


A B C D
1 1 1 1 1 * * * Y1
2 2 2 2 2 * * * Y2

4. Factorial Experiments
5. Fractional Factorial Experiments
RESOURCE DIFFERENCES OF
TRADITIONAL AND TAGUCHI
EXPERIMENTS
NO OF NO OF NO OF EXPERIMENTS
FACTORS LEVELS FULL FACTORIAL TAGUCHI

3 2 8 4
7 2 128 8
15 2 32,768 16
4 3 81 9
13 3 1,594,323 27
STEPS IN EXPERIMENTATION

 State the problem


 Determine the objective
 Determine the measurement method
 Identify the factors influencing the performance
characteristic
 Separate the factors into control and noise factors
 Determine the number of levels and values for all
factors
 Identify control factors that may interact
STEPS CONTD…
 Select the orthogonal arrays and the required linear
graph
 Assign factors and interactions to columns
 Conduct the experiment
 Analyze the data
 Interpret the results
 Select optimum levels of significant factors
 Predict expected results
 Run a confirmation experiment
NOMENCLATURE OF ARRAYS
L - Latin square
a - no of rows
La(bc) b - no of levels
c - no of columns (Factors)
Degrees of freedom- a-1
2-level series 3 -level series 4 -level series Mixed -level
L4(23) L9 (34) L15 (45) *L18 (21,37)
L8 (27) L27 (313) L64 (421) L36 (211,312)
**L12(211) L81 (340) - or
L16 (215) - - L36 (23,313)
L32 (231) - - -

*Interactions cannot be studied


**Can study 1 interaction between the 2-level factor and one
3-level factor
L8(27) ORTHOGONAL ARRAY
COLUMNS

EXPT 1 2 3 4 5 6 7

1 1 1 1 1 1 1 1

2 1 1 1 2 2 2 2

3 1 2 2 1 1 2 2

4 1 2 2 2 2 1 1

5 2 1 2 1 2 1 2

6 2 1 2 2 1 2 1

7 2 2 1 1 2 2 1

8 2 2 1 2 1 1 2

NOTE :
•Eight experimental runs and Balanced number of 1s and 2s
•Any pair of columns have only four combinations (1,1); (1,2); (2,1); (2,2)
•If the same number of combinations occur, then the columns are orthogonal
•In the L8, any pair of columns is orthogonal
•L8 can be applied to 7 or less factors
LOCATION OF INTERACTIONS
1. LINEAR GRAPHS
 Taguchi devised this technique
 Graphic representation of Interaction information
in a matrix experiment
 Helps to assign main factors and interactions to
the different columns of an OA

2. TRIANGULAR TABLES
 Each OA has a set of linear graphs and a
triangular table associated with it
EXAMPLE:
LINEAR GRAPH OF THE L8 OA
A 1
Main factors A,B,C, and D
are assigned to columns
3 5 1,2,4 and 7

B C Interactions AB, AC and BC


should be assigned to
2 6 4 columns 3, 5 and 6

D
7
TRIANGULAR TABLE
These tables give all the possible interacting
column relationships that exist for a given OA
Column no. 2 3 4 5 6 7
1 3 2 5 4 7 6
2 1 6 7 4 5
3 7 6 5 4
4 1 2 3
5 3 2
6 1

L8 TRIANGULAR TABLE
SELECTION OF OA
 Determine the df. Required

 Note the levels of each factor and decide the type of OA


(2-level or 3-level)

 Select the particular OA which satisfies the following


conditions
 df(OA) >= df required for the experiment
 Draw the required graph
 Compare with the standard linear graph of the chosen
OA
 Superimpose the required LG on the standard LG to find
the location of factor columns and interaction columns
The remaining columns (if any) are left out
 Draw the layout indicating the assignment of factors and
interactions
The rows will indicate the no of experiments (trials) to
be executed
COMPUTATION OF DEGREES OF
FREEDOM
DEGREES OF FREEDOM
 Maximum no of independent pair wise comparison

 Df. for each factor with ‘a’ levels = a-1

 Df. of an interaction = product of df of interacting

factors
for factor ‘A’ with ‘a’ levels and factor ‘B’ with ‘b’
levels
 Df. for an experimental design = sum of df’s of

factors and interaction


 Df. available in an OA = no of trials-1

for a L8 OA df = 8-1 = 7
SELECTION OF OA-AN EXAMPLE
An experiment has to be conducted with 4 factors
(A,B,C and D) each of two levels. Also, the interactions
AB, AC and AD are to be satisfied
1. DEGREES OF FREEDOM
FACTOR LEVELS DF.
A 2 2-1 = 1
B 2 2-1 = 1
C 2 2-1 = 1
D 2 2-1 = 1
AB (2-1) (2-1) = 1
AC (2-1) (2-1) = 1
AD (2-1) (2-1) = 1
TOTAL Df. = 7
2. Levels of factors – All at 2-levels
therefore choose 2-level OA
3. Selection of required OA
1. The OA which satisfies the required df is OA, L8
2. Interactions required = 3

Interactions possible in L8 = 3
Therefore the best would be L8

4. Required linear graph


B
A C

D
5. Standard LGS for L8 OA
1 3 2

3 5 1 5 4
6
2 7
4
6 (B)
(A)
B
2
6. Superimpose the 3 AB
required LG with the AC 5 C
A
standard LG 4
1
Linear Graph (B) is 6 AD D
similar to the 7
required LG
7. DESIGN LAYOUT
TRIAL FACTORS
NO. A B AB C AC AD D RESPECTIVE
Y
1 2 3 4 5 6 7

1 1 1 1 1 1 1 1 X X
2 1 1 1 2 2 2 2 X X
3 1 2 2 1 1 2 2 X X
4 1 2 2 2 2 1 1 X X
5 2 1 2 1 2 1 2 X X
6 2 1 2 2 1 2 1 X X
7 2 2 1 1 2 2 1 X X
For conducting the experiment test sheet may be prepared without the
8 2 2 1 2 1 1 2 X X
interacting columns
Interactions are dependent on the main factors and hence cannot be
controlled during experimentation
CONDUCTING THE EXPERIMENT
1. Test sheet
2. Randomization
1. The order of performing the tests should be random
2. Randomization protects the experiment from any
unknown and uncontrolled factors that may vary
during the entire experiment and which influence
the result
3. Two methods of randomization
1. COMPLETE RANDOMIZATION

2. SIMPLE REPETITION
COMPLETE RANDOMIZATION
 Trials are selected randomly
 For repetition, each trial is selected randomly in each
repetition
 Used when change of test setup is inexpensive and easy

SIMPLE REPETITION
 Trials are selected randomly and is repeated for required
nos.
 Used when test setup changes are costly
DATA ANALYSIS
ANALYSIS OF DATA

APPLICATION OF L8 OA
EXPT. A B C D E F G RESPONSE AVG.
1 2 3 4 5 6 7
1 1 1 1 1 1 1 1 5.0 5.1 5.5 4.7 5.08
2 1 1 1 2 2 2 2 6.6 6.8 6.7 6.5 6.65
3 1 2 2 1 1 2 2 5.5 5.3 5.1 5.0 5.23
4 1 2 2 2 2 1 1 7.0 7.3 7.5 7.4 7.30
5 2 1 2 1 2 1 2 6.1 6.3 6.1 6.0 6.13
6 2 1 2 2 1 2 1 2.8 2.5 3.0 3.5 2.95
7 2 2 1 1 2 2 1 5.7 5.4 5.2 6.0 5.58
8 2 2 1 2 1 1 2 6.1 6.3 6.1 6.0 6.13
AVERAGE Y 5.63
DATA ANALYSIS USING ‘R’-GRAPH
RESPONSE TABLE FOR FACTOR EFFECTS

A B C D E F G
LEVEL 1 6.06 5.20 5.86 5.50 4.84 6.15 5.23
LEVEL 2 5.19 6.05 5.40 5.76 6.41 5.10 6.03
DIFF. 0.87 0.86 0.46 0.26 1.57 1.06 0.81
RANK 3 4 6 7 1 2 5

RESPONSE
GRAPH
PREDICTING OPTIMUM CONDITION

For minimization of response the optimum condition is given


by A2 B1 C2 D1 E1 F2 G1

For predicting optimum response (µ) select half the number


of factor effects in the ranking order
E1 F2 A2 B1

µ = Avg. Y+(E1- Avg. Y) +(F2- Avg. Y) +(A2- Avg. Y) +


(B1- Avg. Y)
= (E1 + F2 + A2 + B1) - 3 Avg. Y
= 4.84+5.10+5.19+5.20-3*5.63
= 3.44
DATA ANALYSIS BY ‘ANOVA’
COMPUTATION OF “SS”
SSTotal = 5.02+5.52+…………+6.102+6.02- (180.1)2/32
= 49.56
SSFactor= (T1-T2)2/N
T1= Response total at level 1
T2= Response total at level 2
N = Total no of observations
SSA = (A1-A2)2/32 = (97.04-83.16)2/32 = 6.04
SSB = 5.87 SSC = 1.67 SSD = 0.53
SSE = 19.69 SSF = 8.93 SSG = 5.20
SSError = SSTotal – ‘SS’ of all factors
= 1.66
ANOVA (Initial)
Source SS df V F S’ C (%)
A 6.04 1 6.04 87.43 5.97 12.04
B 5.87 1 5.87 84.93 5.80 11.69
C 1.67 1 1.67 24.11 1.60 3.22
D .53 1 .53 7.61 0.46 .92
E 19.69 1 19.69 285.07 19.62 39.58
F 8.93 1 8.93 129.24 8.86 17.87
G 5.20 1 5.20 75.30 5.13 10.35
ERROR 1.66 24 .07 2.14 4.32
TOTAL 49.56 31 49.56 100.00

V = Variance = SS/df; S’ = Pure sum of squares


S’ = SS (Factor) – df (Factor) * Ve (Factor)
S’A = SSA - dfA * Ve = 6.04-1(1.07)=5.97
S’e = SSTotal –(S’A + S’B +……..+ S’G)
C = Contribution – S’ / SSTotal
F0.05,1,24 = 4.26 POOLING OF ‘SS’
All factors are significant. This result is not useful for engineering
purpose
To avoid over estimation, use half the ‘df’ of OA (here 3 or 4
effects)
Pool the ‘SS’ of factors starting with the smallest ‘SS’ into the
error term

ANOVA (Final)
SOURCE SS df V F S’ C RANK
A 6.04 1 6.04 18.3 5.71 11.52 3
B 5.87 1 5.87 17.8 5.54 11.18 4
E 19.69 1 19.69 59.7 19.36 39.06 1
F 8.93 1 8.93 27.1 8.60 17.35 2
ERROR 9.03 27 0.33 - 10.37 20.92 -
TOTAL 49.56 31 - - 49.56 - -

POOLED SSE approx.< 50% of SSTotal for Half ‘df’ of OA


OPTIMUM FACTOR SELECTION FOR
INTERACTIONS
EXP A B AB C AC D E RESULT
1 1 1 1 1 1 1 1 2.23
2 1 1 1 2 2 2 2 3.25
3 1 2 2 1 1 2 2 2.10
4 1 2 2 2 2 1 1 1.93
5 2 1 2 1 2 1 2 3.28
6 2 1 2 2 1 2 1 2.60
7 2 2 1 1 2 2 1 3.18
8 2 2 1 2 1 1 2 4.10

RESPONSE TABLE OF FACTOR EFFECTS

A B AB C AC D E
LEVEL 1 2.38 2.84 3.19 2.70 2.76 2.88 2.48
LEVEL 2 3.29 2.83 2.48 2.97 2.91 2.78 3.18
DIFF. .91 .01 .71 .27 .15 .10 .70
RANK 1 7 2 4 5 6 3
INTERACTION BREAKDOWN
INTERACTION B1 B2
A1 2.74 * 2.01
A2 2.94 3.64
* Avg. effect of A1B1 = (2.23+3.25)/2 = 2.74
SIGNIFICANT EFFECTS = Half the df of OA = 3*(Rank 1,2 and 3)
= A, AB, E
Assuming smaller-The better quality characteristic
The levels for the significant effects are A1, A1B2, E1

µ Pred = Avg.Y+(Avg.A1 - Avg.Y) + [(Avg.(A1B2) - Avg.Y) -(Avg.A1- Avg. Y)

- (Avg.B2- Avg.Y)] + (Avg.E1- Avg.Y)


= Avg.A1 + Avg.(A1B2) - Avg.A1 - Avg.B2 + Avg.E1
= Avg.(A1B2) - Avg.B2 + Avg.E1
= 2.01 – 2.83 + 2.48 = 1.66
ANOVA FOR ATTRIBUTE DATA (0,1 data)
EXPERIMEN A B AB C AC AD D NO OF CASTINGS
T
1 2 3 4 5 6 7 CRACKED GOOD

1 1 1 1 1 1 1 1 4 20
2 1 1 1 2 2 2 2 1 23
3 1 2 2 1 1 2 2 1 23
4 1 2 2 2 2 1 1 2 22
5 2 1 2 1 2 1 2 4 20
6 2 1 2 2 1 2 1 4 20
7 2 2 1 1 2 2 1 0 24
8 2 2 1 2 1 1 2 0 24
TOTAL 16 176
BAD : 1 GOOD : 0
Thus the data of 1st expt. is considered as four 1s and twenty 0s
SST = 12+12+12+12+02+02+02 +……+02 –(162/192)
Since T = sum of all data 12 = 1; 02 = 0
SST = T – (T2/N) = 16 – (162/192) = 14.667
NOTE: It is analyzed treating the defects as response
Sum of squares of columns are computed as
with variable data
SSA = (8-8)2/192 = 0 SSB = (13-3)2/192 = 0.521
SSC = (9-7)2/192 = 0.021 SSD = (10-6)2/192 = 0.083
SSAB = (5-11)2/192 = 0.188 SSAC = (9-7)2/192 = 0.021
SSAD = (10-6)2/192 = 0.083
SSE = SST - SSCOLUMNS = 13.75
ANOVA
SOURCE SS df MS F
A 0.0 1 0.0
B 0.521 1 0.521 6.94 S
C 0.021 1 0.021
D 0.083 1 0.083
AB 0.188 1 0.188 2.50 N.S
AC 0.021 1 0.021
AD 0.083 1 0.083
E 13.750 184 0.075
TOTAL 14.667 191

You might also like