Anda di halaman 1dari 7

q SHREE DASRAM MAHARAJ PRASANNA q

TAGUCHI TECHNIQUES – A REVIEW

Mr. Kelkar Satej Sudhakar


Assistant Professor
Dept of Mechanical Engg.
Jaywantrao Sawant College of Engg. Hadapsar, Pune-28
Satej_kelkar@yahoo.com [ Contact No. : 9822029364 ]

ABSTRACT

While applying an analytical or a classical design optimization technique, one has to


prepare a mathematical model for the problem. The mathematical model is in form of a set of
ordinary and/or partial differential equations expressing relations between the design
variables and the parameters. If the design variables and the parameters are such that they
cannot be related mathematically then a mathematical model cannot be prepared and none of
the analytical or classical design optimization techniques can be applied for solving the
problem. This is a major limitation of analytical or classical optimization techniques.
The problems for which a mathematical model can not be prepared, are solved with
the help of experimental optimization and designing techniques. Experimental techniques are
also used for studying the effect of different design parameters separately, for confirming the
theoretical approach & calculations and for assuring whether target values have been
achieved or not.
Taguchi Techniques are Fractional Factorial experimental design techniques. Taguchi
Techniques use standard ‘Orthogonal Arrays’ for forming a matrix of experiments in such a
way as to extract maximum important information with minimum number of experiments.
Using Taguchi techniques, number of parameters (each may have number of levels) can be
tested at a time with probably least number of experiments as compared to any of the other
experimental optimization techniques and still the technique provides all the necessary
information required for optimizing the problem. The main advantage of Taguchi Techniques
is not the smallest number of experiments but one can find the best level of each parameter
and share of each parameter towards the problem separately. With some modifications the
same techniques can also be used for achieving sensitivity robustness in the design.
In this paper a review of Taguchi Techniques have been conducted.

Keywords : Taguchi Techniques, Robust Design, Optimum Design.

1. INTRODUCTION
If a design optimization problem is such that, the design variables and the parameter
can not be related mathematically i.e. a mathematical model cannot be prepared for the
problem, then any of the analytical or classical optimization techniques cannot be applied for
solving the problem. This is a major limitation of analytical or classical optimization
techniques.
e.g. Frequently during the volume production of vehicles, bolts that are tightened with
a specific torque have no defined preload. (Probable causes for this problem are the
differences in the friction values and settling phenomenon on coated surfaces). If a design
optimization problem is formulated with an objective of maximizing the preload on the bolts
then the factors like surface finish on the parts, whether lubricant is used or not while making
the joint, which side whether the bolt or the nut is tightened, etc. seems to be controling the
preload induced in the bolts and so these factors are identified as design parameters. Of
course these parameters can not be related mathematically to formulate a mathematical model
for this optimization problem and none of the analytical or classical optimization techniques
is applicable for this problem and a suitable experimental technique it required to be applied
for solving the problem.

2. TAGUCHI TECHNIQUES
Taguchi techniques are experimental design optimization techniques which use
standard ‘Orthogonal Arrays’ for forming a matrix of experiments in such a way as to extract
maximum important information with minimum number of experiments. Besides minimum
number of experiments required (as compared to any other experimental technique), the
advantage of Taguchi Techniques lies in the fact that, one can find the best level of each
parameter and share of each parameter towards the problem separately and also with some
modification the same techniques can be used for achieving sensitivity robustness in the
design. Sensitivity Robustness is achieved in optimum designs by maximizing ‘Signal to
Noise Ratio’.
For understanding the procedure let us solve the same problem stated earlier, that of
bolts which are tightened with a specific torque have no defined preload.

2.1 Step 1 : Problem Definition and Formulating Measurable Target Quantity


Preload on the bolt cannot be measured easily and readily so to convert the target
quantity of preload into a readily measurable quantity one can proceed as follows
The length of the bolt is measured while the bolt is unstressed. After tightening the
bolt with a specified torque, again the length is measured the relative change in the length
reflects the preload. As the relative change in the length is too small (but readily and
accurately measurable) it is multiplied by 100 while entering the reading in the column of
results and thus the problem is redefined in form of a measurable target value as “ Maximize
the relative change in the length of the bolt while it is tightened with a specified torque”

2.2 Step 2 : Problem Analysis and Reduction of Parameters


At first number of parameters are listed which are probably responsible for the
problem. As hardware is available (i.e. the product is already in existence) using a techniques
called as ‘Shainin Technique of Homing In’ the number of parameters listed earlier is
reduced and seven possibly predominant parameters (each with two levels) are finalized. The
parameters are as follows…
Figure 1. Model configuration for Bolt tightening experiment

Factor Factor Specification Factor Level 1 Factor Level 2


No.
1 A : Surface of Part 3 (Top Side) A1 : Polished A2 : Painted
2 B : Tightening Side B1 : Bolt B2 : Nut
3 C : Lubricant C1 :Yes C2 : No
4 D : Surface of the Nut D1 : Phosphatised D2 : Zinc Coated
5 E : Surface of Part 5 (Bottom Side) E1 : Painted E2 : Polished
6 F : Surface of the Bolt F1 : Zinc Coated F2 : Phosphatised
7 G :Surface of Part 3 (Bottom Side) G1 : Polished G2 : Painted

2.3 Step 3 : Planning and Carrying Out Experiments


The above seven parameters with 2 levels each will otherwise (with full factorial
experimental technique) require 27 = 128 number of experiments. If the smallest suitable
Orthogonal Array as suggested by Taguchi is used to plan the conduction of experiments it
will require minimum number of experiments as compared to any other experimental
techniques. For this particular problem L8 Orthogonal Array (Refer Table 1) is suitable which
requires only 8 number of experiments to extract all the necessary information.

Factor
Combination A B C D E F G
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
Table 1. Layout of Experiments Using L8 Orthogonal Array
Now refer say 4th factor combination (row) of the Orthogonal Array which is 1222211
which means A1, B2, C2, D2, E2, F1 and G1 should be the parameter levels for the 4th
experiment. This means the 4th experiment should be carried out with surface of part number
3 polished, tightening from the nut side, no use of lubrication, zinc coated surface of bolt,
polished bottom surface of part 5, zinc coated surface of bolt and polished bottom surface of
part 3. For more accuracy 4 experiments were carried for each of the 8 parameter
combination and results were recorded (Refer Table 2). Here results mean the relative change
in length of the bolt multiplied by 100. While performing the experiments they were
performed at a random sequence.
Factor
Combination
ABCDEFG Expt No. 1 Expt No. 2 Expt No. 3 Expt No. 4 Total
1 1111111 8.26 7.70 7.20 7.27 30.43
2 1112222 9.93 13.10 10.94 10.07 44.04
3 1221122 8.96 7.67 10.17 9.07 35.87
4 1222211 6.33 5.27 7.04 4.53 23.17
5 2121212 8.90 9.84 8.70 9.00 36.44
6 2122121 9.23 9.00 10.70 9.74 38.67
7 2211221 10.06 10.27 9.90 9.76 39.99
8 2212112 9.63 8.93 7.33 6.70 32.59
Grand
281.20
Table 2. Results of Bolt Tightening Experiments Total

2.4 Step4 : Experimental Analysis, ANOVA and Optimization of the Results

2.4.1 Experimental Analysis


Refer Table 2, it can be noted that experiment number 1, 2, 7 and 8 are performed
setting parameter C at level C1 and experiment number 3, 4, 5 and 6 are performed setting
parameter C at level C2. Similarly experiment number 1, 2, 3 and 4 are with parameter A at
level A1 and experiment number 5, 6, 7 and 8 are with parameter A at level A2

For comparing effect of A1 with A2, procedure followed is as follows.


First of all Total result or sum of results for a particular level setting is calculated as
below
∑ A1 = 30.43 + 44.04 + 35.87 + 23.17 = 133.51
∑ A2 = 36.44 + 38.67 + 39.99 + 32.59 = 147.69
Now ∑ A1 and ∑ A2 are divided by 4 x 4 = 16 (four factor combination and four
experiments for each combination) which gives means change in lengths for a particular level
setting A1 and A2

A1 = 133.51 / 16 = 8.34
A2 = 147.69 / 16 = 9.23
Thus Total result (or sum of results) and mean change in length is calculated for rest
of the parameter levels as tabulated in Table 3. Table 3 provides effects of individual factor
levels of the bolt tightening experiment.
The above results show that the preload increases from 8.34 to 9.23 if instead of A1
the parameter A is set at level A2 (i.e. if the top side of part 3 is painted instead of kept
polished). Thus individual parameter levels which will result in optimization (i.e.
maximization of preload) can be determined.
Factor Level Total Mean Change in length
A1 133.51 8.34
A2 147.69 9.23
B1 149.58 9.35
B2 131.62 8.23
C1 147.05 9.19
C2 134.15 8.38
D1 142.73 8.92
D2 138.47 8.65
E1 137.56 8.60
E2 143.64 8.98
F1 122.63 7.66
F2 158.57 9.91
G1 132.26 8.27
G2 148.94 9.31

Table 3. Effect of Individual Factor Level

2.4.2 ANOVA (Analysis of Variances)


Different factors or the parameters affect the result to a different degree. The relative
magnitude of the parameter effect can be judged from Table 3 which gives average value of
result for each factor level. A better feel for the relative effect of the different factors on the
target value can be obtained by decomposition of variance which is called as ‘Analysis of
Variance’ (ANOVA). The total variance in the result, is credited among the parameters with
the help of statistical methods which also helps in calculating the percentage significance of
each factor. Obviously a factor scoring more percentage should be taken care of on priority
basis (some times some unavoidable constraints of time, money, etc. do not permit
optimization of each and every parameters and some of the parameters should be left as they
are to their current levels. ANOVA helps to select the parameters which are badly need
optimization and the parameters which can be left as they are which do not add much
variance and score less percentage of significance).

For the problem, results of ANOVA are as tabulated in Table 4


Sum of Variance Variance ratio
Factor DOF (f)
Squares (s) (V) F- Value
A 1 6.28 6.28 6.83
B 1 10.08 10.08 10.96
C 1 5.2 5.2 5.65
D 1 0.47 0.47 0.51
E 1 1.16 1.16 1.26
F 1 40.37 40.37 43.88
G 1 8.69 8.69 9.45
Error e1 -- -- -- --
Repetition
24 22.17 0.92 1
error (e2)
Correction
1
factor (CF)
Total 32

Table 4. ANOVA Table for Bolt-Tightening Experiment


During Analysis of Variance (ANOVA), F-test is conducted for each of the
parameters. (F-test compares quality of two variances). F value is calculated by taking the
ratio of variance of a particular parameter to variance of error e2 (e2 is error from repetition
to repetition due to unavoidable and unpredictable chance causes). If value of F (i.e. the ratio
of variance) for a particular parameter is greater than the critical value of F (which is referred
from a standard table), then it means, variance added by the parameter to the total result is
significant. It means parameter definitely adds variance to the total result and variance due to
parameter is not merely due to different unavoidable and unpredictable errors and/or chance
causes. It also means the parameter significantly affects the target value and should be taken
care of.
Now Degree of Freedom (DOF) for each parameter is 1 and for e2 it is 24
Using Standard F table it can be noted that
F124 (95%) = 4.26
F124 (99%) = 7.82
Comparing F-values for different parameters (Refer Table 4) and above noted F-
values from standard table it can be concluded that, for parameter B, F and G the F-value is >
F124 (99%). This means probability of the parameter to have significant effect on total result is
99% and on similar lines for parameters A and C, the probability is 95%. Parameters D and E
do not have any significant effect on the total result

2.4.3 Optimization of the Results


The criterion for optimization is maximization of preload or in other words
maximization of relative length change of the bolt. So referring Table 3, the factor level of
the two factor levels which gives more relative length change is declared to be optimum level
setting. As parameters D and E are not significant the choice between levels D1 & D2 and
between E1 & E2 is done on the basis of economy (means the cheaper level setting is
declared to be optimum). Thus the problem is not truly the optimum out of the optimums
because true optimum is the one which is with all parameters set at their optimum level.
Thus the optimum factor combination is A2, B1, C1, D1, E1, F2 & G1
Note that this combination is of its own and it doesn’t match with any of the
combinations from the Orthogonal Array table combinations (i.e. rows of the Orthogonal
Array matrix).
Thus after determining the optimum parameter levels for each parameter and
determining the response or result under these conditions an experiment with parameter
setting at said optimum levels is conducted to conform whether an optimum design is
achieved.
After optimizing the problem if it is realized that still improvement in the result is
necessary then a refinement optimiation is conducted in which the current optimum levels of
parameters are taken as starting levels and two, three or more levels assigned to the same
parameters which are set near by the starting level and again optimization is carried out
which gives refinement of the optimum design.

2.5 Robust Optimum Design


Robust optimum design is the optimum design which tolerates the variation in the set
values of parameters. Using Taguchi Techniques it is possible to achieve sensitivity
robustness in a design which means the sensitivity of design towards the variations is reduced
or variations cause the change in target value (objective function) as small as possible.
Taguchi has defined a term called ‘Signal to Noise Ratio’ and formulae for ‘Signal to
Noise Ratio’ based on different conditions have been formulated. If optimization problem is
modified such that instead of optimizing the target quantity aim of the design is kept as to
maximize the ‘Signal to Noise Ratio’ (as applicable for that particular problem), then the
optimum design achieved is sensitivity robust optimum design. Except the target quantity the
procedure followed in robust optimization will be same as that of simple optimization. Thus
this is an added advantage of Taguchi Techniques that with some modification the same
technique is useful for achieving robustness in optimum designs.

3. SUMMARY AND CONCLUSION

In Taguchi Method, the word "optimization" implies "determination of BEST levels


of control factors". Taguchi techniques are fractional factorial experimental optimization
techniques where Orthogonal Arrays are used for forming matrix of experiment. Using
Taguchi Techniques with minimum number of experiments maximum amount of information
is extracted. Using ANOVA the results of the experiments are processed and importance of
each and every single factor is identified also share of each individual parameter towards the
target is identified. Parameter levels which render optimum result can be identified and set of
such optimum parameter levels is the optimum design. With some modifications same
optimizing techniques are applicable for achieving sensitivity robustness in the design

REFERENCES

1. A. Parkinson, “Robust Mechanical Design using Engineering Model”, ASME Journal


of Mechanical Design, June 1995, vol. 117, pp 48 to 55
2. G. Emch & A. Parkinson, “Robust Optimum Design for Worst Case Tolerances”,
ASME Journal of Mechanical Design, Dec. 1994, vol. 116, pp 1019 to 1025
3. Genichi Taguchi, Subir Chowdhury, Shin Taguchi, 2004, Robust engineering, Tata
Mac-Graw Hill, New Delhi .
4. Krottmaier J., 1993, Optimizing Engineering Designs, McGraw-Hill, Singapore.
5. Marcel Proust, 2005, Design of Experiments, SAS Institute Inc., Cary, NC, USA
6. Phadke M. S., 1989, quality engineeringusing robust design, prentice hall, Engle-
wood cliffs, New Jersey.
7. Ranjit K. Roy, 1990, A primer on the Taguchi method,Van Nostrand Reinhold, New
York
8. Ray C. Johnson, 1961, Optimum Design of Mechanical Elements, John Wiley &
Sons, Inc., New York.

Anda mungkin juga menyukai