Sie sind auf Seite 1von 14

72

CHAPTER 4

WEIGHT-BASED DESIRABILITY METHOD TO


SOLVE MULTI-RESPONSE PROBLEMS

4.1 INTRODUCTION

Optimizing the quality of a product is widespread in the industry.


Products have to be manufactured such that they best fit some quality
properties. Varying the product settings leads to different product qualities
and the aim of the manufacturer is to find the factor settings that
simultaneously optimize the quality properties.

The classical approach to solve such optimization problem is based


on response surface methodology. First, a designed experiment is used to
collect data and to adjust models capturing the relationship between the
responses of interest and the factor settings. Those fitted models can then
predict the quality properties for any design point of the experimental domain.
Secondly, a desirability index is built to combine the predicted properties into
a value belonging to the [0; 1] interval. This index provides a ranking of
possible factor settings in the solutions space and the optimum can be found
by an adequate optimization algorithm. But model predictions are suited with
error; so are the desirability index and the optimal solution found. In practice,
in the related literature and design of experiment software, this error is
neglected.
73

4.2 DESIRABILITY APPROACH

The desirability function approach is one of the most widely used


methods in industry for dealing with the optimization of multiple response
problems. It is based on the idea that the quality of a product that has multiple
quality characteristics is completely unacceptable if one of the characteristics
lies outside the desired limits. This method assigns a score to a set of
responses and chooses factor settings that maximize that score.

Desirability function approach was first proposed by Harrington


(1965) and then revised by Derringer and Suich (1980). The multi-response
optimization problem has been addressed by several authors by using
different expressions of the desirability function. The method finds operating
conditions xl... xk that provide the "most desirable" response values. For each
response Y1, a desirability function di assigns numbers between 0 and 1 to the
possible values of Y1, with di=0 representing a completely undesirable value
of x1…xk, and di=1 representing a completely desirable or ideal response
value. The individual desirability is then combined using an aggregation
criterion such as a geometric mean, which gives the overall desirability (D).
Different forms of the desirability function are proposed according to the
nature of the optimization problem: minimization, maximization or tracking a
target. For each case, a different expression is proposed and shown in
Figure 4.1.

The desirability of the response increases as it becomes closer to its


target value Ti. It reaches the maximum value of 1 only if the response value
is equal to the target T. The overall desirability is then given by the
desirability index which is the geometric mean of the individual desirability.
It is noticed that the definition of the desirability function does not depend on
any distribution assumption. In what follows the nonconformity ratio is
considered as a response variable and the property that the desirability value
74

increases when the response becomes closer to its target is used to make the
”higher the better” rule hold for any type of distribution and for any type of
specification limit.

There are many statistical techniques for solving multiple response


problems like overlaying the contour plot for each response, constrained
optimization problems and desirability approach. The desirability method is
recommended due to its simplicity, availability in the software and provides
flexibility in weighting and giving importance for individual response.
Solving such multiple response optimization problems using this technique
involves using a technique for combining multiple responses into a
dimensionless measure of performance called the overall desirability function.
The desirability approach involves transforming each estimated response, Yi,
into a unit less utility bounded by 0 < di < 1, where a higher d i value indicates
that response value Yi is more desirable, if di=0; this means a completely
undesired response. The individual desirability function for the various
responses under consideration is then incorporated into a single function
which gives the overall assessment of the desirability of the combined
response. Then, the solution x is found that maximizes the geometric mean of
the individual response desirability. There are three categories of response,
and the desirability function can be defined as follows.

The combined desirability can be calculated by using the equation

D = [d 1w1 d2w2 d3w3 d 4w4] / Σwi (4.1)

where D is the overall desirability index and wi represents weight assigned to


each response. Here weights are taken from Eigen values of responses.
75

4.3 DESIRABILITY FUNCTIONS AND DESIRABILITY INDEX

The concept of desirability was introduced by Harrington (1965) to


provide a solution to multi-response optimization problems. It allows to
balance the optimized properties Yi’s against one another, taking into account
their target value, their relative importance and their scale.

Harrington proceeds in two steps First, each response Yi is


transformed to the same scale using a desirability function, denoted by dj,
such that di(Yi)[0,1]. If d i(Yi)=0, the product is not at all acceptable
according to the specifications of the ith property and if di (Yi) =1, the product
fullfils them perfectly. The most well-known desirability functions are those
of Harrington (1965) j based on the exponential function of a linear
transformation of the yi’s and those of Derringer and Suich (1980) based on
power of a liner transformation of the Yi. Gibb et al. (2001) and Govaerts
et al (2005) proposed smoother and differentiable desirability functions based
on the logit function normal density and normal distribution functions. These
three types of desirability functions are presented in Table 4.1 for the cases
where the response must be maximized, minimized or reach a target value.
The target value T and the parameters ‘a’ and ‘b’ have to be adjusted
according to the cumulative distribution function of the standard normal.
76

Table 4.1 Different types of desirability functions

Maximum Minimum Target Value


n
 YT 
Harrington (1965) exp (-exp(-a-bY)) 1-exp(-exp(-a-bY) exp   
 b 
0 if Y<a
0 if Y<a 1 if Y>b
if a1<Y<T
Derringer and Suich  Y  a s aY
s

  if a<Y<b   if <Y<a  a 2  Y s
(1980)  ba   ab    if T < Y<a2
 a2  T 2
1 if Y>b 0 if Y<a
0 if Y>a2

 Y  a1  Y  a2
Le Bailly and  Ya  Ya     (1   ( ))
(   1    b
 1  b2
Govaerts (2005)  b   b 

4.4 OPTIMIZATION STEPS USING DESIRABILITY


FUNCTION METHOD

4.4.1 Determining the Optimal Parameter Combination

Step 1: Calculate S/N Ratio for the corresponding responses using


the following formula.

4.4.1.1 The Nominal-the-Best (NTB)

When the value of y equals T (the target value), the desirability


value equals 1; if the departure of y exceeds a particular range from the target,
the desirability value equals 0, and such a situation represents the worst case.
The desirability function of the-nominal-the-best scenario can be written as:
77

a1
Yi  Li
, Li  y  Ti
Ti  Li

a2
di(Y1 )  Yi  Ui
, Ti  y  Ui
Ti  Ui
(4.2)
0, Y < Li, or , y>Ui

4.4.1.2 The Larger-the-Better (LTB)

When the value of y exceeds a particular criterion value, which can


be viewed as the required value, the desirability value equals 1; if the value of
y is less than a particular criterion value, which is unacceptable, the
desirability value equals 0. The desirability function of the-larger-the-best
scenario can be written as,

0 , yˆ i  Li
 a2
 Yi  Ui
di(Y1 )   , Li  yˆ i  Ui (4.3)
 Ti  Ui
1 , yˆ i  U i

4.4.1.3 The Smaller-the-Better (STB)

When the value of y is less than a particular criterion value, the


desirability value equals 1; if the value of y exceeds a particular criterion
value, the desirability value equals 0.

The desirability function of the-smaller-the-best can be written as,

0 , yˆ i  Li
 a2
 Yi  Ui
di(Y1 )   , Li  yˆ i  Ui (4.4)
 Ti  Ui
1 , yˆ i  U i

78

where Li is ith lower limit and Ui is ith upper limit, Ti is the ith target of the
response and yi is the ith response and a1, a2 represent the weight of the
response.

Step 2: The individual desirability index value has been calculated


using the formula as di= [(Yi-a)/ (b-a)]r Here the value of r = 0.5 .

Step 3: The overall desirability (D) has been computed using the
formula as,

D = [∑ di .wi]/ ∑wi . Here, the value of wi is taken based on Eigen


value.

Step 4: Then the mean desirability index can be calculated and also
optimal combinations will be chosen. The weight based DM can be calculated
as,

WBDM = D1ij* weight 1 +D2ij * weight 2 (4.5)

Step 5: ANOVA can be determined from the desirability index


value in order to find the significance.

4.5 DETERMINING THE OPTIMAL COMBINATION FOR


WEDM OPERATIONS BY USING EXISTING
DESIRABILITY INDEX METHOD

The normalized values are taken from Table 3.4, and then the
individual desirability index value has been calculated using the formula
di= [(Yi-a)/ (b-a)]r and given in Table 4.2. The overall desirability (D) has
been computed using the formula D = [∑ di .wi]/ ∑wi. Here, the value of wi is
arbitrarily taken as 0.5. The mean desirability value for each factor has been
worked out using the L18 OA and the results are given in Table 4.3 and the
factor effects are shown in Figure 4.1. From the main effects the optimal
79

combination by the existing method using desirability index is A1 B1 C1 D3 E2


F2 G1.

Table 4.2 Desirability index and overall desirability value for Case-1

Exp.
Normalized S/N Ratio d1 d2 d3 D
No.
1 0.769256 0.929138 0.263298 0.877072 0.963918 0.513126 0.784705
2 0.326742 0.819085 0.792619 0.571614 0.905033 0.890292 0.788979
3 0 1 0.405736 0 1 0.636974 0.545658
4 0.429902 0.795559 0.956122 0.655669 0.891941 0.977815 0.841808
5 0.430627 0.331032 0.552614 0.656222 0.575354 0.74338 0.658319
6 0.793537 0.933624 0.329279 0.890807 0.966242 0.573828 0.810293
7 0.436857 0.573658 0.531833 0.660952 0.757402 0.729269 0.715874
8 1 0.883847 0.394641 1 0.940131 0.628205 0.856112
9 0.858632 0.547791 0.945295 0.926624 0.740129 0.972263 0.879672
10 0.793537 0.942572 1 0.890807 0.970861 1 0.953889
11 0.059665 0.942572 0.895497 0.244264 0.970861 0.946307 0.720478
12 0.584127 0.888412 0.511257 0.764282 0.942556 0.715022 0.807287
13 0.563768 0.68451 0.508334 0.750845 0.827351 0.712975 0.763724
14 0.756435 0.955938 0.206908 0.869733 0.977721 0.454871 0.767442
15 0.266994 0.915633 0.508334 0.516715 0.956887 0.712975 0.728859
16 0.989613 0.833099 0.857185 0.994793 0.912743 0.925843 0.944459
17 0.792101 0.752668 0 0.890001 0.867564 0 0.585855
18 0.710694 0 0.439379 0.843027 0 0.662857 0.501961

Table 4.3 Mean desirability value for Case-1

Levels
Parameters Max-Min
1 2 3
A 0.764602 0.752662 - 0.01194
B 0.766833 0.761741 0.747322 0.019511
C 0.834077 0.729531 0.712288 0.121789
D 0.756896 0.706024 0.812976 0.106952
E 0.716792 0.827918 0.731186 0.111126
F 0.767996 0.788959 0.718941 0.070018
G 0.806981 0.763958 0.704957 0.102024
80

0.95
0.9

Mean Desirability Index


0.85 A
0.8 B
C
0.75
D
0.7
E
0.65 F
0.6 G
0.55
0.5
1 2 3
Levels

Figure 4.1 Factor effects on desirability index for Case-1

Table 4.4 Results of the pooled ANOVA on desirability for Case-1

%
Factors SS Dof MS F
Contribution
C 0.052122 2 0.026061 1.20323698 19.61007103
D 0.034345 2 0.017173 0.79285473 12.92175837
E 0.043828 2 0.021914 1.01176989 16.48958584
Error 0.135497 11 0.012318 50.97858476
Total 0.265792 17

The result of the pooled ANOVA given in Table 4.4 indicates that
all factors have almost equal contribution towards affecting the multiple
quality characteristics.
81

4.6 DETERMINING THE OPTIMAL COMBINATION FOR


WEDM OPERATIONS BY USING THE PROPOSED
WEIGHT-BASED DESIRABILITY METHOD

The normalized values are taken from Table 4.2 and then the
individual desirability index (d) value has been calculated using the formula
di= [(Yi-a)/ (b-a)]r and given in Table 4.5. The overall desirability (D) has
been computed using the formula D = [∑ d i .wi]/ ∑wi. Here, the value of wi is
the Eigen value computed using Principal Component Analysis (PCA). By
applying PCA on WBDM, the optimal weights are obtained. The optimal
weights are 0.3351, 0.317667, and 0.347233. The Mean desirability value for
each factor has been worked out using the L18 OA and the results are given in
Table 4.6. The factor effects are shown in Figure 4.2.

Table 4.5 Normalized S/N values, individual desirability value and


overall desirability values for Case-1

Exp. No Normalized S/N Ratio d1 d2 d3 D


1 0.769256 0.929138 0.263298 0.877072 0.963918 0.513126 0.750277
2 0.326742 0.819085 0.792619 0.571614 0.905033 0.890292 0.771462
3 0 1 0.405736 0 1 0.636974 0
4 0.429902 0.795559 0.956122 0.655669 0.891941 0.977815 0.830637
5 0.430627 0.331032 0.552614 0.656222 0.575354 0.74338 0.657224
6 0.793537 0.933624 0.329279 0.890807 0.966242 0.573828 0.78465
7 0.436857 0.573658 0.531833 0.660952 0.757402 0.729269 0.714159
8 1 0.883847 0.394641 1 0.940131 0.628205 0.834405
9 0.858632 0.547791 0.945295 0.926624 0.740129 0.972263 0.877305
10 0.793537 0.942572 1 0.890807 0.970861 1 0.953
11 0.059665 0.942572 0.895497 0.244264 0.970861 0.946307 0.605996
12 0.584127 0.888412 0.511257 0.764282 0.942556 0.715022 0.798236
13 0.563768 0.68451 0.508334 0.750845 0.827351 0.712975 0.760558
14 0.756435 0.955938 0.206908 0.869733 0.977721 0.454871 0.720755
15 0.266994 0.915633 0.508334 0.516715 0.956887 0.712975 0.70277
16 0.989613 0.833099 0.857185 0.994793 0.912743 0.925843 0.944115
17 0.792101 0.752668 0 0.890001 0.867564 0 0
18 0.710694 0 0.439379 0.843027 0 0.662857 0
82

Table 4.6 Mean desirability index values for WBDM for Case-1

Levels
Factors Max-Min
1 2 3
A 0.691124 0.609492 - 0.081632
B 0.646495 0.742765 0.561664 0.181101
C 0.825458 0.598307 0.52716 0.298298
D 0.627831 0.61694 0.706154 0.089214
E 0.595973 0.813604 0.541348 0.272256
F 0.654688 0.692193 0.604044 0.088149
G 0.791319 0.65434 0.505266 0.286053

0.85

0.8
Mean Desirability Index

A
0.75 B
0.7 C
D
0.65 E
0.6 F
G
0.55

0.5
1 2 3
Levels

Figure 4.2 Factor effects on desirability index for Case-1

Therefore from Table 4.5 and Figure 4.2, the optimal parameter
combination is A1 B2 C1 D3 E2 F2 G1.

Table 4.7 shows that the controllable factors C, E and G contribute


23.63%, 15.04 % and 17.25%, respectively.
83

Table 4.7 Results of the Pooled ANOVA on WBDM for Case-1

%
Factors SS Dof MS F
Contribution

A 0.02999 1 0.029987 0.3143434 1.81174459


B 0.09852 2 0.049262 0.5163966 5.95259026
C 0.39127 2 0.195635 2.0507745 23.6396207
D 0.02842 2 0.014212 0.1489795 1.71731178
E 0.24894 2 0.124471 1.3047815 15.0404345
G 0.28563 2 0.122812 1.2873986 17.2567962
Error 0.57237 6 0.095396 34.5815019
Total 1.65515 17

4.7 COMPARISON OF SOLUTIONS FOR WEDM CASE-1

The initial settings for the WEDM process were A2B1C1D2E2F2G1.


The optimal factor settings based on the selected techniques are A2B1C1D3
E2F2G1 (Grey), A1 B1 C1 D3 E2 F2 G1 (DM) and the combinations for the
proposed methodologies are A1 B2 C1 D3 E2 F2 G1 (WBDM) and A2 B3 C2 D3
E2 F1 G1 (WGBRA).

To find the improvements under the optimum condition, S/N ratios


for all the responses are determined using the additive model. The overall
improvement percentage is calculated as the ratio between sum of the
improvement values of all the responses and sum of S/N ratios of initial
conditions of all responses. Table 4.8 presents the comparison of results.
From Table 4.8, it is seen that the results from the weight-based Desirability
method and weight-based GRA have shown improvements of 2.053 % and
3.18 %, respectively, from the initial condition. It is also seen that the results
obtained on the existing methods are compared with the solutions obtained
84

with the proposed methods. It is also to be noted that the techniques discussed
in this research have used arbitrary method in assigning weights to the
response variables. Whereas in the proposed weight-based desirability method
and weight-based Grey relation analysis, the weights are scientifically derived
using Principal Component Analysis (PCA).

Table 4.8 Comparison of solutions for Wire EDM process parameters

Grey Weight-Based
Desirability
Relation Desirability Grey Relation
Responses Initial Method
Analysis Method Analysis
(WBDM) (WBGRA)
MRR 17.02362 18.201233 18.20123333 18.49808 19.4093
SF 89.00231 90.939333 90.93933333 90.38083 90.782972
KERF 11.11263 10.160816 10.16081667 10.66508 10.672883
Optimal Setting A2 B1C1D2 A2B1C1D3 A1 B1C1 D3 A1B2C1D3 A2B3C1 D3
E2F2G1 E2F2G1 E2 F2 G1 E2F2 G1 E2F1 G1
Improvements in SN ratios
MRR 1.1776133 1.177613333 1.4744633 2.38568
SF 1.9370233 1.937023333 1.3785233 1.780662
KERF -0.9518133 -0.95181333 -0.4475466 -0.439747
Overall Improvement (%) 1.846 1.846 2.053 3.18

4.8 DISCUSSION

This chapter has presented the use of desirability function for


optimizing the multi-response problems in Taguchi method by comparing the
existing method with weight-based desirability method. The method of
averaging for obtaining the overall desirability value has been done
scientifically; therefore, human judgment has been completely eliminated.
Thus, the weight-based desirability method has improved the solution. Three
85

case studies are solved through this methodology and results are presented
and discussed in chapter 5. The proposed procedure has the following merits:

1. The optimal solution can be easily obtained, since desirability


can effectively resolve problems that have incomplete
information.

2. It does not have any complicated mathematical theory or


computation and can be easily understood by the engineers
having limited knowledge of statistics.

Das könnte Ihnen auch gefallen