Anda di halaman 1dari 10

Libardo V.

Vanegas
Lecturer, Mechanical Engineering Department, Universidad Technologica de Pereira, A. A. 97 Pereira, Columbia e-mail: ivanegas@utp.edu.co

Fuzzy Approaches to Evaluation in Engineering Design


As variables in the design evaluation stage are imprecise (fuzzy), tools, such as fuzzy set theory, should be used. This paper presents several fuzzy approaches to design evaluation. Weights of criteria and performance levels are captured by fuzzy numbers, and the overall performance of an alternative is calculated through the new fuzzy weighted average. Different approaches to express and calculate performance are discussed. Two metrics for measuring performance levels per criterion of a solution are proposed. Also, a new type of fuzzy goal is suggested. Finally, a novel way of comparing the overall performance of a design candidate by drawing an aggregate prole of performance is proposed. It is concluded that fuzzy set theory is a powerful and exible tool for dealing with the imprecision in different types of problems in design evaluation. DOI: 10.1115/1.1814639

Ashraf W. Labib

Senior Lecturer, School of Mechanical, Aerospace & Civil Engineering, University of Manchester, P.O. Box 88, Manchester M60 1QD, UK e-mail: ashraf.labib@manchester.ac.uk

Introduction

Engineering design is an important wealth-creating activity in todays companies and infrastructure; however, the process of designing is very complex and not well understood, and the information managed during the conceptual stage is incomplete, imprecise, and vague fuzzy . Within this stage, several design solutions have to be generated and correctly evaluated and selected. In order to select the best design, several criteria, such as cost, customer satisfaction, perceived value, appearance, and ease of use, have to be considered. Each alternative has a performance degree to which an alternative satises a requirement or criterion with respect to each criterion; then, designers have to determine the overall performance of each alternative, taking into account all the criteria and their relative weights relative importance levels . This decision-making process is not straightforward because assigning performance levels and weights is rather difcult, and because the information for doing that is usually imprecise. Conventional evaluation design methods, such as the weighted objectives method, are not exible enough to deal with imprecise information; thus, improved evaluation approaches are necessary, and these may be based on fuzzy set theory. Fuzzy set theory has been applied in design by several researchers. Wood and Antonsson 1 , Antonsson and Otto 2 , Giachetti et al. 3 and Vanegas and Labib 4 are just a few examples. Particularly, fuzzy set theory has been introduced by some researchers in design evaluation 511 . However, the research is in its infancy; while some investigations have very useful proposals, others have failed to treat their problems effectively. This paper reviews and analyzes some applications of fuzzy set theory in engineering design evaluation and mainly aims at making some contributions to current practices and developing new procedures. Different ways in which fuzzy set theory can be applied in design evaluation are discussed. Two metrics are proposed for nding the performance per criterion of a solution. Weights of criteria can be obtained through the Analytical Hierarchy Process AHP 12 and then converted into fuzzy numbers by using a technique called fuzzy line segment 7 . The New Fuzzy Weighted Average NFWA 11 is proposed for nding the overall performance of alternatives. A new approach to express the required performance of an alternative based on linguistic description is suggested. Finally, a novel alternative way of comparing designs through the calculation of an aggregate fuzzy set is proposed. Section 2 presents a short review of applications of fuzzy
1 To whom correspondence should be addressed. Contributed by the Design Theory and Methodology Committee for publication in the JOURNAL OF MECHANICAL DESIGN. Manuscript received January 26, 2003; revised April 21, 2004. Associate Editor: C. Dym.

set theory and fuzzy logic to design evaluation. Sections 3 8 present ve approaches to fuzzy design evaluation. Section 9 summarizes the approaches. Section 10 presents a case study. Finally, Section 11 concludes this paper. It is assumed that the reader is familiar with the basics of fuzzy sets and fuzzy numbers.

Related Work

Some research works have proposed applications of fuzzy set theory to design evaluation; this section presents a brief survey of some of these research works. Antonsson and Otto 2 compare several methods for incorporating and manipulating the imprecision encountered in engineering design. Probability and Bayesian inferencing, Dempster-Shafer theory, fuzzy set theory and triangular norms, in general, and utility theory are among the existing computable methods for representing uncertainty 2 . The conclusion is that fuzzy set analysis enables the formalization of imprecision in more types of engineering problems because their operations can be either compensated or noncompensated. Similarly, Thurston and Carnahan 5 compare two approaches to multiplecriteria design evaluation, fuzzy set theory, and multiattribute utility analysis. It is concluded that multiattribute utility analysis is preferred in the later stages of preliminary design, where performance levels can be expressed quantitatively, whereas fuzzy set theory is appropriate in the earliest stages of conceptual design evaluation or in situations that are based on linguistic nonquantitative performance levels. Thurston and Carnahan 5 propose the application of fuzzy set theory to multiple-criteria design evaluation. In order to nd the overall desirability of each alternative, the Fuzzy Weighted Average FWA is applied, but they do not use normalized weights such that the extended division is not needed in the calculation. The use of fuzzy set theory is described in two ways. In the former, the candidate with the highest fuzzy rating is selected, and in the latter, the best alternative is the one that is closest to a fuzzy goal. Other authors, such as Carnahan et al. 7 , Chen 13 , Khoo and Ho 14 , Zhou 15,16 , and Wang 17 have followed a similar procedure in design evaluation and Quality Function Deployment QFD . In these applications, the linguistic terms are characterized by fuzzy numbers, and the overall desirability is calculated by the FWA without the extended division. Performing the FWA without this division, however, greatly and unnecessarily increases the imprecision of the results. A previous work 11 has shown this and proposed the use of the NFWA in order to reduce the imprecision. Hsiao 10 and Kaymak and van Nauta Lemke 6 have applied the generalized weighted mean 18 to multiple-criteria design for assembly and decision making, respectively. In the second work 6 , the desirability of an alternative is represented by the degree of membership of its predicted performance in a fuzzy set, and the Transactions of the ASME

24 Vol. 127, JANUARY 2005

Copyright 2005 by ASME

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

importance of weights are real crisp numbers, which is not appropriate, as no uncertainty is considered. In Hsiao 10 , weights of objectives and performance levels are quantied with the membership functions of fuzzy sets; however, contrary to other works 5,7 , the scores assigned during the evaluation process are crisp numbers degrees of membership of fuzzy numbers . Thus, this method is less powerful and credible, as the results do not represent the imprecision of the judgements. In some works, such as 5,7,13 , the obtained performance level of each candidate is a fuzzy number. After this is done for all the alternatives, it follows their ranking or rating for the selection of the best one. Several authors have proposed different ranking or outranking methods for fuzzy numbers. Each of these appears to have some advantages as well as disadvantages, and the selection of a method depends on the particular application 19 . Klir and Yuan 19 describe two outranking methods and one ranking method; one of these is based on -cuts and is inconsistent 19 . The others are an outranking method based on the Hamming distance and a ranking method based on the extension principle 20 . Wang 21 developed two outranking and one ranking method based on the fuzzy preference relation. Wang 17 developed an outranking method based on possibility theory. Alternatively, Chen 22 proposed the determination of a utility real value by using the concept of the maximizing set and minimizing set. Similarly, Masud and Dean 23 proposed the calculation of an equivalent crisp number balance point of the membership region for each fuzzy number. Therefore, in these methods 22,23 , the fuzzy ranking becomes a simple ordering of real numbers. Finally, Carnahan et al. 7 have applied the Analytical Hierarchy Process AHP 12 for determining the weights of alternatives or criteria and developed the concept fuzzy line segment, which is described in Section 3.2, to give more exibility to the assignment of fuzzy values. As a conclusion, both performance levels and weights of criteria in design evaluation are sometimes fuzzied, being mostly represented as triangular fuzzy numbers. When performance levels are expressed as fuzzy numbers, the Fuzzy Weighted Average FWA is usually used for nding the overall performance; however, most authors have avoided the extended division to facilitate the calculations, but the imprecision is increased. When performance levels and weights are real numbers, the geometric mean or weighted geometric mean is normally used. There are some research works that propose metrics in engineering design problems 24 28 . Although they are not directly related to the scope of this work, it may be worth mentioning them. Farhang-Mehr and Azarm 24 and Wu and Azarm 26 present several metrics for assessing the quality or goodness of a set of solutions. McAdams and Wood 25 and Kota et al. 27 introduce metrics for nding similarity or commonality in different designs or products. Finally, Shah et al. 28 deal with metrics for assessing the effectiveness of different creativity methods.

Fig. 1 Fuzzy number representing

c. d.

e.

f.

g.

performance parameters 1 . An example of performance requirement is the weight of the motor must be less than about 50 kg. Weight: importance level of a criterion relative to other criteria Performance or desirability: degree to which an alternative satises a criterion; in other words, degree to which a particular value of an attribute is acceptable or adequate. For example, an alternative motor that has a weight of 55 kg could have 70% of desirability with respect to the criterion low weight. Predicted performance: expected performance of a design with respect to a criterion. For example, the weight of a particular solution is expected to be in the interval 45 kg, 50 kg ; this is the predicted performance of that solution with respect to the criterion weight. Overall performance: degree of desirability of a design, considering all the criteria. The performance levels per criterion are aggregated in a particular way in order to obtain the overall desirability. Attribute: characteristic of a design, related to a criterion i.e., the attribute weight is related to the criterion low weight

Fuzzy Approach to Design Evaluation

Several procedures can be followed during the evaluation of design candidates against multiple criteria through the use of fuzzy set theory. This paper presents different ways in which design evaluation can be carried out, based on different preferences of expressing required and predicted performance levels; some of them have already been proposed, whereas others are proposed and developed here. Each approach is appropriate for a particular case. Some denitions and explanations of terms are given before presenting the following approaches: a. Criteria or performance parameters: those against which a design solution or idea is evaluated, e.g., low weight and good appearance b. Performance requirements: each of these is related to a criterion. They are the specied target values for each of the Journal of Mechanical Design

3.1 Performance Requirements. In design evaluation, alternatives are evaluated against several criteria, such as low cost and short lead time, and generally, there are minimum requirements to be satised. The specication of quantitative performance requirements is necessary in order that the evaluation can be carried out systematically. However, precise crisp performance requirements are normally not adequate; for example, it could be stated that the maximum cost of a design is 1000, but 1010 may be still acceptable. Because a requirement like this is imprecise fuzzy , a fuzzy set or fuzzy number is appropriate for capturing it. Fuzzy sets enable smooth transitions between completely acceptable values to completely unacceptable ones. Qualitative terms, such as attractiveness, may be expressed linguistically, and the linguistic concepts very attractive, . . . , not attractive can also be captured by appropriate fuzzy numbers. The design requirements that are quantitative can be captured by membership functions of fuzzy numbers, which represent the desirability of the possible values of an attribute. In order to illustrate the assignment of a fuzzy number to a requirement, consider the example of a washing machine in which the requirement durable has been converted into a more detailed requirement the washing machine has to operate at least three years without breakdowns. However, as previously stated, a crisp boundary that represents the limit between acceptability and unacceptability is not appropriate. Then, design team members dene the membership function of a fuzzy number as depicted in Fig. 1, which represents the desirability of the minimum operating time without breakdowns. Note that three years is acceptable, but it has just 22% desirability, whereas total desirability is only obtained when the washing machine operates 3.7 years without breakdowns. This approach enables smooth transitions of desirability, from unacceptability to a performance of 100%. JANUARY 2005, Vol. 127 25

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 2 Fuzzy performances

numbers

for

predicted

and

required

When required and predicted performances are expressed linguistically, labels, such as very low VL , low L , medium low ML , medium high MH , high H , and very high VH can be dened. In order to carry out systematically the evaluation process, and taking into account that these labels are imprecise, fuzzy numbers similar to those depicted in Fig. 2 can be dened. Required performance is usually treated by researchers as an expression such as very high or at least high. However, a more exible approach of dening generalized fuzzy numbers, such as the one shown in Fig. 2 shaded area , is proposed here. It indicates the desirability throughout the universe of discourse, whereas the former type just indicates the limit of the requirement. According to the example in Fig. 2, a high performance or better seems completely desirable, a very low performance is unacceptable, and a performance in between is partially desirable, with desirability given by the degree of membership of the fuzzy number. 3.2 Relative Importance of Criteria. The evaluation of design candidates is carried out against the performance requirements, which are dened as fuzzy numbers as described previously. Because some criteria are more important than others e.g., durable may be more important than low weight , different weights should be assigned to them. Also, because during paired comparisons 12 decision makers do not have a precise knowledge about relative importance there is uncertainty during the judgements , fuzzy weights are more appropriate than precise weights. The amount of fuzziness, which may be represented by the scope of the membership function, depends on the degree of uncertainty of the judgements. This uncertainty is related to the consistency of the judgements and the imprecision of the linguistic scale that is used for the comparisons. The Analytical Hierarchy Process AHP 12 is an effective method for assigning weights, and several authors have proposed the integration of fuzzy principles with AHP 7,29,30 . The weighting method proposed by Carnahan et al. 7 , which is based on the concept fuzzy line segment, can be applied. In this method, the priorities are obtained through the AHP and then are converted into triangular fuzzy numbers. 3.3 Calculation of the Overall Performance of Design Candidates. Several steps follow the denition of fuzzy numbers for representing performance requirements and weights of criteria:

Fig. 3 Desirability m r of a design for the criterion weight

1. calculation or estimation of the predicted performance per criterion of each candidate 2. calculation of the desirability per criterion of each candidate, which is based on the predicted and required performances 3. calculation of the overall performance of each solution 4. comparison of the overall performances of the candidates, and selection of the best design Different procedures for carrying out these steps can be followed, depending on the type of requirements, goals, and predicted performances. Predicted performance may be captured by a fuzzy number that represents a range of values that can be achieved with a particular design solution; for example, the power of a machine that certain solution can transmit varies within a certain interval, but some values are more difcult to achieve with that solution less desirable . Predicted performance may also be captured by a fuzzy number that represents stochastic uncertainty, such as that of material properties. According to this, ve approaches are considered in this paper, which are described in Sections 4 8.

4 Approach 1: Predicted Performance as a Real Number


When some attributes of an alternative are known precisely, the predicted or estimated performances with respect to those attributes are real numbers. For the reasons stated in Section 3.1, performance requirements are imprecise and, therefore, are expressed quantitatively through fuzzy numbers. The calculation of the performance level per criterion step 2 in Section 3.3 is straightforward. Consider the example in Fig. 3 for the attribute weight which is related to a criterion, such as low or medium weight . A fuzzy number that captures the required performance desirability of values of weight , and a real number r that represents the predicted performance weight of a design candidate, with respect to that criterion are shown. The desirability of the candidate performance level with respect to that criterion is, then, the degree of membership m(r) of r in the fuzzy number, p m r (1)

For illustrating steps 2 4, listed in Section 3.3, consider the example in Table 1. The second column shows the ve attributes

Table 1 Example of evaluation of a design candidate No. C1 C2 C3 C4 C5 Criteria required performance Weight 8,14,24,30 N Appearance 0,0,5,10 Operating cost 0,0,6,10 Manufacturing cost 0,0,80,120 Durability 0,0,20,32 months Importance 0.3,0.4,0.5 0.1,0.2,0.3 0.5,0.6,0.7 0.7,0.8,0.9 0.6,0.7,0.8 Predicted performance 26.4 N 6.5 5 11.2 23.6 months Performance level 0.6 0.7 1.0 0.2 0.7

26 Vol. 127, JANUARY 2005

Transactions of the ASME

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 4 Required performance of appearance and predicted performance of an alternative

Fig. 6 Overall performance of the alternative in Table 1, calculated through the NFWA

that have to be controlled in the design problem: weight, appearance, operating cost, manufacturing cost and durability, and quadruplets (a,b,c,d), which represent trapezoidal fuzzy numbers that capture required performances. As an example, Fig. 4 shows the trapezoidal fuzzy number for the required performance of appearance. The third column presents the triplets (a,b,c) of triangular fuzzy numbers, which capture the fuzzied relative importance of the different criteria. These fuzzy numbers are depicted in Fig. 5 and can be obtained through the method described in Carnahan et al. 7 . The fourth column lists the predicted performances of the candidate; Fig. 4 shows the predicted performance for the attribute appearance. Finally, the fth column shows the performance level of the alternative for each criterion, which is obtained by Eq. 1 . Figure 4 illustrates the performance level for appearance. In engineering design, a simplied fuzzy weighted average, without the extended division, is normally used for calculating the preference of an alternative when fuzzied preferences per criterion are available. Although, this reduces the complexity of the calculations, the imprecision of the results is greatly and unnecessarily incremented. The authors strongly recommend not to utilize the FWA without the extended division. The overall desirability can be calculated through the NFWA 11 , which produces less imprecise and more meaningful results. The NFWA is calculated as follows: let P be the overall performance on an alternative fuzzy number ; and p 1 , p 2 , . . . ,p n be the performance levels per criterion of n criteria with weights fuzzy numbers W 1 , W 2 , . . . ,W n . The -cut P of the overall performance P is given by P where
n i 1 n

In Eqs. 2 4 , P a and P b are the lower and upper limits, respectively, of the -cut P . p i a and p i b are the lower and upper limits, respectively, of the -cut p i . W i a and W i b are the lower and upper limits, respectively, of the -cut W i . The min operator represents the minimum value that can be obtained by combining the w i in all the possible ways. The max operator represents the maximum value that can be obtained by combining the w i in all the possible ways. Note that the set of w i used in the numerator has to be the same as the one in the denominator. As an example, Eq. 3 is applied for the data in Table 1, for 0, as follows: P P 0.6 0.3
a

0.7 0.1 1 0.5 0.2 0.9 0.3 0.1 0.5 0.9 0.6 0.7 0.3 1 0.7 0.2 0.7 0.3 0.3 0.7 0.7 0.8

0.7 0.6 0.7 0.8

0.5625 0.6393

0.6 0.3
b

,P

(2)

p i a w i
n

min

and wi

max

i 1

p i b w i
n

i 1

i 1

wi (3)

where wi Wi
a

,W i

, for all i

1,2, . . . ,n

and all

0,1 (4)

Figure 6 shows the result obtained by applying the NFWA to the example in Table 1. Note that although the fuzzy numbers of the weights of criteria are relatively very imprecise, the NFWA operation produces an overall performance that is relatively less imprecise this does not occur with the FWA without the extended division . The value of performance with degree of membership of one in the obtained fuzzy number is equal to 0.604, which is very similar to the balance point x coordinate of the centroid of the area under the membership function curve. Then, the overall performance or desirability of the alternative solution in Table 1 is about 60%. This procedure is utilized for calculating the overall performance of each alternative. The step that follows is the ranking or rating of the designs with respect to the overall performance, which can be carried out in different ways 19,2123 . This particular step is, in itself, a subject of research, but this paper is not intended to deal with it. The fuzzy numbers, in this case, represent overall performance levels that can be interpreted as possibility distributions; therefore, the centroid method 22,23 seems to be adequate. In this method, a balance point real number is calculated for each resulting fuzzy number, and, therefore, the fuzzy ranking is reduced to the ranking of real numbers. The alternative with the greatest balance point may be the best.

5 Approach 2: Predicted Performance as a Fuzzy Number Representing Stochastic Uncertainty


Sometimes, the predicted performance of an alternative with respect to an attribute is an uncertain value, whose uncertainty is of stochastic nature. For example, some properties of a material, such as strength and thermal conductivity, are attributes of this type. Similar to the rst approach, the performance requirements are expressed quantitatively through fuzzy numbers, but the predicted performances are now expressed as fuzzy numbers representing stochastic uncertainty. For this case, the authors propose the metric described by the following example. JANUARY 2005, Vol. 127 27

Fig. 5 Weights of criteria of the example in Table 1

Journal of Mechanical Design

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 8 Jones and Hua 26 metric for nding desirability

Fig. 7 Desirability m A r with respect to the criterion weight of a value r with possibility represented by m B r

A trapezoidal fuzzy number, representing the required performance with respect to the attribute weight, and a triangular fuzzy number, representing the possibility distribution of the predicted performance of an alternative predicted weight , are shown in Fig. 7. A value r is mapped by the membership functions m A (x) required performance and m B (x) predicted performance to the degrees of membership m A (r) and m B (r), respectively. Let us assume that the actual weight of the alternative is r. Then, the desirability of the alternative would be equal to m A (r) as in approach 1 ; however, the actual weight is unknown. The desirability of the alternative with respect to weight is obtained by integrating the desirability of each possible weight x, where x X universe of discourse , taking into account the possibility of occurrence of that value r. The effect of each x is the product of the possibility of x with the desirability of x. Hence, m B x dx
X Xm B

motors are less available. Therefore, a fuzzy number can be used for representing this type of predicted performance. For example, a trapezoidal fuzzy number (a,b,c,d) may be used; the interval b,c pessimistic range represents a power range than can be easily obtained, whereas a,d optimistic range represents a power range that can be obtained, but with difculty. Similar to approaches 2 and 3, the performance requirements are expressed quantitatively through fuzzy numbers. The predicted performance is, in this approach, represented by a fuzzy number or fuzzy set of possible values that can be obtained with the investigated design solution. Since values at the boundaries tend to be more difcult to obtain than values within the pessimistic range, fuzzy boundaries are appropriate to represent smooth transition between completely desirable values to totally undesirable ones. This case has been addressed by two research works, but not for evaluation in design. Jones and Hua 31 suggest a metric that consists of the maximum degree of membership of the intersection of the two fuzzy numbers representing predicted and required performances. That is, value m(r) in Fig. 12, which is determined by p m r sup m A
X B

mA x

(5) x sup min m A x ,m B x


X

x dx

(7)

where p is the performance of the alternative with respect to the criterion weight. Note that the former term is the possibility of x ratio of the innitesimal area under m B (x) and the total area , and the latter term is the desirability of x. Because the integral in the denominator is constant with respect to the rst integral, Eq. 5 can be rewritten as

Xm A

x m B x dx (6)

where p is the performance level of the alternative with respect to a criterion, and sup takes the maximum value of m A B (x), for x X. Giachetti 32 also proposes a metric, which is based on possibility theory. The necessity and possibility measures are calculated. Possibility assesses to what extent the alternative design possibly satises the required performance. This is an optimistic selection strategy 32 . The degree to which the predicted performance possibly satises the requirement is given by

p
X m B x dx

Equation 6 is used to nd the desirability of the alternative for each criterion, then the overall performance is calculated through the procedure described for the approach 1. Similarly, the comparison of the alternatives is carried out by calculating balance points.

6 Approach 3: Predicted Performance as a Fuzzy Number Representing Possible Values to be Achieved of a Design Solution
In some cases, a design solution can have a range of values of an attribute. For example, the same solution idea that utilizes electric motors can have a range of values of power. Also, some values of power may be more difcult to achieve, as some electric 28 Vol. 127, JANUARY 2005

Fig. 9 Necessity measure of the predicted durability under its requirement

Transactions of the ASME

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Poss x satises the requirement sup m A


X B

dicted performance, 1 m B (x). The degree to which the predicted performance certainly satises the requirement is given by Nece x satises the requirement (8) inf m A
X B

sup min m A x ,m B x
X

x (9)

where sup takes the maximum value of m A B (x), for x X. Note that this equation is the same as 7 . Therefore, the possibility of x satisfying the requirement is equal to m(r) in Fig. 8. Necessity assesses to what extent the alternative design certainly satises the required performance. It is calculated as the impossibility of the opposite event. This is a pessimistic selection strategy 32 . The opposite event is the complement of the pre-

inf max m A x ,1 m B x
X

where inf takes the minimum value of m A B (x), for x X. An example of the necessity measure is shown in Fig. 9. The desirability per attribute is calculated by combining these two measures, using a factor that represents the level of optimism or pessimism of the designer 32 ,

Poss x satises the requirement

1 2

Nece x satises the requirement

(10)

where p is the performance level and 0,1 . An optimistic designer would use toward 1 and a pessimistic designer toward 0. The metric proposed by Giachetti 32 seems to be inappropriate. Consider the example shown in Fig. 10. The possibility measure is equal to one and the necessity measure is equal to zero. Combining Poss and Nece through Eq. 10 produces a value less than one unless 1 . However, the desirability of this alternative with respect to the attribute durability should be equal to one, as there are several values the interval a,b ) within the pessimistic range of the predicted performance (x a) that satisfy completely the requirement. Therefore, the necessity measure produces an undesirable effect. If 1, only the possibility is taken into account, and the metric is equal to the one proposed by Jones and Hua 31 , which also seems to be decient. For the example in Fig. 8, when x r, the overall desirability of x should be equal to the desirability achieved by x times the preference of taking that value. That is, if the desirability of x is 0.67 and the preference is 0.67, the overall desirability should be 0.67 times 0.67; that is, 0.45. In order to cope with the deciencies of these metrics 31,32 , this paper proposes a new metric. The performance level p of an alternative design with respect to a criterion is given by p sup m A x m B x
X

the example in Fig. 8. The maximum degree of membership of this new fuzzy number is the performance level of the alternative with respect to the criterion durability. Equation 11 is utilized for calculating the desirability per criterion of the alternative. The calculation of the overall desirability and the comparison of the alternatives are carried out in the same way as in approaches 1 and 2.

7 Approach 4: Linguistic Predicted and Required Performances


Contrary to the previous approaches, the approach in which the performance level per criterion is expressed linguistically and the preferred performance is either the highest or the lowest for example, very high or very low , for all the criteria, has been dealt with widely 5,7,1317 . The linguistic terms, indicating performance levels, are captured by fuzzy numbers. In these works, the overall desirability is calculated through the fuzzy weighted average without the extended division, then, they have the problem of increasing imprecision. The resulting fuzzy numbers overall performance levels are ranked, and the alternative with the highest or lowest performance is chosen. If the criteria are, for example, durable and easy of maintaining, then higher performances are preferred; whereas if the criteria are low cost and small number of breakdowns per year, then lower performances that is, lower costs and smaller numbers of breakdowns per year are more desired. This approach is not described here; the reader interested is referred to 5 . As mentioned in the previous paragraph, the comparisons of alternatives is carried out through the FWA without the extended

(11)

where sup takes the maximum value of the product m A (x) m B (x), for x X. This metric Eq. 11 deals with the two preferences required and predicted performance in calculating the performance level and takes the maximum value that can be achieved in order to optimize the design solution. Figure 11 shows the membership function of the fuzzy number m A (x) m B (x) for

Fig. 10 Required versus predicted performance

Fig. 11 A new metric for approach 3

Journal of Mechanical Design

JANUARY 2005, Vol. 127 29

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 13 Examples of a new type of fuzzy goal for linguistic predicted performances Fig. 12 Aggregate fuzzy sets, overall performances of two design solutions

division and, then through the ranking of the alternatives. This paper also proposes a novel way of calculating and comparing overall performance levels that has not been explored for this approach at least in engineering design . An aggregate fuzzy set, of all the fuzzy numbers representing performance levels, is calculated for each alternative through the generalized weighted mean 18
n 1/

fuzzy number. However, in this approach the preferred performance with respect to a criterion is not necessarily the highest or lowest possible. Then, a fuzzy goal is dened for each criterion, and the performance of an alternative with respect to that criterion is determined based on that goal. Two ways of expressing fuzzy goals are considered: a. The fuzzy goal can be dened by a linguistic expression, such as medium-low, which is captured by an appropriate fuzzy number. The performance level is measured according to the fuzzy distance of each predicted performance from the fuzzy goal. The overall desirability is calculated by the NFWA of the fuzzy distances from the fuzzy goals. Therefore, the alternative that is closest to the fuzzy goal is selected. This approach is described in the literature 5,7 . b. A new type of fuzzy goal, such as those in Figs. 13 and 2, is dened. It has the advantage that it denes not only the more desired value of an attribute, but also the desirability of each value in the universe of discourse. Figure 13 depicts two fuzzy goals, A and B. The most desired value of performance of both fuzzy goals is medium-high MH , but goal B is more demanding than goal A, as its desirability decreases rapidly toward both sides, whereas goal A permits a wider range of performance values from 0.12 to 0.92, approximately . If these goals were expressed as in approach a , they would be no difference between them. Therefore, this type of fuzzy goal is more generic approach a is a particular case and more powerful. The procedure for nding the overall desirability and the comparison of alternatives is the same as the one of approach 2, because each linguistic predicted performance can be interpreted as a possibility distribution.

h a 1 ,a 2 , . . . ,a n where
n i 1

i 1

w ia i

(12)

w i 1,

and

wi

0,1

where h is the degree of membership of x in the aggregate fuzzy set, a i m i (x) that is, the degree of membership of x in the ith fuzzy number being aggregated and w i is the defuzzied through centroid method weight of the ith criterion. The parameter , which represents level of optimism or pessimism 6 , is chosen according to the particular application. Large negative values of indicate high level of pessimism. Large positive values of indicate high level of optimism. When 1, the equation reduces to the weighted average operation 19 . Comparison of the prole of each fuzzy set is carried out in order to ascertain which alternative is the best. This comparison may be an additional way of selecting an alternative or may replace the ranking through the NFWA. The shape and the balance point of the aggregate fuzzy set can be used as a means of comparison. The balance point can be used for ranking the alternatives, and the shape or prole represented by the aggregate fuzzy set of each alternative can be visualised in order to see how the alternatives differ from each other. For example, an alternative can have similar performance levels for all the criteria, while another alternative can have low, medium, and high performance levels. The application of the generalized weighted mean, Eq. 12 , produces fuzzy sets such as those presented in the example in Fig. 12. An aggregate fuzzy set A of a design and its balance point are shown in Fig. 12. As can be noted, most of the area under its membership function is concentrated toward high performance, and only a small area toward lower performance. The aggregate fuzzy set B of another solution is also shown. This has the same balance point 0.705 , but the areas are distributed in a different fashion; some criteria are very highly satised whereas satisfaction of others are medium low and very low. The comparison of these two shapes can be served as a basis for the nal decision. For example, designers may opt for the former alternative A because its performance is consistently high, except for a small are that is low, which can be further studied to revise its implications on the overall performance.

Final Comments

Table 2 presents a summary of the approaches described in the previous sections along with some characteristics and limitations. For the rst three approaches, predicted performances are based on numerical scales, whereas for the last two, they are based on linguistic scales, which are also matched to numerical scales usually from 0 to 1 . All approaches, except approach 4, have metrics for calculating performance levels based on required and predicted performances and can be combined in a single design evaluation problem. On the contrary, approach 4 does not need a metric, and the predicted performance is taken as the performance level. The reasons for this are that, for all the criteria, predicted performances are based on the same linguistic scale and the preferred performance is the highest or the lowest possible. Finally, from Eqs. 5 and 11 it can be concluded that the rst approach is a special case of either the second or the third approach.

10

Case Study: A Handle for Closing a Window

Approach 5: Fuzzy Goal

Similarly to approach 4, the performance of an alternative, per criterion, is expressed by a linguistic term, which is captured by a 30 Vol. 127, JANUARY 2005

This example considers the selection of a single manufacturing process for obtaining a large batch of handles for closing a window. According to technical knowledge, the design team members dene the design requirements of the window handle, which are shown in Table 3, as well as the importance of each criterion. As Transactions of the ASME

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Table 2 Summary of the fuzzy approaches to design evaluation App. 1 2 3 Predicted perform. Real number Fuzzy number stochastic uncertainty Fuzzy number of values that can be achieved Required perform. Fuzzy number Fuzzy number Fuzzy number The highest or the lowest Fuzzy number fuzzy goal based on a linguistic scale Metric Eq. 1 Fig. 3 Eq. 6 Fig. 7 Eq. 11 Fig. 11 Not needed a Fuzzy distance not described here b Eq. 6 Fig. 13 Notes The predicted performance has to be known precisely as may be, for example, the number of operations that a certain alternative machine can perform. The uncertainty of the predicted performance is or may be taken as of stochastic nature. For attributes such as material properties, cost, appearance, and ease of manufacture. When the predicted performance has a fuzzy range of possible values that can be achieved by the alternative. For example, the surface roughness that can be obtained with a machining operation may be represented as a fuzzy number. The preferred performance has to be either the highest or the lowest for all criteria. The overall performance is calculated directly with the predicted performances a metric is not needed . It is adequate when the preferred performance is not the highest or the lowest for all criteria. For example, for the attribute duration of the machine, one may need it to have a certain fuzzy life, and both a shorter and a longer life are less desirable. When for some criteria the fuzzy goal is not the highest or the lowest, but the preferred performance is, the fuzzy goal has to be set to be the highest or the lowest; however, approach 5(b) is more appropriate in this case. Calculation of overall performance is similar to approach 2, but the required and predicted performances are, as in approaches 4 and 5(a), based on a linguistic scale.

Fuzzy number capturing a linguistic term

mentioned earlier, design requirements are often fuzzy, then some of them are captured by trapezoidal fuzzy numbers; the importance levels of the criteria are also uncertain and are captured by triangular fuzzy numbers. According to Table 3, for example, the most important criterion is section, as it has an importance of about 0.8; the preferred value of section is 3 mm, but values within the interval 2.5,5 mm have some degree of acceptability. As an example, Fig. 14 shows the trapezoidal fuzzy number that captures the required performance for the attribute weight. Similarly, Fig. 15 shows the required performance for the attribute appearance; instead of dening the required performance as, for example, high H , the fuzzy number represents the desirabil-

Table 3 Window handle: design requirements Attribute Material class Shape class Weight Section Tolerance Relative cost Appearance Batch size Importance 0.1,0.2,0.3 0.7,0.8,0.9 0.2,0.3,0.4 0.6,0.7,0.8 0.2,0.3,0.4 Required performance Zinc-based alloy Three-dimensional solid 0,0,0.08,0.12 kg 2.5,3,3,5 mm 0,0,0.2,0.25 mm 0,0,5,80 * 0.5,0.8,1,1 1,000,000

ity of the values in the universe of discourse. This is a new way of expressing linguistic performance requirements as proposed in Section 8. The designers and the manufacturing engineers work together in order to obtain the necessary information about the manufacturing alternatives and write down the information in a table. Table 4 shows the available manufacturing processes that can be used and the predicted performance of each option for each criterion. Trapezoidal fuzzy numbers, which represent sets of values that can be achieved, are dened for representing predicted performances for the attributes weight, section, and tolerance. Consider the trapezoidal fuzzy number of the predicted performance of machining with respect to the attribute weight, which is shown in Fig. 14. The range 0.001, 1000 kg is a set of values that can be easily achieved with the machining process, but values out of this interval and within the range 1e-5,1e4 kg can also be achieved, but with more degree of difculty. Triangular fuzzy numbers and linguistic labels are dened for the predicted performances with respect to relative cost and appearance, respectively, and the linguistic terms are captured by triangular fuzzy numbers. The uncertainty involved in the judgements for these two attributes may be taken as of stochastic na-

*This requirement is based on the data for the criterion relative cost of the manufacturing processes considered in this example.

Fig. 14 Trapezoidal fuzzy numbers that represent the required performance for the attribute weight and the predicted performance of machining with respect to this attribute

Fig. 15 Fuzzy numbers for predicted and required performances for the attribute appearance

Journal of Mechanical Design

JANUARY 2005, Vol. 127 31

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Table 4 Predicted performance of manufacturing options for each criterion* Manufacturing process Closed-die forg. Die cast Machining Dic press P/M forging CLA/CLV Investment Predicted performance Weight kg 0.001,0.01,10,100 0.01,0.5,10,50 1e-5,1e-3,1e3,1e4 0.002,0.01,8,30 0.002,0.01,8,30 0.01,0.06,10,80 0.01,0.06,10,80 Section mm 2,5,110,200 0.8,2,9,25 0.06,0.1,1e3,1300 1,2,8,20 1,2,100,200 0.5,0.8,10,25 0.6,1.1,10,80 Tolerance mm 0.05,0.1, 0.05,0.15, 0.01,0.1, 0.05,0.1, 0.04,0.06, 0.08,0.15, 0.05,0.1, , , , , , , , Relative cost 10,15,60 15,25,100 5,20,100 10,18,60 50,110,700 150,300,800 100,200,1e4 Appearance MH MH H MH MH H H

*Some data in this table are based on the information presented in 28 .

Table 5 Summary of results Manufacturing process M1 M2 M3 M4 M5 M6 M7 Closed-die forg. Die cast Machining Die press P/M forging CLA/CLV Investment Performance level Weight 1 0.154 1 1 1 1 1 Section 0.375 1 1 1 1 1 1 Tol. 1 1 1 1 1 1 1 Cost 0.689 0.450 0.516 0.676 0.003 0 0 Appearance 0.347 0.347 0.889 0.347 0.347 0.889 0.889 NFWA 0.54,0.60,0.66 0.60,0.67,0.75 0.80,0.84,0.87 0.76,0.82,0.86 0.52,0.61,0.70 0.59,0.68,0.75 0.59,0.68,0.75 Balance point 0.60 0.67 0.84 0.82 0.61 0.68 0.68

ture. Figure 15 illustrates the predicted performance levels for the manufacturing processes with respect to the attribute appearance. According to the characteristics of this case study, approach 2 is applied for relative cost, approach 5(b) for appearance, and approach 3 for the other attributes. The results are shown in Table 5; performance levels are calculated by using Eq. 6 or 11 , with the data in Tables 1 and 2 predicted and required performances , the NFWA is calculated through Eq. 2 with the importance levels of the criteria Table 3 and the performance levels Table 5 . Because most of the requirements of the window handle are very easy to obtain, many of the cells for performance levels columns 25 have 1, which indicates that for a particular attribute, a manufacturing processes can easily achieve the required performance. Besides, no performance level is equal to zero excluding cost , which indicates that all the processes can produce the handle. It can also be observed that relative cost and appearance are the attributes that produce the most signicant effects on the ranking of the alternatives. The NFWA of each alternative is practically a triangular fuzzy number. Accordingly, the seventh column in Table 5 shows the triplets of those fuzzy numbers, which are also shown in Fig. 16. Finally, last column in Table 5 shows the balance points of these fuzzy numbers. As mentioned in Section 3, the ranking of the alternatives can be done by comparing the balance points of the fuzzy numbers, then machining would be the best option, as its balance point, 0.84, is the highest. However, from Fig. 16, it can also be inferred that die press could be the best option, as it overlaps with machining. The usefulness of fuzzy methods is then evident here, as the

balance points crisp values give one unique winner, whereas fuzzy numbers, and the uncertainty involved in them, suggest that either machining or die press may be the best option. It is clear from Fig. 16 that the other processes are less desirable for this case. The next step in the selection could be to carry out a more accurate analysis for the rst two processes; detailed information, such as the capabilities of the machines available and the costs, may be drawn in order to ascertain which is the best option.

11

Conclusions

A survey of approaches to fuzzy engineering design evaluation was presented; some of them have been proposed in the literature, whereas others have been proposed in this paper. The designer can adopt a particular approach depending on the conditions of the problem and/or his/her preferences. As demonstrated with the case study in Section 10, a combination of approaches is also possible and necessary, as the criteria of a design problem are normally of different nature. This paper has also proposed some improvements for some approaches. Two metrics for determining the desirability per criterion of an alternative were proposed: the former for the case in which fuzzy numbers represent stochastic uncertainty, and the latter when they represent ranges of values that can be achieved. It was shown that the latter metric is more suitable than others proposed in other works 3133 . A new fuzzy goal was proposed, which not only considers the more desirable value of the performance of an alternative with respect to a criterion, but also considers the desirability of other values different from the preferred one. A novel way of comparing alternatives through the calculation of an aggregate fuzzy set, both for ranking alternatives or as an additional tool for deciding among alternatives, has been proposed. Because fuzzy set theory has the exibility to manage different types of problems in engineering design evaluation, it is a powerful tool for dealing with the imprecision in design.

Acknowledgments
L.V.V.U. and A.W.L. would like to thank their respective institutions for the support for carrying out this investigation. The authors would also like to thank Professor David Hayhurst UMIST and both anonymous referees for their valuable comTransactions of the ASME

Fig. 16 Ranking of manufacturing options

32 Vol. 127, JANUARY 2005

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

ments and criticism on an earlier version of the manuscript. The authors would also like to thank the European Commission and the partners of the Innovative Production Machines and Systems I*PROMS Network of Excellence www.iproms.org .

References
1 Wood, K. L., and Antonsson, E. K., 1990, Modeling Imprecision and Uncertainty in Preliminary Engineering Design, Mech. Mach. Theory, 25 3 , pp. 305324. 2 Antonsson, E. K., and Otto, K. N., 1995, Imprecision in Engineering Design, ASME J. Mech. Des., 117B, pp. 2532. 3 Giachetti, R. E., Young, R. E., Roggatz, A., Eversheim, W., and Perrone, G., 1997, A Methodology for the Reduction of Imprecision in the Engineering Process, Eur. J. Oper. Res., 100, pp. 277292. 4 Vanegas, L. V., and Labib, A. W., 2001, A Fuzzy Quality Function Deployment FQFD Model for Deriving Optimum Targets, Int. J. Prod. Res., 39 1 , pp. 99120. 5 Thurston, D. L., and Carnahan, J. V., 1992, Fuzzy Ratings and Utility Analysis in Preliminary Design Evaluation of Multiple Attributes, ASME J. Mech. Des., 114, pp. 648 658. 6 Kaymak, U., and Van Nauta Lemke, H. R., 1993, A Parametric Generalized Goal Function for Fuzzy Decision Making With Unequally Weighted Objectives, IEEE International Conference on Fuzzy Systems, March, pp. 1156 1160. 7 Carnahan, J. V., Thurston, D. L., and Liu, T., 1994, Fuzzing Ratings for Multiattribute Design Decision-Making, ASME J. Mech. Des., 116, pp. 511 521. 8 Durr, H., and Schramm, M., 1997, Feature Based Feedback Into the Early Stages of Design, Eur. J. Oper. Res., 100, pp. 338 350. 9 Muller, K., and Sebastian, H.-J., 1997, Intelligent Systems for Engineering Design and Conguration Problems, Eur. J. Oper. Res., 100, pp. 315326. 10 Hsiao, S.-W., 1998, Fuzzy Logic Based Decision Model for Product Design, Int. J. Ind. Ergonom., 21, pp. 103116. 11 Vanegas, L. V., and Labib, A. W., 2001, Application of New Fuzzy-Weighted Average NFWA Method to Engineering Design Evaluation, Int. J. Prod. Res., 39 6 , pp. 11471162. 12 Saaty, T. L., 1994, Fundamentals of Decision Making and Priority Theory With the Analytical Hierarchy Process, RWS Publications, Pittsburgh. 13 Chen, Y. H., 1996, Fuzzy Ratings in Mechanical Engineering Design Application to Bearing Selection, Proc. Inst. Mech. Eng., Part B: J. Eng. Manuf., 210, pp. 4953. 14 Khoo, L. P., and Ho, N. C., 1996, Framework of a Fuzzy Quality Function Deployment System, Int. J. Prod. Res., 34, pp. 299311. 15 Zhou, M., 1998, Fuzzy Logic and Optimization Models for Implementing QFD, Comput. Ind. Eng., 35, pp. 237240.

16 Zhou, M., 1997, Fuzzy Logic Based Models for Quality Planning and Improvement, ASME Conference: Intelligent Engineering Systems Through Articial Neural Networks, ASME, New York, Vol. 7, pp. 311316. 17 Wang, J., 1999, Fuzzy Outranking Approach to Prioritize Design Requirements in Quality Function Deployment, Int. J. Prod. Res., 37, pp. 899916. 18 Klir, G. J., and Floger, T. A., 1988, Fuzzy Sets, Uncertainty and Information, Prentice-Hall, Englewood Cliffs, NJ. 19 Klir, G. J., and Yuan, B., 1995, Fuzzy Sets and Fuzzy Logic: Theory and Applications, Prentice-Hall, Englewood Cliffs, NJ. 20 Zadeh, L. A., 1965, Fuzzy Sets, Information and Control, Academic Press, 8, pp. 338 353. 21 Wang, J., 1997, A Fuzzy Outranking Method for Conceptual Design Evaluation, Int. J. Prod. Res., 35, pp. 9951010. 22 Chen, S. H., 1985, Ranking Fuzzy Numbers With Maximising Set and Minimizing Set, Fuzzy Sets Syst., 17, pp. 113129. 23 Masud, A. S. M., and Dean, E. B., 1993, Using Fuzzy Sets in Quality Function Deployment, Proc. the 2nd Industrial Engineering Research Conference, pp. 270274. 24 Farhang-Mehr, A., and Azarm, S., 2003, An Information-Theoretic Entropy Metric for Assessing Multi-Objective Optimization Solution Set Quality, ASME J. Mech. Des., 125 4 , pp. 655 663. 25 McAdams, D. A., and Wood, K. L., 2002, A Quantitative Similarity Metric for Design-by-Analogy, ASME J. Mech. Des., 124 2 , pp. 173182. 26 Wu, J., and Azarm, S., 2001, Metrics for Quality Assessment of a Multiobjective Design Optimization Solution Set, ASME J. Mech. Des., 123 1 , pp. 18 25. 27 Kota, S., Sethuraman, K., and Miller, R., 2000, A Metric for Evaluating Design Commonality in Product Families, ASME J. Mech. Des., 122 4 , pp. 403 410. 28 Shah, J. J., Kulkarni, S. V., and Vargas-Hernandez, N., 2000, Evaluation of Idea Generation Methods for Conceptual Design: Effectiveness Metrics and Design of Experiments, ASME J. Mech. Des., 122 4 , pp. 377384. 29 Buckley, J. J., 1984, The Multiple-Judge, Multiple Criteria Ranking Problem: A Fuzzy-Set Approach, Fuzzy Sets Syst., 13, pp. 2537. 30 Weck, M., Klocke, F., Schell, H., and Ruenauver, E., 1997, Evaluating Alternative Production Cycles Using the Extended Fuzzy AHP Method, Eur. J. Oper. Res., 100, pp. 351366. 31 Jones, J. D., and Hua, Y., 1998, A Fuzzy Knowledge Base to Support Routine Engineering Design, Fuzzy Sets Syst., 98, pp. 267278. 32 Giachetti, R. E., 1998, A Decision Support System for Material and Manufacturing Process Selection, J. Intel. Manuf., 9, pp. 265276. 33 Esawi, A. M. K., and Ashby, M. F., 1998, Computer-Based Selection of Manufacturing Processes: Methods, Software and Case Studies, Proc. Inst. Mech. Eng., Part B: J. Eng. Manuf., 212 B8 , pp. 595 610.

Journal of Mechanical Design

JANUARY 2005, Vol. 127 33

Downloaded 29 Mar 2012 to 117.211.91.58. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Anda mungkin juga menyukai