Professional Documents
Culture Documents
S.Boobalan(1), Dr.N.Mohandas Gandhi(2) Student(1), Professor(2), Department of Mechanical Engineering (Kumaraguru college of Technology Coimbatore) E-mail: balan1085@gmail.com mob-9597805793
Abstract: Two supplier ranking models that use linguistic variables are presented. The first model was developed using the traditional methods of rule definition of fuzzy logic, while Combs method was used in the second. Combs method maps the problem space using a significantly reduced rule base, hence making the size of the model manageable in terms of computing time and resources. This paper will show how these models use natural language (i.e. linguistic variables) to rate and rank suppliers in order to facilitate the supplier selection task. Comparison of both models is presented along with the results of statistical analysis. Key-Words: Fuzzy Logic, Combs method, Supplier selection
1 Introduction
In the face of global competition, buyers must maintain a dynamic list of criteria for the selection of suppliers. These criteria have traditionally been quantitative values such as price and on-time delivery. More recently qualitative criteria such as quality and service have been added to this list [1]. Many leading companies list value-added criteria among their selection criteria [2-5]. The following is a list of most commonly shared criteria in the supplier selection process: 1) value-added to buyers business (i.e. willingness to have a longterm, mutually beneficial working relationship, dedication to buyers success), 2) quality management and commitment to continuous improvement, 3) technological competitiveness, 4) financial viability, 5) knowledge of buyers line of business, 6) customer service and satisfaction, 7) competitive pricing, and 8) sharing similar values and ethics as the buyer. Most of the criteria are in fact vague. For example how would one measure the suppliers ethics and values? Or how would one determine quantitatively if the supplier were willing to invest in a mutually beneficial relationship with the buyer? It is criteria like these that depend on the human instinct, gut feeling, or experience. As mentioned in [4], Selecting a supplier is a quantitative and qualitative process. A supplier should offer more than just parts that meet spec". It is these types of data that have been very difficult to incorporate in mathematical models. In fact, modeling decision making scenarios has captured engineering interest for over two decades [6]. Several techniques were developed for modeling decision-making [7]. Until the inception of fuzzy logic and the analytic hierarchy process (AHP), the available models only considered the
tangible data (i.e. data that we can quantify). However, systems based solely on quantitative data will not reproduce the desired behavior. This is due to the fact that systems in the real world usually rely on quantitative as well as on qualitative data. The decision makers need models that mimic human decision making to be able to evaluate suppliers using both qualitative and quantitative data. The more comprehensive the decision-making models are and the closer they are to human decision-making behavior, the more flexibility the buyers have at their disposal. Fuzzy logic is a method that models imprecise, ambiguous, or vague data [8]. It is based on human common sense. It can easily handle vagueness, imprecision, or what is commonly known as partial truths in the data. Fuzzy logic has been used in similar decision making applications. Wang and Lin propose a system for ranking different configuration items for software development [9]. They further develop consensus measures to determine group acceptability of the ranking orders. Another application of ranking using fuzzy logic is that performed by Beccali et al. where they present fourteen sets of actions (alternatives) to be ranked according to twelve criteria [10]. Out of the twelve criteria, seven are qualitative. They use fuzzy logic on the seven qualitative criteria in order to come up with a rank for each alternative. This paper aims at modeling the behavior of evaluation and selection of criteria used in the supplier selection from the buyers side. Two models are developed using fuzzy logic demonstrating the power of this approach in dealing with qualitative data. The structure of the remainder of the paper is as follows. In Section 2 background literature on
fuzzy logic will be reviewed. The models will be discussed in Section 3. Research results will be discussed in Section 4 followed by conclusions and recommendations for future research in Section 5.
predict stream/watershed quality for fish habitat [19]. Fuzzy logic has also been used in the banking sector [18]. Banks review an applicant history and decide on the level of risk s/he poses with the aid of a fuzzy logic decision making system. DeTurris et al. reports on the use of Fuzzy Logic in the control of a flying eye that floats down carrying a camera sending pictures of its surroundings to the user [20].
Criteri Suppli a er
Decision maker D1 M G V G M G G M G G V G G G G M G G D2 M G V G M G G M G G V G G G V G M G G D3 M G V G G G M G G V G M G G V G M G G
S1
C1
S2
S3 S4 S1 S2 C2 S3
S4 S1 S2 C3 S3 S4
S1 S2 C4 S3 S4
G G G G
G G G G
G G G V G
C1 P A G E P M-N M-N M-Y M-Y
C2 A M-N M-Y M-Y M-Y G M-Y M-Y M-Y M-Y E M-Y M-Y M-Y Y
3.1
This model consists of four inputs C1, C2, C3, and C4 and one output (score). The inputs are the criteria used in evaluating the supplier, and the output is the score obtained for each of the suppliers prior to ranking them. Examples of criteria could be factors such as trust, relationship, flexibility, quality, service, and ethics [21]. Fuzzy-set values for the input and output are specified as shown in Table 1. Table 1. Fuzzy set values of inputs and output Input Output C1 C2 C3 C4 Score P (Poor) P P P No A (Average) A A A M-N (Maybe Not) G (Good) G G G M-Y (Maybe Yes) E (Excellent) E E E Yes Rules are then developed to describe relationships between inputs and outputs. example rule is provide below: the An
Computer Model: A computer model was developed using the Fuzzy Logic Toolbox of Matlab. The four criteria C1 through C4 were entered into the system as fuzzy set inputs and score as an output set, as shown in Figure 1.
Criteria
Decision maker D1 (0.8, D2 (0.8, 0.9, 1.0) (0.8, 0.9, 1.0) (0.9, 1.0, 1.0) (0.8, 0.9, 1.0) D3 (0.8, 0.9, 1.0) (0.9, 1.0, 1.0) (0.9, 1.0, 1.0) (0.9, 1.0, 1.0)
C1
C2
C3
If Relationship is Excellent AND Flexibility is Good AND Quality is Excellent AND Service is Good Then Score is Yes. In developing the model the following methodology was used: FLS Mapping: In order to cover all the possible combinations of rules a mapping of the FLS was developed. There are 256 rules in this model determined by the complete enumeration of four membership functions in each of four criteria. Table 2 shows a decision rule matrix example where C1 and C2 vary between "Poor" and "Excellent", C3 in this case is "Average" and C4 is "Excellent". As seen, the output values vary from "Maybe not" to "Yes". 16 tables such as this example were developed to cover all possible combinations. Table 2. Sample Decision Rule Matrix
C4
0.9, 1.0)
Figure 1. Computer Model I Membership Functions: Each criterion was assigned its set of membership functions as appropriate (see example in Figure 2). The membership functions are all triangular functions with a range from 0 to 1. Figure 2. Membership Functions for Criteria i Rules: Rules were entered in accordance with the mapping discussed earlier. The "And" method used is based on "product", the aggregation on "maximum" and the defuzzification on "centroid".
3.2
William E. Combs developed an alternative rule configuration for mapping problem spaces in control situations. He presented his findings on a truck backer-upper comparison of the two methods: the standard method (or as Combs refers to it the Intersection Rule Configuration), and the Union Rule Configuration method [22]. The latter, which will be referred to as Combs' method, is simpler to build and faster to process. The idea is based on the following equivalence: [(p q) => r] [(p => r) V (q => r)] where p and q are the antecedents and r is the consequent. In the above equation represents the intersection or the logical AND and V represents the union or the logical OR (for further detail see [2226]. In Combs method each implication has only one antecedent and one consequent. An example of an implication using Combs' method is: If Relationship is High Then Score is High As recommended by Mendel in his comments [24], this paper aims at going past the mathematical proof of the logic and verifying whether it is possible to use Combs' method in areas other than controls (i.e. in the area of supplier selection). To do so, the inputs (C1, C2, C3, and C4) and the output (score) from Model I were maintained. Similarly, the same fuzzy set values for both inputs and output were maintained. In order to maintain consistency and a basis for comparison, the methodology used in Model I was used again with the exception of the FLS mapping. FLS Mapping: In Combs' method, every antecedent
consequent. In other words, every membership function of every criterion implies an output. These implications are then aggregated using the logical OR. The problem space mapping is shown in Table 3.
Criter ia
Decision maker D1 (0.437 9, D2 (0.210 2, 0.263 4, 0.262 7) (0.210 2, 0.263 4, 0.262 7) (0.263 4, 0.262 7, 0.262 7) (0.210 2, 0.263 4, 0.262 7) D3 (0.151 9, 0.170 9, 0.189 9) (0.170 9, 0.189 9, 0.189 9) (0.170 9, 0.189 9, 0.189 9) (0.170 9, 0.189 9, 0.189 9)
C1
C2
C3
C4
0.492 7, 0.547 4)
is directly mapped to a
It should be noted that we are not using Combs method with a strict sense of mathematical completeness. As pointed out in [25], for Combs method to be universally equivalent to the traditional FLS, it must be modified with corrective rules if the rule matrix is not additively separable. In our
example we violate this condition and do not use the corrective rules, however we believe we obtain reasonable results without absolute mathematical rigor. Table 3. Combs' Method Decision Rule Matrix
If Or Or C3 Or C4 Then Output No M-N M-Y Yes P A G E P A G E C1 C2 P P A A G G E E
H1: D 0 The t statistic was computed and found to be equal to 1.918469. For n = 50, and level of significance = 0.05, t49 = 2.0096. Therefore, since t < t49, there was failure to reject the null hypothesis and it was concluded that there was evidence to suggest that there was no difference in the population means. The three dimensional control surfaces in Figure 3 show the output Score that corresponds to select combinations of sets of inputs.
Figure 3. Surface Plot of Output When the scores from both models were ranked, the results showed some differences. For instance, if the first 5 ranks were to be examined, it would be noted that Model I provided the following sequence: S25, S33, S20, S16, and S18. Model II provided the following sequence of ranking: S20, S25, S18, S33, S31. S16 was ranked as # 4 in Model I but ranked as # 8 in Model II. S31 was ranked # 5 in Model II and equally # 5 in Model I (S18 and S31 were both ranked # 5). Although the models do not provide perfect duplication in rankings, the same suppliers are clustered at the top for each method. Similarly, when considering the lowest ranking suppliers, Model I provided the following sequence: S49, S50, S47, S2, and S5. Model II ranks the lowest ranking suppliers in the following sequence: S2, S4, S10, S49, and. S4 and S10 both ranked equally as # 34 in Model I. S47 and S50 ranked as # 32 and # 40 respectively in Model II. Both the statistical analysis and the ranking of the results suggest that there is an association between the two models. Both methods show the same 4 out of 5 selected suppliers in the top 5 group.
http://www.watlow.com/reference/guides/0206.cfm [5] Verizon Wireless, Website at http://verizonwireless.com/jsp/aboutus/supplier_div/ criteria.jsp [6] Saaty, Thomas L., The Analytic Hierarchy Process, McGraw-Hill, 1980. [7] Muralidharan, C., N. Anantharaman, and S.G. Deshmukh, "A Multi-Criteria Group Decisionmaking Model for Supplier Rating", The Journal of Supply Chain Management, Vol. 38 No. 4, Fall 2002. [8] From Computing with Numbers to Computing with Words -- From Manipulation of Measurements to Manipulation of Perceptions, IEEE Transactions on Circuits and Systems, 45, 105-119, 1999. [9] Wang, J. and Y. I. Lin, A fuzzy multicriteria group decision making approach to select configuration items for software development, Fuzzy Sets and Systems, March 2003. [10] Marco, Beccali, Cellura Maurizio, and Ardente Davide, Decision Making in Energy Planning: The ELECTRE Mutlicriteria analysis approach compared to a fuzzy-sets methodology, Energy Conversion Management, Vol. 29, No 16-18, 1998. [11] Zadeh, Lotfi A., Inf. Control 8, pp: 338-353, 1965. [12] Mendel, Jerry M., Uncertain Rule-Based Fuzzy Logic Systems Introduction and New Directions, Prentice-Hall, Upper Saddle River, NJ, 2001. [13] Passino, Kevin, and Stephen Yurkovich, Fuzzy Controls, Addison-Wesley. [14] Jamshidi, M., A. Titli, L. Zadeh, and S. Boverie, Applications of Fuzzy Logic, Prentice Hall, 1997. [15]Yen, J., R. Langari, and L. Zadeh, Industrial Applications of Fuzzy Logic and Intelligent Systems, IEEE Press, 1995 [16] Rinzel, Lawrence, Fuzzy Logic Query in a Medical Database, MS Thesis, California Polytechnic State University, September 1999. [17] Martin, Hilary, An Expert Systems Alternative to Using Hurdle Rates for Project Rate of Return, MS Thesis, California Polytechnic State University, March 2001. [18] Von Altrok, Constantin, Fuzzy Logic and Neurofuzzy Applications in Business and Finance, Prentice Hall, Upper Saddle River, NJ, 1997. [19] Pacific Northwest National Laboratory, Resource and Ecosystems Management, Environmental Technology Division, Richland WA, Science and Technology for Land and Ecosystems Management, available online at http://www.pnl.gov/ecology/Science/land.pdf, last accessed June 2003.
[20] DeTurris, D., J. Ervin and S. Alptekin, Development of an Autonomous Tactical Reconnaissance Platform, Proceedings of 17th AIAA Aerodynamic Decelerator Systems Technology Conference, May 20-22, 2003, Monterey, California. [21] Sarkis, Joseph, Srinivas Talluri, "A Model for Strategic Supplier Selection", The Journal of Supply Chain Management, Winter 2002. [22] Combs, William E. and James Andrew, Combinatorial Rule Explosion Eliminated by a Fuzzy Rule Configuration, IEEE Transactions on Fuzzy Systems, Vol. 6, No. 1, 1998. [23] Dick, S. and Abraham Kandel, A comment on Combinatorial Rule Explosion Eliminated by a Fuzzy Rule Configuration by William E. Combs and James Andrews, IEEE Transactions on Fuzzy Systems, Vol. 7, No. 4, pp: 475-477, 1999. [24] Mendel, J. M. and Qilian Liang, Comments on Combinatorial Rule Explosion Eliminated by a Fuzzy Rule Configuration, IEEE Transactions on Fuzzy Systems, Vol. 7, No. 3, pp: 369-373, 1999. [25] Weinschenk, J. J., W. E. Combs and R. J. Marks II, Avoidance of Rule Explosion by Mapping Fuzzy Systems to a Union Rule Configuration, FUZZIEEE, accepted for publication, 2003.