Meta-Heuristic Approach for Supporting Design-for- Disassembly towards Efficient Material Utilization

Transcription

1 Meta-Heuristic Approach for Supporting Design-for- Disassembly towards Efficient Material Utilization Yoshiaki Shimizu *, Kyohei Tsuji and Masayuki Nomura Production Systems Engineering Toyohashi University of Technology, Aichi Japan Abstract Efficient material utilization is a rational way to facilitate sustainable technology, because it can promote conservation of resource and energy while decreasing the environmental loads at the same time. Especially, recycling and remanufacturing are its important forms that involve product disassembly to retrieve the desired parts and/or subassemblies. With this point of view, in this paper, we have developed a prototype system supporting a strategic decision making on disassembly for recycling at design stage of product life cycle. Based on the information given at the design stage, it decides automatically the optimal disassembly sequence through a meta-heuristic method known as Genetic Programming (GP). Its implementation is taken place by using Java programming language, and an illustrative example is provided to show some features. Keywords: design for disassembly, genetic programming, material re-utilization, sustainable technology, recycle 1. Introduction Due to our conventional economic activities and life style such as mass-production, mass-consumption, and mass-throw-away, we have been confronting to various environmental problems today. Accordingly, many efforts have been begun with demonstrating a new idea for reflecting the environmental consciousness to manufacturing activities. Among them, LCA (Life Cycle Assessment) standardized in IS series is being applied in various industries recently. Some efforts have been also paid to provide an information infrastructure for deploying a decision-aid in quick and understandable manners (Shimizu et al., 2001, 2002). However, its resolution is becoming very complicated and complex under the global competition of economy and the egoism referring to what is known as the south-north problem. Under such circumstances, efficient material utilization through product disassembly to retrieve the desired parts and/or subassemblies becomes a rational way to be taken since it can promote the conservation of resource and energy while decreasing the environmental loads at the same time. However, because multiple parts made of different materials are to assemble products, disassembly load evaluated by time and/or cost becomes a key factor that may obstruct the recovery of the materials. * Author to whom correspondence should be adressed : shimizu@tutpse.tut.ac.jp

2 With this point of view, this paper focuses on product recycling and remanufacturing, and addresses an approach associated with the design-for-disassembly (DfD; Navinchandra, 1991), a reverse engineering considering easiness of disassembly for recycling at the design stage of product life cycle. After general statement about whole idea, we will describe a resolution procedure in detail, and finally show some features of a developed prototype system through an illustrative example. 2. Problem Statement Since the DfD task will not only increase product value at the end of life cycle but also bring about the consciousness about life cycle cost, there is a keen need to develop a user friendly evaluation tool for it. Navinchandra (1991) developed a system that assists designers in identifying recycling-friendly design. A graph search method in terms of ANA/OR graph is used to identify the optimal recycling strategy (Zussman, Kriwet, and Seliger, 1994). For the modelling of disassembly planning, a generic Petri net model is applied and analysed based on the reachability graph regarding all possible disassembly processes by Xirouchakis and Kiritsis (1997). A geometrical analysis is taken place to find out a possible disassembly sequence for the objects whose parts are made of unit cubes through an idea called face coupling (Beasley and Martin, 1993). Similarly, generating a hierarchical disassembly tree from geometrical information, branch and bound algorithm is applied to reduce a tree size, and to derive a disassembly sequence plan (Askiner and Gapta, 2001). Moreover, a multi-objective optimisation model for a disassembly to order is proposed and solved using integer goal programming by Kongar and Gupta (2002). Under mild conditions, we will present below a practical solution method to cope with this kind of problem that seems to stay still at the state of concept. GUI Input Information on product part/subassembly disassembling facility Disassembling time D B Evaluation Systtem for DfD GUI Output Disassemble sequence Score for DfD Disassembling time Disassembling cost (Environmentally benign score) Environment factor DB Disassembling cost DB On going 2.1 General idea of Fig.1 Scheme of the proposed system. system development The aim of the study is to develop a prototype system supporting strategic decision making for efficient material utilization. Through simulating the disassembly sequence based on the information given at the design stage, it decides automatically a disassembly sequence that is optimal under the prescribed objective function. Moreover, by examining the result in detail, we can reveal some issues necessary for improving the disassembly activities and reflect them to a new product design. In Fig.1, we show a scheme of such system that will realize our general idea. By giving some product specific data, in the aid of the equipped DBs, the evaluation system outputs both the optimum disassembly sequence and a few quantitative indices helpful for the reverse engineering.

3 2.2 Genetic programming as a solution method Although evaluation procedure, as the core of the above system, can refer mathematically to an optimization problem regarding disassembly sequence, its solution will become extremely difficult according to the expansion of problem size due to its combinatorial nature. Hence we apply an approximate method that is possible to obtain a near optimal solution within a reasonable computation efforts. In addition, because the disassembly sequence can be represented conveniently as a list of function (disassembly method) and terimal (part/subassembly), the algorithm possible to manage easily such structural expression (tree structure) is relevant. From all of these, we adopt a meta-heuristic method known as Genetic Programming (GP) presently. GP [Koza 1992] is a branch of genetic algorithm (GA), and works better for the particular type of problems trying to find the most efficient solution from a large list of variables. GP creates a computation scheme as the solution, i.e., compositions of the functions and terminals of the present disassembly sequence problem while GA creates a string of numbers that represent decision variables. A usual algorithm of GP is composed of the following four steps quite similar to GA: 1) Generate an initial population with random compositions of the functions and terminals. 2) Evaluate each scheme in the population and assign it a fitness value. 3) Create a new population of schemes. i) Reproduce the superior schemes. ii) Create new schemes by crossover. iii) Create new schemes by mutation. 4) The best scheme obtained after the prescribed generation is chosen as the result of GP. Preliminary statements Noticing the tree structure is composed of branches and leafs, we can represent the disassembly sequence conveniently by viewing the branch as disassembling action and the leaf as disassembled part. In addition, associated with the solution method like GP, we make each of them correspond to the function and terminal respectively. Presently, we classify * parts (terminals) into six categories, i.e., uniform material, unit part, reusable unit, part necessary for treatment before disposal, part hard to destroy, and disposable part while disassembly actions (functions) into five operations, i.e. no action, snap off, screw off, turn off, and cut off. Figure 2 shows an example of such representation. p6 d3 d4 p3 p2 p4 p5 d2 d1 d5 d i F Function (i=1,,5) p i F Terminal (i=1,,6) Fig.2 Tree representation of disassembly sequence. Components of GP and solution algorithm Based on the foregoing discussion, we code the chromosome for the tree in Fig.2 d1 d2 d3 d4 p6 p5 p3 d5 p2 p4 Fig.3 Chromosome denoting tree in Fig.2. * Since these are instances of each class, we can add easily anyone depending on the problem for further consideration.

4 as shown in Fig.3 where each gene is given according to the depth first rule by looking at the tree from the left hand side. Decoding is realized just by the reverse procedure of the coding. We apply a roulette strategy with the elitism as the reproduction rule. Regarding the crossover and mutation, they are to be taken place on the basis of subtree structure. Hence, it needs to satisfy such condition that both number and elements involved in the subtrees must be identical with each other before and after these genetic operations. Under this condition, the crossover is done as follows (See also Fig.4). 1. By taking one of the parents, cut down the sub-tree randomly. 2. Count the number of terminals involved in the chosen sub-tree. 3. Decide the sub-tree with the same number of terminals as the above for the other parents. 4. Exchange the sub-trees between thus decided ones. 5. If there happens the case where some terminals duplicate with each other, go to step 6. Otherwise stop. 6. Replace the duplicated terminal with the missing one. Apply this in order of the sequence of genotype in the case of multiple duplications. On the other hand, we adopt two kinds of mutation to bring about new changes that are stiff only by the crossover operation. One of them can generate a new p2 P2 p3 p6 p3 p4 p5 p6 structure even with different number of terminals between before and after, and the other with drastic change regarding the terminals. Moreover, based on the weighted sum of dimensionless disassembly cost and time, we define the score F by Eq.(1) where superscript and subscript of the asterisis denote an ideal and nadir values respectively. To turn it into a fitness, we degrade it like Eq.(2) by introducing a penalty against physical infeasiblility arising from the priority on the disassembly sequences. Though the priority is assumed to be given a priori, we are planning to derive it from the geometric information of the product. There v represents a binary value that takes 1 if there involves a disassembly sequence violating the given priority and 0 otherwise. Moreover, coefficient a denote a constant, and L level in the tree (L root = 0) where such violation is appeared. Hence the fitness F can stand for the situation where the more and at the shallower level the violations occur, the more seriously the infeasibility will encounter. p4 Parents p5 p5 [p2: duplicated, p5: missing] Replace p2 with p5 [p2: missing, p5: duplicated] Replace p5 with p2 Fig.4 Employed crossover operation. P2 p6 p3 p4 p5 p6 Offsprings p3 p4 p2

5 * * Time Time Cost Cost = w1 + w 100 Time * Time Cost * Cost * * F 2 (1) av (1 ) L F' = F (2) 2.3 Implementation of prototype system By using Java programming language, we design a few GUIs to input data of the product to be assessed, and one to output the results, e.g., visual display of the disassembly sequence and the evaluated score. Through such GUI, we can input name, number, material, and category of each subassembly, and prescribe the relation among the subassemblies regarding the method of connection and priority relation on the disassembly sequence. After setting such information properly and pushing ok bottum from the main menu, output display shown in the later will appear as a result of the optimization through GP. 3. Illustrative Example We examined the effectiveness of the proposed system through numerical experiments taking a few example products. In what follows, we show a case study taking a personal computer for example. Setting parameters of GP as shown in Fig.5, and giving some additional data, we carried out the numerical experiment and confirmed the effectiveness of the proposed approach. We Fitness Generation population size:100 total generation:1000 crossover rate:0.75 mutation rate:0.25 show a feature of convergence Fig.5 Convergence feature in GP. along with generation in Fig.5, and the output GUI in Fig. 6 The metric in the value column is helpful for evaluating the desireability from the DfD viewpoint among the candidate products and deciding the final design as a green product. 4. Conclusion As a key issue for the environmentally conscious manufacturing, we have focused on product recycling and remanufacturing, and developed a prototype system considering disassembly strategy for recycling. We applied GP as a resolution method, and defined the score of disassembly or an achievability to compare quantitatively the features of disassembly among candidates of DfD products. Furthermore, we designed some GUIs to input data and to review the results efficiently. Finally, we have shown the significance of the proposed system through numerical experiments.

6 Fig.6 Output GUI for computation result. References Navinchandra, D., 1991, Design Theory and Methodology, 31, 119. Beasley, D. and R. R. Martin, 1993, Disassembly sequences for objects built from unit cubes, Computer-Aided Design, 25, 751. Zussman, E. A. Kriwet and G. Seliger, 1994, Disassembly-oriented assessment methodology to support design for recycling, Annals of the CIRP, 43, 9. Xirouchakis, P. and D. Kiritsis, 1997, Petri net modelling for disassembly process planning, ASME, De-94/Med-5, 255. Gungor, A. and S. M. Gupta, 2001, Disassembly sequence plan generation using a branch-and-bound algorithm, Int. J. Prod. Res., 39, 481. Shimizu, Y., K. Kainuma, T. Kitajima, 2001, A Prototype System for Evaluating Life Cycle Engineering of Chemical Products, J. Chem Engng. Japan, 34,.676. Shimizu, Y., Y. Miyata and E. Ishihara, 2002, An Infrastructure for Integrating Element Technologies of Life Cycle Engineering, J. Chem Engng. Japan, 34, 810. Kongar, E. and S. M. Gupta, 2002, A multi-criteria decision making approach for disassembly-to-order systems, J. Electronics Manufacturing, 11, 171.