یک رویکرد یکپارچه به طراحی و راه اندازی برای قطعات یدکی سیستم های لجستیک

This paper attempts to solve a comprehensive design problem for a spare part logistic system. The design factors encompass logistic network design, part vendor selection, and transportation modes selection. Two approaches to solve the problem were proposed. In Approach 1, we simultaneously considered all the design factors and proposed two algorithms (SGA-1 and TGA-1). In Approach 2, the design problem was solved in two stages. Firstly, we aimed to find a near-optimal logistic network. Secondly, with the obtained logistic network, we proposed three algorithms (SGA-2, TGA-2, and NN-GA-Tabu) to find optimal combinations for part vendor and transportation modes selection. Numerical experiments indicate that Approach 2 outperforms Approach 1, and the NN-GA-Tabu outperforms all the other four algorithms. The proposed NN-GA-Tabu might also be a good solution architecture for solving other comprehensive space search problems.

Spare part management is a very important issue for capitally-intensive industries (e.g., semiconductor manufacturing, aerospace, defense, and high-speed train). Building a leading-edge semiconductor wafer fab may cost up to 2 billion dollars; and the associated spare parts inventory may need 10–15% of the total expenditure. Other capitally-intensive industries also reveal the same characteristics. Thus, the design and operation of a spare part logistic system is very important for these industries. A spare part logistic system (also called a logistic network) typically involves a group of stations that are hierarchically structured as shown in Fig. 1. In the hierarchy, terminal stations, essentially designed to repair machines in the service field, are equipped with machine-repairing staffs and spare parts inventory. Other higher-layer stations are designed to store and repair spare parts in order to supply spare parts to terminal stations. Parts delivery between any two stations needs a transportation time. In literature, such a logistic network is characterized as a multi-echelon system (Sherbrooke, 1968) As shown in Fig. 2, a machine typically comprises a hierarchical assembly of parts – called bill of materials (BOM). In literature, a spare part logistic system that considers only one kind of part is called a single-indenture system. In contrast, a multi-indenture system is a spare part logistic system that considers a BOM hierarchy involving many kinds of parts. This research is concerned with a multi-indenture, multi-echelon (simply called MIME) spare part supply chain system. Several survey papers on spare part logistics in a MIME system have been published (Guide and Srivastava, 1997 and Kennedy et al., 2002). Prior studies could be essentially grouped in two categories. One category aimed to find optimal operation policies for a given spare part logistic system; that is, how to determine optimal inventory level and repair-staff level for each station in order to reduce the total operational cost. Some assumed that each station is equipped with an infinite staffing capacity for repairing parts; and paid attention to the decision of stocking levels. The pioneer one is the METRIC model developed by Sherbrooke (1968); many of its extensions have been developed (e.g., Graves, 1985, Muckstadt, 1973 and Sherbrooke, 1986). Given a finite staffing capacity for repairing parts, some others investigated the decision for optimum stocking levels (e.g., Diaz and Fu, 1997, Kim et al., 2000 and Perlman et al., 2001). Extending the frontier, Sleptchenko, van der Heijden, and van Harten (2003) aimed to solve a more complex problem – finding an optimum combination for both repair-staff capacities and stocking levels. The other category attempted to find an optimal design for a spare part logistic system. Some aimed to design an optimal logistic network (Candas and Kutanoglu, 2007, Jeet et al., 2009 and Rappold and van Roo, 2009); some focused on optimal selection of part vendors (Wu & Hsu, 2008); and some others examined optimal selection of transportation modes (Kutanoglu & Lohiya, 2008). Such design factors were only partially addressed in prior studies. Their obtained solutions might leave a space for further improvement if more design factors are simultaneously addressed. Yet, such a comprehensive inclusion of design factors may require formidable computational efforts. In this paper, we attempt to solve a comprehensive design problem for a spare part logistic system. The design factors encompass logistic network design, part vendor selection, and transportation modes selection. Two approaches to solve the problem were proposed. In Approach 1, all the design factors are simultaneously considered. That is, a new solution could be generated by varying the selection for any of the design factors. Based on such a solution representation, two meta-heuristic algorithms were proposed to solve the design problem. The two algorithms, adapted from literature ( Goldberg, 1989 and Tsai et al., 2004), are respectively called SGA-1 (simple genetic algorithm in Approach 1) and TGA-1 (Taguchi genetic algorithm in Approach 1). Approach 2 decomposes the design problems into two sub-problems. That is, we solve the design problem in two stages. In stage 1, we focus on finding a near-optimal logistic network, by the application of a technically sound heuristic rule. In stage 2, with the obtained logistic network, we proposed three meta-heuristic algorithms to find optimal combinations for part vendor and transportation modes selection. The three algorithms are called SGA-2 (simple genetic algorithm in Approach 2), TGA-2 (Taguchi genetic algorithm in Approach 2), and NN-GA-Tabu (neural network-genetic algorithm-tabu-search). Numerical experiments indicate that Approach 2 outperforms Approach 1. This advocates the use of a problem-decomposition approach in solving a large-scale problem, if a technically sound heuristic rule can be found. Of the three algorithms in Approach 2, the NN-GA-Tabu outperforms the other two both in solution quality and computation time. We developed the NN-GA-Tabu based on two ideas. First, we develop an efficient yet rough performance evaluator to quickly justify a solution. Second, we use GA to find a quality solution and then use a tabu-search (a local tuning process) to obtain an improved one. The remainder of this paper is organized as follows: Section 2 describes the problem in more detail. Section 3 formulates the comprehensive design problem and analyzes possible ways to solve the problem. Section 4 describes the two algorithms in Approach 1. Section 5 describes the solution architecture of Approach 2 and the proposed NN-GA-Tabu algorithm. Experiment results of all the five algorithms are compared in Section 6. Concluding remarks are in the last section.

Several survey papers on spare part logistics in a MIME system have been published (Guide and Srivastava, 1997 and Kennedy et al., 2002). Prior studies could be essentially grouped in two categories. One category aimed to find optimal operation policies for a given spare part logistic system; that is, how to determine optimal inventory level and repair-staff level for each station in order to reduce the total operational cost. Some assumed that each station is equipped with an infinite staffing capacity for repairing parts; and paid attention to the decision of stocking levels. The pioneer one is the METRIC model developed by Sherbrooke (1968); many of its extensions have been developed (e.g., Graves, 1985, Muckstadt, 1973 and Sherbrooke, 1986). Given a finite staffing capacity for repairing parts, some others investigated the decision for optimum stocking levels (e.g., Diaz and Fu, 1997, Kim et al., 2000 and Perlman et al., 2001). Extending the frontier, Sleptchenko, van der Heijden, and van Harten (2003) aimed to solve a more complex problem – finding an optimum combination for both repair-staff capacities and stocking levels. The other category attempted to find an optimal design for a spare part logistic system. Some aimed to design an optimal logistic network (Candas and Kutanoglu, 2007, Jeet et al., 2009 and Rappold and van Roo, 2009); some focused on optimal selection of part vendors (Wu & Hsu, 2008); and some others examined optimal selection of transportation modes (Kutanoglu & Lohiya, 2008). Such design factors were only partially addressed in prior studies. Their obtained solutions might leave a space for further improvement if more design factors are simultaneously addressed. Yet, such a comprehensive inclusion of design factors may require formidable computational efforts. In this paper, we attempt to solve a comprehensive design problem for a spare part logistic system. The design factors encompass logistic network design, part vendor selection, and transportation modes selection. Two approaches to solve the problem were proposed. In Approach 1, all the design factors are simultaneously considered. That is, a new solution could be generated by varying the selection for any of the design factors. Based on such a solution representation, two meta-heuristic algorithms were proposed to solve the design problem. The two algorithms, adapted from literature ( Goldberg, 1989 and Tsai et al., 2004), are respectively called SGA-1 (simple genetic algorithm in Approach 1) and TGA-1 (Taguchi genetic algorithm in Approach 1). Approach 2 decomposes the design problems into two sub-problems. That is, we solve the design problem in two stages. In stage 1, we focus on finding a near-optimal logistic network, by the application of a technically sound heuristic rule. In stage 2, with the obtained logistic network, we proposed three meta-heuristic algorithms to find optimal combinations for part vendor and transportation modes selection. The three algorithms are called SGA-2 (simple genetic algorithm in Approach 2), TGA-2 (Taguchi genetic algorithm in Approach 2), and NN-GA-Tabu (neural network-genetic algorithm-tabu-search). Numerical experiments indicate that Approach 2 outperforms Approach 1. This advocates the use of a problem-decomposition approach in solving a large-scale problem, if a technically sound heuristic rule can be found. Of the three algorithms in Approach 2, the NN-GA-Tabu outperforms the other two both in solution quality and computation time. We developed the NN-GA-Tabu based on two ideas. First, we develop an efficient yet rough performance evaluator to quickly justify a solution. Second, we use GA to find a quality solution and then use a tabu-search (a local tuning process) to obtain an improved one. The remainder of this paper is organized as follows: Section 2 describes the problem in more detail. Section 3 formulates the comprehensive design problem and analyzes possible ways to solve the problem. Section 4 describes the two algorithms in Approach 1. Section 5 describes the solution architecture of Approach 2 and the proposed NN-GA-Tabu algorithm. Experiment results of all the five algorithms are compared in Section 6. Concluding remarks are in the last section.