CHAPTER 2 LITERATURE SURVEY

Transcription

1 13 CHAPTER 2 LITERATURE SURVEY 2.1 INTRODUCTION The importance given to the supply chain management over the past two decades can be understood from the surge in the number of publications in that area. The literature on supply chain management was reviewed by Schmidt and Wilhelm (2000) based on strategic, tactical and operational decisions and they elaborated on various supply chain modeling issues. Croom et al (2000) presented a framework for the categorization of literature linked to supply chain management based on content and a methodology-oriented criterion. Supply chain management is a vast area comprising many key elements which involves demand allocation, manufacturing decisions, inventory management, facilities location and determination of transportation policy. Among these crucial elements, production and distribution planning is the area of focus in this research Research on Production and Distribution Planning Production planning and distribution planning were considered separately by the researchers since In most of the industries, production planning was devised initially based on demand and then the distribution plan for the devised production plan was determined. In the later part of 20 th century, when industries started adopting multi-site manufacturing in a move to reduce distribution expenses, the theme shifted to reduction of both

2 14 production and distribution cost simultaneously. Some of the literature on production planning and distribution planning are discussed in the detail in the following sections Survey on production planning Abundant of research articles have been published in the areas of aggregate production planning (APP). Since Holt, Modigliani, Muth and Simon proposed the HMMS (Holt, Modigliani, Muth and Simon) rule in 1955, researchers have developed numerous models to solve the APP problem, each with their own pros and cons. Pradenas et al (2004) defined APP as the simultaneous determination of production, inventory and the workforce levels of a company on a finite time horizon. Nam and Logendran (1992) surveyed the range of techniques available for setting aggregate plans and classified each methods in terms of their ability to produce either an exact optimal or a near - optimal solution. An offline planning and online control strategy was proposed by Wang et al (1994) for solving the APP model in the case of meeting seasonal demands for multi-product scenario. An optimal control approach to continuous-time APP problem was presented by Kogan and Khmelnitsky (1995). This approach allowed production smoothing, subcontracting / outsourcing and capacity evolution to be modeled on one hierarchical level. The above problem was tested in a soap manufacturing company using a new fast numerical algorithm. Wang and Fang (2001) presented a novel fuzzy linear programming (FLP) method for solving the APP problem. The other techniques involved are linear programming, heuristics approaches, spread sheets etc. Motivated by pinch analysis used in heat and mass exchanger network synthesis, Singhvi et al (2004) applied it for the production planning problem in the supply chains. Two case studies were solved, one involving single

3 15 product and the other involving multi-products. It is concluded that plans obtained by pinch analysis provide either the best aggregate plans or excellent starting points to reduce the computational time for solutions by mixed integer programming formulations. An APP was generated for a saw mill industry by Pradenas et al (2004) using a heuristic procedure based on tabu search. The overall revenue of the industry was maximized in the above work for a multi-product scenario. Kumar and Haq (2005) used a hybrid genetic-ant colony algorithm to solve the basic production plan. The performance of the algorithm was also compared with genetic algorithm and ant colony algorithm separately. The hybrid algorithm was reported to be superior to the original versions of the meta-heuristics. Jain and Palekar (2005) formulated an APP for a continuous manufacturing process producing different products using dissimilar machines. But similar operations were performed using those machines at different production rates. Also, different production lines could be formed by connecting the equipment differently. The problem was solved using CPLEX and heuristic approaches. The application of fuzzy multi-objective linear programming APP was studied by Wang and Liang (2004). This model considers the time value of money while minimizing the total production costs, carrying and backordering costs and rate of change in worker levels. Holt, Modigliani, Muth and Simon s Works were surveyed by Jaya and Kalyan (2007). The survey dealt with the application of Holt, Modigliani, Muth and Simon s Works in various companies. The survey also highlighted the importance of APP in managing the supply chain. Leung et al (2003) developed a multi-site APP for a lingerie manufacturing company without

4 16 considering distribution cost. The multi-objective problem was solved using goal programming approach. All the above mentioned literature deals with production planning part and does not focus on the distribution planning aspect of an industry. Also, the problems solved using heuristic and meta-heuristic approaches are over simplified Survey on distribution planning Distribution planning, on the other hand, represents an active research field. The distribution of products all along the supply chain, that directly influences the supply chain cost, is a key driver of the overall profitability of the firms. The cost of distribution plays a vital role in determining the price of the product. Hence, distribution is an important consideration for industrial firms in the supply chain network. In distribution planning, one of the strategic decisions is the allocation of products / supplies from production centres to terminal points in a cost effective manner. Often, unit transportation cost is used to find the optimal distribution schedule. A basic assumption in any transportation problem is that the cost of transportation is directly proportional to the number of units transported (Diaby 1991). Such approaches consider the total cost of distribution as purely quantity dependent and proportional to the amount transported between a source and a destination. Tripwise transportation costs were also considered by researchers. Many practical transportation and distribution problems were also modeled as fixed charge transportation problems (Adlakha and Kowalski 1999; Kim and Pardalos 1999; Sun et al 1998). Adlakha and Kowalski (1999) discussed the more-for-less analysis, which helped the decision maker to identify the markets and ship more or less products in fixed charge

5 17 transportation problems. Kim and Pardalos (1999) and Sun et al (1998) used heuristic approach and tabu search algorithm, respectively, to solve NP-hard transportation problems. But considering the production and distribution aspects of the industry separately may not yield the best results. Hence, the theme of research has shifted to integrated production-distribution planning since the late 1990s Survey on Integrated Production-Distribution Planning Many researchers have proposed different integrated productiondistribution models to minimize the overall expenses incurred in the supply chain. Most of the production-distribution problems focus on the gross production, production set-up, inventory holding and distribution costs or just inventory holding and distribution expenses. But production in any plant can be carried out through regular hours, overtime hours and also through outsourcing. In multi-site manufacturing problems, regular time production in most of the plants may be cheaper than overtime and outsourced production in all the plants. Allotting the demand to the plants and planning the distribution of products to the customers based on the average / gross production cost per unit of product using any logic or method will lead to allotting the demand completely to one plant before assigning to other plants resulting in incorrect demand allocation. Some of the production distribution models that do not categorize production as regular, overtime and outsourced are presented below. Ozdamar and Yazgac (1999) proposed a hierarchical aggregatedisaggregate model considering inventory, backorder and transportation costs. GAMS software was used to solve this MILP model. The effect of inflation and exchange rates on a multinational production-distribution problem was

6 18 modeled by Mohammed (1999). They framed a LP model to minimize the total cost which included manufacturing, inventory holding and distribution costs along with capacity expansion, retaining and decreasing costs. Yilmaz and Catay (2006) solved an MILP production-distribution problem using CPLEX and three linear relaxation heuristics. They minimized the costs associated with the production, transportation, inventory holding and capacity expansion. Rizk et al (2006) developed a multi-item flow planning problem between a manufacturing location and distribution centre for the dynamic demand (deterministic and time-varying) case. Three variants of the problem considering product changeover cost, inventory cost and transportation cost involving piecewise linear economics of scale were formulated as MILP models and solved using CPLEX. Bolduc et al (2006) formulated an MILP problem to minimize the transportation cost and the distribution centre s inventory holding cost. They considered a distribution centre and several retailers. The problem was solved using heuristic approaches and also CPLEX. The production plant infrastructure cost, distribution centre infrastructure cost, production cost (regular & outsourced), material handling cost at distribution centres and transportation cost along with duties cost were minimized by formulating an MILP model for a three echelon supply chain by Tsiakis and Papageorgiou (2008). The problem was solved using CPLEX and other similar software packages. Same results were obtained in all cases. Also, it can be understood that most of the above mentioned productiondistribution models belong to LP / MILP/ IP category. Vidyarthi et al (2007) proposed a nonlinear Mixed Integer Programming model that minimized the sum of fixed facility location costs, unit production and transportation costs along with safety-stock costs. A lower bound was obtained by the Lagrangean relaxation, while the heuristic proposed used the solution of the sub problems to construct an overall feasible solution within 5% of the optimal solution. It can be absorbed that almost all

7 19 the above mentioned production-distribution models were solved using commercial softwares like CPLEX, LINGO, LINDO, etc. Some of the researchers also used meta-heuristic and heuristic approaches to solve the production-distribution models. Keskin and Uster (2007) formulated an MILP model to minimize the total cost of a two stage production-distribution problem. The cost of locating distribution centres among plants and retailers along with transportation costs were minimized using meta-heuristic and heuristic approaches. Production set-ups, inventory and distribution costs were minimized for multi-period Integer Programming (IP) models by Boudia (2008). The single product case was solved using quick heuristic approaches (Boudia et al 2008) and meta-heuristic approaches like GRASP (Boudia et al 2007) and MA. The MA approach was found to yield good results (Boudia and Prins 2009). All the above mentioned models solved using heuristic / meta-heuristic approaches were simple in nature and do not replicate the real time industry scenario. Bilgen and Ozkarahan (2004) surveyed the production-distribution models and classified them in terms of the solution methodology used. They were: optimisation based models, meta-heuristic based models, information technology driven models and hybrid models. Some of the researchers also framed multi-objective productiondistribution models with the objectives of distribution time minimization, service level maximization, resource utilization maximization, etc., apart from total cost minimization (Chan and Chung 2004, 2005; Chan et al 2005; Liang 2008; Liang and Cheng 2009). Chan et al (2005) solved a linear productiondistribution model using hybrid GA and AHP. Gross production and distribution costs were considered by them along with other criteria like service level and resources utilization. They allotted the demand from the demand centres to the respective plants based on the size of the manufacturing

8 20 unit. The largest manufacturing facility is utilized completely before allocating the demand to the next bigger one and so on. If a demand could not be satisfied by any one facility, then that demand is split and allotted to the plants based on the availability of units in those plants. In addition to the gross production and distribution costs, inventory holding cost was also considered by Chan and Chung (2004, 2005) later in their work. They also considered a three stage supply chain model in contrast to their earlier two stage model. Liang (2008) and Liang and Cheng (2009) framed multi-criteria models with total cost minimization and distribution time minimization as objectives. Even in those fussy multi-objective (LP) models developed by Liang (2008) and Liang and Cheng (2009) production was classified into regular and outsourced alone. Very few production-distribution models were solved considering varying demand situations. Most of these models are over simplified or they do not classify the production into regular, overtime and outsourced. Lee et al (2002) applied a hybrid simulation analytic approach to minimize the production distribution cost in the supply chain. The production system consists of multi-shops making different products and the distribution system comprises of vehicle supplying to warehouses and retailers. The machine capacity and distribution capacity constraints in the analytic model were considered as stochastic factors and adjusted according to the results from independently developed simulation model. A similar approach was adopted by Almeder et al (2009). They found a robust plan for production, stocking, and transportation considering stochastic cost factors. However, both the above mentioned models do not consider the demand to be uncertain. 2.2 OPTIMIZATION OF PRODUCTION-DISTRIBUTION PLAN The integrated production-distribution models are either deterministic or probabilistic in nature. The probabilistic models are more

9 21 difficult to solve as the uncertainties associated with the parameters / decision variables need to be considered. The survey of the various optimization techniques applied for solving the models are given in the following sections Traditional and Meta Heuristic Approaches As mentioned earlier in section 2.1.2, most of the productiondistribution models belong to LP/MILP/IP category (Bolduc et al 2006; Boudia 2008; Boudia and Prins 2007; Boudia et al 2007, 2008; Chan et al 2005; Gen and Syarif 2005; Keskin and Uster 2007; Rizk et al 2006; Tsiakis and Papageorgiou 2008; Yilmaz and Catay 2006). The deterministic production-distribution models are mostly solved using commercial softwares like CPLEX, LINGO, LINDO, etc., or using heuristic methods. Nontraditional search techniques / meta heuristic approaches like GA, SA, MA and tabu search were also used by many researchers for solving the production-distribution models. But the production-distribution models solved using non-traditional search techniques were over simplified (Boudia and Prins 2009; Chan and Chung 2004, 2005; Chan et al 2005; Gen and Syarif 2005; Keskin and Uster 2007). Particle swarm optimization (PSO) / DPSO algorithms are the recently popular meta-heuristic approaches. Kennedy and Eberhart (1995) devised the PSO algorithm in the year Since then it has gained considerable importance and has been applied to wide variety of problems with excellent success rate (Chen 2011; Kuoa et al 2011). PSO and DPSO algorithms were mostly applied for solving scheduling problems by many researchers. Xia and Wu (2005) proposed a hybrid PSO and SA algorithm to multi-objective flexible job shop scheduling problems. Tseng and Liao (2008) applied the DPSO algorithm for the flow-shop scheduling problem involving lot splitting. Lei (2008) minimized the tardiness and total makespan of jobs for a job shop scheduling problem using PSO algorithm. Onut et al (2008)

10 22 used the PSO algorithm for designing the layout of multi-level warehouse problem and compared its solutions with that of LINGO solutions. Kannan et al (2009) applied PSO for minimizing variations in selective assembly. Bachlaus et al (2008) alone used PSO algorithm for solving productiondistribution problems. A hybrid taguchi-pso was proposed for solving a weight based multi-criteria production-distribution problem involving total cost, distribution volume flexibility and plant flexibility. It is evident from the above survey that few researchers have applied PSO / DPSO algorithms for solving production-distribution models. AHP proposed by Saaty (1980), is a well-proven multi-criteria decision making methodology, especially powerful for complex problems with a set of highly interrelated decision factors. AHP was used for evaluating suppliers by Chiang (2005). Lee et al (2006) applied AHP for evaluating the performance of IT department in the manufacturing industry. AHP was used for supply chain network design by Sha and Che (2006a, 2006b). AHP can also be clubbed with heuristics, meta-heuristics and other problem solving tools efficiently. Chan and Chung (2004, 2005) and Chan et al (2005) integrated AHP with GA to solve multi-criteria production-distribution problems. Wang et al (2010) and Che (2010) used AHP for framing weights to solve multi-criteria problems using PSO. But a complete integration with PSO algorithm was not done. Integrating AHP with PSO is different from integrating AHP with GA (as done by Chan and Chung (2004, 2005) and Chan et al (2005). This is because PSO works over a different logic as compared to GAs. Hence, a novel means of integrating AHP with DPSO algorithm is also proposed in this research Simulation and Other Hybrid Techniques Many researchers have proposed simulation / simulationoptimization approaches for modeling the uncertainties in the supply chain.

11 23 The advantages and disadvantages of analytic and simulation models were discussed by Shanthikumar and Sargent (1983). The importance of simulation in supply chain modeling was insisted by Stefanovic and Stefanovic (2008). Several simulation based models were proposed for inventory optimization in supply chains. Some of the reasons attributed to the choice of simulation to analyze supply chain inventory decisions were the ability for comprehensive modeling and the flexibility to incorporate uncertainty and dynamics (Ettl et al 2000; Petrovic et al 1998; Souza et al 2000). Rao et al (2000) developed integrated model to analyze different supply chain configurations for caterpillar s new line of compact construction equipment. They used simulation-based optimization method to establish inventory levels. Fuzzy logic based methods and hybrid simulation based metaheuristic approaches were also used by some researchers to incorporate uncertainties in the supply chains. Petrovic et al (1998) used fuzzy modeling and simulation for representing supply chain in an uncertain environment. Customer demand and supply of raw material were interpreted and represented by fuzzy sets and a supply chain simulator was developed. The simulator provided a dynamic view of the supply chain and assessed the impact of decisions recommended by the supply chain fuzzy models on supply chain performance. Truong and Azadivar (2003) developed an environment for solving supply chain design problems, where they combined simulation with genetic algorithms and mixed integer programs. Strategic decisions regarding facility location and partner selection were considered. Wang and Shu (2005) proposed a hybrid fuzzy genetic algorithm model to handle uncertainties in the supply chain. Inventory strategies were determined using the hybrid fuzzy genetic algorithm model. Xie et al (2006) designed a two-level hierarchical method to inventory management and control in serial supply chains, in which the supply chain operated under

12 24 imprecise customer demand and was modeled by fuzzy sets. A hybrid algorithm combining mathematical programming and simulation was framed by Byrne and Bakir (1999) for a multi-period multi-product production planning problem. Inventory cost in a serial supply chain was optimized using a hybrid genetic-simulation approach by Daniel and Rajendran (2004). Even though many researches proposed various models considering the stochastic elements in the supply chain, very few production-distribution models were solved. Few production-distribution models that consider stochastic nature of the supply chain are available (refer Lee et al. (2002) and Almeder et al. (2009) works discussed in section 2.1.2). However, both the above mentioned models did not consider the demand to be uncertain. In conclusion, a mathematical model that completely integrates regular time production, overtime production, outsourced production, inventory, backorder, hiring / laying-off of labour and distribution of products from manufacturing plants to demand centers do not exist. Also, a need for the development of a versatile algorithm that deals with multi-objective production-distribution problems for both deterministic and stochastic demand scenarios is also identified.