| In an increasingly competitive world, a major problem confronted by management is the design of their supply chain networks so as to minimize costs and time to market while meeting stringent customer requirements. Once a design or a set of alternative designs is in place, management is still faced with the task of detailed analysis of their supply chains in a variety of scenarios, involving hundreds or thousands of items. The first portion of this thesis presents a methodology for supply chain design to generate viable network alternatives which can then be subject to further analysis. In the second part of the research, a clustering methodology to identify groups of similar items which can be used to support efficient inventory analysis of supply chains is provided.; The approach used for the design of supply chain networks is based on genetic algorithms (GA) where the goal is to identify the set of locations and the flow of material in the network. The objectives included in this model are the minimization of cost and cycle time. Extensive experiments to demonstrate the effectiveness of the approach and the ability to generate diverse design alternatives are presented. The main contributions of this methodology are (i) the ability to consider multiple objectives explicitly, and (ii) the ability to incorporate the stochastic elements inherent in supply chains.; The clustering methodology identifies groups of similar items which can then be used to determine very good approximations of inventory levels required to support a given service level. Two distinct features of this approach are the ability to consider items spread across multiple locations and the ability to capture the relationships between the items using a set of heuristics. Examples are provided to demonstrate the effectiveness of the methodology and the performance of the heuristics, by comparing the results obtained with the optimal solution. Applications of this methodology presented in this research include inventory-service level tradeoff analysis, forecast variability analysis, and commonality analysis. This thesis also includes a case study using data drawn from the computer industry to demonstrate the usefulness of the methodology and its advantages over the commonly used ABC classification method. |
