| In this thesis we address the problem of supply chain management from a decentralized and a centralized perspective. The analysis is based on a general dynamic characterization of supply chains and on a novel modeling framework that considers the simultaneous flow of information (orders) and materials (products) within the supply chain, a feature that differentiates this study from the common available planning models. In the decentralized approach we propose a novel hybrid modeling framework that models the distribution networks with differential equations and the manufacturing sites as finite states or switching systems. The proposed hybrid model is a simulation tool to evaluate the performance of different control laws for the distribution channels and heuristic scheduling rules. We use this tool to show that it is possible to reduce the “bullwhip effect” with simple control strategies in the distribution network, as long as the delay times are not too long. For long delays, it is necessary to use some kind of anticipatory control or share information about the demand to achieve this goal.; Regarding the centralized approach, we propose a multiperiod MILP representation that takes into account the intrinsic delay times in processing orders. For this approach we propose using a rolling horizon algorithm, such as the one used in Model Predictive Control (MPC), for the evaluation of robust and effective action plans to control supply chains with long delays. Here we show that the centralized approaches yield better results than decentralized ones. Finally, by means of a sensitivity analysis we establish a series of heuristic rules for tuning the MPC parameters in such a way that they define a problem of tractable size that gives meaningful solutions and reduces the solution time and the inherent myopic nature of the algorithm. |
