Supply chain network design in the presence of disruption risks.

摘要

Supply chain disruptions come with catastrophic consequences in spite of their low probability of occurrence. One of the most effective ways to hedge against network disruptions is to have a robustly designed supply chain network since contingency plans in the event of a disruption are often significantly limited. This dissertation focuses on systematically understanding the impact of disruptions and exploring robust supply chain network designs for which the impact of disruption is minimized.;This dissertation is comprised of three main problems. The first problem considers a facility location problem in the presence of random facility disruptions where the facilities can be protected with additional investments. The problem is first formulated in a discrete (mixed integer programming) model and an effective solution method for large size of problem instances is developed. Next the problem is formulated in a continuous approximation model which enables us to extract valuable insights. We derive a threshold disruption probability above which the optimal policy is to harden all facilities, and below which a mix of hardened and non-hardened facilities are deployed. Further, we investigate the impact of misestimating the disruption probability and show that underestimating the disruption probability can be very dangerous. We derive an optimal estimate to use when the true disruption probability is difficult to obtain.;The second problem considers a robust supply chain network design in which the links and nodes are susceptible to random disruptions. Based on the concept of chaining introduced by Jordan and Graves (1995), we show that the networks with longer chains are not always superior to the one with shorter chains in the presence of disruptions. In particular, we show that a network with shorter chains is more robust against a single link failure than the one with longer chains, while a network with longer chains is more robust against a single node failure. For the case in which multiple failures are likely, we show that having shorter chains is beneficial. We also show that investing in some level of excess capacity is essential as doing so makes the network less fragile with respect to disruptions.;The last problem considers premeditated risks in designing supply chain networks. Such a problem, a game involving two players, is often formulated in a bilevel optimization model and is quite challenging to solve. Using a defender-interdictor location problem introduced by Scaparra and Church (2008), we develop an effective and efficient genetic algorithm heuristic to solve these problems. We show that the proposed algorithm is quite promising for solving large-scale problems. The proposed algorithm does not employ any problem-specific features and thus can be easily extended to a wide range of applications.