Abstract

The thesis develops novel and proactive optimal control policies for a partially observable facility, which is subject to stochastic disruptions. Unlike traditional Supply Chain
Networks (SCN), where established facilities are considered to be continuously available, a more practical scenario is developed. More specifically, in the proposed frameworks, the
aforementioned assumption is relaxed such that the facilities are subject to stochastic disruptions potentially leading to costly failures. In such practical scenarios, it is critical and of paramount importance for the established facilities to operate with the highest achievable
reliability in the presence of disruptions and degradation. In this regard, this thesis provides a conceptual framework to obtain an optimal control policy for an already established facility subject to stochastic disruptions/degradation such that the disruptions have a direct effect on the connection links within the SCN. The level of degradation of a facility is modeled as a N state continuous time hidden-Markov process with N −1 operational and unobservable states together with one observable failure state. The facility is monitored periodically to observe the level of degradation. If the degradation level exceeds a critical state, a preventive action, namely partial fortification, will be performed. On the other hand, when the degradation level exceeds the failure state, a corrective action, namely full fortification, will be performed which brings the facility to the healthy state. The model is extended to the scenario where an integrated model of Statistical Process Control (SPC) and maintenance planning is considered and the optimal control limit policy is achieved based on a novel Bayesian control chart. The control problems under consideration are formulated in a Partially Observable Markov Decision Process (POMDP) framework to find the optimal preventive level in order to minimize the long-run expected average cost. A comprehensive sensitivity analysis is performed to evaluate the performance of proposed
models.