| This thesis focuses on the optimal production control of deterministic and failure-prone manufacturing systems. It is composed of three parts, the optimal control of reliable manufacturing systems, the optimal control of failure-prone manufacturing systems and adaptive hedging point control of failure-prone manufacturing systems.;In the first part, the optimal control for deterministic pull manufacturing systems with general buffer holding costs is investigated. By decomposing the machines in the system into different sections and layers, a set of constrained optimization subproblems is formulated. Then, the optimal control is obtained by solving these problems. This is a significant extension of previous results, which considered only increasing buffer level costs.;In the second part, the optimal control problem for scheduling failure-prone manufacturing systems under several classes of scheduling policies is considered. Although most of the literature concentrates on examining these systems by means of the Hamilton-Jacobi-Bellman (HJB) equation for a given cost criterion, our focus is on the determination of the steady-state joint probability density of the buffer levels, which can be described by a system of partial differential equations. By discretizing these equations, it is possible to reformulate the original problem as a complementarity problem. Although the dimension of the complementarity problem will often be large, an efficient sparse matrix-based numerical technique is provided to determine a near-optimal control for the system. A major contribution is that, unlike HJB-based methods, this numerical technique can be extended to obtained results for large manufacturing systems.;In the third part, the optimal control of a failure-prone, single machine manufacturing system producing multiple part-types is investigated. Under a class of control policies called prioritized hedging point (PHP) policies, a robust, simulation-based technique is introduced, which adaptively determines the optimal values for the hedging points, even if some of the system parameters are unknown. This novel method is of particular interest because it can be applied to machines producing an arbitrary number of part-types, it is independent of the particular values of the system parameters, it is robust to perturbations, and it is computationally efficient. |
