Content area
Abstract
Assembly systems are prevalent in many manufacturing environments. An assembly system is a production system with two or more tandem lines feeding an assembly station. In the assembly systems considered in this thesis, a different kind of subassembly is produced in each tandem line. When there is a subassembly available at the assembly station from each tandem line, the assembly station produces an item using all of the subassemblies. Two kinds of uncertainty, time and type, commonly impact the operation of such systems. Time uncertainty is due to the fact that the time between the production of two consecutive subassemblies can be random. The other kind of uncertainty is due to the fact that subassemblies produced by the same tandem line need not be of the same type. This thesis addresses the problem of how to optimally control assembly systems subject to one or both kinds of uncertainty.
Four different assembly system control problems are investigated in this thesis. The first problem considers an assembly system where there are only two machines feeding the assembly station and there is time uncertainty. The problem is formulated as a Markov Decision Process and the structure of the optimal policy is characterized for the case where all the machines in the system have exponential processing time distributions. The second problem concerns assembly systems with two or more tandem lines feeding assembly, where there are one or more machines in each tandem line, and there is time uncertainty. The focus is on understanding how kanban can be used to control such an assembly system. The third problem examines an assembly system where there are only two machines feeding the assembly station, and there is type uncertainty. The problem is formulated as a Markov Decision Process and the structure of the optimal policy is characterized for special cases. A heuristic policy is presented for the more general case. The fourth problem treats the case of an assembly system where there is both time and type uncertainty. The problem is formulated as a Markov Decision Process and two heuristics are presented for the problem.
