Single machine changeover scheduling

James Douglas Blocher, Purdue University

Abstract

In a multi-product, discrete-item manufacturing environment, it is often more economical to have one flexible machine to produce the products than to have a dedicated machine for the production of each product. In the flexible-machine environment, it becomes necessary to produce one product, changeover the facility to another product to produce it, and so on. In this thesis, we focus on this changeover. Each product will likely have a unique machine setup--or machine configuration--associated with it. A changeover is the switching of the machine configuration from one setup to another. In almost all cases, the changeover will be costly, and thus, it is an activity which must be managed. Managing this type of activity requires a plan of how much of a product to produce after a changeover and in what order to make the changeovers to various products. Such a plan is referred to as a schedule. In this thesis we look at how the batching and scheduling decision is made for a single machine in a dynamic, deterministic demand environment. Managers of many firms try to avoid late shipments at almost any cost, implying a very high backorder cost; thus we assume all demand must be met on time (if possible). Although we assume demand is deterministic, we also assume demand is known only for a finite time horizon. We assume that the only "relevant costs" are the cost of the changeover. We consider both the case of negligible and non-negligible changeover times. In this thesis, we develop procedures to optimally solve the changeover scheduling problem, as we refer to the problem described above. We develop results for the problem for sequence-independent changeover costs for two cases of changeover times: (1) when non-negligible, positive changeover times exist and (2) when they are negligible. We provide results for obtaining lower bounds on the optimal solution. We develop a forward-time implicit-enumeration algorithm to optimally solve the problem as an alternative to the backward-time approaches and because the backward-time approaches presented thus far do not look very promising when changeover times are present. In addition, not only are forward-time scheduling methods generally more intuitive, but forecast horizon results are readily obtained from forward-time algorithms. We present such results in this thesis.

Degree

Ph.D.

Advisors

Chand, Purdue University.

Subject Area

Operations research|Management

COinS