A heuristic method for scheduling and dispatching of factory production using multiclass fluid networks

Abstract

This dissertation describes a two-phase heuristic method for scheduling and dispatching production in a factory. In the first phase, the production flow is modeled as a multiclass fluid network. This fluid queueing model is a relaxation of the deterministic factory scheduling problem (in addition to being a limit of the stochastic queueing model) so it functions as an approximation of a discrete flexible job-shop with WIP and ongoing inputs. However, buffer levels are allowed to have non-integer values, equipment processing can be simultaneously shared between different products, and a single lot can begin processing at a downstream step before it completely finishes at the previous step. By solving a finite series of quadratic (or linear) programs, an optimal (or nearly optimal) control policy is found for this fluid relaxation problem (with a weighted holding cost objective). In the second phase, production in the discrete factory queueing network is scheduled ahead of time or dispatched in real time by minimizing the deviation of the production from the optimal fluid control policy. Starting assignments are set with a mixed-integer program, and special techniques are used to deal with batching and to avoid sequence-dependent set-ups.