Simulation-based online scheduling of a make-to-order job shop
Thesis (MScEng (Industrial Engineering))--University of Stellenbosch, 2009.
Scheduling is a core activity in the manufacturing business. It assists with efficient and effective utilization of capital-intensive resources and increased throughput, thus increasing profitability. The focus in this thesis is on scheduling of manufacturing orders in a make-to-order job-shop enterprise. It is widely accepted that manufacturing of large volumes and production with as few as possible product variants is the most cost-effective business approach, but the need for low volume, once-off engineering parts will always exist. Many approaches to scheduling exist, including translation of a scheduling problem to a Travelling Salesman analogue, while Discrete-event computer simulation is well established as a means to assist with scheduling. Simulation is appealing in the manufacturing environment, as it can realistically imitate dynamic, stochastic processes while being descriptive in forecasting the future. In this thesis, the development and testing of a simulation-based scheduler is described. The scheduler was developed for, and in collaboration with a South African make-to-order job-shop enterprise. A supporting information system was also developed and it is required that the enterprise changes some of its business processes if this scheduler is implemented. The scheduler considers the status of the enterprise each time a new order is received, and the current schedule is reviewed and may be revised at such a point in time, making it a real-time scheduler. Several classic scheduling dispatching rules and –measures were incorporated in the scheduler. These include First-in First-out, Earliest Due Date, Longest Processing Time, Shortest Processing Time, Smallest Slack and Critical Ratio (dispatching rules), while the performance measures are Makespan, Earliness, Lateness, Average Flow Time and Machine Usage. The proposed scheduler has been verified and validated using test data and designed confidence building tests, and its performance was also compared to an actual, historical schedule. The functioning of the scheduler is finally demonstrated using a stochastic test environment. The scheduler has generally performed satisfactorily and should be implemented as the final phase of this project.