Optimising Machine Selection Rules for Sequence Dependent Setups with an Application to Cartoning

Conventional scheduling technology, which tries to optimise performance metrics such as utilisation and makespan, works well in environments where there is a high degree of stability, and hence certainty. However, in uncertain situations, schedules that try to achieve optimality have trouble achieving this target. An often used technique to circumvent the issues with uncertainty that optimal policies display, is the use of rules that are triggered when a job enters the system or a machine becomes available. Such rules are by definition reactive, and can thus deal very well with uncertainty. The downside of these rules is that they make local decisions, which may result in non-optimal behaviour. This becomes especially apparent in light of significant sequence dependent setup times. These are, by nature, dependent upon the global sequence, something that rules cannot deal well with. In this paper, we investigate a method to generate custom machine selection rules that lead to improved setup times. We illustrate the advantages of this new type of rule by presenting an experimental analysis of using these rules at the cartoning department of a large manufacturing company.