An Application of Hyper-Heuristics to Flexible Manufacturing Systems

Optimizing the productivity of Flexible Manufacturing Systems requires online scheduling to ensure that the timing constraints due to complex interactions between modules are satisfied. This work focuses on optimizing a ranking metric such that the online scheduler locally (i.e., per product) chooses an option that yields the highest productivity in the long term. In this paper, we focus on the scheduling of a re-entrant Flexible Manufacturing System, more specifically a Large Scale Printer capable of printing hundreds of sheets per minute. The system requires an online scheduler that determines for each sheet when it should enter the system, be printed for the first time, and when it should return for its second print. We have applied genetic programming, a hyper-heuristic, to heuristically find good ranking metrics that can be used in an online scheduling heuristic. The results show that metrics can be tuned for different job types, to increase the productivity of such systems. Our methods achieved a significant reduction in the jobs' makespan.