Hybrid Monte Carlo tree search based multi-objective scheduling

As markets demand targeted products for highly differentiated use cases, the number of variants in production increases, whilst the volume per variant decreases. Different product variants result in differences in work content on workstation level which cause takt time losses and result in a poor utilization. In this context, matrix-structured production systems with neither temporal nor spacial linkage emerged to reduce the effects of different work content on the entire production system. However, matrix-structured production systems require far more complex production control. To that end, this paper presents a scheduling approach. The proposed scheduling system considers variable process sequences and their allocation to different workstations in order to optimize scheduling objectives. This contribution presents a Monte Carlo tree search based optimizer combined with local search as post optimizer to derive schedules in a short time span to enabling reactive scheduling. The application of the scheduler to a benchmark problem and an industrial scheduling problem demonstrates the quality of the results and illustrates how the scheduler reassigns the work content dynamically.