Reconfigurable Scheduling as a Discrete-Event Process: Monte Carlo Tree Search in Industrial Manufacturing

Abstract

: In this paper we introduce a theoretical basis for reconfigurable makespan scheduling that is computationally-efficient and general purpose in manufacturing. A full-scale scale case study for batch production in the aerospace industry is shown. A knowledge-based Discrete-Event System, based on a Timed Petri Net, is injected with the initial - current - state and simulated to generate trajectories that represent valid possible schedules or policies analogous to the Monte-Carlo Tree Search (MCTS) planning algorithm. A new, concise, evolutionary metaheuristic is proposed called Elitist Trajectory Mutation (ETM) in order to exploit high performing schedules in localising search and optimisation. The advantage of this approach is reconfigurability, extensibility and ability to be parallelised to enable satisficing performance for real-time applications such as intelligent industrial cyber-physical systems scheduling, autonomous control of distributed systems and active industrial informatics.