Reinforcement Learning-based Job Shop Scheduling for Remanufacturing Production

Abstract

With the development of industry 4.0 and intelligent manufacturing, higher requirements for the manufacturing industry in terms of green, low-carbon and sustainability are highly desired. Hence, maintenance and remanufacturing industry has become a new focus. Different from traditional manufacturing process, remanufacturing factory encountered with incoming "raw material" of different quality which required various non-uniform production processes. It brought a big challenge for remanufacturing job shop scheduling on production efficiency. To tackle this problem, this paper deeply analyzes production processes in remanufacturing workshop, and establishes a mathematical model with the minimum total production time as the objective function. Due to the advantages of reinforcement learning (RL) in solving the job shop scheduling problem (JSP), this paper adopts Q-learning and DQN to solve the remanufacturing scheduling problem, where system states are extracted and five common-used heuristic scheduling rules are selected as the action set, and the reward function was designed consistent with the objective function. Comparison study was carried on with heuristic rules alone, genetic algorithm (GA) and RL and the benchmarking results prove the superiority of RL in solving this problem.