Multi-Agent reinforcement learning for dynamic flexible job shop scheduling with fuzzy transportation time

Self-organizing manufacturing systems (SOMS) represent an advanced type of flexible manufacturing system, which introduces variability in material transportation time. Previous research has predominantly treated logistics transportation and maintenance as separate concerns in the flexible job shop problem, often overlooking the impact of uncertain transportation times. This oversight can compromise simulation accuracy and increase tardiness of jobs. To address these challenges, this paper proposes a dynamic flexible job shop scheduling method that integrates machine reliability with transportation time uncertainty in SOMS. The proposed approach formulates a partially observable Markov decision process, incorporating both machine reliability and logistical constraints to minimize average weighted tardiness of jobs. Additionally, the multi-agent soft actor-critic algorithm is employed to improve adaptability to transportation fluctuations. Simulation experiments demonstrate that the approach significantly reduces average tardiness, achieving a 50% reduction compared to rule-based approaches and a 27% reduction relative to reinforcement learning-based methods.