Multi-Agent Scheduler for the Continuous Dynamic Flexible Job Shop Scheduling Problem

Dazzle Johnson, Gang Chen, Yuqian Lu

Abstract: This work addresses the unique production challenges in dynamic smart manufacturing environments caused by the shift towards mass personalisation. Traditional scheduling methods do not perform reliably in a continuously operating environment with unpredictable job order arrivals, a variety of products and urgent orders, leading to increased tardiness and inefficiencies. To reflect the challenges dynamic manufacturing environments face, we introduce a new problem – the Continuous Dynamic Flexible Job Shop Scheduling Problem (C-DFJSP). In response, we propose MAC-Sched – a MultiAgent Continuous Scheduler. MAC-Sched is a Graph-Based Multi-Agent Reinforcement Learning model that takes a graph representation of the manufacturing environment as input, thus enabling the continuous and reliable production of personalised products. Demonstrating its reliability in multiple unseen continuous operating environments, MAC-Sched outperforms popular heuristic rules, effectively minimising the average tardiness over time amid evolving factory dynamics.