Deep Reinforcement Learning for multiobjective Scheduling in Industry 5.0 Reconfigurable Manufacturing Systems

In today’s manufacturing environment, the need to respond quickly to changing market demands is critical. Reconfigurable Manu- facturing Systems (RMS) represent a monumental step towards realis- ing this requirement, providing an agile and cost-effective framework to accommodate changing production needs. The dynamic nature of RMS requires the integration of robust learning algorithms to continuously op- timise system configurations and scheduling. This study posits Reinforce- ment Learning (RL), specifically the Double Deep Q-Network (DDQN) algorithm, as a viable solution to navigate the complex, multi-objective optimization landscape of RMS. Among the multiple objectives, key con- siderations include minimising tardiness costs, ensuring sustainability by minimizing wasted liquid and gaz emissions of production, optimising makespan, and reducing operator intervention during system reconfigu- ration, thereby improving ergonomics. Our contribution is two-layered: first, we propose a modular, hierarchical architecture for RMS that in- corporates a multi-agent environment at the reconfigurable machine tool (RMT) level, encouraging the interaction of agents in a better man- ner to achieve global optimisation. Secondly, we use DDQN to navigate the multi-objective space cleverly, enabling more efficient and ergonomic system reconfiguration and scheduling. The results highlight the effec- tiveness of using RL to unlock the complicated optimization challenges inherent in modern manufacturing paradigms, paving the way for Indus- try 5.0.