A Deep Reinforcement Learning Based Scheduling Policy for Reconfigurable Manufacturing Systems
Jiecheng Tang, Konstantinos Salonitis
Abstract
Reconfigurable manufacturing systems (RMS) is one of the trending paradigms toward a digitalised factory. With its rapid reconfiguring capability, finding a far-sighted scheduling policy is challenging. Reinforcement learning is well-equipped for finding highly efficient production plans that would bring near-optimal future rewards. For minimising reconfiguring actions, this paper uses a deep reinforcement learning agent to make autonomous decision with a built-in discrete event simulation model of a generic RMS. Aiming at the completion of the assigned order lists while minimising the reconfiguration actions, the agent outperforms the conventional first-in-first-out dispatching rule after self-learning.