Smart Scheduling for Flexible and Hybrid Production with Multi-Agent Deep Reinforcement Learning

As a consequence of growing personalized consumption, the demand for customized production processes is steadily increasing. Therefore, production systems should have a flexible structure, redundant resources, and dynamic scheduling mechanism to incorporate hybrid production of small-lot or even one-item orders. A greater flexibility will result in a higher uncertainty of production performance which can be mitigated by the introduction of smart scheduling mechanisms. In this paper, the scheduling problem of flexible and hybrid production is modelled as a workpiece-machine and workpiece-workpiece interaction problem. As an approach to problem solving, the Multi-Agent Deep Reinforcement Learning algorithm is applied. Simulation results confirm that the scheduling algorithm converges in the training process resulting in a clear performance improvement with respect to processing time.