Scalable cooperative multi-agent-reinforcement-learning for order-controlled on schedule manufacturing in flexible manufacturing systems

To operate flexible manufacturing systems efficiently, a robust and optimal production control is crucial. With an increasing number of workpieces being processed in parallel, ensuring guaranteed lead times represents a complex optimization tasks, better known as the flexible scheduling problem. Cooperative multi-agent reinforcement learning approaches have recently shown their potential in production control. However, ensuring guaranteed lead times in flexible manufacturing systems with these approaches remains an open problem. In this work, an existing cooperative multi-agent framework for flexible job-shop scheduling is transferred and modified to optimize production control in flexible manufacturing systems. Using a centralized training for decentralized execution multi-agent deep reinforcement learning approach, the goal is to optimize order agents to ensure guaranteed lead times. Furthermore, a comprehensive simulation study investigates the effect of common knowledge on facilitating cooperation, and empirically evaluate the frameworks scalability to a range of challenging scenarios.