Modern production systems are increasingly dynamic and difficult to optimize using static or linear control approaches. Online reinforcement learning enables systems to adapt decisions continuously based on real-time feedback, making it a promising approach for modern manufacturing environments.
This session reviews recent advances in online reinforcement learning applied to manufacturing use cases such as predictive maintenance, dynamic scheduling, and process optimization. It covers key methodologies, including deep reinforcement learning, policy-based approaches, and hybrid methods that incorporate domain knowledge to improve learning stability and scalability. The session also discusses current limitations, including computational requirements, algorithm stability, limited industrial validation, and the lack of standardized benchmarks.
Attendees will learn where online reinforcement learning is currently being evaluated, what performance improvements have been demonstrated in research and pilot studies, and what barriers must be addressed before broader production adoption. Practical research and implementation pathways are outlined to support future deployment.
Learning Objectives:
Upon completion, participants will be able to explain how online reinforcement learning enhances predictive maintenance, dynamic scheduling, and process optimization in modern production systems.
Upon completion, participants will be able to analyze the challenges and research pathways of online reinforcement learning, including algorithmic stability, scalability, and real-world industrial validation.
