Order Acceptance and Scheduling in Manufacturing Systems: A Reinforcement Learning Approach

This research introduces Smart-OAS (Order Acceptance and Scheduling), a Reinforcement Learning-based methodological framework for the integrated OAS process. The research addresses the limitations of traditional heuristic rules, whose efficiency decreases significantly in the presence of critical bottleneck resources. The system uses an intelligent agent structured on the basis of Markov Decision Processes (MDP) for the dynamic shop floor status assessment, facilitating the integration of data flows from ERP (Enterprise Resource Planning) and MES (Manufacturing Execution Systems) systems. The novelty lies in the design of a reward function that balances marginal profit with opportunity cost and delay-related penalties. Conceptual validation attests the algorithm’s ability to develop superior decision-making policies by strategically prioritizing high-value-added orders and protecting production capacity on critical machines. The study establishes a roadmap for integrating the model into Industry 4.0 ecosystems, exploring the potential for expansion towards digital twin technologies and sustainability indicators.