Proceedings of the
9th International Conference of Asian Society for Precision Engineering and Nanotechnology (ASPEN2022)
15 – 18 November 2022, Singapore
doi:10.3850/978-981-18-6021-8_OR-08-0197
A Case Study: Dynamic Dispatching Framework for Multi-Product Production System via Reinforcement Learning Approach
1Singapore Institute of Manufacturing Technology, 2 Fusionopolis Way, #08-04, Innovis, 138634, Singapore
ABSTRACT
Contemporary manufacturing systems often exhibit characteristics such as multi-product, multi-objective and increasing types and frequencies of uncertainties during production execution. It is non-trivial to consider a novel dispatching strategy adaptable to the ever-changing environment while maintain satisfactory production performance. In this work, we propose an end-to-end framework which consists of: 1) a simulated multi-product production system that can be configured with comprehensive constraints and uncertainties (the Environment), 2) a dispatcher that makes progressive decisions based on the tracked status of the Environment (the Agent), and 3) a messenger that transmits tracked status, dispatcher's decisions and feedback to decisions in-between the Environment and the Agent. The three-component dynamic dispatching framework, enabled by Reinforcement Learning (RL) approach, is among the first to be demonstrated on such a sophisticated dispatching problem, extracted from a real precision engineering production system. We wish the work would be valuable to advance further development and implementation of RL systems for production control applications.
Keywords: Multi-Product Production System, Reinforcement Learning Controller, Dynamic Dispatching, Multi-Objective Optimization
1Singapore Institute of Manufacturing Technology, 2 Fusionopolis Way, #08-04, Innovis, 138634, Singapore
ABSTRACT
Contemporary manufacturing systems often exhibit characteristics such as multi-product, multi-objective and increasing types and frequencies of uncertainties during production execution. It is non-trivial to consider a novel dispatching strategy adaptable to the ever-changing environment while maintain satisfactory production performance. In this work, we propose an end-to-end framework which consists of: 1) a simulated multi-product production system that can be configured with comprehensive constraints and uncertainties (the Environment), 2) a dispatcher that makes progressive decisions based on the tracked status of the Environment (the Agent), and 3) a messenger that transmits tracked status, dispatcher's decisions and feedback to decisions in-between the Environment and the Agent. The three-component dynamic dispatching framework, enabled by Reinforcement Learning (RL) approach, is among the first to be demonstrated on such a sophisticated dispatching problem, extracted from a real precision engineering production system. We wish the work would be valuable to advance further development and implementation of RL systems for production control applications.
Keywords: Multi-Product Production System, Reinforcement Learning Controller, Dynamic Dispatching, Multi-Objective Optimization
