On Reinforcement Learning in production control and its potentiality in manufacturing

In the new paradigm of industry 4.0, one of the open issues is to configure production and logistics systems to meet the increasingly customized market demand in a shorter time and at lower cost. It is suitable to implement measures and actions to make the production system much more resilient and self-organized to face all adversities. The technologies typical of Industry 4.0 help to meet these tasks. In the current era, it is therefore necessary to make even greater use of these tools. Considering the increasing interest in AI and the promising results of its application in industrial scenario, this paper proposes a new approach in production control using Reinforcement Learning (RL). A literature review is made to highlight the potential of RL application in production systems and how it could help in the decision making process. Among the applications found in the literature, an emphasis is placed on those specifically related to the world of manufacturing. The goal is to train a network to achieve a throughput target, keeping a certain amount of WIP constant on a Flow Shop line. A new approach where the state, action space and reward function are formulated. The system performance is compared with the known results of an analytical model (Practical Worst Case, PWC).