A Model-Free Reinforcement Learning Approach Using Monte Carlo Method for Production Ramp-Up Policy Improvement - A Copy Exactly Test Case

Abstract

Abstract Production Ramp-up is a phase in the production timeline which has gained interest from the industry in the literature in order to decrease time-to-market. Intelligent systems and machine learning (ML) techniques have been applied in manufacturing lines and have demonstrated their potential to support knowledge capturing which can aid decision making. However, they mostly focus on supervised learning techniques which require prior knowledge and data pairs are classified without a systematic framework. This work approaches ramp-up as an episodic problem with a clear final target. Ramp-up is formalised as a decision process and a reinforcement learning approach is followed for deriving a policy, for a copy-exactly test case. Finally, a test case of an assembly station ramp-up by, different users is presented. A Monte Carlo approach is used to apply Reinforcement Learning (RL) and an improved policy is generated and evaluated.