Autonomous production control for matrix production based on deep Q-learning

Abstract Matrix production refers to a highly flexible production system based on independent production cells that are linked by a flexible transportation system. The production control system decides on the sequence of the production steps of each order and their allocation to specific time slots on the available machines. This paper presents an approach based on deep Q-learning that is able to cope with the dynamic events of the system. The performance of the machine learning-based production control is compared to a static rule-based approach. Additionally, the effects of coordination between the independent agents on throughput time is shown.