Automated Production Ramp-up Through Self-Learning Systems

Abstract

The ramp-up of production systems is characterised by situations that arise for the first time. Due to the unpredictability of system behaviour in such situations, instabilities occur that lead to reduced production effectiveness. In order to deal with the resulting uncertainty, this paper presents an approach for self-directed systems capable of "learning", that is, they adapt their behaviour depending on the signals and changes of the circumfluent world. The advantages of such systems are significant, as they can react to changing products, production equipment and process constraints, and are able to function in exceptional situations. The presented concept makes use of reinforcement learning, one of the most general approaches to learning control. Simulations of three different ramp-up processes are used, where, as a demonstration, robots have to assemble windscreens on a moving truck.