Optimization in distributed production systems

Machine learniung with game theory on the PLC

Abstract

The paper presents a novel approach for the development of self-learning algorithms for distributed production systems with special focus on the field-level implementation on PLCs. We employ a fully distributed system architecture in the form of a multiagent system and define game theory-based learning algorithms which allow self-learning control functions and coordination among the modules from scratch. We focus on the implementation of the approaches on the PLC with their evaluation in a real
world application and present a framework which allows for efficient training and resource-saving transfer from simulation to reality. The distributed architecture results in small scale models which can be efficiently implemented as a digital twin within
the PLC program itself. The results show the applicability and performance of the proposed approach and its potential for real
world implementations.