Modular reconfiguration of flexible production systems using machine learning and performance estimates

Abstract

This paper presents an agent-based framework for reconfiguring modular assembly systems using machine learning and system performance estimates based on previous reconfigurations. During a reconfiguration, system integrators and engineers make changes to the machine to meet new production requirements by increasing capacity or manufacturing new product variants. The framework provides a method for automatically evaluating these changes in terms of impact on the performance of the production system, and building a knowledge base. Such knowledge is used to support future reconfigurations by recommending changes that are likely to improve the performance based on previous reconfigurations. The agent architecture of the framework has two levels, one for individual assembly stations and one for the entire production line. Knowledge bases of changes are built and utilised at both levels using machine learning and performance estimates. A prototype implementation of the proposed framework has been evaluated on an assembly production system in an industrial scenario. Preliminary results show that framework helps to reduce the time and resources required to complete a system reconfiguration and reach the desired production objectives.