An Intelligent Framework for Multi-Agent System Based on Dynamic Balancing of Production 4.0

Abstract

Balancing assembly lines is one of the biggest problems facing industrial processes. Conventionally, this balancing consists of optimally allocating production tasks between different workstations or stations, taking into account precedence, capacity and cycle time constraints. However, in the face of today’s dynamic environment (fluctuations in demand, technical hazards, labor shortages, etc.), static balancing is no longer sufficient. Companies need dynamic rebalancing solutions, designed to adapt task distribution in real time as soon as a perturbation appears. Our study based on the use of multi-agent systems (MAS) offers a potentially attractive approach. MAS enable decision-making to be decentralized, with each agent representative of a workstation, a set of tasks, an operator or a piece of equipment. Each agent assesses its situation locally (workload, availability, etc.) and negotiates with the others to achieve a more flexible, proactive assignment of tasks, in line with the principles of Industry 4.0 (connectivity, distributed intelligence and adaptability). The approach proposed in our study is a dynamic framework based on the use of three agents, each of which executes and cooperates with the other agents in order to determine parameters with high variability, propose adjustments and implement them. The results found show that the use of the dynamic balancing environment facilitates monitoring and decision-making through a dynamic dashboard including a set of KPIs specific to balancing. The operational impact of implementing this environment shows a significant increase in terms of productivity and availability, implicitly the synthetic rate of return (‘OEE’) of the production line. As a result, the financial aspect of reconfiguring the line has been reduced thanks to the recommendation agent, which proposes future adjustments without impacting the production line. The line balancing no longer depends on a single a priori measurement, but is instead adapted on an ongoing process, giving greater strength in the face of the uncertainties of a VUCA world.