Enabling Adaptation in Dynamic Manufacturing Environments with Decentralized Agent-Based Systems and Local Perception

Modern industrial landscapes, like Industry 4.0, pose a formidable challenge: how to maintain high-performance levels in dynamic manufacturing environments while minimizing downtime for adaption? Control systems have to be capable of making real-time decisions between adaptation and task fulfillment yet to external, changing production goals. Centralized control systems promise ideal decisions based on global data, but need clean and extensive data. Decentralized multi-agent systems (MAS) instead may decide on local data, yet, the efficiency of decentralized MAS to adapt in dynamic manufacturing settings remains largely unexplored. Thus, we study the potential of a MAS of decentralized agents characterized by adapting exclusively with local information, in addressing the challenges posed by dynamic manufacturing environments. We illustrate the problem using a driverless transportation system for a manufacturing scenario where a control system guides Autonomous Guided Vehicles (AGVs) in response to unforeseen disturbances and unknown product transportation tasks. We compare the decentralized MAS approach to an omniscient centralized control system, measure the performance of both systems and consider the communication efficiency among the MAS population. We show that decentralized agents with local information can adapt efficiently and shed light on their effectiveness in dynamic environments.