A Multi-Layer Approach for Failure Detection in a Manufacturing System Based on Automation Agents

Automatically detecting component failures and unexpected behaviors is an essential service for achieving fault-tolerant robust manufacturing systems. The application of a multi-agent system is regarded as a promising approach for designing complex systems such as manufacturing systems due to its distributed nature. In this paper we present an agent-based control system with diagnostic capabilities on several layers for a pallet transport system. Local diagnostic tasks for detecting failures are performed by the automation agents each controlling one physical component such as a diverter. In order to observe the correct behavior of the components on a system-wide scale, an approach based on Hidden Markov Models linked to the agents is employed.