Design and implementation of a real-time multi-agent system

We present an experimental multi-agent system for model-based real-time fault diagnosis in modular production systems. The architecture comprises a set of hierarchically organized process agents, a simulation agent and a user agent. We discuss the most interesting characteristics of this new real-time AI architecture. A key component in model-based diagnosis is simulation; in our solution a distributed Petri-net simulator and a parallel continuous simulator, both based on multiple TMS320C40 signal processors, form the core of the process agents and the simulation agent, respectively.