MULTI-AGENT INTELLIGENT DIAGNOSTIC SYSTEM.

Principles are reported for a multiagent diagnostic system in which continuous monitoring and diagnosis of subsystems is carried out in parallel by local diagnostic experts. These experts encode the causal knowledge of system behavior as a structure of expectations of and commitments to other agents in the system,, and a group of local tasks. Each diagnostic expert monitors how other specialists in the environment are meeting their commitments to it. Any subsystem is monitored by others which depend on its function. The more other subsystems which depend on the faulty system, the more evidence will exist pointing to its failure. The prototype system was built using the experimental distributed AI testbed and simulator MACE (Multi-Agent Computing Environment), on an Intel Hypercube 16-node concurrent processor.