An agent-based approach for fault diagnosis in manufacturing systems

This dissertation presents a new hybrid approach for fault diagnosis of complex systems that efficiently incorporates both deep reasoning and shallow reasoning diagnostic modes. It includes an episodic experience cache to enable the reuse of recent diagnostic results and an agent-based architecture to facilitate gathering and analysis of the relevant sensory system information. This agent-based architecture allows analysis of diagnostic data from a variety of viewpoints, not unlike human diagnostic reasoning.; One of the key considerations in employing an agent-based architecture is the process of reconciling disparate beliefs from the different agents. This dissertation investigates the use of fuzzy numbers to represent agent recommendations, and presents new aggregation and ranking procedures for these fuzzy recommendations. These new procedures are shown to be computationally efficient.; A case study of a small manufacturing cell is presented which demonstrates the effectiveness of this new diagnostic approach.