An evaluation of a real-time fault diagnosis expert system for aircraft applications

Several aspects of the aircraft domain make inflight diagnosis difficult. Many stem from the fact that the aircraft is in operation during and after the occurrence of a fault. These aspects include failure propagation, operator compensation, and lack of complete information. Still other aspects include responding rapidly to a failure, recognizing multiple failures, and predicting the effect of the failure on the aircraft. To address these concerns, a fault monitoring and diagnosis expert system called Faultfinder was conceived and developed to detect and diagnose inflight failures in an aircraft. Faultfinder is an automated intelligent aid whose purpose is to assist the flight crew in fault monitoring, fault diagnosis, and recovery planning. The present implementation of this concept performs monitoring and diagnosis for a generic aircraft´s propulsion and hydraulic subsystems. This implementation is capable of detecting and diagnosing failures of known and unknown (i.e., unforeseeable) type in a real-time environment. Faultfinder uses both rule-based and model-based reasoning strategies which operate on causal, temporal, and qualitative information. This paper describes a preliminary evaluation of the diagnostic concepts implemented in Faultfinder. The evaluation used actual aircraft accident and incident cases which were simulated to assess the effectiveness of Faultfinder in detecting and diagnosing failures. Results of this evaluation, together with the description of the current Faultfinder implementation, are presented.