Spurious Symptom Reduction in Fault Monitoring Using a Neural Network and Knowledge Base Hybrid System

Abstract

An approach to reduce number of spurious symptoms in aircraft engine fault monitoring is investigated. Two strategies were utilized. A set of rules designed to filter spurious symptoms was created. Then a neural network was designed to generate expectation value for each of the sensors monitored. The neural net was trained for a specific engine during normal operation. After capturing patterns for normal engine behavior in the neural net, an expectation value for the sensor is predicted. The success of this approach relies on generating better expectation values which in turn produce smaller variation from actual operating behavior and hence generate fewer spurious symptoms. Resulting hybrid system of neural networks and rule-based model demonstrates a drastic reduction of overall spurious symptoms.