Abstract

Aircraft engines are complex systems that require high reliability and adequate monitoring to ensure flight safety and performance. Moreover, timely maintenance has necessitated the need for intelligent capabilities and functionalities for detection and diagnosis of anomalies and faults. In this thesis, fault diagnosis in aircraft jet engines is investigated by using intelligent-based methodologies. Two different artificial neural network schemes are introduced for this purpose.
The first fault detection and isolation (FDI) scheme for an aircraft jet engine is based on the multiple model approach and utilizes dynamic neural networks (DNN). Towards this end, multiple DNNs are constructed to learn the nonlinear dynamics of the aircraft jet engine. Each DNN represents a specific operating mode of the healthy or the faulty conditions of the jet engine.

The inherent challenges in fault diagnosis systems is that their performance could be excessively reduced under sensor fault and sensor degradation conditions (such as drift and noise). This thesis proposes the use of data validation and sensor fault detection to improve the performance of the overall fault diagnosis system. In this regard the concept of nonlinear principle components analysis (NPCA) is exploited by using autoassociative neural networks.

The second FDI scheme is developed by using autoassociative neural networks (ANN). A parallel bank of ANNs are proposed to diagnose sensor faults as well as component faults in the aircraft jet engine. Unlike most FDI techniques, the proposed solution simultaneously accomplishes sensor faults and component faults detection and isolation (FDI) within a unified diagnostic framework.

In both proposed FDI approaches, by using the residuals that are generated from the difference between each network output and the measured jet engine output as well as selection of a proper threshold for each network, criteria are established for performing the fault diagnosis of the jet engines. The fault diagnosis tasks consists of determining the time as well as the location of a fault occurrence subject to the presence of disturbances and measurement noise. Simulation results presented, demonstrate and illustrate the effective performance of our proposed neural network-based FDI strategies.