Abstract

Jet engine related costs and the need for high performance reliability have resulted in considerable interest in advanced health and condition-based maintenance techniques. This thesis attempts to design fault prognosis schemes for aircraft jet engine using intelligent-based methodologies to ensure flight safety and performance. Two different artificial neural networks namely, non-linear autoregressive neural network with exogenous input (NARX) and the Elman neural network are introduced for this purpose. The NARX neural network is constructed by using a tapped-delay line from the inputs and delayed connections from the output layer to the input layer to achieve a dynamic input-output map. Consequently, the current output becomes dependent on the delayed inputs and outputs. On the other hand, the Elman neural network uses the previous values of the hidden layer neurons to build memory in the system.

Various degradations may occur in the engine resulting in changes in its components performance. Two main degradations, namely compressor fouling and turbine erosion are modelled under various degradation conditions. The proposed dynamic neural networks are developed and applied to capture the dynamics of these degradations in the jet engine. The health condition of the engine is then predicted subject to occurrence of these deteriorations.

In both proposed approaches, various scenarios are considered and extensive simulations are conducted. For each of the scenarios, several neural networks are trained and their performances in predicting multi-flights ahead turbine output temperature are evaluated. The difference between each network output and the measured jet engine output are compared and the best neural network architecture is obtained. The most suitable neural network for prediction is selected by using normalized Bayesian information criterion model selection. Simulation results presented, demonstrate and illustrate the effective performance of the proposed neural network-based prediction and prognosis strategies.