Gas path fault detection and isolation for aero-engine based on LSTM-DAE approach under multiple-model architecture

Abstract

Gas path fault diagnosis plays a critical role in the security guarantee and maintenance of aero-engines. In this paper, an approach based on a fusion neural network under multiple-model architecture for gas path fault detection and isolation is proposed. We develop a multi-channel long short-term memory network based on a sliding window to explore temporal and spatial relationships of data and capture the residuals of sensor measurements between predicted and observed values. Additionally, denoising autoencoders under a multiple-model architecture are introduced so as to perform fault detection and isolation based on the comparison of reconstructed prediction errors and isolation thresholds. Several simulation results verify that the diagnostic model has excellent robustness and diagnostic ability. The proposed method is compared with other common methods, and the advantages and functions of this method are presented.