Aircraft bleed air system fault detection based on MSE of LSTM and informer

Abstract

The bleed air system is an important part of the aircraft, and the normal operation of the bleed air system has an important impact on the safety and comfort of the aircraft. A deep learning-based method was proposed for the fault diagnosis of the precooler and pressure regulating valve (PRV) in the aircraft bleed air system. This method used long short-term memory network (LSTM) and Informer as prediction models. It also used the mean square error of the predicted and actual values as an anomaly detection indicator. The QAR data of the Airbus A320 series aircraft were used for experimental verification, and the model was evaluated and analyzed from the aspects of prediction performance, fault detection rate, false alarm rate, miss rate, etc. The results showed that the accuracy of our method reached more than 92%, and compared with LSTM, the accuracy of informer increased by 0.5%, the false alarm rate decreased by 0.4%, and the miss rate decreased by 6.7%, proving the effectiveness and superiority of the method of this paper.