BSVAR-based remaining useful life prediction method for aircraft engines

Accurate prediction of remaining useful life (RUL) is critical to the stability, reliability, and safety of aircraft engines. A new deep learning model called BSVAR is suggested for RUL prediction in order to address the issue that current RUL prediction techniques are unable to properly utilize the deterioration information of sensor data. The deep degradation information of sensor data is extracted using a bidirectional long-short-term memory (Bi-LSTM) networks and self-attention based variational autoencoder (SVAE). With the utilization of variational inference, the sensor data is clustered according to the implied degradation information, meanwhile, the latent space can be generated. The combination of the Bi-LSTM, the SVAE, and the regressor is used to establish a RUL prediction model to sufficiently extract the degradation features of sensor data to improve the prediction accuracy. Results from experimental validation on the aero-engine C-MAPSS dataset demonstrate that the suggested approach outperforms the current RUL prediction approaches in terms of prediction performance and can identify the engines’ degree of degradation in the latent space.