A Method for Predicting the Remaining Useful Life of Aircraft Engines Based on NLSTM and Feature Optimization Strategy

To address the issue of diverse monitoring data types, high dimensionality, and sparse values, which significantly affect the accuracy of mechanical equipment's remaining useful life (RUL) prediction, this study proposes a novel aircraft engine RUL prediction method utilizing a feature selection strategy. Initially, based on the monitoring data sequences from different sensors, a feature selection criterion was developed to screen data of high-contribution as inputs for the prediction model. Subsequently, a regression variational autoencoder network was constructed for latent space interpretability in feature extraction, to intuitively express the latent space mapping form and confirm the contribution of the preferred features to the prediction and the representation ability of the degraded features. Finally, the dilated causal convolution network and nested Long Short-Term Memory (LSTM) network were utilized to achieve aircraft engine RUL prediction using the C-MAPSS dataset. In comparison with existing research, this method has effectively reduced prediction errors, achieving the lowest RMSE values in the FD002 and FD004 datasets. Additionally, it has also achieved favorable outcomes in the FD001 and FD003 datasets.