An Attention-Based Deep Learning Model for Remaining Useful Life Prediction of Aero-Engine

Abstract

The prognosis of remaining useful life (RUL) for the aero-engine plays an indispensable role in enhancing the safety of aircraft and diminishing maintenance expenses. Several deep learning (DL) methods have been recently employed to handle this prediction problem due to their adaptable architectures and superior performance in handling the nonlinear characteristics of prediction problems. However, those models still need further improvement to better predict the status of aircraft engines. Therefore, this paper presents a new DL model, dubbed attention residual block-adaptive long short-term memory (ARB-ALSTM), based on integrating the attention mechanism with the long short-term memory network (LSTM) model and residual block to better comprehend the characteristics of this problem, thereby aiding in achieving better prediction. In a more general sense, the attention residual block is responsible for understanding the input dataset and extracting the most effective features, which significantly affect the model’s accuracy. Then, those extracted features are given to an LSTM with an adaptive attention mechanism to effectively capture and analyze long-term dependent information. This proposed model is evaluated using the NASA CMAPSS dataset and compared to several DL models to showcase its effectiveness. The experimental findings reveal that ARB-ALSTM is a strong alternative for predicting the RUL of aircraft engines because it could achieve better outcomes than all the compared models.