Channel attention & temporal attention based temporal convolutional network: A dual attention framework for remaining useful life prediction of the aircraft engines

The health of the aircraft engines is of great concern. And it is a key task to predict the remaining useful life (RUL) of the aircraft engines accurately. However, there are still challenges in RUL prediction, such as the flaw in incomplete sensor signals acquired, difficulty in determining the importance of sensor signals, and neglect of the key time points with significant performance degradation information in the sensor signals when performing RUL prediction. To tackle these challenges, a dual attention framework named Channel Attention & Temporal Attention based Temporal Convolutional Network (CATA-TCN) is proposed for the RUL prediction of the aircraft engines. Specifically, channel attention is integrated into TCN to focus on sensor signals with critical impact on RUL prediction and suppressing unimportant ones in long-term horizon. Next, the processed sensor signals are fed into temporal attention module, which enhances the impact of the key time points and generates the critical degradation features. Finally, the CATA-TCN outputs the predicted RUL by performing non-linear mapping on the extracted features. Turbofan engine degradation simulation data set (C-MAPSS dataset) and real flight data are used to validate the CATA-TCN framework. The experimental results show that the proposed method is significantly more accurate on overall prediction performance (Score and RMSE) than other state-of-the-art methods, especially under changeable operation conditions and complex fault modes.