A Temporal Channel Attention Network for Turbofan Engine Remaining Useful Life Prediction

Aero-engines are exposed to a harsh environment of high temperature and pressure for long periods of time and are prone to performance degradation or even failure. Therefore, it is imperative to estimate the remaining useful life of the engines to prevent accidents from occurring. Among some of the existing RUL prediction methods, time-series features that are highly correlated with equipment degradation within different cycles cannot be effectively extracted. To address this problem, a temporal channel attention network (TCAN) is presented to predict the RUL of turbofan engines. To extract the long-term dependencies in the equipment degradation process, a temporal convolutional network is constructed to process the time-series data. A channel attention mechanism is designed to enhance the degradation trend information within important cycles of the extracted long-term dependencies to better fit the comprehensive degradation process. In the study, the TCAN model is trained and tested on the C-MAPSS dataset, and the RMSE and Sore values of 16.30 and 1273 are achieved, respectively. Compared to some advanced RUL prediction methods, the TCAN has better prediction performance.