Fusion Network Combined With Bidirectional LSTM Network and Multiscale CNN for Remaining Useful Life Estimation

To meet the needs of society and enterprises for high-reliability equipment, Remaining Useful Life (RUL) estimation has received extensive attention. Traditional statistical methods are significantly affected by model and parameter selection. Deep Learning has powerful data processing capabilities and does not require exact physical models and expert prior knowledge. Therefore, Deep Learning has shown a broad application prospect in the field of remaining life prediction. This paper proposes a data-driven method based on Bidirectional Long Short-Term Memory Network (BiLSTM) and Multiscale Convolutional Neural Network (MSCNN) for RUL estimation. Data are generated by sliding time window (TW) to the form of short-term sequences, then the data are filled into two base models in the same batch. Piecewise RUL functions are applied for label setting instead of traditional linear degradation functions. Proposed method is validated on the C-MAPSS dataset provided by NASA. Compared with similar methods, this proposed RUL prediction method has a better prediction capability.