Rolling bearing degradation stage division and RUL prediction based on recursive exponential slow feature analysis and Bi-LSTM model

Accurately predicting the remaining useful life (RUL) of rolling bearings is essential for effective system health management and maintenance in mechanical systems. Traditional RUL prediction methods often suffer from susceptibility to noise, leading to instability in feature extraction and inadequate capture of long-term change trends. To address this challenge, this paper proposes a rolling bearing RUL prediction method based on recursive exponential slow feature analysis (RESFA) and bidirectional long short-term memory (Bi-LSTM) network. Initially, the vibration signal is input into a convolutional neural network for health state classification, and the "3/5" principle is applied to determine the degradation starting (DS) point. Subsequently, features are extracted based on an autoencoder. Additionally, RESFA is utilized to extract long-term degradation trends within the system. Finally, the features extracted from the autoencoder and the slow feature are integrated, and the fused features are inputted into a Bi-LSTM model for accurate bearing RUL prediction. The efficacy of the proposed approach is validated using datasets from the IEEE PHM Prognostic Challenge, the XJTU-SY and ABLT-1A dataests. The prediction accuracy of the method proposed in this paper exceeds that of other state-of-the-art methods, highlighting the effectiveness of the RESFA-based approach in the field of rolling bearing RUL prediction.