Remaining Life Prediction of Rolling Bearings Based on Unsupervised Anomaly Detection and STA-BiLSTM

Bearings are critical components in mechanical systems, susceptible to wear and fatigue failure during prolonged operation, which can disrupt the normal functioning of mechanical equipment. Predicting the remaining useful life (RUL) of bearings is essential to prevent unexpected failures and ensure safe and reliable equipment operation. This article presents a novel approach for RUL prediction, consisting of two stages: determining the degradation start (DS) point and predicting the RUL. In the first stage, an unsupervised anomaly detection method is introduced to accurately identify the DS point in the full life cycle of the bearing. In the second stage, a spatiotemporal attention (STA) mechanism combined with bidirectional long short-term memory (BiLSTM) is proposed for RUL prediction. Raw vibration signals are first processed through an autoencoder (AE) to automatically extract fault features. These features are then fed into the STA model for a deep-weighted fusion of spatial and temporal information, capturing comprehensive insights from both dimensions. Finally, the BiLSTM model predicts the bearing’s RUL. Experimental validation using the PHM2012 and ABLT-1A datasets demonstrates the effectiveness of our proposed method. The RUL prediction results conducted on the ABLT-1A experimental platform indicate that, compared to LSTM, RNN, GRU, and DCNN, the proposed method achieved RMSE reductions of 23.8%, 16.9%, 22.8%, and 14.7%, respectively; MAE reductions of 63.7%, 55.7%, 62.5%, and 47.7%, respectively; and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">${R} ^{{2}}$ </tex-math></inline-formula> increases of 4.7%, 3.4%, 4.5%, and 2.4%, respectively.