Rolling bearings are pivotal components within rotating mechanical systems, and accurately predicting their remaining service life holds significant practical importance. This paper addresses issues prevalent in common deep learning methods for predicting remaining useful life (RUL), notably inadequate feature extraction and low prediction accuracy resulting from reliance solely on short-term or long-term dependent features.In this paper, we introduce a residual useful life prediction method for bearings, named TcLstmNet-CBAM. Compared to conventional deep learning-based bearing life prediction methods, the proposed approach leverages a temporal convolutional network (TCN) to extract long-term temporal dependencies and higher-level spatial features from historical data, while employing a long short-term memory (LSTM) network to capture short-term temporal dependencies and sequence relationships. Additionally, a Convolutional Block Attention Module (CBAM) is incorporated to perform multi-dimensional weighting on the extracted features, assigning greater importance to critical features. Consequently, this method enables more comprehensive feature extraction and enhances RUL prediction accuracy by emphasizing key features. Finally, to validate the effectiveness of the proposed approach, we conducted experiments on the PHM2012 and XJTU-SY rolling bearing datasets, comparing its performance against several other prevalent deep learning prediction methods. Experimental results demonstrate that the proposed TcLstmNet-CBAM method can effectively predict the remaining useful life (RUL) of bearings, achieving a mean absolute error (MAE) of 2.287 and a root mean square error (RMSE) of 3.123. These results strongly validate the effectiveness and superiority of the proposed method.
Keywords: CBAM; Feature extraction; LSTM; Parallel network; Remaining useful life; Rolling bearings; TCN.
© 2025. The Author(s).