Method for predicting remaining useful life of rolling bearings based on the CBGT model

Abstract Accurate prediction of the remaining useful life (RUL) of rolling bearings is essential for ensuring the operational safety of industrial equipment. However, Transformer-based models tend to overlook local degradation details when directly processing raw time-series data, which can lead to suboptimal prediction performance. To address this limitation, this paper proposes a novel hybrid model, termed CBGT, for RUL prediction. Specifically, convolutional neural networks are first employed to extract local features, thereby enhancing the model’s ability to capture short-term degradation patterns. Subsequently, stacked bidirectional gated recurrent units are utilized to model both short- and long-term temporal dependencies, as well as contextual information. Finally, a Transformer module is incorporated to learn global degradation characteristics and long-range dependencies. By integrating local feature extraction, temporal dependency modeling, and global representation learning within a hierarchical framework, the proposed CBGT model significantly improves prediction accuracy. Experimental evaluation on the XJTU-SY dataset demonstrates that the CBGT model achieves a minimum 16.27% reduction in root mean square error relative to the Transformer baseline. These results validate the effectiveness of the proposed approach for RUL prediction.