<p>The structural design of nuclear reactors is increasingly trending toward modularization and miniaturization, making the health status of internal bearings a critical factor in ensuring reactor operational safety. However, the fault evolution process of bearings is highly complex, and no established digital method currently exists to accurately simulate their performance degradation patterns. To address this challenge, this study proposes a dual-network prediction model that integrates a deep convolutional neural network (DCNN) and a gated recurrent unit (GRU). This model effectively captures degradation characteristics by extracting features from both time-series signals and continuous wavelet transform (CWT) images, providing a comprehensive approach to bearing health assessment. Additionally, a Bootstrap framework was introduced to further quantify the prediction interval for bearing remaining useful life (RUL). Finally, the proposed method’s effectiveness was validated through accelerated life testing of bearings. The performance of the proposed method was evaluated by introducing cumulative relative accuracy and convergence metrics. The results indicate that, in comparison to traditional models, the proposed method not only enhances the accuracy of RUL estimation but also accelerates the convergence of prediction errors. These findings further substantiate the superiority of the proposed approach in terms of both predictive precision and reliability.</p>