<p>Rolling bearings are crucial components in industrial systems, and their health status is vital for system performance and stability. To address the limitations of traditional models in extracting spatiotemporal features and processing multisensor data, this paper proposes a remaining useful life (RUL) prediction method for rolling bearings using the Mamba-SDP framework. First, a channel fusion strategy efficiently integrates multisensor data to enhance feature representation under complex operating conditions. Second, the Mamba module extracts spatiotemporal deep features, while the improved channel fast Fourier transform (ICFFT) module processes non-stationary signals for improved frequency analysis. In addition, a scaled dot product (SDP) attention mechanism combined with cross-normalization mitigates numerical fluctuations in high-dimensional spaces, improving model stability and computational efficiency in capturing local and global dependencies. Furthermore, residual connections integrate SDP-weighted shallow features with deep features extracted by Mamba and ICFFT, maximizing the exploitation of both prior and posterior information. Finally, the fused features are fed into fully connected layers for RUL prediction. Extensive experiments on PHM 2012 and XJTU-SY datasets demonstrate that our method achieves MAE, RMSE, and score of 8.99, 11.50, 0.82 and 9.20, 11.40, 0.81, respectively, outperforming several state-of-the-art benchmarks with up to 7.1 % and 8.3 % lower MAE and RMSE, respectively, and a 4 % improvement in score. These results validate the effectiveness and generalization of the proposed approach, providing a promising solution for practical RUL prediction and industrial health management.</p>