<p>The degradation prediction of mechanical rotating components serve as the foundation for intelligent maintenance. Accurate degradation prediction can prevent catastrophic failures, reduce costly unplanned maintenance, and enhance equipment reliability. To address the limitation that degradation indicators based on single-modal data fail to fully and comprehensively describe the degradation process of components, this study employs multi-modal data to develop a novel integrated degradation indicator using gamma distribution and Bayesian inference. This approach facilitates a more thorough and complete characterization and representation of the component degradation process. Meanwhile, to overcome the limitations of traditional convolutional or time-series prediction model architectures in extracting spatiotemporal features, a novel prediction model based on Mamba network is introduced. This model effectively captures complex degradation trend features, enabling accurate prediction of the degradation trends of rotating components. This study employs two datasets for experimental validation. In the real-world scenario dataset, the proposed prediction model achieves optimal RMSE and MAE of 0.02159 and 0.01599, respectively, achieving a maximum RMSE reduction of 0.27576 compared to commonly used convolutional and time-series network frameworks. Therefore, the proposed model exhibits clear advantages in predictive performance.</p>