Bearing degradation prediction based on local correction deep adaptive state space model

Accurate prediction of bearing degradation trends is essential for assessing equipment health and optimising maintenance strategies. However, the multi-stage progressive damage and sudden failures during the nbvprocess may lead to prediction bias and response delays, affecting the timeliness and accuracy of nm,maintenance decisions. To address these challenges, this study introduces a Local Correction Deep Adaptive State Space Model (LCDA-SSM) for bearing degradation prediction. The proposed method integrates the flexibility of state space modelling with the nonlinear representation capabilities of deep learning, thereby enhancing the versatility of state space equations. The proposed method incorporates a dynamic weight adjustment mechanism for long-term memory and a dynamic parameter regulation network within the state transition framework, enabling adaptive modelling of long-term degradation processes. A correction module based on a local attention mechanism is developed within the observation equation to improve the capture of high-entropy signals and local variability. Finally, we evaluated the proposed model on the PRONOSTIA benchmark and a real-world bearing dataset to verify its effectiveness in degradation prediction. The results show that the proposed method outperforms existing approaches in bearing degradation prediction accuracy, thereby demonstrating the superior performance of the proposed model in this task.