A semi-observable state-driven self-tuning remaining useful life prediction method for mechanical systems

Remaining useful life (RUL) prediction of machinery is crucial for ensuring the reliability of mechanical systems and managing uncertainty in their operational states. In most industrial scenarios, the degradation of the health state is typically unable to be measured directly, which imposes significant challenges for RUL prediction, introducing considerable uncertainty in both state estimation and degradation modeling. To address this issue, this article proposes a semi-observable state-driven self-tuning RUL prediction method for mechanical systems. The method first establishes a regression model between multi-sensor data and the health state using historical training data. A self-tuning correction mechanism is then introduced to dynamically compensate for state estimation deviations by leveraging semi-observable state information. Furthermore, a real-time degradation stage division algorithm is developed to identify the current degradation phase of the health state. Finally, the compensated health state is integrated into a particle filtering (PF) framework, enabling online model parameter updating and RUL prediction. Experimental evaluations conducted on a tool wear dataset demonstrate that the proposed method improves prediction accuracy and exhibits strong robustness and adaptability under varying operational conditions.