An interactive remaining useful life prediction framework based on collaborative optimization of health indicators and stochastic degradation model

Abstract Remaining useful life (RUL) prediction is a core technology for implementing health management and predictive maintenance in complex devices. With the increasing complexity of device structures and the diversification of operating environments, the performance degradation process often presents nonlinear characteristics. Most existing methods rely on the assumption of linear degradation, and there is a disconnection between the construction of health indicators (HI) and the matching of degradation models, making it difficult to depict the nonlinear degradation behavior of an actual device accurately. Therefore, this paper proposes an interactive RUL prediction framework for the nonlinear degraded device based on the collaborative optimization of HIs and degradation models. In the offline stage, the HIs are constructed by sensor importance screening and data smoothing, and the nonlinear Wiener degradation model is established. Furthermore, the Bayesian optimization algorithm is used to automatically reverse-adjust the sensor fusion coefficient, failure threshold, and nonlinear factor to match the HIs to the degradation model. In the online stage, HIs are calculated from real-time sensor data, and the model parameters are dynamically updated using the particle filter algorithm, ultimately yielding the probability distribution of RUL. Experimental results on the C-MAPSS dataset show that the proposed method outperforms mainstream methods in terms of RUL prediction accuracy and stability, thereby verifying its effectiveness.