A Multi‐Path Degradation Modeling Framework With Adaptive Fractional Brownian Motion for Robust RUL Prediction in Mechanical Systems

Equipment condition monitoring and performance degradation prediction are key aspects of predictive maintenance in mechanical systems. However, the degradation process in mechanical systems often exhibits memory effects, which makes traditional Markov processes inadequate for effectively capturing the degradation process. Furthermore, individual differences and uncertainties in the degradation process often make a single degradation path insufficient to fully represent the entire degradation process. At the same time, the commonly used 3 criterion for early degradation point identification often struggles with handling outliers and minor fluctuations. To address these issues, the present article introduces a novel method for performance degradation and remaining useful life (RUL) prediction: MSM‐MDAFBM. This method combines the multi‐degradation adaptive fractional Brownian motion model (MDAFBM) with a multi‐path state matching (MSM) strategy that incorporates predictive loss constraints and state similarity. The MSM‐MDAFBM model considers multiple degradation paths and accurately captures the long‐range dependence (LRD) characteristics and fractal properties of the degradation process. Moreover, based on the extended particle filter, the drift coefficient is dynamically updated with the evolution of monitoring data. The path switching method, based on predictive loss constraints and state similarity facilitates efficient switching between degradation paths, thereby enhancing RUL prediction accuracy. Additionally, the present article proposes a dynamic early degradation point identification method based on Chebyshev's inequality, which effectively mitigates the impact of outliers and minor fluctuations. The effectiveness and precision of our prediction framework are demonstrated using two popular prognostic bearing datasets and full‐life test data from a heavy‐duty truck's rear axle.