A method combining dynamic path matching with multipath adaptive drift Lévy stable motion for performance degradation prediction

Characterizing equipment performance degradation and predicting remaining useful life (RUL) are critical aspects of predictive maintenance in mechanical systems. The foundation of effective RUL prediction lies in constructing health indicator (HI) based on condition monitoring signals that accurately reflect equipment degradation and health status. In addition, the individual variability and uncertainty in the degradation process often make it challenging for a single degradation path to represent the entire process fully. To address these issues, this article introduces a novel framework for performance degradation characterization and RUL prediction. Initially, we constructed the HI using the Wasserstein distance and the Cumulative sum (CUMSUM) control chart. This approach not only captures changes in the signal probability distribution during degradation but also exhibits strong monotonicity, trendability, and robustness. Next, we propose a dynamic first prediction time (FPT) dynamic identification method based on Chebyshev’s inequality, which effectively mitigates the influence of outliers and minor fluctuations. Additionally, we develop a dynamic path matching and multipath adaptive drift linear multifractional Lévy stable motion (DPM-MPALMLSM) model for RUL prediction. The MPALMLSM model incorporates multiple degradation paths that accurately capture the non-Gaussian characteristics, long-range dependence features, and multifractal properties of the degradation process, with drift coefficients dynamically updated as monitoring data evolves. The dynamic path matching method, grounded in performance evaluation, facilitates efficient switching between degradation paths, enhancing RUL prediction accuracy. The effectiveness and precision of the proposed framework are demonstrated using full-life testing data from heavy truck transmissions, the XJTU-SY and IMS benchmark bearing datasets.