Remaining useful life prediction for stochastic deteriorating Devices: A direct approach via inverse degradation modeling

• Published in: Mechanical systems and signal processing Vol. 228; p. 112431
• Main Authors: Li, Tianmei, Cai, Zhenyu, Zeng, Zhaoju, Zhang, Zhengxin, Si, Xiaosheng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2025
• Subjects: Data-driven; Lifetime prediction; Prognostics and health management; Remaining useful life; Reverse degradation modeling; Lifetime prediction; Data-driven; Remaining useful life; Prognostics and health management; Reverse degradation modeling
• ISSN: 0888-3270
• DOI: 10.1016/j.ymssp.2025.112431

Remaining useful life (RUL) prediction has been extensively recognized for its fundamental and significant value in enhancing safety, improving reliability, and reducing cost for industrial devices. The advancement of condition monitoring (CM) for degrading devices stimulates the development and prosperity of data-driven prognosis approach for RUL prediction, among which the stochastic-data-driven methods have garnered much favor of researchers for its capability to characterize the uncertainty within the predicted RUL through a probability distribution. However, despite their urgent desirability for supporting efficient decision making on management activities, analytical RUL probability distributions remain challenging to obtain, except in exceptionally limited cases. To overcome the difficulty in deriving the RUL probability distribution, a novel direct RUL prediction approach based on an inverse degradation modeling framework has been presented in this paper. A semi-parametric inverse degradation model integrating population degrading characteristics via a parametric model and the unit-to-unit variability through a non-parametric model has been constructed. A two-stage approach, which identifies unknown parameters based on historical CM data in off-line stage and updates the model when new CM data of individual device are available in on-line stage, has been developed to achieve RUL prediction for an in-service degrading device. The proposed approach has been illustrated and validated by a case study of milling cutters.