A degradation-related slow feature analysis for equipment health indicator extraction and remaining useful life prediction

Predicting the Remaining Useful Life (RUL) of equipments has recently become a crucial technology for assessing operational safety and assisting maintenance decision-making. Numerous studies have demonstrated that a low-dimensional Health Indicator (HI) can be constructed from multidimensional sensor readings related to degradation, and the prediction of RUL can be based on similarities of HI. However, existing approaches for HI construction ignore neither the slow and monotonic nature of a degradation feature nor correlations between HI and RUL. To address this issue, this paper proposes a degradation-related slow feature analysis (DRSFA) method for extracting HIs and applying them in RUL prediction. Specifically, an objective function and its corresponding closed-form solution are proposed, aiming at extracting a health indicator from multidimensional degradation parameters to represent the slow degradation trend of an equipment and is correlated with its RUL. In DRSFA, HIs of each segment of lifecycle data is extracted separately rather than by a unified model, thereby enhancing its scalability as new data become available. As an HI extractor, DRSFA can serve as a plug-and-play module for RUL prediction based on similarity matching. Finally, experiments conducted on the CMAPSS dataset for aero-engine RUL assessment from NASA validate that the proposed method effectively balances RUL prediction accuracy, interpretability, and scalability.