Change point severity-aware remaining useful life prediction

• A severity-aware framework for change point-based RUL estimation is proposed. • Normalized Euclidean Distance is used to quantify severity of the changepoint. • Severity levels guide model specialization in LSTM-based RUL prediction. As a cornerstone of the upcoming industrial revaluation, predictive maintenance leverages intelligent systems and real-time data processing to transform traditional industrial operations. A fundamental part of this maintenance strategy is estimating the Remaining Useful Life (RUL), which is the time remaining before a component or system requires maintenance or replacement. The onset of the degradation process, termed the change point, serves as a significant preliminary step for reliable RUL estimation of complex machinery. Existing works are often limited to only detecting the change point, without considering the intensity or severity of the change that occurs. In this work, we propose a novel framework that integrates severity-aware change detection into the RUL estimation process. The Normalized Euclidean Distance (NED) metric is introduced to quantify the severity of each change point, enabling a more accurate estimation of system degradation. This information is incorporated into a long short-term memory (LSTM)- based RUL estimation approach, allowing predictions to adapt based on the intensity of degradation in change points. Experimental evaluations on C-MAPSS benchmark datasets show that our method enhances prediction accuracy against state-of-the-art LSTM methods.