Multi-Interval-Aggregation Failure Point Approximation for Remaining Useful Life Prediction

Dear Editor, This letter focuses on the remaining useful life (RUL) prediction task under limited labeled samples. Existing machine-learning-based RUL prediction methods for this task usually pay attention to mining degradation information to improve the prediction accuracy of degradation value or health indicator for the next epoch. However, they ignore the cumulative prediction error caused by iterations before reaching the failure point. Therefore, we propose a performance improvement framework called multi-interval-aggregation failure point approximation (MFA) consisting of multi-interval-aggregation prediction and failure point approximation strategy. Multi-interval-aggregation prediction aggregates prediction models with multiple intervals to predict the degradation value or health indicator after any epoch under our proposed minimum iteration principle. The failure point approximation strategy roughly locates the failure point by the model with maximum interval, then continuously approximates the real failure point by aggregating the prediction models with multiple intervals. We conducted numerous experiments in two actual cases, and the experiments demonstrate that our proposed MFA can significantly improve the accuracy and reliability of RUL prediction tasks. Codes will be available at https://github.com/foryichuanqi/IEEE-CAA-paper-2025-Multi-interval-aggregation-failure-point-approximation-for-RUL-prediction after the letter is online.