CRULP: Reliable RUL Estimation Inspired by Conformal Prediction

Reliable Remaining Useful Life (RUL) estimation is vital for the Prognostics and Health Management (PHM) of industrial equipment. Despite the significant breakthroughs regarding the prediction accuracy offered by data-driven methods, the challenge of quantifying the reliability of these predictions remains. A trustworthy and reliable RUL estimation framework, named Conformalized RUL Predictor (CRULP), is proposed for industrial equipment by integrating the deep learning-based RUL prediction model and Uncertainty Awareness Conformalized Quantile Regression (UACQR). The proposed CRULP method innovatively converts the single-point RUL prediction model into construct uncertainty set (i.e. interval in RUL estimation) base on the epistemic uncertainty through a collaborative process of ensemble Conformalized Quantile Regression and Cross Conformal Prediction. Experiments conducted on the two wildly used aero engine datasets demonstrate that, the CRULP framework ensures the predicted RUL intervals are not only have desire coverage but also as narrow as possible, reflecting a high level of precision and reliability. This research has the potential to significantly impact the field of PHM by providing a robust and reliable approach to RUL estimation that can be applied across various scenarios.