Uncertainty-aware remaining useful life prediction for predictive maintenance using deep learning

Abstract

Reliably predicting Remaining Useful Life (RUL) is crucial for reducing asset maintenance costs. Deep learning emerges as a powerful data-driven method capable of predicting RUL based on historical operating data. However, standard deep learning tools typically do not account for the uncertainty inherent in prediction tasks. This paper presents an uncertainty-aware approach that predicts not only the RUL but also the associated confidence interval, capturing both aleatoric and epistemic uncertainty. The proposed approach is evaluated on publicly available datasets of aircraft turbofan engines, showing its ability to estimate accurate RUL and well-calibrated uncertainties that are robust to out-of-distribution data.