Deep learning of survival functions with hazard constraints for semi-supervised RUL prediction from time-series data

Abstract

Accurate prediction of remaining useful life (RUL) is crucial for efficient maintenance of industrial equipment. Although various deep learning-based RUL prediction methods have been studied in recent years, effectively utilizing unlabeled data and long RUL data remains challenging. In this study, we apply a previously proposed method to the semi-supervised RUL prediction using time-series data. In this approach, a survival function modeled by neural networks is learned under hazard constraints. Introducing the survival function allows for consistent probabilistic handling of both labeled and unlabeled data. Experimental results on the CMAPSS dataset demonstrate that the proposed method outperforms baseline approaches.