A predictive maintenance framework based on real-time credibility evaluation of remaining useful life prediction results

Abstract

The increasing availability of remaining useful life (RUL) prediction methods has incentivized the development of predictive maintenance (PdM) for engineering systems. The performance of RUL prediction results is often expected to improve as more condition monitoring data are collected. However, achieving a credible RUL prediction result remains a critical challenge that is often overlooked in current PdM literature. This article proposes a PdM framework to optimize maintenance plans by a PdM utility model correlates the expected maintenance net revenues and losses with the credibility of RUL prediction result to determine the optimal PdM timing. In addition, considering the dynamic characteristics of PdM decision-making driven by condition monitoring data and on the corresponding RUL prediction results, an updating strategy that control the updating frequency is proposed to minimize computational resource waste and avoid decision redundancy. Finally, the proposed PdM framework is validated using the C-MAPSS dataset of turbofan engines.