Data-driven predictive maintenance strategy considering the uncertainty in remaining useful life prediction

Remaining Useful Life (RUL) prediction and maintenance decision-making are two key tasks within the framework of Prognostics and Health Management (PHM) of system. However, existing works are performing the two tasks separately and hierarchically. Besides, the uncertainty in RUL prediction caused by cognitive level and measurement capabilities has not aroused wide concern and this may reduce the credibility of point prediction. To address these issues and finally ensure the safe and reliable operation of the system, this paper proposes a novel data-driven predictive maintenance strategy. The proposed strategy is a complete process from implementing the RUL prediction with uncertainty to making maintenance decision. Considering the prediction aspect, a Local Uncertainty Estimation (LUE) model with Bidirectional Long-Short Term Memory (Bi-LSTM) is proposed to characterize the uncertainty in RUL prediction. Regarding the post-prediction aspect, the Maintenance Cost Rate (MCR), namely maintenance cost per unit operational time, function is constructed by linking the constructed RUL distribution with maintenance-related costs. Oriented towards the economic requirements of operation management, the time for taking maintenance activities can be determined by optimizing the MCR function. The whole proposition is validated on a case study of the aero-engine health monitoring. The comparison with recent publications and the corresponding analysis results indicate that the proposed method is a promising tool in predictive maintenance applications, which can reduce system maintenance costs. (C) 2022 Elsevier B.V. All rights reserved.