Dynamic Predictive Maintenance Decisions Based on System Remaining Useful Life Prediction and Three Inspection Strategies
Lubing Wang, Zhengbo Zhu, Xufeng Zhao
Abstract
Remaining useful life (RUL) prediction and predictive maintenance decisions are two significant research problems in system prognostics and health management (PHM). However, most existing studies have executed these two problems separately and hierarchically, and rarely integrate the two. To solve this problem, this study presents a complete dynamic predictive maintenance decision framework that integrates RUL prediction based on a hybrid deep learning model into the predictive maintenance decisions. Considering the RUL prediction aspect, an advanced hybrid deep learning model is developed to predict system RUL effectively. Meanwhile, a Bayesian optimization method is proposed to further improve RUL prediction performance. Regarding the post-prognostic aspect, we consider periodic, descending, and state-based inspection strategies to help decision-makers formulate predictive maintenance decisions. Besides, a function related to the maintenance cost rate is developed to link the predicted RUL with three inspection strategy costs, which aims to reduce the maintenance cost of the system. Finally, experimental results show that the proposed dynamic predictive maintenance decision framework has superior performance comparing existing prediction methods and can provide decision-makers with optimal maintenance plans.