Dynamic predictive maintenance strategy for the multi-state system based on remaining life prediction

In research on health management of complex systems, most predictive maintenance approaches focus on a single aspect of it, often lacking a holistic approach. To address this gap, we propose a comprehensive dynamic predictive maintenance strategy based on the remaining useful life (RUL) prediction method, designed to enable real-time system monitoring, dynamic forecasting, and optimized maintenance decision-making. Firstly, an integrated 1D-CNN-Informer prediction framework is introduced, which combines one-dimensional convolutional neural networks (1D-CNN) with the Informer model to predict RUL effectively. Secondly, based on the RUL predicted by the hybrid model, a dynamic predictive maintenance strategy for the system is developed. This strategy encompasses several key components, including spare parts ordering, inventory management, and maintenance decision-making, thereby forming a closed-loop maintenance decision system. For spare parts ordering, we further proposed a multi-state spare parts ordering strategy to optimize procurement decisions. This strategy dynamically determines the ordering status by evaluating the expected costs of inventory and out-of-stock, ensuring that overall costs are minimized while maintaining system reliability. Ultimately, the results of the experiment based on the turbofan engine dataset reveal that, compared to existing predictive maintenance strategies, the dynamic predictive maintenance framework we propose not only achieves more precise predictions but also demonstrates significant advantages in optimizing maintenance decisions.