Predictive maintenance based on remaining useful life (RUL) prediction has attracted increasing attention in recent years due to its ability to leverage future equipment states for decision-making. Moreover, in practical engineering scenarios, imperfect maintenance is often adopted because of its cost-effectiveness in reducing the high cost of equipment replacement. However, existing methods that consider RUL prediction under imperfect maintenance involve complex solution procedures, thereby limiting the real-time capabilities of predictive maintenance decision-making. Furthermore, research in maintenance decision-making considering imperfect maintenance is often confined to solving for fixed decision variables, which restricts the flexibility and availability of decisions. To address these limitations, this paper proposes an online predictive maintenance decision-making framework considering imperfect maintenance. Firstly, we propose a data-driven prediction method considering the impact of imperfect maintenance, which can provide more accurate prediction results as state input for subsequent decisions in real time. Secondly, the decision-making problem including imperfect maintenance is modeled as a deep reinforcement learning problem, and real-time dynamic decision-making output is achieved by the trained agent. The proposed framework is applied to the maintenance of aircraft turbofan engines, demonstrating superior economic benefits compared to other maintenance strategies.

Imperfect maintenance,Predictive maintenance,Deep reinforcement learning,Remaining useful life