Intelligent global dynamic maintenance strategy for multi-component systems considering imperfect maintenance using deep reinforcement learning

Predictive maintenance (PdM), as a core technology of prognostics and health management (PHM), has increasingly been recognized for its potential to reduce operational costs and support condition-based maintenance decision-making. Nevertheless, conventional approaches are often constrained by insufficient real-time adaptability and limited robustness, particularly in complex systems with multiple heterogeneous components. These challenges become more pronounced in the presence of dynamically evolving monitoring data, uncertain maintenance effects, and resource-coupled constraints. To overcome these limitations, this study introduces an intelligent global dynamic maintenance strategy that integrates dynamic opportunistic imperfect maintenance with rolling-horizon-based group maintenance optimization. By leveraging condition monitoring data, the proposed framework enables the joint optimization of predictive and opportunistic maintenance decisions over an infinite time horizon. A case study based on degradation data from the main drivetrain of a wind turbine demonstrates that the proposed strategy substantially decreases average maintenance costs while effectively utilizing maintenance opportunities. The results highlight the effectiveness of integrating adaptive health prediction, imperfect maintenance, and dynamic group maintenance optimization for multi-component systems.