Enhanced Deep Reinforcement Learning for Selective Maintenance with Learning Effects

This paper introduces a novel framework for addressing selective maintenance problems with imperfect repairs over a finite horizon. The proposed approach relies on two deep Q-learning architectures to optimize system reliability while minimizing maintenance costs, incorporating both learning effects on maintenance durations and energy consumption optimization. To further enhance the proposed framework, we integrate case-based reasoning, allowing the system to draw insights from previously solved cases. Applied to a real-world coal transportation system, the proposed framework outperforms state-of-the-art algorithms, delivering high-quality solutions with minimal computational effort. The results support the significant role of case-based reasoning and learning effects in improving overall system performances.