Unexpected failures in industrial asset networks—like offshore wind farms, hospital imaging systems, and semiconductor equipment—can cause major financial and operational disruptions. To reduce unplanned downtime, preventive maintenance strategies are essential. This thesis presents a scalable framework for optimizing maintenance scheduling in complex, geographically dispersed asset networks, where predicting failures is difficult and resources are limited.nnThe research is divided into four parts. The first introduces the “dynamic traveling maintainer problem with alerts,” proposing heuristic and deep reinforcement learning (DRL) methods to dispatch maintenance resources efficiently. The second part extends this to multiple maintainers, demonstrating that DRL can scale to larger networks and adapt to changes. The third part incorporates asset interdependencies and uncertainty in degradation processes, using sensor data and Bayesian decision models to guide opportunistic maintenance. The fourth part shifts focus to average cost optimization, introducing statistical methods to estimate cost-effectiveness ratios and confidence regions, enhancing the reliability and explainability of DRL-based policies.nnCollectively, the work shows that DRL offers robust, scalable solutions for maintenance scheduling, integrating real-time data, managing uncertainty, and improving cost-efficiency and reliability across industrial sectors.