Policies for the dynamic traveling maintainer problem with alerts

• We develop the Dynamic Traveling Maintainer Problem with Alerts on asset networks. Information levels rank quality of information retrieved from alerts. Proposed heuristics yield competitive policies for each level. A reinforcement learning heuristic outperforms others, using least information. Downtime industrial assets such as wind turbines and medical imaging devices comes at a sharp cost. To avoid downtime costs, companies seek to initiate maintenance just before failure. Unfortunately, this is challenging following two reasons: On one hand, because failures are notoriously difficult predict, even in presence real-time monitoring which signal early degradation. other available resources serve network geographically dispersed typically limited. In paper, we propose novel model referred dynamic traveling maintainer problem alerts that incorporates these challenges provide three solution approaches how dispatch limited resources. Namely, propose: (i) Greedy urgency, proximity economic risk; (ii) approach optimizes short-term costs; (iii) deep (DRL) long-term costs. Each has different requirements concerning alert Experiments small networks show all methods can approximate optimal policy when given access complete condition For larger networks, proposed policies, DRL consistently achieving lowest