Maintenance Optimization for Offshore Wind Turbines using POMDP

In this work, a Partially Observable Markov Decision Process (POMDP) is used for decision support for offshore wind turbines. The optimal decision policies for inspection and repair are obtained for each time step dependent on the belief state for the damage state after monitoring, and thus dynamic programming can be used. Seasonal weather variations are included through their influence on weather constrains for inspections and repairs, as well as costs to lost production, when corrective repair is not possible after failure. Application of the model is illustrated through an example, where the main bearing is considered. Optimization is initially performed for one component, and decision making for an entire wind farm is considered by using revised decision policies, when mobilization costs are already paid for another repair. The total costs are calculated for an entire wind farm using simulation and by using the identified decision policies.