A Hybrid Machine Learning Framework for Predictive Maintenance Routing Optimization in Floating Offshore Wind Farms

Abstract

Operation and maintenance (O&M) activities for floating
offshore wind farms remain one of the main techno-economic barriers to large-scale commercial deployment. Long offshore distances, severe metocean conditions, degradation of critical components, limited vessel
availability, and the increasing adoption of tow-to-port repair strategies significantly reduce the effectiveness of conventional time-based preventive maintenance approaches. This study presents an integrated decision
support framework combining machine learning, stochastic degradation modelling, and maritime routing optimization to minimize the expected maintenance cost of floating offshore wind farms. The main contribution lies in the formulation of a disaggregated cost function that takes into account different items. The proposed framework is solved through a hybrid scheme. First, an XGBoost predictor estimates the short-term probability of failure for each turbine. Second, a cost priority index is used to identify candidate turbines for intervention. Finally, feasible maintenance routes for a heterogeneous fleet composed of Crew Transfer Vessels (CTV), Service Operation Vessels (SOV), and towing vessels are planned. A simulated case study based on Hywind Scotland floating wind farm and consisting of 14 floating wind turbines rated at 15MW and located 30km from the base port is developed to evaluate the proposed methodology. Three maintenance strategies are compared: purely corrective maintenance, threshold-based maintenance and Heuristic index-based maintenance. Results indicate that the proposed heuristic policy reduces thetotal annual maintenance cost by 30%-45% compared with conventionalmaintenance strategies. These improvements are mainly driven by lower
production losses, reduced deterioration state, and more efficient vessel utilization