Posters

Abstract

Wind farm operators face significant economic losses not only from unavoidable grid limitations but also from preventable causes such as inefficiencies, maintenance delays, and especially turbine component failures. Anticipating these failures before they escalate is critical to improving efficiency, reducing downtime, and ensuring profitability.  This work introduces an early detection system based on Quantile Regression models applied to SCADA data. These models establish adaptive performance ranges under varying operating conditions—such as power output and ambient temperature—allowing the system to detect anomalies in turbine component behavior. Being unsupervised, the approach eliminates the need for pre-labeled events, enabling faster implementation and broader scalability.  The methodology involves systematic feature selection, model training, and the computation of a Health Index (HI) for each component. By monitoring deviations from expected performance ranges, the system can anticipate failures and calculate each component’s Remaining Useful Life (RUL). These indicators are then integrated into preventive maintenance planning, with configurable thresholds that trigger warnings and critical alerts.  Results demonstrate the model’s ability to both identify component health status and accurately estimate useful life, providing operators with actionable insights. The integration of HI and RUL forecasts into preventive maintenance workflows reduces unplanned outages, optimizes resource allocation, and enhances cost-effectiveness in wind farm management.  Overall, this research validates the potential of quantile regression as a practical tool for predictive maintenance. By transforming raw SCADA data into health insights, the proposed system empowers operators to move from reactive to predictive strategies, achieving smarter asset management and greater operational reliability.