Physics‑informed clustering and ESD‑based outlier detection for wind turbine SCADA data

Outlier detection (OD) in wind turbine supervisory control and data acquisition (SCADA) data is essential for reliable condition monitoring, as missed anomalies or false alarms can increase operation and maintenance costs. However, the complex interdependencies and strong operational variability of SCADA data limit the effectiveness of conventional univariate and multivariate OD methods, which often fail to capture contextual dependencies or overlook subtle single‑feature deviations. This study presents a physics‑informed, clustering‑based OD framework that incorporates operational knowledge of wind turbines into statistical hypothesis testing. SCADA data are first partitioned into operationally coherent clusters using physically meaningful variables, and the Generalized Extreme Studentized Deviate (ESD) test is subsequently applied within each cluster to identify context‑dependent outliers. This approach enables adaptive detection while maintaining sensitivity to subtle anomalies that may be masked in complex multivariate spaces. The proposed method is evaluated using 26 SCADA datasets from one wind farm and validated on 14 additional datasets from an independent site. Results demonstrate an average detection sensitivity of 98.15 %, consistently outperforming 11 classical and state‑of‑the‑art OD techniques. Its robust performance across turbines without hyperparameter retuning highlights the generalization capability of the framework and its suitability for large‑scale, data‑driven condition monitoring in modern wind energy systems.