Early Anomaly Detection in Wind Turbine Generators Using SCADA Data

Effective health monitoring of Wind Turbine Generators (WTGs) using anomaly detection techniques can significantly enhance operational reliability. This study presents an unsupervised learning framework designed to extract degradation signatures from Supervisory Control and Data Acquisition (SCADA) electrical time-series data, enabling early fault detection for condition-based maintenance. The proposed approach utilizes a long short-term memory autoencoder to compute a Health Index (HI), quantifying the deviation in operational behavior from a learned healthy baseline. This HI facilitates the precise identification of degradation onset, allowing timely intervention before severe faults occur. The framework was validated using a real-world wind turbine SCADA electrical signal dataset, demonstrating its effectiveness in capturing early signs of abnormality. Experimental results highlight the model's strong potential for supporting proactive maintenance strategies and minimizing unplanned downtime.