ME Thesis Presentation: Real-Time Anomaly Detector for Wind Turbines

Student: Zakir Ahmed

Advisor: Vinay Vaishampayan

Abstract: Wind energy is a rapidly growing power source, but wind turbines are vulnerable to operational anomalies and cyber-physical attacks that can compromise reliability. This thesis presents ZCAD (Zonal Condition Anamoly Detector), a real-time wind turbine anomaly detection system that combines a Q-Blade based wind turbine simulation environment with an advanced machine learning pipeline. Instead of relying on physical sensors or hardware prototypes, ZCAD uses Q-Blade to simulate key turbine parameters such as blade pitch angle, yaw position, wind speed, generator voltage, rotor speed, and power output. Normal behavior is modeled using turbulent wind field and hub-height input profiles, while controlled modifications of yaw, pitch, rotor speed, and torque parameters are used to inject fault scenarios including stuck conditions, imbalances, and overspeed events. The resulting labeled time-series datasets enable rigorous testing of anomaly detection methods across realistic operational and fault conditions.

At the core of ZCAD is a hybrid voting-based machine learning ensemble composed of multiple supervised and unsupervised models. Supervised models (e.g., Random Forest, XGBoost, SVM, Logistic Regression) are trained using Q-Blade data to identify known fault signatures. Unsupervised models (e.g., Isolation Forest, LOF, DBSCAN, Mahalanobis Distance, PCA) learn the normal operational profile and flag statistical outliers. These models operate in parallel, and ZCAD applies a majority-voting mechanism to produce real-time anomaly decisions. This edge-optimized system simulates local turbine monitoring, with immediate detection capability and minimal bandwidth requirements compared to centralized SCADA monitoring. Simulation-based experiments using QBlade-generated datasets demonstrate that ZCAD can detect a wide range of anomalies with over 94% accuracy and low false alarm rates, outperforming traditional single-method detectors. The results support ZCAD’s viability as a robust anomaly detection layer for smart wind energy systems.