Wind turbine anomaly detection using a support vector machine optimized with a Bayesian algorithm

Abstract

Wind energy has experienced significant growth in recent years; however, it still faces challenges in operation and maintenance, which impact energy efficiency and lead to high costs. This study proposes an anomaly detection model for wind turbines based on a support vector machine (SVM), optimized using Bayesian search. The model was trained using vibration data from the Gearbox Reliability Collaborative (GRC) project of the National Renewable Energy Laboratory (NREL), specifically from the High-Speed Shaft Upwind Bearing Radial and High-Speed Shaft Downwind Bearing Radial sensors, with 40,000 records evenly distributed between normal and anomalous conditions. The proposed model achieved an overall fault detection accuracy of 78.95% and 78.50% for the respective sensor data. Bayesian optimization facilitated the fine-tuning of the hyperparameters in the classification technique, enhancing the model's anomaly detection capability. Furthermore, the use of vibration data enabled the identification of critical fault patterns in turbine operation, contributing to the improvement of wind turbine efficiency and reliability.