Bearing Fault Identification for High-Speed Wind Turbines using CNN

Wind turbines condition monitoring and fault warning have significant importance in terms of reducing maintenance costs and improving operation levels of wind farms. Since high-speed shaft bearings account for a major share of wind turbine drivetrain failures particularly offshore ones, there is an urgent need to monitor the health status of wind turbine elements accurately. In this paper, Spectral Kurtosis (SK) data driven approach is utilized to provide a vibration-based prognostic and health monitoring methodology for wind turbine high-speed shaft bearing. Three data-driven machine learning models are applied: Support Vector Machine (SVM), Convolutional Neural Network (CNN), and Random Forest (RF) to detect and classify remaining useful time in bearing through vibration condition monitoring. The results of the study carried out in this paper using actual wind turbine vibration dataset reveal that the CNN approach surpasses SVM and RF techniques in achieving higher accuracy and better classification performance.