Posters

Abstract

Digital monitoring technologies and changes in EU law (Data Act) have significantly expanded the diversity of data available from wind turbines. Blade vibration data is one such example and it is increasingly used in monitoring systems to detect blade-related issues including structural damage (e.g. cracks or delamination), mass imbalance, and environmental effects such as icing. This work expands the use of blade vibration data far beyond the blades themselves: we show blade vibration data can be used as a turbine-level health indicator. We are motivated by the observation that blade vibrations reflect both blade integrity and the coupled movements of the rotor, nacelle, and drivetrain, meaning that changes in drivetrain condition can manifest in vibration patterns measured at the blade. We use a data-driven approach through which an artificial neural network simulates the expected blade vibration behaviour based on patterns learned from historical operational data. Deviations from the simulated normal blade vibrations then indicate systemic anomalies such as turbine component failures. Our approach is not limited to a specific sensor or hardware type and can be applied using existing blade-level accelerometers already deployed for purposes such as ice detection, without relying on proprietary data or predefined fault signatures. In a real-world case study, our AI system identified sustained deviations between the expected blade vibrations simulated by the neural network and actual blade vibrations measured on site. The data pattern was indicative of a drivetrain issue. The deviations started months in advance of routine inspections. A planned inspection later confirmed our finding by detecting material wear due to friction in the drivetrain. Overall, the results demonstrate how state-of-the-art AI techniques can leverage blade vibration data not only to detect blade faults, but also to provide insights beyond the blades, including drivetrain-related anomalies.