Model updating of rotating wind turbines in operation for blade condition assessment

Abstract

As maintenance strategy becomes a central concern in the development of wind energy infrastructure, ensuring the reliability and availability of wind turbines requires increasingly accurate structural models. The growing use of sensors, including those embedded in the blades, provides access to rich vibration data that can be exploited to monitor the structural state of turbines in operation. In this context, model updating —the process of refining numerical model parameters using measurement data —plays a key role not only in improving model accuracy but also in structural health monitoring, enabling the detection and characterization of faults. However, the structural complexity of modern wind turbines, combined with numerous uncertain or unknown parameters, makes accurate predictive modelling challenging. Moreover, the time-periodic dynamics induced by rotor rotation violate key assumptions of classical system identification techniques, which are based on Linear Time-Invariant (LTI) formulations. As a result, model updating approaches for rotating wind turbines remain limited. This study proposes a dedicated framework for model updating of rotating wind turbines based on an equivalent Linear Time-Invariant (LTI) system, derived from a Fourier decomposition of the Floquet modes of the rotating system. This equivalent LTI system allows classical Operational Modal Analysis (OMA) techniques to be directly applied to measured time series, even for time-periodic systems. The modes identified through this approach correspond to those of the equivalent LTI system. By comparing the measured and simulated modes within a cost function, the parameters of the numerical model can be updated to better match the real structure. The full framework is summarized in the corresponding schematic. The main contribution of this study is the introduction of a novel model updating framework for rotating wind turbines, enabling the estimation of blade parameters and thus the detection of potential structural faults during operation. This approach makes it possible to apply classical identification techniques to systems with time-periodic dynamics. It therefore allows to improve digital twin accuracy and enable condition-based maintenance in wind energy systems. To demonstrate its effectiveness, the method is first validated on a simplified 5 Degree of-Freedom (DoF) wind turbine model, showing its ability to accurately update individual blade stiffness parameters using synthetic vibration data. Building on this preliminary case, the framework is then applied to the NREL 5 MW reference turbine, simulated with the aeroelastic code OpenFAST, involving about 80 DoF. This large-scale application enables testing under realistic and computationally demanding conditions. Challenges include noisy, multi-sensor data and reduced interpretability of the LTI equivalence. This study illustrates that the proposed approach is effective for analyzing complex turbine models and monitoring blade health in operation.