Identification of linear fast-varying natural frequencies for offshore wind turbines

Offshore Wind Turbines (OWTs) play a crucial role in mitigating climate change and reaching net-zero goals. Their health status is assessed based on the modal parameters (e.g., natural frequencies) obtained with Operational Modal Analysis (OMA). However, OMA relies on the assumption that the underlying system, and therefore the natural frequencies, are time-invariant. In practice, natural frequencies vary due to operational conditions. Typically, these variations are considered as slow-varying, and vibration measurements are divided into sufficient durations that the behaviour is time-invariant. This approximation fails when the natural frequencies are fast-varying. For instance, in a two-blade OWT, the natural frequencies are highly sensitive to the blade azimuth angle due to the lack of rotational symmetry, causing them to vary continuously as the blades rotate. These variations can be easily misinterpreted as damage or confounding influences if the natural frequencies are not accurately identified. State-of-the-art OMA techniques fail to identify real-time natural frequencies, highlighting the need for advanced techniques. Data-driven methods have been previously proposed in the literature to overcome the time-invariant assumption. Nonetheless, the accuracy of these methods is limited by insufficient and unlabelled sensor data, unobserved operational conditions, and a lack of physical interpretation. Moreover, these methods fail to identify fast-varying natural frequencies. This work proposes a hybrid OMA method that integrates domain knowledge of operating conditions with a data-driven method to identify fast-varying natural frequencies. Synthetic data is generated from a finite element model of a two-blade OWT under realistic environmental and operational conditions. The results demonstrate that the hybrid OMA method improves the accuracy of time-varying natural frequency identification.