Presentations

Abstract

A digital twin is a virtual counterpart of a physical asset, such as an offshore wind farm, allowing close monitoring of the asset and systematic evaluation of operational scenarios. For applications such as energy yield assessment and long-term evaluation of turbine loads, it is crucial to ensure transparency, reliability and accuracy of the underlying sub-models. Therefore, proper calibration of these sub-models is necessary to generate actionable insights. To this end, an integrated framework for constructing digital twins of offshore wind farms is presented, which is calibrated using multiple types of field measurements. The framework is designed to support scenario analysis focused on annual expected energy production (AEP) and medium- to long-term load assessments, where wake interactions and the impact of turbine control play a central role. First, scalable hyperparameter tuning is used to calibrate analytical wake models, e.g., the Gauss-Curl Hybrid model, using field data. Calibration methods based on SCADA anemometer data and scanning lidar measurements are compared. For the scanning lidar, the wind field is reconstructed to estimate horizontal wind speeds from line-of-sight velocity data. These estimates are combined with SCADA wind direction data and the wind farm layout to identify wake-induced velocity deficits, which are then used for hyperparameter tuning. In the case of SCADA data, a time-series-based calibration approach is applied to anemometer wind speed and active power measurements of the turbines. Second, accelerometer sensors are installed on drivetrain components. The vibration measurements are combined with SCADA data to perform operational modal analysis to extract the eigenfrequencies and compare modal behaviors. The resulting modes are used to calibrate a turbine model in OpenFAST, obtained through scaling of open-source reference turbines, such as the LEANWIND 8MW and DTU 10MW turbines. Analytical models are then used to reflect the simulated high-level loading to the level of individual machine components (e.g., main bearing, pitch bearing). Third, the wake and load models are integrated into a single digital twin of the wind farm, which allows to predict both wear-out component loads and produced powers under various scenarios, such as normal operation or controller actions (e.g., grid curtailments). The full digital twin is calibrated on data of multiple offshore wind parks within the Belgian-Dutch North Sea area, covering data from in-house accelerometers, lidars, and SCADA-1s. To demonstrate the framework, the long-term wind resource at a wind farm site is captured using ERA5 reanalysis data. The digital twin is then used to estimate the long-term expected production using a combination of field data and reanalysis data. These outcomes are then compared against the pre-construction energy yield budget. In addition, damage equivalent loads of drivetrain components are estimated during curtailments and compared with the counterfactual conditions of uninterrupted normal operation. The results highlight discrepancies between initial model assumptions and the lessons learned during the asset’s project lifetime, and quantifies the net impact of curtailments on component lifetime.