Abstract Summary
The rapid growth of the wind industry sector has motivated increasing sizes of wind turbines, with corresponding natural frequencies becoming smaller, and eventually shifted into the dominant portions of the wind and wave loading spectra. As a result of this effect and the increasing regularity of extreme weather events, the fatigue assessment of offshore wind turbines has grown in importance. A key parameter in determining the fatigue life of offshore wind turbines is the stiffness of the foundation. However, high uncertainty is associated with the soil properties, which results in inaccurate estimation of the fatigue life. In order to update these estimates from information from the structure as is, measured data from tower sensors (accelerometers, strain gauges, etc.) can be exploited to update the foundation parameters of offshore wind turbine models. Some challenges however arise when attempting to do so, including the implied identifiability of different parametrizations of the foundation parameters, e.g. whether the stiffness matrix of the foundation should be represented using the stiffness or compliance parameters. The practical unidentifiability of parameters may lead to highly diverging optimal parameter estimates under use of different tools. Combined with the associated uncertainty of the soil properties, this can severely throw off accumulated damage estimates related to remaining useful life. In this work, the NREL 15 MW reference wind turbine has been modelled via use of the Simscape environment. The developed Simscape model has the advantage of being able to separately model the different components of a wind turbine, resulting in a coupled servo-hydro-aero-elastic model of a fixed-bottom offshore wind turbine. The foundation is modelled using linear coupled springs, which comprises translational, rotational, and cross-coupling springs in a series configuration. An additional advantage of the Simscape model is that the representation of the tower uses a greater number of degrees of freedom in comparison to industry standards such as OpenFast. The model is used to generate synthetic data to match those corresponding to an instrumented offshore wind turbine, assuming deployment of accelerometers at four different levels near the nacelle. An operational modal analysis (OMA) method, SSI-cov, is used on noise corrupted data to obtain estimates for the modal parameters (natural frequencies and mode shapes) of the wind turbine. These identified modal parameters are then contrasted to model estimated modal parameters, within a Bayesian model updating approach, to determine not only the optimal estimates of the stiffnesses in the coupled spring foundation but also their associated uncertainties. The results show that the maximum a posteriori (MAP) estimates for all three parameters are close to their true values and that the translational spring stiffness is the most identifiable, whilst the rotational stiffness is characterized by the highest uncertainty. Discussions are produced regarding the parametrizations of the foundation stiffness matrix and the consequences of this choice to the identifiability of the related parameters.
