Kriging meta-models for damage equivalent load assessment of idling offshore wind turbines

Abstract. Lifetime reassessments of offshore wind turbines are very time consuming due to the large number of required simulations. As a result, the use of meta-models as surrogate models of the aeroelastic simulation model could offer a suitable alternative to simulations in the time domain (e.g., Kriging, artificial neural networks, or polynomial chaos expansion). Meta-models for the approximation of fatigue loads, i.e., damage equivalent loads, of wind turbines in normal operation have been researched comprehensively in recent years. Especially for offshore wind turbines, however, the downtimes, i.e, the times when the wind turbine is idling, also have a significant impact on the lifetime. For this reason, the creation of meta-models, more precisely Kriging meta-models, for an idling offshore wind turbine is investigated comprehensively in this paper. To ensure that the fatigue loads for the training and test data are not influenced by the initial transients at the start of the simulation, the run-in times are determined first. The subsequent investigation of meta-modelling shows that for the approximation of the rotor blade root bending moments, two additional input parameters have to be considered in addition to the input parameters that are used for the creation of a meta-model for the same offshore wind turbine in normal operation. The comprehensive investigation of the Kriging meta-models shows that the meta-models trained with 2000 data points represent the simulation model with an acceptable approximation quality when choosing suitable Kriging settings.