Calibration and Reduction of Large-Scale Dynamic Models - Application to Wind Turbine Blades

This thesis investigates the validity of structural dynamics models of wind turbine blades. An outlook on methods for model calibration to make models valid for their intended use is presented in the thesis. The intention is to make the models valid for robust predictions. The model validity is here assessed to be of hierarchical dual level. On one hand, a detailed structural dynamics model needs to be substantiated by good correlation between experimental results of wind turbine testing and theoretical simulation results using that model. On the other hand, after that detailed model has been validated, a model of significantly low order based on the detailed model has been validated by a good model-to-model correlation. With the connection between models, this implies that also the low order model is implicitly validated by testing. The development of a highly detailed structural dynamics model provides real physical insights to observation made during testing. This model is often developed using finite element analysis. A model verification and validation activity is done to create a three dimensional finite element model that is capable to predict the dynamics of wind turbine blade with sufficient accuracy. Integration of such large-scale models of wind turbine blades in aeroelastic simulations places an untenable demand on computational resources and, hence, means of speed-up become necessary. The common practice is to develop, calibrate and validate an industry-standard beam model against the simulated data obtained from the validated highly detailed rotor blade model. However, the validated beam model cannot well capture the coupling features of the highly detailed model because of its inherent limitations. Our scientific hypothesis is that it is possible to create low-order rotor blade models which preserve the vibrational pattern of the baseline model at its eigenfrequecies and also closely mimic its input-output behavior. Toward this end, a quasi optimal modal truncation algorithm is developed to yield reduced models which have the eigenmodes with highest contribution to the input-output map of the large-scale model. The predictive capability of the created reduced model is compared with that of the validated beam model