Data-Driven Modeling Adaptive Aerostructures

Abstract In the growing domain of wind energy, morphing aerostructures provide an avenue of research to improve aerodynamic performance and turbine control. Branching off previous work by the authors, several data-driven models which relate the actuation of a linearly segmented active morphing blade to an arbitrary performance parameter, aerodynamic thrust load in the case examined, are developed, presented, and compared to evaluate accuracy. Models are developed for the Danish Technical University (DTU) 10 MW and National Renewable Laboratory (NERL) 15 MW reference turbines to ensure generalizability. The models are trained using OpenFAST simulations to generate training and test data. State vector machine (SVM), neural network (NN), decision tree, and gaussian process regression (GPR) models are developed. Comparing root-mean-square error (RMSE), the GPR method produces the least error, in most cases by an order of magnitude or more. The developed model is then used in a demonstrative application, an optimization of the target parameter. Particle swarm optimization is used to find the global minimum input space of the GPR model at discrete wind speeds. The references turbines show a respective maximum improvement in the target performance parameter of 3.15% and 2.66%. Optimization results indicate that refinement of the developed model may improve results.