Gaussian Process-based surrogate modeling for aerodynamic performance prediction of upright and tilted VAWTs

Fostering sustainable solutions is essential in the current energy transition landscape. H-type vertical axis wind turbines have shown great potential in possible future offshore installations but innovative solutions are still needed. Experiments and numerical simulations are currently used to address the complex, inherent unsteadiness, resulting from dynamic stall, flow separation, and reduced lift, ultimately lowering efficiency. In this work, a machine learning approach is dedicated to the performance prediction of VAWTs operating in tilted position. Gaussian processes have been adopted to predict the local aerodynamics of VAWTs. The methodology has been developed through coupling with 3D CFD simulations relying upon a well-established numerical setting, already validated against experimental data. A set of test cases has been considered to assess the reliability of the machine learning code against varying TSR values and tilt angles. Results indicated that the adopted Gaussian-based regression models ensure very good fitting ability. Results at different operating conditions are presented to characterize the capability of the algorithm also with respect to operating conditions outside the data-driven field, with strong link with physical issues occurring in different rotor regions.