CFD 기반 인공신경망을 이용한 로터 블레이드 에어포일 공력 예측 모델

Abstract

Rotor blade analysis and design process typically use airfoil aerodynamic tables with two-dimensional quasi-steady flow assumption. Thus, its performance highly depends on the accuracy of the airfoil table. However, constructing an airfoil table is not trivial for airfoil shape design problems where the number of design parameters significantly increases. In this study, a surrogate model based on artificial neural networks is developed for airfoil aerodynamic tables in rotorcraft applications. The shape of airfoils is parameterized through Class-Shape Transformation(CST) using coefficients of a Bernstein polynomial expansion. The training data is comprised of 530 shapes sampled from perturbations of the CST coefficients around ten baseline airfoils. The aerodynamic coefficients for each shape are calculated using fully-turbulent flow simulations in the subsonic and transonic regimes. The range of angles of attack also includes stall angle to include any nonlinearities in model prediction. Overall, the trained model accurately predicts aerodynamic coefficients of non-trained airfoil shapes. The variation of model accuracy is also analyzed between interpolated and extrapolated spaces.