Unsteady and nonlinear aerodynamic prediction of airfoil undergoing large-amplitude pitching oscillation based on gated recurrent unit network

In this paper, a reduced-order model (ROM) based on data-driven machine learning algorithm is constructed to identify the aerodynamic forces of airfoil undergoing large-amplitude pitching oscillation. Strong nonlinearity and unsteadiness in aerodynamics is a major challenge in the prediction of aerodynamic forces. To deal with this problem, the recurrent neural network (RNN) with gated recurrent unit (GRU) is applied for nonlinear and unsteady aerodynamic identification. A motion input signal which covers a wide range of frequency and amplitude is designed to enable the ROM with generalization capability. Shear stress transport (SST) model with low-Reynolds number modification is introduced into the computational fluid dynamics (CFD) method to calculate the aerodynamic forces as the training data. The time step size and lag order of the model are determined by the frequency domain characteristics of the training data. The results suggest that the proposed ROM has a high identification precision on nonlinear unsteady aerodynamics. The well-trained ROM could accurately predict the aerodynamic forces of airfoil undergoing sinusoidal oscillations with various frequencies and amplitudes. The proposed ROM shows advantages in accuracy over other ROM techniques. The calculation speed of ROM is 69 times faster than that of CFD method on the premise of accuracy, which can be expected a good application in engineering.