Reconstruction of Fluid Flows Past Airfoils Using Neural Network

Present paper attempts to reconstruct 2D unsteady flow field using neural network when fluid flows past airfoils at low Reynolds number. First of all, the details of fluid domain are obtained by solving incompressible fluid governing equation using our own developed fluid solver based on local radial basis function (LRBF) method, and then some randomly selected tempo-spatial points (with velocities and pressure information) are fed into neural network to train. The training process of first step is to learn Reynolds number, continuing with reconstruction of fluid field and comparison with numerical results. The flow Reynolds number is set as 200, while angle of attack is 20°. Besides, the locally refined nodes distribution of spatial domain is to globally reduce the computing resource.