Fast flow field prediction over airfoils using deep learning approach

In this paper, a data driven approach is presented for the prediction of incompressible laminar steady flow field over airfoils based on the combination of deep Convolutional Neural Network (CNN) and deep Multilayer Perceptron (MLP). The flow field over an airfoil depends on the airfoil geometry, Reynolds number, and angle of attack. In conventional approaches, Navier-Stokes (NS) equations are solved on a computational mesh with corresponding boundary conditions to obtain the flow solutions, which is a time consuming task. In the present approach, the flow field over an airfoil is approximated as a function of airfoil geometry, Reynolds number, and angle of attack using deep neural networks without solving the NS equations. The present approach consists of two steps. First, CNN is employed to extract the geometrical parameters from airfoil shapes. Then, the extracted geometrical parameters along with Reynolds number and angle of attack are fed as input to the MLP network to obtain an approximate model to predict the flow field. The required database for the network training is generated using the OpenFOAM solver by solving NS equations. Once the training is done, the flow field around an airfoil can be obtained in seconds. From the prediction results, it is evident that the approach is efficient and accurate.