An investigation of neural networks for aerodynamic predictions

This work proposes a novel multi-output neural network for the prediction of the lift coefficient of aerofoils using inviscid compressible flow data. Contrary to existing neural networks that are designed to predict aerodynamic quantities of interest, the proposed network considers as output the pressure at a number of selected points on the aerofoil surface. The proposed approach is compared against the more traditional network where the lift coefficient is directly the only output of the network. Furthermore, a detailed comparison of the proposed neural network against the popular proper orthogonal decomposition method is presented. The numerical results, involving high dimensional problems with flow and geometric parameters, show the benefits of the proposed approach.