Deep Learning for prediction of Aerodynamic Simulations

The motivation of this work is to conjugate the potential of Artificial Neural Networks and the Aerodynamic field. Artificial Neural Networks (ANNs) are a constantly growing field of study, every day new applications in different fields such as Engineering, Medical, Data Science, or even ’Everyday Life’ are found. On the other hand, Computational Fluid Dynamic (CFD) is the ’state of the art’ of the Aerodynamic analysis field, nowadays there’s plenty of CFD software, both commercial and open-source, capable of simulating the most complex fluid model and geometry. The goal is to develop a Neural Network (in this specific case a Multi-Layer Perceptron) that can predict the results of CFD simulations of a 2D airfoil giving as input three parameters: the angle of attack of the airfoil with respect to the free-stream velocity, the Mach number and the Reynolds number. Two Neural Networks have been implemented: a Multi-Layer Perceptron has been trained to predict the two aerodynamic coefficients CL (Lift coefficient) and CD (Drag coefficient) of the airfoil; then, a second Multi-Layer Perceptron has been trained to predict the velocity field of the fluid around the airfoil.