Surrogate Modeling of Flight Behavior With Deep Neural Networks

This paper explores the efficacy of generating aerodynamic data using a neural network as a surrogate model. We show that for the NASA N2A Hybrid Wing Body (HWB) aircraft and the NASA Generic T-tail (GTT) aircraft, we can generate highly accurate approximations of certain aerodynamic features, once that surrogate model has been trained on FlightStream® data. We suggest the use of a neural network surrogate model as a new approach for calculating approximate aerodynamic coefficients efficiently [ 5] and in a comparatively short time frame compared to vorticity solving software such as FlightStream® [2].