Machine Learning to Reconstruct Aeronautical Databases with Deep Neural Networks

In this work, two different interpolation methods are compared with the aim at reconstructing complete aeronautical fields. The first method consists on applying singular value decomposition to the database, combined with linear interpolation. In the second method, the linear interpolation will be replaced by a neural network with the aim at improving the results. The two methods have been applied to reconstruct an atmospheric boundary layer field. The results show that using neural networks improves the error made in the interpolation by two orders of magnitude. In addition, the neural network provides quite accurate results, reconstructing the stationary variables of the atmospheric boundary layer.