Recent Advances in Data-Driven Modeling for Aerodynamic Applications using DLR's SMARTy Toolbox

From aircraft design to certification a huge amount of aerodynamic data is needed. In order to fulfil different requirements this data covers the entire flight envelope and includes pressure and shear stress distributions, global coefficients as well as derivatives. Moreover, data is typically gathered from different sources including flight tests, wind tunnel experiments or numerical simulations, and they are often available at various levels of fidelity, ranging from simple hand book methods to high-fidelity simulations. Within the past few years, the demand for efficient exploitation and exploration of these data sources became evident to further enhance existing designs, evaluate new technical capabilities and foster the availability of high-fidelity aerodynamic data in closely related disciplines. Driven by this, the aim of data-driven methods is to provide consistent aerodynamic data models based on various data-sources but with lower evaluation time and storage than the original models. Especially the increased availability of tools together with hardware in the field of machine learning in general and deep learning in particular has further accelerated demands as well as developments. In this paper we will show some recent advances in the field of data-driven modeling with a focus on applied aerodynamic challenges. This contains multi-fidelity modeling for aero-performance and stability and control databases, reduced-order modeling based on graph neural networks for the prediction of surface pressure distributions as well as data fusion to combine results from numerical and experimental analysis with different spatial resolutions. For all methods results will be shown for an industrial-relevant military configuration.