Progress on the Field Inversion and Machine Learning Approach with the DLR TAU Code

Kurzfassung

Accurate prediction of turbulent flows is important due to their typical key roles in engineering and in science. However, scale-resolving methods like LES and DNS are often still too expensive for daily use and Reynolds-averaged Navier-Stokes (RANS) simulations are used instead. Here, the influence of turbulence on the mean flow is modeled via a turbulence model. RANS turbulence models have always been largely empirical and their development supported by data. Given the increasing availability of highly accurate data from DNS and LES simulations as well as from experiments, and the increasing maturity of machine learning techniques it is argued that RANS turbulence models could be improved further using machine learning.