Design of a parametrized numerical experiment for a 2D turbulent boundary layer flow with varying adverse pressure gradient and separation behaviour

Kurzfassung

The present work is part of the European project HiFi-TURB, which is a joint effort aiming to improve the turbulence models available and their capabilities to predict complex, industrially-relevant flows using novel machine learning techniques. One industry-relevant flow phenomenon, which is still difficult to predict using state-of-the-art Reynolds-Averaged Navier-Stokes (RANS) based turbulence models, is pressure induced flow separation. This report presents the design process and configuration of a parametrized numerical experiment for a 2D turbulent boundary layer flow subjected to an adverse pressure gradient. This design study, set up using RANS turbulence models, represents the foundation for subsequent highfidelity simulations such as Direct Numerical Simulation (DNS). Generated data sets, both from RANS computations and from high-fidelity compuations, are afterwards compared and utilized in the machine learning step to deduce novel correlations for improved predictions with RANS models. The focus of this work is thus to create several configurations (by varying geometry and inlet condition) in order to obtain varying flow conditions. Providing a range of configurations is essential for the training process to avoid overfitting and obtaining a more generalized model during the machine learning step. In the framework of the HiFi-TURB project, the Center for Computer Applications in AeroSpace Science and Engineering (C2A2S2E) at the DLR Institute of Aerodynamics and Flow Technology (DLR AS) is focusing especially on improving Differential Reynolds Stress Models using high-fidelity computations performed for the test case described in this work alongside selected machine learning algorithms.