Evaluation of physics constrained data-driven methods for turbulence model uncertainty quantification

In order to achieve a virtual certification process and robust designs for turbomachinery, the uncertainty bounds Computational Fluid Dynamics have be known. The formulation of turbulence closure models implies major source overall Reynolds-averaged Navier-Stokes simulations. We discuss common practice applying physics constrained eigenspace perturbation Reynolds stress tensor in account model form models. Since basic methodology often leads overly generous estimates, we extend recent approach adding machine learning strategy. application data-driven method is motivated by striving detection flow regions, which are prone suffer from lack prediction accuracy. this way any user input related choosing degree supposed become obsolete. This work especially investigates an approach, tries determine priori estimation confidence, when there no accurate data available judge prediction. around NACA 4412 airfoil at near-stall conditions demonstrates successful framework. Furthermore, highlight objectives limitations underlying methodology.