Turbulence modeling is one of the main sources of uncertainty in CFD simulations of technical flows. This is not surprising, as turbulence is the most complex phenomenon in classical physics. Turbulent flows pose a multi-scale problem, where the dimension of the technical device is often of the order of meters (or even meters in case of airplanes and ships) whereas the smallest turbulence vortices are of the order of - meters for high Reynolds number flows. Direct Numerical Simulation (DNS) of turbulence is therefore restricted to very small flow domains and low Reynolds numbers. Even the reduction in scales through Large Eddy Simulation (LES), does not lead to acceptable turn-around times for most technical flow simulations, especially in cases where wall boundary layers are important, as is most often the case (see Scale-Resolving Simulations in Ansys CFD ).

A practical solution to this dilemma is offered through the concept of Reynolds-Averaged Navier-Stokes (RANS) equations. Instead of resolving the turbulent structures in time and space and then averaging the solution to obtain the desired engineering mean flow quantities, one first averages the equations and solves directly for the time-mean (or ensemble mean) variables. While this is much more economical, it eliminates the turbulence-related physics from the equations. Turbulence models are then required to feed back that information to allow physically correct simulations. When using RANS turbulence models, one should not forget that these models are tasked to bridge many orders of magnitude in computing power relative to DNS. From this perspective, it is not surprising that RANS computations are prone to modeling errors of significant size. Unfortunately, the RANS-related uncertainty cannot be quantified reliably, as there are not only quantitative errors, but there is also the potential for qualitative failure, when predicting incorrect flow topologies. Still, properly selected RANS models work very well for many technical applications, and it is therefore essential to understand the strength and weaknesses of different models to achieve optimal solution accuracy.

The current document is not intended as a textbook on turbulence modeling. The reader is referred to the available textbooks [2] – [7] the ERCOFTAC best Practice Guidelines [8] as well as the Ansys Fluent and the Ansys CFX Theory and User documentation for deeper studies and for more details. The goal here is to guide the user through the process of optimal RANS model selection within the Ansys CFD codes, especially Ansys Fluent and Ansys CFX. Nevertheless, some of the equations required for the discussion are provided in 'Appendix A: Theory' so that the reader does not have to continuously revert to the Ansys Theory Documentation. Another document of relevance is Generalized k-omega Two-Equation Turbulence Model (GEKO) in Ansys CFD .