The Innovative CFD FAQ: Turbulence Modeling Questions
I have gotten a lot of questions about turbulence models over the years. I guess that's a reflection of the fact that, even after decades of research and development, we still don't have a single model which will reliably produce good results for a wide variety of geometries and flow condition. So, here are my thoughts on a few turbulence-related topics:
How can I compare the results of a steady-state RANS simulation with those from a Large Eddy Simulation (LES)?
Since LES is, by definition, an unsteady simulation, you must compute a time-averaged flowfield in order to make an apples-to-apples comparison with a RANS simulation. To do this properly, you must include enough time steps in your average that the mean flow reaches a statistically steady-state (i.e. the average stops changing as you add more snapshots to the average). If you compute averages of higher order quantities (such as uu, uv, vv, etc.) then you can also compute the resolved Reynolds stresses and compare that with the RANS model's predicted values.
What turbulence model should I use for [a specific flow configuration]?
The choice of turbulence model can become almost religious in the CFD community. Some people always use a Spalart-Allmaras model. Others swear by Menter's SST model or a Wilcox k-omega variant. For simple attached flows, you can even get good results from the old algebraic models. In general, I recommend using a Spalart or SST model (or similar k-omega model) unless you know that there is a model in the solver you are using which has been tuned for your specific class of problem. These do a pretty good job for most production applications, but keep in mind that no model is perfect, so make sure you evaluate the results carefully before declaring that you have “the answer”.
When you are finished here, you can return to the
main FAQ page.
Or you can head head back to
Innovative CFD home page
and browse the other topics.