Choosing Turbulence Models for Your Applications

There are so many different turbulence models that no single code can contain even a small subset of them. As a result, when choosing a model for an application, you will be constrained by what your solver has implemented.

Likewise, no web page can possibly discuss every model, and my ability to give advice on how to make the choice is limited. However, even though I can't know which code you want to use or your specific case, there are still some useful rules of thumb I can pass along.

Know Your Physics

Overall, my best advice for choosing between turbulence models is to try to match up the physics of your particular application with the models available in your solver. By this I mean, you use your engineering judgment to predict what flow features are likely to be present in your application, and which ones are likely to have the most impact on the information you are seeking to obtain from the CFD. Then, examine the models which are available in your solver and the different options available for them.

A General Rule of Thumb

If you're really at a loss, however, or your case is just too geometrically complex to use a model that has been optimized for one specific situation, the SST model or the Spalart-Allmaras models are usually fairly safe choices. They are generally robust and provide reasonable accuracy over a wide variety of configurations.

Personally, I would incline to the SST model, but, especially for external flows, the Spalart model often does a very good job. If your case is heavily influenced by free shear layers, then you may want to use a general purpose k-epsilon model instead of the SST model.

Take Advantage of Specialized Models

Many turbulence models, especially two-equation models, have been optimized for a particular class of flow. With all this specialization, it is hard to make blanket statements about these models. As a rule of thumb, however, a general purpose k-epsilon model will often do somewhat better than an SST model at capturing free shear layers, but not as well at capturing near wall behavior. A k-epsilon model is also frequently less stable than an SST model (or Spalart), making it a bit more difficult to use.

If your particular code has a model which has tweaked specifically for your class of problems, however, it may well give better results than anything else, notwithstanding any rules of thumb to the contrary. At Lockheed-Martin, for example, they have tailored a k-l model especially to predict separated flows over fighter wings, and on a good grid, they get remarkably accurate results (according to a recent presentation I attended).

Similarly, there are k-epsilon models which have been tailored specifically for planar shear layers. Other models have been tuned for round jets. Still others have been adapted to better cope with the recirculation regions around backward-facing steps. If your case fits one of these categories and you have a tailored model handy, it makes sense to use it.

On the other hand, do not use a model which has been optimized for a specific class of flows outside of that class. There are general purpose models and there are specialized models. If you try to use a specialized model like a general purpose model, you are asking for trouble.

Additional Reading

You can continue learning about turbulence models on the overview of turbulence models page or take a look at how these models worked for an example case on the model comparison page.

You can also leave the turbulence model tips page and look over the other tips and tricks.

Alternatively, you can head to the Innovative CFD home page to browse among the other topics.


Have you got a favorite model that I didn't mention? Got a horror story (or a success) you want to share? Do you have a question about turbulence modeling that wasn't answered? Send me a message I'm always looking for new ideas to write about for the site, so this is your chance to direct its development.

New! Comments

Have your say about what you just read! Leave a comment in the box below.