CFD Online Discussion Forums

CFD Online Discussion Forums (
-   Main CFD Forum (
-   -   k-e Turbulence model & Separation (

James Date July 3, 2000 09:33

k-e Turbulence model & Separation
Hi there,

I'm modelling the flow about a NACA0012 aerofoil with a RANS code using the standard k-e turbulence model. I'm finding that separation and hence stall are failing to be predicted correctly, as found by many other researchers. Can anyone give me a reference which details why this is the case and ways of improving the result, i.e. by changing the equation coefficients.

Thanks in advance.

James Date

John C. Chien July 3, 2000 09:48

Re: k-e Turbulence model & Separation
(1).Standard k-epsilon model with wall function treatment is not a good choice for separated flow, even though in some cases you can obtainsome solutions. (2). For this reason, Low Reynolds number version has long been developed to take care of this problem, by solving the whole flow field including the near wall region. (3). The suggestion is : try to use a low Reynolds number version of the model. Or you can also try a two-layer model, which also covers the whole flow field. (4). Make sure that Y+=<1. rule is observed in creating the mesh, because fine mesh is required for the wall region.

Dave Hunt July 3, 2000 11:20

Re: k-e Turbulence model & Separation

predicting stall angle is v. difficult. I saw a few papers at last years AIAA Norfolk conference trying to do this and failing.

My understanding is this; Turbulence models are not up to the job. However, certain models are worse than others. For this class of problems you would be better using k-w than k-e. Further the Menter SST model tends to do the best job of all linear 2 equation models.

Going to non-linear 2-eqn may help but most of the advantage is in the SST like Cmu terms. Going to 7/8 equation DRSMs may be good. I haven't seen results for stall prediction so I wouldn't like to guess at which DRSM models are most suited to this problem.

In terms of wall-functions, the evidence that I've seen is that although they are usually blamed for poor performance, the fault generally lies with the parent model. Going to a lo-Re k-e model may give worse results!

In summary, I would recommend the Menter SST k-w model and would expect it to give results as good as if not better than most models upto DRSM. It will not however, give you the correct answer. Other models may do better on your particular case by fluke.

Hope this is useful.

Dave Hunt

John C. Chien July 3, 2000 12:30

Re: k-e Turbulence model & Separation
(1). Yes,Going to a lo-Re k-e model may give worse results! That is exactly right. (2). But I think, by going this way, he will have the "hands-on" experience first, and be able to feel the difficulties involved. (3). By following what was printed on paper is perhaps the worst way to learn. (4). So, the right approach is still this: move on to the two-layer model, and a low Reynolds number of k-epsilon model first. In this way, it will eliminate the uncertainty associated with the use of the wall function. (5). Then systematically refine the two-layer model or the low Reynolds number model used, until you get the right answer. (6). This requires hands-on experience and know how, that is, I am assuming that you are capable of doing the modeling or model change. (7). There are "many" low Reynolds number k-epsilon models on the market, so you need to do a review first. Some models from Japan and Taiwan have shown to give good results. (8). The original low Reynolds number model is somewhat inconsistent due to epsilon equation. The k-w model may have advantage in this area, but no one can say that their models were developed for separated flows. (9). If you are a commercial code user type, then these suggestions are useless. You will be limited by the options available in the code. (10). The right model is the one which you can exercise the control to give you the right answer. (not the one others were able to get the right answer claimed in their paper)

All times are GMT -4. The time now is 01:20.