Dear cfd members,

my problem is the following. I am using the modified FEATFLOW program, where the standard k-epsilon turbulence equation is implemented with standard wall functions. Well, for the backward facing step at Re 44 000 it reproduced data, which are very nicely correlating with the profiles found (experimentally/numerically) for the problem. But. When I am trying to test the program for the channel flow problem for developed turbulence, let us say for Re_tau=180/Re=2800 the results are very strange, because the curving of the velocity profile shows quite a different tendency, then the experimental was found. I have started to look after some computed data by K-eps, but I have found no results. Ok, for the DNS, there is a great number of publications for this problem, but what is with the K-eps? Is it possible to compute with stisfactory precision the profile with this two equation model? By the way the y+ caculated by the wall function was about 15 for the obtained developed turbulence results.

I would appreciate any response ;-)....

Assuming your "standard wall function" is an empirical relation obtained from flow over either a flat plate or channel flow, and most turbulence models are tuned for such flows, it seems odd that you have trouble reproducing that kind of flow. Regardless of the turbulence model, and regardless of the accuracy of the shear stress or dimensional velocity, you should at least see that the non-dimensional u+(y+) profile is correct. At the very least, the first point off the wall must be exactly on the wall function that you used. Could the pressure gradient in the channel be an issue? Maybe you should try a flat plate first (zero pressure gradient).

Dear Mani,

the problem is not with the point at the boundary of the computational domain, that is calculated almost precisely, but the rest of the profile tends to be more laminar then turbulent. Hard to explain without pictures. The predicted profile compared to the one I am calculated has steeper tendency in the vicinity of the wall and flatter in the center of the channel.

Well, the pressure, the pressure for the developed flow in the normal direction to the streamline is almost flat. And, as far as I know, it should be flat, so that is almost correct. My question was rather that kind, if the k-epsilon model is really able to compute the velocity profile accurately? Can somebody send me any link or paper, containing results calculated by k-eps in channel flow?

thanks,

Otto

I didn't mean the pressure gradient in normal direction, but in streamwise direction. Try a flat plate with zero streamwise pressure gradient. That's usually the benchmark case for turbulent boundary layers, and I am sure you can find an abundance of reports for any turbulence model, including k-epsilon. Having validated that case you can proceed to a slightly more difficult case of channel flow, where you need to apply a streamwise pressure gradient. In that case, your wall function should take account of the streamwise pressure gradient. I would also try to run the turbulence model by fully resolving the boundary layer, not using the wall function. If your implementation of the model is correct, this will most definitely work. If it does, there must be something wrong with your wall function approach.