Hi everyone!
What would be a reasonable value of YPLUS when using Reynolds Stress Model for turbulence? And how many cells do I need near these walls ? What about 10 cells near the YPLUS region ?


Regards,
cfd guy

The wall model in cfx is a standard law of the wall for high reynoldsnumber (Only) Damn' two-layer models will be available in release ??.?? year 20??, thats for sure.

y+ for the near wall cell should be >> 11.4,

I think it is standard to use y+ > 30 and less than 100-200.

What do you mean with how many cells you need at the wall? You only need 1, where y+ >> 11.4. You are not going to resolve the liminar sublayer, so only one.

Hope it was a little help.

Jan

Thanks for answering me Jan.
Well, as I'm not interested to solve the laminar sublayer, a y+ in the range that you suggested, I guess it's fine to me.
But it confuses me when people say that you need very fine meshing near the walls when you work with RSM. (I'm working with strongly swirling flows)
In fact, I've noted that I could not get a converged solution with a very refined grid near the walls, so I started to check out my y+ values and I saw the most of them was lower than 11 (checking in the cfx-4 output file).
By the way, can I estimate y+ using the flat plate correlation for a cylindrical body ?


Regards,
cfd guy

(1). When you say "strongly swirl flows", you are dealing with difficult flow problems. (2). And for wall confined flows, you are going to get reverse flows, that is flow separations. (3). With flow separations, the wall function is not a valid approach. (although people have been using it for separated flows for a long time)

Hi,

If you are dealing with swirling flows and recirculation be careful about y+, because the velocity component parallel to the wall at the re-attachment point is going to be zero, which means that your y+ collapses in any case indpendently of the grid resolution (and the law-of-the-wall models revert to the laminar case). If you have too many regions of low y+ due to the above-mentioned effect at the wall, maybe you should consider the Low-Re k-e model available in CFX. Contrarily to what its name implies this is a model specifically designed to deal with the development of the B.L. I think.

Herve

I agree with Herve, the naming Low-Re ke is a bit confusing, it is also used for high Re number, it just depends on the range of y plus. Outside the boundary layer a standard k-epsilon is used.

And that might be the problem, because then you do not maintain the information about the anisotropic structure of the turbulence as with the RSM.

So if it is important to resolve the reynolds stresses the Low-Re k-e can not be used.

A wall model for RSM with y+ < 11 does not exist, but I hope the more is written about that problem, the faster it will be implemented. !!!!!!!

Jan

... (following up on Jan) which might mean you have to decide what to trade-off, and what is really important in your simulation.

However, if wall effects matter (inc. for the development of turbulence further away), and that the law of the wall is an issue it might be worth looking at the Low-Re k-e model or maybe the k-w (Wilocox, 1993) available in CFX-4, for a try. It will be cheaper that running the RSM anyway!

Bear in mind that the standard RSM included in CFX is also a simplified version of the 1975 LRR model and does not include terms such as the wall-reflection or pressure-echo in the pressure-strain correlation which are important at walls (the rapid part is also simplified), so depending on what you are looking at remember it is not the full RSM you are using either. Away from the wall a 1994 paper by Schwarz et al. (if I remember correctly!) shows it is sufficient as it is tough...

Hope this helps, Herve

Thanks for the information, I was not aware that the RSM model in CFX is a "simplified model". Do you know if it is standard, to use this simplified model , in universities and other commercial codes?

Is there any large difference in the result obtained depending on simplified or not for swirled flow.

Just a comment, It is so true, one has to consider what is important for the simulation, if one has the information/knowledge

Jan

Thanks John,
What do you mean that the wall function is not a valid approach in confined flows? What would be a valid one? LES ? Could you expand on it please?
Regards,
cfd guy

Thanks again Mr. Jan and thanks to you too Mr. Herve
The biggest problem using k-e, k-w models are due its isotropic behavior, which is not observed in RSM. As I said before, I'm not interested in solve the boundary layer, but is a big issue dealing with the mesh refining against y+ < 11.
We are dealing with cyclone flows, and we always wondered how the wall-reflection terms would improve (or not) our simulations. Why is so dificult to implement it in CFX? The manual brings that there's not a generic way to implement it, however two examples using a 2-D flow are available. I tried once to implement this terms in a 3D problem, but no success was reached.


Regards,
cfd guy

My understanding is that the wall reflection terms are available in the /extras/ directory in your CFX packages. However, being the LRR wall reflection it needs to calculate the normal to the walls at all nodes at the walls to be implemented, which is quite difficult to do, even more in 3D (this is not a CFX problem, but more a turbulence model problem). Therefore the extra code available assumes a particular geometry case I think to simplify theis calculation of the normal. As a matter of fact, the wall reflection term has been ignored for a long time because people didn't look into the B.L. and its akward modelling in the LRR model has lead to further research where the wall reflection has been accounted for in an extension of the so-called rapid part (SSG model, Sarkar, Speziale and Gatski, 1994?) but it was pointed out by the authors that there was still a lot of "blur" so as to how much "better" it was...

Coming back to the model in CFX, if you compare the formulation in the Mathematical Formulation Chapter (used to be Chapter 12) with the original paper of LLR (1975) you will notice that the rapid and wall reflection terms have been truncated. The formulation that is available however is a common one in industrial CFD which has been globally quite successful (see Schwarz et al., 1994) in all but a few flows configurations (Cokljat and Younis, 1995; Straight Two-Stage Channels).

Regards, Herve

P.S. Sorry to be so long in my answers; this topic was part of my PhD research not so long ago...

P.P.S. cfdguy, were are you based in Brazil?

Herve,
You don't need apologizes for being so long in your answers, the more I read more I learn.
I've just entered in the master graduation here at the University of Campinas (UNICAMP), Brazil.
So Herve, have you ever noticed someone who did some simulations envolving strongly swirling flows, like cyclones, using RSM with wall reflection terms ?
Regards,
cfd guy

Hi,

I'm not a cyclone specialist but I've got two papers on my desks at the moment:

Erdal, F.M., Shirazi, S.A., Mantilla, I., Shoham, O., (1998), "CFX Study of Bubble carry-Under in gas-Liquid Cylindrical Cyclone Separators", J. Soc. Petroleum Engineers Prod. & Facilities, Vol. 15(4), pp. 217-222

They used CFX with both k-e and RSM and decided to favor a simple k-e based on its efficienciency and the little effects t had on the overall flow compared to RSM

The second paper is an internal piece of work related to high order turbulence in cyclones; it says that by modifying the k-e constant Ce,2 based o the Richardson number some "curvature correction" could be achieved.

C'e,2 = Ce,2 (1+0.2Cu) with Cu = 0.5* (k/e)^2* (dui/dxj)* (duj/dxi)

which you can code in CFX. It also used the truncated RSM using german results (Obi, 1991) which showed that the wall reflection terms were not always shown an improvement of the overall solution. Regarding the wall treatment, they mention the Low-RE k-e model as a good alternative if you can afford the grid at the walls, or the k-e and RSM with ref. to Grotjans and Menter (1998; CFD'98, Vol. 1(2), ed. K.D. Papailiou, pp. 1112-1117, Chichester, ECCOMAS, John Wiley&Sons) for modified wall functions. This paper showed that higher order turbulence model was needed to capture the correct vortex-type (and pressure).

Therefore, (1) why don't you try the Low-Re k-e with the modified C'e,2 constant. That would be my suggestion based on the above two pieces of research to evaluate the importance of the wall treament while still somewhat accounting for curvature effects. In that fashion you would have a good wall representation, curvature effects would be accounted for in the Low-Re k-e model by modifying the constant, and the whole thing should be cheap to run. Then (2) look at the Grotjnas and Menter paper, use their wall function with the CFX RSM and compare for yourself. I think you can esaily modify the wall function in USRWTM.

Herve

P.S. No sou brasileiro mas tenho familia em Campinas. Meu cunhado esta estudando na PUC e conheco gentes que trabalham na UNICAMP tb em computacao. Abraco, Herve.

Herve, thank you very much!
I'll try your suggestions as soon as possible and I'll contact you by e-mail discusting the papers you have mentioned.
Regards,
cfd guy


P.S.: Confesso que fiquei surpreendido quando vi palavras suas em português, muita coincidência mesmo.