CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   ANSYS (http://www.cfd-online.com/Forums/ansys/)
-   -   SST / turbulence model (http://www.cfd-online.com/Forums/ansys/77780-sst-turbulence-model.html)

PCFD July 3, 2010 07:31

SST / turbulence model
 
Hi,

I have a few questions concerning the expert-parameters for the SST model.

A) First of all, can someone tell me what is the "BC TKI Factor" = 1000 and how does it impact the turbulence model. I cannot find any description of that parameter in the manual.

B) The blending function from the variable arg_1. I would like to know if the model-constant "500" in the second term (in the following) equation is within the expert parameters as well.
arg_1 = min\left(max\left(\cfrac{\sqrt{k}}{\beta'\omega y^2},\cfrac{500\nu}{y^2\omega}\right),\cfrac{4\rho k}{CD_{kw}\sigma_{\omega 2}y^2}\right)
And what is the name of the parameter?

C) My third question is quite general - I would like to know if my understanding of Buoyancy Turbulence is correct. Is this the turbulence due to free convection?


Thx.

Josh July 5, 2010 15:02

Hi -

a) As far as I know, the BC TKI factor is a dimensionless quantity that has to do with the turbulence intensity. It isn't mentioned in the original SST model formulation, nor have I ever seen it mentioned outside of CFX. I recall reading a paper (though I can't find it) that altered SST parameters to see their effect on solution accuracy, but the TKI factor wasn't one of them. I would leave it as is.

b) I doubt this is in the expert parameters. It is simply a function that prevents F1, the blending function, from going to zero within the viscous sublayer. In the original formulation, it isn't given a name. Why would you want to change it?

c) Your understanding is correct in my eyes!

PCFD July 6, 2010 02:00

Hi Josh,

first of all, thanks for your answer.

I would be quite interested to read the paper which you mentioned. It would be great if you remembered the name of it.

I do have a case where the SST seems to predict a totally different flow field compared to the k-omega and BSL-model. I doubt that it is correct.

Have you made expierences with the new curvature correction model in version 12.1? Especially with the CScale factor?

And do you have expierences with the reynolds stress models (e.g. BSL EARSM)? In my case the best predictions are made with the k-omega model but it is even too optimistic. ( I have to admit that my flow is quite complex = many vortices and very 3-dimensional....)

Regards.

Josh July 6, 2010 15:50

Hi PCFD -

I did some digging and found the paper - Drag Prediction of Engine-Airframe Interference Effects with CFX 5 (Langtry et al., 2004). It's available from AIAA. I was a little optimistic about the paper's contents. They only alter the a1 coefficient to make the model comparable to the Spalart-Allmaras model. As a1 is increased, the separation increased, if I recall correctly. Still, it's worth a read.

I am in the process of downloading 12.1, so I haven't used the new curvature correction options.

I have some experience with the Reynolds stress models, though not much. They are typically recommended for high-vorticity flows, so it seems reasonable to use them with your simulation. Most of my experience comes from the SST models and its derivatives, though.

PCFD July 6, 2010 17:51

Hi,

it is the other way around, the separation decreases when a1 is increased.

Could you give a few more details on your experiences with reynolds stress models?
a) what kind of flows do you investigate? Turbomachineries (compressors / turbines) ....?
b) what kind of interfaces were incorporated in your simulations (e.g. stage interfaces)???
c) which model would you call as very robust? which criteria do you use for your model selection?
d) do you have advices for initialization? my tests (using meshes with y+-values between 5-15) did all diverge.
e) can you briefly describe some differences which you saw when you used a RS-model compared to the SST?

Thx for your patience since I have so many questions.

Regards.

Josh July 6, 2010 18:17

Hello again -

I have very little experience with the RSMs, but will try to answer your questions.

a) Mostly, I study low Reynolds number flows over airfoils, though I have simulated a turbine passage. I used the RSMs briefly for the low Re flow over a NACA 0012 with varying angles of attack. I was more interested in the location and extension of separation than I was in the wake characterization, so I stuck with the transitional SST model mostly. The RSMs provided good wake visualization.

b) Not sure what you mean.

c) I assume you mean in reference to only the RSM turbulence models. I haven't used or studied them enough to make an opinion on that. I have studied the k-e, k-w, SA, and SST models thoroughly, however. In those studies, I have come across papers that achieved good results with the RSMs and EARSMs, particularly the Wallin EARSM (Yuan et al., 2006) in low Re flow. None of the RSM models are very robust - they are pretty complicated models, especially when compared to those based on the eddy viscosity. In terms of what I select for my simulations, I read a lot of literature in the area I'm studying and see what models showed success in capturing desired phenomena. In the case of low Re flow, I've seen that the k-e, k-w, SST, SA, and RSM models are capable of capturing the laminar separation bubble. I then read literature specific to those models and refine further. Finally, I choose a couple of the models and create an appropriate simulation. Right now, for example, I'm testing the standard SST model vs. the SST-SAS model for low Re flow. I'll see if the extra costs associated with the SST-SAS model are worth the increased accuracy.

d) Divergence could be caused by many things, e.g., improper flow physics, steadiness vs. unsteadiness, domain size, 2D vs. 3D... Also, a grid cannot be explained only in terms of y+. Spanwise and streamwise refinement are important, as are the type of mesh (unstructured, structured, O-grid, C-grid...), points within the boundary layer, quality of mesh (orthogonality of structured grid, aspect ratio...), etc. I'm unsure if the RSMs offer wall functions. If not, your y+ should be smaller than or equal to 1.

e) The RSM definitely captured phenomena in the wake I didn't see with the SST. The flow had stronger curvature with the RSM, for example. I didn't have experimental data, so it's difficult to tell if this was the RSM properly capturing the wake. The SST seemed to better capture the location and extent of separation. Once again, however, I have very little experience with the RSM models.

I recommend reading some papers on the RSMs, especially if you can find some pertaining to your particular field of work.

Good luck!

PCFD July 12, 2010 01:40

Curvature Correction
 
Hi,

do you have some papers in mind?

I have tried the BSL EARSM model - I could see that the separation in my turbomachinery is oscillating from one side to the other but it is still too early to make a decision whether it is going to work out.

Anyway, the BSL EARSM offers some case-sensitation parameters and it would be good to have some literature which describes the impact of these on the flow.

In my case, the k-omega model has the best convergence-properties. Nevertheless, the losses are extremely underpredicted even though the typical separation in the diffusing system is on the right side. To my mind the dissuion is overpredicted because I see that the mach numbers are too low. I am open for further proposals.

Has anyone expierences with the CC-model in version 12.1? In particular with the CScale-coefficient?

derz July 12, 2010 23:47

While I don't have much to add, I find this conversation enlightening.

I too am most happy with the SST model, but am very interested in developing my knowledge on these RSMs for the reason mentioned earlier - the more detailed wake characteristics.

Any information on your findings for the SST-SAS model would be appreciated.

Josh July 13, 2010 16:28

PCFD -

Try reading the one paper I mentioned (Yuan et al., 2006). Can't recall the title, but it involved low Re flow with several turbulence models. I recall at least one EARSM model providing excellent results.

derz -

Contact me in a few months, perhaps at the end of the summer. By then, my paper on the SST and SST-SAS models should be submitted and I can show you my findings.


All times are GMT -4. The time now is 01:48.