Hello Everyone:

I am trying to validate a Low Reynolds number version of the Turbulence model. As you all know, the mesh close to the wall becomes finer as the Reynolds number increases. I need the first computational node at y+=1. For accuracy reasons, I would need about 5 nodes within y+ = 5.

Can you please help me in getting experimental data where the Reynolds numbers are less than about 20,000? (lower the better) That way, I need not create meshes so large that it becomes computationally too expensive.

All your help is appreciated,

Thomas

Since there are so many low Reynolds number two-equation k-epsilon turbulence models developed in the last 30 years, it would be easier to compare the results with other models. At least, for the zero pressure gradient case, every model should produce nearly the same results. The skin friction distribution on a flat plate can be used as a reference case.

Hi Chen,

My ultimate goal is to look for non-linear Low Reynolds number Turbulence models. I would need cases where the non-linear effects like the secondary flows are dominant.

Thanks, Thomas

Dear Thomas The question is why do you want to avoide the fine grid solution (even in the beginning). Your ultimate goal is to apply the two equation model to fully turbulent flows. Another question is which model are you using? If you are using Jones and Launder model and if you are using an implicit scheme, the trouble lies in the numerical treatment of additional terms in the epsilon equation (as compared to high Reynolds number model). If you treat only these terms explicitly, you can overcome many troubles. If you are using an explicit scheme, there should be no problem, if everything else is in order. Please do compare your model results with the experimental or DNS data only. Comparing the model resluts with those of other models is useless. For example, it is very well known today that in the near wall region k-omega model is superior to k-epsilon and in the outer layer vice versa. What are you supposed to get from the comparison of two model results, how would you say which one is better?

Hi Ahmad,

I am using the Low Reynolds number Turbulence model developed by Dr. Shih of NASA Glenn (formerly called NASA Lewis). Good thing about this model is that it satisfies certain realizability conditions (which basically makes sure that the solution is physical like the normal turbulent stresses cannot be negative). The scheme is implicit.

I am working on the non-linear version of the above model to capture turbulent anisotropy. This creates Prandtl's 2nd kind of secondary flows. An accurate prediction of these flows would help in a better prediction of thermal characteristics especially in a turbine blade.

I agree with you 100% that the comparison with other turbulence models does not make sense. I need to do that with experimental data or DNS.

Low Reynolds number Turbulence models predict the near wall physics better. The downside of these models is that they are computationally too expensive. When the Reynolds number increases, the distance of the first computational node where y+=1 becomes very small. Even though one may not need too many nodes in the flow direction for accuracy, one will be forced to increase the number of nodes in that direction to reduce the aspect ratio. So, one would end up with a huge mesh especially if the flow length is too long.

So, I am looking for cases where the Reynolds number is not too large and the secondary flow physics is dominant.

Regarding the query, how I would conclude which Turbulence model is better, definitely the one which matches the experimental data better especially in the regions where the secondary flows are dominant.

Best Regards, Thomas

(1). ask NASA to run some tests for you, so you can do your low Re turbulence modeling, (2). create your model first and check against available heat transfer or skin friction data.(3). if NASA's model is physically correct, then why do you have to re-model it? So, a model is a model, do whatever you like. If you think the experimental data is more correct than the CFD results, then perhaps we should all give up the CFD and starting doing experimental fluid dynamics ( it's not a bad idea at all). (4). the secondary flow is basically 3-D, so, it will be hard to obtain the mesh independent CFD solutions. But, you can still create a good model for future use or verification.

Dear Thomas I have been working on the comparison of low Reynolds number k-epsilon models with a special reference to their application in coastal bottom boundary layers. Recently I have slightly changed my direction towards sediment transport. So my field of interest is basically hydraulic and coastal engineering. I have compared so far a number of models proposed till 1992. All of the present day models claim various benefits, but my experience shows that it is not true always. Sometimes, I found some typographical mistakes in the proposed equations published in AIAA journal, can you believe that? How could I know that the model equations those I am using have some difference of positive or negative sign? I came to know it just by chance when I was just going through the later journal issues that I saw an errata published in there regarding that model. I do not want to disclose the name of the researcher who proposed it, because he is someone well known regarding the keyword Realizability. Well! My suggestion is to consider original model by Jones and Launder also despite its inaccurate limiting behavior near the wall, it is good in many cases. As far as the computational economy is concerned, if you are not using exponential grid spacing please try it. Please do not expect too much from the modern models, because you may be disappointed in the end. For the data why don't you ask someone in NASA like F.R. Menter of NASA Ames, he has proposed low Reynolds number models also and they are being used extensively.

Hi Ahmad, Thanks for your feedbacks and comments.

Is F.R. Menter still at NASA, Ames? I tried getting his information through the website of NASA, Ames but was unsuccessful.

Thanks,

Thomas