[lakeat]

Dear all,

I can't get a good and consistent result from my simulations, so this makes me to think about in general the capability of turbulence modeling in more detail. Here are some big questions from an engineer's viewpoint:

Take a flow around square cylinder as an example.

1. To date, which model is the best RANS turbulence modeling? Best as it is so reliable, that you would go for it definitely if you need accuracy, OR best as it is reliable and also a good efficiency.

2. Does 2D-LES make any sense? I think mathematically, it is wrong, but I found it do can produce some not-too-bad results for turbulence comparing with 2D-(U)RANS. And also a better spectrum than the RANS even for 2D simulations.

3. So, in 2D simulations (mesh is 2D, I mean), Will you go 2D-LES, or 2D URANS?

4. The difficulty is this, I (in civil engineering) am not JUST interested in mean Cd, but also rms values of Cl and Strouhal number and even the high frequency signal as in wind tunnel, which I found it is so easy to get a smooth curve of lift coeff from the RANS and also a low RMS value of Cl.


Any experience and advice?


Many many thanks

[aerospaceman]

Hi,

Thats a very good question, without a definite answer.

Suggest you look at the "European Research Community on Flow, Turbulence and Combustion" database of validation results:

http://uriah.dedi.melbourne.co.uk/w/...ular_obstacles


Then there is also G. Iaccarinho et al. / Int. J. Heat and Fluid Flow, 24, (2003) 147-156 "Reynolds averaged simulation of unsteady separated flow".

They obtain good results from RANS using a v2-f model.

I understand your concern in obtaining the actual flow field rather than the statistics, which would point your away from RANS. The contributor to the first link is interested in environmental flows, which is why LES is the least in the computational science.

I don't think that 2D LES is worthwhile pursuing due to the highly three dimensional nature of turbulence, especially the flow around these objects.

Thats my opinion, anyway.

[lakeat]

Thanks for your reply. I noticed that recently, the LRN turb models are becoming popular.

1. I wonder is there any better turb models than the v2f so far?
2. If LRN turb moel prooves to be so successful, why do people still use the HRN model, do you have any experience that HRN turb model is 100% percent better than the LRN turb modeling?

[aerospaceman]

Hi Daniel,

Turbulence modelling is a huge field. I understand that you are looking for a simple answer, that 'X' turbulence model is the one for you.

At least from my experience, it is very difficult to answer that question. No turbulence model will ever work unless you thoroughly scrutinize it and understand the underlying assumptions and limitations of it.

Low Reynolds Number models, High Reynolds Number models, linear eddy viscosity models, non-linear eddy viscosity models, models not employing the RANS method, DES, LES, etc are all out there.

The way I would go about your problem would be to find a paper that has a similar geometry and starting out from there. I've quoted two good references, and there are plenty more out there.

Maybe someone more into environmental flows may provide a more specific paper.

Additionally, turbulence models are very sensitive to boundary conditions, so hopefully the paper you start out from specifies their boundary conditions.

Hope this helps

[lakeat]

Thank you very much, aerospaceman.

I will find some good quality papers to start.

[Rami]

The ultimate solution, of course, is using DNS, which avoids turbulence model altogether... This is however rarely practical. Check if your problem can be analyzed by DNS (you did not mention your Re No.). If not - find the best experience of others on a problem as near as possible to yours.

[heidar]

Originally Posted by lakeat

Thanks for your reply. I noticed that recently, the LRN turb models are becoming popular.

1. I wonder is there any better turb models than the v2f so far?
2. If LRN turb moel prooves to be so successful, why do people still use the HRN model, do you have any experience that HRN turb model is 100% percent better than the LRN turb modeling?

yes i have, i have a paper on a journal and observed that SST as a low reynolds model is weak in reynolds number around 7000 to 20000 in comparison with HRN in my case of study

[heidar]

Hello
I have a good exprience on Rans models and i suggest the RSM as the accurate and the RNG as the more optimized model

[heidar]

Originally Posted by aerospaceman

Hi Daniel,

Turbulence modelling is a huge field. I understand that you are looking for a simple answer, that 'X' turbulence model is the one for you.

At least from my experience, it is very difficult to answer that question. No turbulence model will ever work unless you thoroughly scrutinize it and understand the underlying assumptions and limitations of it.

Low Reynolds Number models, High Reynolds Number models, linear eddy viscosity models, non-linear eddy viscosity models, models not employing the RANS method, DES, LES, etc are all out there.

The way I would go about your problem would be to find a paper that has a similar geometry and starting out from there. I've quoted two good references, and there are plenty more out there.

Maybe someone more into environmental flows may provide a more specific paper.

Additionally, turbulence models are very sensitive to boundary conditions, so hopefully the paper you start out from specifies their boundary conditions.

Hope this helps

yes you are true but Genrally we can discuss and recommend one model but it is dependant on respective cases.

[lakeat]

Originally Posted by heidar

Hello
I have a good exprience on Rans models and i suggest the RSM as the accurate and the RNG as the more optimized model

1. Using RSM on a LRN mesh mode or HRN mesh mode?
2. Using RNG on a LRN mesh mode or HRN mesh mode?
3. Is RSM better than v2f in accuracy?
4. Is RNG better than v2f in mesh efficiency and time efficiency?
5. You mean bluff bodies or stramlined bodies?
6. Concerning accuracy, besides drags, how about RMS(Cl) and St number?

[heidar]

Originally Posted by lakeat

1. Using RSM on a LRN mesh mode or HRN mesh mode?
2. Using RNG on a LRN mesh mode or HRN mesh mode?
3. Is RSM better than v2f in accuracy?
4. Is RNG better than v2f in mesh efficiency and time efficiency?
5. You mean bluff bodies or stramlined bodies?
6. Concerning accuracy, besides drags, how about RMS(Cl) and St number?

1- i recommended that in equal conditions
2- as same as 1
3- RSM is better
4- no V2f is more better
5- as same as 1
6- you must rely on 5 equation model (in 2D) in comparison with 2 equation models (i your only concern is accuracy)

[adrin]

You didn't mention what Reynolds number you're interested in. Transitional, low, and high Re flows require different modeling capabilities. Also, I don't know whether you intend to compare your results with experiments, but I'd be very careful - 2D turbulence is a misnomer and is quite a bit more energetic than (realistic) 3D. So, you end up having much more coherent vortical structures (nicely observed eddies) in 2D than you would otherwise (there is no vorticity stretching mechanism in 2D and the energy cascade is different). Finally, running unsteady simulations, you can get the primary Strouhal number with good success, but high frequency signals will be impossible to obtain _unless_ your computational timestep size is small enough to resolve the flow field at that frequency level. So, even if we assume that your time integration is infinitely accurate, your timestep size should be in the order of 1/f (f: frequency of your interest) to capture the physics correctly. That's just physics even if you were to perform DNS. However, for increasingly smaller timestep sizes, I'm not sure whether traditional turbulence models will even be valid (they are based on the assumption that we're looking at long enough of time period for time averaging of turbulence statistics to be valid)

Adrin

[lakeat]

First, thank you all.

I am interested in flow around bluff bodies. More specifically rectangular prisms, I found they are still very challenging.

As concerning the Re, I did not mention because, it ranges from 1E1~1E7. Incompressible flow.

-> A sidenote: I found St is not that easy to get in LRN modeling, I have been testing them for a month. I am not sure if they could be more easily got from HRN modeling+wall functions. Sorry, my experience is very limited.

I am very interested to study and capture the Strouhal number changing with the Re.

But along this way, I found myself lost in the various turbulence modeling. I know some of them prove to be powerful in some cases, in some Re(s) and in some geo-configurations. But my luxury hope is that there is certain (2D URANS) turb model that has GOOD consistency and numerical stability for a various Re(s).

Thank you again.

[heidar]

Dear all colleague
In calculation the transmitting flow through bluff bodies you must consider the more important thing that is vortex shedding and its fluctuating effects (Frequencies) on flow behavior. As ARDIN said in a turbulent flow analysis having experimental data for new cases is vital and it could conduct the user to achieve the best numerical method via confirmation with the experimental method. For example at first you must find a mesh density which the solution through it is independent to mesh (MESH DEPENCY); afterwards, you should find the best available wall treatment; ultimately, you should make a comparison between your available models to detect your best efficient approach.
It is how a turbulent researcher conducts its solution, in turbulent each case with its different dimensions, fluid and flow characteristics has its unique approaching route.
In some cases 2D simulation may get the efficient solution, for the 2D nature of the problem. For instances the flow structures at the center of a straight duct is far from the effects of lateral walls and the researchers only has the up and down effects of walls.
Best regards
Jafarizadeh

[Rami]

A scientist once said he is puzzled by two great question:

1. Is there God?
2. What is the best way to model a turbulence flow?

His conclusion was that at least when he dies he'll know the answer to one of them.