Hi, friends,

two questions to ask you

1.) for simulation of unsteady turbulent flow, the grid sizes used in LES should be larger than that in using RANS?

2.)

in order to resolute the turbulence structure( e.g., near wall region), the requirement on the minimum grid space is basically the same? (i mean they should have the same order of magnitude of the minimum grid space),right?

thanks in advance

Hello,

While using the RANS approach to account for turbulence, it all depends upon whether you are using wall functions or not. If you are using wall functions, the first node from the wall should be in the log region of the boundary layer, between y+=30 and 200. If you are not using wall functions, the first node from the wall should lie well within the boundary layer, about y+=1.

If you are using LES to account for turbulence, the mesh needs to be fine enough so that only the isotropic scales need to be modeled.

It so happens that the scales that are left to be modeled using LES are really small. So, the small size of the cell you would have while using LES would be much smaller than if you were to use the RANS approach.

Thanks,

Thomas

Absolutely right. The grid size of LES should be small enough. But it is not easy to get the exact size of the smallest cells. Some people use y+, other people use Kolmogorov scales,etc. As you know, the basic assumption is that the flow with SGS is isotropic in the case of using Smagorinsky model.

Based on my experience, not only the size of smallest grids but also there are many other facts to be considered in LES. Computational domain size, cell cluster, boundary conditions, wall treatment, etc.

So if you want to do LES, some compromise would be necessary. At first, please use reasonable size of computational domain, number of cells and run the problem in your computer. If the result is not accurate enough, then increase the resolution. But even though the resolution is not enough (in the LES viewpoint), you may get satisfactory results.

Good luck

Kim.

Thank Dr. Abraham and Dr. Kim for your excellent answer to my questions.

i want to further know what you mean by the your following words:

"But even though the resolution is not enough (in the LES viewpoint), you may get satisfactory results."

for your conviennet explaining it, let take an example,

for a oscillation problem involing turbulent flow, one get results( here refer to osicillation time history of pressure at a monitor point) by using RANS with fine grid (including considering minimum grid space in near wall region). another guy obtains results (again refer to osicillation time history of pressure at the monitor point) by using LES with coarse grid(e.g, only half grid size of RANS's grid. but don't know its minimum gird space in near wall region)

so which result of both is correct?

if their minimum grid spaces are basically same, what is your opinion?

thanks in advance

Ideally, while doing LES you want to have a mesh which resolves all the anisotropic scales leaving only the isotropic scales to be modeled. The modeling of the isotropic scales is similar to the RANS approach.

In reality, it is not possible to satisfy all the mesh resolution requiremnts all the time with LES. For the question whether a coarse mesh LES gives a good solution, it depends on the level of accuracy you are lookiing at. This is where your engineering intuition should provide you the answer.

Thanks,

Thomas

As Thomas mentioned, depends on the results, you have to decide your grid resolution. More important fact is what is your physical or real flow condition. So, considering simultaneously your problems and the results you want to get from the simulation, the numerical scheme should be decided.

I'm trying to compare my LES/DNS results with experimental data. In turbulent flow, SGS of the flow is modelled with standard Smagorinsky model. In laminar flow, rough/lower order (i.e. 2nd order) DNS is used. The flow parameters(Drag, Lift coefficients, Strouhal number, etc.) of LES are agree well with experiments at high Reynolds number.

In laminar flow, Re=150 ~ 250, my results show some discrepancy to other numerical simulations and experiment. Actually, it is not easy to say that my results are wrong. Because some groups show similar results of mine. One of the serious problem is experimental data is not credible (that is my viewpoint).

Conclusively,

depends on the flow characteristics, capacity of your computer available, your time-table, your results to see,

please decide your numerical schemes.

1. Computational domain size and Two/Three dimensional analysis

2. Inlet/Outlet/Wall boundary conditions

3. Grid resolution : Number of cells, cell clustering

4. Order of accuracy : 2nd or higher, upwind, ...

5. Turbulence model : two-equation, RSM, LES, etc.

Actually there is no exact answer in the grid resolution. Try !!