Hi all,

I'm a university student who now is doing a project about CFD. Can anyone tell me the differences between LES and DNS and introduce them briefly? Thank you for your contribution!

Sara

Hello Sara,

The so called Navier-Stokes Equations are reckoned to be the equations which govern the flow physics. It contains all the information about the laminar and the turbulent flows.

For accurately computing the turbulent flows, we need to resolve all the scales from the largest eddy to the smallest one present in the flow (called the Kolmogorov length scale).

If you resolve all the scales of the flows, it is called Direct Numerical Simulation (DNS). The downside of this approach is that it is computationally very expensive because the scales in a turbulent flow is widely varying, getting worser with the increasing Reynolds number.

Instead of resolving all the scales, sometimes what is done is that only the largest scales (which contain most of the information) are computed and the small scales are modelled. Such an approach is referred to as Large Eddy Simulation (LES). This approach is now getting importance with the increase in the computing power.

Thanks,

Thomas

One way to think of LES, is that it should produce the same results as DNS as you refine the grid. Other models like URANS do not necessarily make sense when you the mesh is too fine. With LES, everything smaller than your cell size (or filter size) is modelled (i.e. the NS equations are not solved for fluid characteristics smaller than your filter). The model can be "static" (e.g. Smagorinsky), and uses little or none of the information contained in the flow. Other LES models are dynamic and use the data contained in the larger scales. DNS: no model, everything is computed via the NS equations. The problem: DNS is limited to small Reynolds numbers, and many people make the mistake of comparing DNS to LES, at Reynolds numbers where LES is not supposed to work: LES relies on the separation of scales in the flow, and that assumption makes little sense at small Reynolds numbers... Good luck!

HI,

Could someone explain Unsteadny RANS to me....I can't understand how a time averaged system can be unsteady.

Use an average other than time averaging, such as the ensemble average.

You need to look closely on how the time averaging is done. The time scale used for the time averaging process is much larger compared to the time scales of turbulence (which are very small) but is much smaller compared to the time scales of the unsteadiness of the gross mean flow (which are very large).

The above averaging process allows the algorithm to capture the unsteadiness of the mean flow leaving the turbulence models to model the turbulence effects.

Thanks,

Thomas

Hi Thomas, could you elaborate on that last point for me. When would you want to solve the unsteady RANS ? If you were looking at pressures on the roof of say a stadium, would that then be a case for looking at the unsteady solution ?

Hello Dave,

There would be times when getting an unsteady solution is not possible. BasicallY, that is part of the physics. In such a situation, one has to solve unsteady RANS.

Thanks,

Thomas

Hi Thomas, is it a case that you can get a steady state answer, but it would just be very incorrect compared to say the RMS result from an unsteady problem ? Dave

Hello Dave,

I would think that it would be incorrect. If a problem is unsteady physically, a correct numerical method should reflect that reality.

If the numerical scheme ends up getting a steady state solution, there could be something wrong with the method. One probable reason is that excessive numerical viscosity could be smearing the unsteadiness, hence able to steady state solution even though the correct numerical approach should have given an unsteady solution.

Thanks,

Thomas

Thomas, This must be a problem when using comercial codes as you would be unsure as to what is going on behind the sceenes ? This must be an even bigger problem when undertaking say LES (or other unsteady) simulations using the comercial codes ?? I have very little LES expeience, just what I have read. What order of discritisation would you aim for when undertaking a LES simulation ?? or is that a silly question ?? Dave