Hi, I use the large eddy simulation to simulate the channel flow first I use the smoginsky model the skin friction coefficient is 0.0085. second i use the dynamic model, it is 0.0073, I don't know, which problem is it. Kim and Moin with DNS got it 0.00818, who can help me?

Remember, in LES you need to model the unknown terms. Since models do not describe the physics of flow exactly, you should definitely ended up having different results depending on the model you are using.

Hi. It is clear that the Smagorinsky model is too dissipative. In particular, it will be very inappropriate in areas where the levels of turbulence are low, and in areas of transition. The Germano dynamic model should give you a better evaluation of the subgrid-scale energy cascade. However, I think in your case the big problem might be the wall model. In LES this is one of the trickiest parts. You may want to get in touch with one of the French experts in LES : Dr. Pierre Sagaut (sagaut@onera.fr).

Hi,

At first, don't forget that in your case you 've found using the dynamic model a skin friction coefficient within ~10 % of the exact solution ... and whitout fitting any magical parameter. I am sure you have used a Smagorinsky constant which is something like an optimal value for the channel flow. This value has been found after a lot of trials (Deardoff, Moin, ...). If you use this model without any changes to simulate an other flow you will often get bad results.

Secondly before trying to answer your question, can you tell me if you use wall damping function with your Smagorinsky model and if the other results (mean velocity, two point correlations, energy spectrum ...) are also worse with the dynamic model than with the Smagorinsky one.

Gary

I think you are probably right to suspect that something needs attention (unless the Reynolds number is very low). There are several places to start looking:

- how accurately are simulating the convection terms (are you using a general purpose CFD code or an LES code where effort has been spent to accurately simulate the convection process)

- how are you damping the sub-grid eddy viscosity in the Smagorinsky case

- where is the first grid point (y+<1?)

- are you expanding the grid spacing

- what model are you using to parameterise the sub-grid stresses for the "dynamic" model?

In general (ie not always) the "dynamic" model is going to misbehave more than the Smagorinsky model if the resolution is less than sufficient.