CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)

 Bin Li February 15, 2004 09:44

Dear all

Some questions about LES confused me. 1. Generally, in les code the codes for filter is not represented, is that right? If yes, then where the defferent effect of different filter,for example the Gauss filter and the box filter, are incarnated in LES? In the filter width?

2. In order to validate my LES code, the flow around cylinder and the flow in square duct were evaluated. Both results are not exact. Did somebody validate his code? which example?

3. For compressible flow, the Favre average was used in les equations. In code, do i have to add the Favre average effect? that is to say, i have to make much codes?

4. About the LES equations. LES equations were educed by using certain filter to Navier Stokes equations. So the equations is the equations about filtered density, filtered velocity, filtered pressure, that is to say, about resolved quantity, denote: f_bar. After the discrete LES equations was solved by iterating method ONCE, are the quantities(density, velocity and pressure) resoved(large scales) ones or instantaneous(full scales, denote: f_ins). The unresolved quantity is defined as followed, f_flu = f_ins - f_bar. If I wanna compute the turbulent intensity, can i use f_flu? that is, E_int = 0.5*U_flu*U_flu (one dimension)?

5. In compressible LES equations, if the Favre average is not used, we have to use model in continuous equation. Anybody did research this aspect?

6. In real computation, how to choose the number of the grid points ? as more as my computer capability?

best regards

 Lionel Larcheveque February 16, 2004 12:49

Hi,

1) One generally considers the filter effect only through its width. However there are some paper dealing with the compatibility between subgrid models and filter types (see for instance http://les.colorado.edu/~vasilyev/Pu...ns/PF2002a.pdf )

2) To validate the code I used, I performed computations of cavity flow based on Forestier, Jacquin and Geffroy configuration (JFM 475, pp 101-145, 2003). It's a massively separated flow at quite high Reynolds number (Re_L=8.10^5) with a highly detailed database (pressure spectra, velocity and phase-averaged velocity measurements) and configuration information. Try to ask the corresponding author for the datafiles. Nonetheless note that this test case requires using either wall functions (alternatively coarse grid near the wall) or a huge super-computer.

3) Favre filtering is a convenient way to obtain compressible filtered N-S equations with a LHS that closely matches the one from standard compressible N-S equations. Simply compute the new RHS terms and consider that the code solves for favre filtered variable rather than conservative variables (i.e. <rho>, <rho*u_i>/<rho>,<rho*E>/<rho> rather than <rho>, <rho*u_i>,<rho*E>, with <.> a filtered variable)

4) I'm rather confused with the sentence "After the discrete LES equations was solved by iterating method ONCE". LES is an intrinsic unsteady method, so you have to compute many timesteps, and average results over time to obtain turbulent intensity. At each timestep, the quantities you're solving for are filtered quantities. If your computation is a true LES, most of the turbulent kinetic energy comes from resolved scales, so turbulent intensity can be approximated by k=0.5*{time_mean[<U_ins_i^2>-<U_ins_i>^2]*time_mean[<U_ins_i^2>-<U_ins_i>^2]}

5) Don't know, but Favre filtered subgrid terms are still difficult to modelize and not to use this approach results in numerous new ones.

6) It's really problem-dependant. My advice is to perform two or three computations you can easily afford with various cell sizes, and then look at classical indicators (for instance mesh resolution near wall in wall units, inertial subrange in the TKE spectra in turbulent free flow regions, ...) to estimate if your case requires finer meshes.

Best regards

 Bin Li February 20, 2004 10:58