Hi,

Can anybody give me some ideas on how to set up an LES simulation? I have used RANS now for 3 years on hybrid grids of up to 15 million cells. However, now I want to look at unsteady simulations.

How do you decide on what cell sizes to use and where? I am simulating vehicle aerodynamics so this is a huge problem for me. I have a computing cluster to carry out the CFD solves but establishing the grid size and time step size is the most tricky for me. Any advice/experience anybody can pass onto me?

If for instance, the characteristic length of my vehicle was 9m, then from duct flows the largest turnulence length scale woulf be about 7% of this = 0.66m. One question is, how many cells do you need to resolve this, 10, 100, 1000?

Also, do you need this grid density in the whole domain with no expansion in the node density? Is 15 million cells enough to carry out LES simulations? Obviously, the near wall grid needs to be much finer but I don't know where to start.

I would really appreciate some advice as I would love to be able to produce good simulations from the start. Hope to hear from somebody soon.

Regards,

Carlos.

Dear Carlos,

It is quite difficult to tell you how many cells you will need in order to sove your problem, specially that LES is very suitable to the wall dynamics, where it is the source of all turbulence in a over body simulations.

In FLUENT there is the Dynamic SGS which should give the best results, but it needs some wall discretizations very close. It is advisable that the cell height give you at least y+ = 1 in order to get enough resolution there. Also after that you should expand the grid to save computational efforts, but it is hard to say the ratio, 5% is usually taken. And it will depends on your simulation.

As I can see you are capable to simulate a huge case (15 million) so, try to get y+ = 1 and expand the grid out in 5% and compare with some experimental results, such turbulence spectrum in some point or average streamwise velocity.

Hope it helps

Best Regards

Lourival Mendes