Hello everyone. I am using ANSYS CFX to do the simulations of the air flow in a simplified combustor geometry. I know that RSM is good for this simulation, but as you know, in ANSYS CFX, there are several types of RSM, like, LES, BSL, SSG. Could anyone tell me which one is most suitable for my case, i.e, which one is most accurate to get the results similar with the experimental results. Thanks very much for your help.

Eric

Can anyone help me with this, please?

Hi Eric,

The RSM models you mentioned (RSM-SSG and RSM-LRR) differ in the closure approximations used in the reynolds stress transport equations. LRR uses a linear relation for the pressure-strain correlation, which is only valid for small turbulence anisotropy. SSG uses a non-linear relation and is more accurate (especially for higher anisotropy). Therefore RSM-SSG should be your standard choice when considerin a RSM-model.

LES is a different story altogether as it is not a RANS-method (reynolds averaged Navier-Stokes), but a filtering of the NS equations. Basically the eddies down to the size of the grid spacing are resolved directly, while smaller eddies are modeled. Employing this method requires an approach specific to LES (finer grids, smaller timesteps compared to RANS).

Hope this helps.

I'd like to add that RSM-BSL might offer some advantages when flow separation is of major importance. RSM-BSL is an SST-like version of the stress-omega RSM model (omega-equation based as opposed to epsilon-equation based as the other models are), which is better suited to wall-treatment.

As turbulence anisotropy is probably of more importance in a combustor geometry, I think SSG should still be your first pick, but you can always play around a bit to see the impact of the specific models.

Hi Timon. Thanks very much for your kind reply.

I tried RSM SSG model, but the results do not converge, while if I use K-epsilon model, the results will become converged. As k-epsilon is not very accurate and not good for the large recirculation areas prediciton, I cannot use k-epsilon for my project.

Can you help with me with this?

Could you please give me your email address so that I could you write something more detaily and send you some files I have done.

Many thanks Eric

Hi Eric,

Convergence is always a bit more troublesome with RSM, as it is far more sensitive to the grid than 2-equation models. A few tips:

- make sure you have a very good grid (good quality elements) - restart from a converged SST solution - use local timestepping (gradually increase the timescale factor, from 1 up to about 5 more or less) - use first order discretization first and switch to second order when residuals become more smooth.