[siw]

Hi,

I'm modelling an external aerodynamics body with some components that are a bluff shape, which is lead to in my post http://www.cfd-online.com/Forums/cfx...ge-models.html.

I don't have enough hardware resources or time for DES and LES simulations. Also I don't have any experimental or other numerical data for validation. So need to evaluate the sensitivities to get a some level of confidence in the results.

To get this simulation started I will use a steady-state SST-RANS simulation and although it's not going to be physically correct (I assuming that on engineering judgement but could be wrong and SST-RANS is enough) will give me an initial flowfield to start the unsteady simulations with. I then plan to use the SAS-SST model to better capture the turbulent structure near the bluff components.

With the SAS-SST would it be better to get a mesh independent solution first at a given time step size and then vary the time step size to see its effects? Since there's no experimental data I have no idea about turbulent scales and time step sizes in the volume of interest so may have to take a crude guess at those. I've not used SST-SAS before so have read up about it in the CFD guides and anything I can find here and elsewhere online.

Thanks

[ghorrocks]

Your approach sounds good.

Note that mesh independant solutions are very difficult to achieve for LES or any of the related models (DES, SAS etc). This is because the turbulent fluctuations modelled in these models will have subtle differences due to the mesh, and this will lead to bifurcations causing large differences. To get around this you should try to achieve mesh sensitivity on either an avergaed flow or a turbulence spectrum.

Alternately, you can do a validation case with high quality experimental/numerical/analytical answers such as the Ahmed body, turbulent decay. If your model is similar to the Ahmed body and you successfully model the Ahmed body against experimental results then you have a strong case to show your model is good on your geometry, even though you have no experiments to compare against.

[siw]

Thanks for the reply Glenn, as always your advice is hoped for.

In light of your text I think a sensible modification to my approach is to get a mesh insensitive solution on the steady SST-RANS solution - or at least 3 simulations of finer meshes even if solution independence is not achieved due to the flowfield being unsteady.

Assessing mesh sensitivity based on the turbulent specturm is new to me - I've always based it on avereged flow and some global integrated parameter such as lift and drag coefficients. I'll need to read up on doing similar with the turbulent spectrum - any leads to point me in the right direction?

Unfortunately the geometry is not similar to any common cases such at the Ahmed body or any other open literature case.

[ghorrocks]

You should be able to get mesh insensitive results for SST-RANS and that is definitely a good first step.

Do some reading into LES and turbulence theory - things like the -5/3 rule is a good one - if you can get model turbulence with an energy spectrum slope of -5/3 then you can be pretty sure you are on the money.

But in your case if lift, drag or some other similar parameter is important then just run it long enough so that the time averaged value is good, then comapre against different mesh densities. This will be much simpler than turbulence energy spectrums.

Unfortunately the geometry is not similar to any common cases such at the Ahmed body or any other open literature case.

Ideally the geometry is similar, but if that is not possible then go for something which is in a similar flow regime. Are you high or low Re and Mach numbers? The Ahmed body is intermediate Re and low Ma, scramjet work is high Re and Ma, and there will be other examples. Just choose something in the same Re and Ma number regimes.