[nikos]

Hi everybody,
I'm trying to investigate airflow around a rectangular cylinder.
The choosen mesh is an hexaedral unstructured mesh appropriate for a low reynolds number modelling at wall. I've tested this case under turbfoam and everything is OK.
I've started the computation with smagorinsky approach and same parameters as those provided in pitzdaily case conserning the numerical approach.
A turbulentinlet is provided at inlet and pressure outlet at outlet.
Around the body, the airflow is developping fine, however I can see a pressure oscillation in time (from -0.2 to 0.2 in a p* formulation ) travelling from inlet to outlet where p* remains 0.
Solver is GAMG with GS smoothing.
CFL max is 2 and mean value is quasi 0.
I've tried to reduce the timestep but similar phenomena occurs.

Can it be due to delta coefficients, because my mesh is relatively coarse at inlet and I have an agressive refinement around the body ?

I've also tried to increase non orthogonal corrections but nothing has changed ?

What do you think of using GAMG as preconditionner for PCG approach because there is a large time reduction during time step ?

ps : i've applied modified backwarddtscheme provided and recompiled.

[eugene]

what do you mean a "turbulentInlet" is provided at the inlet?

CFL max should be 1. Anything larger can destabilise the solution.

Increasing non-orthogonal correction will make any problem worse. More stability is gained by decreasing non-orthogonal correction, i.e. Gauss linear limited 0.333;

[nikos]

Thank you for responding so quickly Eugene

I've setup a turbulentInlet condition with low velocity fluctuation scale (0.005) even it is not the best condition ... but for a sample test it's better than nothing

I've forgot to say that in the region where mesh is refined quickly, there are particularly elevated values for the Nusgs

When you say destabilize do you mean diverges ? because in this case there is no order changes on all variables, there are pretty good except that pressure around the cylinder is modulated by the global pressure oscillation travelling in the domain.

[eugene]

1. Skip the turbulentInlet, it is worse than nothing.

So you have pressure oscilations traveling from your inlet to your outlet messing up the pressure measurements as they pass?

I never use Smagorinsky so I cant help you there. Try SpalartAllmaras or oneEqEddy.

[nikos]

Ok , but I've tried without and set up a classical inlet ... same problem

Concerning oneq eddy, if I don't init my case with smagorinksy during few time steps computation won't start, but after that using solution coming from smagorinsky, I have the same behaviour with oneeqeddy

DES is doing the job on a coarse mesh.

I will try it changing the discretisation scheme as proposed

If I find where te problem comes from I will post a message

thank you Eugene

did someone use turbulence generator like works done in Chalmers ?

[ralph]

Hello Eugene,
where is the orthogonal correction used?
Just in the laplacianSchemes and the snGradSchemes, or anywhere else?

Ralph

[eugene]

Yes, the snGradScheme correction feeds into a lot of other stuff though.

[ralph]

Thanks