[Sereff]

Hi everyone,

Just a quick and easy check, If I were to run a periodic boundary flow simluation, I should implement the following bounary conditions:


Inlet
> velocity: cyclic; > pressure: cyclic; > nut: cyclic;
outlet
> velocity: cyclic; > pressure: cyclic; > nut: cyclic;

ground(wall)
for y+ < 5
> velocity: noSlip; > pressure: zeroGradient; > nut: zeroGradient;
for y+ > 30
> velocity: noSlip; > pressure: zeroGradient; > nut: nutUWallFunction;

upperAtm
> velocity: zeroGradient; > pressure: zeroGradient; > nut: zeroGradient;
sideAtm
> velocity: cyclic; > pressure: cyclic; > nut: cyclic;


Am I correct? With the BCs above I am not able to get the simulation to match the DNS data, by a bit margin. Any advise? Much appreciated


Kind regards,

[Ardali]

Is it a channel flow or free surface channel flow?
In general yes. It depends very much on your resolution.
I assume you talk about channel flow so check available tutotrail in OFm channel395.

[Sereff]

Hi Ardali,

Thank you for your reply. That was the case I based on and I validated against. I think the problem might lies in the fvScheme, where I implemented

> div(phi,U) LUST grad(U); instead of > div(phi,U) Gauss linear;

If you have any knowledge to share on the effect of different numerical scheme, it would be amazing.


Kind regards,

[Ardali]

That is a long story. LUST is a TVD scheme with conformal coefficient of 75% centeral differencing. If your case works with 'linear', I recommend to use it. Otherwise TVD with conformal coefficient of 0 is good too.



If you want to do DNS at moderately high Reynolds number, I recommend to put sometime to implement a higher order scheme. It helps you to have less expensive simulation.
Ardalan

[Sereff]

Hi again, I changed the numerical schemes, but it seems to be not helping.
I attached my plot of Rij (averaged from UPrime2Mean).

kind regards

[Ardali]

Is it only the resolved part? Please add the modelled part and check again.

[Sereff]

yes... so the modeled Rij can be found by
> #includeFunc R;

and the target one is:
Rij = UMean2Prime + RMean?

or is it,
Rij = UMean2Prime + RMean2Prime?

Sorry for my stupid question...

Kind regards,

[Ardali]

No, You have to take nutMean and calculate any term using Boussinesq.