Dear all,

Running a transient simulation of mixing in a T junction using the Oodles solver extended with a passive scalar, the velocity field shows checkerboard oscillations.

These oscillations occur in the two inlet ducts before the mixing zone. Downstream the mixing zone the velocity seems smooth (but of course turbulent).

I use the linear scheme for div(phi,U) and the other div schemes are set to limitedLinear. The mesh is built of hexahedrals, while the time step is such that Co_max = 1.4 or something. Co_mean is 0.2.

Regarding the extended oodles solver: I did not change any solution algoritm.

Anyone having a clue?

Many thanks,

Bram

[ville]

Hi,
I have seen same kind of checkerboard oscillations
when running implicit LES (LES with no sgs model) of a free jet that enters a chamber i.e. when using the "linear" scheme. This was at a situation where Co_max = 0.10. My understanding is that if
the cell Reynolds number ( U*dx/nu ) is smaller than two, oscillations will most probably not
develop. Otherwise they will probably be seen.
I guess this
would need some kind of a further comment...
I don't know if the available discretization schemes are
actually a limitation of doing e.g. DNS with OF.
Best regards,
Ville

Hej,

Regarding the cell reynolds number as mentioned by Ville: this would force me to use a very fine mesh, I think. But it depends on your definition of nu (subgrid viscosity or the fluid viscosity).

Nevertheless, it seems that spatial discretization is the main source of 'checkerboard' oscillations in the case I described. The filteredLinear scheme does not show the oscillations while the linear scheme does.

I found out however that keeping Co_max < 1 helps the solver anyway to keep stable for both schemes, despite the checkboarding of the latter.

Does someone has a clue regarding the order of accuracy of the filteredLinear scheme? From the header-file in the source it seems that it should be 2nd order for 80% of the solution.


Regards,

Bram

[ville]

Hi,
I don't know about the filteredLinear scheme but it would be very interesting to give it a try.
If the cell Reynolds number is required to be < 2 then applying "the rumor that oscillations will probably not develop if Re_cell < 2" supports the
use of a sgs model i.e. doing LES with the linear
scheme.This should effectively decrease the cell
reynolds number by increasing the effective viscosity. Considering e.g. normal channel flow, for me it seems that some people get "good resuls" with
no oscillations etc using the linear scheme whereas other people might see these oscillations.
Is the difference between these results in using
a proper sgs model (no oscillations?) vs
e.g. dns (oscillations?). Would anybody be interested commenting?

Regards,
Ville