[kandelabr]

I tried to run a DNS case (i managed to get cell size small enough and geometry is fairly simple) only to find out dnsFoam only solves 'box' geometry.

I guess there's a reason for this?

[piu58]

dnsFoam uses a spectral method for solving the case. This requires a rectangular mesh, all cell sizes equal and the cell count a power of 2.

You may use paraFoam instead and make the spatial and time resolution as small as the Kolmogorov scale requires. The dissipation rate eps disappears in this case (= gets very small).

[kandelabr]

Oh I see.
Thanks for the tip - but are you sure you meant paraFoam?

Could that be the reason that simpleFoam with k-Epsilon model blows up with very small cell sizes? I'm not really sure what's going on because everything was fine with triangular mesh with slightly bigger cells (just above Kolmogorov scale).
Now I used blockMesh with sizes around 0.1x Kolmogorov scale and strange anomalies in pressure and velocity occur in one spot.

[piu58]

simple Foam is not suited for transient cases. icoFoma doesn't have turbulence at all. pimpleFoam makes large time steps and corrects afterwards.

From the standard incompressible solvers only pisoFoam remains.

[kandelabr]

pisoFoam it is then.
I guess I can't use simpleFoam even though it is a steady-state case (a gap in plain bearings)?

Thanks for your help.

[ShantanuSG]

Originally Posted by piu58

dnsFoam uses a spectral method for solving the case. This requires a rectangular mesh, all cell sizes equal and the cell count a power of 2.

You may use paraFoam instead and make the spatial and time resolution as small as the Kolmogorov scale requires. The dissipation rate eps disappears in this case (= gets very small).

Greetings Sir,

How do you find that it is a spectral code? or that it uses spectral method?
I also want to know what is time discretization scheme in this solver?

[RobertHB]

Originally Posted by ShantanuSG

I also want to know what is time discretization scheme in this solver?

The time discretization schemes are chosen in the fvOptions dictionary. An overview can be found here.

[Santiago]

Originally Posted by piu58

dnsFoam uses a spectral method for solving the case. This requires a rectangular mesh, all cell sizes equal and the cell count a power of 2.

The statement is partially true, in the sense that the second phrase is true: an inverse FFT is done, so the properties on the mesh described by @piu58 are needed. On the other hand, dnsFoam IS NOT A SPECTRAL CODE. Is just a variant of icoFoam. Still 2nd order FVM, using PISO for integrating the flow field.

The requirements on the mesh come from the Kmesh class, that is used to generate the K field, which is used to generate a pseudo-random force.