[saidc.]

Hi,

I'm trying to solve natural convection laminar flow of close-packed equal size 1000 of spheres in a cylinder with bouyantPimpleFoam.

Flow is not logically and numerically complicated. In this case, the geometry (png is attached) is complex and causes some restrictions. The overlapping assumption was made to avoid infinite volumes of mesh when generating the geometry. Unfortunately, even with this approach, it becomes difficult to produce a very good mesh even if the number of cells is excessively large.

Some issues about mesh that I wrote on another thread Post

checkMesh:

Mesh stats
    points:           6182008
    faces:            58780252
    internal faces:   52886104
    cells:            27,916,589
---------------------------------
Checking geometry...
    Overall domain bounding box (-0.0305 -0.0305 -0.0305) (0.0305 0.0305 0.0305)
    Mesh has 3 geometric (non-empty/wedge) directions (1 1 1)
    Mesh has 3 solution (non-empty) directions (1 1 1)
    Boundary openness (1.99962127762e-16 1.22694797709e-17 -9.87824964054e-17) OK.
    Max cell openness = 3.84341179621e-16 OK.
    Max aspect ratio = 52.5849496127 OK.
    Minimum face area = 7.93767293804e-13. Maximum face area = 1.98285487187e-07.  Face area magnitudes OK.
    Min volume = 3.171338e-19. Max volume = 2.1346491e-11.  Total volume = 7.147417e-05.  Cell volumes OK.
    Mesh non-orthogonality Max: 88.09473404 average: 17.4884805521
   *Number of severely non-orthogonal (> 70 degrees) faces: 5263.
    Non-orthogonality check OK.
  <<Writing 5263 non-orthogonal faces to set nonOrthoFaces
    Face pyramids OK.
    Max skewness = 3.2348321860 OK.
    Coupled point location match (average 0) OK.

Mesh OK.

End

Time step continuity errors is constantly increasing to really high numbers so probably there are problems with my boundary conditions besides Mesh.

IC:

T: 293.15K, U: (0 0 0), p: 1atm, p_rgh: 101593.94 Pa

BCs of cylinder:

T: Adiabatic (zeroGradient), U: noSlip
p: calculated (1 atm), p_rgh: fixedFluxPressure (101593.94 Pa)

BCs of spheres:

T: fixedValue (393.15K), U: noSlip
p: calculated (1 atm), p_rgh: fixedFluxPressure (101593.94 Pa)

The solution diverges even if I relax the schemes to the limits, using upwind schemes and increase the tolerances to 1e-4 for p_rgh.

I'm open to any suggestion.

Kind regards,
Said.

[JulioPieri]

Have you tried running on a Euler-Lagrange solver? Then you model your spheres as heavy material, impose a max packing alpha and let the flow adapt. Not sure if it's gonna work, but it can be a good test.

You can either use MPPICFoam (without energy equation), or you could implement kinematicCloudProperties on a compressible solver like rhoSimpleFoam. swak4Foam can do this for you.

To solve directly the mesh, I'd for first for a 2d axissymetric case to quickly generate the intricate mesh. When you have good snappyHexMesh parameters, jump to the 3d case

[saidc.]

Originally Posted by JulioPieri

Have you tried running on a Euler-Lagrange solver? Then you model your spheres as heavy material, impose a max packing alpha and let the flow adapt. Not sure if it's gonna work, but it can be a good test.

You can either use MPPICFoam (without energy equation), or you could implement kinematicCloudProperties on a compressible solver like rhoSimpleFoam. swak4Foam can do this for you.

To solve directly the mesh, I'd for first for a 2d axissymetric case to quickly generate the intricate mesh. When you have good snappyHexMesh parameters, jump to the 3d case

Hi Julio,

Thanks for your reply. I couldn't get the idea you mentioned. The solvers you said are used for kinematic particle transport as far as I know. But there is no particle movement in my case. Spheres are stand still, gas will flow around them. So I wonder how can I implement these solvers to my case and is it necessary? I don't any experience for these solvers.

If I misunderstood you, I would appreciate it if you could explain your idea a little more.

Kind regards,
Said.

[JulioPieri]

So, the solvers are indeed designed to study lagrangian particles. However, I think (I'm not sure) you could run it with heavy particles, so they don't move. If you couple the solution between Euler and Lagrange phases (i.e., the flow sees the particles and change its way to get around them) you could mimmic the effect you want.

It's just an idea, and very simple to implement. The lagrangian data is concentrated in a single file, called kinematicCloudProperties. There you create your clouds (set of particles) and specify things like maximum packing, quantity, and diameter.

Start from the MPPICFoam/Goldschmidt tutorial, run it take a look in paraview. To post process lagrangian data, the info will appear in the "Properties" tab in paraview below the mesh/patches stuff.

The idea is to create a cylinder, put the, say, 20 spheres you have, with the diamater they have. Then calculate the maximum packing (a simple volume ratio) and put that number in the file. The spheres will set on the floor randomly but keeping the maximum packing you specified. Then the flow will just pass them as if they were obstructions.

Again, this is just an idea. I've never tried something like that, but could be a test worth trying.