[doerschlott]

Hey guys,

I came across an odd behaviour with hex-mesh and bent geometries, briefly explaining the process now:

I have a hot plate with two adjunct cooling channels (water) on the left- and right-hand side of the plate and two structural plates at the face of the plate (think of a double T-girder).
In CAD, I modelled the plate with an exact curve (therefore, also the cooling channels).
Afterwards this model with separate volume bodies was exported as a .STEP file and loaded into ICEM CFD, a ANSYS meshing tool.
Herewith a hex-mesh of the cooling channels and the plate was generated, with elongated cells into the direction of flow, due to a high aspect ratio.
Note that the radial cell size at the interface to solid parts is small to precisely model boundary layers.
The mesh looked good and 'checkMesh' didn't complain about the mesh, except some skewed faces at the outside of the solid, where actually no calculation happens.

The considered solver for that problem is ‘chtMultiRegionFoam’, from OpenFOAM v1806. In theory a compressible solver, but with the option of ‘rhoConstant’ in ‘thermophyiscalProperties’ the water should be calculated as incompressible.
Coming to the calculation, I always observe divergence of pressure (and following the temperature) at the interface between water and structural plates. The geometry is strongly bent at this position, so I had the concern that this could cause the problem, as reducing time steps didn’t help.
In order to confirm this assumption, I did some calculations with a straight plate.
Surprise: convergence.
Later I tried the simplest incompressible solver ‘icoFoam’ with the same meshes (only water obviously), to exclude the influence of a compressible solver on an incompressible fluid. Same results: straight plate did converge, the bent one did diverge.
However, a tet-mesh of the bent geometry runs smoothly and converges surprisingly fast.
Additionally, for ‘p_rgh’ I observed jumps from the final residual to the next initial residual of a nCorrector in the PIMPLE loop by two orders of magnitude. The used solver for ‘p_rgh’ was ‘GAMG’ with a ‘DIC’ smoother. Have a look inside the appended log file from the chtMultiRegionFoam calculation, just before it crashes. (note that the same behaviour takes place in the first time step)

Solving for fluid region WATER
diagonal:  Solving for rho, Initial residual = 0, Final residual = 0, No Iterations 0
DILUPBiCGStab:  Solving for Ux, Initial residual = 0.18658, Final residual = 0.00139597, No Iterations 1
DILUPBiCGStab:  Solving for Uy, Initial residual = 0.133811, Final residual = 0.0127596, No Iterations 1
DILUPBiCGStab:  Solving for Uz, Initial residual = 0.11622, Final residual = 0.000953427, No Iterations 1
DILUPBiCGStab:  Solving for h, Initial residual = 0.151693, Final residual = 0.00657475, No Iterations 1
Min/max T:-0.897606 561.517
GAMG:  Solving for p_rgh, Initial residual = 0.270376, Final residual = 0.00219579, No Iterations 32
GAMG:  Solving for p_rgh, Initial residual = 0.147484, Final residual = 0.00118562, No Iterations 32
diagonal:  Solving for rho, Initial residual = 0, Final residual = 0, No Iterations 0
time step continuity errors (WATER): sum local = 0.0796439, global = 0.00043864, cumulative = 0.00261569
GAMG:  Solving for p_rgh, Initial residual = 0.139599, Final residual = 0.00115698, No Iterations 34
GAMG:  Solving for p_rgh, Initial residual = 0.0933175, Final residual = 0.000917169, No Iterations 32
diagonal:  Solving for rho, Initial residual = 0, Final residual = 0, No Iterations 0
time step continuity errors (WATER): sum local = 0.072617, global = 0.000614247, cumulative = 0.00322994
DILUPBiCGStab:  Solving for epsilon, Initial residual = 0.140317, Final residual = 0.000689391, No Iterations 1
bounding epsilon, min: -2.69034 max: 1.27591e+12 average: 1.15774e+08
DILUPBiCGStab:  Solving for k, Initial residual = 0.101415, Final residual = 0.000854208, No Iterations 1

In case you need a look into fvSchemes or fvSolutions, tell me. Although I don't think the problem emerges from here, as similiar calculations with straight geometries did converge.




Greatly appreciate any advice, in case someone had a similar problem with strongly bent geometries and hex-mesh.

Best regards
doerschlott