[Quentin Chevalier]

Ok this is linked to the case detailed in this other topic but related to a different problem.

I have a Re=10^6 axisymmetric case solved with simpleFOAM (not in SIMPLEC format) on a quadrilateral structured mesh. This a steady-state computation using limited schemes. I'm doing a convergence analysis on my results (i.e. increasing the number of points and checking it doesn't affect the results too much).

This is a velocity-driven jet flow with a annular straight nozzle that acts as a wall. Top boundary condition is no slip and corresponds to coflow maximum.

I have unexpected results ; increasing the mesh size from 143k element (called coarse) to 1606k (called fine) reduces result quality ; I have spurious plumes appearing above the nozzle and a laminar-like jet.

Now I made my mesh with gmsh and tried to keep the nozzle area both refined and as continuous as possible in terms of elements size. This meant fine-tuning Progression parameters to obtain the desired result, which is attached for a quick visual appraisal of the difference ; I can guarantee that the fine mesh has more elements, but since they are not spread evenly I'm not sure how much refined in terms of length scale is the new mesh. I chose quads so I could have larger aspect ratios for my elements and solve the viscous lengthscale without too many points in x.

I've tried adding more iterations, complexifying the mesh, and going slower during the calculation to pinpoint the origin of the problem, but to no avail. I find it slightly disturbing that adding cells can create a problem - especially since the fine mesh looks all right to me...

My gmsh is 4.7.1, I installed OpenFOAM v2112 in a container.

Calculations are becoming long with these meshes on my machine. Anyone here had a similar problem ?