[Regis_]

I'm running into a weird issue with moveDynamicMesh. I first have a coarse grid that I conform to an STL geometry using moveDynamicMesh. That works fine. Here's a snapshot of that stage


Later on I refine the whole grid (and solution) using the topoSet/refineHexMesh combo. The grid refines as expected and passes checkMesh. The grid is still conformed as it were before on the lower resolution grid:


Now, I want to re-conform the higher resolution grid to the same geometry. (In the meantine I do some local grid refinement, but that shouldn't matter.) At this point, what I expect is that the points will move very little, since only the new points need to conform to the STL file. This is where I run into issues. On the very first time step of the moveDynamicMesh solver, the new grid goes bad -- extremely high negative volume, extremely high skew -- and the moveDynamicMesh solver quickly crashes.

I managed a find one very specific case that I can sort of see what is going on:


In this specific run, the grid "recovers" and conforms as expected:


However, I can't get it to work with any other STL or grid size. Even in this case that sort of worked, if I do one more global refinement before I call moveDynamicMesh for the second time, it crashes.

In constant/dynamicMeshDict I have:

dynamicFvMesh dynamicMotionSolverFvMesh;
motionSolverLibs ("libfvMotionSolvers.so");
motionSolver displacementSBRStress;
diffusivity  quadratic inverseDistance 1(lower);

where `lower` is my bottom patch (the one it is conforming). I tried to reduce greatly the `velocity` in `processor*/<timeDir>/pointDisplacement.boundaryField.lower.velocity`, as well as smaller time steps, but no luck. One thing that I noticed is that if I have some local refinement regions that touches the `lower` patch, then the grid goes bad quicker.

My question is--
What settings in the dynamicMeshDict can I experiment with so that I get properly conformed mesh all the time? Has anyone else experienced this behavior before? I'm running OpenFOAM 6.