MLMG solver fails to converge when AMR grid is used

[taiyaa]

I have a 3D model car simulation problem. I adapted the model center so the model problem could be computed on the base grid. But when I set the amr.max_level=1, the program will complain "amrex::Abort::0::Failed to build required coarse EB level 1 !!!". (For my current problem, for instance, 'eb2.stl_center = -0.10000001 -0.000001 -0.01'). If I move the center a little bit, 'eb2.stl_center = -0.05000001 -0.000001 -0.01', the problem will be "amrex::Abort::0::MLMG failing so lets stop here !!!" .

I tried to shift the solid center, but I found myself always fall into either the two above situations. I am not sure what caused these issues. I understand for curved surfaces, multiple cuts may occur on a single cell so the program may not work, see AMReX-Codes/amrex#608. For my case, the program did not complain the multi-cut issue. Besides, the base grid does not have multi-cut issue. When I refine the grid, I am not expecting the refined grid will suffer the multi-cut issue. Not sure why MLMG does not work on my case. Could anybody have a look at my problem?

I attached the input file and model (ASCII STL format, remove the '.txt' should work) below.
input.txt
whole_car.stl.txt