Holy Thread Resurrection, Batman!
My quest continued unabated, but I had enough hints to make progress. It was a pretty slow job that entailed:
- Starting with the STEP file I originally received of a somewhat complex fabrication
- Turning off STEP Compound Merge to make separate parts and reading into FreeCAD
- Defeaturing it and quartering it in FreeCAD (it has two planes of symmetry)
- Writing it out as another STEP file
- Reading into a standalone copy of GMSH
- Manually identifying the surfaces I wanted (gah, this took hours!)
- Putting the surfaces into physical groups
- Making a 2D mesh
- Exporting the mesh from GMSH in the physical groups as a UNV file
- Importing the UNV file into Salome
- Working at length with Salome to stitch the mesh manually (gah! this took days!)
- Exporting the msh again as a UNV file
- Re-importing the mesh to FreeCAD as a UNV file
- Exporting the mesh as an ABAQUS INP file
- Writing a shell script that looked for the groupings in the UNV file and reformatted them into ELSET commands
- Editing in the different thicknesses for the different components
- Running a structured experiment looking for the influence of each panel on the important primary modes of the structure, using Calculix from the command line
- Exporting results of interest to VTK files using ccx2paraview.py (and Python, obviously)
It was quite the labour of love, I can tell you. However, it did add value because it has shown that the performance of the whole structure is really held back by one particular part, as is often the case. So we can lift the performance of the whole structure by upgauging that part, or save weight hand over fist by thinning every other part.
The shell mesh had about 1.7 million degrees of freedom, and took about 60 minutes to solve on my Dell G5 laptop with 16GB of RAM. It was pretty much maxed out for RAM most of the time, so it would have been quicker if I had more. There’s no way a solid mesh would have solved on this laptop, I think; a recent cantilever example I did to teach someone about different meshes went from about 400 DOF as a shell mesh to about 40000 DOF as a tetrahedral mesh, so a factor of about 100 for switching away from shells.
It has been a pretty steep learning curve, and you can question my sanity in all of it. However, it seems with some determination there is a genuine path to free and meaningful finite element analysis out there.
Going forward, I want to refine the process and am open to any and all of your collective suggestions, ridicule, etc.