Hi there,
I tested last year extensively exactly what you are after: assemblies in structural FEA in Calculix.
It won’t do it as a general method for assemblies & joints.
If you read the ccx help docs carefully, you’ll find that the couplings (2 x types):
‘soft’ [ RBE3 (Nastran) or distributing (Abaqus) ] &
‘rigid’ [ RBE2 (Nastran) or kinematic (Abaqus) ]
don’t handle properly rotations / moments for both loads and constraints, or merged at ref_nodes to other mesh.
That means that all joints will be incorrect as a general method, only translations will be handled.
But it gets worse: Calculix, I believe, doesn’t handle either 6DoF_springs [ CBUSH (Nastran) & connector cartesian+rotation (Abaqus) ].
This means that we cannot connect reference nodes on pairs of couplings with 6 K values for rivets, bolted areas, etc.
Even if you try beams instead of 6DoF_springs they’ll disconnect the rotations at the shared reference node on couplings either end. Some codes have MPCs, but these will most likely over-constrain the model up at some DoF with some assumed inf. stiffness; they can also fail later when moving into non-linear FEA.
Below is a fig showing a tiny test model with a bearing and a shaft that Calculix couldn’t handle, while all other commercial codes will do via couplings + 6DoF_springs.
I’ve recently reviewed all this for structural assemblies with joints (only way to build realistic FEA models) with several open source FEAs and none of them do it, except code_aster.
I hope this helps.