Several issues at play here.
First: If you perform a normal linear analysis, small displacements, a rotation of a body will look like inflation when you exaggerate the deformation. Imagine displacement along a tangent to the cylinder. If you scale that, the cylinder wil "grow". You could live with that, maybe keep the deformation scale at 1.0 but the right solution is: perform a large deformation analysis.
In the linear case, all loads and contacts are applied at what is the original undeformed shape. When something actually moves, or deflects significantly, you need large deformation analysis.
You also require output in several (many) steps. in any case, keep deformation scale at 1.0 (otherwise in contact analysis objecty appear to move through each other).
I do not know exactly the consequence of your modeling approach "they are defined as 3D contacts in
mechanisms and then dragged into place"
I would see to it that the rtolers are constrained in some sort of form to prevent them flying away to the sides (singularity) and define contact in structure.
Below is a video of an example with really lartge deformations (no contact). There is no expansion here.