After seeing my
Rhombic Dodecahedron Puzzle Bram Cohen casually suggested a puzzle in which one particular piece has to be put in last. That is to say, a puzzle in which the pieces must be assembled in a particular sequence.
The picture above shows how the idea might be adapted to one of my swirly 'marble' puzzles: the four pieces must be assembled in the sequence illustrated above because each new piece blocks the movements of the previously installed pieces.
Anxious to try it out, I built a set of pieces without checking their dimensions. But I encountered some delicate, thin walls while I was peeling the models out of the support material.
VisCAM revealed some astonishingly thin, wispy structures on the thinnest piece. Some of the walls taper down to less than 0.1mm near some of the edges. It held together only because the super-thin portions are anchored to the thicker 'backbone' up the center!
It took a lot of slow, cautious work to free this thin part from the leftover plastic but it came out intact! Dipping the piece in plastic-welding solvent made it strong enough for functional testing, despite a few frayed-looking edges where the walls taper down to nothing.
Reviewing the measurements it became clear that the super-thin walls had occurred as a result of how two spiral-shaped channels gradually converged inside the puzzle. Luckily the thin piece would be amply protected by the thicker piece when the puzzle is assembled.
As I'd hoped, the pieces can only be assembled in a specific sequence. But unfortunately they fit together too tightly, so the last piece wouldn't go in.
I tried to improve the fit by sanding the pieces, but the PVC is just too soft for hand sanding so I resorted to a high-speeed abrasive wheel. This improved the fit and gave the pieces a nice smooth feel.
Perhaps it's a bad habit to keep building models with such thin walls, as they're extremely vulnerable to breakage while cleaning the model. But even though they could break during manual cleanup, they don't pose any sort of danger to the build process so the main penalty is the tedious labor. The SD300 can safely attempt any sort of ridiculous geometry, even if the STL file is riddled with defects, and the worst that might happen is the model might be disintegrate or consume excessive labor during post-build cleaning.