We did this a lot last year thanks to a poorly thought out design. You can see the pertinent stuff in the student's reveal video here:
Leopards 2015 Reveal
Gory details below:
That giant roller arm is supported by a leadscrew in tension on the back side of the pivot. If I recall correctly, it was a 1/2-10 2:1 screw. I would estimate a static load of about 40-lbs. Dynamic loads much higher if someone hit the arm and forced it down. We originally coupled it by turning the screw down to 8mm and using a ruland rigid coupling to a CIM output shaft. The litany of failure reported to me from out first regional:
- Coupling failed by slipping axially on CIM output shaft. Fixed by set screwing coupling to pin it axially.
- Retaining ring failed. Partially due to improper installation (it's HARD to seat it correctly). Even after seating properly, they could still fail it running into something. Fixed by screwing a retaining plate to the front of the VP. Hole was just big enough to let the 8mm shaft through.
- CIM output shaft was completely pulled out of the retained part of the output shaft assembly. I had assumed the output shafts were atomic parts, so I had no idea what the team was were saying over the phone until I pulled up the cad and realized the CIM shaft was press fit into the other half of the assembly. I believe they had someone hit the tail end with a bit of weld for this one.
The ultimate solution was to conclude that VPs simply aren't meant to support axial loads. As Chris said, the only actual solution is a 1:1 torque transfer to a shaft that's actually designed to support an axial load. I ended up spending a weekend with a lathe and CNC mill doing a full-on leadscrew end prep with a seat for an angular contact bearing, threads for a lock nut, and keyed shaft. The bearing and shaft went into a block that bolted to the front of the VP. The keyed 8mm shaft slid into a hole that was conveniently left in one of the shaft assemblies a CIM output was forcibly removed from.
That whole solution is one step further than Bryce's. His solution would work fine until you started separating the two parts of the output assembly. I think I moved the failure point to the 10-32 screws holding the block to the VP.
TL;DR. VPs aren't designed for significant axial loading. Especially in tension. Design for a shaft that can support the axial load and a 1:1 torque transfer to said shaft, or your VP is going to disappoint you at some point.