Float Table: redesigning how a drive belt fails
The old connection passed every required test. The field found what the tests did not.
- When
- 2021–2022
- My role
- Design Engineer II, sustaining engineering
The situation
The Float Table is Humanscale's mechanically counterbalanced sit-stand desk, two sizes, with a drawing package of more than 200 parts and assemblies that I maintained. Two jobs ran under sustaining engineering at the same time: a request for a left-handed release paddle, and a spike in returned units.
shock-hero.jpgThe left-handed option
- Sales wanted a left-handed release paddle with as few new release-cable lengths as possible. I ran the cable-length optimization across all eight crossbeam, worksurface and side combinations. It needed one additional length.
- Designed a new brake release bracket and cable extender so the mechanism could be released from the left side.
The field failures
Root cause analysis traced the returns to a shock load hitting the top or bottom of the travel with momentum. It damaged two things: a bent sheet-metal hard-stop tab and the riveted aluminum drive-belt end connections.
The part that surprised me was that the rivet design had passed every required test. By spec it was valid. What only the field showed was that the rivet holes created a stress concentration. Static overload was fine. A jostling impulse could spike past the weakened point and snap it.
The fix avoided piercing the belt at all. Two fixture plates clamp the belt around a small rod under tension, so there is no stress concentration anywhere in the belt. I also built in a controlled slip of about 5 mm on purpose. An overload now bleeds off energy gradually instead of breaking all at once.
- Replaced the sheet-metal hard stop with a more rigid riveted steel part.
- Tested old and new designs to failure. The new connection held roughly three times the load, failed by slipping instead of rupturing, and ran 22,500 cycles without failure.
- Handed off to operations with drawings, tooling quotes and tooling validation samples. Supplier samples passed at 100%.
What I took from it
Making it stronger was not the real fix. The fix was deciding how the assembly should fail and designing for that.
shock-compare.pngshock-failure.jpg