Post-Mortem: The Axis That Lost Its Edge
What happened. A high-speed pick-and-place axis on an assembly cell started showing positioning drift after eleven months. The integrator had run the axis at 1200 mm/s with a C7 rolled ball screw, no preload. At warranty time, the backlash measured 0.18 mm — three times the original spec. The customer rejected the acceptance re-test. We spent three weeks on-site replacing the nut and re-dating the machine. Here’s what the numbers actually said, and where we went wrong.
The spec sheet vs the floor
C7 rolled screws ship with a backlash spec of 0.05 mm (0.002 in). That number is true at room temperature, with a new nut, mounted on a granite surface plate. On an aluminum extrusion frame, running 20 hours a day, the balls and race wear in. The wear rate isn’t linear — the first 500 hours settle things, then the wear creeps. By month eleven, the effective backlash can triple without anyone noticing until the CMM report flags it.
The root cause wasn’t the screw quality. It was the selection logic. We’d treated “positioning axis” as a single category and defaulted to C7 because it was cheaper. That choice belongs on a conveyor pusher, not on a dispensing axis that needs ±0.02 mm repeatability.
What the preload actually does
Preload presses the balls against both raceways simultaneously. There is no free-play zone. When you reverse direction, the nut moves instantly. The cost is friction torque, which turns into heat, which turns into thermal growth. That tradeoff isn’t a footnote — it’s the entire design decision.
Three preload levels show up in catalogs:
| Level | Typical F0 / Ca | Temp rise | Where it belongs |
|---|---|---|---|
| P0 (light) | ~5% | 5–10°C | Most automation axes |
| P1 (medium) | ~10% | 15–20°C | Cutting, grinding, high rigidity |
| P2 (heavy) | ~15% | 20–30°C | Ultra-precision (watch the motor and cooling) |
The friction torque from preload is small but continuous. For a 16 mm lead, 500 N preload: T = F0·lead / (2π·η) = 500 × 0.016 / (2π × 0.9) ≈ 1.4 N·m. The motor has to produce that torque 100% of the time, even at standstill. Size for it. If the motor is already at 85% load at P0, going to P1 pushes it into overload.
Where preload doesn’t earn its keep
Not every axis needs it. A clamp that presses one way and retracts, a feed that never reverses position, a conveyor stop — these see the load take up the backlash themselves. Specifying a preloaded nut there adds cost, heat, and failure modes for zero benefit. The original machine that started this story was a reversing axis on a dispensing head. It should have had P0 double-nut preload from day one.
How we re-spec’d the replacement
We swapped the C7 nut for a C5 ground, double-nut, P0 preload. Lead stayed 10 mm. The motor torque budget got re-checked: the additional 0.4 N·m of preload friction was well within the servo’s RMS rating. We added a thermocouple on the nut housing as part of the FAT checklist — baseline temperature at 20°C ambient, alarm at 45°C. The frame end changed from fixed-free to fixed-supported to handle the thermal growth from the preload heat.
Six months later, the backlash is still 0.01 mm. No drift.
The line that should have been in the spec: “If the axis reverses direction and accuracy is in the spec, preload it. Full stop. The C7-without-preload option exists for unidirectional loads, not for every axis in the machine.”
What I’d tell the next engineer
The C7 vs C5 decision is not a cost tradeoff you make once at the BOM stage. It’s a function of how the axis moves, what it carries, and how tight your accuracy spec is. If you’re unsure, spec C5 P0 double-nut. The price difference is small, the heat is manageable, and you won’t be back in eleven months explaining why the machine can’t hold tolerance.
