Backlash: the free play
Backlash is mechanical clearance: the angle the output can move while the input is locked, under essentially no load. It's measured by reversing a tiny probing torque and reading the dead band. In a conventional gearbox this is the dominant error term; in a strain wave gear the preloaded multi-tooth mesh removes designed-in clearance, which is why our harmonic series spec backlash from ≤ 15 to ≤ 40 arcseconds depending on frame, and why the figure we actually measure and log on the bench is typically better than the published limit.
An arcsecond is 1/3600 of a degree. At a 500 mm robot reach, 20 arcseconds subtends about 0.05 mm at the tool (the width of a human hair).
Lost motion: the practical number
Lost motion is measured differently: apply ±4% of rated torque to the output — a realistic disturbance, not a feather touch — and read the total angular deflection between the two directions. It bundles residual clearance and low-torque elastic wind-up and hysteresis into one figure, typically a small number of arcminutes for a strain wave stage. We measure it on the bench per serial; ask for the figure for your frame alongside the published repeatability spec.
Torsional stiffness: the spring you're driving through
Load the output with the input locked and the drive winds up like a torsion spring. Datasheets publish this as a stiffness curve, usually piecewise (K1, K2, K3) because the mesh stiffens as more tooth flanks come under load. Stiffness matters twice:
- Static droop. A gravity load on an arm deflects the joint by torque ÷ stiffness. This deflection is repeatable, so controls can compensate, but only if the stiffness is consistent unit to unit.
- Dynamics. The drive's spring against the load inertia sets a resonance that caps your servo bandwidth. A stiffer drive (or a bigger frame size) raises that ceiling.
Hysteresis: what doesn't spring back
Load a drive, release it, and it doesn't quite return to zero; friction in the mesh eats a small angle. That residual is hysteresis loss, and it's already included in the lost-motion measurement. It's the reason lost motion can never be zero even when backlash effectively is.
How we test, and what ships with the unit
Every drive that leaves our works is measured, not sampled: backlash and lost motion on the bench, no-load running torque across a full rotation to catch tight spots, and ratio verification. The results are logged against the serial number, and the test record ships with the unit. If a number matters enough to print in the catalogue, it matters enough to measure on every serial.
Related: What is a strain wave gear? · How to size a harmonic drive