How strain wave gears work, how to size them, and how to read the numbers on a datasheet, written by the engineers who machine and measure them.
Average load torque is a cube-weighted mean, not an arithmetic one. A worked index-and-hold cycle where sizing on the peak, on the time-average, and on the cube rule lands on three different frames.
Grease usually ends a drive before the steel does, yet almost no catalogue commits to an interval. Why that is, the four variables that actually set it, and the failure signature that arrives first.
Interface, ratio, torque class, backlash grade: what has to match for a drop-in, what forces a full re-qualification, and where the datasheet stops being able to tell you.
Bearing convention trains you to read a life figure as L10 — 90% survival. Harmonic drive ratings are commonly referenced to mean life instead, and the part that wears out is not the flexspline.
One kN·m/rad figure hides a stiffening spring with hysteresis. At rated torque our gear series winds up 291 to 400 arcseconds — several times its published transmission accuracy.
A thin steel cup, an elliptical cam, and a two-tooth difference: how the mechanism delivers 100:1 reduction with zero backlash in a package the size of a hockey puck.
A practical sizing walkthrough: peak vs rated torque, choosing a ratio, checking input speed, and a worked example that lands on a real part number.
Three numbers that get confused constantly. What each one measures, how we test them on the bench, and which one actually limits your positioning accuracy.
Five questions that decide which reducer family belongs in your joint, with real catalogue numbers, including why 15 arcseconds vs 12 arcminutes is 0.07 mm vs 3.5 mm at the tool.
One failure mode is a clock, one is a cliff, one is an accident. What accelerates each, what to monitor, and the operating decisions that set your service life.
Rated efficiency is true at one operating point. The loss mechanisms, the three torque numbers nobody reads, and why gravity axes always get brakes.
A motor-side encoder sees the motor, not the load. What bit-depth actually buys, why resolution isn't accuracy, and when dual-absolute feedback earns its cost.
The bus decides where your control loops can live. Cycle times, sync, wiring, and cost, plus the one question that settles it: where does interpolation live?
Motor current is a rumour of output torque, filtered through friction. When the rumour is good enough, and when only a calibrated sensor will do.
Five purchase orders and an intern-year of integration, or one part number. The honest make-vs-buy arithmetic, including the three cases where building wins.
A humanoid carries its actuators and its battery in its own hands. The mass cascade, the energy budget, force limits, and the right joint per axis.
The joint met its spec on our bench. Whether it still meets it on your machine is decided by a surface, a bolt pattern, a cable, and the first hour after power-on.
Send us your peak torque, output speed, and envelope. We'll size the drive and quote, usually the same working day.
Request a quote