Harmonic-Drive Acceptance Checklist: 7 Checks Before You Approve the Reducer

If you're the person who signs the acceptance sheet for a precision reducer—for a robotic joint, a wafer-handling axis, or a telescope mount—this article is your checklist. It's the procedure I run in our motion-control integration shop before a harmonic-drive unit reaches a customer. It takes roughly one hour per unit, and it catches more real defects than any software dashboard I've seen.

Most of the harmonic-drive systems news today focuses on AI condition monitoring and higher torque density. That's fine. But as a quality/compliance manager who reviews roughly 200 units a year—and rejected about 5% of supplier first deliveries in 2024 for dimensional or runout non-conformance—I can tell you the gaps are still in basic physical checks. The fundamentals haven't changed: backlash, stiffness, mounting, and traceability. What has changed is the execution, and that's where mistakes hide.

This checklist has seven checkpoints. Use it for incoming inspection, first-article validation, or final acceptance before you mount an expensive gearbox onto a much more expensive machine.

Stage 1: Three bench checks before assembly

1. Confirm the mechanism and the matching components

I get the appeal of the harmonic-drive mechanism: one stage gives you reduction ratios from about 30:1 to 160:1, near-zero backlash, and a compact coaxial package. But what gets overlooked is that the wave generator, flexspline, and circular spline are a serial-matched set. Never mix components from different kits, even if the tooth counts look the same.

Confirm the ratio manually before install. Lock the circular spline, rotate the wave-generator input by hand. A 50:1 reducer should require exactly 50 input revolutions for one output revolution. If the number doesn't match the part number, stop and call your supplier.

2. Measure output-flange runout before you bolt anything to it

Support the unit by its circular-spline housing. Run a dial indicator against the output-flange pilot and rotate the input at least two full revolutions. Record the total indicated runout and compare it with the tolerance on your drawing—not with a headline number from the brochure. In one 50 Nm-class line we accept 0.025 mm on the output pilot, but the exact pass/fail limit belongs to your design.

A high reading after shipping usually means the flexspline was deformed by rough handling or the flange was dented during installation. If it's high, don't try to pull it in with bolt torque. Return it.

3. Measure breakaway torque while the unit is still cold

This is the step most people skip because motorized units make it easy to command a slow speed and watch the current. That gives you running torque, not breakaway torque. Use a torque wrench on the input and rotate slowly in both directions. The unit should break free smoothly.

Here's something vendors won't tell you: published no-load torque is usually measured after the grease has been distributed by a run-in cycle, often at 20–25 °C. A fresh unit that sat in a cold warehouse will feel stiff for the first few rotations. Let it warm up and rotate it 20–30 times before you decide. If the breakaway torque is still ragged or far above the stated limit, that's a genuine problem—usually excess wave-generator preload or poor lubrication.

Stage 2: Under torque and motion

4. Test your mounting system, not just the gearbox

The next mistake shows up after assembly. A harmonic drive can look perfect on the bench and still create oscillations on your machine because the circular-spline housing is bolted to a plate that flexes under load. The flexspline is a thin-walled cup by design; it depends on the mounting interface for support.

Mount the unit to the same plate or bracket it will see in production. Clamp that bracket to the bench. Apply 50% and then 100% of rated torque to the output flange with a calibrated lever arm, and measure the deflection at the flange with a dial indicator. If the reading at your load radius is more than a few hundredths of a millimetre, the mount is the problem. Put another way: don't spend a fortune on a low-backlash reducer and then bolt it to a sheet-metal bracket. I've seen an entire line blamed for oscillation that was actually a flexing adapter plate.

5. Measure lost motion at the output after mounting

The catalogue says “zero backlash,” and the unit probably did have negligible backlash at the factory. But after you bolt the output to a coupling, load flange, or gear stage, the effective lost motion at the payload can be significantly larger.

Procedure: attach a lever arm of known length R to the output. Set a dial indicator at the end of the arm. Apply a small alternating torque—a few percent of rated is usually enough—and measure the indicator movement between reversals. Convert to angular value using 206,265 × indicator movement / R in arc-seconds.

Then compare that number to the acceptance criterion in your purchase order. A harmonic drive often shows less than an arc-minute of lost motion when new, sometimes much less in precision classes, but your number is the one that matters. As a QC person, I'll also say this: under ISO 9001, the acceptance criteria you communicate to your supplier are what count. A sales brochure doesn't define a contract.

6. Establish a baseline on motorized units with AI monitoring

Since around 2022, more specifications have asked for a brushless DC motor and AI-driven condition monitoring packaged with a harmonic-drive reducer. The AI part is only as useful as the baseline data you give it. Without a healthy reference, an “anomaly detected” alert just means “something changed”—and someone still has to tear the machine apart to find out what.

For these integrated units, run a speed sweep from 10% to 100% rated speed under load, and save the raw current waveform, drive position error, and case temperature. It takes about 20 minutes. That file is what lets you later tell the difference between a motor-commutation issue and a reducer issue. It's also the exact data that prevents a 2 a.m. phone call six months from now.

Stage 3: Final sign-off data

7. Reach thermal steady state and document traceability

Last physical check: run the unit at a representative duty cycle—75% rated load is usually enough—until the case temperature stops rising. Compare the stabilized temperature to the supplier limit. In a harmonic drive, that reading is dominated by the wave-generator bearing and grease condition, so it deserves more attention than motor winding temperature alone.

Then create the acceptance record. Photograph the serial numbers, note the date code and grease batch if visible, and write down the test values, test equipment, and the reviewer's name. A reducer without a traceable serial number isn't a spare part—it's a paperweight. When a field failure shows up later, that record is the only thing that tells you whether to repair the gearbox or redesign the mount.

What I still see rejected at final review

These are not rare edge cases. They show up in our monthly quality reviews:

• “Zero backlash” gets treated as a permanent property. It's the result of elastic preload in a flexible mechanism, and it can change with tooth wear and wave-generator bearing condition. If a supplier promises zero backlash for infinite cycles, ask to see the test data behind that sentence.
• Coaxial and right-angle solutions get mixed up. For example, ask what uses a bevel gear and you'll get automotive differentials, machine-tool right-angle heads, and older servo gearboxes. Bevel gears are the right answer when the axis must turn 90°. A harmonic drive is a coaxial mechanism by design. If you add a bevel stage to fix the layout, don't blame the harmonic drive for the backlash that the bevel stage introduces.
• Tuning around a mechanical defect. If an axis starts oscillating in the low-frequency range, look at the reducer and mounting before you spend a day re-tuning PID gains. A genuinely backlash-free drive removes one entire class of instability. It doesn't fix a flexing bracket or a worn coupling.

One more thought, and I'll step off the bench. If your commissioning engineer has started researching natural blood pressure reducers because the axis won't stop hunting, check the mechanical path before you rewrite the motion controller. A properly applied harmonic-drive reducer doesn't just improve positioning accuracy; it removes a whole class of late-night retuning sessions.