Harmonic Drive Housing Machining: Keeping Three Bores on One Axis
A strain-wave reducer does not transmit torque through a fixed mesh the way a spur gearbox does. An elliptical wave generator deflects a thin-walled flexspline into two lobes, and because the flexspline carries two fewer teeth than the circular spline, each input revolution advances the output by two teeth. At a 100:1 ratio the flexspline cup flexes about 2,000 times per minute at 1,000 rpm input. The circular spline is bolted to your housing and never moves. So the housing is not a container. It is the datum that the entire gearset trusts.
That is why harmonic drive housing machining is judged on what a general shop will not measure unless you ask. Bore diameters usually pass. What fails is coaxiality between bores, the clearance envelope that lets the flexspline breathe, and the axial face the output flange runs against.
What the housing is actually holding
A typical cup-type unit has three functional interfaces in the housing: the input bearing bore that supports the wave generator, the pilot diameter and bolt circle that locate the circular spline, and the output bearing bore that carries the output flange. All three must share an axis within a few thousandths of a millimeter, and the shoulder faces that clamp each bearing must be perpendicular to that axis.
Errors compound in a way that is not linear. Circular spline eccentricity of 0.01 mm produces a once-per-revolution error component that appears directly as lost motion and torque ripple. Output flange face runout of 0.01 mm at a 60 mm radius tilts the payload by roughly 34 arcseconds, which a controller cannot compensate because it does not repeat on reversal.
The features that decide whether the assembly passes
These are the callouts worth writing explicitly, with the failure mode each one prevents.
| Feature | Typical callout | What happens if it drifts |
|---|---|---|
| Input bearing bore to output bearing bore | Coaxiality 0.010 mm, evaluated over the full bore length | Once-per-rev error, measurable as lost motion above 1 arcmin |
| Circular spline pilot diameter | Ø tolerance H7, concentric to bearing axis within 0.015 mm | Tooth mesh offset, uneven load across the two lobes, whine at speed |
| Bearing shoulder face | Perpendicularity 0.005 mm to the bore axis | Bearing cocked on its race, cage wear, axial play that grows with temperature |
| Output flange mounting face | Axial runout 0.008 mm at the flange OD | Payload wobble, angular error that reverses with load direction |
| Flexspline clearance envelope | 0.3 to 0.5 mm radial, 0.5 mm at cup bottom | Cup contact, grease breakdown, heat, premature flexspline fatigue |
| Bore roundness | 0.004 mm at three axial stations | Bearing outer race distortion, higher starting torque, shorter life |
The clearance envelope is the callout most often left off the drawing. It comes from the counterbore depth behind the circular spline. Reduce that depth to gain stiffness and the cup bottoms out under axial load, failing in hours rather than months.
See also: Gearbox Housing Machining: Bore Alignment, Seal Bores and Material Choice and Battery Housing Machining: Gasket Grooves, Flatness and Leak Test Prep.
Holding coaxiality when the wall is 3 mm thick
A 7075 housing for a 60 mm frame size commonly has 3 to 4 mm walls and a 90 mm overall length. Boring that in two setups will not hold 0.01 mm: refixturing adds 0.01 to 0.02 mm and the part moves when the jaws release.
The sequence that works starts with stress. Rough all bores and the outer profile leaving 0.8 mm on diameters, then release the blank and let it sit. A 24-hour natural age is enough for most 7075-T651 geometries; a 2-hour soak at 120 °C gets there faster. Semi-finish to 0.15 mm, then finish every bore in one setup — on a 5-axis machining center that reaches both ends without reorienting, or by line-boring from one side with a bar that spans the housing.
Clamping is the other half of the problem. A three-jaw chuck on a thin ring distorts it into a triangle that springs back after the cut, so the bore measures round on the machine and out-of-round on the CMM. Clamp on the flange face with six or eight low-pressure points, or bore in a soft-jaw pocket machined to the actual OD at the actual clamping pressure. If you must hold on the OD, keep clamping force under 2 kN.
Finish with light passes: 0.15 mm radial depth, 0.05 mm/rev feed, a sharp positive-geometry insert, coolant aimed at the cut, and a spring pass to remove the last 0.005 mm of deflection.
Material choice: 7075, 6061 or 4140
7075-T6 is the default for robot joints. At 71 GPa modulus and roughly 505 MPa yield it gives the most stiffness per kilogram, and it machines cleanly at 400 to 600 m/min with carbide. Its weakness is thread durability — an M4 thread survives maybe 30 to 50 assembly cycles before the flanks deform, so use inserts or through-bolts on anything that gets serviced.
6061-T6 is the cost-down option: 30 percent lower yield strength, cheaper per kilogram, and it anodizes more evenly. For a 30 mm frame size reducer under 3 N·m output it is adequate. Avoid it for the circular spline pilot when the mating part is steel and the joint thermally cycles; 6061 at 23.6 µm/m·K against steel at 11.5 µm/m·K will loosen a light interference fit over a 40 °C swing.
4140 pre-hardened to 28 to 32 HRC is the choice when the housing is also the structural member of the joint, when bearing seats see fretting, or when the customer wants steel threads. It machines at 180 to 220 m/min and moves more after roughing than aluminum does, so leave 1.0 mm and stress-relieve before finishing. Weight triples against 7075.
Inspection that predicts assembly behavior
Checking bore diameters with a bore gauge tells you almost nothing about whether the reducer will pass a lost-motion test. Ask for a CMM report that evaluates coaxiality as a cylindrical tolerance zone between the two bearing bores, not as separate circular runout at each end, with roundness at three stations per bore. Measure at 20 °C after a four-hour soak; a 7075 housing read at 28 °C on a 50 mm bore comes out about 0.008 mm large against the drawing temperature.
Then blue the circular spline pilot, assemble dry, and turn the input by hand through 20 revolutions. Contact should be even around the pilot and absent on the counterbore floor. Finish by indicating output flange axial runout at 0.001 mm while the input turns — this catches shoulder face error a CMM report on the bore alone will miss.
Send the gearset drawing with the housing drawing when you source these parts. You can request a quote with both files and get DFM notes back within 24 hours.
Frequently asked questions
How tight does coaxiality need to be for a 1 arcmin reducer?
For most 50:1 to 100:1 cup-type units, 0.010 mm coaxiality between input and output bearing bores supports 1 arcmin lost motion with margin. Pushing to 0.005 mm helps only if the bearing seats and the circular spline pilot are held to the same level; a single tight feature in a loose stack changes nothing.
Can you machine the housing after the bores are hard anodized?
Yes, but plan the allowance. Type III hard anodize at 25 µm grows roughly 25 µm per surface in both directions, so the bore closes by 0.025 mm per side and will not take an H7 bearing. Mask the bores, or leave 0.03 mm on the diameter and finish-bore after coating with a diamond insert.
Why does the housing distort after boring even with light cuts?
Residual stress from the billet and the roughing operation redistributes as material is removed. The fix is sequence, not force: rough, release, stress-relieve, semi-finish, then finish clamped on a face rather than an interrupted OD. A 3 mm 7075 wall still moves 0.005 to 0.010 mm if you skip the release step.
What surface finish should the bearing bores have?
Ra 0.8 µm is the practical target for a pressed bearing seat in aluminum, with a feed pattern that avoids a lead pushing the outer ring off-axis. Going below Ra 0.4 µm adds cost without improving fit and can reduce the friction that keeps a light press fit from creeping.
Related reading
Send a drawing for a quote and DFM review.
STEP, IGES, X_T, PDF or DWG up to 50 MB. Tell us quantity and finish, and the quote comes back with tolerance and process notes, not just a price.