The drawing says coaxiality 0.02 mm, runout 0.01 mm, the CNC shop nods, and the first 50 housings come back with drift, noise and backlash over spec. The tolerance didn't fail — the chain that produces it did. Here are the root causes we see in robot reducer housings, and the drawing and supplier changes that make ±0.02 mm actually hold.
A robot OEM designed a joint reducer housing with bearing-bore coaxiality at 0.02 mm and face runout at 0.01 mm. The first article passed. Then the 50-piece production run shipped, and 30% came back with excessive drift, noise and backlash over spec. The shop said “the tolerance is too tight”; the engineer said “you approved it”; the truth was that the tolerance was achievable — but only with a process chain that none of the parties had specified.
A harmonic drive housing has three features that must stay in one reference frame: the flexspline pocket (roundness), the circular-spline bore (coaxiality with the output bearing seat), and the mounting face (perpendicularity to both). Machined in three separate setups, each setup re-introduces alignment error — and the errors add, not average.
Industry practice for motion-critical housings is to finish the bore, face and bolt circle in one setup so every datum traces to a single zero point. Suppliers report holding total runout to 0.005 mm or better and bolt-circle true position to the same band when bores and faces are machined from one reference (source: Bishen Precision). When a shop re-fixtures for the production run — breaking the reference that made the prototype good — first-article performance drifts.
Reducer housings are increasingly thin — wall sections of 1–2 mm are common in lightweight robot joints (source: EPRO MFG). A thin wall under a 3-jaw chuck or a vice distorts elastically while clamped, then springs back when released: the part measures in-spec on the machine and out-of-spec off it.
Aluminum at 6061-T6 is the usual culprit because it is easy to cut and light, but its low stiffness and high thermal growth (CTE ~23.6 µm/m·°C) make it sensitive to both clamping force and cutting heat. The fix is a fixture strategy that supports the wall (pot chucks, custom mandrels, low-clamp-force soft jaws) and light finishing passes with coolant, so the part leaves the machine in its free state.
A housing machined from cast or bar stock carries residual stress. Rough machining redistributes it; over the following days or weeks the part “creeps” — bores drift out of round, faces lose flatness — and the CMM on day 1 disagrees with the CMM on day 10.
The standard countermeasure is stress relief between rough and finish machining: artificial aging or vibration stress relief after roughing, then finish machining the precision features. For cast-iron and alloy-steel RV reducer housings this is published practice for holding coaxiality at 0.005–0.008 mm (source: laitujia). If the supplier has no stress-relief step between rough and finish, dimensional creep is a live risk.
A part can pass on the shop's CMM and fail on the customer's for the same feature. Three causes dominate:
A housing that hits 0.015 mm coaxiality at machine warm-up may drift to 0.025 mm after hours of running, as spindle and ball-screw growth changes the tool tip position. Batch repeatability is a machine-condition story, not a first-article story.
This is why qualified robotics suppliers run the tolerance band with CMM verification on every first article and periodic SPC on production, and why buyers should ask for process-capability data (Cpk on the coaxiality/runout features) rather than a single good sample. Machine-side compensation — in-process probing with a touch probe to re-zero the coordinate system before critical features — is the modern countermeasure and is now standard on precision cells (source: laitujia).
| Stage | Check | Why |
|---|---|---|
| Drawing release | Datum scheme explicit; bearing-bore coaxiality and face runout called out to one reference | Kills cause 4 (datum mismatch) |
| Drawing release | Features ≤ ±0.02 mm state measurement method + 20°C reference | Kills cause 4 (method/temperature) |
| RFQ | Bores to be finished in one setup; probe verification on critical features | Kills cause 1 (fixture stacking) |
| RFQ | Wall thickness ≤ 2 mm flagged for low-clamp-force fixturing | Kills cause 2 (thin-wall spring-back) |
| RFQ | Stress relief (aging/vibration) between rough and finish | Kills cause 3 (dimensional creep) |
| FAI | Production fixture/program identical to FAI; Cpk on coaxiality/runout from a real run | Kills cause 5 (drift in volume) |
| Production | CMM first-article per batch + periodic SPC; deviation logged on 8D | Catches drift before assembly |
Bearing-bore coaxiality, flexspline-pocket roundness, and face perpendicularity — all to one datum. Published practice for motion-critical housings: bearing-bore-to-motor-register concentricity near 0.013 mm, total runout to 0.005 mm in one-setup machining, bores H6–H7, and face runout ≤ 0.01 mm. Which number is right for your design depends on the drive's backlash and noise budget, but every one of them must reference the same datum frame.
Because the production run breaks the reference that made the first article good. The top causes: re-fixturing between prototype and production (fixture stacking), thin-wall spring-back from clamping, internal stress release over time, machine thermal drift, and measurement disagreement. Check the production setup against the FAI first — reference drift between prototype and production is the #1 hidden cause.
For the tight coaxiality/runout values, effectively yes. Finishing the bore, face and bolt circle in one setup makes every datum trace to a single zero point and eliminates second-setup error. Three-axis shops that index between setups can hold the numbers on a good day but rarely batch after batch — which is exactly the drift pattern buyers see.
If the housing is cast or bar stock and holds coaxiality tighter than ~0.01 mm, yes. Rough machining redistributes residual stress, and the part creeps afterward. Artificial aging or vibration stress relief between rough and finish machining is published practice for RV reducer housings holding 0.005–0.008 mm coaxiality. Ask the supplier for their stress-relief step in the process sequence.
Ask for Cpk data on the coaxiality/runout features from a real production run, plus a CMM report measured against your datum scheme. One good first article proves nothing. Also ask: are the two bearing bores machined in one setup? Is there a stress-relief step? What fixture supports the thin walls? The answers separate a shop that can hold ±0.02 mm from one that quotes it.
Four things: (1) one explicit datum scheme for coaxiality, runout and perpendicularity; (2) measurement method and 20°C reference on features ≤ ±0.02 mm; (3) a note that critical bores must be finished in one setup; (4) a stress-relief requirement between rough and finish. Each maps to a root cause that otherwise comes back as drift in production.
Send your reducer housing drawing — we'll review the datum scheme and process chain and return a DFM review with Cpk data within 3 business days.
Request a Quote