The Check We Did Not Have
The machine shop's production review caught a part that could not be assembled, and caught it before cutting. The adapter fit its seat perfectly and could never have reached it. The revision that fixes it matters less than the check that should have found it first.
Every fit check the program had run asked the same question: with the parts in their final position, do they overlap? The adapter answered cleanly. Seated in the body it reported zero interference, and on that basis it was released and ordered.
The shop asked the other question. Their production review came back with two numbers and a conclusion: the body's internal hex measures 16.02 mm across flats, the adapter's cylindrical feature is 18.00 mm, and the second cannot enter the first. They asked for revised models before they would proceed.
They were right, and the finding was confirmed independently against the existing model before any geometry changed. The reason it was missed is geometry that had simply never been examined. To reach its seat the shank has to pass through the hex pocket, and a round feature can only pass a hexagonal opening through the circle that fits inside it. Measured that way the opening is 16.020 mm, not the 18.498 mm you get corner to corner. The shank was 1.98 mm too wide to enter a space it fit inside perfectly. The seated position was real and unreachable, a sealed cavity with no path in. Swept along the insertion axis in 0.25 mm steps, the pair is blocked for nearly the whole journey, peaking at almost 200 mm3 of interference, and clear at exactly two places: fully outside, and fully home.
This was the same mistake twice, one level up. An earlier revision trapped the retaining screw inside the adapter. This one trapped the adapter inside the body. Both passed review for the same reason: the only test was a static fit at the destination, and a test that looks only at the endpoint cannot see a wall in the corridor.
The fix is taken entirely on the adapter, and that choice was deliberate. Widening the body's pocket to admit an 18 mm shank would have forced the relief behind it outward too, dropping the neck wall from 5.500 mm to 4.596 mm. That would have spent an approved dimension on two parts to avoid changing one number on one part. The shank was turned down to 15.9 mm instead. The body did not change at all, though it was reissued at the new revision so the drawing pack stays internally consistent.
The honest lesson is not the new diameter. It is that this program had no way to find this on its own. The check that would have caught it did not exist, so it was written: insertion is now simulated as a path rather than a destination, with a deliberate negative control that has to fail, and a geometry that cannot be assembled no longer exports. That check is the actual deliverable of this revision.
It is worth being clear about who found what. A vendor's review is not a substitute for engineering, and it caught this one because the design reached them with a defect in it. The value of the swept check is that the next error of this kind gets caught here, before a supplier has to spend their time on it.
Technical Notes
- Adapter plain shank turned down from Ø18.0 mm to Ø15.9 mm. Every other feature on the part is unchanged.
- The body's hex pocket measures 16.020 mm across flats. That, not the 18.498 mm across corners, is the usable aperture for a cylindrical feature.
- Swept at 0.25 mm stations, the superseded pair is blocked from 13.75 mm below the seat, peaking at 197.26 mm3 of interference.
- At Ø15.9 the shank clears the same aperture by 0.120 mm on diameter.
- Adapter mass recalculated from 42.35 g to 39.748 g, being 5.129 cm3 of 17-4 PH. The turn-down removed 0.335 cm3.
- Assembly now calculates to 214.255 g against the 205–220 g target, with 41.899% of mass below the midplane against the 40–48% band.
- Body geometry is untouched. Its exported solid is byte-identical to the previous revision apart from the timestamp written into the file header.
- Insertion was re-verified at 251 stations of 0.10 mm, then again at 801 stations of 0.02 mm through the critical window, clear at every station with no boolean errors.
- The superseded shank was fed to the same check as a negative control and correctly reported blocked, which is what makes the check worth anything.
- Mass and balance were cross-checked on a second, independent geometry kernel: 214.265 g against 214.255 g, and 41.892% against 41.899%.
- The revised pack carries a note that the pocket dimension applies after anodize, with masking offered as an acceptable way to hold it.
- All figures here are results from the digital model. Nothing in this entry has been measured on metal.
Decisions Made
- The correction is made on the adapter alone, to protect the 5.500 mm neck wall on the body.
- Ø15.9 mm is a ceiling set by the fastener head and the pocket aperture, not a preferred number.
- Export is gated on a swept insertion path instead of a static fit at the seat, with a negative control that must fail.
- The superseded revision was withdrawn in writing, with an explicit instruction not to quote or machine from it.
- The unchanged body is reissued at the new revision so the pack cannot be mixed across revisions, while the shop was told that body work already done remained valid and only the adapter needed re-quoting.
Open Questions
- The turn-down leaves a 4.200 mm wall at the fastener head pocket. No minimum wall for this feature is specified anywhere in the engineering record, so the number is unvalidated rather than approved.
- Does the fit survive anodize on the pocket, where coating build-up eats into a clearance measured in hundredths?
- What else in the model is validated only in its final position? The swept check covers this joint. It does not yet cover everything that moves.
- What other classes of defect would currently reach a supplier before they reach a check here?
- Does the calculated 214.255 g hold up on a real assembly, or does it drift the way the earlier mass figure did?
Next Validation Step
The reordered article has not arrived yet. When it does, the first measurement is the shank diameter, because that single reading identifies which revision was actually cut and whether the two parts go together at all.