Fastener Pattern Stack-Up: A Functional Gauge That Accepts Every Part and an Assembly That Jams
A pump housing in 6061-T6 has six through-holes Ø6.6 +0.12/0 on a 100 x 80 mm rectangular pattern and bolts to a stainless manifold with six M6 tapped holes. The housing drawing gives the hole pattern a position tolerance of Ø0.3 at MMC to datum A (mounting face) and datum B (a bearing bore). The manifold drawing gives its tapped bosses a position tolerance of Ø0.2 at MMC to its own datum face and a different bore. The fastener is a standard M6 bolt, maximum major diameter 6.00 mm including plating. In the first 20 builds, 2 units would not assemble without opening the holes and a further 3 needed force. A functional gauge built to the housing drawing accepts every housing. The housing flange is 20 mm thick, the manifold bosses are 25 mm deep, and both drawings also hold the hole and tapped-hole axes perpendicular to Ø0.1 over their full thickness.
1. Compute the worst-case clearance available for floating-fastener assembly and compare it with the sum of the stated positional tolerances. Is the drawing stack valid as written?
Read the worked answer
Floating-fastener rule: the sum of the positional tolerances on the two parts must not exceed the guaranteed clearance, T1 + T2 <= H_min - F_max.
- H_min = 6.60 mm, F_max = 6.00 mm, so guaranteed clearance = 0.60 mm.
- Stated tolerances: 0.30 + 0.20 = 0.50 mm.
- Apparent leftover: 0.10 mm.
On those numbers the drawings are valid and even have margin, which is why the drawing check and the functional gauge both pass. And they are both wrong to pass, because this calculation is only valid if position is the only contributor - the formula assumes the hole and the bolt axes are parallel at the interface.
They are not. Both drawings separately allow the axes to be perpendicular to Ø0.1 over the full thickness. A hole whose axis is tilted within a Ø0.1 zone has its centre displaced by up to 0.05 mm at the far face, and the mating tapped hole can be displaced by 0.05 mm in the opposite direction. At the interface that is an extra 0.10 mm of relative offset - exactly the leftover. Add any plating in the through-hole (which can take the hole below 6.60) and any thread-axis drift in the tapped boss, and the stack goes negative. That is precisely the observed 25% jam rate.
The second, deeper flaw is the datum scheme: the two parts are located to different bores, so the pattern-to-pattern relationship that actually governs assembly is not captured by either drawing at all. The functional gauge only proves the housing is good to its own datums, so it can never detect this.
2. If the stack is the root cause, why do only 2 of 20 builds jam badly rather than all of them? Explain what the distribution tells you and what the 0.12 mm bonus does.
Read the worked answer
Why not all 20
The stack just computed is a worst-case combination: every contributor at its extreme, in the same direction. Worst case tells you whether interference is possible, not how often it occurs. Real builds draw each contributor from its own distribution - position error on the housing, position error on the manifold, perpendicularity tilt, plating thickness, thread drift, bolt straightness, and the operator's willingness to tap a part home with a mallet. A jam needs several of them to align in the same direction simultaneously, so a small probability per build is exactly what a zero-margin worst case produces. A 10% hard-jam rate plus 15% needing force is consistent with a stack that is at zero margin at worst case and roughly 2 to 3 sigma into interference in practice.
What the bonus tolerance does
The MMC modifier gives bonus equal to the departure of the feature from MMC. A hole at the top of its size range, 6.72 mm, adds 0.12 mm of bonus to its Ø0.3 position tolerance. That bonus is genuinely useful - but it works the wrong way around here. Hole size variations add clearance in the same proportion, so oversized holes help only if the positional errors do not also grow. The builds that jam are typically the ones where the hole came out at the low limit (6.60 or below after plating) and the position error used the bonus. Bonus does not create margin; it transfers it between size and position.
Practical consequence: do not quote a 25% jam rate as evidence that the design is marginal. Quote it as evidence that the design has zero margin at worst case and that the true variation is larger than either drawing admits.
3. Recommend the drawing change and the verification that would let you release the next build. Compare at least two options with numbers.
Read the worked answer
The governing relationship must be restored so that the worst-case stack has positive margin, including orientation.
Option A - tighten position and use projected tolerance zones. Give the manifold Ø0.15 and the housing Ø0.25, and replace the perpendicularity callouts with projected tolerance zones of Ø0.1 projected over the full 25 mm stack height. Then the sum is 0.40 against 0.60 clearance, giving 0.20 mm of worst-case margin, and the orientation error is by construction contained inside the projected zone rather than added on top of it. Cost: modest tightening of the manifold pattern, which is a machined boss and easily held.
Option B - enlarge the housing holes to Ø6.9 +0.15/0. Clearance becomes 0.90 mm against the same 0.50 + 0.10 = 0.60 mm of contributors, giving 0.30 mm of margin with no tightening at all. Cost: bearing area under the bolt head drops and the joint becomes more slip-prone; if the joint carries shear, this is a bad trade unless shoulder bolts or dowels take the shear. For a pump housing subject to vibration and preload loss, I would not choose this alone.
Recommended: A, plus a shared datum scheme. Both drawings must locate the pattern to the same functional feature (the pump bore or the manifold spigot) so that the assembly stack is actually controlled by the drawings. Publishing a common datum reference frame and a projected tolerance zone is the fix that makes the two drawings agree with each other.
Verification before release: a functional gauge is not sufficient and never was - it only tests one part to its own datums. Build a matched assembly check (housing plus manifold plus six bolts, no forcing allowed), run at least 30 builds, and record the assembly force. Then add a CMM check of the pattern-to-pattern relationship using a shared datum simulator. Accept the drawing only when the 30-build sample assembles with zero forced joints.