Every Part Passes, the Assembly Doesn't Fit: Where the Tolerances Add Up
An assembly plant sends a batch back: it will not fit. We check the report part by part — every dimension is inside tolerance, every inspection stamp in place. The only reply from their line is: "Your parts are fine. We just can't build with them."
Both sides believe they are right in a situation like this, because they are not looking at the same thing. The inspection report proves that an individual part passes. What happens on the assembly line is that several parts have to pass together. That second statement appears nowhere in the report.
Start with how tolerance stacks up. A tolerance on a drawing is written for one part: this dimension, within this range, is acceptable. But assembly is not a matter of dropping one part into place. It is usually a chain of parts arranged along one direction — housing, gasket, bracket, cover, one bearing against the next. Each part takes the end of its tolerance that suits it best. Each one is acceptable on its own; stacked into the assembly, they drift to the other end. Worse, assembly does not average anything out. It does not pick the part that happens to sit at mid-tolerance; it takes the most extreme combination exactly as it comes. So a batch can pass inspection part by part and still refuse to assemble. Nobody measured wrongly — the inspection scope simply never included this layer.
A practical question follows: have the mating dimensions been singled out? A drawing is dense with dimensions, and many carry tolerances, but only a few actually decide whether the assembly closes — the height of a boss, the distance from a hole to an edge, the depth of a step. If the drawing review does not mark those few, inspection tends to spread its effort evenly: every dimension gets measured once, and "full inspection passed" looks good on the report. Whether the critical dimensions were measured more densely or tracked over time is something nobody can say. The dimensions that mattered were not watched, and effort went where it was not needed. Passing full inspection and having the key dimensions under control are two different things.
The hardest case is the one that used to fit. The part has not changed, the drawing has not changed, the mould has not moved — but one part in the chain arrived from a new batch or a new supplier, or an upstream process was adjusted, and its dimensional distribution shifted slightly. Every part is still inside tolerance, but its position in the chain has moved, and the accumulated result now reaches the end. This is difficult to trace precisely because nothing on paper changed; the shift is hidden in the movement of a distribution. Only long-term trend records on the mating dimensions reveal it. By the time the assembly fails and the reports are pulled, every part reads "pass" — which is no clue at all.
Then there is the datum. How the part was held during measurement, and which face locates it in the assembly, are often two different things. "Pass" on the report is stated against the measurement datum; "will not fit" on the line is stated against the assembly datum. When the two do not coincide, error migrates from one part to the next. The rule of three-same — same datum, same measurement method, same judgement rule — is how we keep single-part inspection honest. For a multi-part assembly, three-same on each part is not enough; the assembly datum has to come into the picture too. How the part is located in the final product should be reproduced as closely as possible when it is measured. Otherwise the more beautiful the report, the bigger the gap at assembly.
A third group of problems appears only after the part is fitted. Measure it free and the dimensions are fine; tighten a screw or clip it into place and it deforms under load, and the gap appears. Change the assembly sequence or the tightening effort and the same part gives a different result. Layers of different materials add thermal expansion to the mix — plastic against metal, or one plastic against another, each changing by a different amount with temperature. Something that fits snugly at ambient conditions may loosen or jam once the temperature moves. None of this means the dimension was made wrong. It means the part changed state between the inspection bench and the final product. Inspection captured only the instant it spent on the bench.
That is where the argument starts. The part maker lays out the full inspection report: every part passes, so the claim is rejected. The assembly plant pushes over the finished unit that will not close: whatever the report says, it cannot be built. Both sides have evidence, and neither can convince the other — because the starting point is wrong. Single-part acceptance has been treated as a sufficient condition for assemblability. It is only a necessary one. What should have been done at the drawing stage is to break the assembly requirement down onto each part: which dimensions are mating dimensions, in which direction they stack, what share goes to each part, and which datum locates it. If nobody draws up that tolerance-stack allocation before tooling begins, production can only discover the problem by trial fitting — and the fix usually lands on the mould and the delivery date.
So what is the practical route? When a new part arrives, do not start with its own tolerance. Ask three questions first: what does it mate with, along which direction does it stack, and where is it located in the assembly. Write those answers into the drawing review record and mark the key mating dimensions separately instead of burying them among general dimensions. At sample sign-off, do not sign only the single part — fit it with its mating parts and sign an assembly sample to keep as a physical reference for later disputes. In production, do not limit sampling to individual parts; run a periodic fit check with mating parts, which catches problems earlier than measuring a few more singles. This is general experience; specifics follow the technical agreement and assembly requirements. But none of these actions needs to wait for a returned batch.
A part passing inspection only means it cleared its own gate. Assembly is a different examination, and it examines the whole tolerance chain. Tolerance is not drawn on one part; it is drawn along an assembly chain. Jingmu clarifies the assembly relationship at the tooling review, writes key mating dimensions and locating datums into the drawing, and verifies against an assembly sample.