Tolerance stack up analysis and manufacturing tolerance planning

All three commercial bands are routable — ±0.125 mm (±0.005 in) standard, ±0.025 mm (±0.001 in) precision, ±0.005 mm (±0.0002 in) ground — so the useful question is not what is possible but what is worth buying. Standard practice costs nothing extra; each step tighter adds passes, inspection and eventually a second process. Spend it on the relationships that carry function, and note that a relationship spanning two setups inherits roughly 0.05 mm of positional error, so cutting both features in one setup is often cheaper than tightening either. Name the feature and its measurement condition and the band is confirmed on the quote.

Start from the assembly rather than the part. Identify the clearance or fit that must never fail, then allocate variation across the contributing features rather than applying a tight band everywhere. Concentrating the tolerance on one feature and leaving the rest at ±0.125 mm is almost always cheaper than splitting it evenly across four. And consider absorbing the variation instead: a slot or a larger clearance hole removes the need for the tolerance entirely.

State it explicitly, because the amounts are decisive at precision bands. Anodize adds 5 to 25 µm growing about half outward, powder adds 50 to 100 µm per surface, zinc adds 5 to 25 µm, and electropolishing removes 10 to 40 µm. On a ±0.025 mm fit, any of these is larger than the tolerance. Mark the final condition for each critical dimension and say which features are masked or machined after finishing.

Prescribe it where the method is part of the acceptance requirement, and always check the tolerance is measurable. A gauge is conventionally expected to resolve about a tenth of the tolerance it verifies, so a ±0.025 mm feature wants roughly 0.005 mm of capability. Metrology is also referenced to 20 °C, and aluminium moves about 23.6 µm per metre per °C, so state the temperature and the free-state condition on anything tight.

On the relationships, not the individual dimensions. A bore held to ±0.025 mm is often less important than that bore being perpendicular to its mounting face — and the cheapest way to get that is to cut both in one setup, since each re-fixture adds roughly 0.05 mm of positional error between faces. Mark which relationships must hold together, and the tolerances on the individual features can usually relax.

Worst case sums every contributor at its limit and guarantees the assembly works even when every part sits at its worst permitted value. It also demands the tightest part tolerances, which is the expensive half. RSS takes the square root of the sum of the squared contributors, which returns a looser assembly tolerance because every contributor reaching its limit in the same direction is improbable. That improbability is the trade: a small fraction will not assemble. Use worst case where a single failure is unacceptable or volumes are low enough that the tighter tolerances cost little. Use RSS where volumes are high, contributors are independent and you have real distribution data. Do not use RSS because worst case gave an inconvenient answer.