Tolerance Stack-Up on Turned Parts: Why Good Parts Don't Fit

By CHX Metal Engineering Team · Published 2026-11-13

Stack-up is the sum of every tolerance in a chain. A shaft with five features, each held to plus or minus 0.02 mm, can drift 0.2 mm end to end with every dimension in spec - which is why a critical fit needs one controlled datum and one tight call, not five loose ones.

Three good parts and one stuck assembly

Last spring an actuator assembly came back to us in a courier box: a stainless shaft we had made, an aluminium housing bought elsewhere, and a sealed bearing off the shelf. Every part gauged good against its own print. Put together, the shoulder bottomed out 0.08 mm early and the bearing ran tight. The three suppliers spent a week pointing at each other, which is the normal outcome, because each print was individually satisfied and none of them carried a stack note.

We were the ones with the least incentive to be wrong and still took two days and a set of slip gauges to prove it: the housing bore sat at the top of its band, the shaft shoulder sat at the bottom of its band, and the chain between them was never anyone's job. The fix cost one line on the drawing. The week cost everybody more than the parts did.

The arithmetic nobody runs until it bites

On a turned part the chain is usually lengths between shoulders, stacked along the axis. Each length carries its own tolerance, and worst case they all land on the same side at once. Five features, each at ±0.02 mm, is a potential 0.2 mm of drift on the overall envelope — and no single dimension on the print ever shows it, because the print tolerances parts, not fits.

This is why we read an RFQ for the chain before the numbers. If a drawing marks every length the same, the fit the buyer actually cares about has no protection at all: the drift hides in whichever band happens to be loosest. Our general tolerance for unmarked features is ISO 2768-m, and that is a sensible default for a single dimension — it just never was designed to guarantee a stack. Our machining tolerances page covers what the grades mean in practice; this page is about what happens between them.

Worst case or statistics

The statistical answer is to stop adding bands linearly. If the processes are independent and roughly centred, the combined drift grows with the square root of the count, not the count itself, and a five-feature chain behaves closer to ±0.045 mm than ±0.1 mm. That arithmetic is legitimate and we use it — but only when it is true. It needs volume, centred processes, and features that do not share a setup error.

A one-piece prototype is none of those things. Neither is a short run where every part comes off the same bar in the same collet, where the errors are correlated by construction. I once argued this with a customer's quality engineer for most of an afternoon and we settled it the honest way: we ran ten parts, plotted the shoulder-to-shoulder distance, and the distribution sat where worst case said it would, not where the spreadsheet hoped. On small lots I plan for worst case and treat statistics as a promise about the future, not a fact about the batch.

Where the one tight call belongs

The practical fix is rarely tightening everything. It is naming the functional fit — the bearing seat, the sealing face, the gap the nut closes across — and putting the tight number only there, against a named datum, so inspection measures the same thing the assembly feels. Everything else stays in the general band. Over-tolerancing a whole shaft costs money on every operation and still leaves the real question unanswered, because the stack is a property of the chain, not of any single diameter.

On our floor the critical features are held to ±0.005 mm and the rest falls back to ISO 2768-m unless the drawing says otherwise. That split is deliberate: one defended number beats five vague ones, and it is cheaper. Where a fit is genuinely uncertain, the minimum order is one piece — a single part with a first-article check, in 3–7 days, answers the stack question before a batch is committed.

What we check before quoting a stacked part

Three questions, in order: which dimension actually closes a fit, what datum it is measured from, and whether the chain around it is bounded anywhere on the print. If the drawing answers none of the three, the quote comes back with questions attached instead of just a price — the same habit that caught the actuator above, and cheaper for everyone than finding it at assembly. Our tolerance capability hub lists what we hold by process and machine.

The plant behind this runs 220 machines in Shajing, Bao'an, Shenzhen — 100 machining centres, 30 of them 3+2, 50 CNC lathes and 70 Tsugami Swiss-type lathes — across 5,000 m² with 100+ staff, producing up to 5 million pieces a month on repetitive work. Production runs 7–15 days once a design is settled. Terms are EXW Shenzhen only - we are happy to recommend reliable freight forwarders.

Thirteen years in, the drawings that arrive with a stack note are still the minority. The assemblies that come back in courier boxes are fewer each year. The difference is one line on a print, and it is usually the buyer who writes it.

Machining capacity behind this article

Address: Building B, 3rd Floor, Haosan Maozhoushan Industrial Zone, Buyong Community, Shajing Street, Bao'an District, Shenzhen  |  Tel: +86-755-27427365  |  Email: mc@chxmetal.com

Frequently asked questions

What is tolerance stack-up on a turned part?

It is the accumulation of individual tolerances along a chain of dimensions, usually shoulder-to-shoulder lengths on the axis. Each dimension can be inside its own band while the overall envelope drifts by the sum of the bands, which is how an assembly interferes with every component individually in spec.

When is statistical stack-up (RSS) safe to use?

Only with volume, roughly centred processes and independent features. On a one-piece prototype or a short run where all parts come from the same bar and setup, the errors are correlated and the worst-case sum is the safer plan. We treat RSS as a high-volume assumption, not a default.

How tight should the critical dimension be called?

Only as tight as the fit needs, against a named datum. We hold critical features to +/-0.005 mm; everything unmarked falls to ISO 2768-m. Tightening every feature adds cost to every operation without protecting the stack, because the drift lives in the chain, not in one diameter.

Can a fit be proven before committing a batch?

Yes. The minimum order is one piece: a prototype with a first-article check runs in 3-7 days, so the shoulder-to-shoulder distance or bore fit is measured on a real part before production of 7-15 days is scheduled.

Does tightening everything solve stack-up?

Almost never. It raises the price on every operation and leaves the functional fit undefined, since no datum says which measurement matters. One tight, datumed call plus a general band on the rest costs less and answers the assembly question directly.

Send your drawings for a DFM review and a quotation

STEP, IGES, DXF, DWG or PDF — note the material, quantity and finish. Quotes usually go back within one working day. Email mc@chxmetal.com or use the quote form.