A 20 µm tolerance band is not a property of the metal. Temperature, probe contact, clamping and datum choice each take a slice of it — and the slices add up faster than most drawings assume.
A bracket comes off the machine at 09:40. The supplier's CMM says the bore spacing is 99.996 mm. The part flies to the customer, and their incoming inspection reports 100.009 mm. Both machines carry a valid calibration certificate. Both operators followed the drawing. The tolerance is ±0.01 mm, so one of them accepted the part and the other rejected it.
In most of these disputes nobody is lying and nobody is incompetent. The two measurements are answering slightly different questions, because a ±0.01 mm callout is only meaningful together with a temperature, a fixturing state and a datum scheme. When those three are left to the shop floor, two competent labs will disagree by more than the tolerance allows.
Here is the arithmetic behind that claim.
The band is only 20 µm wide
±0.01 mm means a total permissible spread of 20 µm. To put that next to the instruments: a mid-range bridge CMM is typically specified around MPE_E ≈ 1.8 + L/350 µm, which at 100 mm gives roughly 2.1 µm. A machine tool's positioning repeatability claim is often ±2–3 µm. So the measuring equipment already occupies a meaningful fraction of the band before the part is even clamped.
The MSA convention used across automotive and medical supply chains is blunt about this: measurement system variation should sit at or below 10% of the tolerance, 10–30% is conditional, and above 30% the gauge is not capable of the job. Ten percent of 20 µm is 2 µm. That is a hard target — one that most inspection rooms, measured honestly, do not hit on aluminium parts.
1. Five degrees of room air is twelve micrometres of aluminium
Aluminium expands about 23.6 µm per metre per kelvin. Over a 100 mm feature, a 5 °C difference between the shop and the inspection room moves the measurement by 11.8 µm — more than half the entire tolerance band, produced by nothing but air conditioning.
| Material | CTE (µm/m·K) | Change over 100 mm at ΔT = 5 K |
|---|---|---|
| 6061-T6 aluminium | 23.6 | 11.8 µm |
| 7075-T6 aluminium | 23.6 | 11.8 µm |
| C36000 free-machining brass | 20.5 | 10.3 µm |
| 304 stainless steel | 17.3 | 8.7 µm |
| 1018 carbon steel | 11.7 | 5.9 µm |
| Ti-6Al-4V | 8.6 | 4.3 µm |
Two consequences follow.
First, aluminium is the worst common material for this argument, and aluminium is also where most prototype and enclosure work happens. A part that is comfortably inside tolerance on a 26 °C afternoon can be outside it at 21 °C, with no change to the part at all.
Second, the fix is procedural, not instrumental. ISO 1 defines 20 °C as the reference temperature for geometric specification, so the question is never "is the part good" but "is the part good at 20 °C, measured in a defined state". A working shop rule: let aluminium parts under roughly 200 mm soak for 30 minutes, steel parts for about 60. Parts with a large thermal mass, or parts that were just ground, need longer.
2. An honest uncertainty budget for a 20 µm band
Add up the realistic contributions on a 100 mm aluminium part with two through bores:
| Contribution | Typical magnitude | Under your control? |
|---|---|---|
| CMM MPE at 100 mm (1.8 + L/350) | ~2.1 µm | Partly — a smaller measuring volume helps |
| 3 °C temperature mismatch between part and machine | ~7.1 µm | Yes — soak, and log the room temperature |
| Stylus contact, bending and form error on a bore | 1–3 µm | Partly — qualified stylus, scanning instead of single hits |
| Clamping distortion on the fixture | 1–5 µm | Yes — measure in free state, or clamp exactly as assembled |
| Datum interpretation / operator choice | 2–8 µm | Yes — write the datum scheme on the drawing |
| Surface roughness affecting contact | 0.5–2 µm | Partly — specify Ra and the probing method |
Combined by root-sum-square, those figures land near 9–10 µm — around half the tolerance band, well past the 30% MSA limit. That is the real explanation for the 99.996 versus 100.009 dispute: the two labs did not measure the same thing to different accuracy, they measured different things to similar accuracy.
The useful output of this table is priority, not precision. Temperature dominates, datum choice comes second, and the CMM itself is rarely the problem. Buying a better machine while inspecting parts straight off the shop floor is spending money in the wrong column.
3. Clamping is a measurement, not a preparation step
A thin-wall bracket pulled onto three points on a granite plate will read flat. Released, it springs back and the flatness callout fails at the customer. This is not a metrology subtlety — it is a common root cause of "it was good when it left here".
Two practices remove most of it. Either measure in free state, supported only where the drawing's primary datum contacts the mating surface, or define explicitly that the part is to be measured restrained, and state the clamping force. The one answer that guarantees a dispute is leaving the question unasked.
The same logic applies to the machine: a part machined in a three-jaw chuck and measured off the machine carries the chuck's distortion into the number, and a second operation that removes the material that was holding the shape can move a bore by more than 10 µm in aluminium.
4. The argument is usually about datums, not about numbers
When two labs differ by 15 µm on a hole position, the number itself is almost never the issue. What differs is which face was treated as A.
- A large machined face versus three pads. A full face as primary datum constrains rock and roll over its whole area; three pads constrain it over the pads. On a part with 0.03 mm of face flatness the two readings diverge.
- Position without a datum reference frame. "Distance between hole centres" with no datum callout means each lab picks its own origin. Add -A- | -B- | -C- and the ambiguity disappears.
- Mating clearance ignored. For bolted interfaces, a projected tolerance zone or MMC modifier expresses what actually matters — whether the fastener passes — rather than the theoretical axis location.
- Angular callouts interpreted in two planes. A perpendicularity of 0.02 mm means different things measured over the feature length versus over the part length.
Writing the frame costs a few minutes at quoting and removes an entire category of NCR.
5. What to write on the drawing and the purchase order
For any characteristic at or below ±0.01 mm, six lines do most of the work:
- Reference temperature and soak rule — "measured at 20 °C ± 2 K after stabilisation".
- A complete datum reference frame, with at least three planes, not just a note that the part "sits on its base".
- Free state or restrained state, stated explicitly, with clamping force if restrained.
- Measurement method for key characteristics: CMM scanning versus hard gauge, stylus diameter, number of points per feature.
- Gauge R&R expectation, typically ≤10% of tolerance for the characteristics that gate acceptance.
- An arbitration rule for disputes — for example CMM at 20 °C with the ISO 14253-1 default decision rule, which accounts for measurement uncertainty instead of pretending it is zero.
None of this is exotic. It is the difference between a tolerance that can be verified and a tolerance that can only be argued about.
The conclusion worth carrying into the next RFQ
A ±0.01 mm callout describes a part at 20 °C, in a defined state, referenced to a defined datum frame. Leave any of those three open and the number on the drawing stops being a specification and becomes an opinion — one that two calibrated labs can hold simultaneously and disagree about.
The practical order of attack is: control the temperature, write the datum frame, decide free state or restrained, and only then talk about the measuring machine. In our experience most tolerance disputes are settled in that order, and most of them are settled before the first article is measured, not after.
CHX Precision Metal is a contract machining supplier in Shenzhen, China, founded in December 2013, operating a 5,000 m² plant with 100+ staff and 220 CNC machines — 100 five-axis machining centres, 50 CNC lathes and 70 Tsugami Swiss-type sliding-head lathes. Standard tolerance is ±0.01 mm, tightened to ±0.005 mm on ground features; the quality system is certified to ISO 9001:2015 and ISO 14001:2015. Prototypes ship in 3–7 days, production runs in 7–15 days, with no minimum order quantity above one piece.
If you are working through a tolerance stack on a part that has already caused a measurement argument, send the drawing and the inspection report to mc@chxmetal.com, or call +86 755 2742 7365. More technical notes on five-axis work, Swiss-type turning and DFM are at chxmetal.com.
