CMM measurement uncertainty: a practical guide for quality engineers.
Your CMM reads 5.012 mm. The drawing says ⌀5.00 ±0.05. It passes — but does it really? Not until you know the uncertainty of that 5.012. This guide shows where CMM measurement uncertainty comes from, how to estimate it, and how it decides every borderline call you make.
What CMM measurement uncertainty actually means
Every measurement is an estimate, not a perfect truth. Measurement uncertainty is the band around your CMM reading within which the true value is expected to lie, at a stated confidence — almost always 95%. You report it as result ± U, where U is the expanded uncertainty. A reading of 5.012 mm with U = 0.004 mm means the true size is very likely between 5.008 and 5.016 mm.
This matters because a CMM does not hand you the true value. It hands you a number with a fog around it. If the fog is wide compared to your tolerance, you cannot trust the pass/fail decision — and in my 14 years running plants, the borderline characteristics are exactly the ones that come back as customer rejections.
Where the uncertainty comes from
An uncertainty budget adds up independent contributors. For a typical shop-floor CMM measurement, these are the ones that matter:
| Source | Typical contributor | What drives it |
|---|---|---|
| Machine geometry (MPE_E) | Largest at long lengths | Scale, squareness, calibration |
| Probing error (MPE_P) | 1–3 µm | Stylus, qualification, point count |
| Temperature | Often dominant in India | Deviation from 20 °C, part vs scale CTE |
| Fixturing / datum setup | 2–10 µm | Rigidity, datum realisation |
| Part form error | Varies | Roundness, surface finish, sampling |
| Operator / strategy | 2–8 µm | Point density, filtering, alignment |
The single biggest contributor on most Indian shop floors is temperature. The CMM is specified at 20 °C. A workshop at 32 °C, with a steel part whose coefficient of thermal expansion is about 11.7 µm per metre per °C, can drift several micrometres on a 200 mm feature before you touch anything. This is why accredited labs are air-conditioned to 20 ±1 °C.
Building an uncertainty budget
You combine contributors in quadrature — root sum of squares — because they are independent. The recipe follows ISO/IEC Guide 98 (the GUM):
- List each source and estimate its standard uncertainty u_i. Convert any "± limit" Type B value to a standard uncertainty by dividing by √3 for a rectangular distribution.
- Combine: u_c = √(u_1² + u_2² + … + u_n²).
- Expand: U = k × u_c, with coverage factor k = 2 for roughly 95% confidence.
A worked example for a 100 mm length on a workshop CMM:
| Source | Limit / spec | u_i (µm) |
|---|---|---|
| Machine (MPE_E at 100 mm) | ±2.8 µm | 1.62 |
| Probing | ±2.0 µm | 1.15 |
| Temperature (±3 °C, steel) | ±3.5 µm | 2.02 |
| Fixturing | ±3.0 µm | 1.73 |
| Operator / strategy | ±2.0 µm | 1.15 |
Combined: u_c = √(1.62² + 1.15² + 2.02² + 1.73² + 1.15²) ≈ 3.5 µm. Expanded: U = 2 × 3.5 ≈ 7 µm (0.007 mm). So your 100 mm reading should be reported as value ± 0.007 mm at 95% confidence.
The 4:1 rule and why it protects you
The classic test accuracy ratio says the measurement uncertainty should be no more than a quarter of the tolerance you are verifying. For a ±0.05 mm tolerance (0.10 mm band), you want U ≤ 0.0125 mm. Our 0.007 mm budget passes comfortably. For a tight ±0.005 mm feature, the same CMM would fail the 4:1 rule — you would need a better environment or a more capable machine.
Uncertainty and gauge capability are two views of the same problem. Run a Gauge R&R study with our calculator to quantify how much of your observed variation is the measurement system rather than the part. A %GRR under 10% is good; over 30% means the system cannot be trusted for that tolerance.
Uncertainty and the pass/fail decision
ISO 14253-1 governs how uncertainty enters conformance decisions. The key idea is the guard band. Instead of accepting any reading inside the tolerance, you shrink the acceptance limits inward by the expanded uncertainty U:
- Accept only if the reading lies inside (tolerance limit − U).
- Reject only if it lies outside (tolerance limit + U).
- Uncertain zone within ±U of the limit — neither proven conformant nor proven non-conformant.
Take ⌀5.00 +0.05/−0.05 with U = 0.007 mm. A reading of 5.049 is inside tolerance, but it sits within U of the upper limit, so guard-banding flags it as not proven conformant. By default the supplier carries the uncertainty: you only accept what you can prove good. This is exactly why borderline FAI characteristics should be noted, not silently passed.
How to reduce CMM uncertainty
- Fix the environment first. Temperature usually dominates. Soak parts to room temperature, get as close to 20 °C as you can, and avoid measuring a part fresh off a hot machine.
- Qualify the probe properly. Requalify the stylus with the calibration sphere before a critical job; a stale qualification inflates probing error.
- Add points on form features. Three points define a circle but tell you nothing about roundness. Use enough points to capture the real form.
- Fixture to the datums. Realise the datum reference frame the drawing specifies, rigidly, so the alignment matches the design intent.
- Tackle the biggest contributor first. The budget tells you where to spend effort — halving a small term is wasted work.
Common mistakes
- Quoting MPE_E as uncertainty. It is one term in the budget, not the whole answer.
- Ignoring temperature. The most common and most expensive omission on a non-air-conditioned floor.
- Passing borderline readings silently. A reading 1 µm inside the limit with 7 µm uncertainty is not a clean pass — guard-band it.
- Under-sampling form features. Too few points understate roundness and cylindricity error.
- Never running a Gauge R&R. Without it you are trusting a measurement system you have never validated.
For tolerance interpretation behind these decisions, see our guides on ISO 286 fits and tolerances and position tolerance with MMC. For ready-to-use inspection forms, browse the MetricMech templates library.