Gauge R&R for a vision measuring station: how to run it, and how to read the result

What does a gauge R&R show for a vision station?

Gauge R&R, or measurement system analysis, splits the variation of a measuring system into repeatability (the same part, measured again) and reproducibility (the same parts, measured under different conditions). The result is compared either with the tolerance or with the variation between production parts.

For a vision measuring station the question is the same as for a micrometer: is the station’s own variation small enough, compared with the tolerance, to sort good parts from bad? How the station is built and what sets its accuracy is explained on machine vision dimension measurement.

How should the study be designed?

The usual study uses several parts, measured several times, under several conditions. On an in-line station, the conditions are different from a hand-gauge study:

• Reload every time. Take the part out and put it back for each repeat, as production would present it. A static repeat mainly characterises variation under fixed presentation; it does not replace a study with production reloading.
• Cover the range. Include parts near both tolerance limits, from different batches, tools or cavities.
• Run it on the line. Use the fixture, light, cycle and environment of production, not a bench set-up.
• Fix the recipe. Lock the inspection recipe and calibration before the study, and record their versions.

How do you read the result?

Many quality teams use the guideline in the AIAG Measurement System Analysis manual (4th edition): a %GRR under 10 % is generally considered adequate, 10 % to 30 % may be acceptable for some applications, and over 30 % is considered unacceptable (BPI Consulting summary of AIAG MSA). The same source argues these thresholds can mislead and recommends Dr Donald Wheeler’s classification of measurement systems instead (BPI Consulting). Which criterion applies is in the end agreed between you and your customer, so agree it before the study, not after.

Report the result against the tolerance for sorting decisions, and against the process variation if the station will also be used to monitor the process.

Why add reference parts?

A station can be repeatable and still read every part a little too large. Repeatability does not show that. Measuring the same parts on your CMM or calibrated gauges, and comparing, finds any systematic difference so it can be corrected in calibration. Keep those reference parts: the same comparison, run at set intervals, is the routine check that the station is still right.

The acceptance report for a measuring station then contains three things: the repeatability study, the comparison with your reference measurements, and the calibration procedure. How stations are accepted in general is on how a station is tested and accepted.

Define the factors rather than renaming them operators

NIST’s gauge R&R framework starts with the measurement process and the sources of variation to be studied. (NIST e-Handbook, 2.4 Gauge R&R studies) In a camera station, loading method, fixture, shift and temperature may matter, but they are not automatically interchangeable with the operator factor in a conventional crossed study. Define which factors are crossed with parts, which are nested and which cannot be randomised. Use an analysis appropriate to that design; do not place a collection of different conditions into an “operator” column without considering confounding.

Distinguish static image repeats from full measurement repeats. Repeatedly measuring an unmoved part helps characterise image and algorithm noise under fixed conditions. Removing and reloading the part includes presentation variation. Neither experiment alone describes every source of production variation, and the static spread is not guaranteed to be smaller in every finite dataset. Keep the study types separate and use their comparison diagnostically. NIST’s repeatability section discusses estimating and pooling variability under specified conditions. (NIST e-Handbook, 2.4.4.1 Analysis of repeatability)

Select parts that cover the intended measurement range and include conditions relevant to the decision boundary. An artificially narrow group of nearly identical parts can make a process-relative percentage look poor even when the tolerance-relative result answers a different question. Conversely, choosing an excessively wide range can make process-relative performance look flattering. State the part selection and denominator so the percentage is interpretable.

A worked tolerance-relative calculation

Hypothetical arithmetic example: suppose a suitable study estimates an independent repeatability standard deviation of 0.006 mm and a between-condition component of 0.008 mm. Combining those assumed components by root-sum-of-squares gives σGRR = √(0.006² + 0.008²) = 0.010 mm. With a full drawing tolerance width of 0.20 mm and a stated six-standard-deviation convention, the study variation is 6 × 0.010 = 0.060 mm, or 30% of the tolerance. These are invented teaching inputs, not addaScan measurements or an acceptance recommendation.

Do not mix a percentage based on standard deviations with a percentage contribution based on variances. Nor is tolerance-relative GRR the same as GRR relative to total process variation. Name the convention and show the formula in the report. The commonly quoted AIAG-style thresholds discussed earlier are guidelines in their stated context; the customer’s accepted method and risk determine the actual release criterion.

Add bias, uncertainty and a monitoring plan

A station with a small GRR can still have a systematic error. Compare its results with reference values on the same defined features and inspect whether the difference changes with size or position. NIST treats bias separately from variation. (NIST e-Handbook, 2.4.5 Analysis of bias) Calibration can address a demonstrated systematic effect, but the reference itself has uncertainty and the correction must be validated rather than adjusted to a convenient part.

JCGM 100 sections 4–6 describe evaluating and combining uncertainty contributions. (JCGM 100:2008, Guide to the expression of uncertainty in measurement) Gauge R&R can inform that work, but it is not automatically a complete uncertainty budget. Include relevant reference, calibration, environmental and presentation contributions and avoid counting the same effect twice. South Africa’s measurement legislation provides the national framework for measurement standards, not a ready-made tolerance or universal GRR limit. (South Africa: Measurement Units and Measurement Standards Act 18 of 2006)

After acceptance, keep the approved study design, raw data, analysis method, fixture revision and recipe together. Define reference checks after lens replacement, camera movement or fixture repair. A passing historical study does not validate a materially changed station. Use the dimensional-inspection solution for system scope; this article explains how to make the measurement evidence interpretable and reproducible.