Gauge R&R for Checking Fixtures: How to Run an MSA Study That Means Something

A checking fixture that has never been through a measurement system analysis is an opinion, not a gauge. The uncomfortable part is that most fixture studies do get run — and still produce a number nobody should trust, because they were designed to measure the indicator rather than the fixture. This article covers how to structure a Gauge R&R study that reflects what actually happens at the line, how to read %GRR and ndc without fooling yourself, and what the failing numbers are telling you.

A checking fixture is a system, not an instrument

When you run an MSA on a micrometer, the measurement system is essentially the instrument plus the person holding it. A checking fixture is not like that. The reading at the indicator is the end of a chain that begins the moment the operator drops the part onto the locators, and every link in that chain contributes variation. The indicator is usually the smallest contributor in the chain — which is precisely why studying it alone gives such a flattering answer.

• The fixture itself — locator wear, datum pad flatness, clamp repeatability, frame stability.
• Part seating — whether the part is fully home against every locator before the clamps close.
• The indicator or probe — resolution, mounting stiffness, zero drift.
• The operator's method — clamp sequence, clamp force, where and when the reading is taken.
• The environment — temperature, floor vibration, coolant and swarf on the datum pads.

The mistake that invalidates most fixture studies

The single most common error is leaving the part clamped between trials. If an operator clamps once and takes three readings, the study has measured the repeatability of the indicator — a number that will look excellent and mean nothing. Part loading is normally the largest source of variation in a fixture, so a study that excludes it reports on the smallest contributor and then certifies the whole system on that basis.

Designing the study

The AIAG convention is 10 parts, 3 appraisers and 3 trials — 90 readings. The count matters less than the choice of parts. The 10 pieces have to span the real process spread, including parts near both tolerance limits. Ten good parts pulled from one shift will compress the part-to-part variation, and because %GRR is a ratio, compressing the denominator makes an adequate measurement system look unacceptable.

Reading the result: %GRR and ndc

Two numbers come out of the study and both have to pass. %GRR expresses the measurement system's variation as a percentage of the reference spread. The number of distinct categories, ndc, is calculated as 1.41 x (part variation / gauge variation) and answers a different question: how many separate groups this system can actually tell apart. A fixture can scrape past on %GRR and still fail ndc, which means it can confirm that parts differ but cannot reliably sort them.

The middle band is where judgement lives. A %GRR between 10% and 30% may be accepted with documented customer approval, and the decision turns on how critical the characteristic is and what improvement would cost. It is a conditional pass, not a quiet one: if it is not written down and agreed, it will be found in an audit.

Percentage of what? Tolerance vs total variation

This is the detail that most often turns a good-looking report into a misleading one. %GRR can be expressed against the total observed variation of the study parts, or against the tolerance. They answer different questions and can differ substantially on the same data.

• Against total variation — answers "can this system distinguish one part from another?" Use it when the study feeds a process capability argument.
• Against tolerance — answers "can this system decide pass or fail?" That is the question a checking fixture exists to answer, so it is usually the right basis on the shop floor.
• Reporting a number without stating which basis was used — the most common way a report misleads without anyone lying. Always state the basis on the report itself.

European supply chains often ask a different question again. ISO 22514-7 and VDA 5 approach measurement capability through measurement uncertainty rather than through study variance, and report capability indices instead of %GRR. If your customer is a German OEM or tier one, expect the request to be phrased in those terms — the underlying fixture work is the same, but the acceptance arithmetic is not interchangeable.

Go/no-go fixtures need a different study

If the fixture answers a binary question — a pin enters or it does not, a trim gauge closes or it does not — there is no variable reading to analyse and a standard Gauge R&R does not apply. The equivalent is an attribute agreement analysis: the same parts assessed repeatedly by several appraisers, scored for agreement with each other and against a known reference. The design trap here is the mirror of the reload rule. A study built from obviously good and obviously bad parts will show near-perfect agreement and prove nothing; borderline parts have to be included deliberately, because they are the only ones that test the system.

The error nobody logs: temperature

Dimensional specifications are defined at a standard reference temperature of 20 degrees Celsius (ISO 1). A checking fixture, by design, does not live at that temperature — it lives next to the press or the weld cell. Aluminium expands at roughly twice the rate of steel, so an aluminium fixture holding a steel part is a differential expansion problem, and the differential grows with the size of the part. On a long body panel this is not a rounding error.

The practical consequence is simple: record the temperature with the study, and record it again in any dispute. If the study ran at eight in the morning and the disagreement happened at three in the afternoon, temperature may be the entire disagreement, and no amount of re-measuring will resolve it until someone writes the two numbers down.

When the number comes back bad

A failing %GRR is not one problem, it is a family of problems, and the split between repeatability and reproducibility tells you which one you have before you touch the fixture. Work through it in order rather than changing things and re-running.

When not to run a Gauge R&R

An MSA study consumes parts, operators and production time. There are situations where running one is a waste, or where the result will not mean what the report claims it means.

• The fixture is attribute-only — run an attribute agreement analysis instead; a variable study does not apply.
• The characteristic has no meaningful tolerance — there is nothing to judge the result against.
• A known fault has not been fixed yet — studying a fixture with a worn locator measures the wear, and the study is void the moment you replace it.
• The fixture is a one-off prototype check with no production life — the study can cost more than the fixture.
• The part itself is unstable — thin sheet or a flexible moulding that changes shape under clamp force. That variation is real, but it is a part and process discussion, not a measurement system one.

Accepting a new fixture from a supplier

Most of the arguments about fixture capability happen after delivery, because the acceptance basis was never agreed before the build. Settling these six points in the purchase documents costs nothing and removes the argument entirely.

1. 1Agree the acceptance basis in writing before the fixture is built — %GRR against tolerance or against total variation, and the threshold that constitutes a pass.
2. 2Agree who runs the study, on whose parts, and who witnesses it.
3. 3Require full unclamp and reload cycles in the written protocol.
4. 4Require the study temperature to be recorded on the report.
5. 5Require the raw readings, not only the summary — a summary cannot be re-analysed if it is challenged.
6. 6Repeat the study after installation on your own floor. A supplier study proves the fixture left in good condition; it does not prove it works in your environment with your operators.