Gauge R&R for sheet metal measurement systems: scope and limits

A gauge R&R study on a caliper comes back at 9%, comfortably inside the 10% acceptance limit, and the part is signed off. Six months later the line is rejecting 4% of parts that the same gauge passes, and the study is blamed. The study was arithmetically correct. The parts were not: all ten were hand-picked across three lots and spread over 0.8 mm, while the process actually produces a 0.12 mm spread. Measured against that real spread, the same gauge reports 51% and an ndc of 2.

Gauge R&R is the study that separates the variation in the part from the variation in the measurement. It is the second half of a traceability system: calibration proves the gauge is right on the day, and a gauge R&R proves the measurement system can tell one part from another. The standards and compliance hub covers the rest of the cluster.

This page sets out what the study separates, why the denominator decides the answer, how to design a study that survives a customer review, and the features on sheet metal parts where the numbers usually fail.

What does a gauge R&R study actually separate?

The study separates two sources. Repeatability is the variation when one operator measures the same part twice with the same gauge — the gauge's own consistency, including the fixture and the part location. Reproducibility is the variation between operators using the same gauge on the same part. Together they form the measurement system variation, usually written GRR.

What the study does not separate matters just as much. Bias — a gauge that reads 0.05 mm high every time — is invisible to GRR because it does not create variation. Linearity, the change in bias across the measuring range, is also invisible. Stability, the drift over weeks, needs a different check. Those three are covered by calibration and by periodic re-measurement of a known master, which is why a calibration certificate and a GRR study answer different questions and neither replaces the other.

Resolution is the other input. A gauge should resolve to about one tenth of the tolerance band; a caliper with 0.01 mm resolution on a 0.30 mm band meets that, and a tape measure on a ±2 mm callout does not. Resolution is a design property, not a study result, but a study run on a gauge with too coarse a resolution will show excellent repeatability simply because the readings cannot change.

Measurement variation also distorts capability. Observed spread combines the process and the gauge, so a process with a standard deviation of 0.035 mm on a ±0.15 mm band shows a true Cpk of 1.43 and an observed Cpk of 1.35 once a gauge with a 0.012 mm standard deviation is added. The gauge makes the process look worse, never better, and the gap widens as the ratio of gauge to process variation grows. That is the case for using a capable gauge rather than a tolerant one: a supplier reporting Cpk from a weak gauge is reporting a number that understates the process and cannot be improved by working on the press.

The definitions of repeatability and reproducibility come from ISO 5725, which is written for interlaboratory experiments on a measurement method. A shop study borrows the definitions and applies them inside one plant, with operators in place of laboratories. That is a legitimate adaptation, and it is worth saying so in the report: a customer who reads ISO 5725 and expects laboratories will otherwise question the design.

The denominator decides the answer

The same GRR number can be expressed against the tolerance or against total variation, and the two give different verdicts. Take a gauge with a GRR of 0.072 mm at six sigma and a tolerance of 20.0 ±0.15 mm, so a band of 0.30 mm.

Nothing about the gauge changed between the last two rows. The parts did. This is the single most common way a gauge R&R study misleads a buyer, and the fix is procedural: state which denominator was used, report both where possible, and always report ndc.

The acceptance bands used in most OEM programmes are 10% or less to accept, 10–30% conditional with the characteristic still monitored, and above 30% to reject the measurement system regardless of how good the parts are. Those bands are conventions rather than a standard, and the standard that governs acceptance near a limit is ISO 14253-1, which introduces a guard band equal to the measurement uncertainty. A gauge at 24% of tolerance is a gauge that will spend a quarter of the band on ambiguity.

Which sheet metal features usually fail?

The table below is the pattern we see across cut, formed, welded and coated parts. The percentages are of the tolerance band on a typical drawing, not of a selected study spread.

Two rows explain most arguments. The hole position across a bend fails because the caliper measures to whatever edge the operator chooses, and the bend radius makes that edge ambiguous. The fix is not a better caliper; it is a fixture that locates on the two datums and measures the feature in one setup. The bend angle fails because a protractor with 0.1° resolution and 0.5° repeatability cannot judge a ±0.5° callout at all, and the measurement belongs on a fixture with a dial indicator against a known angle.

The stamped part inspection guide covers sampling for press parts, and the progressive die tolerance guide sets out what the tool can hold, which is the number a study should be judged against.

Coating thickness is the one characteristic where within-part variation dominates. A single powder-coated panel can read 60 µm in the middle and 95 µm near an edge, because the electrostatic field concentrates powder on the edge and the film builds there. A gauge R&R study that lets each operator choose the spot will report poor reproducibility, and the cause is the spot rather than the operator. Fix five reading points per part, one per face and one on an edge, write them into the control plan, and the study becomes a measure of the gauge and the technique. The paint shop capability describes the line those readings come from, and the coating failure guide covers what a thin edge film does outdoors.

Running a study that survives a customer review

1. Pick the characteristic that decides fit. One feature per study. A study that averages five dimensions hides the one that fails.
2. Select 10 parts from the process, across shifts or lots, covering the normal spread. Record the lot for each part.
3. Number the parts so they can be identified by the study leader but not by the operator.
4. Use 3 operators who normally do the job, not the quality manager.
5. Randomise the order and hide the previous readings. An operator who can see the last result will converge on it.
6. Run 3 trials per operator, 90 readings in total, and record them as measured.
7. Calculate, then look at the range chart before the percentage: a single trial outside the control limit points to a handling problem, not a gauge problem.

The calculation can be the average-and-range method or an ANOVA. The range method is easier on a shop floor and adequate for 10 × 3 × 3. ANOVA separates the operator-by-part interaction, which matters when one operator loads a part differently from another, and that is common on fixtures with a clamp that has to be felt rather than seen.

If the result is above 30%, the useful next step is to find the source rather than repeat the study. Compare repeatability and reproducibility: high repeatability points at the gauge, the fixture or the part location; high reproducibility points at training, the measuring point or the clamping force. The calibration and traceability guide covers the instrument side.

What Ganesh runs, and where a study fits

Ganesh works inside ISO 9001:2015 and ISO 14001:2015 across five units, with first article inspection and PPAP available where the programme requires them. On repeat parts, the measurement system is usually a checking fixture built in the in-house tool room, which is the same shop that builds press tools, progressive dies and welding fixtures.

That choice is deliberate. A fixture that locates on the datums and presents the feature to a dial indicator turns a 35% caliper study into a 10% fixture study, and it does so by removing operator judgement rather than by buying a more expensive instrument. The fixture's own pins and pads wear, so the study has to be repeated on a defined interval and after any repair.

For press parts, the stamping capability describes the press range from 20 T to 300 T and the 250 T punching station, and the working tolerance of ±0.5 mm on custom parts is the band the standard measurement route supports. Tighter features are handled by agreement with a named method and a named gauge, which is exactly the input a gauge R&R study needs.

What to put in the drawing and the inspection plan

• The characteristic and its tolerance band, so the gauge can be chosen with an uncertainty under a quarter of the band.
• The datum scheme and the measurement point, especially on a formed flange.
• The gauge or fixture for each characteristic, named rather than left to the inspector.
• The study requirement: 10 parts, 3 operators, 3 trials, tolerance denominator, ndc of 5 or more.
• The re-study interval, and the trigger for an early re-study after a fixture repair.
• Whether the fixture ships with the tooling and who owns its maintenance.
• The record to include in the PPAP or first article pack.

The quotation request takes the drawing and the inspection requirements together; naming the study at enquiry is what lets the fixture be designed into the route rather than added after the first rejection.

When a gauge R&R is the wrong study

Four cases. A go/no-go plug gauge produces attribute results, so the study is an attribute agreement analysis, not a GRR. A destructive test, such as a weld macro or a coating adhesion cross-cut, cannot be measured twice on the same part and needs a nested design. A process that is not yet in statistical control has no stable part variation to measure, so the control chart comes first. And a feature with no agreed measurement point will fail any study until the drawing is fixed, because the operators are measuring different things.

Ganesh does not run a gauge R&R on a customer's instrument, and we do not certify a measurement system for a process we do not run. Where a drawing calls for a study on a feature we cannot gauge without a third-party instrument, that instrument and its study sit with an accredited laboratory. Say so at enquiry, because it changes the record pack and the timing of the first article.

Send the drawing with the characteristics that decide fit and the tolerance bands. The engineering team will return the gauge, the fixture and the study design for each one, and flag any callout that no shop-floor measurement can honestly support.