Engineering Tools

Gage R&R Calculator

Measurement system repeatability and reproducibility from a parts x operators x trials study, AIAG Average-Range method.

Enter measurements for 2-10 parts, each checked by 2 or 3 operators over 2 or 3 trials, to break down measurement variation into equipment and appraiser components.

Splitting measurement noise from part-to-part reality

A gage never reads a perfectly true value — every reading mixes the part's actual dimension with noise contributed by the measurement process itself. Before a Cpk study, an SPC chart, or a go/no-go inspection can be trusted, that noise needs to be sized relative to the real variation between parts. That sizing is what a Gage R&R study produces.

The Average-Range method used here separates two sources of measurement noise. Equipment Variation (EV) is estimated from how much a single operator's repeat readings on the same part disagree with each other — it isolates the gage's own repeatability. Appraiser Variation (AV) is estimated from how far apart different operators' average readings land on the same parts — it isolates disagreement caused by technique or setup rather than the instrument. Combining EV and AV as a root-sum-square gives GRR, the total measurement system variation.

Part Variation (PV) is estimated separately, from the spread of each part's average reading across all operators and trials. Total Variation (TV) combines GRR and PV the same way GRR combines EV and AV. %GRR — GRR expressed as a percentage of TV — is the headline number: it answers "of everything this study saw vary, how much was the gage rather than the parts?" A second number, ndc (number of distinct categories), estimates how many separate part-size groups the system can reliably tell apart; it is derived from PV and GRR, not from %GRR.

When to Use This

  • Qualifying a gage or fixture before it is used to release Cpk/Ppk capability data or feed an SPC chart.
  • Comparing two candidate instruments, or one instrument across two shifts/operators, to see which produces less measurement noise.
  • Investigating a process that looks capable on paper but keeps producing inspection disputes between operators — a high AV relative to EV points at technique or fixturing, not the parts.

Common Mistakes

  • Letting operators see each other's readings, or their own prior reading on the same part, during the study — the Average-Range method assumes independent, blind trials, and the AIAG K-constants are only valid under that assumption.
  • Choosing parts that do not span the process's real spread — if all sampled parts happen to be nearly identical, Part Variation collapses and %GRR looks artificially bad even for a perfectly good gage.
  • Judging the system by %GRR alone when a tolerance is available. %GRR (GRR against total observed variation) and %Tolerance (GRR against the spec width) can disagree, and a system that passes one while failing the other is still an unreliable inspection tool for that spec — this calculator reports the worse of the two as the overall status.
  • Running the study with fewer than 2 or more than 3 trials, or with a part count outside 2-10 — the published K1/K2/K3 constants this method depends on are only tabulated for those counts.

FAQ

Why does the calculator only accept 2 or 3 operators and 2 or 3 trials?

The Average-Range method converts a range of readings into a variation estimate by multiplying it by a constant (K1 for trials, K2 for operators) published in the AIAG Measurement Systems Analysis manual. Those tables only cover 2 or 3 operators and 2 or 3 trials, so counts outside that range have no published constant to use and are rejected before calculation.

What counts as an acceptable %GRR?

Under the AIAG convention this calculator follows: %GRR at or below 10% is acceptable, 10-30% is marginal (may be usable depending on the application and cost of a better gage), and above 30% is unacceptable. When a tolerance is entered, the same bands are applied to %Tolerance as well, and the reported status is whichever of the two bands is worse.

AV came back as exactly 0 — is that a bug?

No. AV is derived from the difference between operators' average readings, adjusted downward for the repeatability noise already captured by EV; if that adjustment would make the underlying value negative, AV is floored at 0 rather than reported as an impossible negative variation. It means the operators' averages were close enough that no reproducibility component could be separated out beyond what repeatability already explains — not that operators read identically.