Gage R&R
Every operator measures every part, more than once. The study splits what you measured into the gauge, the operators and the parts, and says whether the gauge can be trusted on this dimension — before a Cpk or any other study leans on it.
Readings by operator and trial
Parts
Operators
Trials
Tolerance (USL − LSL)
| Op · trial | Part 1 | Part 2 | Part 3 | Part 4 | Part 5 | Part 6 | Part 7 | Part 8 | Part 9 | Part 10 |
|---|---|---|---|---|---|---|---|---|---|---|
| A · 1 | ||||||||||
| A · 2 | ||||||||||
| A · 3 | ||||||||||
| B · 1 | ||||||||||
| B · 2 | ||||||||||
| B · 3 | ||||||||||
| C · 1 | ||||||||||
| C · 2 | ||||||||||
| C · 3 |
Enter every operator’s readings on every trial for at least two parts.
- Each row is one operator’s pass over every part.
- Measure in a random order, blind to earlier readings.
- Enter the tolerance to rate the gauge against the drawing.
How to run this study
Find out how much of the variation you measure comes from the measurement system itself — the gauge (repeatability) and the people using it (reproducibility) — before trusting a Cpk, a gate-seal weight, or any study built on a dimension. A gauge that eats a third of the tolerance makes every later number about the gauge.
Procedure
- 01Check the gauge resolves at least a tenth of the tolerance. A gauge that reads in steps as large as the spread cannot pass, whatever the operators do.
- 02Select ten parts that represent the full range of the process, not ten consecutive parts. Number them discreetly.
- 03Operator A measures all ten parts in a random order and records each reading against its part number. Operators B and C do the same, without seeing each other’s results.
- 04Repeat for the second and third trials, re-randomising the order each time so no one remembers a reading.
- 05Enter each operator’s readings, one row per trial, under the part numbers — the layout of the AIAG data sheet.
- 06Enter the tolerance to rate the gauge against the drawing. The tool runs the ANOVA, pools the operator × part interaction when it is not significant, and reports %GRR, the repeatability / reproducibility split and the number of distinct categories.