A gage R&R study measures how much of the variation in your data comes from the measurement itself instead of from the parts. Repeatability is the gauge disagreeing with itself when one person measures the same part again and again. Reproducibility is different people getting different answers from the same gauge. A common rule: under 10 percent of the variation is acceptable, 10 to 30 percent may be acceptable depending on the use, and over 30 percent is not acceptable. Also look for at least 5 distinct categories.
Why it matters before anything else
Every capability study, control chart and inspection record is built on measurements. If the gauge adds a lot of noise, good parts get rejected, bad parts pass and your Cpk is wrong in both directions. A gage R&R tells you how much of what you see is real. That is why it comes before the capability study in a PPAP, and why you run it before you trust a control chart.
The two parts
- Repeatability (equipment variation). One person measures the same part several times with the same gauge. The spread in those readings is the gauge’s own noise. Poor repeatability usually points at the gauge: wear, a loose fixture, too coarse a resolution.
- Reproducibility (appraiser variation). Different people measure the same parts with the same gauge. The spread between people is reproducibility. Poor reproducibility usually points at technique, the work instruction or training.
This split tells you where to look. A fix for a gauge problem and a fix for a technique problem are completely different.
How to set up a study
Check the basics first
The gauge is calibrated, the resolution is fine enough (a common rule is one tenth of the tolerance or better), and the fixture holds the part the same way every time.
Pick the parts
The classic study uses 10 parts. They must cover the range of variation you see in production, including some near each limit. Ten identical parts tell you nothing.
Pick the people
Three operators who normally use the gauge, not the best one.
Number the parts and hide the numbers
Operators should not know which part is which or see their earlier readings.
Measure in random order
Each operator measures each part three times, with the order shuffled each round. That gives 90 readings.
Record every reading to the gauge’s full resolution
Do not round. Rounding hides the very noise you are trying to measure.
Analyze it
Use software or a workbook. The ANOVA method is usually preferred, because it also shows whether operators and parts interact.
A made-up result, and how to read it
The bore gauge, tolerance 0.100 mm. These are example numbers for reading the output.
| Measure | Value | What it says |
|---|---|---|
| Repeatability (equipment variation) | 0.0021 mm, 19.5 percent of total variation | The gauge’s own noise |
| Reproducibility (appraiser variation) | 0.0012 mm, 11.1 percent of total variation | Differences between operators |
| Gage R&R total | 0.0024 mm, 22.4 percent of total variation | Between 10 and 30: marginal |
| Part variation | 0.0105 mm, 97.4 percent of total variation | The real differences between parts |
| Number of distinct categories | 6 | Above 5: the gauge can sort parts into enough groups |
| Gage R&R as a share of the tolerance | 14.5 percent | Judged against the tolerance instead of the process spread |
Read it from the top. Total gage R&R is 22.4 percent, in the gray zone. Repeatability is larger than reproducibility, so the problem is mostly the gauge and not the people. Retraining operators would waste a month. Check the gauge: resolution, wear, the fixture. Fix that and run the study again.
Software and handbooks differ on the multiplier used for the spread, 5.15 or 6. The percent of study variation is a ratio, so it does not change. The percent of tolerance does. Pick one method and keep using it.
The 10 and 30 percent rule
| Result | Common reading |
|---|---|
| Under 10 percent | Acceptable |
| 10 to 30 percent | May be acceptable depending on the use, the cost of the gauge and the cost of mistakes. Improve it if you can |
| Over 30 percent | Not acceptable. Fix the measurement system before you trust the data |
| Distinct categories under 5 | The gauge cannot separate parts well enough for control or capability work |
Your customer may set its own limits, and some applications, such as a characteristic with a very tight tolerance, need much better than 10 percent.
Mistakes in running a study
- Using only good parts. Parts from a narrow range make the gauge look worse in percent, and hide how it behaves near the limits.
- Letting operators see the part numbers. People remember the last reading. The result becomes a memory test.
- One operator measuring the parts three times. That is a repeatability study at best. It cannot show differences between people.
- Rounding readings. You lose the noise you set out to measure.
- Confusing it with calibration. Calibration checks the gauge against a standard. A gage R&R checks the whole measurement system in use. You need both.
- Accepting a 40 percent result because it is all you have. Every decision built on that gauge is shaky. Fix it or add a better gauge.
- Never repeating it. Repeat it after repairs, after a gauge change and at a set interval.
For go/no-go and visual checks, where the answer is pass or fail, an attribute agreement study is used instead. The Gage R&R Calculators workbook covers the variable-data study, and the Calibration Log keeps the calibration side current.
Questions people ask
What is the difference between gage R&R and calibration?
Calibration compares a gauge to a known standard. Gage R&R measures the repeatability and reproducibility of the whole measurement system in use, including the operators and the method.
How many parts, operators and trials do I need?
The classic study uses 10 parts, 3 operators and 3 trials each, which is 90 readings. Your customer may require something different.
What is a good gage R&R percentage?
Commonly, under 10 percent is acceptable, 10 to 30 percent may be acceptable depending on the application, and over 30 percent is not. Look for 5 or more distinct categories too.
What if my gage R&R fails?
Look at which part is large. High repeatability points at the gauge or fixture. High reproducibility points at technique or training. Fix that, then run the study again.
Do I need a gage R&R for every gauge?
You need one for the gauges that control characteristics that matter, especially special characteristics and anything in a PPAP package. Follow your customer’s requirements.