Gage R&R: Proving a Measurement System
Run a simple gage study, tell repeatability from reproducibility, and decide whether a gauge and its users are good enough for the tolerance.
Table of contents
When two people measure the same part and get different numbers, who is right? A Gage R&R study answers that question with data. It separates the variation in the parts from the variation added by the gauge and the people. If the measuring system is noisy, you may scrap good parts, pass bad ones, or chase problems in a process that was fine all along.
What You Will Be Able to Do
- Explain repeatability and reproducibility.
- Set up a basic gage study.
- Read a percent study result and say what it means.
- Tell where to look when a study fails.
- Run a quick shop check on your own technique.
What Is Being Tested
A measurement system is more than a tool. It includes the gauge, the person, the method, the fixture, the part, and the room. Every reading has two parts:
Reading = true size of the part + measurement error
Gage R&R (repeatability and reproducibility) measures how big that error is.
| Term | Question it answers | Also called | Mostly about |
|---|---|---|---|
| Repeatability | Does one person get the same number on the same part, over and over? | Equipment variation | The gauge |
| Reproducibility | Do different people get the same number on the same part? | Appraiser variation | The people and the method |
Calibration and Gage R&R are different. Calibration asks whether a tool reads true against a standard. A gage study asks how much the readings scatter in real use. A freshly calibrated gauge can still fail a gage study if it is hard to use the same way twice.
Running a Basic Study
The usual setup in the auto industry's Measurement Systems Analysis (MSA) manual is 10 parts, 3 appraisers, and 2 or 3 trials each. Software or a template does the math, but you need to run it properly.
- Choose parts that span the real range of the process, not ten that look alike.
- Number the parts where the appraisers cannot see the number.
- Pick appraisers who really use this gauge on the job.
- Check that the gauge is calibrated and clean, and decide exactly where to measure on each part.
- Have the first appraiser measure all the parts in random order. Record the readings.
- Have the other appraisers do the same, without seeing the earlier results.
- Repeat for the next trial, in a new random order.
- Enter the data and read the results.
Keep the conditions honest. Measuring more carefully than you do on a normal day gives a pass that does not reflect the real job.
Reading the Results
The main result is percent GRR, which is how much of something the measurement error uses up. There are two common versions:
- Percent of tolerance tells you whether the gauge can make good pass or fail calls.
- Percent of study variation tells you whether the gauge can see real changes in the process, which matters for control charts.
The common guidelines are:
| Percent GRR | Meaning |
|---|---|
| Under 10 percent | Generally acceptable |
| 10 to 30 percent | May be acceptable, depending on the use, cost, and risk |
| Over 30 percent | Not acceptable, improve the system |
Customers can set tighter limits, especially on critical features. Check customer requirements first.
Worked example: a shaft has a tolerance band of 0.020 inch. A study finds a measurement system standard deviation of 0.0004 inch. Measurement spread is usually taken as 6 standard deviations, so 6 x 0.0004 = 0.0024 inch.
Percent of tolerance = 0.0024 / 0.020 x 100 = 12 percent
That falls in the middle band. It may be fine for this feature, and the quality engineer decides.
You may also see number of distinct categories (ndc). It says how many different groups of parts the system can reliably tell apart. A common rule of thumb is that it should be 5 or more. If it is 1 or 2, the gauge can barely do more than sort good from bad.
When a Study Fails
The results show where to look.
| What you see | Likely problem | What to try |
|---|---|---|
| Repeatability is high | Gauge is worn or loose, or the fixture is poor | Repair or replace it, improve the fixture, clean the part |
| Reproducibility is high | People measure in different ways | Train, write the method down, mark the spot |
| One person differs from the rest | That person's technique | Coach them, and compare side by side |
| Readings wander by time of day | Temperature | Let parts soak, control the room |
| Resolution is too coarse | Gauge cannot see small changes | Use a finer tool |
Do not blame the appraiser first. High reproducibility usually means the method is unclear. For example, one person measures near the shoulder of a shaft and another measures at its middle.
Other Measurement Checks
Gage R&R looks at scatter. A full analysis can also look at:
- Bias: the average reading compared with the true value.
- Linearity: whether bias changes across the gauge's range.
- Stability: whether the system holds steady over weeks and months, often tracked by measuring a master on a schedule.
For go/no-go gauges there are no numbers to average, so shops run an attribute agreement study. Appraisers judge the same known parts more than once, and the study checks how often they agree with themselves, each other, and the known answer.
A Quick Check on Your Own Technique
You do not need software to test yourself.
- Choose a part and mark one spot.
- Measure that spot ten times, lifting the tool off each time.
- Find the spread, which is the largest reading minus the smallest.
- Compare it with the tool's smallest step. A spread of about two steps is good. On a 0.0001 micrometer, that means about 0.0002 inch or less.
- Ask a coworker to do the same. Compare your averages.
A big spread means your own repeat is poor, so look at your force, your cleaning, and whether you are square. A difference in averages means you and your coworker are measuring differently, so watch each other and work out why.
This is a useful habit, not a replacement for a formal study.
Why It Matters to the Floor
- A noisy gauge can make a stable process look out of control. Operators then adjust a good process and make it worse.
- It can hide a real shift that you need to see.
- Parts near a limit may pass on one reading and fail on the next. People end up re-measuring until they get an answer they like, and that is a data integrity problem.
If you notice that readings change by who takes them, say so. That is valuable information.
Check Your Understanding
1. Three operators each get nearly the same number over and over, but each operator's number differs from the others. Is the issue repeatability or reproducibility?
Reproducibility. Each person repeats well, but they do not match each other. Look at the method.
2. A study shows 38 percent of tolerance. Can you keep using the gauge for this feature as it is?
Not as it stands. Over 30 percent is not acceptable. Improve the gauge or method, or use a better tool.
3. Why pick study parts from across the process range?
Because the study has to reflect real part variation. Parts that are all alike can make a sound gauge look poor, or hide how it does at the edges.
4. A gauge has a current calibration, but it fails its gage study. How can both be true?
Calibration shows it reads true against a standard under controlled conditions. The study shows too much scatter in real use, with real parts and real people.
Key Takeaways
- A measurement system includes the gauge, the person, the method, and the environment.
- Repeatability is one person's scatter. Reproducibility is the difference between people.
- Under 10 percent is generally fine, 10 to 30 percent depends on the use, and over 30 percent needs fixing.
- High repeatability points at the gauge. High reproducibility points at the method.
- A quick ten-reading check is a good way to test your own technique.