Key takeaways
- A measurement system includes the gauge, the fixture, the operator, the method and the environment.
- Gauge R&R separates repeatability, the equipment, from reproducibility, the appraisers.
- Report %GRR against the tolerance when the tolerance is what the decision depends on.
- A calibrated gauge can still be unsuitable for a tight tolerance.
Why measurement systems are studied
Every measurement carries variation. The gauge, the fixture, the operator, the method, the temperature and the part itself all contribute. Measurement system analysis quantifies that variation so a pass or fail decision is based on evidence rather than on the confidence of the person holding the caliper.
A gauge can be calibrated and still be unsuitable. Calibration confirms that the gauge reads a master within its own tolerance. It does not confirm that the gauge can separate good parts from bad ones at the tolerance on the drawing.
A study also protects the supplier. If a part is rejected near the limit, the supplier can show whether the gauge or the process caused the decision. Without that evidence, the argument is about opinion, and the buyer usually wins.
The five variable studies
Bias is the difference between the average of repeated readings and a reference value. Linearity is how that bias changes across the measuring range. Stability is how the bias behaves over time, checked against the same master.
Repeatability is the variation when one operator measures the same part with the same gauge several times. Reproducibility is the variation between operators using the same gauge and the same parts. A gauge R&R study combines the two and reports them as one number.
The studies are not alternatives. Bias and linearity are usually checked when a gauge is introduced or repaired. Stability is a periodic check. Gauge R&R is the study that most programmes ask for before a part is approved.
Setting up a gauge R&R
A common variable study uses ten parts, three operators and three trials, which gives ninety readings. The parts should span the process range, not ten good parts from the same hour. If every part is near the nominal, the study cannot see how the gauge behaves near the limits.
Randomise the order and keep the operators blind to the part number. Use the same fixture and the same measuring position that production will use. If the study uses a different clamp or a temperature controlled lab, it describes that setup, not the line.
Record the gauge, its resolution, the fixture, the temperature and the calibration status. If a reading looks like an outlier, keep it and investigate it. Removing a reading because it looks wrong is how a study becomes fiction.
Reading the result: %GRR and ndc
The %GRR is the gauge variation divided by a reference, multiplied by 100. The reference can be the tolerance or the study variation. The two answers are different, and the report should say which one was used. For an inspection decision, the tolerance is usually the right basis.
The number of distinct categories, or ndc, estimates how many levels the measurement system can tell apart. A value of five or more is the usual target. A low ndc means the gauge compresses the parts into a few groups, so a capability study built on it will flatter the process.
The common bands are under 10 per cent acceptable, 10 to 30 per cent conditional and over 30 per cent unacceptable. Those bands assume the study is sound. A clean number from a bad part set is still a bad study.
Choosing the study
The table separates the questions. Use it to select the study before collecting data, because the data collection plan follows from the question.
| Study | Question | Typical method |
|---|---|---|
| Bias | Does the gauge read the reference correctly? | Measure a standard, compare the average with the reference |
| Linearity | Does the bias change across the range? | Measure standards across the range, plot the bias |
| Stability | Does the bias drift over time? | Measure the same master at set intervals, chart the result |
| Repeatability | Does one operator get the same result? | Same part, same operator, repeated trials |
| Reproducibility | Do operators get the same result? | Same parts, several operators, randomised order |
| Gauge R&R | How much of the tolerance is measurement noise? | Ten parts, three operators, three trials |
Attribute agreement
A go or no-go gauge, a visual check or a colour match needs an attribute study. The usual design uses fifty parts, three appraisers and three trials, with a known reference for every part. The study reports agreement within each appraiser, agreement between appraisers and agreement with the reference.
A high within appraiser agreement with a low agreement against the reference is a warning. The inspectors are consistent with each other but consistently wrong, which usually means the limit sample or the written standard is wrong.
Agreement is not the same as accuracy. A team can agree on a wrong limit, so the reference matters as much as the repeatability.
Where studies go wrong
The most common mistake is a part set that does not represent the process. The second is a study run by the quality engineer with the gauge in a quiet office, using a fixture that the line does not have. The third is a report that gives a single %GRR without saying what the denominator was.
Other traps are a gauge with poor resolution, an operator who has practised on the study parts, and a gauge that was adjusted between trials. Each one makes the measurement look better than it is, which is the opposite of the purpose.
Finally, a study is not a one-off. A new operator, a repaired gauge or a moved fixture changes the system. Repeat the study when the system changes, and keep the report with the part approval records.
What to ask a supplier
Ask for the study report, not just the number. It should name the parts, the operators, the trials, the fixture, the gauge and the basis of the %GRR. It should show the raw readings or the analysis output, and it should state the action taken if the result is marginal.
If the gauge is marginal, the supplier can improve the gauge, add a fixture, increase the sample or guard band the tolerance. Guard banding is a commercial decision as well as a technical one, because it rejects parts that are inside the drawing. The customer should agree it in writing.
Calibration is not capability
A calibration certificate says the gauge is within its own tolerance. A gauge R&R says whether it can do the job on this part. Ask for both.
Frequently asked questions
What is an acceptable gauge R&R percentage?
The usual guidance is under 10 per cent is acceptable, 10 to 30 per cent may be acceptable depending on the characteristic, the cost and the risk, and over 30 per cent is unacceptable. The customer or the control plan can set a tighter rule.
Should %GRR use the tolerance or the study variation?
Use the tolerance when the measurement decides whether a part is inside the drawing. Use the study variation when the purpose is to understand the process. Report both if the audience needs them, and always state the denominator.
How many parts are needed for a gauge R&R?
Ten parts, three operators and three trials is the common variable study. The parts must span the process range. For an attribute study, fifty parts with a known reference is the usual starting point.
Does a gauge R&R need to be repeated?
Yes, after a gauge change, a fixture change, a new operator group, a process change or a move to another site. A study is evidence for a setup, not a permanent property of the gauge.
Sources and further reading
- AIAG MSA, Measurement Systems Analysis reference manual, fourth edition
- ISO 14253-1, Inspection by measurement of workpieces and measuring equipment
- IATF 16949, Quality management system requirements for automotive production
Qumek Quality Desk
Quality team
The quality team responsible for first article inspection, PPAP packs, supplier audits and the records shipped with production parts.
Checking a gauge study before approval?
Send the study report and the drawing tolerance. The quality desk will check the basis of the %GRR, the part set and the action plan.
Send the study