Gauge R&R
Can the gauge even see the signal you're charting?
Before you trust a measurement, ask whether the gauge can even see the signal. Gauge R&R splits observed variation into the part-to-part variation you care about and the measurement system's own repeatability and reproducibility. If %GRR is high, you are charting noise.
What it measures
Every measurement is the true value plus measurement error, and a measurement-systems analysis (MSA) quantifies how much of what you observe is which:
- Repeatability: variation when the same operator measures the same part repeatedly with the same gauge. This is the equipment's own noise.
- Reproducibility: variation between operators (or setups) measuring the same parts. This is the human-and-method contribution.
- Part-to-part: the real variation between parts, the signal you actually want to measure and control.
The headline numbers: %GRR, the ratio of measurement-system variation to total variation (a common acceptance rule is under 10% good, 10 to 30% marginal, over 30% unacceptable), and the number of distinct categories, ndc = 1.41 × (σ_part / σ_gauge), which should be at least 5 for the gauge to resolve the process into useful levels. A study uses several parts spanning the real range, a few operators, and repeat trials, analyzed by ANOVA to separate the components.
How to read the output
Read %GRR against the variation you are trying to control, not in the abstract. A gauge with 25% GRR might be fine for a wide tolerance and useless for a tight one, because what matters is whether the measurement noise is small relative to the spec or the process spread you need to resolve. A high %GRR has two cures pointing in different directions: high repeatability means fix the equipment (calibration, fixturing, a better instrument); high reproducibility means fix the method (operator training, a clearer procedure, automation of the read). And ndc is the resolution check, below about 5, the gauge cannot tell process levels apart, so any control chart built on it is sorting noise.
A real use case
An electrode-thickness control chart looks alarmingly noisy, points jumping around inside the limits with no stable pattern, and the team is about to chase a phantom process problem. A Gauge R&R study on the thickness gauge tells the real story: %GRR comes back at 45%, almost all of it repeatability, and ndc is 2. The gauge simply cannot resolve the thickness variation the process actually has, so the chart was plotting measurement noise dressed as process drift. The fix is the measurement system, a better-fixtured gauge with finer resolution, not the coating line. Once the gauge can see the signal, the chart settles and the real, much smaller process variation becomes visible and controllable.
Common mistakes
- Charting and reacting to a characteristic before validating the gauge. A high-GRR measurement turns a stable process into a noisy chart and sends teams chasing ghosts.
- Judging %GRR in the abstract instead of against the tolerance or process spread you need to resolve.
- Ignoring the repeatability-versus-reproducibility split, which points at equipment versus method, two different fixes.
- Skipping ndc. A gauge can have an acceptable-looking %GRR and still fail to resolve the process into enough distinct levels to control it.
- Running the study on parts that do not span the real process range, which inflates the apparent part-to-part variation and flatters the gauge.
Validate the gauge before trusting the chart
Niobia runs the measurement-systems analysis from a Gauge R&R study, separating observed variation into repeatability (equipment), reproducibility (operator), and part-to-part by ANOVA, and reports %GRR with the number of distinct categories so you know whether the measurement system can actually see the signal. Because every control chart and capability index downstream inherits the gauge's noise, this is the validity check that keeps the rest of the SPC layer honest, the same place-the-gate-first discipline that spec compliance applies to data quality. A high-GRR result tells you to fix the measurement before reading anything into the process.
Frequently asked
What is the difference between repeatability and reproducibility?
Repeatability is the gauge measuring the same part twice and not agreeing with itself: equipment noise. Reproducibility is two operators (or setups) measuring the same part and not agreeing with each other: method and human variation. The split matters because the fixes differ, equipment for one, procedure or training for the other.
What counts as an acceptable %GRR?
A common rule is under 10% good, 10 to 30% marginal (acceptable depending on the application and cost), over 30% unacceptable. But judge it against the tolerance or process spread you need to resolve, and check the number of distinct categories (ndc should be at least 5) alongside it.
Why run Gauge R&R before SPC?
Because a control chart cannot distinguish process variation from measurement noise. If the gauge contributes a large share of the observed variation, the chart plots that noise and any signal you read into it is unreliable. Validating the measurement system first is what makes the downstream charts and capability numbers trustworthy.