Measurement systems analysis
Resolution, bias, linearity, stability, repeatability and reproducibility, how to run a Gage R&R, and how to read it.
Every measurement contains error. Before you chart or judge capability, confirm the measurement system can tell good parts from bad and see real changes.
Properties of a measurement system
- Resolution (discrimination): the smallest step the gauge can show. Aim for at least one tenth of the tolerance or of the process spread.
- Bias: the average difference from a reference value.
- Linearity: whether bias changes across the measuring range.
- Stability: whether bias and variation change over time (check with a control chart of a master part).
- Repeatability: variation when one person measures the same part repeatedly with one gauge (equipment variation).
- Reproducibility: variation between different people (appraiser variation).
Running a crossed Gage R&R
- Choose about 10 parts that cover the full process range, not just good parts.
- Choose 3 operators who normally do the measurement.
- Each operator measures each part 2 or 3 times in random order without seeing earlier results.
- Analyze with the ANOVA method (preferred) or the average and range method.
Reading the results
- %Study variation (%GRR): under 10% acceptable, 10% to 30% may be acceptable, over 30% not acceptable (AIAG).
- %Tolerance: measurement spread compared with the tolerance; use it when the gauge is used to accept parts.
- ndc (number of distinct categories): . 5 or more is needed.
- Charts by operator: an operator with larger ranges needs retraining or a better method; an operator whose averages are offset may read the gauge differently.
Other studies
- Nested Gage R&R: for destructive tests where each part can be measured only once.
- Type 1 study: repeated measurements of one reference part to check bias and repeatability (Cg, Cgk).
- Attribute agreement: for go/no-go gauges and visual inspection, using kappa.
Questions and answers
Our %GRR is 35% but the tolerance is generous. Can we use the gauge?
Possibly for accepting parts if %Tolerance is acceptable, but not for SPC or capability: the gauge hides real process variation. Improve the gauge or method for process control.
ndc is 2. What does that mean?
The gauge can only sort parts into about two groups within the process spread, like a go/no-go gauge. It cannot support a control chart.
