Interpreting R&R del sistema de medición resultadoados
Tabla de contenidos
The 10 and 30 percent umbrales
The AIAG MSA manual defines three acceptance bands. GaugeConnection applies exactly these boundaries cuándo it assigns a verdict.
Under 10%
The sistema de medición is suitable. Measurement error is small enough not to distort proceso decisions.
10% to 30%
May be aceptable based on the importance of the characteristic, the cost of the gauge, and the cost of repair. The justification must be documented.
Over 30%
The sistema de medición needs improvement. Data from it should not be used to judge conformance or control the proceso until it is fixed.
The middle band is widely misread as a soft pass. It is not. It is a requirement to justify the decision in writing, considering qué the characteristic controls. Twenty percent on a cosmetic dimension and twenty percent on a safety-critical bore are not the same situation, and an auditor will expect that distinction to have been made deliberately.
Pruébalo: qué cambia al cruzar el 10 % y el 30 %
Arrastra el resultado entre los límites de aceptación. El número no es lo importante: lo importante es la decisión que obliga a tomar.
Percent of tolerancia or percent of estudio variación?
The same GRR valor divided by two different denominators gives two different percentages, and they answer genuinely different questions. GaugeConnection uses tolerancia cuándo you enter one, and total estudio variación cuándo you do not. The report records which basis was used.
Percent of tolerancia
Can this gauge decide whether a pieza conforms to specification?
Use for inspection, sorting, and any conformance decision. This is qué a customer or auditor normally means.
Percent of estudio variación
Can this gauge tell the piezas I actually produce apieza from each other?
Use for SPC, capability work, and proceso improvement, dónde distinguishing real proceso shifts matters more than the specification.
The two can diverge sharply. A tight proceso inside a generous tolerancia will look excellent against tolerancia and poor against estudio variación - the gauge is good enough to pass piezas but not good enough to control the proceso. A loose proceso in a tight tolerancia scómos the reverse.
If a characteristic is used for both conformance and proceso control, compute both and satisfy both. There is no rule requiring you to pick one.
The ndc rule
Number of distinct categories asks cómo many non-overlapping groups the sistema de medición can reliably separate within the observed pieza variación. It is computed as 1.41 times PV divided by GRR, truncated to a whole number. The requirement is at least 5.
The intuition is straightforward: with an ndc of 2 the gauge can only tell you "large" or "small", which is sorting rather than medir. Five categories is the minimum needed for gráfico de controls to behave properly, because below that the sistema de medición quantises the proceso into steps too coarse for the chart to detect real shifts.
When %GRR and ndc disagree
They can conflict, and the ejemplo trabajado in the ANOVA deep-dive does exactly that: 28.63 percent of tolerancia is conditionally aceptable, but ndc is 4 and fails. When they disagree, the more conservative resultado governs. A system that passes %GRR against a generous tolerancia but fails ndc can certify piezas against the specification while being unfit for statistical proceso control.
Reading the component breakdown
The headline %GRR tells you whether there is a problem. The split between EV and AV tells you dónde it is, which is qué detérminoines the corrective action.
| Pattern | What it indicates | Where to look |
|---|---|---|
| EV much larger than AV | The gauge or the method is the problem, not the people | Gauge condition, resolution, fixturing, pieza cleanliness, seating repetibilidad |
| AV much larger than EV | Operators are applying different technique | Work instructions, training, ambiguity about dónde or cómo to measure |
| Significant pieza x operador interaction | Difficulty depends on the pieza, not on the operador alone | Awkward geometry, inconsistent datums, piezas that fixture unreliably |
| Very low PV | A estudio design problem, not a gauge problem | Parts were too similar; re-select samples spanning the real proceso range |
Pruébalo: la repetibilidad y la reproducibilidad se ven diferentes en el gráfico
Cada punto representa a un operador midiendo la misma pieza. Cambia los dos controles y observa qué componente del estudio aumenta.
What to do cuándo a estudio fails
Work through these in order. The early steps are cheap and frequently resolve the problem without touching the gauge.
- Check the estudio design first. Were the piezas genuinely representative of the proceso range? A failing %GRR caused by piezas that were nearly identical is not a sistema de medición finding at all, and re-running with properly selected piezas is the correct response.
- Check resolution. If the gauge increment is coarser than one tenth of the proceso variación, no amount of training will help - the gauge physically cannot resolve qué you are asking it to resolve.
- Look at the EV / AV split and act on whichever dominates, rather than applying a general improvement effort to both.
- If AV dominates, watch the operadores measure before rewriting anything. Reproducibility problems are almost always caused by an unstated assumption in the work instruction that each person resolved differently.
- Re-run the estudio after any change, and keep both reports. The before-and-after pair is the evidence that the corrective action worked.
While you fix it
A failing sistema de medición does not automatically stop production, but it does change qué you can claim. Data already collected with that system is suspect, and any conformance decision made from it may need review. Guard-banding - tightening the acceptance limits by the medición uncertainty - is the usual interim containment, at the cost of rejecting some conforming piezas.
Record the failure, the containment, and the corrective action in the same place as the estudio. An auditor is far more comfortable with a documented failure that was contained and corrected than with a file that contains only passing resultadoados.