GaugeConnection
¿Es usted consultor de calidad? Perfil público, referidos de clientes y acceso gratuito a todas las herramientas.
Más información

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.

Acceptable

Under 10%

The sistema de medición is suitable. Measurement error is small enough not to distort proceso decisions.

Conditionally aceptable

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.

Unaceptable

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.

These are guidelines, not law
The 10 and 30 percent boundaries are conventions that have proven useful, not statistical constants. Customer-specific requirements frequently tighten them - some automotive programmes require under 10 percent for all special characteristics with no conditional band at all. Check the contract before relying on the default bands.

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.

Aceptable con condiciones
El sistema es utilizable, pero solo con una justificación. La aceptación en esta banda es una decisión documentada, no una opción predeterminada.
Qué hacer
Registra por qué se acepta el sistema: criticidad de la característica, costo de reemplazo y controles compensatorios como bandas de protección o un muestreo más frecuente.
Qué pregunta un auditor
¿Quién aprobó la aceptación condicional, con qué fundamento y dónde está registrada esa decisión?
10 % y 30 % son convenciones, no leyes de la naturaleza. Un resultado de 9,8 % y uno de 10,2 % describen sistemas de medición casi idénticos; por eso, trata un valor en el límite como una invitación a revisar ndc y cómo se eligieron las piezas, no como un aprobado o rechazado decidido por redondeo.
Las bandas de 10 % y 30 % mostradas aquí son convenciones ampliamente utilizadas, no una regla de cumplimiento. Los criterios de aceptación de tus piezas los establecen tus propios procedimientos y tu cliente, y deben ser confirmados por una persona calificada.

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.

La misma persona repite la misma medición. Esto es repetibilidad, reportada como EV.
Cada operador es consistente, pero está centrado en un lugar diferente. Esto es reproducibilidad, reportada como AV.
Lecturas repetidas de una pieza
Operador A
Operador B
Operador C
Lee bajo Valor verdadero Lee alto
EV (repetibilidad)
0.0103
AV (reproducibilidad)
0.0084
EV y AV son comparables. Corrige primero el instrumento: un instrumento que no es repetible hace imposible evaluar la capacitación de los operadores.
Dos estudios pueden reportar el mismo %GRR por razones completamente distintas. La separación entre EV y AV indica qué debes corregir, por lo que el total por sí solo nunca es un plan de acción.
Los puntos y las cifras de EV/AV son ilustrativos y se generan únicamente a partir de las posiciones de los controles. No representan un instrumento real ni sustituyen un estudio realizado correctamente y revisado por una persona calificada.

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.

  1. 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.
  2. 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.
  3. Look at the EV / AV split and act on whichever dominates, rather than applying a general improvement effort to both.
  4. 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.
  5. 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.

Rejoining the server...

Rejoin failed... trying again in seconds.

Failed to rejoin.
Please retry or reload the page.

The session has been paused by the server.

Failed to resume the session.
Please retry or reload the page.