7.2 Gauge Repeatability & Reproducibility (GR&R)

Gauge Repeatability & Reproducibility (GR&R)

Quick Answer: Gauge Repeatability and Reproducibility (GR&R) is the primary statistical study used in the Six Sigma Measure phase to evaluate measurement system precision for continuous data. Measurement variation is partitioned into Repeatability (Equipment Variation, EV), which captures variation from repeated measurements by one operator on the same part with the same gauge, and Reproducibility (Appraiser Variation, AV), which captures variation between different operators. Total Gauge R&R is $\sigma^2_{GRR} = \sigma^2_{EV} + \sigma^2_{AV}$. Industry benchmarks classify a system as acceptable if %GR&R is under 10%, conditionally acceptable between 10% and 30%, and unacceptable if over 30%. In addition, the Number of Distinct Categories ($ndc$) must be 5 or greater. Qualitative inspection is evaluated through Attribute Agreement Analysis and Kappa statistics. Independent CSSYB study guide by OpenExamPrep.

Components of Measurement System Precision: EV and AV

While calibration and bias studies address whether a gauge is centered on the true master value (accuracy), Gauge Repeatability and Reproducibility (GR&R) evaluates whether the system produces consistent readings upon repeated use (precision).

Total measurement system variation ($\sigma_{GRR}$) is partitioned into two distinct components:

$$\sigma^2_{GRR} = \sigma^2_{repeatability} + \sigma^2_{reproducibility} = \sigma^2_{EV} + \sigma^2_{AV}$$ $$\sigma_{GRR} = \sqrt{\sigma^2_{EV} + \sigma^2_{AV}}$$

Repeatability: Equipment Variation (EV)

Repeatability represents the inherent variability of the measurement device. It is defined as the variation observed when one operator measures the same part multiple times using the same gauge in the same environment:

• Root Causes of High EV: Mechanical play, loose bearings, worn indicator dials, electrical sensor noise, or flexing fixtures.
• Corrective Actions for High EV: Rebuilding or servicing the gauge, improving part clamping rigidity, and reducing mechanical vibration.

Reproducibility: Appraiser Variation (AV)

Reproducibility represents the human and procedural variability between operators. It is defined as the variation observed among the average measurements made by different operators using the same gauge to evaluate the same parts:

• Root Causes of High AV: Ambiguous operational definitions, differing clamping force, improper probe angles, or inconsistent visual alignment.
• Corrective Actions for High AV: Developing clear Standard Operating Procedures (SOPs), conducting operator training, and implementing visual fixturing to standardize part positioning.

---

Designing a Variable Gauge R&R Study

A variable Gauge R&R study must be carefully structured to isolate equipment variation from appraiser variation across a representative production sample.

Standard Study Protocol

A typical study, following the AIAG Measurement Systems Analysis manual, uses:

• 10 Parts ($p = 10$): Selected from standard production to represent the entire process variation spread, including parts near nominal and near both specification limits.
• 2 to 3 Operators ($o = 2\text{ or } 3$): Trained technicians who routinely operate the equipment in daily production.
• 2 to 3 Trials ($r = 2\text{ or } 3$): Each operator measures every part 2 or 3 separate times.
• Blinded and Randomized Order: Parts must be secretly coded so operators cannot identify parts or recall previous readings. Measurement order must be fully randomized across trials.

Calculation Methods: Average and Range vs. ANOVA

Two mathematical approaches calculate GR&R components:

1. Average and Range ($\bar{X}-R$) Method: A traditional approach using range calculations and statistical correction factors ($d_2, K_1, K_2$) to estimate $\sigma_{EV}$ and $\sigma_{AV}$. While simple for manual calculation, it assumes no interaction between operators and parts.
2. Analysis of Variance (ANOVA) Method: The preferred modern method. ANOVA breaks down total variance into Part-to-Part, Operator (AV), Equipment (EV), and the Operator-by-Part Interaction. A significant interaction reveals that specific operators measure certain part geometries differently, highlighting targeted training or fixturing issues.

---

%GR&R Acceptance Criteria and Interpretation

Total measurement standard deviation ($\sigma_{GRR}$) is expressed as a percentage of total process variation (%Study Variation) or specification tolerance (%Tolerance / P/T ratio):

$$\%GR\&R = \frac{\sigma_{GRR}}{\sigma_{total}} \times 100\% \quad \text{and} \quad \%P/T = \frac{6 \times \sigma_{GRR}}{\text{USL} - \text{LSL}} \times 100\%$$

AIAG Acceptance Thresholds

Number of Distinct Categories ($ndc$)

The Number of Distinct Categories ($ndc$) quantifies the measurement system's discrimination relative to process spread:

$$ndc = 1.41 \times \left(\frac{\sigma_{part}}{\sigma_{GRR}}\right)$$

The resulting value is truncated to an integer:

• $ndc \ge 5$: Acceptable measurement system. The gauge has sufficient discrimination to monitor process variation and support control charting.
• $2 \le ndc \le 4$: Conditionally acceptable for coarse grouping (e.g., low, medium, high), but inadequate for estimating process capability ($C_p, C_{pk}$).
• $ndc < 2$: Unacceptable. The gauge acts only as an attribute go/no-go checker and cannot distinguish part-to-part variation.

---

Attribute Measurement System Analysis (Attribute Agreement Analysis)

When inspection is qualitative (e.g., visual inspection for cosmetic flaws, solder joint pass/fail, or invoice auditing), Attribute Agreement Analysis assesses inspection consistency.

Study Design and Metrics

• Study Structure: 30 to 50 parts with known reference standards, including borderline conforming and defective units, evaluated by 2 to 3 inspectors in randomized order across 2 to 3 trials.
• Within-Appraiser Agreement (Repeatability): Consistency of an individual inspector across repeated evaluations of the same parts.
• Between-Appraiser Agreement (Reproducibility): Agreement among all inspectors evaluating the same parts.
• Appraiser vs. Standard Agreement (Accuracy): Percentage of inspector evaluations matching the certified master standard.

Cohen's and Fleiss' Kappa ($\kappa$) Statistics

Because random chance produces some agreement, attribute systems rely on Kappa ($\kappa$):

$$\kappa = \frac{P_{observed} - P_{chance}}{1 - P_{chance}}$$

• $\kappa > 0.75$: Strong to excellent agreement beyond chance; the visual inspection system is acceptable.
• $0.40 \le \kappa \le 0.75$: Moderate agreement; requires boundary samples ("golden boards") and retraining.
• $\kappa < 0.40$: Poor agreement; the inspection process is unreliable and requires immediate standard overhaul.