6.6 Gage R&R for Variable Measurement Systems

Measurement System Analysis (MSA) is a mandatory phase of the Measure stage in DMAIC continuous improvement. Before analyzing process performance, estimating baseline capability ($C_{pk}$), or making capital equipment decisions, a Six Sigma Black Belt must prove that the measurement system is accurate, precise, stable, linear, and capable. If measurement error accounts for a large portion of observed variation, process data cannot be trusted for statistical inference.

---

Components of Measurement System Variation

Total observed variance in process measurements ($\sigma^2_{\text{total}}$) is the sum of actual process variance ($\sigma^2_{\text{part}}$) and measurement system error variance ($\sigma^2_{\text{msa}}$):

$\sigma^2_{\text{total}} = \sigma^2_{\text{part}} + \sigma^2_{\text{msa}}$

$\sigma^2_{\text{msa}} = \sigma^2_{\text{repeatability}} + \sigma^2_{\text{reproducibility}}$

1. Accuracy vs. Precision

• Accuracy (Location Error): The closeness of sample measurements to the true reference standard value established by a master calibration laboratory. Accuracy encompasses bias, linearity, and stability.
• Precision (Width Error): The closeness of repeated measurements of the exact same physical part to each other under specified operating conditions. Precision encompasses repeatability and reproducibility.

2. The 5 Categories of Measurement Error

1. Bias: The systematic difference between the observed average of measurements and the true reference master standard value. Bias is corrected through instrument zero-point adjustment or calibration offsets.
2. Linearity: The change in bias across the entire operating measurement range of the gauge. Linearity measures whether the instrument maintains equal accuracy across small, medium, and large dimensions.
3. Stability (Drift): The change in measurement bias over extended time periods when measuring the exact same reference standard at scheduled intervals. Loss of stability indicates environmental drift, electronic aging, or mechanical wear.
4. Repeatability (Equipment Variation - EV): Variance observed when one operator measures the same part multiple times using the same physical gauge under identical environmental conditions. It represents inherent equipment noise.
5. Reproducibility (Appraiser Variation - AV): Variance observed when different operators measure the same physical part using the same physical gauge in their routine working environment. It represents operational technique differences.

---

Experimental Design for a Continuous Gage R&R Study

A standard crossed continuous Gage R&R study uses a structured factorial setup:

• Parts ($a$): Select 10 physical parts representing the entire operational range of process variation (including parts near specification limits).
• Operators ($b$): Select 2 or 3 operators who routinely perform measurements in daily operations.
• Trials ($n$): Each operator measures all 10 parts 2 or 3 times in fully randomized order.
• Blinding: Operators must be blinded to part numbers to prevent memory bias or conscious rounding.

---

Analysis Methods: ANOVA vs. Average & Range (X-bar & R)

Gage R&R studies are analyzed using two primary mathematical methodologies:

1. Two-Way ANOVA Method (Authoritative Standard)

The ANOVA method decomposes total measurement variance into Part, Operator, Operator $\times$ Part Interaction, and Equipment Error (Repeatability).

• Operational Advantage: Quantifies the Operator $\times$ Part interaction effect ($\sigma^2_{\text{operator} \times \text{part}}$). If interaction is statistically significant ($p < 0.05$), operators measure specific part geometries differently, indicating a need for standardized fixturing and training.

2. Average & Range (X-bar & R) Method

Calculates equipment variation from average range $\bar{\bar{R}}$ across operators using $d_2$ tabular constants. It ignores interaction effects and is less statistically robust than ANOVA.

---

Evaluation Benchmarks for Gage R&R

Gage capability is evaluated using two primary percentage metrics: %GRR and Number of Distinct Categories ($ndc$).

1. Percentage Gage R&R (%GRR)

Defined as the ratio of measurement standard deviation to total process standard deviation:

$\%\text{GRR} = \left( \frac{\sigma_{\text{GRR}}}{\sigma_{\text{total}}} \right) \times 100\%$

Where $\sigma_{\text{GRR}} = \sqrt{\sigma^2_{\text{repeatability}} + \sigma^2_{\text{reproducibility}}}$.

2. Number of Distinct Categories ($ndc$)

Represents the number of non-overlapping confidence groups the gauge can distinguish across process variation:

$ndc = \sqrt{2} \left( \frac{\sigma_{\text{part}}}{\sigma_{\text{GRR}}} \right) \approx 1.41 \left( \frac{\sigma_{\text{part}}}{\sigma_{\text{GRR}}} \right)$

• Benchmark Criterion: $ndc \ge 5$ is required for an acceptable measurement system. If $ndc < 5$, the gauge acts as a discrete binning tool rather than a continuous instrument.

---

Resolution & The 10-to-1 Rule

The 10-to-1 Rule (Rule of 10) states that measurement device resolution (smallest readable scale division) must be at least 1/10th of the process specification tolerance width ($\text{USL} - \text{LSL}$) or process 6-sigma variation ($6\sigma_{\text{part}}$).

• Worked Example: If tolerance width is $0.100\text{ mm}$, the gauge readout must resolve to at least $0.010\text{ mm}$ (preferably $0.001\text{ mm}$).

---

Root Cause Remediation Strategies for MSA Failures

When a Gage R&R study fails, Black Belts isolate whether repeatability or reproducibility is the primary driver:

• High Repeatability Error (EV): Clamping instability, gauge wear, excessive friction, electrical noise, or insufficient device resolution. Remediation: Maintenance, recalibration, or upgrading to optical/digital gauges.
• High Reproducibility Error (AV): Inconsistent operator technique, ambiguous visual alignment standards, or operator parallax error. Remediation: Operator retraining, physical alignment fixtures, and standardized operating procedures (SOPs).

---

Measurement System Stability & Linearity Analysis Workflow

1. Measurement System Stability Protocol

• Select 1 reference master standard part.
• Measure the standard 3 to 5 times per shift across 20 to 30 operating days.
• Plot subgroup averages ($\bar{X}$) and ranges ($R$) on standard control charts.
• Acceptance Criteria: Zero points out of control; no upward or downward trends over time.

2. Measurement System Linearity Protocol

• Select 5 reference standards spanning the entire operating range (e.g., $10\text{ mm}, 30\text{ mm}, 50\text{ mm}, 70\text{ mm}, 90\text{ mm}$).
• Measure each standard 12 times in randomized order.
• Plot Bias vs. Reference Value and perform linear regression: $\text{Bias} = \beta_0 + \beta_1 (\text{Reference Value})$.
• Acceptance Criteria: Slope $\beta_1$ must not be significantly different from zero ($p > 0.05$), and Linearity $% = \left( \frac{|\beta_1| \times \text{Process Variation}}{\text{Tolerance}} \right) \times 100% \le 5%$.'''