Gage R&R Before SPC: AIAG Acceptance Thresholds, Worked Example, and 5 Fixes When It Fails

Gage R&R is the measurement-system analysis that has to happen before any control chart or capability study is trustworthy. The logic is simple: if your measurement system contributes 40% of the variation you observe, then your process variation estimate is inflated, your Cpk is understated, and your control chart signals half of its alarms on measurement noise instead of real process shifts. You can’t monitor a process with a ruler that varies as much as the thing you’re measuring.

This guide walks the full measurement system analysis workflow: the three components of measurement variation, the study design, the AIAG acceptance thresholds, and a worked example that takes raw data through %GRR and the number of distinct categories. The deliverable is a defensible statement—backed by numbers—that the measurement system is adequate for the tolerance it’s measuring against.

The Three Sources of Measurement Variation

Total observed variation in a measurement decomposes into:

Variance Decomposition $$\sigma^2_{total} = \sigma^2_{part} + \sigma^2_{measurement}$$ $$\sigma^2_{measurement} = \sigma^2_{repeatability} + \sigma^2_{reproducibility}$$

Repeatability (EV, Equipment Variation): variation when the same operator measures the same part repeatedly with the same gage. Captures within-operator noise.
Reproducibility (AV, Appraiser Variation): variation between operators measuring the same parts. Captures between-operator bias.
Part-to-part variation (PV): real differences between the parts being measured. This is the signal you want to see.

A Gage R&R study estimates the first two and compares their combined contribution to either the total variation or the tolerance. If measurement variation is a small fraction of either, the gage is adequate.

AIAG Acceptance Criteria (The Numbers Customers Ask About)

AIAG MSA Thresholds

%GRR = (σ_measurement / σ_total) × 100%, or computed against tolerance as (6 × σ_measurement / tolerance) × 100%.

≤ 10%: measurement system is acceptable for all applications.
10% to 30%: may be acceptable depending on application, cost, customer requirements, and risk.
> 30%: measurement system is not acceptable; improve before proceeding.

Number of Distinct Categories (ndc) = 1.41 × (σ_part / σ_measurement), truncated to an integer.

≥ 5: adequate discrimination.
< 5: measurement system cannot resolve the parts well enough for process control.

The two criteria work together. A measurement system passing %GRR but failing ndc usually has low part-to-part variation relative to measurement noise—capability studies using it will be unreliable even if the percentage metric looks acceptable.

Study Design: Sample Size and Structure

The standard AIAG Gage R&R Crossed study uses:

10 parts representing the typical process spread (not 10 parts from one lot—10 parts covering the range of variation the gage will encounter)
3 operators randomly selected from those who actually measure parts in production
2–3 trials per operator per part (3 is standard)

Total measurements: 10 × 3 × 3 = 90. Parts are measured in random order within each operator’s session; part identity is hidden from the operator to prevent bias. Operators measure independently; their sessions are separated in time or space so they don’t influence each other.

Common Mistake

Using 10 parts from a single production lot. The parts need to span the process variation the gage will encounter in production. If all 10 parts are within 2σ of the process mean, the study’s estimate of part-to-part variation is artificially low, which inflates %GRR and fails ndc. Deliberately include parts from the edges of the process distribution.

Worked Example: A Thickness Gage Study

Setup

Characteristic: Plating thickness on a stamped component
Specification: 0.150 ± 0.010 mm (tolerance = 0.020 mm)
Measurement device: Coating thickness gage, resolution 0.001 mm
Study: 10 parts, 3 operators, 3 trials each = 90 measurements

After running the study and computing variance components (via ANOVA method, the AIAG-preferred approach):

σ_{repeatability} = 0.00120 mm
σ_{reproducibility} = 0.00080 mm
σ_measurement = √(0.00120² + 0.00080²) = 0.00144 mm
σ_part = 0.00450 mm (from within-study estimate)
σ_total = √(0.00450² + 0.00144²) = 0.00472 mm

Percent GRR Against Total Variation

$$\%GRR_{total} = \frac{\sigma_{measurement}}{\sigma_{total}} \times 100 = \frac{0.00144}{0.00472} \times 100 \approx 30.5\%$$

Percent GRR Against Tolerance

$$\%GRR_{tol} = \frac{6 \times \sigma_{measurement}}{tolerance} \times 100 = \frac{6 \times 0.00144}{0.020} \times 100 = 43.2\%$$

Number of Distinct Categories

$$ndc = 1.41 \times \frac{\sigma_{part}}{\sigma_{measurement}} = 1.41 \times \frac{0.00450}{0.00144} \approx 4.4$$

Truncated: ndc = 4.

Interpreting the Results

This measurement system fails AIAG acceptance on all three criteria:

%GRR vs total: 30.5% (needs to be < 30%)
%GRR vs tolerance: 43.2% (needs to be < 30%)
ndc: 4 (needs to be ≥ 5)

The measurement variation is too large relative to both the process spread and the tolerance. Using this gage for Cpk reporting would produce a Cpk number that reflects measurement noise mixed into process variation. Using it for X-bar control charting would produce false alarms on measurement drift.

What to Do When Gage R&R Fails

Decompose the failure. If repeatability dominates (σ_repeat >> σ_reprod), the gage itself is noisy—consider a more precise instrument or improve fixturing to reduce part-to-gage positioning variation.
If reproducibility dominates, operators are using the gage differently. Invest in standard operating procedures, operator training, and fixturing that constrains how the part is presented to the gage.
Check resolution. The gage must resolve at least 1/10 of the tolerance. A 0.020 mm tolerance requires a gage with 0.002 mm resolution or better. A coarser gage produces floor-limited repeatability.
Fixture the measurement. Many repeatability problems trace to inconsistent part presentation. A jig that locates the part in a repeatable position removes this source.
Validate part range. If parts are too similar, re-select parts to span process variation properly. A valid study needs parts from the full range the gage will see in production.

When to Do Gage R&R

Before any capability study. PPAP submissions require Gage R&R documentation alongside the capability results.
Before implementing SPC on a new characteristic. Control limits computed from measurement-noise-inflated data will be too wide and will miss real shifts.
When introducing a new gage, operator, or fixture. Any change to the measurement process invalidates prior MSA.
Annually for critical characteristics. IATF 16949 requires periodic MSA for key quality characteristics.
After any gage calibration anomaly or failure. Calibration addresses accuracy; MSA addresses precision.

A Note on Attribute Gage R&R

For pass/fail or go/no-go measurements (not continuous), the Attribute Gage R&R is used instead. The acceptance criterion shifts to agreement rates: appraiser-to-standard, appraiser-to-appraiser, and within-appraiser agreement, all typically requiring ≥ 90%. An AIAG Attribute MSA follows a similar 10-part, 3-operator structure but records only pass/fail outcomes.

The Hierarchy MSA Enforces

Measurement systems sit upstream of every other quality tool. An invalid measurement system corrupts control charts, capability studies, FMEAs (where severity scoring depends on detectability), and customer audits (where the first question is often “how do you know that?”). The AIAG MSA Reference Manual documents the standard methodology and is the de facto reference for automotive suppliers—though the method applies equally well outside automotive. Run the MSA first. Everything downstream depends on it.

Once the measurement system is validated, the next step is selecting the right control chart for the data type and subgroup structure—covered in choosing the right control chart. From there, the SPC control chart tool handles the computation.