Nested Gage R&R: Measurement System Analysis for Destructive Tests
When to Use Nested Gage R&R
Conventional Gage R&R requires the same part to be measured repeatedly by multiple operators, which is impossible for destructive measurements (tensile testing, fatigue testing, chemical sampling) because each measurement destroys the part. In these cases a nested design is used: different operators measure different part samples, so measurements are nested within parts and the part x operator interaction cannot be estimated. This applies to destructive tests such as tensile strength, hardness and chemical composition, and is the only feasible Gage R&R scheme for such measurement systems.
How to Set Up the Nested Design
Assign each operator an independent set of parts (for example, 10 parts each) and measure each part once; if each part is measured twice, the two measurements come from different homogeneous samples. Because parts do not cross operators, the data have a hierarchical nested structure: operator, then part, then repeat measurement. The tool decomposes the variance components with a nested variance model: part-to-part variation (PV), operator variation (AV) and measurement error (EV), and combines them into the GRR estimate.
How to Read the Results
The nested design's variance decomposition contains no interaction term, and %GRR is combined from repeatability (EV) and reproducibility (AV). The acceptance criteria match conventional Gage R&R: %GRR < 10% is acceptable, 10% to 30% is conditionally acceptable, and > 30% is unacceptable; ndc should be >= 5. Note that because parts do not cross operators, part variation may include between-batch differences; when part uniformity is poor, PV is inflated and %GRR is underestimated, so interpret conclusions cautiously.
Precautions
For destructive measurements, assess part uniformity first (adjacent locations of the same batch should give close values) and randomize part assignment. Use at least 5 parts per operator, and 10 where possible. If the measurement process includes sampling error (for example, different sampling locations), the sampling variation is mixed into EV and should be stated in the conclusions. Each destructive sample represents an independent individual, so the sampling-location difference is itself part of the measurement variation and should be included in the analysis; the tool outputs the variance-component table and a %GRR report for direct use in MSA audits.