Measure Phase: Data Collection and Process Characterization
Domain 3 tests whether the data you will analyze can be trusted. Expect measurement-system studies, sampling choices, operational definitions and baseline capability calculations.
Measurement system and data plan
- An operational definition fixes the metric, unit, instrument, who measures, when, and how outliers are handled.
- Continuous data carries more information per observation than attribute data; prefer it when the process can be measured, not just judged.
- Sampling: random to avoid bias, stratified to cover shifts and lines, systematic for time-ordered streams. Rational subgroups must contain common-cause variation only.
- MSA covers bias, linearity, stability, repeatability (equipment) and reproducibility (appraiser).
- %GRR = GRR variation ÷ total variation: below 10% acceptable, 10–30% conditional on application and criticality, above 30% unacceptable.
- For attribute systems use agreement analysis with kappa; kappa near 1 means agreement beyond chance.
Baseline capability
- Cp = (USL − LSL) / 6σ measures spread only; Cpk = min(USL − x̄, x̄ − LSL) / 3σ also accounts for centring.
- Pp/Ppk use the overall standard deviation; Cp/Cpk use within-subgroup σ. Indices are valid only for a stable, roughly normal process.
Worked example
Specification 9.5–10.5 mm, mean 10.1 mm, within-subgroup σ = 0.12. Cp = 1.0 / (6 × 0.12) = 1.39. Cpk = min(0.4, 0.6) / (3 × 0.12) = 1.11. Read it as capable but off-centre, so centring the mean buys more than squeezing variation. Confirm stability first - a Cpk from an unstable process is meaningless.
Traps that cost marks
Capability calculated on an uncontrolled process; σ borrowed from another period; %GRR judged against tolerance when the study used process variation; treating attribute data as continuous; ignoring gauge resolution, which should be no more than about 10% of tolerance.