Separate part variation from measurement variation

Separate part variation from measurement variation

A measurement study separates variation due to the measurement system from variation between parts. Select parts and conditions that represent the intended use, and use a design appropriate to whether repeated measurements are possible. Randomized, blinded repeats reduce memory and order effects. Repeatability and reproducibility are distinct components; estimated independent variances combine before taking a square root. Consequently, a percentage based on standard deviations is not the same as a percentage of variance. Do not treat a sample design or a generic percentage as a universal acceptance criterion. Interpret the result against the measurement decision, risk, resolution and other relevant sources of measurement error.

Follow the method

1. Supplied standard deviations (mm)
2. Standard deviation (mm)
3. Measurement component
4. Repeatability
5. Reproducibility
6. Combined gauge
7. Total study
8. Study variation ratio / % of total SD
9. Variance contribution / % of total variance

Explain two honest percentages from one measurement study

Fictional case: a gauge study uses ten representative parts, three operators and two repeated measurements, giving 60 readings. The core exhibit supplies repeatability SD .003 mm, reproducibility SD .004 mm and total study SD .025 mm. Metrologist Eve must explain a report that labels gauge variation both 20% and 4%, without claiming these components were calculated from an unseen raw dataset.

Follow the method

1. Gauge variance
2. Gauge SD
3. Share of total SD
4. Share of total variance
5. Fitness decision

Twenty percent and four percent can both be correct

Distinguish measurement-study design from supplied variance estimates and explain why standard-deviation and variance percentages differ.

Fictional case: a gauge study uses ten representative parts, three operators and two repeated measurements, giving 60 readings. The core exhibit supplies repeatability SD .003 mm, reproducibility SD .004 mm and total study SD .025 mm. Metrologist Eve must explain a report that labels gauge variation both 20% and 4%, without claiming these components were calculated from an unseen raw dataset.

Role: Measurement-system specialist and inspection owner

Normal condition

The study represents intended parts, appraisers and conditions; repeats preserve identity and avoid memory effects; the analysis method and denominator are explicit.

The gap

A manager adds .003 and .004 to obtain .007 mm and treats the larger percentage as evidence that the smaller one is dishonest.

• Components are supplied estimates for an arithmetic illustration.
• No universal acceptance threshold or complete MSA conclusion is supplied.

1. Define the measurement decision

   Eve records the characteristic, range of intended parts, fixture, gauge and appraisers. She asks whether the study represents actual use rather than ten unusually similar parts selected for convenience.

   Why: A precise calculation cannot repair an unrepresentative study. Measurement fitness concerns the decisions the system must support.

   Evidence: The study plan identifies artifacts, operators, conditions and intended use.

2. Protect repeated observations

   The illustrative plan has 60 measurement opportunities. Eve assigns stable part identities and a blinded/randomized order where suitable, with actual observations recorded rather than copied from earlier rounds.

   Why: Repeats are meant to reveal measurement variation, not recall of the previous reading. The design is an example, not a mandated sample size for every measurement system.

   Evidence: A measurement schedule distinguishes part, operator and repeat.

3. Combine supplied variance components

   She squares the SDs, adds .000009 and .000016mm², then takes the square root:.005 mm. She does not add standard deviations directly.

   Why: Independent component variances add under the stated model. Their square root returns the result to measurement units.

   Evidence: Combined gauge variance .000025mm² and SD .005 mm.

4. Label the two denominators

   Gauge SD divided by total SD is .005/.025=.20 or 20%. Gauge variance divided by total variance is .000025/.000625=.04 or 4%.

   Why: The two percentages answer different questions and are mathematically consistent. A common study-variation multiplier cancels when applied to both numerator and denominator.

   Evidence: The report labels percent of total SD separately from percent of total variance.

5. Interpret fitness with the missing context

   Eve requests the fitted-model details, representative data and applicable decision criteria. She also considers bias, resolution, stability and other measurement effects beyond repeatability/reproducibility.

   Why: A small variance ratio alone does not establish every aspect of measurement fitness. Calibration status and repeatability are related evidence, not interchangeable approvals.

   Evidence: The conclusion reports the supplied arithmetic and lists the remaining fitness evidence.

The selected parts cover almost no process range

The ten study parts were all chosen from one narrow group, unlike the range encountered in normal inspection.

Review and redesign the representative-artifact selection before interpreting the ratio as evidence for normal production decisions. Preserve the original study scope.

The denominator and apparent relative variation depend on the study population. A ratio from an artificially narrow range can be misleading for the intended use.

The revised plan explains how selected artifacts represent the measurement task.

Run a practice session

Materials

• Measurement schedule
• Component cards
• Calculator

1. Describe a repeat · 5 minutes

   How would memory of a reading bias the exercise?

2. Combine components · 8 minutes

   Why square before adding?

3. Work the changed estimates · 10 minutes

   How can 20% and 4% agree?

4. Debrief fitness · 5 minutes

   What can this ratio not tell us?

Debrief

• Ask learners to state the denominator aloud.
• Challenge the idea that one example design is a universal minimum.

Draw a units column beside every intermediate calculation.

Transfer into the work

Owner: Measurement-system owner

Record: Study plan, raw readings, model, component labels and fitness decision

Review: Before use for a new decision and after material method changes

Evidence: Representative design and suitable measurement evidence

Improve the method or study design and reassess rather than changing percentage labels to appear acceptable.