Skip to content
Six Sigma | Part 4 of 10

Measure the Measurement First: Gage R&R

A gage repeatability and reproducibility study shows how much of the tolerance the measurement itself uses, before any process data is trusted.

A team collects 125 bore diameters, calculates capability and starts looking for causes. Weeks later someone notices that two operators measure the same part 0.008 mm apart. Part of the variation the team was hunting lived in the gage.

What a gage study measures

Every measurement contains two kinds of variation that have nothing to do with the part:

  • Repeatability: the same person measures the same part with the same gage several times and gets different values. Typical causes: gage design, clamping, measuring point.
  • Reproducibility: different people get different averages on the same parts. Typical causes: methods that are not standardized, training.

Together they form gage R&R. The question is how much of the tolerance they use up.

The standard study

Ten parts that cover the normal range of the process, three appraisers, two trials each. Parts are numbered on the bottom, measured blind and in random order, so nobody remembers the previous value.

The average range method calculates repeatability from the differences between the two trials, and reproducibility from the differences between the appraiser averages. The result is expressed as a share of the tolerance:

share of tolerance = 6 × gage R&R / (upper limit minus lower limit)

Share of tolerance Decision
below 10 percent acceptable
10 to 30 percent acceptable with care, depending on the use
above 30 percent fix the gage first

The example

The bore of a pump housing, tolerance 0.060 mm, was measured by hand with an internal micrometer. The study gave a gage R&R of 4.3 µm, which is 43 percent of the tolerance. Repeatability was the larger part: the measuring depth varied from part to part.

A simple holding fixture with a fixed depth brought the result to 0.9 µm, or 9 percent. Cost: a few hundred euros and two days. Part of the “process variation” disappeared before the process was touched.

Other situations

  • Visual or go and no-go checks: an attribute agreement study with 30 to 50 known good and bad parts, each appraiser twice.
  • Automated in-line gages: repeatability with the same part measured 25 to 50 times, plus a bias check against a reference.
  • Calibration is not a gage study. Calibration checks accuracy at a reference point, not the variation in daily use.

Resolution

The gage must show steps of at most one tenth of the tolerance. A gage that reads in steps of 0.01 mm cannot judge a tolerance of 0.06 mm.

Make it a rule

In a mature program, no data is collected for a critical characteristic until its gage has passed a study. It feels like a delay in week two. In practice it is the fastest way to a result, because every later chart and test is built on numbers that can be trusted.

Newsletter

One Lean idea.
Every two weeks.