Continuing my saga about measurement influences and getting the measurement devices correct (see the previous post).
Here we were, called in to find out why the inspectors in the continuous pickle line were having problems with their “simple” measurements of strip width.
The first thing we did was look at the data. Model 204 is a huge database with a whole new section, at that time, devoted to Measurement Quality Assurance.
Looking at the data, we noted a really unusual fact: after the first few checks made on the tape measures, the differences were zero. Entry after entry showed that for the full range of the reference devices, from about 3 to 5 feet, they all agreed to within three significant figures. There was more than a year’s worth of useless data!
As we say in the measurement business: something that looks unbelievable usually is, or, someone cheated. What was clear was that the mill staff know or figured out the values they thought they were expected to get and “got” them!
So, the “calibration check” results were not sensible and the practice had been developed ad hoc, with little input by a person experienced in calibration or traceability. But, in reality, that was only of secondary importance.
It didn’t explain the measurement results online! What a smoke screen!
Fortunately, we had a corporate SPC manual that taught about measurement repeatability and reproducibility. What’s more, we also had a corporate committment to try to do things correctly.
It wasn’t hard then to devise a simple test following accepted and written corporate practices for R&R testing, known in the rest of the world as Gage R&R. R&R tests measure the variability of both the measurement devices and the practices followed by the people who actually used them.
The first R is for Repeatability. It is the variability due to the devices themselves.
The second R, for Reproducibility, measures the influence of the user (easily remembered if you recall that “people reproduce”). How could people err or introduce variablity?
There’s countless ways, such as, not measuring straight across the strip, not flattening the strip should there be a curl in it, or, simple not reading or recording the measurement result correctly. The latter usually results in “flyers’, eg. 1/4″ read/recorded as 1/2″ or 7/16″ read/recorded as 7/8″, or vice versa. It is, nonetheless, variablity in measurements.
Believe me it happens!
We set up a series of test during a down turn.
Five Inspectors with their tape measures and five sample pieces of strip of unknown width, merely labeled as A, B, C, D and E, that approximately spanned the widths normally produced. Each operator was asked to measure every sample and record the results on their personal form that we had preprinted.
Then they were all given a new form and asked to repeat the measurements. This was done three times.
The results, when run throught the Gage R&R calculations, were a real shock. The best result we could expect on 90% of the measurements taken was ± 1/8″, not ± 1/32″, as required!
Where do you get the procedure and calculations for a Gage R&R test? Depends on your resources. There’s always the AIAG MSA Manual. That costs about $20.
There’s also a free handbook of measurement and statistics on the Web, the NIST/SEMATECH e-Handbook of Statistical Methods, . The Gauge R & R studies Section is a bit heavy on math for the average person, but most engineers and anyone with a good math background should have no problems with it.
(Note, I use the spelling “Gage” while NIST uses “Gauge”, merely a matter of preference. In fact, I prefer not to use the prefix at all, since R&R applies to far more devices than dimension gages that is implied in the term)
There’s more to the saga, however. The response of management to the results and the way we went about fixing the problem.
That’s for another time or two, possibly next week.
