Consistency Charts

SPC for measurement systems

Wed, 03/27/2013 - 11:15

ADVERTISEMENT

What happens when we measure the same thing and get different values? How can we ever use such a measurement system to measure different things? By thinking of measurements as the product and the measurement procedure as the process we can use the techniques of statistical process control (SPC) to address the problems of complex measurement systems.

ADVERTISEMENT

A consistency chart

Churchill Eisenhart, a famous statistician who worked at the National Bureau of Standards, once wrote that a measurement system cannot be regarded in any logical sense as measuring anything at all until it has attained a “state of statistical control.” As I showed in “Three Questions for Success” (QDD, March 1, 2011) a process behavior chart is an operational definition of how to get the most out of any process. The measurement process is no exception.

 …

Want to continue?
Log in or create a FREE account.
By logging in you agree to receive communication from Quality Digest. Privacy Policy.

Comments

What about destructive testing? My background as an adhesive technician leads me to ask for this question. R&R means repeatibility, that's simply out of border of any destructive testing - like car crash-testing, for example. I agree on Consistency Charting: provided the "industry bigs" are made aware of it; to date, they do ignore the meaning of Statistics, they use SPC as a commercial tool to put pressure on prices - not on quality. Thank you.

Yes, a different approach is clearly required with destructive testing. I make a distinction between destructive testing in which paired samples can be obtained and destructive testing where paired samples are unobtainable. In the former case we can place an upper bound on measurement error. In the latter it is impossible to identify measurement error as a component of the overall variation. However, since measurement error is automatically included in the computations for the limits, we can still use these destructive tests to obtain data for a process behavior chart for the production process.

Destructive testing can be difficult to analyze for measurement error.  In our process, a destructive test is part of the quality assessment of our product.  We have found using an Average-Range chart with n=3 to be effective.  The average result of 3 samples is reported as a single measurement and the range chart is used to identify measurement errors.  The 3 samples are taken from the same moment in production to support homogeneity.  The range chart gives a reasonable approximation of R&R.

This has been the best approach we have found.  Obviously, this would be cost prohibitive for crash testing vehicles but can be effective for other types of destructive testing.

Yes, the range chart includes the variability between samples.  However, the difference between samples should be very small compared to the measurement error.  Given the nature of our process, variation over time is much higher than between samples at a given moment in time.  Therefore, in our case, the range chart is dominated by measurement error.  This is not ideal, but it's the best approximation of measurement error we have found.

Add new comment