Verifiying New Measurement Methods Via Statistical Analysis

[Rogue909]

I know this is going to appear as a silly question and I believe that I'm over thinking or missing the obvious. We've brought on a couple new pieces of metrology equipment and, via our QA manual, are required to perform a statistical analysis to verify that the equipment being brought online produces repeatable results compared to what is currently done. Our parts are being measured first by the new equipment then by the old equipment. This functionally serves to provide me with lots of data to sift through and analyze for statistical effectiveness. Also, it gives time on the floor to work on 'the kinks' that come along with new equipment.

Now I've gathered enough information to do the analysis. Data wise everything 'feels' good. The problem that I'm running in to is the data does not fit a normal distribution curve. I think this is mainly a resolution issue. We typically hold .00X +/-.001". Measured with a resolution of .001". You don't end up with a very good distribution in the data. Just a histogram with three columns; 2 hits in -0.001, 14 in 0.000, and 9 in +0.001.

Most of the statistical tools that I've looked at (students test and ANOVA mainly) require the normal distribution of the data. What other methods do I have to work around this?

 

[Hansac]

You need to do linear or quadratic regression to check the R-square of your test data.

 

[IRstuff]

As a general rule, your test accuracy ratio (TAR) should be 10x, i.e., 0.00X/resolution should be at least 10, ideally.

As for what you are getting, you could still apply some sort of envelope analysis to determine a normal curve that contains all the measured data. An eye-ball look suggests that you possibly have a tiny bit of offset error, but you really don't present much data to work work here.

TTFN (ta ta for now)
I can do absolutely anything. I'm an expert!

https://www.youtube.com/watch?v=BKorP55Aqvg

faq731-376 forum1529 Entire Forum list

http://www.eng-tips.com/forumlist.cfm