GRR Flatness with a robot and CMM scan

[Todd_A]

I have read several posts that warn against doing GRR on flatness because of variation of the appraiser. How does that coincide with a robotic cell, where there is no human interaction? A robot loads the part on a CMM, and the flatness is scanned. The robot then removes the part, picks up another, etc. until the sample size is complete.

Run A: 0.0210 0.0242 0.0264 0.0224 0.0247 0.0281 0.0189 0.0221 0.0215 0.0236
Run B: 0.0206 0.0246 0.0258 0.0219 0.0246 0.0269 0.0191 0.0217 0.0212 0.0233
Run C: 0.0207 0.0241 0.0260 0.0217 0.0240 0.0273 0.0193 0.0214 0.0208 0.0226

These results are comprised of 10 pieces, each being ran three times. My total GR&R is coming in at 14.51%! Why? Clearly the variation between the ABC runs is minimal. My equipment variation is coming in at 11.27%. I don't understand why everything looks so poor when the actual variation is so tight. By the way, those numbers are metric.

Below is the advice being given by ANOVA. Furthermore, I tested the data in another program and got similar results.

"Your measuring equipment may need calibration or maintenance."

This is a brand new CMM. The GR&R was performed the day after installation was complete. I'm 100% positive, without question that calibration is good. I went as far as running a probe validation after each 10 piece set.

Can anyone offer some advice?


Thank you

 

[IRstuff]

Were they run in the same order in all three runs? What is the specified performance of the CMM? What is the measurement unit? How are you calculating the GR&R?

The repeatability of your measurements, assuming order was maintained, is something like 1.5%

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

 

[Todd_A]

Thanks IRstuff. Yes, the part run was kept in order. This is metric, and I'm simply using a template

http://econoshift.com/en/gage-rr-template-2/.

Again though, the numbers were run through several different templates. One we usually use in house here, and then whatever I could find to compare. They're consistent. As far as how it's being calculated? This is not my forte', I don't know. I can say that I ran a data set from another feature through the same template and it came in at .39%, and the variation was worse than what these are.

I had to edit for your CMM question. Exactly, I don't know what it's rated for, but I know it's very tight. The company I work for spared no expense for this Hexagon CMM. There are features with max tolerances of .013mm, and it's accurately repeating those.

 

[IRstuff]

The 14.51% GR&R is the percentage of the total measured variation attributable to the measurement itself, and simply means that you choose samples that are borderline for test accuracy ratio. You didn't specify the specific model of CMM, but a typical system on their website

https://www.hexagonmi.com/en-US/products/coordinate-measuring-machines/shop-floor-cmms/tigo-sf

has a basic accuracy of about 2.2 microns. Your sample standard deviation is 2.5 microns and the part measurement standard deviations average to 0.35 microns, hence the 14.5% GR&R. The machine's absolute repeatability is 0.35 microns, so it seems to me to be meeting its specification.

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