What does % study variation of 20% mean in GRR?

QE1993

Involved In Discussions
Hello!

Statistics is not my strong suit, so I would appreciate all the knowledge bombs.

I did a GRR on a drop indicator that measured a counter-bore depth of a tube.

The total %studyvar was 29.89%. I was previously getting almost a 38%, but was able to identify set-up of the inspection/part apparatus as a source of significant variation. I fixed it and was expecting the GRR to be <10%; however, it is still higher than ideal.
The data is attached.

Our acceptance criteria allows 11-30% GRR's to be accepted with a justification. I will be using the gage purely for inspection (i.e. ensure parts are within their tolerances).

Can someone help me interpret these results? I don't know what else I can do to reduce variation even further and I do not believe we have a better method accessible.

Thank you!
 

Attachments

  • Gage R and R.docx
    36.4 KB · Views: 277
Last edited:

Miner

Forum Moderator
Leader
Admin
I have two comments:
  • Since you are using your gage solely for inspection, use %Tol. %SV is only applicable for process control and capability studies.
  • You do not have sufficient gage resolution . Looking at the Range chart, you only have 3 measurement levels within the UCL-R. You need a minimum of 5 possible levels to have adequate resolution.
 

QE1993

Involved In Discussions
Hi! Thanks for the response. Do you mind if I ask some follow up questions?

1. Can you explain why %tolerance is more relevant for inspection only purposes?

2. The gage is resolves to 0.000050 and the feature has a total tolerance of .004. It meets 10-1. Shouldn't it be sufficient enough? Please let me know if I am misinterpreting something.

Thank you!
 

Ninja

Looking for Reality
Trusted Information Resource
1. Can you explain why %tolerance is more relevant for inspection only purposes?

2. The gage is resolves to 0.000050 and the feature has a total tolerance of .004. It meets 10-1. Shouldn't it be sufficient enough? Please let me know if I am misinterpreting something.

My reasons (if there is a discrepancy...go with Miner's):

1. You're essentially measuring pass fail. This test is to ensure your gage can adequately discriminate between pass and fail. Pass and fail are determined by Tolerance.

2. Your gage might have enough resolution, but your parts don't vary enough for the test to be thorough. Sometimes this is called "NDC" for # of distinct categories. Your gage might be fine, but you have three categories of parts...not enough variation among them to truly test the test method.
Having at least 5 categories discernible will help.
 

Golfman25

Trusted Information Resource
Hi! Thanks for the response. Do you mind if I ask some follow up questions?

1. Can you explain why %tolerance is more relevant for inspection only purposes?

2. The gage is resolves to 0.000050 and the feature has a total tolerance of .004. It meets 10-1. Shouldn't it be sufficient enough? Please let me know if I am misinterpreting something.

Thank you!

Welcome to the club of trying to make the math work. You have a standard gage, looking at a standard part, probably without issue for years. Now you need the fancy statistical study to show your gage is "good" which shows your gages are "bad." Get ready to pull your hair out. We have very little luck getting classic gages to "work" statistically. In many cases you need to pick up "variation" that you can't control anyway. Good luck.
 

Ninja

Looking for Reality
Trusted Information Resource
Welcome to the club of trying to make the math work. You have a standard gage, looking at a standard part, probably without issue for years. Now you need the fancy statistical study to show your gage is "good" which shows your gages are "bad." Get ready to pull your hair out. We have very little luck getting classic gages to "work" statistically. In many cases you need to pick up "variation" that you can't control anyway. Good luck.

LOL...thanks...the old "Automotive wants it, so everybody else wants it 'cause auto is perfect, so we have to do it because everyone else does" gig.
I've lived there way too long...

'Auotomotive is perfect' means they have a bunch of money we want...

Love it! :D
 

QE1993

Involved In Discussions
Get ready to pull your hair out. We have very little luck getting classic gages to "work" statistically. Good luck.

What's making me pull my hair out even further is that we keep asking our customer (who randomly required the GR&R for the new PPAP requirements that are being flowed down) for their GRR on the gage, since they use the same method for their RI. They keep avoiding giving it to us and one particularly snotty Quality Engineer over there said "well it doesn't matter if ours pass, you're the experts, that's why we're placing POs." We still have to correlate, sir. We're going to have major issues if we don't correlate.
 

Golfman25

Trusted Information Resource
What's making me pull my hair out even further is that we keep asking our customer (who randomly required the GR&R for the new PPAP requirements that are being flowed down) for their GRR on the gage, since they use the same method for their RI. They keep avoiding giving it to us and one particularly snotty Quality Engineer over there said "well it doesn't matter if ours pass, you're the experts, that's why we're placing POs." We still have to correlate, sir. We're going to have major issues if we don't correlate.

They can't get theirs to "pass" either. :)
 

Proud Liberal

Quite Involved in Discussions
Have you considered any impact that form variances could be influencing your study. I work with parts with tolerances in microns. To eliminate form (roundness, parallelism, etc) from being included in our R&R studies, we mark parts so measurements can be taken at the same place. In our case, the impact can be significant.
 
Top Bottom