The Basis of %GRR (% Gage R&R) and other Statistics - Minitab

[bobdoering]

ndc should be 5 or greater.

This may be true for very generic measurements, but for SPC, the ndc(CL) [(UCL-LCL)/GRR] should be >10. It is an important point missed by the AIAG folks. It allows for 5 statistically significant zones on either side of the mean. Less resolution than that (such as ndc=5) is useless for control.

Unfortunately, you won't find that in Minitab. They're a little behind.

 

[chong3305]

Hi,

I'm new to this GR&R. I've run a analysis of my readings through Minitab. The results come out as usual. I'm not really sure which is the real %GRR that I should take. (%contribution or %study var)?

Thanks

 

[Miner]

There are several metrics that are out put by various software packages. Most include:

• % Tolerance (P/T Ratio) - This is used to determine whether the gauge is suitable for inspection to a tolerance
• % Study Variation (%GRR) - This is used to determine whether the gauge is suitable for use in process control (e.g., SPC) or for statistical studies (e.g., DOE)
• % Contribution - Helps determine which components of variation are candidates for improvement
• ndc - an alternate metric similar to % Study Variation

See my blog for the acceptance criteria used for each.

 

[chong3305]

Hi,

My results generated from Minitab just show the %Contribution and %Study Var. It did not show the %Tolerance. Do I left out anything in the setting? By the way, I'm using the Crossed method and my purpose of study is to check the measurement error in the quality department. In the study, 3 operators were being involved and 10 parts with 3 time of measurement being taken. Correct me if I'm wrong.

Thanks

 

[Miner]

In order to see the % Tolerance, you must first enter a tolerance. Otherwise it will not appear in the session window.

If you want the actual measurement variation, you should look at the s or 6s values. The other metrics are all as a proportion of tolerance, process variation, etc.

 

[BrQ]

Can I get a little help bringing the explanation of the results to a very basic level?
I have heard it stated that a 10%tol Gr&r result means that you can expect your measurement system error to be 10% of your tolerance.
(This sounds more like a statement of MU to me than a GR&R.)
I believe this is untrue (who would think that a 10% error on a tol of .030" would be deemed "excellent"?)
But I am having trouble relaying exactly what the 10% is, in laymans terms.

Thanks always for the help.

 

[Miner]

To be precise, it means that 6 standard deviations of the measurement variation consume 10% of the tolerance. The probable error of a single measurement is 0.67 StdDev, which is a much smaller percentage.

 

[BrQ]

Where does the 0.67 stdDev come from? Was this using my example of a.030 tolerance?
I'm getting closer to understanding this and being able to explain it, always find your help immensly helpful, you are a true asset to the Elsmar community.

 

[Bev D]

In general the %R&R is not calculated in a mathematically correct way. so the %number has no direct correlation to how much error is present given the tolerance. It is more helpful to think of them like p values.

 

[BrQ]

Tried explaining that to PD engineer, he wants more.
Informed him of the fact that R&R is not geared toward giving you measurement uncertainty, just not sure exactly how to explain any better than " these are values used to interpret the study and these are the thresholds to rate those values.
Went into the fact that the outcome is not proportional, imperfect math to give the answer he wants to hear, etc..

Thanks for the reply, I would like to hear more about Miners intrepretation, just a little more explanation, I feel like I'm almost getting to something I can use to pass on. Any thoughts Bev? Miner?
Thanks again to both, always appreciate your inputs.