Dear all,

I would like to have some input on the following:

We have one measurement device that is used by 15 operators for inspection measurements. All the operators have measured 10 parts twice (for five different specifications we produce). What is an appropriate way to perform R&R calculations? Can I perform calculations on the total operator group? Or shall I split them in groups of three (if so, how shall I split them)?

Please let me know your opinion on this matter!

Thanks in advance.

Patric Wessels

If you are doing the checks under the TS-16949/QS 9000 rules, there are others better qualified to answer as to the specific rules. The rules are much simpler than for measurement uncertainty.

If you are doing this for measurement uncertainty, then it is different. You would take the total of 30 measurements to develop your Type A uncertainties. Then there is the various influences that are Type B contributors such as temp at start of the measurements, temp at end of measurements, thermal coefficient of expansion for the calipers and the part, the uncertainties brought in from the calibration of the calipers.....on and on.....

Hope this helps.

Hershal

You cannot mix parts with different nominal dimension (specs) in one GRR study. So here you have data for 5 GRR studies. The operators participating in the study are usually assumed to have equal profociency, skills, training etc. If you have a large number of operators using the same measurement system, IMO, it is enough to select 3 - 4 operators randomly though I think the formula constants are available for more number of them.

For every part (specification), take 10 samples, choose your operators (say 3 different operators for each study), each has two trials on every sample.

You cannot mix parts with different nominal dimension (specs) in one GRR study.

Atul is correct of course.

I did a GR&R on a 6'' caliper. I used several gage blocks from 1/2 to 6 inches.

This is of course completely wrong, but it gives excellent numbers!

It took 3 years before it was noticed. Our registrar finally caught it. He asked if I knew that it was an incorrect method. I said I did. The audit proceeded. He did not even raise it as a nonconformance.

Also of note - this study was submitted dozens of times as part of Big 3 customer PPAPs. It was never commented upon.

I'll get to 17025, Uncertainty, and MSA after we get cost of poor quality down from 20% of Sales.

Don't get me wrong, this is great stuff...but if the customer doesn't know or care other than to tick off the GR&R box on the PPAP....well...I'm gonna have fun with it.

Dear Atul,

Of course you are right. I made R&R calculations for each specification, specifically. And per specification I split the total operator group into groups of three. As an example, for the thickness measurement (thickness 950 ± 7 nanometer) of a particular specification the R&R for the five groups of three operators each was 22.7 - 17.1 - 13.3 - 19.0 - 15.8 respectively. I think these are nice figures. However, what kind of spread can I expect between the groups of three, assuming equal skill etc.? And what factors are influencing the outcome?

The fact is that even if you repeat the same GRR study - same parts & appraisers - you would not get the exact same result. Assuming that all operators are equal, I'm not sure if a difference of as much as 7% in GRR% is resonable or to be expected. IMO, you could look at their charts and see if some are more equal than others!

Dear Atul,

Of course you are right. I made R&R calculations for each specification, specifically. And per specification I split the total operator group into groups of three. As an example, for the thickness measurement (thickness 950 ± 7 nanometer) of a particular specification the R&R for the five groups of three operators each was 22.7 - 17.1 - 13.3 - 19.0 - 15.8 respectively. I think these are nice figures. However, what kind of spread can I expect between the groups of three, assuming equal skill etc.? And what factors are influencing the outcome?

What's important--other than the basic integrity of the data--is knowing how much of the tolerance the process is using. GR&R output numbers are of little value until you have that information. There is an interesting and seldom-mentioned catch-22, chicken-or-the-egg paradox in the AIAG requirements for MSA. On page 73 of the 3rd Editon of the MSA manual there is a poorly written statement that says,

For Process Control situations where the measurement result and decision criteria determine, "process stability, direction and compliance with the natural process variation" (i.e., SPC, process monitoring, capability, and process improvement), the availability of samples over the entire operating range becomes very important. (Emphasis added.)

But how can one reliably determine the variability of the process until a gage study has been done? According to the AIAG, it's necessary to statistically qualify measurement systems before they're used in production, but the basis for the qualification is data derived from use of unproven measurement systems. The result is the Bart Simpson paradox: you're d*mned if you do and d*mned if you don't.

You have to do the best you can to assure that measurement system error and tolerance consumed by the process don't overlap. You do this by using calibrated gages and experienced operators to select parts for the gage study, and do your best to assure that the known operating range of the tolerance is represented in the parts selected for the study.