AIAG MSA and Gage R&R - What does the Process Variation % R&R truly mean?

[Jrodrigu]

AIAG MSA and Gage R&R

I'm implementing Gage R&R and Measurement Process Evaluation at my facility and have a few questions.

I understand the rules of thumb regarding % R&R percentages i.e. <10% is good, 10-30% is ok but may need improvement, >30% measurement is inadequate. How does this equate to the new philosophy of Gage R&R Process control view where process variation is accounted for rather than tolerances. What does the process variation %R&R truly mean? What criteria or rules of thumb should be followed for the process %R&R? How does this correlate to traditional Gage R&R using the tolerances?

Any information is helpful including websites or literature.

 

[Bill Ryan - 2007]

What does the process variation %R&R truly mean? What criteria or rules of thumb should be followed for the process %R&R? How does this correlate to traditional Gage R&R using the tolerances?

It's my understanding (from my Tier 1 customers) that the same acceptance rules apply. That is why I still use the %Tolerance figure. If I were to use the % Process variation I would be spending some pretty hefty bucks on new gaging with no budget or customer help to finance new gaging (funny how, when the customer is approached about helping to improve the gage system for a very reliable/stable process by helping on the financial end, the improvement becomes not as necessary!!!!!).

I have not had an issue by staying with the %Tolerance numbers and as long as the MSA manual uses the term "if" I will continue to do so (unless told otherwise by MY management). Now that I've made that declaration, I'm sure I'll be getting quotes for some super-duper gage measuring to .0000001 mm. tomorrow.

Bill

 

[Al Dyer]

Gage R&R is important, but there are 3 other areas that are now becoming more important and reviewed.

They Are:

1: Bias
2: Linearitty
3: Stability

Of course we need to define in our lab manual which gage gets what test, but that is the good thing about not having definitive requirements for each.

Look in the PDF files and you will find a wealth of information and examples!

 

[Jrodrigu]

So what does it mean if your % Tolerance R&R is 13.27% which is ok and your %Process Variation R&R is 97.6% which is an inadequate measurement system? Will improving the process actually decrease the variation or is it inherent in the test method?

 

[Roger Eastin]

I think it means that with respect to the tolerance, you're ok. However, for any meaningful process improvement, I wouldn't use your measurement system to tell you when you made an improvement. All your "process variation" is eaten up by the measurement method.

 

[Darius]

I am agree with Roger, may be not imposible, but hard to see any improvement with such variation of the measurement system.

Most of the text books about R&R tell you that is the R&R fail, you have to repeat the experiment or try another measurement equipment.

The case looks to me like R&R for destructive testing, wrong measurement equipment scale or sampling methodology (most of the pleople I know, tink that just taking samples with no "rational subgrouping and sampling methodology" will give good results, and in most of the cases the R&R fail).

So, i first recomend to review the sampling methodology, then if there are no other way, try other measurement equipment.

:thedeal:

 

[Atul Khandekar]

Jrodrigu said:
So what does it mean if your % Tolerance R&R is 13.27% which is ok and your %Process Variation R&R is 97.6% which is an inadequate measurement system? Will improving the process actually decrease the variation or is it inherent in the test method?

I believe we are looking at a highly capable process - one with a very high Cp. (cannot comment on Cpk or centering). As Roger Eastin has already pointed out, most of your process variation seen in the SPC study is caused by measurement system variation. As such this measurement system cannot be used to 'improve' the process further.
Question is, with a high Cp, is it worth spending the time, effort and money to improve the measurement system? What is the practical limit?

-Atul.

 

[Ravi Khare]

The purpose of a measurement system is to detect process variation.

Even if we compare the R&R with the tolerance, it may perhaps eat up most of the process variation where Cp is high, but you still will be able to centre the process within the tolerance. If you have 5 or more number of distinct categories within the tolerance range, you are still within the central 20% zone.

For such a highly capable process, insisting on a good R&R to Process variation ratio ( <10%) may be statistically correct, but the cost may not justify the end.

 

[rerich]

GR&R Formulas for Calculation of Standard Deviation

This has been a hot topic around the office lately. We have been using a spreadsheet made in Microsoft Excel. There are 2 options for caclulating standard deviation.
STDEV
STDEVP

STDEV 'estimates' the standard deviaton based on a sample of the population.
STDEVP 'calculates' the standard deviation based on the assumption that you have the whole population.

STDEV uses the N-1 version of the standard deviation formula
STDEVP used the N squared version of the standard deviation formula.

STDEV matches the results as calculated using the AIAG/Ford Method
STDEVP are lower than the results as calculated above

Although our documentation says that we should use the STDEV method, I disagree with the ESTIMATED formula. One could say that we only have a sample of the population of the infinite number of parts that could be manufactured, and these limits are not able to be set in the spreadsheet, i.e. lifecycle of the part is 1000 pieces. The formula estimates the error exponentionally.

I believe that we do have the 'whole' population because our spec says that we use '1' part with 2 operators and 8 loads each. This could also be interpreted as having the 'whole' population because we are doing a 'GAGE' R&R, so is the gage the 'whole' poplulation since it is what is being tested?

Your thoughts would be appreciated.

Thank you,
Rich

 

[Rob Nix]

After going through the same discussions throughout the years, it comes down to this: don't worry about it. Pick a measure of dispersion and use it consistently.

Juran says in "Quality Planning and Analysis" that standard deviation (sample or population) "is only a formula. There is no hidden meaning to the standard deviation, and it is best viewed as an arbitrary index".

Regarding total estimated variation, the MSA manual says on page vi, that you 'the reader can choose' whether to use 5.15s (99% of total estimated distribution) or 6s (99.73%). It is still just an estimate.

And, if comparing a distribution to a print tolerance, the tolerance itself is often an arbitrary zone chosen by an engineer.

In short, it is an exercise in picking nits. You can discuss the accuracy and precision of the statistic til the cows come home, but you will end up correcting a gage if it generally appears to have "too much" variation, and you will leave it alone if it appears to be consistent. It is like saying that if you are standing about 10 feet from the edge of a cliff your are at a safe distance (even if your true distance is 9 ft.). But if you appear 6 inches from the edge when your true distance is 5.2 inches, it is still dangerous - back up!.

I suggest you put the "hot topic" to rest. Pick a measure of dispersion, and use it consistently.