Gage R&R Study - Appropriate sample collection and size for a Gage R&R Study

[Coleman Donnelly]

Perhaps someone here can help us settle a dispute...
We are planning a gage R&R study for a checking fixture that was recently recieved and there is a difference of interpetation as to how the sample pieces should be selected.
The study will be conducted using the 10x3x3 method. We have 100 pcs to choose our samples from all of which have had 100% layouts completed. One opinion is that the 10 pcs should be selected that reflect the mean tollerance band for the measuremeants that will be taken. The other opinion is that the 10 pcs should reflect the greatest range with even distribution along the tollerance band for the measuremeants that will be taken.
I would like to get some further clairification on this topic from people here with greater understanding of the concept than either of us so please expand upon your reasoning as this will help us to gain a greater confidence level in the work that we are doing
TIA

[David DeLong]

Yes there is controversy on this subject but here goes.

I would suggest that you reflect samples that reflect your complete process. This would include both samples that are at your process extremes.

I have seen so many times where an R & R study achieved an 80% value of the process while only 12% of the specification. The samples were selected randomly and not truly reflective of the process width.

[Coleman Donnelly]

Thank you for the input - could you please expand upon your reasons for your recomendation?

[David DeLong]

In Reply to Parent Post by Coleman Donnelly

Thank you for the input - could you please expand upon your reasons for your recomendation?

In the R & R study, there is a formula for calculating the process width using the 10 samples provided. If one was to provide 10 samples there were all around the average, the process width calculated would shrink and your R & R percentage of the process would be relatively large.

If one selected samples that truly reflected the process, your calculated process width using the 10 samples would be larger reflecting a lower gauge percentage.

Hope this helps.

[Coleman Donnelly]

Part of our dispute envolves the fact that the project engineer has the understanding that all parts used for gage R&R have to be within tollerance of the customer specifications in order for the gage R&R to be "good" My thoughts have been that the part being in tollerance did not matter as long as it was within the ability of the gage to read the part.

[Jim Wynne]

In Reply to Parent Post by Coleman Donnelly

Part of our dispute envolves the fact that the project engineer has the understanding that all parts used for gage R&R have to be within tollerance of the customer specifications in order for the gage R&R to be "good" My thoughts have been that the part being in tollerance did not matter as long as it was within the ability of the gage to read the part.

There's a chicken-or-egg/Catch-22 thing here that is seldom recognized (how can you have confidence in the extent of process variation if the measurement system hasn't been qualified?) but nevertheless, I'm not sure why you would want to include nonconforming parts in a variables measurement study that will be submitted to a customer. Get the process in control and capable, then do your MSA.

[Coleman Donnelly]

Anyone else care to chime in on this?

[Bev D]

In Reply to Parent Post by Coleman Donnelly

Anyone else care to chime in on this?

In order to answer effectively I woudl want to know what you are lookign for (I would give different input based on your answer)
do you want to truly improve your process?
OR
are you only interested in submitting the appropriate paperwork to your customer and not raising any questions from them?

whether your parts are in tolerance or out of tolerance has no bearing on the validity of the assessment of the measurement system. Most of the time the more variation you have the more informative the study is, particularly if your system has measuremetn error that is variable dependign on the actaul value...by the way jsut to stir the pot some more, 10 pieces is really too small: we're trying to estimate a standard deviation with subgroups of 2-3 pieces...30 subgroups is a better sample size for this.

Your customer may feel differently: some require you to have out of spec parts to assess your ability to detect our of tolerance parts; some don't want out of tolerance parts in teh study becuase they - wrongly - interpret their presence as an indication that your process is not capable and you will ship them nonconforming parts.