In MSA manual, it mentioned select 10 parts to concduct the GRR, and we'd better to select those parts which can cover the whole process variation. Then my consultant said we can select some parts which are closed to LSL/USL.

My question is how can I select those parts from the production process? The probability is quite low concenring the normal distribution in the area of LSL/USL. Does anybody have a clue or can you show me your method?

Thanks a lot in advanced

I am afraid I can not help and I will complicate this:

This is one of the biggest problems that exist with GRR as if there is very little change between the parts then the GRR cannot be good as statistics deals very badly with "little change". Idealy you want parts that are out of spec as this gives the best results.

If the GRR is for the instruments and the operators why do you actually need to measure the same dimension and not collect 10 seperate dimensions with the same tolerance?

If the GRR is for the instruments and the operators why do you actually need to measure the same dimension and not collect 10 seperate dimensions with the same tolerance?
Because in general, GR&R is used to verify that the system is appropriate for use on a particular part. The subject of "families" has been discussed before, and it's true that sometimes there's no need to use specific parts, but care must be take to insure that surrogate data is reasonably analogous.

If possible, you can purposely MAKE parts which are 'beyond the norm', and which approach (or exceed) the USL/LSL.

I have done this in the past, especially for flatness measurement; on thin parts it is fairly easy to create virtually anything wanted.

I believe that a GRR study requires that parts be selected from the process in such a way that they represent the entire process variation that occures during production. If you have SPC charts, you can study how the process varies over time and select parts over a period of time, not all at once.If possible, you can purposely MAKE parts which are 'beyond the norm', and which approach (or exceed) the USL/LSL.Unfortunately, I have not not been able to find any supporting evidence for this. Question would be: exceed USL/LSL by how much? 10%? 50%? This is sure to reduce the GRR number (variable).

One more wrench for the works. . . . Once you have the parts selected, maintain them for periodic studies for the same gage family. Use the same samples year after year. This will help show degredation in the gaging process.

Of course, if you are making wings for 747's. . ..get a LARGE storage area!

:bonk: