Thanks Miner for your reply. Can you provide a reference to support your answer? I'm just asking you because it's not written anywhere. I was confused between the two approaches.
The reason why I asked this question was because I wasn't getting a right answer from the books. The gage is used for process control and by forcing the variation by selecting the parts a good approach? Can we do that? But collecting parts from different days and shifts seems right but it would take you ages to get the right parts.
You said that your gage is used for process control (e.g., SPC). The answer is very clear (refer to AIAG's MSA manual). The variation of the parts selected for the study MUST match the variation of the process.
If you are comfortable with doing some basic manual calculations, you can substitute the PV (Part Variation) from the parts used in the MSA study with a more accurate Part Variation obtained from a capability study or from SPC charts.
Suppose you randomly collected parts from different days and shifts and you do your gage R&R for process control, what if your %Tolerance is like more 10% or more than 20%? What would you conclude from such a study? Is the gage capable?
Since your gage is used for process control, not inspection, the %Tolerance is irrelevant. Only %Total Variation (or ndc) is relevant.
Suppose that for the same process you have forced the variation and have collected the parts which represents the entire range of possible parts and you do your gage R&R and the results look good which is less than 10%. What would you do? Which one should I accept?
How will I have confidence in the Gage?
I would have no confidence in the gage unless the part variation matched the actual process variation.
One more thing
Suppose there is an ISO Audit, will they even accept such a methodology?
If you follow the AIAG MSA manual, ISO will accept it. If you substitute the study PV with PV from a capability study or from SPC charts, this will depend on the auditor. Most reasonable auditors
should accept it if you can rationally explain and defend it statistically. However, there is always that one auditor that will not accept any deviations.