Ok, here is the update.
I received a few samples. Using the air gage, only one is out by .0001. On the CMM all parts are within spec.
Bob, in speaking with the customer, their gage probe (the part which holds the part for measurement) is .001 smaller diameter than ours. Could that cause the discrepancy?? Some of the parts are tight, so we know we are influencing the part when it is put on the gage. Thus, we like the CMM since there is no part influence.
If the parts are tight, that is a problem. It should be free floating on an air gage.
As far as the probe diameter, how close does the master ring read? How about their gage - how close does it read to their master ring value? That is key.
You are right, the CMM would be a good referee method.
Back to my original question though. We have taken all the Cpk data which shows capability -- not perfection. You're still subject to a "bad" part from time to time. In this case, I see only 1 pc. out by a tenth. So what is the point behind the cpk if you're going to make me sort cause 1 pcs. was slightly out of spec.? Not to mention that the cpk is done using a "bogus" tolerance.
If you are required to show a Cpk of 1.67, it pretty much reduces your usable portion tolerance by 25% - more if you are not centered. That accommodates the portion of your probability that lies out in the tails of the distribution, as well as sampling error (because you sample and do not measure 100%), gage error, etc. (see
total variance equation)
So, do you sort because 1 pc. was slightly out of spec? No, you sort because there is a
probability that there are more pieces out of spec, and by doing that sort you may find some even more out of spec than the relatively small sample you have now. Bad parts are like roaches - if you find one, you can bet there are more.
If you ran to 75% of the spec (or
could), there would be no argument now. With the contributing variations, you never really "own" the whole tolerance on the shop floor. Only dock inspection and customer incoming does - and even then only theoretically because they still have gage error.
Is the tolerance bogus? If they really need it that tight, no. Are you capable? Hard to say. You have gaging issues, but even so, you do not have the kind of process you can "dial in" like machining. Are you running high, low, or are you using the whole spec? It would still be good to see a capability study.
The real problem is, if your company signed up for a job they are not statistically capable on, this will be an ongoing problem....