That [detecting process shifts] in NOT, however, the sole purpose of measurements. The GR&R method is to determine at accuracy of placing the measured units into conforming or nonconforming categories.
I would phrase it differently; the purpose of GR&R is to confirm that a specific measurement system is appropriate to to the application, with the null hypothesis being that the system
is appropriate. It's a type of controlled experiment.
Gage R&R is not about the process!
Of course it's about the process.
Everything in manufacturing is about the process. If processes are properly controlled, the need for categorization of product conditions is eliminated, or at least greatly diminished.
It is about the unit under test, and how much confidence we can place in the determination that this unit is in fact either conforming or nonconforming based on our measurement.
This is indeed one of the purposes, but one we should be doing our best to avoid, and we do process control to avoid it.
This is a completely different issue than what Wheeler is discussing. Notice that GR&R is based on the product tolerance. Since when did product tolerance have anything to do with process stability?
If not for product tolerance, why would we even be concerned with process control, except perhaps for economic reasons? Tolerance and stability are
not independent concepts, even though they may be evaluated independently.
Wheeler makes a basic assumption, and that is that process control is what's really important. A standard AIAG-type GR&R process, sans the questionable arithmetic, can be useful on a number of fronts. It can tell you something about variation between operators, whether it might be physically difficult to use a given device in a given application, etc. Most of what's useful about it can be accomplished through observation and simple charting of the data, without the onerous calculations. The simple fact is that given the AIAG acceptance criteria, much time and effort is wasted, and people are forced into choosing between wasting time and prevarication.
We need to be able to decide when a process shift has taken place so as to avoid using measurement systems to sort good from bad. By and large, the American automotive industry wants its suppliers to engage in statistical play-acting rather than rationally controlling their processes. Any move beyond the edges of the automotive template regardless of the statistical bona fides of the movement, is anathema. The industry is presently teetering precariously on the edge of oblivion because of this kind of thinking.