I disagree with you about the role of "bonus tolerance" that I prefer to call variable limit tolerance. It is defined the way it is because it is the way the real parts work and assemble.
We agree on this. I don't have anything against the use of 'bonus' tolerances to accept and ship parts that will work. that's just good business sense. I am opposed to using for them for statistical analysis (beyond my comment below) and process improvement. Tehre are times when variation reduction is nto cost effective due to technology and organizational maturity issues. However, often we can reduce the variation and then have no need for 'bonus' tolerances...
Is it improperly applied to designs? Absolutely!!! Is it summarily ignored in continuous data statistical conformance predictions? Absolutely!!! Should we continue to ignore it for those reasons? Absolutely not!!!
uhhhh. Here is where I have issues. Using statistical capability models which rely on a fairly accurate assumption on the underlying distribution to predict future conformance is seductive, but ultimately futile in my experience. I have found that we are far better off if we understand the process, characterize the inputs and outputs and where applicable create real models of Y=f(X) form to control our processes. If we need to know how many defects we are making a simple counting of the defects we actually encounter is usually close enough for any business purpose and is far easier than distributional models. (I prefer to spend my statistical energy on
analytic studies that improve and really predict future performance than to waste energy - and time - on precise
enumerative studies of material I've already made and then use that to 'predict' future performance.)
I think you will agree that the real value is realized in examining the data and improving the process so that the probability of a defect is so remote that the process performance can put statisticians out of business.
Absolutely. except that I think statisticians are still needed...
Is Blueplanet's hole shallow or deep? What is causing such a uniform or skewed distribution for both perpendicularity and size? If the holes are shallow is the perpendicularity deviation generated mostly from measurement error or finish? If it is deep is its orientation deviation predictable... can its mean orientation error be adjusted? Why is the size skewed toward the upper limit? Do we even have enough data to make any valid conclusions???
All great questions that go to true capability; the Cpk value and it's calculation have no bearing on these questions....but they are the money making questions...