Mike Clayton

Lighten up!  This was actually NOT from Motorola data, but was APPLIED to characterization efforts as a way of comparing dissimilar processes and ranking them for actions.  They found it in literature on 40 years of industrial research, and it was “typical” shift across many industries. 
So just study ANY process using stat methods, find the var…[Read more]

If I remember correctly, the guru at Motorola had used 40 years of research reports on manufacturing variability from many industries.
It was not Motorola-only data.  What was remarkable was that this shift (roughly) was independent of industry, but common to manufacturing methods OF THAT TIME which were mostly mechanical, mechanized, but not…[Read more]

I agree that hardness can be treated as non-destructive, IF you make multiple measurements near same site (not too close as past test changes its neighborhood slightly), then use the Shainon Plot to show that the family of variation piece to piece is MUCH less than within piece.  Of course if you have a large piece, then make measurements…[Read more]

Single sided specs that originate at ZERO are usually best handled using  Cp = Cpk by definition, then if the data looks symmetrical just use the Cpk = Mean/3 sigma of raw data….which is the distance between the Mean and Target (0) divided by 3 standard deviations.  Report Cp as the same if required, addding a footnote that Cp is really u…[Read more]

USL and LSL are the upper and lower SPECIFICATION limits used in Cp and Cpk.  And they are the Achilles heel of this whole metric.   Often they are “American style” limits, which tend to be simply 10% of nominal.  Useless in many cases, but common.  So Cp and Cpk can be >5 for example in computerized testing, but parts can cause field fail…[Read more]

http://www.itl.nist.gov/div898/handbook/mpc/  is a link to the NIST-Sematech online textbook, this part on measurement systems analysis.
The over-simplified R&R studies often miss the point that measurement systems have many sources of variation that need study and control.   Bias, linearity, stability over time, etc.   So even with one opera…[Read more]