RE: The rationale for substantive knowledge of underlying process

["DeRoeck, Rich" <RDeRoeck@alphaind.com>]

	In the high volume environment I mentioned, there is an abundance of
data and the chart of choice is typically the Xbar,R or Xbar,s charts. Based
on my experience I would estimate that approx. 85-90% of charts used in this
environment are Average/Range charts. You get much more information and
knowledge from collecting small samples of measurement data on key
characteristics upstream in the process than counting defectives or defects
at the back end. Whenever I visit a manufacturing site and see lots of
attribute charts and few variable (if any) data being collected, I know that
I'm auditing an unsophisticated and weak SPC program. The XmR chart is also
not the chart of choice since the Average/Range chart is more sensitive to
process upsets using similar detection rules ASSUMING THE SUBGROUPING AND
SAMPLING FREQUENCY makes sense for that process (process knowledge
required?) If the subgrouping is incorrect, it will NOT lead to a special
cause but a false alarm or missed signal, just the thing Shewhart was trying
to minimize.
	An educated 7th grader educated in what? My 8th grade daughter is
educated but how would she know how to set up a control chart for a
plasma-etch process in a GaAs wafer fab? Do you think a little understanding
of the process would be helpful? Maybe a little knowledge about metrology,
sampling, calibration, measurement method/technique?  Nah....just throw it
on a XmR chart and interpret. Chapters 5 and 6 in Wheeler's Advanced Topics
in SPC are very useful references to a better understanding of these
important topics. 
	I never said that today's world of high speed/volume manufacturing
has altered the underlying theory of understanding variation. Of course it
hasn't, but it has made implementing SPC more changeling (check out JQT for
examples). 

Autocorrelation anyone?

Rich DeRoeck