🚀 go-pugleaf

Bob Drensek says:
>
> The rational is that the underling distribution is only half of a normal
> distribution, and reflects the tolerance.  Replace the Nominal with the Mean
> for the Cpk.  Is this statistically valid, no.  The issue is that the
> underling distribution is not normal.  You are in a region of statistics that
> Standard deviations technically fail.  But for back of the envelope process
> evaluation  purposes, it works reasonably well.
<<snipp>>
================
Some thoughts that may read on this discussion are that there are
some special cases of unitlateral specs which are naturally bounded
by zero.  For example, flatness where you are tracking flatness by
TIR (Total Indicator Reading aka FIM, etc).

In an unperfected process, the distribution could be normal and
displaced far enough from perfect, zero, that there is room for both
tails.  As the process is perfected, the mean starts moving towards
zero, and the below-the-mean-tail starts getting chopped off.  In
a perfect process, the mean would be zero,  your
measurement system is not going to pick up any variation and you
will have essentially no distribution.  Short of that, you can invoke
Wheeler's "Empirical Rule" part 3 (see "Understanding SPC" by Wheeler
and Chambers, p 49 ), which says that  99 to 100%  of the data will
be located within 3 sd of either side of the average.  If going below
aveage is unlikely (Xbarbar= within, say 1 sd of 0.000) then
you only have to worry about the other half of the distribution -
which is what Bob was saying and you can play the Cpk=Zusl/3 game.


<< All opinions, statements, &c are my own>>
L.H. Garlinghouse, C.Q.E.
Pocahontas AR U.S.A.
(870) 892-4586  ext 7659
garlinghou@waterlooindustries.com