🚀 go-pugleaf

Roger Duffy posted (in part):

>I have a real-time SPC system in place on several CNC machines....
> a part produced
>on one machine this month may be made on a different machine the next
>time it is made....
>...
>
>Is there a way to establish one index for each machine that would
>reflect the overall machine performance, say over the tolerance range
>0.0002" to 0.001".  I'd like to have one number that would reflect the
>performance that we could track monthly and use to compare the
>variability and performance of one machine to another.

What is presented here is akin to a blocked, and perhaps nested factorial
experiment. What needs to be established first is the constancy of the
tool(s) (a fixed factor at worst assuming the various speeds are constant
for a given tool) involved. Second, establish that the underlying
distributions are approximately normal. Should that be the case then
shortest path to this "composite" index is to test for equality of
variances for each group of parts, i.e. for each combination of tool,
machine and part. Should that be the case then normalize the data for
each group, pool the data by machine and then calculate the composite
index for each machine.

Be advised that the ground under which Cpk stands is not the most
statistically stable and this composite index will make most
statisticians--and not a few engineers--feel they are astride a San
Andreas fault filled with the remains of Fisher, Gossett, Pearson et. al.

If it is necessary to sell management on the capability of the equipment
based upon single number (or series thereof), a dubious proposition at
best, then a "better" approach would be simply to calculate empirical
distributions and estimate the ppm defective.

Kwenda mzuri,

Frank Isackson
re:SOLUTIONS
205 West Walnut Avenue
Unit A
Monrovia, CA 91016
(626) 358-1340 (voice)
(626) 358-1322 (fax)
FIsackson@aol.com