Hi Phil- In general I would agree, but in the world of semiconductors,
limits that are too wide or too narrow can spell disaster. I usually
like to keep a Cpk of 1.33~2.0 in our supply base. Anything less than
1.33 may suggest a problem exists with the product, and its consistency;
anything greater than 2.00 suggests that the limits are too wide and
that we don't understand the attributes of a good product- thus our
making 'good' product is merely by happenstance and accidental (this
opens up a whole new avenue of discussion- everyone looks at defective
product to understand why it's defective; does any one actually analyze
really great product to understand what was done to make it great?).
Manufacturing, as you rightly pointed out, quite often gets strapped
with trying to meet arbitrary Cpk goals. There are certain processes in
semiconductor manufacturing, for example, that constantly have a Cpk of
.9~1.0, and yet this is the best it will ever get because of the nature
of the equipment and materials at a given cost structure. The suggestion
here is you can have any Cpk you want if you're willing to pay for it...
These instances need to be clearly defined by the manufacturing support
personnel who have to deal with them. This is important, because
"over-managing" a process is even more disastrous than under-managment
of one. Boundary testing, both performance and damage boundary testing,
should be done and the spec limits (and ultimately Cpk's) defined from
the higher level product- this is rarely done however except in perhaps
Japan.
And so it goes....
> Bill Davis, Ph.D.
> Diamond Multimedia Systems
> Senior Scientist
> Tel. 408.325.7868
> Cell. 408.888.5650
> e-mail: [log in to unmask]
>
>
-----Original Message-----
From: Phil Crepeau [mailto:[log in to unmask]]
Sent: Thursday, February 25, 1999 12:02 PM
To: [log in to unmask]; [log in to unmask]
Subject: RE: [TN] Cpk Question
hi,
well, i don't quite get the problem with opening up the specificaton
limits.
if the design can tolerate wider limits, you should use wider limits.
this
is what design for manufacturability is all about. i don't know how
many
times the manufacturing department gets saddled with drawing tolerances
that
are tighter than they need to be. mostly it happens because the
designer
(sorry about that) use canned tolerances and don't consider opening them
up.
in fairness to designers, it's usually because they are under tremendous
time and money constraints to release their designs.
phil crepeau
-----Original Message-----
From: Bill Davis [mailto:[log in to unmask]]
Sent: Thursday, February 25, 1999 11:46 AM
To: [log in to unmask]
Subject: Re: [TN] Cpk Question
And just for what it's worth, unless you understand what you're
measuring, don't use Cp's or Cpk's (he said to the horrors of the
masses!!). This is because of the dependency of Cpk on the
specifications or limits. You can artifically drive the Cpk number
higher merely by widening the limits.
It also has a tendency, bu itself, to suggest a process is in control,
when in fact it is out of control. Take for example a process which has
a Cpk of 2; with a purely bimodal distribution at the plus and minus 1.0
sigma limits, one can get such a Cpk; however, I would submit to you
that the process which produced one mode is different that the process
which produced the other mode. You may get an inkling of bimodality by
including a first order statistic along with Cpk, such as range or
standard deviation. Probably the best method is to understand what the
Pearson number is (or Pearson set, as it's genrally called). This is
nothing more than a plot of skew v. kertosis and will tell you histogram
shape and a value which is akin to Cpk called the limit (either LL or
UL).
Enough of stats101...
> Bill Davis, Ph.D.
> Diamond Multimedia Systems
> Senior Scientist
> Tel. 408.325.7868
> Cell. 408.888.5650
> e-mail: [log in to unmask]
>
>
-----Original Message-----
From: [log in to unmask] [mailto:[log in to unmask]]
Sent: Wednesday, February 24, 1999 2:14 PM
To: [log in to unmask]
Subject: [TN] Cpk Question
Isn't the world of statistics wonderful. At some very large corporations
(often automotive), Cpk is
equated almost to the bottom line. A Cpk simply provides some evidence
of how well a process is
performing relative to whether it is in control (first) and whether it
is capable of producing
product meeting specified requirements (second). This requires a
processes' control limits to be
inside the product's specified acceptance limits. If you were to turn a
Cpk chart on its side and
match its control limits to to those of an X bar R chart, or process
control chart, you would get a
clear picture whether this was true.
The higher the Cpk's numeric value, the better is a process capable of
performing. Also, remember
that when a process is said to be in control, this only means it is
consistent. That is, it may be
effecting product either consistently good - or bad. When within the
Cpk's specified limits, it is
said to be in control and effecting product meeting specified
requirements. That's a very short
version of it.
Earl Moon
################################################################
TechNet E-Mail Forum provided as a free service by IPC using LISTSERV
1.8c
################################################################
To subscribe/unsubscribe, send a message to [log in to unmask] with
following text in the body:
To subscribe: SUBSCRIBE TechNet <your full name>
To unsubscribe: SIGNOFF TechNet
################################################################
Please visit IPC's web site (http://www.ipc.org) "On-Line Services"
section for additional information.
For technical support contact Hugo Scaramuzza at [log in to unmask] or
847-509-9700 ext.312
################################################################
################################################################
TechNet E-Mail Forum provided as a free service by IPC using LISTSERV
1.8c
################################################################
To subscribe/unsubscribe, send a message to [log in to unmask] with
following
text in the body:
To subscribe: SUBSCRIBE TechNet <your full name>
To unsubscribe: SIGNOFF TechNet
################################################################
Please visit IPC's web site (http://www.ipc.org) "On-Line Services"
section
for additional information.
For technical support contact Hugo Scaramuzza at [log in to unmask] or
847-509-9700 ext.312
################################################################
################################################################
TechNet E-Mail Forum provided as a free service by IPC using LISTSERV 1.8c
################################################################
To subscribe/unsubscribe, send a message to [log in to unmask] with following text in the body:
To subscribe: SUBSCRIBE TechNet <your full name>
To unsubscribe: SIGNOFF TechNet
################################################################
Please visit IPC's web site (http://www.ipc.org) "On-Line Services" section for additional information.
For technical support contact Hugo Scaramuzza at [log in to unmask] or 847-509-9700 ext.312
################################################################
|
